BR122025007928A2 - VIDEO DECODING METHOD AND VIDEO CODING METHOD - Google Patents

Abstract

An image decoding method according to the present invention may comprise the steps of: determining whether a combined prediction mode is applied to a current block; when the combined prediction mode is applied to the current block, obtaining first and second prediction blocks relative to the current block; and, based on a calculation of a weighted sum of the first and second prediction blocks, obtaining a third prediction block relative to the current block.

Description

BACKGROUND OF THE INVENTION FIELD OF INVENTION

[001] The present invention relates to a method of encoding and decoding video signal and an apparatus for the same.

FUNDAMENTALS OF THE RELATED TECHNIQUE

[002] As display panels become increasingly larger, even higher-quality video services become increasingly necessary. The biggest challenge facing high-definition video services is the significant increase in data volume, and to address this issue, research is underway to improve video compression. As a representative example, the Motion Picture Experts Group (MPEG) and the Video Coding Experts Group (VCEG) under the International Telecommunication Union-Telecommunications (ITU-T) formed the Joint Collaboration Team on Video Coding (JCT-VC) in 2009. JCT-VC proposed High Efficiency Video Coding (HEVC), which is a video compression standard with a compression performance about twice that of H.264/AVC, and was approved as a standard on January 25, 2013. With the rapid advancement of high-definition video services, the performance of HEVC gradually reveals its limitations.

SUMMARY OF THE INVENTION

[003] An object of the present invention is to provide a combined prediction method that combines a plurality of prediction methods in encoding/decoding a video signal, and an apparatus for performing the method.

[004] An object of the present invention is to provide a method of partitioning a coding block into a plurality of prediction units in encoding/decoding a video signal, and an apparatus for performing the method.

[005] The technical problems to be achieved in the present invention are not limited to the technical problems mentioned above, and other problems not mentioned can be clearly understood by those skilled in the art from the following description.

[006] A method of decoding/encoding a video signal according to the present invention includes the steps of: determining whether or not to apply a combined prediction mode to a current block; obtaining a first prediction block and a second prediction block for the current block when the combined prediction mode is applied to the current block; and obtaining a third prediction block for the current block based on a weighted sum operation of the first prediction block and the second prediction block. At this point, the first prediction block may be obtained based on the motion information of a merge candidate of the current block, and the second prediction block may be obtained based on an intra-prediction mode of the current block.

[007] In the video signal encoding and decoding method according to the present invention, it can be configured that applying triangular partitioning to the current block is not allowed when the combined prediction mode is applied to the current block.

[008] In the video signal encoding and decoding method according to the present invention, the intraprediction mode of the current block can be set as a flat mode.

[009] In the video signal encoding and decoding method according to the present invention, the second prediction block can be obtained based on a reference sample line included in an adjacent reference sample line.

[010] In the video signal encoding and decoding method according to the present invention, when performing the weighted sum operation, weighting values applied to the first prediction block and the second prediction block can be determined based on the prediction coding modes of neighboring blocks adjacent to the current block.

[011] In the video signal encoding and decoding method according to the present invention, when at least one of the width and height of the current block is greater than a threshold value, the combined prediction mode may not be applied to the current block.

[012] In the video signal encoding and decoding method according to the present invention, the blended prediction mode can be configured to be applicable to the current block when a flag indicating that a blending mode is applied to the current block is true.

[013] The features briefly summarized above in relation to the present invention are merely exemplary aspects of the detailed description of the present invention that will be described below and do not limit the scope of the present invention.

[014] According to the present invention, interforecasting efficiency can be improved by providing a combined forecasting method that combines a plurality of forecasting methods.

[015] According to the present invention, interprediction efficiency can be improved by proposing a method of partitioning a coding block into a plurality of prediction blocks, and deriving motion information from each of the prediction blocks.

[016] The effects that can be obtained from the present invention are not limited to the effects mentioned above, and other effects not mentioned can be clearly understood by those skilled in the art from the following description.

BRIEF DESCRIPTION OF THE DRAWINGS

[017] Figure 1 is a block diagram showing a video encoder in accordance with an embodiment of the present invention.

[018] Figure 2 is a block diagram showing a video decoder in accordance with an embodiment of the present invention.

[019] Figure 3 is a view showing a basic coding tree unit in accordance with an embodiment of the present invention.

[020] Figure 4 is a view showing various types of partitioning of a coding block.

[021] Figure 5 is a view showing a partitioning pattern of a coding tree unit.

[022] Figure 6 is a view showing the shape of a basic data unit.

[023] Figures 7 and 8 are views showing examples of partitioning a coding block into a plurality of sub-blocks.

[024] Figure 9 is a flowchart illustrating an interprediction method in accordance with an embodiment of the present invention.

[025] Figure 10 is a flowchart illustrating a process of deriving motion information from a current block using a blending mode.

[026] Figure 11 is a view showing an example of candidate blocks used to derive a merge candidate.

[027] Figure 12 is a view showing examples of partitioning a coding block into a plurality of prediction units using a diagonal line.

[028] Figure 13 is a view showing examples of partitioning a coding block into two prediction units.

[029] Figure 14 is a view showing examples of partitioning a coding block into a plurality of prediction blocks of different sizes.

[030] Figure 15 is a view showing an example of applying diagonal partitioning to a transform unit.

[031] Figure 16 is a view showing neighboring blocks used to derive a triangular merge candidate.

[032] Figure 17 is a view to describe examples of determining availability of a neighboring block for each triangular forecast unit.

[033] Figures 18 and 19 are views showing examples of deriving a forecast sample based on a weighted sum operation of a first forecast sample and a second forecast sample.

[034] Figure 20 is a flowchart illustrating an intraforecast method in accordance with an embodiment of the present invention.

[035] Figure 21 is a view showing intraprediction modes.

[036] Figures 22 and 23 are views showing an example of a one-dimensional array arranging reference samples in a row.

[037] Figure 24 is a view showing an example of angles formed between angular intraprediction modes and a straight line parallel to the x-axis.

[038] Figure 25 is a view showing an embodiment of acquiring prediction samples when the current block has a non-square shape.

[039] Figure 26 is a view showing wide-angle intraprediction modes.

[040] Figure 27 is a view showing an embodiment of PDPC application.

[041] Figure 28 is a view showing application of examples of weighting values.

[042] Figure 29 is a view showing examples of assigning a different weighting value to a sub-block according to whether a merge candidate has bidirectional motion information.

[043] Figure 30 is a view showing examples of assigning a different weighting value to a sub-block according to the intraprediction mode of the current block.

[044] Figure 31 is a view showing an example where a second merge candidate is specified considering the search order of candidate blocks.

[045] Figure 32 is a view showing an example in which a first merge candidate and a second merge candidate are selected from merge candidates derived from non-neighboring blocks.

[046] Figure 33 is a view to describe an example of determining a second merge candidate.

[047] Figure 34 is a view showing an example where weighting values applied to prediction blocks are determined based on the shape of candidate blocks.

[048] Figure 35 is a view showing an example where a merge candidate is replaced.

[049] Figures 36 and 37 are views showing a sub-block in which a second transform will be performed.

[050] Figure 38 is a view to describe examples of determining a transform type of the current block.

[051] Figure 39 is a flowchart illustrating a block strength determination process.

[052] Figure 40 is a view showing predefined filter candidates.

DETAILED DESCRIPTION OF THE PREFERRED MODALITY

[053] Hereinafter, an embodiment of the present invention will be described in detail with reference to the accompanying drawings.

[054] The encoding and decoding of a video is performed by the block unit. For example, an encoding/decoding process such as transform, quantization, prediction, loop filtering, reconstruction, or the like can be performed in a coding block, a transform block, or a prediction block.

[055] Next, a block to be encoded/decoded will be referred to as a “current block”. For example, the current block may represent an encoding block, a transform block, or a prediction block according to a current encoding/decoding process step.

[056] Furthermore, it may be understood that the term “unit” used in this specification indicates a basic unit for performing a specific encoding/decoding process, and the term “block” indicates a sample matrix of a predetermined size. Unless otherwise indicated, “block” and “unit” may be used to have the same meaning. For example, in an embodiment described below, it may be understood that a coding block and a coding unit have the same meaning.

[057] Figure 1 is a block diagram showing a video encoder in accordance with an embodiment of the present invention.

[058] Referring to Figure 1, a video coding apparatus 100 may include a picture partitioning part 110, a prediction part 120 and 125, a transform part 130, a quantization part 135, a rearrangement part 160, an entropy coding part 165, an inverse quantization part 140, an inverse transform part 145, a filter part 150, and a memory 155.

[059] Each of the components shown in Figure 1 is shown independently to represent characteristic functions different from each other in a video encoding apparatus, and this does not mean that each component is formed by the separate hardware configuration unit or single software. That is, each component is included to be listed as a component for convenience of explanation, and at least two of the components can be combined to form a single component, or a component can be divided into a plurality of components to perform a function. Integrated embodiments and separate embodiments of the components are also included in the scope of the present invention, if they do not depart from the essence of the present invention.

[060] Furthermore, some of the components are not essential components that perform essential functions in the present invention, but may be optional components only for improving performance. The present invention may be implemented by including only essential components for implementing the essence of the present invention, excluding components used for improving performance, and a structure including only essential components, excluding optional components used for improving performance, is also included in the scope of the present invention.

[061] The image partitioning part 110 may partition an input image into at least one processing unit. At this point, the processing unit may be a prediction unit (PU), a transform unit (TU), or a coding unit (CU). The image partitioning part 110 may partition an image into a combination of a plurality of coding units, prediction units, and transform units, and encode an image by selecting a combination of a coding unit, a prediction unit, and a transform unit based on a predetermined criterion (e.g., a cost function).

[062] For example, an image may be partitioned into a plurality of coding units. In order to partition the coding units in an image, a recursive tree structure such as a quad tree structure may be used. A video or a coding unit partitioned into different coding units using the largest coding unit as a root may be partitioned to have as many child nodes as the number of partitioned coding units. A coding unit that is no longer partitioned according to a predetermined constraint becomes a leaf node. That is, when it is assumed that only square partitioning is possible for a coding unit, the coding unit may be partitioned into up to four different coding units.

[063] Next, in one embodiment of the present invention, the coding unit may be used as a meaning of a unit that performs encoding or a meaning of a unit that performs decoding.

[064] The prediction unit may be one that is partitioned into a shape of at least one square, rectangle, or the like of the same size within a coding unit, or it may be any one prediction unit, among the prediction units partitioned within a coding unit, that is partitioned to have a shape and/or size different from those of another prediction unit.

[065] If the coding unit is not a smaller coding unit when a prediction unit that performs intraprediction based on the coding unit is generated, intraprediction can be performed without partitioning an image into a plurality of N x N prediction units.

[066] The prediction portions 120 and 125 may include an inter-prediction portion 120 that performs inter-prediction and an intra-prediction portion 125 that performs intra-prediction. It may be determined whether to use inter-prediction or perform intra-prediction for a prediction unit, and specific information (e.g., intra-prediction mode, motion vector, reference image, etc.) may be determined according to each prediction method. At this point, a processing unit for performing prediction may be different from a processing unit for determining a prediction method and specific content. For example, a prediction method and a prediction mode may be determined in a prediction unit, and prediction may be performed in a transform unit. A residual coefficient (residual block) between the generated prediction block and the original block may be input to the transform portion 130. Furthermore, prediction mode information, motion vector information, and the like used for prediction may be encoded by the entropy coding portion 165 together with the residual coefficient and passed to a decoder. When a specific encoding mode is used, an original block may be encoded as is and transmitted to a decoder without generating a prediction block through prediction part 120 and 125.

[067] The interprediction part 120 may predict a prediction unit based on information in at least one picture among pictures before or after the current picture, and in some cases, may predict a prediction unit based on information in a partial area that has been encoded in the current picture. The interprediction part 120 may include a reference picture interpolation part, a motion prediction part, and a motion compensation part.

[068] The reference image interpolation part may receive reference image information from memory 155 and generate pixel information of an integer number of pixels or less from the reference image. In the case of a luminance pixel, a DCT-based 8-coefficient interpolation filter with a variable filter coefficient may be used to generate pixel information of an integer number of pixels or less by the unit of 1/4 pixels. In the case of a chroma signal, a DCT-based 4-coefficient interpolation filter with a variable filter coefficient may be used to generate pixel information of an integer number of pixels or less by the unit of 1/8 pixels.

[069] The motion prediction part may perform motion prediction based on the reference image interpolated by the reference image interpolation part. Various methods, such as a full-search-based block matching algorithm (FBMA), a three-step search (TSS), and a novel three-step search (NTS) algorithm, may be used as a motion vector calculation method. The motion vector may have a motion vector value of a 1/2 or 1/4 pixel unit based on interpolated pixels. The motion prediction part may predict a current prediction unit by varying the motion prediction method. Various methods, such as a skip mode, a blend mode, an advanced motion vector prediction (AMVP) mode, an intra-block copy mode, and the like, may be used as the motion prediction mode.

[070] The intraprediction part 125 may generate a prediction unit based on information about reference pixels in the neighborhood of the current block, which is pixel information in the current image. When a block in the neighborhood of the current prediction unit is a block on which interprediction has been performed, and thus the reference pixel is a pixel on which interprediction has been performed, the reference pixel included in the block on which interprediction has been performed may be used in place of reference pixel information of a block in the neighborhood on which intraprediction has been performed. That is, when a reference pixel is not available, at least one reference pixel among available reference pixels may be used in place of unavailable reference pixel information.

[071] In intraprediction, the prediction mode may have an angular prediction mode that uses reference pixel information according to a prediction direction and a non-angular prediction mode that does not use directional information when performing prediction. A mode for predicting luminance information may be different from a mode for predicting chroma information, and intraprediction mode information used to predict luminance information or predicted luminance signal information may be used to predict the chroma information.

[072] If the size of the prediction unit is equal to the size of the transform unit when intraprediction is performed, intraprediction can be performed for the prediction unit based on a pixel on the left side, a pixel on the top left side, and a pixel on the top of the prediction unit. However, if the size of the prediction unit is different from the size of the transform unit when intraprediction is performed, intraprediction can be performed using a reference pixel based on the transform unit. Furthermore, intraprediction using N x N partitioning can be used only for the smallest coding unit.

[073] The intra-prediction method may generate a prediction block after applying an intra-adaptive smoothing (AIS) filter to the reference pixel according to a prediction mode. The type of the AIS filter applied to the reference pixel may vary. In order to perform the intra-prediction method, the intra-prediction mode of the current prediction unit may be predicted from the intra-prediction mode of the prediction unit existing in the neighborhood of the current prediction unit. When a prediction mode of the current prediction unit is predicted using the mode information predicted from the neighboring prediction unit, if the intra-prediction modes of the current prediction unit are the same as those of the neighboring prediction unit, information indicating that the prediction modes of the current prediction unit are the same as those of the neighboring prediction unit may be transmitted using predetermined flag information, and if the prediction modes of the current prediction unit and the neighboring prediction unit are different from each other, prediction mode information of the current block may be encoded by performing entropy encoding.

[074] Further, a residual block including a prediction unit that performed prediction based on the prediction unit generated by the prediction part 120 and 125 and residual coefficient information, which is a difference value of the prediction unit with the original block, may be generated. The generated residual block may be input to the transform part 130.

[075] The transform part 130 may transform the residual block including the original block and the residual coefficient information of the prediction unit generated through the prediction part 120 and 125 using a transform method such as Discrete Cosine Transform (DCT) or Discrete Sine Transform (DST). Here, the DCT transform kernel includes at least one of DCT2 and DCT8, and the DST transform kernel includes DST7. Whether or not to apply DCT or DST to the transform of the residual block may be determined based on the intraprediction mode information of a prediction unit used to generate the residual block. The transform in the residual block may be skipped. A flag indicating whether or not to skip the transform in the residual block may be encoded. The transform skip may be left for a residual block that has a size less than or equal to a threshold, a luma component, or a chroma component under the 4:4:4 format.

[076] Quantization part 135 may quantize values transformed into the frequency domain by transform part 130. Quantization coefficients may vary depending on the block or importance of a video. A value calculated by quantization part 135 may be provided to inverse quantization part 140 and rearrangement part 160.

[077] The rearrangement part 160 may rearrange coefficient values for the quantized residual coefficients.

[078] The rearrangement portion 160 may shift coefficients of a two-dimensional block format into a one-dimensional vector format by means of a coefficient sweep method. For example, the rearrangement portion 160 may sweep coefficients from DC to high-frequency domain coefficients using a zigzag sweep method, and shift the coefficients into a one-dimensional vector format. According to the size of the transform unit and the intraprediction mode, a vertical sweep of sweeping the coefficients of a two-dimensional block format in the column direction and a horizontal sweep of sweeping the coefficients of a two-dimensional block format in the row direction may be used instead of the zigzag sweep. That is, according to the size of the transform unit and the intraprediction mode, a sweep method to be used may be determined among the zigzag sweep, the sweep in the vertical direction, and the sweep in the horizontal direction.

[079] The entropy coding part 165 may perform entropy coding based on the values calculated by the rearrangement part 160. Entropy coding may use various coding methods such as Exponential Golomb, Context Adaptive Variable Length Coding (CAVLC), Context Adaptive Binary Arithmetic Coding (CABAC), and the like.

[080] The entropy coding part 165 may encode various information such as residual coefficient information and block type information of a coding unit, prediction mode information, partitioning unit information, prediction unit information and transmission unit information, motion vector information, reference frame information, block interpolation information, and filtering information input from the rearrangement part 160 and the prediction parts 120 and 125.

[081] The entropy encoding part 165 may entropy encode the coefficient value of a coding unit input from the rearrangement part 160.

[082] The inverse quantization part 140 and the inverse transform part 145 inversely quantize the values quantized by the quantization part 135 and inversely transform the values transformed by the transform part 130. The residual coefficient generated by the inverse quantization part 140 and the inverse transform part 145 may be combined with the predicted prediction unit via a motion estimation part, a motion compensation part, and an intraprediction part included in the prediction part 120 and 125 to generate a reconstructed block.

[083] The filter portion 150 may include at least one of a deblocking filter, an offset correction unit, and an adaptive loop filter (ALF).

[084] The deblocking filter can remove block distortion generated by the boundary between blocks in the reconstructed image. In order to determine whether or not to perform deblocking, whether or not to apply the deblocking filter to the current block can be determined based on the pixels included in multiple columns or rows included in the block. A strong filter or a weak filter can be applied according to the required deblocking filtering strength when the deblocking filter is applied to a block. Furthermore, when vertical direction filtering and horizontal direction filtering are performed when applying the deblocking filter, the horizontal direction filtering and vertical direction filtering can be processed in parallel.

[085] The displacement correction unit can correct a displacement to the original image by the pixel unit for an image on which deblocking has been performed. In order to perform displacement correction for a specific image, it is possible to use a method of dividing pixels included in the image into a certain number of areas, determining an area to perform the displacement, and applying the displacement to the area, or a method of applying a displacement considering edge information of each pixel.

[086] Adaptive Loop Filtering (ALF) can be performed based on a value obtained by comparing the filtered reconstructed image with the original image. After dividing the pixels included in the image into predetermined groups, a filter to be applied to a corresponding group can be determined, and filtering can be performed differently for each group. A luminance signal, which is the information related to whether or not ALF is applied, can be transmitted to each coding unit (CU), and the shape and filter coefficient of an ALF filter to be applied can vary according to each block. Furthermore, an ALF filter of the same type (fixed type) can be applied regardless of the characteristic of a block to be applied.

[087] The memory 155 may store the reconstructed block or image calculated through the filter part 150, and the reconstructed and stored block or image may be provided to the prediction part 120 and 125 when interprediction is performed.

[088] Figure 2 is a block diagram showing a video decoder in accordance with an embodiment of the present invention.

[089] Referring to Figure 2, a video decoder 200 may include an entropy decoding part 210, a rearrangement part 215, an inverse quantization part 220, an inverse transform part 225, a prediction part 230 and 235, a filter part 240, and a memory 245.

[090] When a video bitstream is input from a video encoder, the input bitstream may be decoded in a procedure opposite to that of the video encoder.

[091] The entropy decoding part 210 may perform entropy decoding in a procedure opposite to that which performs entropy coding in the entropy decoding part of the video encoder. For example, various methods corresponding to the method performed by the video encoder, such as Exponential Golomb, Context Adaptive Variable Length Coding (CAVLC), and Context Adaptive Binary Arithmetic Coding (CABAC), may be applied.

[092] The entropy decoding part 210 can decode information related to the intraprediction and interprediction performed by the encoder.

[093] The rearrangement part 215 may perform rearrangement on the bit stream entropy decoded by the entropy decoding part 210 based on the rearrangement method performed by the encoder. The coefficients expressed in a one-dimensional vector form may be reconstructed and rearranged as two-dimensional block form coefficients. The rearrangement part 215 may receive information related to the coefficient scan performed by the coding part and perform reconstruction by a backward scan method based on the scan order performed by the corresponding coding part.

