CN101860754B - Method and device for encoding and decoding motion vectors - Google Patents

Abstract

The present invention relates to a video encoding method. In particular, it relates to Motion Vector (MV) coding, wherein MV coding is based on edge matching or adaptive template matching, i.e. motion vector coding based on boundary matching (MVCBM) and motion vector coding based on adaptive template matching (mvcmta). In general, the present disclosure defines a prediction candidate set, where the predictor adaptively changes based on the current distribution of neighboring MVs. Subsequently, matching techniques (boundary matching in MVCBM and adaptive template matching in mvcmat) are used to reduce the size of the prediction candidate set to reduce the number of bits used for indexing. Then, the optimal motion vector predictor is selected among the predictors included in the reduced set. The guessing strategy based on the minimum MVD standard further saves the number of index bits.

Description

Method and device for encoding and decoding motion vectors

technical field

The present invention relates generally to digital video processing. In particular, the present invention relates to methods and apparatus including encoding and decoding motion vector encoding.

Background technique

Motion estimation (ME for short) and motion compensation (MC for short) are modern video coding standards (such as MPEG-1, MPEG-2, MPEG-4, H.261, H.263, H.264 and AVS ) to suppress time redundancy and achieve high coding efficiency. In ME/MC, the current frame is divided into non-overlapping blocks. For each current block in the current frame, a search area is defined in the reference frame. Each point in the search area corresponds to a candidate block that is potentially a good predictor of the current block. A similarity measure is defined and a search area is searched to find, based on the similarity measure, a candidate block that has a high, if not maximum, similarity to the current block. The candidate block with the greatest similarity will be the best matching block. The relative displacement between the best matching block and the current block is called a motion vector, which needs to be coded. Coding of motion vector (MV for short) is also called motion vector coding (MVC for short).

On the other hand, in order to implement motion compensation at the decoding end, the encoding end needs to consume a large amount of bit streams for transmitting motion vector information. For example, in H.264, when the bit rate is very low (for example, QP=40), the number of bits consumed by the motion vector information accounts for a high percentage of the total coding number, even up to 50%. Therefore, there is an urgent need in this field for a more effective method for motion vector coding to improve its coding efficiency.

In the latest motion vector coding methods, many tools are introduced to improve the efficiency of motion vector coding. For example, in order to reduce the number of compressed bits for motion vectors, the H.264 standard encodes motion vectors using a simple predictive coding technique. For each current block, H.264 uses one of the three adjacent motion vectors (MV) to form a motion vector predictor (MVP for short). The motion vector difference (motion vector difference, MVD for short) between the MVP and the real MV of the current block is encoded into the bitstream. This "intermediate MVP" effectively reduces the bits used for MV encoding, since it is similar to the real MV of the current block in most cases. However, the intermediate MVP does not always optimally predict the current MV. More efficient MVPs are needed to further reduce the number of coding bits used for MVs.

There have been other attempts to achieve efficient motion vector coding. For example, it is documented in the following public document: "Decoder side motion vector derivation for inter frame video coding" by S. Kamp, M. Evertz and M. Wien, IEEE International Conference on Image Processing, pp .1120-1123, 2008, (hereinafter referred to as "S. Kamp et al. article") and US Patent No. 7,023,919 entitled "Method for coding motionvector using 2-dimentional minimum bitrate predicting technique".

In the article by S. Kamp et al., a method for obtaining motion vectors is described. This method performs motion estimation-like methods at the decoder to obtain motion vectors and reference indices. However, this method cannot adaptively change the MV predictor candidate set, and this method abandons the use of the existing accurate ME method to obtain the MV but directly obtains the MV with a not very accurate template matching method at the decoding end, and the method No guessing strategy is used in . Furthermore, template matching uses templates of fixed size and shape. No reference index or MVD is used in this method. In general, the method is quite complicated, and MV and reference index using template matching are not always reliable by themselves.

In US Patent Application No. 7023919, a method of motion vector coding is described. This method does not use matching techniques to reduce the MV predictor candidate set, nor does it provide an adaptive candidate set. Even if a guessing strategy is used, it is based on the minimum bit rate and not the minimum MVD. No temporal-based MV predictors are used as candidates in this approach. This method lacks efficiency, especially if the MVD is small. Also different MVDs may have the same bit rate, making a minimum bit rate based criterion inefficient.

Still other methods have been tried using boundary matching. For example in the following public documents: "Video Error Concealment Using Spatio-Temporal Boundary Matching and Partial Differential Equation" by Yan Chen, Yang Hu, Oscar C.Au, Houqiang Li, Chang Wen Chen, IEEE TRANSACTIONS ON MULTIMEDIA, VOL. 10, NO.1, January 2008 (hereinafter referred to as "Yan Chen et al. article"), and US Patent No. 5,596,370 entitled "Boundary matching motion estimation apparatus".

In Yan Chen et al., a system and method for concealing errors in a video signal is provided. Boundary matching is limited to error concealment. The disclosed method does not point out that using boundary matching selection can improve the accuracy of the MV predictor to reduce the MVD, nor does it utilize the existing MV obtained from the precise motion estimation at the encoder side. Also, no template matching is used in this method. The disadvantage of this method is that using boundary matching to obtain MV itself is not always reliable.

In US Patent No. 5,596,370, an apparatus for motion estimation is provided. Boundary matching is limited to obtaining motion vectors, and the purpose of this method is to provide an alternative to motion estimation. This method does not use the MV obtained in the motion estimation at the encoder, and the disadvantage is that obtaining the MV itself using boundary matching instead of motion estimation is not always reliable. Boundary matching is only used to obtain motion vectors, and this method does not give the information of motion vector encoding. This approach is not efficient since there is no adaptive candidate set, and the method does not consider other matching techniques such as adaptive template matching.

Therefore, it is necessary to provide a motion vector encoding system and method to solve one or more of the above problems.

2003以及Alan Bovik的The Essential Guide to Video Processing，Academic Press

2009。这些公开的文件都作为参考被引入其中。For other background technologies related to video coding, please refer to the document: Iain Richardson's H.264and M-PEG-4VideoCompression, Wiley&Sons,

2003 and The Essential Guide to Video Processing by Alan Bovik, Academic Press

2009. These published documents are hereby incorporated by reference.

Contents of the invention

In view of the above, the present invention needs a method and device for encoding and decoding motion vectors to overcome the problems of low efficiency and poor reliability of the methods in the above technologies.

Specifically, the present invention provides a motion vector encoding method, comprising the following steps:

Step 1: defining a candidate set of predictors based on adaptive changes in motion vectors adjacent to the motion vector;

Step 2: using a matching method to reduce the number of predictors in the candidate set to reduce the number of bits used by the index;

Step 3: Select the best predictor among the remaining predictors in the candidate set based on the minimum motion vector difference criterion, and obtain the current motion vector difference;

Step 4: Based on the difference between the residual predictor and the current motion vector, use a guessing strategy to detect the motion vector predictor to further save the number of bits used to encode the index;

Step 5: Judging whether the correct motion vector predictor can be detected, if the correct motion vector predictor is detected, encode at least the current motion vector difference, if the correct motion vector predictor cannot be detected, encode at least the current motion vector The difference and index are encoded.

Wherein, the matching method is a boundary matching method or an adaptive template matching method.

Wherein, the boundary matching method is to paste a part of data corresponding to the predictors in the candidate set into the current data space to obtain a boundary matching error, and remove one or more predictors with the largest boundary matching error in the candidate set.

Wherein, the adaptive template matching method is to determine the shape of the adaptive template for the current data space based on the possible correlation between the adaptive template and the current data; or determine the shape of the adaptive template for the current data space based on the consistency in the adaptive template. Dimensions of the adaptive template for the data space.

Wherein, the adaptive template matching method further includes calculating the difference between the template in the current data space and the template in the shifted data space in the reference picture by using the displacement corresponding to the predictor in the candidate set.

Wherein, in the adaptive template matching method, the width and shape of the adaptive template have been specified.

Wherein, after the step 5, if the correct motion vector predictor is detected, at least the encoded current motion vector difference is sent; if the correct motion vector predictor cannot be detected, at least the encoded current motion vector difference is sent and encoded bits as indices.

Wherein, the candidate set of predictors includes several proportional spatial and temporal predictors.

Wherein, in the step 1, the motion vector adjacent to the motion vector is a motion vector adjacent to the motion vector in time, space, time and space, or based on an analysis strategy.