[094] The inverse quantization part 220 may perform inverse quantization based on a quantization parameter provided by the encoder and a coefficient value of the rearranged block.

[095] The inverse transform part 225 may perform inverse transform on the transform, i.e., DCT or DST, performed by the transform part on a result of quantization performed by the video encoder, i.e., inverse DCT or inverse DST. Here, the DCT transform kernel may include at least one of DCT2 and DCT8, and the DST transform kernel may include DST7. Alternatively, when the transform is skipped in the video encoder, even the inverse transform part 225 may not perform inverse transform. The inverse transform may be performed based on a transmission unit determined by the video encoder. The inverse transform part 225 of the video decoder may selectively perform a transform technique (e.g., DCT or DST) according to a plurality of information such as a prediction method, a size of a current block, a prediction direction, and the like.

[096] The prediction part 230 and 235 may generate a prediction block based on information related to generating a prediction block provided by the entropy decoder 210 and information about a previously decoded block or image provided by the memory 245.

[097] As described above, if the size of the prediction unit and the size of the transform unit are the same when intraprediction is performed in the same manner as the video encoder operation, intraprediction is performed on the prediction unit based on the pixel on the left side, the pixel on the upper left side, and the pixel on the top of the prediction unit. However, if the size of the prediction unit and the size of the transform unit are different when intraprediction is performed, intraprediction can be performed using a reference pixel based on a transform unit. Furthermore, intraprediction using N x N partitioning can be used only for the smaller coding unit.

[098] The prediction part 230 and 235 may include a prediction unit determination part, an inter-prediction part, and an intra-prediction part. The prediction unit determination part may receive various information as input—prediction unit information from the entropy decoding part 210, prediction mode information from the intra-prediction method, motion prediction-related information from an inter-prediction method, and the like, identify the prediction unit from the current coding unit, and determine whether the prediction unit performs inter-prediction or intra-prediction. The inter-prediction part 230 may perform inter-prediction on the current prediction unit based on information included in at least one picture among pictures before or after the current picture including the current prediction unit using information necessary for inter-prediction of the current prediction unit provided by the video encoder. Alternatively, the inter-prediction part 230 may perform inter-prediction based on information in a previously reconstructed partial area in the current picture including the current prediction unit.

[099] In order to perform inter-prediction, it may be determined, based on the coding unit, whether the motion prediction method of the prediction unit included in a corresponding coding unit is a skip mode, a merge mode, an advanced motion vector prediction mode (AMVP mode), or an intra-block copy mode.

[100] The intra-prediction part 235 may generate a prediction block based on information about the pixel in the current picture. When the prediction unit is a prediction unit that has performed intra-prediction, the intra-prediction may be performed based on the prediction unit's intra-prediction mode information provided by the video encoder. The intra-prediction part 235 may include an intra-adaptive smoothing (AIS) filter, a reference pixel interpolation part, and a DC filter. The AIS filter is a part that performs filtering on the reference pixel of the current block, and may determine whether or not to apply the filter according to the prediction mode of the current prediction unit and apply the filter. AIS filtering may be performed on the reference pixel of the current block using the prediction mode and AIS filter information of the prediction unit provided by the video encoder. When the prediction mode of the current block is a mode that does not perform AIS filtering, the AIS filter may not be applied.

[101] When the prediction mode of the prediction unit is a prediction unit that performs intraprediction based on a pixel value obtained by interpolating the reference pixel, the reference pixel interpolation part may generate a reference pixel of a pixel unit that has an integer value or a value less than the integer value by interpolating the reference pixel. When the prediction mode of the current prediction unit is a prediction mode that generates a prediction block without interpolating the reference pixel, the reference pixel may not be interpolated. The DC filter may generate a prediction block through filtering when the prediction mode of the current block is DC Mode.

[102] The reconstructed block or image may be provided to the filter portion 240. The filter portion 240 may include a deblocking filter, an offset correction unit, and an ALF.

[103] Information about whether a deblocking filter is applied to a corresponding block or picture and information about whether a strong filter or a weak filter is applied when a deblocking filter is applied may be provided by the video encoder. The deblocking filter of the video decoder may be provided with information related to the deblocking filter provided by the video encoder, and the video decoder may perform deblocking filtering on a corresponding block.

[104] The shift correction unit can perform shift correction on the reconstructed image based on the shift correction type and shift value information applied to the video when encoding is performed.

[105] ALF may be applied to a coding unit based on information about whether or not to apply ALF and information about ALF coefficients provided by the encoder. ALF information may be provided to be included in a specific parameter set.

[106] Memory 245 may store the reconstructed image or block and use it as a reference image or a reference block and may provide the reconstructed image to an output unit.

[107] Figure 3 is a view showing a basic coding tree unit in accordance with an embodiment of the present invention.

[108] A coding block of a maximum size can be defined as a coding tree block. An image is partitioned into a plurality of Coding Tree Units (CTUs). A coding tree unit is a coding unit that has a maximum size and can be termed as a Large Coding Unit (LCU). Figure 3 shows an example where an image is partitioned into a plurality of coding tree units.

[109] The size of the coding tree unit may be set at a picture level or a sequence level. To this end, information indicating the size of the coding tree unit may be signaled via a picture parameter set or a sequence parameter set.

[110] For example, the coding tree unit size for the entire image in a sequence may be set to 128 x 128. Alternatively, at the image level, any of 128 x 128 and 256 x 256 may be determined as the coding tree unit size. For example, the coding tree unit size may be set to 128 x 128 on a first image, and the coding tree unit size may be set to 256 x 256 on a second image.

[111] Coding blocks can be generated by partitioning a coding tree unit. The coding block indicates a basic unit for performing encoding/decoding. For example, prediction or transform can be performed for each coding block, or a prediction coding mode can be determined for each coding block. Here, the prediction coding mode indicates a method of generating a prediction image. For example, the prediction coding mode can include prediction within an image (intra-prediction), prediction between images (inter-prediction), current image reference (CPR) or intra-block copy (IBC), or combined prediction. For the coding block, a prediction block can be generated using at least one prediction coding mode among intra-prediction, inter-prediction, current image reference, and combined prediction.

[112] Information indicating the prediction coding mode of the current block may be signaled via a bit stream. For example, the information may be a 1-bit flag indicating whether the prediction coding mode is an intra mode or an inter mode. Only when the prediction coding mode of the current block is determined to be the inter mode can either the current picture reference or the combined prediction be used.

[113] The current picture reference is to set the current picture as a reference picture and obtain a prediction block of the current block from an area that has already been coded/decoded in the current picture. Here, the current picture means a picture including the current block. Information indicating whether the current picture reference is applied to the current block can be signaled via a bit stream. For example, the information can be a 1-bit flag. When the flag is true, the prediction coding mode of the current block can be determined as the current picture reference, and when the flag is false, the prediction mode of the current block can be determined as interprediction.

[114] Alternatively, the prediction coding mode of the current block can be determined based on a reference picture index. For example, when the reference picture index indicates the current picture, the prediction coding mode of the current block can be determined as the current picture reference. When the reference picture index indicates a picture other than the current picture, the prediction coding mode of the current block can be determined as interprediction. That is, the current picture reference is a prediction method using the information in an area where encoding/decoding has been completed in the current picture, and interprediction is a prediction method using the information in another picture where encoding/decoding has been completed.

[115] Combined prediction represents a coding mode in which two or more of the intra prediction, the inter prediction, and the current picture reference are combined. For example, when combined prediction is applied, a first prediction block may be generated based on one of the intra prediction, the inter prediction, and the current picture reference, and a second prediction block may be generated based on another. When the first prediction block and the second prediction block are generated, a final prediction block may be generated through an average operation or a weighted sum operation of the first prediction block and the second prediction block. Information indicating whether or not combined prediction is applied may be signaled through a bit stream. The information may be a 1-bit flag.

[116] Figure 4 is a view showing various types of partitioning of a coding block.

[117] The coding block may be partitioned into a plurality of coding blocks based on quad tree partitioning, binary tree partitioning, or ternary tree partitioning. The partitioned coding block may be partitioned again into a plurality of coding blocks based on quad tree partitioning, binary tree partitioning, or ternary tree partitioning.

[118] Quad-tree partitioning refers to a partitioning technique that partitions a current block into four blocks. As a result of quad-tree partitioning, the current block can be partitioned into four square-shaped partitions (see, “SPLIT_QT” of Figure 4(a)).

[119] Binary tree partitioning refers to a partitioning technique that partitions a current block into two blocks. Partitioning a current block into two blocks along the vertical direction (i.e., using a vertical line crossing the current block) can be termed as vertical-direction binary tree partitioning, and partitioning a current block into two blocks along the horizontal direction (i.e., using a horizontal line crossing the current block) can be termed as horizontal-direction binary tree partitioning. As a result of binary tree partitioning, the current block can be partitioned into two non-square-shaped partitions. “SPLIT_BT_VER” of Figure 4(b) shows a result of vertical-direction binary tree partitioning, and “SPLIT_BT_HOR” of Figure 4(c) shows a result of horizontal-direction binary tree partitioning.

[120] Ternary tree partitioning refers to a partitioning technique that partitions a current block into three blocks. Partitioning a current block into three blocks along the vertical direction (i.e., using two vertical lines crossing the current block) can be termed as vertical-direction ternary tree partitioning, and partitioning a current block into three blocks along the horizontal direction (i.e., using two horizontal lines crossing the current block) can be termed as horizontal-direction ternary tree partitioning. As a result of ternary tree partitioning, the current block can be partitioned into three non-square-shaped partitions. At this point, the width/height of a partition positioned at the center of the current block can be twice as large as the width/height of the other partitions. “SPLIT_TT_VER” of Figure 4(d) shows a result of vertical direction ternary tree partitioning, and “SPLIT_TT_HOR” of Figure 4(e) shows a result of horizontal direction ternary tree partitioning.

[121] The number of times a coding tree unit is partitioned can be defined as a partitioning depth. The maximum partitioning depth of a coding tree unit can be determined at the sequence or image level. Consequently, the maximum partition depth of a coding tree unit can be different for each sequence or image.

[122] Alternatively, the maximum partitioning depth for each partitioning technique may be determined individually. For example, the maximum allowable partitioning depth for quad tree partitioning may be different from the maximum allowable partitioning depth for binary tree partitioning and/or ternary tree partitioning.

[123] The encoder may signal information indicating at least one of the partitioning type and partitioning depth of the current block via a bit stream. The decoder may determine the partitioning type and partitioning depth of a coding tree unit based on information parsed from the bit stream.

[124] Figure 5 is a view showing a partitioning pattern of a coding tree unit.

[125] Partitioning a cipher block using a partitioning technique such as quad tree partitioning, binary tree partitioning, and/or ternary tree partitioning can be termed as multi-tree partitioning.

[126] The coding blocks generated by applying multi-tree partitioning to a coding block can be termed as lower coding blocks. When the partitioning depth of a coding block is k, the partitioning depth of the lower coding blocks is set to k + 1.

[127] Conversely, for coding blocks that have a partitioning depth of k + 1, a coding block that has a partitioning depth of k can be termed as an upper coding block.

[128] The partitioning type of the current coding block may be determined based on at least one of a partitioning type of a top coding block and a partitioning type of a neighboring coding block. Here, the neighboring coding block is a coding block adjacent to the current coding block, and may include at least one of a top neighboring block and a left neighboring block of the current coding block, and a top-left adjacent neighboring block. Here, the partitioning type may include at least one of a quad tree partitioning or not, whether or not it is a binary tree partitioning, binary tree partitioning direction, whether or not it is a ternary tree partitioning, and ternary tree partitioning direction.

[129] In order to determine a partitioning type of a cipher block, information indicating whether or not the cipher block can be partitioned can be signaled via a bit stream. The information is a 1-bit flag “split_cu_flag”, and when the flag is true, it indicates that the cipher block is partitioned by a quad-tree partitioning technique.

[130] When split_cu_flag is true, information indicating whether the cipher block is quad-tree partitioned can be signaled via a bit stream. The information is a 1-bit flag of split_qt_flag, and when the flag is true, the cipher block can be partitioned into four blocks.

[131] For example, in the example shown in Figure 5, as a coding tree unit is quad-tree partitioned, four coding blocks that have a partitioning depth of 1 are generated. Furthermore, it is shown that quad-tree partitioning is applied again to the first and fourth coding blocks among the four coding blocks generated as a result of quad-tree partitioning. As a result, four coding blocks that have a partitioning depth of 2 can be generated.

[132] Furthermore, cipher blocks that have a partitioning depth of 3 can be generated by applying quad-tree partitioning again to a cipher block that has a partitioning depth of 2.

[133] When quad-tree partitioning is not applied to the coding block, whether binary tree partitioning or ternary tree partitioning is performed on the coding block can be determined by considering at least one of the coding block size, whether the coding block is positioned on the image boundary, the maximum partitioning depth, and the partitioning type of a neighboring block. When it is determined to perform binary tree partitioning or ternary tree partitioning on the coding block, information indicating the partitioning direction can be signaled via a bit stream. The information can be a 1-bit flag of mtt_split_cu_vertical_flag. Based on the flag, whether the partitioning direction is a vertical direction or a horizontal direction can be determined. Furthermore, information indicating whether binary tree partitioning or ternary tree partitioning is applied to the coding block can be signaled via a bit stream. The information can be a 1-bit flag from mtt_split_cu_binary_flag. Based on the flag, whether binary tree partitioning or ternary tree partitioning is applied to the cipher block can be determined.

[134] For example, in the example shown in Figure 5, it is shown that vertical-directed binary tree partitioning is applied to a coding block that has a partitioning depth of 1, vertical-directed ternary tree partitioning is applied to the coding block on the left side among the coding blocks generated as a result of the partitioning, and vertical-directed binary tree partitioning is applied to the coding block on the right side.

[135] When implementing a video encoding or decoding device, there is a problem where a region larger than a threshold value is difficult to process due to hardware performance. For example, when it is possible to simultaneously process up to 4,096 samples based on hardware performance, data units of size 64 x 64 must be accessed and processed redundantly, and data cannot be processed simultaneously for regions with samples larger than 4,096. Thus, a basic data processing unit can be defined as a pipeline-based basic data unit (Virtual Processing Data Unit, VPDU, hereinafter referred to as a basic data unit).

[136] The basic data unit can be classified as square, non-square, or non-rectangular.

[137] Figure 6 is a view showing the shape of a basic data unit.

[138] Basic data units may include samples as large as or smaller than the maximum number of samples that can be processed simultaneously. For example, as shown in the example in Figure 6(a), square blocks with a size of 64 x 64 can be defined as basic data units. Alternatively, non-square blocks can be defined as basic data units. For example, as shown in the example in Figure 6(b) or 6(c), a block that has a size of 32 x 128 or a block that has a size of 64 x 32 can be defined as a basic data unit.

[139] Although not shown, basic triangular, L-shaped, and polygonal data units can be defined.

[140] Information for determining a basic data unit may be signaled via a bit stream. The information may determine at least one of the size and shape of the basic data unit. Based on the information, it may be determined whether or not to allow a non-square basic data unit or whether to allow a non-rectangular basic data unit.

[141] Alternatively, at least one of the size and format of a basic unit of data may be predefined in the encoder and decoder.

[142] Whether or not a type of coding block partitioning is allowed can be determined by considering the size of a basic data unit. For example, when a coding block generated as a result of coding block partitioning is larger than the basic data unit, the partitioning may not be allowed. Alternatively, when a non-square coding block generated as a result of coding block partitioning is larger than the basic data unit, the partitioning may not be allowed. For example, when the width or height of a coding block is greater than a threshold value, or when the number of samples included in a coding block is greater than a threshold value, binary tree or ternary tree partitioning may not be allowed. Consequently, encoding information related to binary tree or ternary tree partitioning may be omitted.

[143] Alternatively, it may be set to necessarily partition a coding block larger than the basic data unit. Alternatively, it may be set to necessarily perform binary tree partitioning or ternary tree partitioning on a coding block larger than the basic data unit. Therefore, for a coding block larger than the basic data unit, although the split_flag flag indicating whether or not to partition a coding block is not encoded, the value of the flag may be derived as 1.

[144] As another example, a coding block larger than the basic data unit may be partitioned into a plurality of subblocks. Here, the subblock may be defined as a prediction unit, which is a basic unit for prediction, or a transform unit, which is a basic unit for transform and/or quantization. At this point, partitioning a coding block into a plurality of prediction units may be defined as VPDU prediction unit partitioning, and partitioning a coding block into a plurality of transform units may be defined as VPDU transform unit partitioning.

[145] At least one of VPDU prediction unit partitioning and VPDU transform unit partitioning may be applied to a coding block. The partitioning type of a coding block according to the application of VPDU prediction unit partitioning may be set to be the same as the partitioning type of a coding block according to the application of VPDU transform unit partitioning.

[146] When only VPDU prediction unit partitioning is applied to a coding block, prediction is performed for each sub-block, but transform and/or quantization may be performed for a coding block. At this point, a prediction mode such as a prediction coding mode, an intra-prediction mode, or an inter-prediction mode may be determined for a coding block.

[147] When only VPDU transform unit partitioning is applied to a coding block, prediction is performed for a sub-block, but transform and/or quantization may be performed for each sub-block.

[148] Figures 7 and 8 are views showing examples of partitioning a coding block into a plurality of sub-blocks.

[149] Figure 7 is a view showing a partitioning pattern when only one square basic data unit is allowed, and Figure 8 is a view showing a partitioning pattern when one square basic data unit and one non-square basic data unit are allowed.

[150] When it is considered that only square basic data units are allowed, in Figures 7(a) and 7(b), CU0 and CU2 are defined as two different VPDUs, and CU1 is defined as four different VPDUs. Consequently, CU0 and CU2 can be partitioned into two sub-blocks, and CU1 can be partitioned into four sub-blocks.

[151] When it is considered that square basic data units and non-square basic data units are allowed, in Figures 8(a) and 8(b), CU0 and CU2 can be defined as one VPDU, while CU1 can be defined using two different VPDUs. Consequently, CU0 and CU2 are not partitioned into sub-blocks, while CU1 can be partitioned into two sub-blocks.

[152] At this point, CU1 can be partitioned into square subblocks or non-square subblocks. For example, CU1 can be partitioned into two square subblocks based on a horizontal line that partitions CU1 up and down. Alternatively, CU1 can be partitioned into two non-square subblocks based on a vertical line that partitions CU1 left and right.

[153] When there are a plurality of partitioning type candidates applicable to a coding block, information indicating any of the plurality of partitioning type candidates may be signaled via a bit stream. For example, the information may indicate whether a coding block is partitioned into square subblocks or whether a coding block is partitioned into non-square subblocks.

[154] Alternatively, partitioning a coding block into square subblocks may be set to have a higher priority than partitioning a coding block into non-square subblocks. For example, partitioning a coding block into non-square subblocks may be allowed when partitioning a coding block into square subblocks is not allowed.

[155] Alternatively, the partitioning type of a cipher block may be determined based on the partitioning type of a parent node cipher block. For example, it may be configured to partition a cipher block into square subblocks when the parent node cipher block is partitioned based on a ternary tree. Conversely, it may be configured to partition a cipher block into non-square subblocks when the parent node cipher block is partitioned based on a binary tree or a ternary tree.

[156] Interprediction is a prediction coding mode that predicts a current block using information from a previous image. For example, a block at the same position as the current block in the previous image (then a placed block) can be defined as the prediction block of the current block. Then, a prediction block generated based on a block at the same position as the current block will be termed as a placed prediction block.

[157] On the other hand, when an object existing in the previous image has moved to another position in the current image, the current block can be effectively predicted using a motion of the object. For example, when the motion direction and size of an object can be known by comparing the previous image and the current image, a prediction block (or a prediction image) of the current block can be generated considering the motion information of the object. Then, the prediction block generated using the motion information can be termed as a motion prediction block.

[158] A residual block can be generated by subtracting the prediction block from the current block. At this point, when there is a motion of an object, the energy of the residual block can be reduced by using the motion prediction block instead of the placed prediction block, and therefore the compression performance of the residual block can be improved.

[159] As described above, generating a prediction block using motion information can be referred to as motion compensation prediction. In most interprediction, a prediction block can be generated based on motion compensation prediction.

[160] The motion information may include at least one of a motion vector, a reference picture index, a prediction direction, and a bidirectional weight index. The motion vector represents the motion direction and size of an object. The reference picture index specifies a reference picture of the current block among reference pictures included in a reference picture list. The prediction direction indicates any one of unidirectional L0 prediction, unidirectional L1 prediction, and bidirectional prediction (L0 prediction and L1 prediction). According to the prediction direction of the current block, at least one of motion information in the L0 direction and motion information in the L1 direction may be used. The bidirectional weight index specifies a weighting value applied to an L0 prediction block and a weighting value applied to an L1 prediction block.

[161] Figure 9 is a flowchart illustrating an interprediction method in accordance with an embodiment of the present invention.

[162] Referring to Figure 9, the interprediction method includes the steps of determining an interprediction mode of a current block (S901), acquiring motion information of the current block according to the determined interprediction mode (S902), and performing motion compensation prediction for the current block based on the acquired motion information (S903).