Wherein, in the guessing strategy in step 4, the number of the remaining predictors is at least 1, and an intermediate parameter is generated for each remaining predictor and the current motion vector difference, and each intermediate parameter is regenerated based on the minimum motion vector difference criterion Select the predictor to be selected, if the reselected predictor to be selected is the same as the remaining predictor that generates the intermediate parameter, then determine that the remaining predictor is an optional predictor, if the number of optional predictors is 1, Then the alternative predictor is the correct motion vector predictor.

The present invention also provides a method of decoding encoded motion vectors encoded by the method of claim 1, comprising the steps of:

Step 1: Decoding the motion vector difference;

Step 2: defining a candidate set of predictors based on adaptive changes in motion vectors adjacent to the motion vector;

Step 3: using a matching method to reduce the number of predictors in the candidate set;

Step 4: Based on the remaining predictors in the candidate set and the decoded motion vector difference, use a guessing strategy to determine whether the motion vector predictor can be determined;

Step 5: If the motion vector predictor can be determined, use the decoded motion vector difference and the determined motion vector predictor to obtain the motion vector; if the motion vector predictor cannot be determined, decode the index and determine the motion vector predictor, use the decoded The motion vector difference and the determined motion vector predictor get the motion vector.

The matching method is the same as that used in encoding.

The present invention also provides a motion vector encoding device, which includes the following parts:

The first selection module is used to select the predictors that form the candidate set;

The first processing module receives the candidate set of predictors of the first selection module, and uses a matching method to reduce the number of predictors in the candidate set;

The second selection module is connected to the first processing module, receives the remaining predictors in the candidate set, and selects the optimal predictor among the remaining predictors based on the minimum motion vector difference criterion;

The second processing module is connected to the second selection module, and obtains the current motion vector difference according to the selected optimal predictor;

A detection module, connected to the second processing module and the first processing module, detects a motion vector predictor using a guessing strategy based on the received remaining predictors in the candidate set and the current motion vector difference;

A judgment module, connected to the detection module, if the detection result of the detection module is that a correct motion vector predictor is detected, the judgment result is sent to the encoding module;

The encoding module receives the determination result, and encodes at least the current motion vector difference or encodes at least the current motion vector difference and an index.

The present invention also provides a motion vector decoding device, which includes the following parts:

a receiving module, receiving at least the encoded motion vector difference;

The first decoding module is connected to the receiving module, and decodes the motion vector difference according to the received at least coded motion vector difference;

The first selection module is used to select the predictors that form the candidate set;

The first processing module receives the candidate set of predictors of the first selection module, and uses a matching method to reduce the number of predictors in the candidate set;

A judging module, connected to the first processing module and the first decoding module, based on the remaining predictors in the candidate set and the decoded motion vector difference, using a guessing strategy to judge whether the motion vector predictor can be determined;

The second decoding module is connected to the judgment module, and decodes the index according to the judgment result;

An analysis module, connected to the first processing module and the second decoding module, determines the motion vector predictor according to the decoded index;

The second processing module is connected to the judgment module, the first decoding module and the analysis module, and according to the judgment result, obtains the motion vector by using the decoded motion vector difference and the decoded and determined motion vector predictor; or by using the decoded The motion vector difference and the determined motion vector predictor result in the motion vector.

The motion vector encoding and decoding method provided by the invention can rapidly, reliably and effectively encode and decode the motion vector, thereby improving the efficiency of the system. And the present invention can be used in any video coding system that employs motion estimation and needs to send motion information to the decoder, such as but not limited to H.264, KTA, AVS, MPEG. Furthermore, the invention can be applied to any coder using hybrid predictive coding.

Other aspects of the present invention are also disclosed through the examples described below.

Description of drawings

These and other objects, aspects and embodiments of the present invention will be described in detail below with reference to the accompanying drawings, in which:

Fig. 1 shows an exemplary representation of a current block and its neighboring blocks according to an embodiment of the invention.

Fig. 2 shows inner boundary pixels and outer boundary pixels of a block according to an embodiment of the invention.

Fig. 3 shows an exemplary template for template matching according to an embodiment of the present invention.

FIG. 4 shows an exemplary template with a template width equal to 4, according to an embodiment of the present invention.

Fig. 5 shows a flowchart of an MVCBM according to an embodiment of the present invention.

FIG. 6 shows a flowchart of MVCATM according to an embodiment of the present invention.

FIG. 7 shows a block diagram of an embodiment of a video encoding system according to an embodiment of the present invention.

FIG. 8 shows a block diagram of an embodiment of a video decoding system according to an embodiment of the present invention.

Fig. 9 shows a schematic diagram of a motion vector encoding device according to an embodiment of the present invention.

Fig. 10 shows a schematic diagram of a motion vector decoding device according to an embodiment of the present invention.

Detailed ways

The present invention relates to video encoding methods. In particular, it relates to motion vector (MV) coding, where MV coding is based on boundary matching or adaptive template matching, namely motion vector coding based on boundary matching (MVCBM) and motion vector coding based on adaptive template matching (MVCATM). In general, the present invention defines a candidate set of predictors, where the predictors vary adaptively based on the current distribution of neighboring MVs. Subsequently, matching techniques (boundary matching in MVCBM and adaptive template matching in MVCATM) are used to reduce the size of the predictor candidate set to reduce the number of coding bits used by the index. Then, the optimal motion vector predictor is selected among the predictors included in the reduced set. The guessing strategy based on the minimum MVD criterion further saves the number of bits in the index.

Specifically, the set of possible MVP candidates includes several proportional spatial and temporal predictors. To increase the diversity of the predictors, the spatial predictors are adaptively changed based on the characteristics of neighboring motion vectors. To select a good predictor from the candidate set (CS), matching techniques such as but not limited to boundary matching (BM) and adaptive template matching (ATM) are used. The optimal MVP of the current block is selected from the predictors selected by the matching technique to minimize the MVD. Since the optimal MVP is selected based on the minimum MVD criterion, a guessing strategy is introduced, which can save the number of bits for signaling the MVP index to the decoder in some cases. The present invention can significantly reduce the bit rate relative to the H.264 standard.

A. Scaled Motion Vector Predictor

The motion vectors of neighboring blocks are used as predictors for the motion vector of the current block. The motion vectors of neighboring blocks may correspond to different reference frames in an embodiment which allows multiple reference frames eg in H.264. When motion vectors of two adjacent blocks correspond to different reference frames, their temporal distances are different relative to the current frame. Thus in embodiments using multiple reference frames, the motion vectors of neighboring blocks are scaled proportionally according to their temporal distance before being used to predict the motion vector of the current block. Figure 1 shows an exemplary representation of a current block and its neighboring blocks. Taking mv _A (the MV of block A on the left of the current block) as an example, it is assumed that the temporal distance between block A and its reference block is d _p , and the temporal distance between the current block and its reference block is d _c . The proportional predictor mv _SA is computed by

$\mathrm{<span\; class="notranslate">\; m</span>}{\mathrm{<span\; class="notranslate">\; v</span>}}_{\mathrm{<span\; class="notranslate">\; SA</span>}}<span\; class="notranslate">\; =</span>{\mathrm{<span\; class="notranslate">\; mv</span>}}_{\mathrm{<span\; class="notranslate">\; A</span>}}<span\; class="notranslate">\; \&times;</span>\frac{{\mathrm{<span\; class="notranslate">\; d</span>}}_{\mathrm{<span\; class="notranslate">\; c</span>}}}{{\mathrm{<span\; class="notranslate">\; d</span>}}_{\mathrm{<span\; class="notranslate">\; p</span>}}}<span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; 1</span>}<span\; class="notranslate">\; )</span>$

mv _SB of block B, mv _SC of block C, etc. can be calculated using the same method. Therefore, each predictor is scaled before use.

B. MVP candidate set

An aspect of the present invention is to provide predictors and encoding methods for parameter generation and representation in data compression, such as encoding motion vectors in video compression. Two or more candidates are adaptively selected according to adjacent data information. "Adjacency" data information includes but not limited to spatial adjacency, time adjacency, adjacency combination and some functions of adjacency based on analysis strategies. Analysis strategies include, but are not limited to, current correlation, distribution, such as selecting neighboring motion vectors that are dissimilar to other neighboring vectors as candidates for encoding the current motion vector in video compression.

The set of possible MVP candidates consists of several proportional spatial and temporal predictors. To increase the diversity of the predictors, the spatial predictors are adaptively changed based on the characteristics of neighboring motion vectors.