[163] Here, the interprediction mode represents various techniques for determining motion information of the current block, and may include an interprediction mode that uses translational motion information and an interprediction mode that uses affine motion information. For example, the interprediction mode using translational motion information may include a blending mode and an advanced motion vector prediction mode, and the interprediction mode using affine motion information may include an affine blending mode and an affine motion vector prediction mode. The motion information of the current block may be determined based on a neighboring block adjacent to the current block or information analyzed from a bit stream according to the interprediction mode.

[164] The motion information of the current block may be derived from the motion information of another block. Here, another block may be a block encoded/decoded by interprediction before the current block. Setting the motion information of the current block to be equal to the motion information of another block may be defined as a blending mode. Furthermore, setting the motion vector of another block to be the prediction value of the motion vector of the current block may be defined as an advanced motion vector prediction mode.

[165] Figure 10 is a flowchart illustrating a process of deriving motion information from a current block using a blending mode.

[166] A merge candidate of the current block may be derived (S1001). The merge candidate of the current block may be derived from a block encoded/decoded by interprediction before the current block.

[167] Figure 11 is a view showing an example of candidate blocks used to derive a merge candidate.

[168] Candidate blocks may include at least one of neighboring blocks including a sample adjacent to the current block or non-neighboring blocks including a sample not adjacent to the current block. Then, samples for determining candidate blocks are defined as reference samples. Furthermore, a reference sample adjacent to the current block is referred to as a neighboring reference sample, and a reference sample not adjacent to the current block is referred to as a non-neighboring reference sample.

[169] The neighboring reference sample may be included in a neighboring column of the leftmost column of the current block or a neighboring row of the topmost row of the current block. For example, when the coordinates of the top-left sample of the current block are (0, 0), at least one of a block including a reference sample at the position of (-1, H-1), a block including a reference sample at the position of (W-1, -1), a block including a reference sample at the position of (W, -1), a block including a reference sample at the position of (-1, H), and a block including a reference sample at the position of (-1, -1) may be used as a candidate block. With reference to the drawing, neighboring blocks of index 0 to 4 may be used as candidate blocks.

[170] The non-neighboring reference sample represents a sample of which at least one of an x-axis distance and a y-axis distance of a reference sample adjacent to the current block has a predefined value. For example, at least one of a block including a reference sample of which the x-axis distance of the left reference sample is a predefined value, a block including a non-neighboring sample of which the y-axis distance of the top reference sample is a predefined value, and a block including a non-neighboring sample of which the x-axis distance and the y-axis distance of the top left reference sample are predefined values can be used as a candidate block. The predefined values can be a natural number such as 4, 8, 12, 16, or the like. Referring to the drawing, at least one of the blocks of index 5 to 26 can be used as a candidate block.

[171] Alternatively, a candidate block that does not belong to the same coding tree unit as the current block may be set to be unavailable as a merge candidate. For example, when a reference sample deviates from the upper bound of a coding tree unit to which the current block belongs, a candidate block including the reference sample may be set to be unavailable as a merge candidate.

[172] A merge candidate may also be derived from a temporally neighboring block included in a different picture from the current block. For example, a merge candidate may be derived from a placed block included in a placed picture. Any of the reference pictures included in the reference picture list may be defined as a placed picture. Index information to identify the placed picture among the reference pictures may be signaled via a bitstream. Alternatively, a reference picture that has a predefined index among the reference pictures may be determined as the placed picture.

[173] The motion information of the merge candidate may be set to be the same as the motion information of the candidate block. For example, at least one of a motion vector, a reference image index, a prediction direction, and a bidirectional weight index of the candidate block may be set to the motion information of the merge candidate.

[174] A merge candidate list including merge candidates may be generated (S1002).

[175] Merge candidate indices in the merge candidate list may be assigned according to a predetermined order. For example, indices may be assigned in the order of a merge candidate derived from a left neighboring block, a merge candidate derived from an upper neighboring block, a merge candidate derived from an upper right neighboring block, a merge candidate derived from a lower left neighboring block, a merge candidate derived from an upper left neighboring block, and a merge candidate derived from a temporarily neighboring block.

[176] When a plurality of merge candidates are included in the merge candidate list, at least one of the plurality of merge candidates may be selected (S1003). Specifically, information for specifying any of the plurality of merge candidates may be signaled via a bit stream. For example, merge_idx information indicating an index of any of the merge candidates included in the merge candidate list may be signaled via a bit stream.

[177] When the number of merge candidates included in the merge candidate list is less than a threshold value, the merge candidate included in the inter-region movement information list may be added to the merge candidate list. Here, the threshold value may be the maximum number of merge candidates that the merge candidate list can include, or a value obtained by subtracting an offset from the maximum number of merge candidates. The offset may be a natural number such as 1, 2, or similar. The inter-region movement information list may include a merge candidate derived based on a block encoded/decoded before the current block.

[178] The inter-region motion information list includes a merge candidate derived from an inter-prediction-based encoded/decoded block in the current picture. For example, the motion information of a merge candidate included in the inter-region motion information list may be defined as the motion information of an inter-prediction-based encoded/decoded block. Here, the motion information may include at least one of a motion vector, a reference picture index, a prediction direction, and a bidirectional weight index.

[179] A coding block may be partitioned into a plurality of prediction units, and prediction may be performed on each of the partitioned prediction units. Here, a prediction unit represents a basic unit for performing prediction.

[180] A coding block may be partitioned using at least one of a vertical line, a horizontal line, an oblique line, and a diagonal line. Information for determining at least one of the number, angles, and positions of lines that partition a coding block may be signaled via a bit stream. For example, information indicating any of the candidates for the partition type of a coding block may be signaled via a bit stream, or information specifying any of a plurality of line candidates for partitioning a coding block may be signaled via a bit stream. Alternatively, information for determining the number or types of candidate lines that partition a coding block may be signaled via the bit stream. For example, whether an oblique line with an angle greater than that of a diagonal line and/or an oblique line with an angle less than that of a diagonal line may be used as a line candidate may be determined using a 1-bit flag.

[181] Alternatively, at least one of the number, angles, and positions of lines partitioning a coding block may be adaptively determined based on at least one of the intraprediction mode of the coding block, the interprediction mode of the coding block, the position of a merge candidate of the available coding block, and a partitioning pattern of a neighboring block.

[182] When a coding block is partitioned into a plurality of prediction units, intraprediction or interprediction may be performed on each of the partitioned prediction units.

[183] Figure 12 is a view showing examples of partitioning a coding block into a plurality of prediction units using a diagonal line.

[184] As shown in the examples in Figures 12(a) and 12(b), a coding block can be partitioned into two triangular prediction units using a diagonal line.

[185] In Figures 12(a) and 12(b), it is shown that a coding block is partitioned into two prediction units using a diagonal line connecting two vertices of the coding block. However, the coding block can be partitioned into two prediction units using an oblique line, at least one end of which does not pass through a vertex of the coding block.

[186] Figure 13 is a view showing examples of partitioning a coding block into two prediction units.

[187] As shown in the examples of Figures 13(a) and 13(b), a coding block can be partitioned into two prediction units using an oblique line, both ends of which are in contact with the upper boundary and the lower boundary of the coding block, respectively.

[188] Alternatively, as shown in the examples of Figures 13(c) and 13(d), a coding block can be partitioned into two prediction units using an oblique line, both ends of which are in contact with the left boundary and the right boundary of the coding block, respectively.

[189] Alternatively, a coding block can be partitioned into two prediction blocks of different sizes. For example, a coding block can be partitioned into two prediction units of different size by defining an oblique line partitioning the coding block to contact two boundary surfaces that form a vertex.

[190] Figure 14 is a view showing examples of partitioning a coding block into a plurality of prediction blocks of different sizes.

[191] As shown in the examples of Figures 14(a) and 14(b), since a diagonal line connecting the upper-left and lower-right corners of the coding block is set to pass through the left boundary, the right boundary, the upper boundary, or the lower boundary, instead of passing through the upper-left corner or the lower-right corner of the coding block, the coding block can be divided into two prediction units with different sizes.

[192] Alternatively, as shown in the examples of Figures 14(c) and 14(d), since a diagonal line connecting the upper-right and lower-left corners of the coding block is defined to pass through the left boundary, the right boundary, the upper boundary, or the lower boundary, instead of passing through the upper-left corner or the lower-right corner of the coding block, the coding block can be divided into two prediction units with different sizes.

[193] Each of the prediction units generated by partitioning a coding block will be referred to as an “nth prediction unit”. For example, in the examples shown in Figures 12 to 14, PU1 may be defined as a first prediction unit, and PU2 may be defined as a second prediction unit. The first prediction unit means a prediction unit including a sample positioned in the bottom left or a sample positioned in the top left in the coding block, and the second prediction unit means a prediction unit including a sample positioned in the top right or a sample positioned in the bottom right in the coding block.

[194] Conversely, a prediction unit including a sample positioned in the upper right or a sample positioned in the lower right in the coding block may be defined as a first prediction unit, and a prediction unit including a sample positioned in the lower left or a sample positioned in the upper left in the coding block may be defined as a second prediction unit.

[195] The embodiments below are described with a focus on examples of partitioning a coding block using a diagonal line. In particular, partitioning a coding block into two prediction units using a diagonal line is referred to as diagonal partitioning or triangular partitioning, and a prediction unit generated based on diagonal partitioning is referred to as a triangular prediction unit. However, it is also possible to apply the embodiments described below to examples of partitioning a coding block using an oblique line of an angle other than that of a vertical line, a horizontal line, or a diagonal line.

[196] Whether or not to apply diagonal partitioning to a coding block may be determined based on at least one of a slice type, the maximum number of merge candidates that the merge candidate list may include, the size of the coding block, the format of the coding block, the prediction coding mode of the coding block, and the partitioning pattern of the parent node.

[197] For example, whether or not to apply diagonal partitioning to a coding block can be determined based on whether the current slice is type B. Diagonal partitioning may be allowed only when the current slice is type B.

[198] Alternatively, whether or not to apply diagonal partitioning to a cipher block may be determined based on whether the maximum number of merge candidates included in the merge candidate list is two or more. Diagonal partitioning may be allowed only when the maximum number of merge candidates included in the merge candidate list is two or more.

[199] Alternatively, when at least one of the width and height is greater than 64 in the hardware implementation, there is an issue where a data processing unit of size 64 x 64 is accessed redundantly. Consequently, when at least one of the width and height of the coding block is greater than a threshold value, partitioning of a coding block into a plurality of prediction blocks may not be allowed. For example, when at least one of the width and height of a coding block is greater than 64 (e.g., when at least one of the width and height is 128), diagonal partitioning may not be used.

[200] Alternatively, diagonal partitioning may not be allowed for a coding block of which the number of samples is greater than a threshold value, considering the maximum number of samples that can be simultaneously processed in the hardware implementation. For example, diagonal partitioning may not be allowed for a coding tree block of which the number of samples is greater than 4,096.

[201] Alternatively, diagonal partitioning may not be allowed for a coding block where the number of samples included in the coding block is less than a threshold value. For example, diagonal partitioning may be configured to not apply to a coding block when the number of samples included in the coding block is less than 64.

[202] Alternatively, whether or not to apply diagonal partitioning to a coding block can be determined based on whether the width-to-height ratio of the coding block is less than a first threshold value or whether the width-to-height ratio of the coding block is greater than a second threshold value. Here, the width-to-height ratio whRatio of the coding block can be determined as a ratio of the width CbW to the height CbH of the coding block as shown in Equation 1. [EQUATION 1]

[203] The second boundary value may be an inverse of the first boundary value. For example, when the first boundary value is k, the second boundary value may be 1/k.

[204] Diagonal partitioning can be applied to a coding block only when the width-to-height ratio of the coding block is between the first boundary value and the second boundary value.

[205] Alternatively, triangular partitioning may be used only when the width-to-height ratio of the coding block is less than the first threshold value or greater than the second threshold value. For example, when the first threshold value is 16, diagonal partitioning may not be allowed for a coding block of a size of 64 x 4 or 4 x 64.

[206] Alternatively, whether or not to allow diagonal partitioning can be determined based on the partitioning pattern of the parent node. For example, when a parent node's coding block is partitioned based on quad-tree partitioning, diagonal partitioning can be applied to a leaf node's coding block. Conversely, it can be configured to disallow diagonal partitioning for the leaf node's coding block when the parent node's coding block is partitioned based on binary tree or ternary tree partitioning.

[207] Alternatively, whether or not to allow diagonal partitioning may be determined based on the prediction coding mode of a coding block. For example, diagonal partitioning may be allowed only when the coding block is encoded by intra-prediction, when the coding block is encoded by inter-prediction, or when the coding block is encoded by a predefined inter-prediction mode. Here, the predefined inter-prediction mode may represent at least one of a blending mode, an advanced motion vector prediction mode, an affine blending mode, and an affine motion vector prediction mode.

[208] Alternatively, whether or not to allow diagonal partitioning may be determined based on the size of a parallel processing region. For example, when the size of a coding block is larger than the size of a parallel processing region, diagonal partitioning may not be used.

[209] Whether or not to apply diagonal partitioning to a coding block can be determined by considering two or more of the conditions listed above.

[210] As another example, information indicating whether or not to apply diagonal partitioning to a coding block may be signaled via a bit stream. The information may be signaled at a sequence, image, slice, or block level. For example, the flag triangle_partition_flag indicating whether triangular partitioning is applied to a coding block may be signaled at a coding block level.

[211] When it is determined to apply diagonal partitioning to a coding block, information indicating the number of lines that partition the coding block or the positions of the lines may be signaled via a bit stream.

[212] For example, when a coding block is partitioned by a diagonal line, information indicating the direction of the diagonal line partitioning the coding block can be signaled via a bit stream. For example, the triangle_partition_type_flag flag indicating the direction of the diagonal line can be signaled via a bit stream. The flag indicates whether the coding block is partitioned by a diagonal line connecting the upper left corner and the lower right corner, or whether the coding block is partitioned by a diagonal line connecting the upper right corner and the lower left corner. Partitioning a coding block by a diagonal line connecting the upper left corner and the lower right corner can be termed a left triangular partition type, and partitioning a coding block by a diagonal line connecting the upper right corner and the lower left corner can be termed a right triangular partition type. For example, when the value of the flag is 0, it may indicate that the partition type of the coding block is the left triangular partition type, and when the value of the flag is 1, it may indicate that the partition type of the coding block is the right triangular partition type.

[213] Additionally, information indicating whether the prediction units are the same size or information indicating the position of a diagonal line for partitioning the coding block may be signaled via a bit stream. For example, when the information indicating the sizes of the prediction units indicates that the sizes of the prediction units are the same, encoding of the information indicating the position of the diagonal is omitted, and the coding block may be partitioned into two prediction units using a diagonal line passing through two vertices of the coding block. Conversely, when the information indicating the sizes of the prediction units indicates that the sizes of the prediction units are not the same, the position of the diagonal line that partitions the coding block may be determined based on the information indicating the position of the diagonal line. For example, when the left triangular partition type is applied to a coding block, the position information may indicate whether the diagonal line touches the left boundary and the lower boundary or the upper boundary and the right boundary of the coding block. Alternatively, when the right triangular partition type is applied to a coding block, the position information can indicate whether the diagonal line is in contact with the right boundary and the lower boundary or the upper boundary and the left boundary of the coding block.

[214] Information indicating the partition type of a coding block may be signaled at a coding block level. Consequently, the partition type may be determined for each coding block to which diagonal partitioning is applied.

[215] As another example, information indicating the partition type may be flagged for a sequence, an image, a slice, a block, or a coding tree unit. In this case, partition types of coding blocks to which diagonal partitioning is applied in a sequence, an image, a slice, a block, or a coding tree unit may be defined to be the same.

[216] Alternatively, information for determining the partition type may be encoded and signaled for the first coding unit to which diagonal partitioning is applied in the coding tree unit, and the second and subsequent coding units to which diagonal partitioning is applied may be set to use a partition type equal to that of the first coding unit.

[217] As another example, the partition type of a coding block may be determined based on the partition type of a neighboring block. Here, the neighboring block may include at least one of a neighboring block adjacent to the top-left corner of the coding block, a neighboring block adjacent to the top-right corner, a neighboring block adjacent to the bottom-left corner, a neighboring block positioned on the top, and a neighboring block positioned on the left side. For example, the partition type of the current block may be set to be the same as the partition type of a neighboring block. Alternatively, the partition type of the current block may be determined based on whether the left triangular partition type is applied to the top-left neighboring block or whether the right triangular partition type is applied to the top-right neighboring block or the bottom-left neighboring block.

[218] The size of a prediction unit may be set to be no larger than that of a transform unit. Furthermore, it may be configured to necessarily include a prediction unit in a transform unit. Consequently, when VPDU transform unit partitioning is applied to a coding block, diagonal partitioning may be applied to the transform unit instead of applying diagonal partitioning to the coding block.

[219] Figure 15 is a view showing an example of applying diagonal partitioning to a unit transform.

[220] As shown in the example of Figure 15, when a coding block is partitioned into a plurality of transform units, whether or not to apply diagonal partitioning to each transform unit can be determined. Alternatively, it can be determined whether diagonal partitioning is applied by the block unit that has a size smaller than the transform unit.

[221] Alternatively, when VPDU transform unit partitioning is applied to a coding block, whether or not to apply diagonal partitioning to the coding block is determined, and all transform units included in the coding block may be set to follow the determination. For example, when it is determined to apply diagonal partitioning to a coding block, all transform units included in the coding block may be partitioned into two prediction units.

[222] In order to perform motion prediction compensation in a first triangular prediction unit and a second triangular prediction unit, motion information of each of the first triangular prediction unit and the second triangular prediction unit can be derived. At this point, the motion information of the first triangular prediction unit and the second triangular prediction unit can be derived from merge candidates included in the merge candidate list. To distinguish a general merge candidate list from a merge candidate list used to derive the motion information of triangular prediction units, the merge candidate list for deriving the motion information of triangular prediction units is referred to as a triangular merge candidate list, and a merge candidate included in the triangular merge candidate list will be referred to as a triangular merge candidate. However, using the method of deriving a merge candidate and the method of constructing a merge candidate list described above because of triangular merge candidates and the method of constructing the triangular merge candidate list is included in the spirit of the present invention.

[223] Information for determining the maximum number of triangular merge candidates that the triangular merge candidate list can include may be signaled via a bit stream. The information may indicate a difference between the maximum number of merge candidates that the merge candidate list can include and the maximum number of triangular merge candidates that the triangular merge candidate list can include.

[224] Triangular merge candidates can be derived from a spatially neighboring block and a temporally neighboring block of a coding block.

[225] Figure 16 is a view showing neighboring blocks used to derive a triangular merge candidate.

[226] A triangular merge candidate may be derived using at least one of a neighboring block positioned at the top of a coding block, a neighboring block positioned at the left side of the coding block, and a block placed included in a different picture from the coding block. The top neighboring block may include at least one of a block including a sample (xCb + CbW-1, yCb-1) positioned at the top of the coding block, a block including a sample (xCb + CbW, yCb-1) positioned at the top of the coding block, and a block including a sample (xCb-1, yCb-1) positioned at the top of the coding block. The left neighboring block may include at least one of a block including a sample (xCb-1, yCb + CbH-1) positioned at the left side of the coding block and a block including a sample (xCb-1, yCb + CbH) positioned at the left side of the coding block. The placed block can be determined as any one of a block including a sample (xCb + CbW, yCb + CbH) adjacent to the upper right corner of the coding block and a block including a sample (xCb/2, yCb/2) positioned at the center of the coding block, in a placed image.

[227] Neighboring blocks can be searched in a predefined order, and triangular merge candidates can be constructed as a triangular merge candidate list according to a predefined order. For example, the triangular merge candidate list can be constructed by searching the triangular merge candidates in the order of B1, A1, B0, A0, C0, B2, and C1.

[228] The motion information of the triangular prediction units can be derived based on the triangular merge candidate list. That is, the triangular prediction units can share a triangular merge candidate list.

[229] In order to derive motion information of the triangular merge unit, information for specifying at least one of the triangular merge candidates included in the triangular merge candidate list may be signaled via a bit stream. For example, the index information merge_triangle_idx for specifying at least one of the triangular merge candidates may be signaled via a bit stream.

[230] The index information may specify a combination of a merge candidate from the first triangular prediction unit and a merge candidate from the second triangular prediction unit. For example, Table 1 shows an example of a combination of merge candidates according to index information merge_triangle_idx. [TABLE 1]

[231] When the merge_triangle_idx index information value is 1, it indicates that the motion information of the first triangular prediction unit is derived from a merge candidate that has an index of 1, and the motion information of the second triangular prediction unit is derived from a merge candidate that has an index of 0. A triangular merge candidate for deriving motion information of the first triangular prediction unit and a triangular merge candidate for deriving motion information of the second triangular prediction unit can be determined through merge_triangle_idx index information.

[232] A partition type of a coding block to which diagonal partitioning is applied may be determined based on index information. That is, the index information may specify a combination of a merge candidate from the first triangular prediction unit, a merge candidate from the second triangular prediction unit, and a partitioning direction of the coding block. When a partition type of the coding block is determined based on index information, triangle_partition_type_flag information indicating the direction of a diagonal line partitioning the coding block may not be encoded. Table 2 expresses partition types of a coding block with respect to the merge_triangle_idx index information. [TABLE 2]

[233] When the TriangleDir variable is 0, it indicates that the left triangle partition type is applied to the coding block, and when the TriangleDir variable is 1, it indicates that the right triangle partition type is applied to the coding block. When combining Table 1 and Table 2, it can be set to specify a combination of the merge candidate of the first triangular prediction unit, the merge candidate of the second triangular prediction unit, and the partition direction of the coding block by index information merge_triangle_idx.

[234] As another example, index information only for any one of the first triangular prediction unit and the second triangular prediction unit may be flagged, and an index from one triangular merge candidate to the other one of the first triangular prediction unit and the second triangular prediction unit may be determined based on the index information. For example, a triangular merge candidate of the first triangular prediction unit may be determined based on the index information merge_triangle_idx indicating an index of any one of the triangular merge candidates. Furthermore, a triangular merge candidate of the second triangular prediction unit may be specified based on merge_triangle_idx. For example, the triangular merge candidate of the second triangular prediction unit may be derived by adding or subtracting an offset to or from the index information merge_triangle_idx. The offset can be an integer such as 1 or 2. For example, a triangular merge candidate that has a value obtained by adding 1 to merge_triangle_idx as an index can be determined as the triangular merge candidate of the second triangular prediction unit. When merge_triangle_idx indicates a triangular merge candidate that has the largest index value among the triangular merge candidates, the movement information of the second triangular prediction unit can be derived from a triangular merge candidate that has an index of 0 or a triangular merge candidate that has a value obtained by subtracting 1 from merge_triangle_idx as an index.

[235] Alternatively, the motion information of the second triangular prediction unit may be derived from a triangular merge candidate that has a reference image equal to that of the triangular merge candidate of the first triangular prediction unit specified by the index information. Here, the triangular merge candidate that has a reference image equal to that of the triangular merge candidate of the first triangular prediction unit may indicate a triangular merge candidate that has at least one of L0 reference image and L1 reference image equal to that of the triangular merge candidate of the first triangular prediction unit. When there are a plurality of triangular merge candidates that have a reference image equal to that of the triangular merge candidate of the first triangular prediction unit, at least one of the triangular merge candidates may be selected based on whether the merge candidate includes bidirectional motion information or a difference value between the index of the merge candidate and the index information.

[236] As another example, index information may be signaled for each of the first triangular prediction unit and the second triangular prediction unit. For example, the first index information 1st_merge_idx for determining a triangular merge candidate of the first triangular prediction unit and the second index information 2nd_merge_idx for determining a triangular merge candidate of the second triangular prediction unit may be signaled via a bit stream. The motion information of the first triangular prediction unit may be derived from the triangular merge candidate determined based on the first index information 1st_merge_idx, and the motion information of the second triangular prediction unit may be derived from the triangular merge candidate determined based on the second index information 2nd_merge_idx.

[237] First index information 1st_merge_idx may indicate an index of any of the triangular merge candidates included in the triangular merge candidate list. The triangular merge candidate of the first triangular prediction unit may be determined as a triangular merge candidate indicated by the first index information 1st_merge_idx.

[238] The triangular merge candidate indicated by the first index information 1st_merge_idx may be set to not be used as a triangular merge candidate of the second triangular prediction unit. Accordingly, the second index information 2nd_merge_idx of the second triangular prediction unit may indicate an index of any one of the remaining triangular merge candidates excluding the triangular merge candidate indicated by the first index information. When the value of the second index information 2nd_merge_idx is less than the value of the first index information 1st_merge_idx, the triangular merge candidate of the second triangular prediction unit may be determined as a triangular merge candidate having the index information indicated by the second index information 2nd_merge_idx. On the other hand, when the value of the second index information 2nd_merge_idx is equal to or greater than the value of the first index information 1st_merge_idx, the triangular merge candidate of the second triangular prediction unit can be determined as a triangular merge candidate that has a value obtained by adding 1 to the value of the second index information 2nd_merge_idx as an index.

[239] Alternatively, whether or not the second index information is signed may be determined according to the number of triangular merge candidates included in the triangular merge candidate list. For example, when the maximum number of triangular merge candidates that the triangular merge candidate list can include does not exceed 2, the sign of the second index information may be omitted. When the sign of the second index information is omitted, a second triangular merge candidate may be derived by adding or subtracting an offset to or from the first index information. For example, when the maximum number of triangular merge candidates that the triangular merge candidate list can include is 2 and the first index information indicates index 0, the second triangular merge candidate may be derived by adding 1 to the first index information. Alternatively, when the maximum number of triangular merge candidates that the triangular merge candidate list can include is 2 and the first index information indicates 1, the second triangular merge candidate can be derived by subtracting 1 from the first index information.

[240] Alternatively, when the flag of the second index information is omitted, the second index information may be set to a default value. Here, the default value may be 0. The second triangular merge candidate can be derived by comparing the first index information and the second index information. For example, when the second index information is less than the first index information, a merge candidate having index 0 may be defined as the second triangular merge candidate, and when the second index information is equal to or greater than the first index information, a merge candidate having index 1 may be defined as the second triangular merge candidate.

[241] When the triangular merge candidate has unidirectional motion information, the unidirectional motion information of the triangular merge candidate can be defined as the motion information of the triangular prediction unit. Conversely, when the triangular merge candidate has bidirectional motion information, only one of L0 motion information and L1 motion information can be defined as the motion information of the triangular prediction unit. Which of the L0 motion information and L1 motion information will be obtained can be determined based on the index of the triangular merge candidate or motion information of another triangular prediction unit.

[242] For example, when the index of the triangular merge candidate is an even number, the L0 motion information of the triangular prediction unit can be set to 0, and the L1 motion information of the triangular merge candidate can be set to the L1 motion information of the triangular prediction unit. Conversely, when the index of the triangular merge candidate is an odd number, the L1 motion information of the triangular prediction unit can be set to 0, and the L0 motion information of the triangular merge candidate can be set to 0. Conversely, when the index of the triangular merge candidate is an even number, the L0 motion information of the triangular merge candidate can be set to the L0 motion information of the triangular prediction unit, and when the index of the triangular merge candidate is an odd number, the L1 motion information of the triangular merge candidate can be set to the L1 motion information of the triangular prediction unit. Alternatively, when the triangular merge candidate is an even number for the first triangular prediction unit, L0 motion information of the triangular merge candidate can be defined as L0 motion information of the first triangular prediction unit, while when the triangular merge candidate is an odd number for the second triangular prediction unit, L1 motion information of the triangular merge candidate can be defined as L1 motion information of the second triangular prediction unit.

[243] Alternatively, when the first triangular prediction unit has L0 motion information, L0 motion information of the second triangular prediction unit may be set to 0, and L1 motion information of the triangular merge candidate may be set to L1 motion information of the second triangular prediction unit. Conversely, when the first triangular prediction unit has L1 motion information, L1 motion information of the second triangular prediction unit may be set to 0, and L0 motion information of the triangular merge candidate may be set to L0 motion information of the second triangular prediction unit.

[244] A triangular merge candidate list for deriving motion information from the first triangular prediction unit and a triangular merge candidate list for deriving motion information from the second triangular prediction unit may be defined differently.

[245] For example, when a triangular merge candidate for deriving motion information of the first triangular prediction unit in the triangular merge candidate list is specified based on the index information for the first triangular prediction unit, motion information of the second triangular prediction unit may be derived using the triangular merge candidate list including the remaining triangular merge candidates excluding the triangular merge candidate indicated by the index information. Specifically, motion information of the second triangular prediction unit may be derived from any of the remaining triangular merge candidates.

[246] Consequently, the maximum number of triangular merge candidates that the triangular merge candidate list of the first triangular prediction unit includes and the maximum number of triangular merge candidates that the triangular merge candidate list of the second triangular prediction unit includes may be different. For example, when the triangular merge candidate list of the first triangular prediction unit includes M merge candidates, the triangular merge candidate list of the second triangular prediction unit may include M-1 merge candidates excluding the triangular merge candidate indicated by the index information of the first triangular prediction unit.

[247] As another example, a merge candidate of each triangular prediction unit is derived based on the neighboring blocks adjacent to a coding block, and availability of the neighboring blocks can be determined by considering the shape or position of the triangular prediction unit.

[248] Figure 17 is a view to describe examples of determining availability of a neighboring block for each triangular forecast unit.

[249] A neighboring block not adjacent to the first triangular prediction unit may be defined as unavailable to the first triangular prediction unit, and a neighboring block not adjacent to the second triangular prediction unit may be defined as unavailable to the second triangular prediction unit.

[250] For example, as shown in the example of Figure 17(a), when the left triangular partition type is applied to a coding block, it may be determined that blocks A1, A0, and A2 adjacent to the first triangular prediction unit among neighboring blocks adjacent to the coding block are available to the first triangular prediction unit, while blocks B0 and B1 are unavailable to the first triangular prediction unit. Consequently, the triangular merge candidate list for the first triangular prediction unit may include triangular merge candidates derived from blocks A1, A0, and A2, and may not include triangular merge candidates derived from blocks B0 and B1.

[251] As shown in the example of Figure 17(b), when the left triangular partition type is applied to a coding block, it can be determined that blocks B0 and B1 adjacent to the second triangular prediction unit are available to the second triangular prediction unit, while blocks A1, A0, and A2 are unavailable to the second triangular prediction unit. Consequently, the triangular merge candidate list for the second triangular prediction unit may include triangular merge candidates derived from blocks B0 and B1, and may not include triangular merge candidates derived from blocks A1, A0, and A2.

[252] Accordingly, the number of triangular merge candidates or the range of triangular merge candidates that the triangular prediction unit can use can be determined based on at least one of the position of the triangular prediction unit or the partition type of the coding block.

[253] As another example, the blending mode may be applied to only one of the first triangular prediction unit and the second triangular prediction unit. Furthermore, the motion information of the other of the first triangular prediction unit and the second triangular prediction unit may be set equal to the motion information of the triangular prediction unit so that the blending mode is applied, or may be derived by refining the motion information of the triangular prediction unit so that the blending mode is applied.

[254] For example, a motion vector and a reference image index of the first triangular prediction unit may be derived based on a triangular merge candidate, and a motion vector of the second triangular prediction unit may be derived by refining the motion vector of the first triangular prediction unit. For example, the motion vector of the second triangular prediction unit may be derived by adding or subtracting a refined motion vector {Rx, Ry} to or from the motion vector {mvD1LXx, mvD1LXy} of the first triangular prediction unit. The reference image index of the second triangular prediction unit may be set to be equal to the reference image index of the first triangular prediction unit.

[255] Information for determining a refined motion vector indicating the difference between the motion vector of the first triangular prediction unit and the motion vector of the second triangular prediction unit may be signaled via a bit stream. The information may include at least one of information indicating the size of the refined motion vector and information indicating the sign of the refined motion vector.

[256] Alternatively, the refined motion vector signal may be derived based on at least one of the position of the triangular prediction unit, the index of the triangular prediction unit, and the partition type applied to the coding block.

[257] As another example, the motion vector and reference image index of either of the first triangular prediction unit and the second triangular prediction unit may be signed. The motion vector of the other of the first triangular prediction unit and the second triangular prediction unit may be derived by refining the signed motion vector.

[258] For example, the motion vector and reference picture index of the first triangular prediction unit may be determined based on the signed information of a bit stream. Further, the motion vector of the second triangular prediction unit may be derived by refining the motion vector of the first triangular prediction unit. For example, the motion vector of the second triangular prediction unit may be derived by adding or subtracting a refined motion vector {Rx, Ry} to or from the motion vector {mvD1LXx, mvD1LXy} of the first triangular prediction unit. The reference picture index of the second triangular prediction unit may be set to be equal to the reference picture index of the first triangular prediction unit.

[259] Motion prediction compensation prediction for each coding block may be performed based on the motion information of the first triangular prediction unit and the motion information of the second triangular prediction unit. At this point, video quality degradation may occur at the boundary between the first triangular prediction unit and the second triangular prediction unit. For example, video quality continuity may be degraded in the vicinity of an edge existing at the boundary between the first triangular prediction unit and the second triangular prediction unit. In order to reduce video quality degradation at the boundary, a prediction sample may be derived via a smoothing filter or a weighted prediction.

[260] Prediction samples in a coding block to which diagonal partitioning is applied may be derived based on a weighted sum operation of a first prediction sample obtained based on the motion information of the first triangular prediction unit and a second prediction sample obtained based on the motion information of the second triangular prediction unit. Alternatively, a prediction sample of the first triangular prediction unit is derived from a first prediction block determined based on the motion information of the first triangular prediction unit, and a prediction sample of the second triangular prediction unit is derived from a second prediction block determined based on the motion information of the second triangular prediction unit, and a prediction sample positioned in the boundary region of the first triangular prediction unit and the second triangular prediction unit may be derived based on a weighted sum operation of the first prediction sample included in the first prediction block and the second prediction sample included in the second prediction block. For example, Equation 2 shows an example of deriving prediction samples from the first triangular prediction unit and the second triangular prediction unit. [EQUATION 2]

[261] In Equation 2, P1 denotes a first forecast sample, and P2 denotes a second forecast sample. w1 denotes a weighting value applied to the first forecast sample, and (1-w1) denotes a weighting value applied to the second forecast sample. As shown in the example in Equation 2, the weighting value applied to the second forecast sample can be derived by subtracting the weighting value applied to the first forecast sample from a constant value.

[262] When the left triangular partition type is applied to a coding block, the boundary region may include prediction samples of which the x-axis coordinate and the y-axis coordinate are equal. Conversely, when the right triangular partition type is applied to a coding block, the boundary region may include prediction samples of which the sum of the x-axis coordinate and the y-axis coordinate is greater than or equal to a first boundary value and less than a second boundary value.

[263] A size of the boundary region may be determined based on at least one of the coding block size, the coding block shape, motion information of the triangular prediction units, a difference value between the motion vectors of the triangular prediction units, an output order of reference images, and a difference value between the first prediction sample and the second prediction sample at the diagonal boundary.

[264] Figures 18 and 19 are views showing examples of deriving a prediction sample based on a weighted sum operation of a first prediction sample and a second prediction sample. Figure 18 shows an example of applying the left triangular partition type to a coding block, and Figure 19 shows an example of applying the right triangular partition type to a coding block. In addition, Figures 18(a) and 19(a) are views showing prediction patterns for a luma component, and Figures 18(b) and 19(b) are views showing prediction patterns for a chroma component.

[265] In the drawings, numbers marked on the forecast samples positioned near the boundary between the first forecast unit and the second forecast unit indicate weighting values applied to the first forecast sample. For example, when a number marked on a forecast sample is N, the forecast sample can be derived by applying a weighting value of N/8 to the first forecast sample and applying a weighting value of (1-(N/8)) to the second forecast sample.

[266] In a non-threshold region, either the first prediction sample or the second prediction sample may be determined as a prediction sample. Referring to the example in Figure 18, in a region belonging to the first triangular prediction unit among regions in which the absolute value of the difference between the x-axis coordinate and the y-axis coordinate is greater than a threshold value, the first prediction sample derived based on the motion information of the first triangular prediction unit may be determined as a prediction sample. Conversely, in a region belonging to the second triangular prediction unit among regions in which the difference value between the x-axis coordinate and the y-axis coordinate is greater than a threshold value, the second prediction sample derived based on the motion information of the second triangular prediction unit may be determined as a prediction sample.

[267] Referring to the example of Figure 19, in a region where the sum of the x-axis coordinate and the y-axis coordinate is less than a first threshold value, the first prediction sample derived based on the motion information of the first triangular prediction unit may be determined as a prediction sample. Conversely, in a region where the sum of the x-axis coordinate and the y-axis coordinate is greater than a second threshold value, the second prediction sample derived based on the motion information of the second triangular prediction unit may be determined as a prediction sample.

[268] A threshold value for determining a non-threshold region may be determined based on at least one of the size of a coding block, the format of the coding block, and a color component. For example, when the threshold value for a luma component is set to N, the threshold value for a chroma component may be set to N/2.

[269] The prediction samples included in the boundary region may be derived based on a weighted sum operation of the first prediction sample and the second prediction sample. At this point, the weighting values applied to the first prediction sample and the second prediction sample may be determined based on at least one of the position of a prediction sample, the size of a coding block, the format of the coding block, and a color component.

[270] For example, as shown in the example of Figure 18(a), prediction samples at the position of the same x-axis coordinate and y-axis coordinate can be derived by applying the same weighting value to the first prediction sample and the second prediction sample. The prediction samples of which the absolute value of the difference between the x-axis coordinate and the y-axis coordinate is 1 can be derived by setting the ratio of the weighting value applied to the first prediction sample and the second prediction sample to (3:1) or (1:3). Furthermore, the prediction samples of which the absolute value of the difference of the x-axis coordinate and the y-axis coordinate is 2 can be derived by setting the ratio of the weighting value applied to the first prediction sample and the second prediction sample to (7:1) or (1:7).

[271] Alternatively, as shown in the example of Figure 18(b), the prediction samples at the position of the same x-axis coordinate and y-axis coordinate can be derived by applying the same weighting value to the first prediction sample and the second prediction sample, and the prediction samples of which the absolute value of the difference between the x-axis coordinate and the y-axis coordinate is 1 can be derived by setting the ratio of the weighting value applied to the first prediction sample and the second prediction sample to (7:1) or (1:7).

[272] For example, as shown in the example of Figure 19(a), the prediction samples of which the sum of the x-axis coordinate and the y-axis coordinate is less than the width or the height of a coding block by 1 can be derived by applying the same weighting value to the first prediction sample and the second prediction sample. The prediction samples of which the sum of the x-axis coordinate and the y-axis coordinate is equal to or less than the width or the height of the coding block by 2 can be derived by setting the ratio of the weighting value applied to the first prediction sample and the second prediction sample to (3:1) or (1:3). The predicted samples of which the sum of the x-axis coordinate and the y-axis coordinate is greater than or less than the width or the height of the coding block by 1 or 3 can be derived by setting the ratio of the weighting value applied to the first prediction sample and the second prediction sample to (7:1) or (1:7).

[273] Alternatively, as shown in the example of Figure 19(b), prediction samples of which the sum of the x-axis coordinate and the y-axis coordinate is less than the width or the height of the coding block by 1 can be derived by applying the same weighting value to the first prediction sample and the second prediction sample. Prediction samples of which the sum of the x-axis coordinate and the y-axis coordinate is equal to or less than the width or the height of the coding block by 2 can be derived by setting the ratio of the weighting value applied to the first prediction sample and the second prediction sample to (7:1) or (1:7).

[274] As another example, the weight value can be determined by considering the position of a prediction sample or the shape of a coding block. Equations 3 through 5 show an example of deriving a weight value when the left triangular partition type is applied to a coding block. Equation 3 shows an example of deriving a weight value applied to the first prediction sample when the coding block has a square shape. [EQUATION 3]

[275] In Equation 3, x and y denote the position of a prediction sample. When a coding block has a non-square shape, a weighting value applied to the first prediction sample can be derived as shown in Equation 4 or 5. Equation 4 shows a case where the width of a coding block is greater than the height, and Equation 5 shows a case where the width of a coding block is less than the height. [EQUATION 4]

[276] When the right triangular partition type is applied to a coding block, a weighting value applied to the first prediction sample can be determined as shown in Equations 6 through 8. Equation 6 shows an example of deriving a weighting value applied to the first prediction sample when the coding block has a square shape. [EQUATION 6]

[277] In Equation 6, CbW denotes the width of a coding block. When the coding block has a non-square shape, the weighting value applied to the first prediction sample can be derived as shown in Equation 7 or Equation 8. Equation 7 shows a case where the width of a coding block is greater than the height, and Equation 8 shows a case where the width of a coding block is less than the height. [EQUATION 7]

[278] In Equation 7, CbH denotes the height of the coding block.

[279] As shown in the example, among the forecast samples in the boundary region, the forecast samples included in the first triangular forecast unit are derived by assigning a weighting value greater than that of the second forecast sample to the first forecast sample, and the forecast samples included in the second triangular forecast unit are derived by assigning a weighting value greater than that of the first forecast sample to the second forecast sample.

[280] When diagonal partitioning is applied to a coding block, it can be configured to not apply a combined prediction mode by combining the intraprediction mode and the blending mode to the coding block.

[281] Intraprediction is to predict a current block using reconstructed samples that were encoded/decoded in the neighborhood of the current block. At this point, the reconstructed samples before an in-loop filter was applied can be used for intraprediction of the current block.