In one embodiment, in order for the MVP to achieve high accuracy, the MVP candidate set includes temporal and spatial MVPs. The MVP candidate set includes a plurality of such MVPs, the number of which varies according to the situation, such as but not limited to 5 MVPs or 8 MVPs.

In a further embodiment, for a given candidate set, the included predictors vary adaptively based on the current distribution of MVPs. For example, without loss of generality, the size of the candidate set is chosen to be 3, and the candidate set with size equal to 3 includes mv _Scol , mv _SH.264 and mv _Snei . mv _Scol is the co-located MV scaled according to formula (1) (referring to the motion vector of the block having the same location as the current block in the previous frame). It is a temporal MVP candidate. mv _SH.264 and mv _Snei are space MVP candidates.

mv _SH.264 is the middle one of 3 adjacent motion vectors (these 3 adjacent motion vectors are scaled according to formula (1)), for example it can be from the middle of mv _SA , mv _SB and mv _SC one gets.

mv _Snei is one of three proportional neighbor motion vectors mv _SA , mv _SB and mv _SC which is farthest from mv _SH.264 :

${\mathrm{<span\; class="notranslate">\; mv</span>}}_{\mathrm{<span\; class="notranslate">\; Snei</span>}}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; arg</span>}\underset{{\mathrm{<span\; class="notranslate">\; mv</span>}}_{\mathrm{<span\; class="notranslate">\; x</span>}}}{\mathrm{<span\; class="notranslate">\; max</span>}}{<span\; class="notranslate">\; |</span><span\; class="notranslate">\; |</span>{\mathrm{<span\; class="notranslate">\; mv</span>}}_{\mathrm{<span\; class="notranslate">\; x</span>}}<span\; class="notranslate">\; -</span>{\mathrm{<span\; class="notranslate">\; mv</span>}}_{\mathrm{<span\; class="notranslate">\; SH</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; 264</span>}}<span\; class="notranslate">\; |</span><span\; class="notranslate">\; |</span>}^{\mathrm{<span\; class="notranslate">\; 2</span>}}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; x</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; SA</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; SB</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; SC</span>}<span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; 2</span>}<span\; class="notranslate">\; )</span>$

The reason for setting this predictor mv _Snei is to try to get a more efficient MVP when only one of the neighboring blocks A, B and C belongs to the object containing the current block. In this case, the correlation between the intermediate MVP and the actual MV of the current block may be low.

Therefore, the MVP candidate set (Candidate Set, referred to as CS) is composed of three MVPs under this specific exemplary condition, wherein the size of the candidate set under this condition is equal to 3 (size-3CS):

CS={mv _SH.264 , mv _Scol , mv _Snei } (3)

C.CS reduce and optimize MVP

The initial set of MVP candidates will contain many MVPs, and the index of the selected one needs many bits to specify. In order to reduce the bits used for the index of the optimal MVP, the CS is reduced to include a smaller number of predictors. In one embodiment, a method called boundary matching is used. In another embodiment, a method called adaptive template matching is used. Based on the matching status, the size of the Reduced Candidate Set (Reduced Candidate Set, RCS for short) is fixed or variable. For example, the best predictor is selected, and if the following predictors are almost good, the following good predictors are also selected. Or we can just choose the best predictor if it is much better than the others. In this particular case of size-3CS with mv _Scol , mv _SH.264 and mv _Snei , we can use a fixed-size RCS of size 2.

D. Boundary Matching

One aspect of the present invention is to: determine the adaptive candidate set; reduce the candidate set by matching techniques to save the number of bits of the index; implement the guessing strategy based on the minimum MVD and the entire MV encoding process; use boundary matching as a matching technique in MV encoding .

A further aspect of the present invention also includes: performing boundary matching, such as copying a part of data corresponding to each candidate into the current data space, and measuring the coordination and continuity between the adjacent data in the current data space and the pasted data part sex.

Boundary matching (BM) has been explained in the following articles: "Video Error Concealment Using Spatio-Temporal Boundary Matching and Partial Differential Equation" by Yan Chen, Yang Hu, Oscar C.Au, Houqiang Li, Chang Wen Chen, From IEEE Journal of Multimedia (IEEE TRANSACTIONS ONMULTIMEDIA), VOL.10, NO.1, January 2008, introduce its disclosure content here.

Boundary matching is widely used in error concealment to find the motion vector pointing to the most probable block in the reference frame to restore the lost block. BM estimation searches for a predictor based on the spatial continuity between a block and its neighbors. If an incorrect predictor is used, there is a high probability that the corresponding reconstructed block does not have high spatial continuity with its neighbors. But if the correct predictor is used, the reconstructed block should be contiguous with neighboring blocks.

The matching criterion is boundary matching error (BME for short), which measures the spatial discontinuity between the inner boundary pixels of the candidate block and the outer boundary pixels of the current block, as shown in Fig. 2. Fig. 2 shows inner boundary pixels and outer boundary pixels of a block. The letters N, W, E, S represent the directions North, West, East and South. BME has two parts: smooth discontinuity distortion (SDD for short) and edge discontinuity distortion (EDD for short):

BME＝α×SDD+(1-α)×EDD (4)

The weight factor α is a real number between 0 and 1.

Note that in this instantiated MV encoding method BM is performed at both the encoder and the decoder to find a suitable predictor. Since the east and south sides of the current block are usually not yet decoded at the decoder, only readily available boundary information is used, such as that of the north and/or west boundaries. SDD and EDD are calculated as follows:

$\mathrm{<span\; class="notranslate">\; SDD</span>}<span\; class="notranslate">\; =</span>\underset{\mathrm{<span\; class="notranslate">\; k</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 0</span>}}{\overset{{\mathrm{<span\; class="notranslate">\; N</span>}}_{\mathrm{<span\; class="notranslate">\; x</span>}}<span\; class="notranslate">\; -</span>\mathrm{<span\; class="notranslate">\; 1</span>}}{\mathrm{<span\; class="notranslate">\; \&Sigma;</span>}}}<span\; class="notranslate">\; |</span>{\mathrm{<span\; class="notranslate">\; p</span>}}_{\mathrm{<span\; class="notranslate">\; rec</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; x</span>}<span\; class="notranslate">\; +</span>\mathrm{<span\; class="notranslate">\; k</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; the\; y</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; -</span>{\mathrm{<span\; class="notranslate">\; p</span>}}_{\mathrm{<span\; class="notranslate">\; rec</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; x</span>}<span\; class="notranslate">\; +</span>\mathrm{<span\; class="notranslate">\; k</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; the\; y</span>}<span\; class="notranslate">\; -</span>\mathrm{<span\; class="notranslate">\; 1</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; |</span><span\; class="notranslate">\; +</span>\underset{\mathrm{<span\; class="notranslate">\; k</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 0</span>}}{\overset{{\mathrm{<span\; class="notranslate">\; N</span>}}_{\mathrm{<span\; class="notranslate">\; W</span>}}<span\; class="notranslate">\; -</span>\mathrm{<span\; class="notranslate">\; 1</span>}}{\mathrm{<span\; class="notranslate">\; \&Sigma;</span>}}}<span\; class="notranslate">\; |</span>{\mathrm{<span\; class="notranslate">\; p</span>}}_{\mathrm{<span\; class="notranslate">\; rec</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; x</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; the\; y</span>}<span\; class="notranslate">\; +</span>\mathrm{<span\; class="notranslate">\; k</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; -</span>{\mathrm{<span\; class="notranslate">\; p</span>}}_{\mathrm{<span\; class="notranslate">\; rec</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; x</span>}<span\; class="notranslate">\; -</span>\mathrm{<span\; class="notranslate">\; 1</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; the\; y</span>}<span\; class="notranslate">\; +</span>\mathrm{<span\; class="notranslate">\; k</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; |</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; 5</span>}<span\; class="notranslate">\; )</span>$