[282] The intraprediction technique includes matrix-based intraprediction, and general intraprediction considering directionality relative to neighboring reconstructed samples. Information indicating the intraprediction technique of the current block may be signaled via a bit stream. The information may be a 1-bit flag. Alternatively, the intraprediction technique of the current block may be determined based on at least one of the position of the current block, the size of the current block, the shape of the current block, and an intraprediction technique of a neighboring block. For example, when the current block exists across an image boundary, it may be configured to not apply the matrix-based intraprediction to the current block.

[283] Intraprediction based on intraprediction matrix is a method of acquiring a prediction block of the current block by an encoder and a decoder based on a matrix product between a previously stored matrix and reconstructed samples in the neighborhood of the current block. Information for specifying any of a plurality of previously stored matrices may be signaled via a bit stream. The decoder may determine a matrix for intraprediction of the current block based on the information and size of the current block.

[284] General intraprediction is a method of acquiring a prediction block for the current block based on a non-angular intraprediction mode or an angular intraprediction mode. Next, a process of performing intraprediction based on general intraprediction will be described in more detail with reference to the drawings.

[285] Figure 20 is a flowchart illustrating an intraforecast method in accordance with an embodiment of the present invention.

[286] A reference sample line of the current block may be determined (S2001). The reference sample line means a set of reference samples included in a k-th line beyond the top and/or left side of the current block. The reference samples may be derived from reconstructed samples that were encoded/decoded in the neighborhood of the current block.

[287] Index information for identifying the reference sample line of the current block among a plurality of reference sample lines may be signaled via a bit stream. For example, index information intra_luma_ref_idx for specifying the reference sample line of the current block may be signaled via a bit stream. The index information may be signaled by the coding block unit.

[288] The plurality of reference sample rows may include at least one of a first row, a second row, a third row, and a fourth row at the top and/or left side of the current block. A reference sample row configured from a row adjacent to the top of the current block and a column adjacent to the left side of the current block among the plurality of reference sample rows is referred to as an adjacent reference sample row, and the other reference sample rows may be referred to as non-adjacent reference sample rows.

[289] Only some of the plurality of reference sample lines may be selected as the reference sample line of the current block. For example, the remaining reference sample lines excluding a third non-adjacent reference sample line among the plurality of reference sample lines may be defined as candidate reference sample lines. Table 3 shows indexes respectively assigned to the candidate reference sample lines. [TABLE 3]

[290] A greater number of candidate reference sample lines or a smaller number of candidate reference sample lines than the candidate reference sample lines described above may be defined. Furthermore, the number or positions of non-adjacent reference sample lines defined as candidate reference sample lines is not limited to the example described above. For example, a first non-adjacent reference sample line and a third non-adjacent reference sample line may be defined as candidate reference sample lines, or a second non-adjacent reference sample line and a third non-adjacent reference sample line may be defined as candidate reference sample lines. Alternatively, the entire first non-adjacent reference sample line, the second non-adjacent reference sample line, and the third non-adjacent reference sample line may be defined as candidate reference sample lines.

[291] The number or types of candidate reference sample lines may be determined based on at least one of the size of the current block, the format of the current block, the position of the current block, whether or not the current block is partitioned into sub-blocks, and the intraprediction mode of the current block.

[292] The reference sample line of the current block may be determined based on at least one of the position of the current block, the size of the current block, the format of the current block, and a predictive coding mode of a neighboring block. For example, when the current block is in contact with the boundary of a picture, a block, a slice, or a coding tree unit, the adjacent reference sample line may be determined as the reference sample line of the current block.

[293] Alternatively, when the current block has a non-square shape, an adjacent reference sample line may be determined as the reference sample line of the current block. Alternatively, when the width to height ratio of the current block is greater than or equal to a threshold value, the adjacent reference sample line may be determined as the reference sample line of the current block.

[294] The reference sample line may include top reference samples positioned at the top of the current block and left reference samples positioned at the left side of the current block. The top reference samples and left reference samples may be derived from reconstructed samples in the neighborhood of the current block. The reconstructed samples may be in a state before the in-loop filter is applied.

[295] Then, the intraprediction mode of the current block can be determined (S2002). As for the intraprediction mode of the current block, at least one of a non-angular intraprediction mode and an angular intraprediction mode can be determined as the intraprediction mode of the current block. The non-angular intraprediction mode includes a planar mode and a Dc mode, and the angular intraprediction mode includes 33 or 65 modes between the lower left diagonal direction and the upper right diagonal direction.

[296] Figure 21 is a view showing intraprediction modes.

[297] Figure 21(a) shows 35 intraforecast modes, and Figure 21(b) shows 67 intraforecast modes.

[298] A greater or lesser number of intraforecast modes than those shown in Figure 21 can be defined.

[299] A Most Likely Mode (MPM) can be defined based on the intraprediction mode of a neighboring block adjacent to the current block. Here, the neighboring block may include a left neighboring block adjacent to the left side of the current block and an upper neighboring block adjacent to the top of the current block.

[300] The number of MPMs included in an MPM list may be predefined in the encoder and decoder. The number of MPMs included in the MPM list may be predefined in the encoder and decoder. For example, the number of MPMs may be 3, 4, 5, or 6. Alternatively, information indicating the number of MPMs may be signaled via a bit stream. Alternatively, the number of MPMs may be determined based on at least one of the prediction coding mode of a neighboring block, and the size, format, and reference sample line index of the current block. For example, when an adjacent reference sample line is determined as the reference sample line of the current block, N MPMs may be used, while when a non-adjacent reference sample line is determined as the reference sample line of the current block, M MPMs may be used. M is a natural number less than N, and for example, N can be 6, and M can be 5, 4, or 3. Consequently, when the reference sample line index of the current block is 0 and the MPM flag is true, any one of 6 candidate intraprediction modes can be determined as the intraprediction mode of the current block, while when the reference sample line index of the current block is greater than 0 and the MPM flag is true, any one of 5 candidate intraprediction modes can be determined as the intraprediction mode of the current block.

[301] Alternatively, a fixed number (e.g., 6 or 5) of MPM candidates may be used regardless of the index of the current block's reference sample line.

[302] An MPM list including a plurality of MPMs is generated, and information indicating whether an MPM that is the same as the intraprediction mode of the current block is included in the MPM list may be signaled via a bit stream. The information is a 1-bit flag and may be referred to as an MPM flag. When the MPM flag indicates that an MPM that is the same as that of the current block is included in the MPM list, index information identifying one of the MPMs may be signaled via a bit stream. For example, index information mpm_idx specifying any one of the plurality of MPMs may be signaled via a bit stream. The MPM specified by the index information may be set as the intraprediction mode of the current block. When the MPM flag indicates that an MPM equal to that of the current block is not included in the MPM list, residual mode information indicating any of the residual intraprediction modes other than MPMs can be signaled via a bitstream. The residual mode information indicates an index value corresponding to the current block's intraprediction mode when indices are reassigned to the residual intraprediction modes excluding MPMs. The decoder can arrange the MPMs in ascending order and determine the current block's intraprediction mode by comparing the residual mode information with the MPMs. For example, when the residual mode information is equal to or less than the MPM, the current block's intraprediction mode can be derived by adding 1 to the residual mode information.

[303] When the intraprediction mode of the current block is derived, the comparison between some of the MPMs and the residual mode information may be omitted. For example, among the MPMs, non-angular intraprediction mode MPMs may be excluded from comparison targets. When non-angular intraprediction modes are defined as MPMs, it is clear that the residual mode information indicates the angular intraprediction mode, and thus the intraprediction mode of the current block can be derived by comparing the residual MPMs excluding the non-angular intraprediction modes and the residual mode information. Instead of excluding the non-angular intraprediction modes from the comparison targets, the number of the non-angular intraprediction modes is added to the residual mode information, and then a comparison result value can be compared with the residual MPMs.

[304] Instead of setting the default mode for the MPM, information indicating whether or not the current block's intraprediction mode is the default mode may be signaled via a bit stream. The information is a 1-bit flag, and the flag may be referred to as a default mode flag. The default mode flag may be signaled only when the MPM flag indicates that an MPM equal to that of the current block is included in the MPM list. As described above, the default mode may include at least one of a flat mode, a Dc mode, a vertical direction mode, and a horizontal direction mode. For example, when the flat mode is set as the default mode, the default mode flag may indicate whether or not the current block's intraprediction mode is a flat mode. When the default mode flag indicates that the current block's intraprediction mode is not the default mode, one of the MPMs indicated by the index information may be set as the current block's intraprediction mode.

[305] When the default mode flag is used, it can be configured to not define an intraprediction mode equal to the default mode as an MPM. For example, when the default mode flag indicates whether the current block's intraprediction mode is a flat mode, the current block's intraprediction mode can be derived using 5 MPMs excluding the MPM corresponding to the flat mode.

[306] When a plurality of intraprediction modes are defined as default modes, index information indicating any of the default modes may be additionally signaled. The intraprediction mode of the current block may be defined as the default mode indicated by the index information.

[307] When the current block's reference sample line index is not 0, it may be configured to not use the default mode. For example, it may be configured to not use a non-angular intraprediction mode such as a Dc mode or a planar mode when a non-adjacent reference sample line is determined to be the current block's reference sample line. Consequently, when the reference sample line index is not 0, the default mode flag is not flagged, and the default mode flag value may be set to a predefined value (i.e., false).

[308] When the intraprediction mode of the current block is determined, prediction samples may be acquired for the current block based on the determined intraprediction mode (S2003).

[309] When Dc mode is selected, prediction samples are generated for the current block based on an average value of reference samples. Specifically, values of all samples in a prediction block may be generated based on an average value of the reference samples. The average value may be derived using at least one of top reference samples positioned at the top of the current block and left reference samples positioned at the left side of the current block.

[310] Depending on the shape of the current block, the number or range of reference samples used to derive the mean value may vary. For example, when the current block is a non-square block whose width is greater than the height, the mean value can be calculated using only the top reference samples. Conversely, when the current block is a non-square block whose width is less than the height, the mean value can be calculated using only the left reference samples. That is, when the width and height of the current block are different, the mean value can be calculated using only reference samples adjacent to the longest one. Alternatively, based on the width-to-height ratio of the current block, whether or not to calculate the mean value using only the top reference samples or whether or not to calculate the mean value using only the left reference samples can be determined.

[311] When planar mode is selected, a prediction sample can be acquired using a horizontal direction prediction sample and a vertical direction prediction sample. Here, the horizontal direction prediction sample is acquired based on a left reference sample and a right reference sample positioned on a horizontal line equal to that of the prediction sample, and the vertical direction prediction sample is acquired based on a top reference sample and a bottom reference sample positioned on a vertical line equal to that of the prediction sample. Here, the right reference sample can be generated by copying a reference sample adjacent to the top right corner of the current block, and the bottom reference sample can be generated by copying a reference sample adjacent to the bottom left corner of the current block. The horizontal direction prediction sample can be acquired based on a weighted summation operation of the left reference sample and the right reference sample, and the vertical direction prediction sample can be acquired based on a weighted summation operation of the top reference sample and the bottom reference sample. At this point, a weighting value can be assigned to each reference sample based on the position of the forecast sample. The forecast sample can be acquired based on an average operation or a weighted sum operation of the horizontal forecast sample and the vertical forecast sample. When the weighted sum operation is performed, the weighting values applied to the horizontal forecast sample and the vertical forecast sample can be determined based on the position of the forecast sample.

[312] When an angular prediction mode is selected, a parameter indicating the prediction direction (or prediction angle) of the selected angular prediction mode can be determined. Table 4 shows the intra direction parameter intraPredAng of each intraprediction mode. [TABLE 4]

[313] When 35 intraforecast modes are defined, Table 4 shows the intra direction parameter of each intraforecast mode that has an index of any one of 2 to 34. When more than 33 angular intraforecast modes are defined, Table 4 is further subdivided to define the intra direction parameter of each angular intraforecast mode.

[314] After arranging the top reference samples and left reference samples of the current block into a row, a prediction sample can be acquired based on the value of the intra direction parameter. At this point, when the value of the intra direction parameter is a negative value, the left reference samples and top reference samples can be arranged into a row.

[315] Figures 22 and 23 are views showing an example of a one-dimensional array arranging reference samples in a row.

[316] Figure 22 is a view showing an example of a one-dimensional vertically directed array arranging reference samples in the vertical direction, and Figure 23 is a view showing an example of a one-dimensional horizontally directed array arranging reference samples in the horizontal direction. The embodiments of Figures 22 and 23 will be described on the assumption that 35 intraprediction modes are defined.

[317] When the intra-forecast mode index is any one of 11 to 18, a one-dimensional horizontally directed array in which top reference samples are rotated counterclockwise can be applied, and when the intra-forecast mode index is any one of 19 to 25, a one-dimensional vertically directed array in which left reference samples are rotated clockwise can be applied. By arranging the reference samples in a row, an intra-forecast mode angle can be considered.

[318] A reference sample determination parameter may be determined based on an intra-direction parameter. The reference sample determination parameter may include a reference sample index for specifying a reference sample and a weighting value parameter for determining a weighting value applied to the reference sample.

[319] Reference sample index iIdx and weighting value parameter ifact can be acquired through Equations 9 and 10 shown below, respectively. [EQUATION 9]

[320] In Equations 9 and 10, Pang denotes an intra-direction parameter. The reference sample specified by the reference sample index iIdx corresponds to an integer pel.

[321] At least one or more reference samples may be specified to derive a prediction sample. Specifically, the position of a reference sample used to derive a prediction sample may be specified by considering the slope of the prediction mode. For example, a reference sample used to derive a prediction sample may be specified using reference sample index iIdx.

[322] At this point, when the slope of the intra-prediction mode is not expressed by a reference sample, a prediction sample can be generated by interpolating a plurality of reference samples. For example, when the slope of the intra-prediction mode is a value between a slope between a prediction sample and a first reference sample and a slope between a prediction sample and a second reference sample, the prediction sample can be acquired by interpolating the first reference sample and the second reference sample. That is, when an angled line following an intra-prediction angle does not pass through a reference sample positioned within a whole pel, the prediction sample can be acquired by interpolating reference samples adjacent to the left and right sides or up and down from a position where the angled line passes.

[323] Equation 11 shows an example of acquiring a prediction sample based on reference samples. [EQUATION 11]

[324] In Equation 11, P denotes a prediction sample, and Ref_1D denotes any one of the one-dimensionally arranged reference samples. At this point, the position of a reference sample can be determined by the position (x, y) of the prediction sample and index iIdx of the reference sample.

[325] When the intraforecast mode slope can be expressed as a reference sample, the weighting value parameter ifact is set to 0. Consequently, Equation 11 can be simplified as shown in Equation 12. [EQUATION 12]

[326] The intraprediction for the current block may be performed based on a plurality of intraprediction modes. For example, an intraprediction mode may be derived for each prediction sample, and the prediction sample may be derived based on an intraprediction mode assigned to each prediction sample.

[327] Alternatively, an intraprediction mode may be derived for each region, and intraprediction for each region may be performed based on the intraprediction mode assigned to each region. Here, the region may include at least one sample. At least one of the size and shape of the region may be adaptively determined based on at least one of the size of the current block, the shape of the current block, and the intraprediction mode of the current block. Alternatively, at least one of the size and shape of the region may be predefined in the encoder and decoder to be independent from the size or shape of the current block.

[328] Alternatively, intraprediction may be performed based on each of a plurality of intrapredictions, and a final prediction sample may be derived based on an average operation or a weighted sum operation of a plurality of prediction samples acquired via the plurality of intrapredictions. For example, a first prediction sample may be acquired by performing intraprediction based on a first intraprediction mode, and a second prediction sample may be acquired by performing intraprediction based on a second intraprediction mode. Thereafter, a final prediction sample may be acquired based on an average operation or a weighted sum operation of the first prediction sample and the second prediction sample. At this point, a weighting value assigned to each of the first prediction sample and the second prediction sample may be determined by considering at least one of the possibility of the first intraprediction mode being an angular/non-angular prediction mode, the possibility of the second intraprediction mode being an angular/non-angular prediction mode, and an intraprediction mode of a neighboring block.

[329] The plurality of intraprediction modes may be a combination of a non-angular intraprediction mode and an angular prediction mode, a combination of angular prediction modes, or a combination of non-angular prediction modes.

[330] Figure 24 is a view showing an example of angles formed between angular intraprediction modes and a straight line parallel to the x-axis.

[331] As shown in the example in Figure 24, angular prediction modes can exist between the lower left diagonal direction and the upper right diagonal direction. Describing in terms of an angle formed by the x-axis and the angular prediction mode, angular prediction modes can exist between 45 degrees (lower left diagonal direction) and -135 degrees (upper right diagonal direction).

[332] When the current block has a non-square shape, according to the intra-prediction mode of the current block, it may be a case of deriving a prediction sample using a reference sample farther from the prediction sample, instead of a reference sample even closer to the prediction sample, among the reference samples positioned on the angular line following the intra-prediction angle.

[333] Figure 25 is a view showing an embodiment of acquiring prediction samples when the current block has a non-square shape.

[334] For example, as shown in the example of Figure 25(a), it is assumed that the current block has a non-square shape of which the width is greater than the height, and the intraprediction mode of the current block is an angular intraprediction mode that has an angle between 0 and 45 degrees. In the above case, when the prediction sample A near the right column of the current block is derived, there may be a case of using the left reference sample L far from the prediction sample, instead of the top reference sample T near the prediction sample, among the reference samples positioned in an angular mode following the angle.

[335] As another example, as shown in the example of Figure 25(b), it is considered that the current block has a non-square shape of which the height is greater than the width, and the intraprediction mode of the current block is an angular intraprediction mode that has an angle between -90 and -135 degrees. In the above case, when the prediction sample A near the bottom row of the current block is derived, there may be a case of using the top reference sample T far from the prediction sample, instead of the left reference sample L near the prediction sample, among the reference samples positioned in an angular mode following the angle.

[336] To solve the problem described above, when the current block has a non-square shape, the intraprediction mode of the current block can be replaced with an intraprediction mode of the opposite direction. Consequently, angular prediction modes that have an angle larger or smaller than the angular prediction modes shown in Figure 21 can be used for a non-square shaped block. The angular intraprediction mode like this can be defined as a wide-angle intraprediction mode. The wide-angle intraprediction mode represents an angular intraprediction mode that does not belong to the range of 45 to -135 degrees.

[337] Figure 26 is a view showing wide-angle intraprediction modes.

[338] In the example shown in Figure 26, intraprediction modes with an index of -1 to -14 and intraprediction modes with an index of 67 to 80 represent wide-angle intraprediction modes.

[339] In Figure 26, although 14 wide-angle intraprediction modes (-1 to -14) that have an angle greater than 45 degrees and 14 wide-angle intraprediction modes (67 to 80) that have an angle less than -135 degrees are shown, a greater or lesser number of wide-angle intraprediction modes can be defined.

[340] When a wide-angle intraprediction mode is used, the length of top reference samples can be set to 2W + 1, and the length of left reference samples can be set to 2H + 1.

[341] Since a wide-angle intraprediction mode is used, sample A shown in Figure 25(a) can be predicted using reference sample T, and sample A shown in Figure 25(b) can be predicted using reference sample L.

[342] By adding existing intraprediction modes and N wide-angle intraprediction modes, a total of 67+N intraprediction modes can be used. For example, Table 5 shows intra-steering parameters of intraprediction modes when 20 wide-angle intraprediction modes are defined. [TABLE 5]

[343] When the current block has a non-square shape and the intraprediction mode of the current block acquired in step S2002 belongs to a transform range, the intraprediction mode of the current block may be transformed into a wide-angle intraprediction mode. The transform range may be determined based on at least one of the size, shape, and aspect ratio of the current block. Here, the ratio may represent a ratio of width to height of the current block.

[344] When the current block has a non-square shape of which the width is greater than the height, the transform range can be defined from an intraprediction mode index (e.g., 66) of the upper right diagonal direction (an index of the intraprediction mode in the upper right diagonal direction - N). Here, N can be determined based on the ratio of the current block. When the intraprediction mode of the current block belongs to the transform range, the intraprediction mode can be transformed into a wide-angle intraprediction mode. The transform can be subtraction of a predefined value from the intraprediction mode, and the predefined value can be the total number (e.g., 67) of intraprediction modes excluding the wide-angle intraprediction modes.

[345] According to the embodiment described above, 66° to 53° intraprediction modes can be transformed into -1° to -14° wide-angle intraprediction modes, respectively.

[346] When the current block has a non-square shape of which the height is greater than the width, the transform range can be defined from an intraprediction mode index (e.g., 2) of the lower left diagonal direction (an index of the intraprediction mode in the lower left diagonal direction + M). Here, M can be determined based on the ratio of the current block. When the intraprediction mode of the current block belongs to the transform range, the intraprediction mode can be transformed into a wide-angle intraprediction mode. The transform can be the addition of a preset value to the intraprediction mode, and the preset value can be the total number (e.g., 65) of angular intraprediction modes excluding the wide-angle intraprediction modes.

[347] According to the embodiment described above, 2° to 15° of intraprediction modes can be transformed into 67° to 80° of wide-angle intraprediction modes, respectively.

[348] Next, the intra-prediction modes belonging to the transform range will be referred to as wide-angle intra-substitution prediction modes.

[349] The transform range can be determined based on the current block ratio. For example, Tables 6 and 7 show a transform range when 35 intraprediction modes and 67 intraprediction modes are defined excluding the wide-angle intraprediction modes, respectively. [TABLE 6] [TABLE 7]

[350] As shown in the examples of Tables 6 and 7, the number of intra-wide-angle substitution prediction modes belonging to the transform range can vary according to the current block ratio.

[351] A transform range as shown in Table 8 can be defined by subdividing the ratio of the current block. [TABLE 8]

[352] It may be configured not to use the wide-angle intraprediction mode when a non-adjacent reference sample line is determined as the reference sample line of the current block or when a multi-line intraprediction coding method of selecting any one of a plurality of reference sample lines is used. That is, although the current block has a non-square shape and the intraprediction mode of the current block belongs to the transform range, the intraprediction mode of the current block may not be transformed to the wide-angle intraprediction mode.

[353] Alternatively, non-adjacent reference sample lines may be configured as not being available as the reference sample line of the current block, or may be configured to not use the multi-line intraprediction coding method of selecting any one of a plurality of reference sample lines when the intraprediction mode of the current block is determined as the wide-angle intraprediction mode. When the multi-line intraprediction coding method is not used, an adjacent reference sample line may be determined as the reference sample line of the current block.

[354] When the wide-angle intraprediction mode is not used, the sum of nTbW and nTbH can be set to refW and refH. Consequently, excluding the top-left reference sample, a non-adjacent reference sample with a distance of i from the current block can include as many (nTbW + nTbH + offsetX[i]) top reference samples and as many as (nTbW + nTbH + offsetY[i]) left reference samples. That is, a non-adjacent reference sample with a distance of i from the current block can include up to (2nTbW + 2nTbH + offsetX[i] + offsetY[i] + 1) reference samples. For example, when the value of whRatio is greater than 1, the value of offsetX can be set to be greater than the value of offsetY. For example, the offsetX value can be set to 1 and the offsetY value can be set to 0. Conversely, when the whRatio value is less than 1, the offsetY value can be set to be greater than the offsetX value. For example, the offsetX value can be set to 0 and the offsetY value can be set to 1.

[355] As wide-angle intraprediction modes are used in addition to existing intraprediction modes, resources required to encode the wide-angle intraprediction modes increase, and thus coding efficiency may be decreased. Consequently, coding efficiency may be improved by coding replacement intraprediction modes for the wide-angle intraprediction modes, rather than coding the wide-angle intraprediction modes as they are.

[356] For example, when the current block is coded using the 67th wide-angle intraprediction mode, the 2nd intraprediction mode, which is the replacement intraprediction mode of the 67th wide-angle intraprediction mode, can be coded using the current block's intraprediction mode. Furthermore, when the current block is coded using the -1st wide-angle intraprediction mode, the 66th intraprediction mode, which is the replacement intraprediction mode of the -1st wide-angle intraprediction mode, can be coded using the current block's intraprediction mode.

[357] The decoder may decode the intraprediction mode of the current block and determine whether the decoded intraprediction mode belongs to the transform range. When the decoded intraprediction mode is a wide-angle substitution intraprediction mode, the intraprediction mode may be transformed into the wide-angle intraprediction mode.

[358] Alternatively, when the current block is encoded using the wide-angle intraprediction mode, the wide-angle intraprediction mode may be encoded as is.

[359] Encoding the intraprediction mode may be performed based on the MPM list described above. Specifically, when a neighboring block is coded using a wide-angle intraprediction mode, the MPM may be set based on a wide-angle replacement intraprediction mode corresponding to the wide-angle intraprediction mode.

[360] When a forecast block is generated as a result of performing intraforecasting, forecast samples may be updated based on the position of each of the forecast samples included in the forecast block. The update method like this may be termed as an intraweighted forecast method based on a sample position or Position Dependent Intraforecast Combination (PDPC).

[361] Whether or not to use PDPC can be determined by considering the current block's intraprediction mode, the current block's reference sample line, the current block's size, or a color component. For example, when the current block's intraprediction mode is at least one of a planar mode, a Dc mode, a vertical direction mode, a horizontal direction mode, a mode that has a smaller index value than the vertical direction mode, and a mode that has a larger index value than the horizontal direction mode, PDPC can be used. Alternatively, only when at least one of the current block's width and height is greater than 4 can PDPC be used. Alternatively, only when the current block's reference picture line index is 0 can PDPC be used. Alternatively, only when the current block's reference picture line index is equal to or greater than a predefined value can PDPC be used. Alternatively, PDPC can be used only for a luminance component. Alternatively, depending on whether two or more of the conditions listed above are met, whether or not to use PDPC can be determined.

[362] As another example, information indicating whether PDPC is applied or not can be signaled via a bit stream.

[363] When a forecast sample is acquired via intra-forecast sampling, a reference sample used to correct the forecast sample can be determined based on the position of the acquired forecast sample. For convenience of explanation, in the embodiment described below, a reference sample used to correct a forecast sample is referred to as a PDPC reference sample. Furthermore, a forecast sample acquired via intra-forecasting is referred to as a first forecast sample, and a forecast sample acquired by correcting the first forecast sample is referred to as a second forecast sample.

[364] Figure 27 is a view showing an embodiment of PDPC application.

[365] The first prediction sample may be corrected using at least one PDPC reference sample. The PDPC reference sample may include at least one of a reference sample adjacent to the top left corner of the current block, a top reference sample positioned at the top of the current block, and a left reference sample positioned at the left side of the current block.

[366] At least one of the reference samples belonging to the reference sample line of the current block may be defined as the PDPC reference sample. Alternatively, regardless of the reference sample line of the current block, at least one of the reference samples belonging to the reference sample line having index 0 may be defined as the PDPC reference sample. For example, although the first prediction sample is acquired using a reference sample included in a reference sample line of index 1 or index 2, the second prediction sample may be acquired using a reference sample included in the reference sample line of index 0.

[367] The number or positions of the PDPC reference samples used to correct the first prediction sample may be determined by considering at least one of the intraprediction mode of the current block, the size of the current block, the shape of the current block, and the position of the first prediction sample.

[368] For example, when the intraprediction mode of the current block is a flat mode or a Dc mode, the second prediction sample may be acquired using a top reference sample and a left reference sample. At this point, the top reference sample may be a vertical reference sample to the first prediction sample (e.g., a reference sample of the same x-coordinate), and the left reference sample may be a horizontal reference sample to the first prediction sample (e.g., a reference sample of the same y-coordinate).

[369] When the intra-forecast mode of the current block is a horizontal intra-forecast direction mode, the second forecast sample may be acquired using an upper reference sample. At this point, the upper reference sample may be a vertical reference sample to the first forecast sample.

[370] When the intra-forecast mode of the current block is a vertical direction intra-forecast mode, the second forecast sample may be acquired using a left reference sample. At this point, the left reference sample may be a horizontal reference sample to the first forecast sample.

[371] When the intraprediction mode of the current block is a lower left diagonal direction intraprediction mode or an upper right diagonal direction intraprediction mode, the second prediction sample may be acquired based on an upper left reference sample, an upper reference sample, and a left reference sample. The upper left reference sample may be a reference sample adjacent to the upper left corner of the current block (e.g., a reference sample at the position of (-1, -1)). The upper reference sample may be a reference sample positioned in the upper right diagonal direction of the first prediction sample, and the left reference sample may be a reference sample positioned in the lower left diagonal direction of the first prediction sample.

[372] In summary, when the position of the first prediction sample is (x, y), R(-1, -1) can be set as the upper left reference sample, and either R(x + y + 1, -1) or R(x, -1) can be set as the upper reference sample. Furthermore, R(-1, x + y + 1) or R(-1, y) can be set as the left reference sample.

[373] A prediction mode may be applied to the current block a plurality of times, or a plurality of prediction modes may be redundantly applied to the current block. Thus, a prediction method using a homogeneous or heterogeneous prediction mode may be termed as a combined prediction mode (or Multi-hypothesis Prediction Mode).

[374] The combined forecast mode includes at least one of a mode combining a blending mode and a blending mode, a mode combining interforecast and intraforecast, a mode combining a blending mode and an advanced motion vector forecast mode, and a mode combining a blending mode and intraforecast.

[375] In the combined prediction mode, a first prediction block may be generated based on the first prediction mode, and a second prediction block may be generated based on the second prediction mode. Then, a third prediction block may be generated based on a weighted sum operation of the first prediction block and the second prediction block. The third prediction block may be set as the final prediction block of the current block.

[376] Whether or not to use combined prediction mode may be determined based on the size or format of the current block. For example, whether or not to use combined prediction mode may be determined based on at least one of the coding block size, the number of samples included in the coding block, the width of the coding block, and the height of the coding block. For example, when at least one of the width and height of the coding block is greater than or equal to 128, or when the number of samples included in the coding block is less than or equal to 64, combined prediction mode may not be applied to the coding block.

[377] Additionally, information indicating whether the combined prediction mode is applied to the current block may be signaled via a bit stream. For example, the information may be a 1-bit flag. For example, flag mh_intra_flag indicating whether the combined prediction mode combining the merge mode and intraprediction is used may be signaled via a bit stream. When mh_intra_flag is 1, it indicates that the combined prediction mode combining the merge mode and intraprediction is used, and when mh_intra_flag is 0, it indicates that the combined prediction mode combining the merge mode and intraprediction is not used.

[378] Only when the prediction coding mode of the current block is interprediction mode and merge mode is applied to the current block, the combined prediction mode combining merge mode and intraprediction can be applied to the current block. That is, mh_intra_flag can be flagged when the value of flag merge_flag indicating whether merge mode is applied to the current block is 1.

[379] In the combined forecast mode combining blending mode and intraforecast, the current block's intraforecast mode can be set to a predefined intraforecast mode. For example, when the combined forecast mode is used, the current block's intraforecast mode can be set to a flat mode.

[380] As another example, when the combined prediction mode combining the blending mode and the intraprediction is used, one of the candidate intraprediction modes may be determined as the intraprediction mode of the current block. Here, the candidate intraprediction mode may include at least one of a non-angular intraprediction mode and an angular intraprediction mode of a specific direction. Here, the non-angular intraprediction mode includes at least one of the Dc mode and the planar mode, and the angular intraprediction mode includes at least one of the intraprediction mode of the horizontal direction, the intraprediction mode of the vertical direction, and the intraprediction mode of a diagonal direction. For example, when the combined prediction mode is used, only the Dc mode, the planar mode, the intraprediction mode of the horizontal direction, or the intraprediction mode of the vertical direction may be set as the intraprediction mode of the current block. Alternatively, when using the combined forecast mode, only the flat mode, the horizontal direction intraforecast mode, or the vertical direction intraforecast mode can be set as the current block's intraforecast mode. Alternatively, when using the combined forecast mode, only the Dc mode, the flat mode, or the vertical direction intraforecast mode can be set as the current block's intraforecast mode. Alternatively, when using the combined forecast mode, only the Dc mode, the flat mode, or the horizontal direction intraforecast mode can be set as the current block's intraforecast mode.

[381] Index information for specifying any of the candidate intraprediction modes may be signaled via a bit stream. For example, index mh_intra_idx specifying any of the candidate intraprediction modes may be signaled via a bit stream. Tables 9 and 10 show intraprediction modes according to the value of mh_intra_idx. An intraprediction mode indicated by mh_intra_idx may be determined as the intraprediction mode of the current block. [TABLE 9] [TABLE 101

[382] The number of candidate intraprediction modes may have a fixed value in the encoder and decoder. Alternatively, the number or types of candidate intraprediction modes may vary according to at least one of the current block size, the current block format, and whether a neighboring block is intraprediction coded.

[383] As another example, when the combined forecast mode combining the merge mode and intraforecast is applied, the intraforecast mode of the current block can be determined based on MPMs. At this point, the number of MPMs that the MPM list includes when general intraforecast is applied may be different from the number of MPMs that the MPM list includes when the combined forecast mode is applied. For example, when the combined forecast mode is not applied, the MPM list may include 6 or 5 MPMs, while when the combined forecast mode is applied, the MPM list may include 4 or 3 MPMs. That is, when general intraforecast is performed, N MPMs are used, while when the combined forecast mode is applied, MPMs less than N may be used.

[384] For example, it is considered that when a general intraprediction mode is applied, MPMs derived based on the intraprediction mode of a neighboring block adjacent to the current block are Flat mode, Dc mode, INTRA_MODE32, INTRA_MODE31, INTRA_MODE33, and INTRA_MODE30. When the combined prediction mode is applied to the current block, the intraprediction mode of the current block can be determined using only some of the 6 MPMs. For example, the intraprediction mode of the current block can be determined using an MPM list including 3 MPMs having the smallest indexes among the MPMs, i.e., Flat mode, Dc mode, and INTRA_MODE32. Alternatively, any one that has a predefined index value among the MPMs can be determined as the intraprediction mode of the current block.

[385] As another example, it is possible to equally define the number of MPMs used to determine the intraforecast mode of the current block when an overall intraforecast is applied, and the number of MPMs used to determine the intraforecast mode of the current block when the combined forecast mode is applied.

[386] When the combined prediction mode is applied, the MPM flag encoding may be omitted, and the MPM flag may be assumed to be true. That is, when the combined prediction mode is applied, any one of the MPMs may be determined as the intraprediction mode of the current block.

[387] It can be configured not to use the multi-line intraprediction coding method when the combined prediction mode combining the blending mode and intraprediction is applied to the current block. When the multi-line intraprediction coding method is not used, an adjacent reference sample line can be used for the intraprediction of the current block.

[388] Alternatively, it may be configured to use the multi-line intraprediction coding method when the blended prediction mode combining the blending mode and intraprediction is applied to the current block. Information for determining the reference sample line of the current block among a plurality of reference sample lines may be signaled via a bit stream. Alternatively, the reference sample line of the current block may be determined by considering at least one of the size of the current block, the format of the current block, the intraprediction mode of the current block, and the reference sample line of a neighboring block.

[389] The number of available candidate intraprediction modes may be set differently according to the reference sample line of the current block. For example, when an adjacent reference sample line is used, the candidate intraprediction mode may include at least one of the planar mode, the Dc mode, the horizontal direction intraprediction mode, the vertical direction intraprediction mode, the lower left diagonal direction intraprediction mode, and the upper right diagonal direction intraprediction mode. Conversely, when a non-adjacent reference sample line is used, the candidate intraprediction mode may include at least one of the horizontal direction intraprediction mode, the vertical direction intraprediction mode, the lower left diagonal direction intraprediction mode, and the upper right diagonal direction intraprediction mode.

[390] A first prediction block can be generated using a merge candidate selected in merge mode, and a second prediction block can be generated using intraprediction mode. At this point, it can be configured to not apply triangular partitioning to the current block when the first prediction block is generated. That is, triangular partitioning may not be allowed in the combined prediction mode combining merge mode and intraprediction. Furthermore, it can be configured to not apply PDPC to the current block when the second prediction block is generated. That is, PDPC may not be allowed in the combined prediction mode combining merge mode and intraprediction.

[391] A third prediction block may be generated based on a weighted sum operation of the first prediction block and the second prediction block. For example, Equation 13 shows an example of generating a third prediction block based on a weighted sum operation of the first prediction block and the second prediction block. [EQUATION 13]

[392] In Equation 13, Pmerge denotes a first prediction block obtained based on the merge mode, and Pintra denotes a second prediction block obtained based on the intraprediction. Pcomb denotes a third prediction block obtained by combining the first prediction block and the second prediction block. w denotes a first weighting value applied to the first prediction block. A second weighting value applied to the second prediction block can be derived by subtracting the first weighting value w from the constant N. Here, N can have a predefined value at both the encoder and the decoder. For example, N can be 4 or 8. Alternatively, the constant N can be derived based on at least one of the size, the format, and the intraprediction mode of the current block.

[393] Unlike the example shown in Equation 13, w can be set as the second weight value, and a value obtained by subtracting the second weight value from the predetermined constant N can be set as the first weight value.

[394] The weighting values applied to the first prediction block and the second prediction block may be determined based on at least one of the current block's intraprediction mode and whether the merge mode has bidirectional motion information. For example, the first weighting value w when the current block's intraprediction mode is angular prediction mode may be greater than the first weighting value w when the current block's intraprediction mode is non-angular prediction mode. Alternatively, the first weighting value w when bidirectional prediction is performed by the merge candidate may be greater than the first weighting value w when unidirectional prediction is performed by the merge candidate.

[395] As another example, weight values may be determined by considering prediction coding modes of neighboring blocks adjacent to the current block. Here, the neighboring block may include at least one of a top neighbor block adjacent to the top of the current block, a left neighbor block adjacent to the left side of the current block, and an top left neighbor block adjacent to the top left corner of the current block. A weight value may be determined based on the number of intraprediction-coded neighboring blocks among neighboring blocks adjacent to the current block. For example, as the number of intraprediction-coded neighboring blocks among neighboring blocks adjacent to the current block increases, the first weight value may have a smaller value, and the second weight value may have a larger value. Conversely, as the number of non-intraprediction-coded neighboring blocks among neighboring blocks adjacent to the current block decreases, the first weight value may have a larger value, and the second weight value may have a smaller value. For example, when both the left neighbor block and the upper neighbor block are intraprediction encoded, the second weighting value can be set to n. When only one of the left neighbor blocks and the upper neighbor blocks is intraprediction encoded, the second weighting value can be set to n/2. When neither the left neighbor block nor the upper neighbor block is intraprediction encoded, the second weighting value can be set to n/4. Here, n represents an integer equal to or less than N.

[396] The weighting value may be determined by the sample unit or sub-block in the current block. For example, the current block may be partitioned into a plurality of sub-blocks, and a weighting value applied to each sub-block may be set to have a different value. At this point, the number of sub-blocks may have a fixed value in the encoder and decoder. Alternatively, the number of sub-blocks may be adaptively determined based on at least one of the current block size, the current block format, the current block intra-prediction mode, and the predictive coding mode of a neighboring block.

[397] Figure 28 is a view showing application of examples of weighting values.

[398] The current block may be partitioned into a plurality of sub-blocks, and a weighting value applied to each sub-block may be set to a different value. For example, as shown in the examples of Figures 28(a) and (b), for a sub-block including a sample positioned in the upper left corner of the current block, the second weighting value is set to be greater than the first weighting value. Conversely, for a sub-block including a sample positioned in the lower right corner of the current block, the first weighting value may be set to be greater than the second weighting value.

[399] A third prediction sub-block for a sub-block can be obtained by performing weighted prediction on the first prediction sub-block and the second prediction sub-block corresponding to the sub-block. At this point, as shown in the example of Figure 28, the weighting values used for the weighted prediction can be set differently for each sub-block.

[400] A weight value assigned to a sub-block may be determined differently depending on whether the merge candidate has bidirectional motion information.

[401] Figure 29 is a view showing examples of assigning a different weighting value to a sub-block according to whether a merge candidate has bidirectional motion information.

[402] When a merge candidate has bidirectional motion information, the first weight value w assigned to at least one of the sub-blocks may be set to a larger value compared to a case where the merge candidate has unidirectional motion information. For example, in Figures 29(a) and (b), it is shown that for all sub-blocks, the weight value w applied when bidirectional prediction is applied has a larger value than the weight value w applied when unidirectional prediction is applied.

[403] Alternatively, a weighting value applied to a sub-block may be determined based on the intraprediction mode of the current block.

[404] Figure 30 is a view showing examples of assigning a different weighting value to a sub-block according to the intraprediction mode of the current block.

[405] When the intraprediction mode of the current block is in the vertical direction, the second weighting value of a sub-block positioned at the top of the current block may be set to be greater than the second weighting value of a sub-block positioned at the bottom of the current block.

[406] For example, the second weighting value may be set to be greater than the first weighting value for a sub-block positioned at the top of the current block, while the first weighting value and the second weighting value may be set equally, or the first weighting value may be set to be greater than the second weighting value for a sub-block positioned at the bottom of the current block.

[407] When the intraprediction mode of the current block is in the horizontal direction, the second weighting value of a sub-block positioned on the left side of the current block may be set to be greater than the second weighting value of a sub-block positioned on the right side of the current block.

[408] For example, the second weight value is set to be greater than the first weight value for a sub-block positioned on the left side of the current block, while the first weight value and the second weight value may be set equally, or the first weight value may be set to be greater than the second weight value for a sub-block positioned on the right side of the current block.

[409] As another example, when the intraprediction mode of the current block is in the vertical direction, the first weight value of a sub-block positioned at the top of the current block may be set to be greater than the first weight value of the sub-block positioned at the bottom of the current block, or when the intraprediction mode of the block is in the horizontal direction, the first weight value of a sub-block positioned at the left side of the current block may be set to be greater than the first weight value of a sub-block positioned at the right side of the current block.