$\mathrm{<span\; class="notranslate">\; EDD</span>}<span\; class="notranslate">\; =</span>\underset{\mathrm{<span\; class="notranslate">\; k</span>}<span\; class="notranslate">\; -</span>\mathrm{<span\; class="notranslate">\; 0</span>}}{\overset{{\mathrm{<span\; class="notranslate">\; N</span>}}_{\mathrm{<span\; class="notranslate">\; w</span>}}<span\; class="notranslate">\; -</span>\mathrm{<span\; class="notranslate">\; 1</span>}}{\mathrm{<span\; class="notranslate">\; \&Sigma;</span>}}}<span\; class="notranslate">\; |</span>\frac{<span\; class="notranslate">\; \&dtri;</span><span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; \&Delta;p</span>}}_{\mathrm{<span\; class="notranslate">\; rec</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; x</span>}<span\; class="notranslate">\; +</span>\mathrm{<span\; class="notranslate">\; k</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; the\; y</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; )</span>}{<span\; class="notranslate">\; |</span><span\; class="notranslate">\; \&dtri;</span><span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; \&Delta;</span>}{\mathrm{<span\; class="notranslate">\; p</span>}}_{\mathrm{<span\; class="notranslate">\; rec</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; x</span>}<span\; class="notranslate">\; +</span>\mathrm{<span\; class="notranslate">\; k</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; the\; y</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; )</span><span\; class="notranslate">\; |</span>}<span\; class="notranslate">\; \&Center\; Dot;</span>\frac{{<span\; class="notranslate">\; \&dtri;</span>}^{<span\; class="notranslate">\; \&perp;</span>}{\mathrm{<span\; class="notranslate">\; p</span>}}_{\mathrm{<span\; class="notranslate">\; rec</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; x</span>}<span\; class="notranslate">\; +</span>\mathrm{<span\; class="notranslate">\; k</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; the\; y</span>}<span\; class="notranslate">\; )</span>}{<span\; class="notranslate">\; |</span>{<span\; class="notranslate">\; \&dtri;</span>}^{<span\; class="notranslate">\; \&perp;</span>}{\mathrm{<span\; class="notranslate">\; p</span>}}_{\mathrm{<span\; class="notranslate">\; rec</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; x</span>}<span\; class="notranslate">\; +</span>\mathrm{<span\; class="notranslate">\; k</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; the\; y</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; |</span>}<span\; class="notranslate">\; |</span><span\; class="notranslate">\; \&times;</span><span\; class="notranslate">\; |</span><span\; class="notranslate">\; \&dtri;</span>{\mathrm{<span\; class="notranslate">\; p</span>}}_{\mathrm{<span\; class="notranslate">\; rec</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; x</span>}<span\; class="notranslate">\; +</span>\mathrm{<span\; class="notranslate">\; k</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; the\; y</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; |</span>$

$<span\; class="notranslate">\; +</span>\underset{\mathrm{<span\; class="notranslate">\; k</span>}<span\; class="notranslate">\; -</span>\mathrm{<span\; class="notranslate">\; 0</span>}}{\overset{{\mathrm{<span\; class="notranslate">\; N</span>}}_{\mathrm{<span\; class="notranslate">\; h</span>}}<span\; class="notranslate">\; -</span>\mathrm{<span\; class="notranslate">\; 1</span>}}{\mathrm{<span\; class="notranslate">\; \&Sigma;</span>}}}<span\; class="notranslate">\; |</span>\frac{<span\; class="notranslate">\; \&dtri;</span><span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; \&Delta;p</span>}}_{\mathrm{<span\; class="notranslate">\; rec</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; x</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; the\; y</span>}<span\; class="notranslate">\; +</span>\mathrm{<span\; class="notranslate">\; k</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; )</span>}{<span\; class="notranslate">\; |</span><span\; class="notranslate">\; \&dtri;</span><span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; \&Delta;</span>}{\mathrm{<span\; class="notranslate">\; p</span>}}_{\mathrm{<span\; class="notranslate">\; rec</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; x</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; the\; y</span>}<span\; class="notranslate">\; +</span>\mathrm{<span\; class="notranslate">\; k</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; )</span><span\; class="notranslate">\; |</span>}<span\; class="notranslate">\; \&Center\; Dot;</span>\frac{{<span\; class="notranslate">\; \&dtri;</span>}^{<span\; class="notranslate">\; \&perp;</span>}{\mathrm{<span\; class="notranslate">\; p</span>}}_{\mathrm{<span\; class="notranslate">\; rec</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; x</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; the\; y</span>}<span\; class="notranslate">\; +</span>\mathrm{<span\; class="notranslate">\; k</span>}<span\; class="notranslate">\; )</span>}{<span\; class="notranslate">\; |</span>{<span\; class="notranslate">\; \&dtri;</span>}^{<span\; class="notranslate">\; \&perp;</span>}{\mathrm{<span\; class="notranslate">\; p</span>}}_{\mathrm{<span\; class="notranslate">\; rec</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; x</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; the\; y</span>}<span\; class="notranslate">\; +</span>\mathrm{<span\; class="notranslate">\; k</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; |</span>}<span\; class="notranslate">\; |</span><span\; class="notranslate">\; \&times;</span><span\; class="notranslate">\; |</span><span\; class="notranslate">\; \&dtri;</span>{\mathrm{<span\; class="notranslate">\; p</span>}}_{\mathrm{<span\; class="notranslate">\; rec</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; x</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; the\; y</span>}<span\; class="notranslate">\; +</span>\mathrm{<span\; class="notranslate">\; k</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; |</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; 6</span>}<span\; class="notranslate">\; )</span>$

是梯度算子，

是与梯度方向正交的方向的正规算子，

是拉普拉斯算子。在一个典型实施例中，这些相关的算子可以计算如下：where p _rec (x, y) refers to the current reconstructed pixel value at position (x, y) if (x, y) points to the outer boundary, and if (x, y) points to the inner boundary, p _rec ( x, y) refer to possible candidate blocks.

is the gradient operator,

is the normal operator in the direction orthogonal to the gradient direction,

is the Laplacian operator. In a typical embodiment, these related operators can be calculated as follows:

$<span\; class="notranslate">\; |</span>{<span\; class="notranslate">\; \&dtri;</span>}^{<span\; class="notranslate">\; \&perp;</span>}{\mathrm{<span\; class="notranslate">\; p</span>}}_{\mathrm{<span\; class="notranslate">\; rec</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; x</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; the\; y</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; |</span><span\; class="notranslate">\; =</span><span\; class="notranslate">\; |</span><span\; class="notranslate">\; \&dtri;</span>{\mathrm{<span\; class="notranslate">\; p</span>}}_{\mathrm{<span\; class="notranslate">\; rec</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; x</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; the\; y</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; =</span>\sqrt{{<span\; class="notranslate">\; [</span>\frac{<span\; class="notranslate">\; \&PartialD;</span>{\mathrm{<span\; class="notranslate">\; p</span>}}_{\mathrm{<span\; class="notranslate">\; rec</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; x</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; the\; y</span>}<span\; class="notranslate">\; )</span>}{<span\; class="notranslate">\; \&PartialD;</span>\mathrm{<span\; class="notranslate">\; x</span>}}<span\; class="notranslate">\; ]</span>}^{\mathrm{<span\; class="notranslate">\; 2</span>}}<span\; class="notranslate">\; +</span>{<span\; class="notranslate">\; [</span>\frac{<span\; class="notranslate">\; \&PartialD;</span>{\mathrm{<span\; class="notranslate">\; p</span>}}_{\mathrm{<span\; class="notranslate">\; rec</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; x</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; the\; y</span>}<span\; class="notranslate">\; )</span>}{<span\; class="notranslate">\; \&PartialD;</span>\mathrm{<span\; class="notranslate">\; the\; y</span>}}<span\; class="notranslate">\; ]</span>}^{\mathrm{<span\; class="notranslate">\; 2</span>}}}$

$\frac{<span\; class="notranslate">\; \&PartialD;</span>{\mathrm{<span\; class="notranslate">\; p</span>}}_{\mathrm{<span\; class="notranslate">\; rec</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; x</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; the\; y</span>}<span\; class="notranslate">\; )</span>}{<span\; class="notranslate">\; \&PartialD;</span>\mathrm{<span\; class="notranslate">\; x</span>}}<span\; class="notranslate">\; =</span>\frac{{\mathrm{<span\; class="notranslate">\; p</span>}}_{\mathrm{<span\; class="notranslate">\; rec</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; x</span>}<span\; class="notranslate">\; +</span>\mathrm{<span\; class="notranslate">\; 1</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; the\; y</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; -</span>{\mathrm{<span\; class="notranslate">\; p</span>}}_{\mathrm{<span\; class="notranslate">\; rec</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; x</span>}<span\; class="notranslate">\; -</span>\mathrm{<span\; class="notranslate">\; 1</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; the\; y</span>}<span\; class="notranslate">\; )</span>}{\mathrm{<span\; class="notranslate">\; 2</span>}}$

$\frac{<span\; class="notranslate">\; \&PartialD;</span>{\mathrm{<span\; class="notranslate">\; p</span>}}_{\mathrm{<span\; class="notranslate">\; rec</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; x</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; the\; y</span>}<span\; class="notranslate">\; )</span>}{<span\; class="notranslate">\; \&PartialD;</span>\mathrm{<span\; class="notranslate">\; the\; y</span>}}<span\; class="notranslate">\; =</span>\frac{{\mathrm{<span\; class="notranslate">\; p</span>}}_{\mathrm{<span\; class="notranslate">\; rec</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; x</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; the\; y</span>}<span\; class="notranslate">\; +</span>\mathrm{<span\; class="notranslate">\; 1</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; -</span>{\mathrm{<span\; class="notranslate">\; p</span>}}_{\mathrm{<span\; class="notranslate">\; rec</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; x</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; the\; y</span>}<span\; class="notranslate">\; -</span>\mathrm{<span\; class="notranslate">\; 1</span>}<span\; class="notranslate">\; )</span>}{\mathrm{<span\; class="notranslate">\; 2</span>}}$

$\frac{{<span\; class="notranslate">\; \&PartialD;</span>}^{\mathrm{<span\; class="notranslate">\; 2</span>}}{\mathrm{<span\; class="notranslate">\; p</span>}}_{\mathrm{<span\; class="notranslate">\; rec</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; x</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; the\; y</span>}<span\; class="notranslate">\; )</span>}{{<span\; class="notranslate">\; \&PartialD;</span>}^{\mathrm{<span\; class="notranslate">\; 2</span>}}\mathrm{<span\; class="notranslate">\; x</span>}}<span\; class="notranslate">\; =</span>{\mathrm{<span\; class="notranslate">\; p</span>}}_{\mathrm{<span\; class="notranslate">\; rec</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; x</span>}<span\; class="notranslate">\; +</span>\mathrm{<span\; class="notranslate">\; 1</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; the\; y</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; +</span>{\mathrm{<span\; class="notranslate">\; p</span>}}_{\mathrm{<span\; class="notranslate">\; rec</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; x</span>}<span\; class="notranslate">\; -</span>\mathrm{<span\; class="notranslate">\; 1</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; the\; y</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; -</span>\mathrm{<span\; class="notranslate">\; 2</span>}{\mathrm{<span\; class="notranslate">\; p</span>}}_{\mathrm{<span\; class="notranslate">\; rec</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; x</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; the\; y</span>}<span\; class="notranslate">\; )</span>$

$\frac{{<span\; class="notranslate">\; \&PartialD;</span>}^{\mathrm{<span\; class="notranslate">\; 2</span>}}{\mathrm{<span\; class="notranslate">\; p</span>}}_{\mathrm{<span\; class="notranslate">\; rec</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; x</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; the\; y</span>}<span\; class="notranslate">\; )</span>}{{<span\; class="notranslate">\; \&PartialD;</span>}^{\mathrm{<span\; class="notranslate">\; 2</span>}}\mathrm{<span\; class="notranslate">\; the\; y</span>}}<span\; class="notranslate">\; =</span>{\mathrm{<span\; class="notranslate">\; p</span>}}_{\mathrm{<span\; class="notranslate">\; rec</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; x</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; the\; y</span>}<span\; class="notranslate">\; +</span>\mathrm{<span\; class="notranslate">\; 1</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; +</span>{\mathrm{<span\; class="notranslate">\; p</span>}}_{\mathrm{<span\; class="notranslate">\; rec</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; x</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; the\; y</span>}<span\; class="notranslate">\; -</span>\mathrm{<span\; class="notranslate">\; 1</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; -</span>\mathrm{<span\; class="notranslate">\; 2</span>}{\mathrm{<span\; class="notranslate">\; p</span>}}_{\mathrm{<span\; class="notranslate">\; rec</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; x</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; the\; y</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; 7</span>}<span\; class="notranslate">\; )</span>$

In Equation (6), (▽(Δ ))/(|(▽(Δ ))|) represents the normalized gradient estimated by Laplacian, (▽ ^⊥ )/(|▽ ^⊥ |) is the normalized vector along the tangent direction. If the structure across the boundary is properly matched, the two terms should be orthogonal to each other and the inner product is zero. However, if the matching is not proper, the absolute value of the inner product of the two terms tends to become larger, causing the inner product to become larger. Furthermore, for each pixel in Equation (6), the inner product is multiplied by the gradient magnitude |▽·|, which tends to make the numerical range of EDD and SDD comparable. The inner product also reflects the condition of the blocks at the boundary.

For each predictor in CS, the corresponding reference block is used to compute BME. Those predictors with less BME were included in RCS. For example, in the special case of a CS of size -3 with mv _Scol , mv _SH.264 and mv _Snei , the predictor with the largest BME is removed and the other two predictors are included in the RCS.

E. Adaptive Template Matching

Another aspect of the present invention is to: determine the adaptive candidate set; reduce the candidate set by matching techniques to save the number of bits of the index; develop an adaptive template shape and width strategy; perform a guessing strategy based on the minimum MVD and the entire MV encoding process; MV Adaptive template matching is used in encoding as a matching technique. Furthermore, the present invention provides a strategy for determining template shape and width based on correlation and similarity criteria, which is suitable for a variety of applications such as but not limited to motion vector coding.

Further aspects of the invention also include:

The template shape for the current data space is determined based on possible dependencies between the template and the current data, e.g., in a block-based video compression system like H.264, if the macroblocks are divided into different partitions by the rate-distortion criterion ( This means that the correlation between different partitions can be very low), the template of this partition may not include pixels in another partition;

Determine the template size for the current data space based on the consistency within the template, e.g., calculate the possible intra-template differences, if pixels far from the current data space are not similar to pixels close to the current data space, then Pixels that are far away can be removed from the template; while

Template matching techniques may include, but are not limited to, using candidate-based displacements to calculate the difference between the template in the current data space and the displacement-based template in the reference picture.

Figure 3 shows an exemplary template for template matching. Template matching has been described in: "Decoder side motion vector derivation for inter frame video coding" by S. Kamp, M. Evertz and M. Wien, in IEEE International Conference on Image Processing, pp.1120 -1123, 2008, incorporated herein by reference. In order to obtain a good prediction of the current block, an L-shaped template region (TR) is defined around the left and upper boundaries of the target block. The template widths M_left and M_up are defined to be the width of pixels extending to the left and top of the object of the template area. To test the similarity between the template of the current block and the template of the candidate block corresponding to the MVP, only the sum of absolute difference (SAD for short) of the two templates needs to be calculated. When the template belongs to the same object as the current block, if the template SAD given by the MVP is small, it is reasonable to speculate that the block corresponding to the MVP can also provide a good prediction of the current block.

However, if some parts of the template area belong to another object with a different motion, this will still result in a large template SAD even if the MVP is good. In one embodiment, to avoid this situation, instead of using fixed sizes of M_left and M_up, an adaptive template shape and width are used based on the possible correlation between the template and the target block and the similarity inside the template. In the preferred embodiment, large templates are used since they give a more meaningful template SAD. But any pixels belonging to objects belonging to a different current block shall not be included in the template.

The L-shaped template consists of two parts: the upper part and the left part. The term "left formwork" refers to the left part of the L-shaped formwork. Similarly, the term "upper formwork" refers to the upper portion of the L-shaped formwork. Typically, we allow templates to be L-shaped templates, left templates, or upper templates.

In an exemplary embodiment of a size-3 CS with mv _Scol , mv _SH.264 and mv _Snei , we define the template shape selection strategy based on the correlation between the current block and its neighbors as follows:

if(blocktype=P16*16)

then use the left template and the upper template;

Else if (blocktype=P16*8) /* has a partition of 16x8 above and a partition of 16x8 below */

Then the upper partition uses the left template and the upper template, and the lower partition only uses the left template;

Else if (blocktype=P8*16)/*has a left 8x16 partition and a right 8x16 partition*/then

The left partition uses the left template and the upper template, and the right partition only uses the upper template;

else if (blocktype=P8*8) /* has four 8x8 partitions, some of which can be subdivided into smaller partitions of size 4x8, 8x4, 4x4 */

Then the partition in the upper left corner of the current MB uses the left template and the upper template,

Partitions at the left boundary of the current MB use only the left template,

The partition at the upper boundary of the current MB only uses the upper template,

The other partitions use the left template and the upper template.