[410] As another example, a weight value assigned to each sub-block may be determined based on the prediction coding modes of neighboring blocks adjacent to the current block. For example, when the current block's top neighboring block is intra-prediction coded and the current block's left neighboring block is not intra-prediction coded, the second weight value of a sub-block positioned at the top of the current block may be set to be greater than the second weight value of a sub-block positioned at the bottom of the current block. Conversely, when the current block's top neighboring block is not intra-prediction coded and the current block's left neighboring block is intra-prediction coded, the second weight value of a sub-block positioned at the left side of the current block may be set to be greater than the second weight value of a sub-block positioned at the right side of the current block.

[411] For at least one of the plurality of subblocks, the first weight value or the second weight value may be set to 0. That is, for at least one of the plurality of subblocks, the first prediction subblock or the second prediction subblock may be set to the third prediction subblock. Consequently, combined prediction may not be performed on at least one of the plurality of subblocks. For example, for at least one of the plurality of subblocks, the first weight value w may be set to 0, or the first weight value w may be set to constant N.

[412] A block to which the combined prediction mode combining the blending mode and intraprediction is applied can be considered as being interprediction coded. Consequently, the intraprediction mode of a block encoded in the combined prediction mode can be set to be unavailable for deriving an intraprediction mode of a block that will be coded/decoded after that. When an MPM is derived, the intraprediction mode of an unavailable block can be considered as the flat mode.

[413] On the other hand, motion information of a block encoded by the combined prediction mode can be defined as available to derive motion information of a block that will be encoded/decoded thereafter.

[414] In the mode combining a blending mode and a merging mode, motion compensation prediction may be performed using a plurality of merging candidates. Specifically, a first prediction block may be generated using a first merging candidate, and a second prediction block may be generated using a second merging candidate. A third prediction block may be generated based on a weighted sum operation of the first prediction block and the second prediction block.

[415] The information for specifying the first merge candidate and the second merge candidate may be signaled via a bit stream, respectively. For example, the merge_idx index information for specifying the first merge candidate and the merge_2nd_idx index information for specifying the second merge candidate may be signaled via a bit stream. The second merge candidate may be determined based on the merge_2nd_idx index information and the merge_idx index information.

[416] The merge_idx index information specifies any of the merge candidates included in the merge candidate list.

[417] The merge_2nd_idx index information may specify any of the remaining merge candidates excluding the merge candidate specified by merge_idx. Consequently, when the value of merge_2nd_idx is less than merge_idx, a merge candidate whose index is the value of merge_2nd_idx may be set as the second merge candidate. When the value of merge_2nd_idx is equal to or greater than the value of merge_idx, a merge candidate that has a value obtained by adding 1 to the value of merge_2nd_idx as an index may be set as the second merge candidate.

[418] Alternatively, the second merge candidate can be specified by considering the search order of candidate blocks.

[419] Figure 31 is a view showing an example where a second merge candidate is specified considering the search order of candidate blocks.

[420] In the example shown in Figure 31, marked indices on neighboring samples and non-neighboring samples indicate the search order of candidate blocks. For example, candidate blocks can be sequentially searched from position A0 to position A14.

[421] When block A4 is selected as the first merge candidate, a merge candidate derived from a candidate block that has a search order close to A4 can be specified as the second merge candidate. For example, a merge candidate derived from A5 can be selected as the second merge candidate. When the candidate block at position A5 cannot be used as a merge candidate, a merge candidate derived from a nearby candidate block can be selected as the second merge candidate.

[422] The first merge candidate and the second merge candidate may be selected from merge candidates derived from non-neighboring blocks.

[423] Figure 32 is a view showing an example where a first merge candidate and a second merge candidate are selected from merge candidates derived from non-neighboring blocks.

[424] As shown in the example in Figure 32, merge candidates derived from a first candidate block and a second candidate block that are not adjacent to the current block can be selected as the first merge candidate and the second merge candidate, respectively. At this point, the block row to which the first candidate block belongs and the block row to which the second candidate block belongs may be different. For example, the first merge candidate can be derived from any of candidate blocks A5 through A10, and the second merge candidate can be derived from any of candidate blocks A11 through A15.

[425] Alternatively, it may be configured such that the first candidate block and the second candidate block are not included in the same row (e.g., line or column).

[426] As another example, the second merge candidate can be specified based on the first merge candidate. At this point, the first merge candidate can be specified by the signed merge_idx index information of a bitstream. For example, a merge candidate that is adjacent to the first merge candidate can be specified as the second merge candidate. Here, the merge candidate that is adjacent to the first merge candidate can mean a merge candidate that has an index difference of 1 from the first merge candidate. For example, a merge candidate that has an index value of merge_idx + 1 can be defined as the second merge candidate. At this point, when the value of merge_idx+1 is greater than the maximum index value (or when the index value of the first merge candidate is the maximum index), a merge candidate that has an index value of merge_idx-1 or a merge candidate that has an index value of a predefined value (e.g., 0) can be set as the second merge candidate.

[427] Alternatively, the merge candidate adjacent to the first merge candidate may mean a merge candidate derived from a candidate block spatially adjacent to a candidate block used to derive the first merge candidate. Here, the candidate block's neighboring candidate block may mean a block adjacent to the candidate block in the left, right, top, bottom, or diagonal direction.

[428] As another example, the second merge candidate may be specified based on the motion information of the first merge candidate. For example, a merge candidate sharing the same reference image with the first merge candidate may be selected as the second merge candidate. When there are a plurality of merge candidates sharing the same reference image with the first merge candidate, a merge candidate having the smallest index among the plurality of merge candidates, a merge candidate having the smallest index difference from the first merge candidate, or a merge candidate having the smallest motion vector difference from the first merge candidate may be selected as the second merge candidate. Alternatively, the second merge candidate may be selected based on the index information specifying any one of the plurality of merge candidates.

[429] Alternatively, when the first merge candidate is unidirectional prediction of a first direction, a merge candidate including motion information in a second direction may be defined as the second merge candidate. For example, when the first merge candidate has L0 direction motion information, a merge candidate having L1 direction motion information may be selected as the second merge candidate. When there are a plurality of merge candidates having L1 direction motion information, a merge candidate having the smallest index or a merge candidate having the smallest index difference from the first merge candidate among the plurality of merge candidates may be defined as the second merge candidate. Alternatively, the second merge candidate may be selected based on the index information specifying any one among the plurality of merge candidates.

[430] As another example, one among the merge candidates derived from neighboring blocks adjacent to the current block may be set as the first merge candidate, and one among the merge candidates derived from non-neighboring blocks non-adjacent to the current block may be set as the second merge candidate.

[431] As another example, one among the merge candidates derived from candidate blocks positioned at the top of the current block may be set as the first merge candidate, and one among the merge candidates derived from candidate blocks positioned at the left side may be set as the second merge candidate.

[432] In the combined prediction mode combining a blending mode and a blending mode as described above, the blending mode may mean a blending mode based on a translational motion model (hereinafter referred to as a translational blending mode) or a blending mode based on an affine motion model (hereinafter referred to as an affine blending mode). That is, motion compensation prediction may be performed by combining a translational blending mode and a translational blending mode or by combining an affine blending mode and an affine blending mode.

[433] As another example, when a neighboring block used to derive the first merge candidate is encoded based on affine motion information, a merge candidate derived from the neighboring block encoded based on the affine motion information may be defined as a second merge candidate. For example, when the first merge candidate is an affine merge candidate, or when the first merge candidate is derived based on the motion information of a sub-block in a coding block encoded based on the affine motion information, the affine merge candidate or the merge candidate derived based on the motion information of a sub-block in a coding block encoded based on the affine motion information may be selected as a second merge candidate. Here, the second merge candidate may include at least one of a merge candidate closest to the first merge candidate in the search order, a merge candidate having the smallest index difference from the first merge candidate, a merge candidate having the smallest index, and a merge candidate having the smallest motion vector difference from the first merge candidate, among the merge candidates that satisfy the above conditions.

[434] Unlike the example described above, when a neighboring block used to derive the first merge candidate is encoded based on translational motion information (i.e., non-affine motion information), a merge candidate derived from the neighboring block encoded based on translational motion information may be defined as a second merge candidate. For example, when the first merge candidate is a non-affine merge candidate, a non-affine merge candidate may be selected as the second merge candidate. Here, the second merge candidate may include at least one of a merge candidate closest to the first merge candidate in the search order, a merge candidate having the smallest index difference from the first merge candidate, a merge candidate having the smallest index, and a merge candidate having the smallest motion vector difference from the first merge candidate, among the non-affine merge candidates.

[435] Alternatively, the second merge candidate is derived by adding or subtracting an offset to or from the index of the first merge candidate, and when the neighboring block encoding methods used to derive the first merge candidate and the second merge candidate are different, the second merge candidate may be redefined to another merge candidate.

[436] Figure 33 is a view to describe an example of determining a second merge candidate.

[437] When the first merge candidate is selected, a merge candidate that has an index derived by adding or subtracting an offset to or from the index of the first merge candidate may be selected as the second merge candidate. For example, a merge candidate that has a value indicated by a bitstream's signed merge_idx as an index may be selected as the first merge candidate, and a merge candidate having merge_idx + 1 as an index may be selected as the second merge candidate.

[438] At this point, when the first merge candidate is an affine merge candidate or is derived from a sub-block belonging to a coding block encoded by an affine motion model, an affine merge candidate or a merge candidate derived from a sub-block belonging to a coding block encoded by an affine motion model shall be defined as the second merge candidate. When a merge candidate having merge_idx + 1 as an index is not an affine merge candidate or is not derived from a sub-block belonging to a coding block encoded by an affine motion model, another merge candidate may be defined as the second merge candidate. Here, another merge candidate may be the first merge candidate or a merge candidate having the smallest difference of merge_idx + 1 among the affine merge candidates or the merge candidates derived from a sub-block belonging to a coding block encoded by an affine motion model.

[439] Conversely, when the first merge candidate is a non-affine merge candidate, a non-affine merge candidate should be set as the second merge candidate. When a merge candidate having merge_idx + 1 as an index is an affine merge candidate or is derived from a sub-block belonging to a coding block encoded by an affine motion model, another merge candidate may be set as the second merge candidate. Here, another merge candidate may be the first merge candidate or a merge candidate having the smallest difference of merge_idx + 1 among the non-affine merge candidates. For example, in the example shown in Figure 33, it is shown that merge candidate A3 having merge_idx + 2 as an index is set as the second merge candidate.

[440] As another example, motion compensation prediction can be performed by combining a translational blending mode and an affine blending mode. That is, either one of the first blending candidate and the second blending candidate can be an affine blending candidate, and the other one can be a non-affine blending candidate.

[441] A combined prediction block may be acquired through a weighted sum operation of the first prediction block derived from the first merge candidate and the second prediction block derived based on the second merge candidate. At this point, the weighting value applied to the first prediction block may be set to be greater than the weighting value applied to the second prediction block.

[442] Alternatively, the weighting values may be determined based on the motion information of the first merge candidate and the motion information of the second merge candidate. For example, the weighting values applied to the first prediction block and the second prediction block may be determined based on the difference in the output order of the reference image and the current image. Specifically, as the difference in the output order of the reference image and the current image increases, the weighting value applied to the prediction block may be set to a smaller value.

[443] Alternatively, the weighting values applied to the first prediction block and the second prediction block may be determined by considering the sizes or shapes of the candidate block used to derive the first merge candidate (then referred to as the first candidate block) and the candidate block used to derive the second merge candidate (then referred to as the second candidate block). For example, a weighting value applied to a prediction block derived from the first candidate block or the second candidate block that has a similar shape to the current block may be set to a larger value. Conversely, a weighting value applied to a prediction block derived from the first candidate block or the second candidate block that has a dissimilar shape to the current block may be set to a smaller value.

[444] Figure 34 is a view showing an example where weighting values applied to prediction blocks are determined based on the shape of candidate blocks.

[445] The current block is considered to have a non-square shape in which the width is greater than the height.

[446] A first prediction block and a second prediction block may be derived based on the first merge candidate and the second merge candidate, and a combined prediction block may be generated based on a weighted sum operation of the first prediction block and the second prediction block. At this point, the weighting values applied to the first prediction block and the second prediction block may be determined based on the shapes of the first candidate block and the second candidate block.

[447] For example, in the example shown in Figure 34, the first candidate block has a square shape, and the second candidate block has a non-square shape of which the width is greater than the height. Since the shape of the second candidate block is the same as that of the current block, the weighting value applied to the second prediction block can be set to be greater than the weighting value applied to the first prediction block. For example, a weighting value of 5/8 can be applied to the second prediction block, and a weighting value of 3/8 can be applied to the first prediction block. Equation 14 shows an example of deriving a combined prediction block based on a weighted sum operation of the first prediction block and the second prediction block. [EQUATION 14]

[448] P(x, y) denotes a combined prediction block, P1(x, y) denotes a first prediction block, and P2(x, y) denotes a second prediction block.

[449] As another example, the weighting values applied to the first prediction block and the second prediction block may be determined based on the shape of the current block. For example, when the current block has a non-square shape of which the width is greater than the height, a greater weighting value may be applied to a prediction block generated based on a merge candidate derived based on a candidate block positioned at the top of the current block between the first merge candidate and the second merge candidate. When both the first merge candidate and the second merge candidate are derived from candidate blocks positioned at the top of the current block, weighting values applied to the first prediction block and the second prediction block may be set equally. Conversely, when the current block has a non-square shape of which the height is greater than the width, a greater weighting value may be applied to a prediction block generated based on a merge candidate derived based on a candidate block positioned at the left side of the current block between the first merge candidate and the second merge candidate. When both the first merge candidate and the second merge candidate are derived from candidate blocks positioned to the left of the current block, weighting values applied to the first prediction block and the second prediction block can be set equally. When the current block is square-shaped, weighting values applied to the first prediction block and the second prediction block can be set equally.

[450] As another example, a weighting value applied to each prediction block can be determined based on the distance between the current block and a candidate block. Here, the distance can be derived based on an x-axis coordinate difference, a y-axis coordinate difference from the current block, or a minimum value between them. A weighting value applied to a prediction block derived from a merge candidate that has a small distance from the current block can be set to be greater than a weighting value applied to a prediction block derived from a merge candidate that has a large distance from the current block. For example, in the example shown in Figure 31, the first merge candidate is derived from a neighboring block adjacent to the current block, and the second merge candidate is derived from a non-neighboring block that is not adjacent to the current block. In this case, since the x-axis distance between the first candidate block and the current block is smaller than the x-axis distance between the second candidate block and the current block, the weighting value applied to the first prediction block can be set to be larger than the weighting value applied to the second prediction block.

[451] Alternatively, when both the first merge candidate and the second merge candidate are derived from non-neighboring blocks, a larger weighting value may be assigned to a prediction block derived from a non-neighboring block at a closer distance to the current block among the non-neighboring blocks. For example, in the example shown in Figure 32, since the y-axis distance between the first candidate block and the current block is smaller than the y-axis distance between the second candidate block and the current block, the weighting value applied to the first prediction block may be set to be larger than the weighting value applied to the second prediction block.

[452] Integrated motion information may be derived based on the first merge candidate and the second merge candidate, and motion compensation prediction for the current block may be performed based on the integrated motion information. For example, a motion vector of the current block may be derived based on an average operation or a weighted sum operation of the motion vector of the first merge candidate and the motion vector of the second merge candidate. At this point, the weighting value applied to the motion vector of the first merge candidate and the weighting value applied to the motion vector of the second merge candidate may be determined by the embodiments described above.

[453] When the first merge candidate is a non-affine merge candidate and the second affine merge candidate is an affine merge candidate, a motion vector of the current block can be derived by scaling the motion vector of the second merge candidate. Equation 15 shows an example of deriving a motion vector of the current block. [EQUATION 15]

[454] In Equation 15, (mvX, mvY) denotes the motion vector of the current block, (mv0x, mv0y) denotes the motion vector of the first merge candidate, and (mv1x, mv1y) denotes the motion vector of the second merge candidate. M denotes a scale parameter. M can be preset in both the encoder and decoder. Alternatively, the value of the scale parameter M can be determined according to the size of the current block or a candidate block. For example, when the width or height of the second candidate block is greater than 32, M can be set to 3, and in other cases, M can be set to 2.

[455] The motion information of the first merge candidate or the second merge candidate may be stored as the motion information of a block to which the combined prediction mode combining a merge mode and a blend mode is applied. The stored motion information may be used to derive motion information of a block that will be encoded/decoded thereafter.

[456] Alternatively, the block may be partitioned into a plurality of sub-blocks, and any one of the motion information of the first merge candidate and the motion information of the second merge candidate may be stored as the motion information of each sub-block. At this point, the motion information of one part of the plurality of sub-blocks may be defined as the motion information of the first merge candidate, and the motion information of the other part may be defined as the motion information of the second merge candidate.

[457] Alternatively, the integrated motion information derived based on the motion information of the first merge candidate and the second merge candidate may be stored as the motion information of a block to which the combined prediction mode combining a merge mode and a blend mode is applied.

[458] In the prediction mode in which a blending mode and an advanced motion vector prediction mode are combined, a first prediction block is generated using the motion information derived from a blending candidate, and a second prediction block is generated using a motion vector derived from a motion vector prediction candidate.

[459] In the advanced motion vector prediction mode, a motion vector prediction candidate may be derived from a neighboring block adjacent to the current block or a placed block in a placed image. Thereafter, any one of a plurality of motion vector prediction candidates may be specified, and the specified motion vector prediction candidate may be set as the motion vector predictor of the current block. Thereafter, a motion vector of the current block may be derived by adding the motion vector predictor of the current block and a motion vector difference value.

[460] In prediction mode where a merge mode and an advanced motion vector prediction mode are combined, the merge candidate and the motion vector prediction candidate may be derived from the same candidate block. For example, when the merge candidate is specified by merge_idx, a motion vector from a candidate block used to derive the specified merge candidate may be set as the motion vector predictor. Alternatively, when the motion vector prediction candidate is specified by mvp_flag, a merge candidate derived from a candidate block used to derive the specified merge candidate may be selected.

[461] Alternatively, the candidate block used to derive the merge candidate and the candidate block used to derive the motion vector prediction candidate may be different. For example, when a merge candidate derived from a candidate block positioned on top of the current block is selected, it may be configured to select a motion vector prediction candidate derived from a candidate block positioned on the left side of the current block.

[462] Alternatively, when the merge candidate selected by the index information and the motion vector prediction candidate selected by the index information are derived from the same candidate block, the motion vector prediction candidate may be replaced with a motion vector prediction candidate derived from a neighboring candidate block adjacent to the candidate block, or the merge candidate may be replaced with a merge candidate derived from a candidate block adjacent to the candidate block.

[463] Figure 35 is a view showing an example where a merge candidate is replaced.

[464] In the example shown in Figure 35(a), it is shown that a merge candidate and a motion vector prediction candidate derived from the candidate block at position A2 are selected. As shown in the figure, when a merge candidate and a motion vector prediction candidate are derived from the same candidate block, a merge candidate or a motion vector prediction candidate derived from a candidate block adjacent to the candidate block may be used instead of the merge candidate or the motion vector prediction candidate. For example, as shown in the example in Figure 35(b), a merge candidate at position A1 may be used instead of the merge candidate at position A2.

[465] A first prediction block may be derived based on the current block's merge candidate, and a second prediction block may be derived based on the motion vector prediction candidate. Thereafter, a combined prediction block may be derived via a weighted sum operation of the first prediction block and the second prediction block. At this point, the weighting value applied to the second prediction block generated using the advanced motion vector prediction mode may be set to be greater than the weighting value applied to the first prediction block generated using the merge mode.

[466] A derived residual image can be derived by subtracting a prediction video from an original video. At this point, when the residual video is shifted to the frequency domain, the subjective quality of the video is not significantly reduced, although the high-frequency components between the frequency components are removed. Therefore, when the values of the high-frequency components are converted to small or the values of the high-frequency components are set to 0, there is an effect of increasing compression efficiency without generating significant visual distortion. By reflecting this characteristic, the current block can be transformed to decompose a residual video into two-dimensional frequency components. The transform can be performed using a transform technique such as Discrete Cosine Transform (DCT) or Discrete Sine Transform (DST).

[467] DCT decomposes (or transforms) a residual video into two-dimensional frequency components using a cosine transform, and DST decomposes (or transforms) a residual video into two-dimensional frequency components using a sine transform. As a result of the residual video transform, the frequency components can be expressed as a basis video. For example, when the DCT transform is performed on a block of size N x N, as many basic pattern components as N2 can be acquired. The size of each of the basic pattern components included in a block of size N x N can be acquired through the transform. Depending on a transform technique used for this, the size of the basic pattern component can be referred to as a DCT coefficient or a DST coefficient.

[468] The DCT transform technique is mainly used to transform a video in which many non-zero low-frequency components are distributed. The DST transform technique is mainly used for videos in which many high-frequency components are distributed.

[469] The residual video can be transformed using a transform technique other than DCT or DST.