The reason for this strategy is also why we want to divide macroblocks (MBs) into different partitions. Different partitions often have relatively low correlation or different motion situations. So when using the template SAD to select a good MVP, including pixels from another partition in the template may risk the correctness of the selected good MVP. But other template shape selection strategies can also be used.

If the current block is smooth, the pixels in the template are required to have similar brightness and smoothness. If the current block is a texture block, the pixels in the template are required to have a similar texture. If the current block has an edge, it is required that the pixels in the template have an edge that is an extension of the edge within the block, and that the pixels on both sides of the edge in the template are similar to the pixels within the block. In other words, set the stencil width based on the consistency or similarity between the stencil and the block.

In an exemplary embodiment of a size-3 CS with mv _Scol , mv _SH.264 , and mv _Snei , the construction criteria for the template width are defined as follows:

FIG. 4 shows an exemplary template with a template width equal to four.

The maximum width of the template is N=4; /*that is, the template has N rows of pixels*/

Computes the difference between adjacent rows in a template. Take the above template as an example, as shown in Figure 4, calculate:

SAD12=sum of absolute difference between row 1 and row 2 (SAD);

SAD23=SAD between row 2 and row 3;

SAD34=SAD between row 3 and row 4;

if (SAD12>SAD23+threshold1)

The upper template includes only 1 row; otherwise

if (SAD23>SAD12+threshold2‖SAD12>SAD34+threshold3)

The upper template includes 1 row and 2 rows; otherwise

if (SAD34>SAD12+threshold2)

The upper template includes 1 row, 2 rows and 3 rows;

otherwise

The upper template includes 1 row, 2 rows, 3 rows, 4 rows;

Similarly, use the same method to get the width of the left template. Once the template shape and width are set for the current block, the template of the reference block with the same shape and width is used to calculate the template SAD. A better matching template region in the reference picture has a smaller template SAD to the current template. Since the template has a high spatial correlation with the current block, it is reasonable to assume that blocks corresponding to better matching templates provide good predictions for the current block.

In the present invention, for each predictor in CS a corresponding template SAD is calculated. Those predictors with fewer template SADs are included in RCS. For example, in the special case of a CS of size-3 with mv _Scol , mv _SH.264 , and mv _Snei , the predictor with the largest template SAD is removed and the other two predictors are included in the RCS.

F. Final MVP decision

Another aspect of the invention is to perform a compare and count process on all candidates in the reduced set of candidates, each of which compares the difference between the candidate and the current parameter and encodes the smallest difference.

In the present invention, according to the set strategy, the candidate set reduced by the matching technique may contain one or more MVPs. For example, in the special case of a CS of size-3 with mv _Scol , mv _SH.264 and mv _Snei , the RCS contains two predictors. So it is also necessary to select the final predictor from the two predictors. Now limit the pmv _opt of the final MVP of the current block as follows:

${\mathrm{<span\; class="notranslate">\; pmv</span>}}_{\mathrm{<span\; class="notranslate">\; opt</span>}}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; arg</span>}\underset{{\mathrm{<span\; class="notranslate">\; pmv</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}}{\mathrm{<span\; class="notranslate">\; min</span>}}\mathrm{<span\; class="notranslate">\; D.</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; mv</span>}<span\; class="notranslate">\; -</span>{\mathrm{<span\; class="notranslate">\; pmv</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; 8</span>}<span\; class="notranslate">\; )</span>$

Wherein pmv _i refers to the i-th MVP included in the RCS, D(·) is a function used to measure the distance, and mv represents the motion vector of the current block. The final MVD of the current block is (mv-pmv _opt ).

G. Index coding

Another aspect of the invention is to encode the index of the candidate with the smallest difference compared to the current parameters by a guessing strategy. The guessing strategy is based on the encoded smallest difference and reduced candidate set: in general, N possible parameters are generated for N candidates, and the best candidate is then reselected for each possible parameter based on the smallest difference criterion and Get the new minimum difference. If the best candidate is the same as the candidate that produced the possible parameter (the new minimum difference is also equal to the original encoded minimum difference), then the candidate is possible; otherwise, the candidate is set as impossible. If there is only one possible candidate, no encoding index is required, otherwise an encoding index is required.

When all MVPs in the RCS are the same, no index notification is required for selection. Otherwise, one or more bits are required to inform the index of the final MVP. However, the optimal MVP can in some cases be obtained at the decoder without the need to transmit the index. The following illustrates how the optimal MVP is obtained: "Method for coding motion vector using 2-dimentional minimum bitrate predicting technique" by HyunDuk Cho, Sung Deuk Kim, US Patent No. 7023919B2, April 2006, the disclosure of which is incorporated herein by reference content. A guessing strategy is proposed. The difference is that this embodiment uses the minimum MVD as the criterion instead of the minimum bit rate, which is not efficient when different MVDs have the same bit rate.

In this embodiment, it is assumed that after the matching technique, the number of remaining predictors in the candidate set is N, and the best predictor and MVD have been determined. For each predictor MVP_i, it is first assumed to be the final predictor, so that a new motion vector (the value of MVP_i plus MVD) can be obtained. In this regard, the new motion vector selects the optimal one among the N predictors according to the standard with the smallest difference. If MVP_i is selected, it means that MVP_i is possible, otherwise MVP_i is impossible.

Taking as an example an exemplary embodiment of a size-3 CS with mv _Scol , mv _SH.264 and mv _Snei (with 2 MVPs mv _BM1 and mv _BM2 in RCS) to show how this works:

Step 1: Obtain two possible motion vectors MV1=MVD+mv _BM1 and MV2=MVD+mv _BM2 ;

Step 2: Assuming MV1 is the real motion vector, select the best predictor for MV1 according to Equation 8. If the best predictor for MV1 is mv _BM1 , set flag_BM1=1, otherwise flag_BM1=0;

Step 3: Assuming MV2 is the real motion vector, select the best predictor for MV2 according to Equation 8. If the best predictor for MV2 is mv _BM2 , set flag_BM2=1, otherwise flag_BM2=0;

Step 4: If ((flag_BM1=1 && flag_BM2=1)‖(flag_BM1=0&&flag_BM2=0)), the index of the final MVP will be transmitted. Otherwise no index is transmitted and the best predictor can be obtained according to flag_BM1 and flag_BM2.

Figure 5 shows a flowchart for MVCBM. In an exemplary embodiment using boundary matching as a matching technique for motion vector coding:

on the encoder side

MVCBM (encoder):

Step 501: Compute the scaled intermediate predictor (mv_SH.264), the co-located predictor (mv_Scol) and the most dissimilar predictor (mv_Snei) from the intermediate predictor in the neighbor predictors, but by the equation ( 1) A similar approach scales the co-location predictor (mv_Scol) and the neighbor predictor.

Step 503: Use boundary matching to select two predictors (mv_bm1 and mv_bm2) into the reduced candidate set (RCS) with less boundary matching error. It is of course also possible to select more than two detectors.

Step 505: Based on the minimum MVD standard, select the final MVP for the current block between mv_bm1 and mv_bm2 to obtain the MVD.

Step 507: Based on the current MVD, use the guessing strategy to judge whether the correct MVP can be detected; if the correct MVP can be detected, it is not necessary to encode and send bits as an index subsequently; otherwise, it is necessary to encode and send 1 bit as an index subsequently ( After the boundary match, the number of predictors is greater than 2, then the bit as index can be greater than 1);

Step 509: Encoding the MVD and other information, if the correct MVP cannot be detected, an encoding index is also required. Other information includes, for example, a coded block pattern (Coded Block Pattern, CBP for short), a quantized residual, and the like.

In the step 509, in the actual encoding process, the MVP indexes of all the partitions in the macroblock are coded and output together, and it is at the end of the encoding output of the entire macroblock. The reason for this is that at the decoding end, the decoding of the MVP index is only considered during the reconstruction process, and other information of the macroblock has been decoded before.

on the decoder side

MVCBM (decoder):

Step 511: Decode the MVD and other information, and start to reconstruct the current block. Other information includes, for example, CBP, quantized residuals, and the like.

Step 513 and step 515 are the same as steps 501 and 503 of the encoder.

Step 517: Based on the decoded MVD, use the guessing strategy to determine whether the MVP can be determined; if it cannot be determined which one is the MVP, decode 1 bit as an index.

Step 519: Obtain the MV of the current block by using MVP and MVD.

Fig. 6 shows a flowchart of MVCATM. In another exemplary embodiment using adaptive template matching as a matching technique for motion vector coding:

encoder side

MVCATM (encoder)

Step 601: Compute the scaled intermediate predictor (mv_SH.264), the co-located predictor (mv_Scol) and the most dissimilar predictor (mv_Snei) from the intermediate predictor among neighbor predictors, but by the equation ( 1) A similar approach scales the co-location predictor (mv_Scol) and the neighbor predictor.

Step 603: Based on the possible similarity and correlation between the template and the target block, define a template with adaptive shape and width for the current block;

Step 605: Use adaptive template matching to select two predictors (mv_tm1 and mv_tm2) with fewer template SADs.

Step 607: Based on the minimum MVD standard, select the final MVP for the current block between mv_tm1 and mv_tm2 to obtain the MVD.

Step 609: Based on the current MVD, use the guessing strategy to judge whether the correct MVP can be detected; if the correct MVP can be detected, there is no need to encode and send a bit as an index; otherwise, it is necessary to encode and send a bit as an index ;

Step 611: Encode the MVD and other information, and if the correct MVP cannot be detected, the index needs to be encoded. Other information includes, for example, CBP, quantized residuals, and the like.

decoder side

MVCATM (decoder):

Step 621: Decode the MVD and other information, and start to reconstruct the current block. Other information includes, for example, CBP, quantized residuals, and the like.

Step 623, Step 625 and Step 627: Same as Steps 601, 603, 605 on the encoder side.

Step 629: Based on the decoded MVD, use the guessing strategy to determine whether the MVP can be determined; if it cannot be determined which one is the MVP, decode 1 bit for the index.

Step 631: Obtain the MV of the current block by using MVP and MVD.

Figure 7 shows a block diagram of one embodiment of a video encoding system. The present invention performs motion vector coding (MVC) after motion estimation (ME). The motion vector will be provided to the macroblock if the macroblock is of inter-prediction (Inter-Pred for short) type. If the macroblock is of the intra-prediction (Intra-Pred) type, a motion vector is also unnecessary. Transforming the residual by transform (T) and quantizing the residual by quantization (Q) gives a quantized residual. The MVD is encoded together with the quantized residual and the index to be transmitted by entropy coding, and the output for transmission is sent to the decoder. Dequantization (Q-1) and inverse transform (T-1) are performed on the quantized residual. The reconstructed residual is added to the result of the prediction block obtained by the ME to give the reference image.

Figure 8 shows a block diagram of one embodiment of a video decoding system. When the decoded macroblock is an inter-prediction (Inter-Pred) type, MVD is further decoded. The bitstream is decoded by entropy decoding to obtain a decoded residual. The decoded residual is dequantized by dequantization (Q-1) and inverse transformed by inverse transform (T-1) to give the signal. Decoding of the motion vector coding method (DMVC) is performed during reconstruction. If the guessing strategy in DMVC cannot determine the motion vector predictor (MVP), further decoding of additional indices may be required. The MVs generated by DMVC are used in motion compensation (MC). The result of MC is added to the signal to give the reconstructed frame.

Figure 9 shows a motion vector encoding device according to an embodiment of the present invention, including the following parts:

901. A first selection module, configured to select predictors forming a candidate set;

903. The first processing module receives a candidate set of predictors of the first selection module, and uses a matching method to reduce the number of predictors in the candidate set;

905. The second selection module is connected to the first processing module, receives the remaining predictors in the candidate set, and selects the optimal predictor among the remaining predictors based on the minimum motion vector difference criterion;

907. The second processing module is connected to the second selection module, and obtains the current motion vector difference according to the selected optimal predictor;

909. The detection module is connected to the second processing module and the first processing module, and uses a guessing strategy to detect the motion vector predictor based on the difference between the remaining predictors obtained from the first processing module and the current motion vector;

911. The judgment module is connected to the detection module, and if the detection result of the detection module is that a correct motion vector predictor is detected, send the judgment result to the encoding module;

913. The encoding module receives the judgment result, and encodes at least the current motion vector difference or encodes at least the current motion vector difference and an index.

Fig. 10 shows a kind of motion vector decoding device of the embodiment of the present invention more, comprises the following parts:

1001. The receiving module receives at least the coded motion vector difference;

1003. The first decoding module is connected to the receiving module, and decodes the motion vector difference according to at least the received coded motion vector difference;

1005. A first selection module, configured to select predictors forming the candidate set;

1007. The first processing module receives the candidate set of predictors of the first selection module, and uses a matching method to reduce the number of predictors in the candidate set;

1009. A judging module, connected to the first processing module and the first decoding module, based on the difference between the remaining predictors in the candidate set and the decoded motion vector, using a guessing strategy to judge whether the motion vector predictor can be determined;

1011. The second decoding module is connected to the judgment module, and decodes the index according to the judgment result;

1013. The analysis module is connected to the first processing module and the second decoding module, and determines the motion vector predictor according to the decoded index;

1015. The second processing module is connected to the judgment module, the first decoding module, and the analysis module, and according to the judgment result, obtains a motion vector by using the decoded motion vector difference and the decoded and determined motion vector predictor; or by using The decoded motion vector difference and the determined motion vector predictor result in a motion vector.

Generally speaking, various implementations of video codec and motion vector codec can be implemented in various terminal equipment or user equipment, including but not limited to mobile phones and other wireless communication devices, personal digital assistants (PDAs) ), portable and desktop computers, imaging/video devices (such as digital cameras), audio-visual (AV) equipment (such as video players), gaming devices, Internet or local area network (LAN) devices that allow access and possibly browsing, and A movable unit or device combining these functions.

Embodiments of the present invention may be implemented in the form of software, hardware, application logic or a combination of software, hardware and application logic. Software, application logic and/or hardware may reside on integrated circuit chips, modules or memory. Portions of software, hardware, and/or application logic may reside on integrated circuit chips, portions of software, hardware, and/or application logic may reside on modules, and portions of software, hardware, and/or application logic may reside in memory, if desired. superior. In one example, the application logic, software or an instruction set resides on any of a variety of conventional computer-readable media. In the context of the present invention, a "computer-readable medium" may be any medium or device that contains, stores, communicates, propagates or transmits instructions for use by or in connection with an instruction execution system, device or device (such as a computer) common use of devices or equipment. A computer readable medium may include a computer readable storage medium, which may be any medium or device that contains or stores instructions for use by or in conjunction with an instruction execution system, device or device, such as a computer.

Different functions discussed herein may be performed in different orders from each other and/or concurrently, if desired. Furthermore, one or more of the functions described above may be optional or may be combined, if desired.

Although various aspects of the invention are indicated in the independent claims, other aspects of the invention include the features of the above-described embodiments and combinations thereof and/or dependent claims having features of the independent claims, not only those explicitly stated in the claims. given combination.

It is also noted herein that while the above describes exemplary embodiments of the invention, these descriptions should not be viewed in a limiting sense. Various changes and modifications can also be made without departing from the scope of the present invention defined in the claims.

Claims (10)

1. a motion vector encoder method, is characterized in that, comprises the following steps:

Step 1: the Candidate Set that limits fallout predictor, wherein said fallout predictor is based on the motion vector adaptive variation adjacent with described motion vector, in the situation that use a plurality of reference frames, adjacent motion vector before being used as predicting described motion vector, according to its time gap by convergent-divergent pro rata;

Step 2: the figure place that the number that utilizes matching process to reduce the fallout predictor in described Candidate Set uses to reduce index;

Step 3: based on selected optimum fallout predictor in the remaining predicted device of standard in described Candidate Set of minimum movement phasor difference, and it is poor to obtain current motion vector, the poor minimum of the fallout predictor that this is optimum and current motion vector value;

step 4: poor based on remaining predicted device and current motion vector, utilize guess strategy to survey motion vector predictor further to save the figure place for code index, wherein guess strategy refers to that the number of described remaining predicted device is at least 1, to each remaining predicted device and intermediate parameters of the poor generation of current motion vector, N remaining predicted device produced N intermediate parameters, standard based on the minimum movement phasor difference reselects fallout predictor to be selected and obtains new minimum movement phasor difference for each intermediate parameters, if the fallout predictor to be selected that reselects is identical with the remaining predicted device that produces this intermediate parameters, judge that this remaining predicted device is optional fallout predictor, otherwise set this remaining predicted device and be not optional fallout predictor, if the number of optional fallout predictor is 1, this optional fallout predictor is correct motion vector predictor,

Step 5: judge whether to detect correct motion vector predictor, if detect correct motion vector predictor, encoding, current motion vector is poor at least,, if can't detect correct motion vector predictor, to current motion vector is poor at least, with index, encodes;

Wherein said matching process is the adaptive template matching process, described adaptive template matching process is, be identified for the shape of the adaptive template in current data space based on correlation possible between adaptive template and current data, if the correlation between the different subregions in current data space is very low, the template of this subregion may not comprise the pixel in another subregion; Perhaps based on the consistency in adaptive template, be identified for the size of the adaptive template in current data space, wherein said consistency refers to the difference of the interior pixels of described adaptive template.