[470] In the following, the transform of a residual video into two-dimensional frequency components will be referred to as the two-dimensional video transform. In addition, the size of the basic pattern components acquired as a result of the transform is referred to as a transform coefficient. For example, the transform coefficient can mean a DCT coefficient or a DST coefficient. When the first transform and the second transform described below are applied, the transform coefficient can mean the size of a basic pattern component generated as a result of the second transform.

[471] The transform technique may be determined for each block. The transform technique may be determined based on at least one of the current block's prediction coding mode, the current block size, and the current block format. For example, when the current block is coded in intraprediction mode and the current block size is less than N x N, the transform may be performed using the DST transform technique. On the other hand, when the above conditions are not satisfied, the transform may be performed using the DCT transform technique.

[472] The two-dimensional video transform may not be performed for some blocks of the residual video. Not performing the two-dimensional video transform can be termed as a transform skip. When the transform skip is applied, quantization can be applied to the residual coefficients that were not transformed.

[473] After the current block is transformed using DCT or DST, the current transformed block can be transformed again. At this point, the DCT- or DST-based transform can be defined as a first transform, and transforming again a block to which the first transform is applied can be defined as a second transform.

[474] The first transform may be performed using any of a plurality of transform kernel candidates. For example, the first transform may be performed using any of DCT2, DCT8, or DCT7.

[475] Different transform kernels may be used for the horizontal direction and the vertical direction. Information indicating a combination of a transform kernel from the horizontal direction and a transform kernel from the vertical direction may be signaled via a bit stream.

[476] Units for performing the first transform and the second transform may be different. For example, the first transform may be performed on an 8 x 8 block, and the second transform may be performed on a sub-block of a size of 4 x 4 among the transformed 8 x 8 block. At this point, the transform coefficients of the residual regions that were not performed in the second transform may be set to 0.

[477] Alternatively, the first transform can be performed on a 4 x 4 block, and the second transform can be performed on a region of size 8 x 8 including the transformed 4 x 4 block.

[478] Information indicating whether the second transform has been performed or not can be signaled by means of a bit stream.

[479] Alternatively, whether or not to perform the second transform can be determined based on whether the horizontal direction transform kernel and the vertical direction transform kernel are the same. For example, the second transform can be performed only when the horizontal direction transform kernel and the vertical direction transform kernel are the same. Alternatively, the second transform can be performed only when the horizontal direction transform kernel and the vertical direction transform kernel are different from each other.

[480] Alternatively, the second transform may be allowed only when both the horizontal direction transform and the vertical direction transform use a predefined transform kernel. For example, when a DCT2 transform kernel is used for both the horizontal direction transform and the vertical direction transform, the second transform may be allowed.

[481] Alternatively, whether or not to perform the second transform may be determined based on the number of non-zero transform coefficients of the current block. For example, it may be configured to not use the second transform when the number of non-zero transform coefficients of the current block is less than or equal to a threshold value, and it may be configured to use the second transform when the number of non-zero transform coefficients of the current block is greater than the threshold value. It may be configured to use the second transform only when the current block is intraprediction encoded.

[482] Based on the current block shape, the size or shape of a sub-block on which the second transform will be performed can be determined.

[483] Figures 36 and 37 are views showing a sub-block in which a second transform will be performed.

[484] When the current block has a square shape, the second transform can be performed on a sub-block that has a size of N x N in the upper left part of the current block after the first transform is performed. For example, when the current block is a coding block of a size of 8 x 8, the second transform can be performed on a sub-block that has a size of 4 x 4 in the upper left part of the current block after the first transform is performed on the current block (see Figure 36).

[485] When the current block has a non-square shape of which the width is at least 4 times greater than the height, the second transform can be performed on a sub-block of a size (kN) x (4kN) in the upper left part of the current block after the first transform is performed. For example, when the current block has a non-square shape of a size of 16 x 4, the second transform can be performed on a sub-block of a size of 2 x 8 in the upper left part of the current block after the first transform is performed on the current block (see Figure 37(a)).

[486] When the current block has a non-square shape of which the height is at least 4 times greater than the width, the second transform can be performed on a sub-block of a size (4kN) x (kN) in the upper left part of the current block after the first transform is performed. For example, when the current block has a non-square shape of a size of 16 x 4, the second transform can be performed on a sub-block of a size of 2 x 8 in the upper left part of the current block after the first transform is performed on the current block (see Figure 37(b)).

[487] The decoder may perform inverse transform of the second transform (second inverse transform) and may perform inverse transform of the first transform (first inverse transform) on a result of the second inverse transform. As a result of performing the second inverse transform and the first inverse transform, residual signals for the current block may be acquired.

[488] Information indicating the transform type of the current block may be signaled via a bit stream. The information may be tu_mts_idx index information indicating one of a combination of a transform type for the horizontal direction and a transform type for the vertical direction.

[489] A transform kernel for the vertical direction and a transform kernel for the horizontal direction can be determined based on the transform type candidates specified by the tu_mts_idx index information. Tables 11 and 12 show transform type combinations according to tumtsidx. [TABLE 11] [TABLE 12]

[490] The transform type can be determined as any one of DCT2, DST7, DCT8, and jump transform. Alternatively, a transform type combination candidate can be configured using only transform colors other than the jump transform.

[491] In the case where Table 11 is used, when tu_mts_idx is 0, jump transform can be applied in the horizontal direction and in the vertical direction. When tu_mts_idx is 1, DCT2 can be applied in the horizontal direction and in the vertical direction. When tu_mts_idx is 3, DCT8 can be applied in the horizontal direction, and DCT7 can be applied in the vertical direction.

[492] In the case where Table 12 is used, when tu_mts_idx is 0, DCT2 can be applied in the horizontal direction and in the vertical direction. When tu_mts_idx is 1, jump transform can be applied in the horizontal direction and in the vertical direction. When tu_mts_idx is 3, DCT8 can be applied in the horizontal direction, and DCT7 can be applied in the vertical direction.

[493] Whether or not to encode index information may be determined based on at least one of the size, format, and number of non-zero coefficients of the current block. For example, when the number of non-zero coefficients is equal to or less than a threshold value, index information is not flagged, and a default transform type may be applied to the current block. Here, the default transform type may be DST7. Alternatively, the default mode may vary depending on the size, format, or intraprediction mode of the current block.

[494] The threshold value may be determined based on the current block size or format. For example, when the current block size is less than or equal to 32 x 32, the threshold value may be set to 2, and when the current block is greater than 32 x 32 (for example, when the current block is a coding block of a size of 32 x 64 or 64 x 32), the threshold value may be set to 4.

[495] A plurality of lookup tables may be previously stored in the encoder/decoder. In the plurality of lookup tables, at least one of the index value assigned to the transform type combination candidates, the type of the transform type combination candidates, and the number of transform type combination candidates may be different.

[496] A lookup table for the current block may be selected based on at least one of the current block size, the current block format, the current block prediction coding mode, the current block intraprediction mode, whether the second transform is applied, and whether transform hopping is applied to a neighboring block.

[497] For example, when the current block size is 4 x 4 or smaller or when the current block is interprediction encoded, the lookup table in Table 11 can be used, and when the current block size is larger than 4 x 4 or when the current block is intraprediction encoded, the lookup table in Table 12 can be used.

[498] Alternatively, information indicating any of the plurality of lookup tables may be signaled via a bit stream. The decoder may select a lookup table for the current block based on the information.

[499] As another example, the index assigned to the transform type matching candidate may be adaptively determined based on at least one of the current block size, the current block format, the current block prediction coding mode, the current block intraprediction mode, whether the second transform is applied, and whether transform jumping is applied to a neighboring block. For example, when the current block size is 4 x 4, the index assigned to the transform jumping may have a value less than the index assigned to the transform jumping when the current block size is greater than 4 x 4. Specifically, when the current block size is 4 x 4, index 0 may be assigned to the transform jumping, and when the current block is greater than 4 x 4 and less than or equal to 16 x 16, an index greater than 0 (e.g., index 1) may be assigned to the transform jumping. When the current block is larger than 16 x 16, a maximum value (e.g. 5) can be assigned to the transform jump index.

[500] Alternatively, when the current block is interprediction encoded, index 0 may be assigned to the transform jump. When the current block is intraprediction encoded, an index greater than 0 (e.g., index 1) may be assigned to the transform jump.

[501] Alternatively, when the current block is an interprediction-encoded block of size 4 x 4, index 0 may be assigned to the transform jump. Conversely, when the current block is not interprediction-encoded or the current block is larger than 4 x 4, an index of a value greater than 0 (e.g., index 1) may be assigned to the transform jump.

[502] Transform type combination candidates other than the transform type combination candidates listed in Tables 11 and 12 can be defined and used. For example, a transform type combination candidate can be used in which transform jump is applied to either of the horizontal direction transform and the vertical direction transform, and a transform kernel such as DCT7, DCT8, or DST2 is applied to the other. At this time, whether or not to use transform jump as a transform type candidate for the horizontal direction or for the vertical direction can be determined based on at least one of the size (e.g., width and/or height), the format, the prediction encoding mode, and the intraprediction mode of the current block.

[503] Alternatively, information indicating whether a particular transform type candidate is available may be signaled via a bit stream. For example, a flag indicating whether the transform jump can be used as a transform type candidate for the horizontal direction and the vertical direction may be signaled. Based on the flag, whether a particular transform type combination candidate is included in a plurality of transform type combination candidates may be determined.

[504] Alternatively, whether a specific transform type candidate is applied to the current block may be signaled via a bit stream. For example, flag cu_mts_flag indicating whether or not to apply DCT2 for the horizontal direction and for the vertical direction may be flagged. When the value of cu_mts_flag is 1, DCT2 may be set as a transform core for both the vertical direction and for the horizontal direction. When the value of cu_mts_flag is 0, DCT8 or DCT7 may be set as a transform core for both the vertical direction and for the horizontal direction. Alternatively, when the value of cu_mts_flag is 0, the tu_mts_idx information specifying any of the plurality of transform type combination candidates may be flagged.

[505] When the current block has a non-square shape of which the width is greater than the height or a non-square shape of which the height is greater than the width, the encoding of cu_mts_flag may be omitted, and the value of cu_mts_flag may be assumed to be 0.

[506] The number of available transform type combination candidates may be set differently according to the size, shape, or intraprediction mode of the current block. For example, when the current block has a square shape, three or more transform type combination candidates may be used, and when the current block has a non-square shape, two transform type combination candidates may be used. Alternatively, when the current block has a square shape, only transform type combination candidates of which the transform type for the horizontal direction is different from the transform type for the vertical direction may be used among the transform type combination candidates.

[507] When there are three or more transform type combination candidates that the current block can use, index information tu_mts_idx indicating one among the transform type combination candidates may be flagged. Conversely, when there are two transform type combination candidates that the current block can use, flag mts_flag indicating any one among the transform type combination candidates may be flagged. Table 13 shows a process of encoding information for specifying transform type combination candidates according to the current block format. [TABLE 13]

[508] Indices of transform type combination candidates may be rearranged (or reordered) according to the shape of the current block. For example, the indices assigned to transform type combination candidates when the current block has a square shape may be different from the indices assigned to transform type combination candidates when the current block has a non-square shape. For example, when the current block has a square shape, a transform type combination may be selected based on Table 14 shown below, and when the current block has a non-square shape, a transform type combination may be selected based on Table 15 shown below. [TABLE 14] [TABLE 15]

[509] A transform type can be determined based on the number of horizontal direction nonzero coefficients or the number of vertical direction nonzero coefficients of the current block. Here, the number of horizontal direction nonzero coefficients indicates the number of nonzero coefficients included in a 1 x N block (where N is the width of the current block), and the number of vertical direction nonzero coefficients indicates the number of nonzero coefficients included in an N x 1 block (where N is the height of the current block). When the maximum value of the horizontal direction nonzero coefficient is less than or equal to a threshold value, a first transform type can be applied in the horizontal direction, and when the maximum value of the horizontal direction nonzero coefficient is greater than the threshold value, a second transform type can be applied in the horizontal direction. When the maximum value of the vertical direction non-zero coefficient is less than or equal to a threshold value, the first type of transform can be applied in the vertical direction, and when the maximum value of the vertical direction non-zero coefficient is greater than the threshold value, the second type of transform can be applied in the vertical direction.

[510] Figure 38 is a view to describe examples of determining a transform type of the current block.

[511] For example, when the current block is intraprediction encoded and the maximum value of the horizontal direction nonzero coefficient of the current block is 2 or less (see Figure 38(a)), the transform type of the horizontal direction can be determined as DST7.

[512] When the current block is intraprediction coded and the maximum value of the vertical direction nonzero coefficient of the current block is greater than 2 (see Figure 38(b)), DCT2 or DCT8 can be determined as the transform type of the vertical direction.

[513] Information indicating whether or not to explicitly determine the transform type of the current block based on the flagged information of a bit stream may be flagged via a bit stream. For example, the sps_explicit_intra_mts_flag information indicating whether explicit transform type determination is allowed for an intraprediction-encoded block and/or the sps_explicit_inter_mts_flag information indicating whether explicit transform type determination is allowed for an interprediction-encoded block may be flagged at a sequence level.

[514] When explicit transform type determination is allowed, a transform type of the current block may be determined based on the signed tu_mts_idx index information of a bitstream. Conversely, when explicit transform type determination is not allowed, the transform type may be determined based on at least one of the current block size, the current block format, whether transform of a sub-block unit is allowed, and the position of a sub-block including a non-zero transform coefficient. For example, the horizontal direction transform type of the current block may be determined based on the current block width, and the vertical direction transform type of the current block may be determined based on the current block height. For example, when the current block width is less than 4 or greater than 16, the horizontal direction transform type may be determined as DCT2. Otherwise, the horizontal direction transform type may be determined as DST7. When the height of the current block is less than 4 or greater than 16, the vertical transform type can be determined as DCT2. Otherwise, the vertical transform type can be determined as DST7. Here, in order to determine the horizontal transform type and the vertical transform type, threshold values compared to the width and height can be determined based on at least one of the size, shape, and intraprediction mode of the current block.

[515] Alternatively, when the current block has a square shape of which the height and width are equal, the horizontal direction transform type and the vertical direction transform type are set to be the same, while when the current block has a non-square shape of which the height and width are different, the horizontal direction transform type and the vertical direction transform type can be set to be different from each other. For example, when the width of the current block is greater than the height, the horizontal direction transform type can be determined as DST7, and the vertical direction transform type can be determined as DCT2. When the height of the current block is greater than the width, the vertical direction transform type can be determined as DST7, and the horizontal direction transform type can be determined as DCT2.

[516] The number and/or type of transform type candidates or the number and/or type of combination transform type candidates may vary according to whether explicit transform type determination is allowed. For example, when explicit transform type determination is allowed, DCT2, DST7, and DCT8 may be used as transform type candidates. Consequently, each of the horizontal direction transform type and the vertical direction transform type may be set to DCT2, DST8, or DCT8. When explicit transform type determination is not allowed, only DCT2 and DST7 may be used as transform type candidates. Consequently, each of the horizontal direction transform type and the vertical direction transform type may be determined to be DCT2 or DST7.

[517] When the encoder performs transform and quantization, the decoder can acquire a residual block through inverse quantization and inverse transform. The decoder can acquire a reconstructed block for the current block by adding a prediction block and the residual block.

[518] When a reconstructed block of the current block is acquired, information loss occurring in the quantization and coding process can be reduced through in-loop filtering. An in-loop filter may include at least one of a deblocking filter, an adaptive sample shift (SAO) filter, and an adaptive loop filter (ALF). Then, a block reconstructed before the in-loop filter is applied is referred to as a first reconstructed block, and a block reconstructed after the in-loop filter is applied is referred to as a second reconstructed block.

[519] The second reconstructed block can be acquired by applying at least one of the deblocking filter, the SAO, and the ALF to the first reconstructed block. At this point, either the SAO or the ALF can be applied after the deblocking filter is applied.

[520] The deblocking filter is to mitigate video quality degradation (blocking artifact) at the boundary of a block, which occurs as quantization is performed on individual blocks. In order to apply the deblocking filter, a block resistance (BS) between the first reconstructed block and a neighboring reconstructed block can be determined.

[521] Figure 39 is a flowchart illustrating a block strength determination process.

[522] In the example shown in Figure 39, P denotes a first reconstructed block, and Q denotes a neighboring reconstructed block. Here, the neighboring reconstructed block can be a block adjacent to the left side or the top side of the current block.

[523] In the example shown in Figure 39, it is shown that the block resistance is determined by considering the P- and Q-prediction coding modes, whether a non-zero transform coefficient is included, whether interprediction is performed using the same reference image, and whether the difference value of motion vectors is greater than or equal to a threshold value.

[524] Whether or not to apply the deblocking filter can be determined based on the block resistance. For example, when the block resistance is 0, filtering cannot be performed.

[525] SAO is intended to mitigate a ringing artifact that occurs when quantization is performed in the frequency domain. SAO can be performed by adding or subtracting an offset determined by considering the pattern of the first reconstructed video. The method for determining the offset includes an edge offset (EO) or a band offset (BO). EO denotes a method of determining an offset of a current sample based on the pattern of surrounding pixels. BO denotes a method of applying a common offset to a set of pixels with similar brightness values within a region. Specifically, the pixel brightness can be divided into 32 equal sections, and the pixels with similar brightness values can be defined as a group. For example, four adjacent bands among 32 bands can be defined as a group, and the same offset value can be applied to the samples belonging to all four bands.

[526] ALF is a method of generating a second reconstructed video by applying a filter that has a predefined size/shape to the first reconstructed video or a reconstructed video to which a deblocking filter has been applied. Equation 16 shows an example application of ALF. [EQUATION 16]

[527] Any of the predefined filter candidates can be selected by an image unit, a coding tree unit, a coding block, a prediction block, or a transform block. Each of the filter candidates can be different in size or shape.

[528] Figure 40 is a view showing predefined filter candidates.

[529] As shown in the example of Figure 40, at least one of the diamond shapes of 5 x 5, 7 x 7, and 9 x 9 can be selected.

[530] Only a diamond shape of a size of 5 x 5 can be used for a chroma component.

[531] Applying the above-described embodiments focusing on a decoding process or an encoding process to an encoding process or a decoding process is included within the scope of the present invention. Changing the described embodiments in a predetermined order into an order different from the described order is also included within the scope of the present invention.

[532] Although the above embodiments have been described based on a series of steps or flowcharts, this does not limit the order of the time series of the present invention, and they may be performed simultaneously or in a different order as needed. Furthermore, each of the components (e.g., units, modules, etc.) that constitute the block diagram in the embodiments described above may be implemented as a hardware or software device, or a plurality of components may be combined to be implemented as a single hardware or software device. The embodiments described above may be implemented in the form of program commands that can be executed by multiple computer components and recorded on a computer-readable recording medium. The computer-readable recording medium may include program commands, data files, data structures, and the like, independently or in combination. Computer-readable recording media include, for example, magnetic media such as hard disks, floppy disks, and magnetic tape, optical recording media such as CD-ROMs and DVDs, magneto-optical media such as floppy disks, and hardware devices specifically configured to store and execute program commands, such as ROM, RAM, flash memory, and the like. The hardware devices described above may be configured to operate using one or more software modules to perform the method of the present invention, and vice versa.

[533] The present invention can be applied to an electronic device that encodes and decodes a video.

Claims (8)

1. A VIDEO DECODING METHOD comprising the steps of: determining whether to apply a combined prediction mode to a current block; obtaining a first prediction block and a second prediction block for the current block when the combined prediction mode is applied to the current block, wherein the first prediction block is obtained based on the motion information of a merge candidate of the current block, and the second prediction block is obtained based on an intra-prediction mode of the current block; and obtaining a third prediction block for the current block based on a weighted sum operation of the first prediction block and the second prediction block, wherein when at least one of a width and a height of the current block is greater than a threshold value, the combined prediction mode is not applied to the current block. 2. METHOD, according to claim 1, characterized in that the intraprediction mode of the current block is defined as a flat mode. 3. METHOD, according to claim 2, characterized in that the second prediction block is obtained based on a reference sample line included in an adjacent reference sample line. 4. The method of claim 1, wherein when the combined prediction mode is applied to the current block, triangular partitioning is disabled for the current block. 5. A VIDEO CODING METHOD comprising the steps of: determining whether to apply a combined prediction mode to a current block; obtaining a first prediction block and a second prediction block for the current block when the combined prediction mode is applied to the current block, wherein the first prediction block is obtained based on the motion information of a merge candidate of the current block, and the second prediction block is obtained based on an intra-prediction mode of the current block; and obtaining a third prediction block for the current block based on a weighted sum operation of the first prediction block and the second prediction block, wherein when at least one of a width and a height of the current block is greater than a threshold value, the combined prediction mode is not applied to the current block. 6. METHOD, according to claim 5, characterized in that the intraprediction mode of the current block is defined as a flat mode. 7. METHOD, according to claim 6, characterized in that the second prediction block is obtained based on a reference sample line included in an adjacent reference sample line. 8. The method of claim 5, wherein when the combined prediction mode is applied to the current block, triangular partitioning is disabled for the current block.