2. a kind of motion vector encoder method according to claim 1, it is characterized in that, described adaptive template matching process also comprises and utilizing based on displacement corresponding to fallout predictor in Candidate Set, calculates poor between the template of the data space after displacement in the template in current data space and reference picture.

3. a kind of motion vector encoder method according to claim 1, is characterized in that, in the adaptive template matching process, the shape of described adaptive template is specified.

4. a kind of motion vector encoder method according to claim 1, is characterized in that, after described step 5,, if detect correct motion vector predictor, sends the current motion vector of at least having encoded poor; , if can't detect correct motion vector predictor, send the position as index that the current motion vector of at least having encoded is poor and encoded.

5. a kind of motion vector encoder method according to claim 1, is characterized in that, the Candidate Set of described fallout predictor comprises some proportional room and time fallout predictors.

6. a kind of motion vector encoder method according to claim 1, it is characterized in that, in described step 1, the motion vector adjacent with described motion vector be adjacent with the described motion vector time, space is adjacent, the time and space is adjacent or based on the adjacent motion vector of analysis strategy.

7. the method for the decoding motion vector of having encoded, described motion vector of having encoded is encoded by the method for claim 1, it is characterized in that, comprises the following steps:

Step 1: decoding motion vectors is poor;

Step 2: limit the Candidate Set of fallout predictor, wherein said fallout predictor is based on the motion vector adaptive variation adjacent with described motion vector;

Step 3: utilize matching process to reduce the number of the fallout predictor in described Candidate Set, obtain the remaining predicted device;

step 4: based on the remaining predicted device in described Candidate Set and the motion vector difference of decoding, utilize the guess strategy judgement can determine motion vector predictor, wherein guess strategy refers to that the number of described remaining predicted device is at least 1, motion vector difference to each remaining predicted device and decoding produces an intermediate parameters, N remaining predicted device produced N intermediate parameters, standard based on the minimum movement phasor difference reselects fallout predictor to be selected and obtains new minimum movement phasor difference for each intermediate parameters, if the fallout predictor to be selected that reselects is identical with the remaining predicted device that produces this intermediate parameters, judge that this remaining predicted device is optional fallout predictor, otherwise set this remaining predicted device and be not optional fallout predictor, if the number of optional fallout predictor is 1, this optional fallout predictor is correct motion vector predictor,

Step 5:, if can determine motion vector predictor, utilize motion vector difference and the fixed motion vector predictor of decoding at least to obtain motion vector; If can not determine motion vector predictor, the index and judge motion vector predictor of decoding, utilize the motion vector difference of decoding at least and the motion vector predictor of having judged to obtain motion vector;

Wherein said matching process is the adaptive template matching process, described adaptive template matching process is, be identified for the shape of the adaptive template in current data space based on correlation possible between adaptive template and current data, if the correlation between the different subregions in current data space is very low, the template of this subregion may not comprise the pixel in another subregion; Perhaps based on the consistency in adaptive template, be identified for the size of the adaptive template in current data space, wherein said consistency refers to the difference of the interior pixels of described adaptive template.

8. the method for a kind of decoding according to claim 7 motion vector of having encoded, is characterized in that, the matching process that described matching process adopts during with coding is identical.

9. motion vector encoder device is characterized in that comprising following part:

First selects module, in order to select to form the fallout predictor of Candidate Set, wherein said fallout predictor is based on the motion vector adaptive variation adjacent with described motion vector, in the situation that use a plurality of reference frames, adjacent motion vector before being used as predicting described motion vector, according to its time gap by convergent-divergent pro rata;

The first processing module, receive the described first Candidate Set of selecting the fallout predictor of module, in order to utilize matching process, reduces the number of the fallout predictor in described Candidate Set;

Second selects module, with described the first processing module, is connected, and receives the remaining predicted device in described Candidate Set, based on standard selected optimum fallout predictor in described remaining predicted device of minimum movement phasor difference;

The second processing module, select module to be connected with described second, obtains current motion vector according to the fallout predictor of selected optimum poor, the fallout predictor that this is optimum and current motion vector value differ from minimum;

Detecting module, with described the second processing module be connected processing module and be connected, poor based on remaining predicted device and current motion vector in the described Candidate Set that receives, utilize guess strategy to survey motion vector predictor; Wherein guess strategy refers to that the number of described remaining predicted device is at least 1, to each remaining predicted device and intermediate parameters of the poor generation of current motion vector, N remaining predicted device produced N intermediate parameters, standard based on the minimum movement phasor difference reselects fallout predictor to be selected and obtains new minimum movement phasor difference for each intermediate parameters, if the fallout predictor to be selected that reselects is identical with the remaining predicted device that produces this intermediate parameters, judge that this remaining predicted device is optional fallout predictor, otherwise set this remaining predicted device, be not optional fallout predictor; If the number of optional fallout predictor is 1, this optional fallout predictor is correct motion vector predictor;

Judge module, be connected with described detecting module, judges that the result of detection of detecting module, whether as detecting correct motion vector predictor, sends to coding module with judged result;

Coding module, receive described judged result, if judged result is to detect correct motion vector predictor, encoding, current motion vector is poor at least,, if judged result is to detect correct motion vector predictor, to current motion vector is poor at least, with index, encode;

Wherein said matching process is the adaptive template matching process, described adaptive template matching process is, be identified for the shape of the adaptive template in current data space based on correlation possible between adaptive template and current data, if the correlation between the different subregions in current data space is very low, the template of this subregion may not comprise the pixel in another subregion; Perhaps based on the consistency in adaptive template, be identified for the size of the adaptive template in current data space, wherein said consistency refers to the difference of the interior pixels of described adaptive template.

10. motion vector decoding apparatus is characterized in that comprising following part:

Receiver module, receive the motion vector difference of at least having encoded;

The first decoder module, connect described receiver module, and according to the motion vector difference of at least having encoded that receives, decoding motion vectors is poor;

First selects module, in order to select to form the fallout predictor of Candidate Set, wherein said fallout predictor is based on the motion vector adaptive variation adjacent with described motion vector, in the situation that use a plurality of reference frames, adjacent motion vector before being used as predicting described motion vector, according to its time gap by convergent-divergent pro rata;

The first processing module, receive the described first Candidate Set of selecting the fallout predictor of module, in order to utilize matching process, reduces the number of the fallout predictor in described Candidate Set, obtains the remaining predicted device;

Can judge module, connect described the first processing module and described the first decoder module,, based on the remaining predicted device in described Candidate Set and the motion vector difference of decoding, utilize the guess strategy judgement determine motion vector predictor; Wherein guess strategy refers to that the number of described remaining predicted device is at least 1, motion vector difference to each remaining predicted device and decoding produces an intermediate parameters, N remaining predicted device produced N intermediate parameters, standard based on the minimum movement phasor difference reselects fallout predictor to be selected and obtains new minimum movement phasor difference for each intermediate parameters, if the fallout predictor to be selected that reselects is identical with the remaining predicted device that produces this intermediate parameters, judge that this remaining predicted device is optional fallout predictor, otherwise set this remaining predicted device, be not optional fallout predictor; If the number of optional fallout predictor is 1, this optional fallout predictor is correct motion vector predictor;

The second decoder module, connect described judge module, if can not determine motion vector predictor, and the index of decoding;

Analysis module, connect described the first processing module and the second decoder module, according to the index of decoding, judges motion vector predictor;

The second processing module, connect described judge module, described the first decoder module and described analysis module,, if can determine motion vector predictor, utilizes motion vector difference and the fixed motion vector predictor of decoding to obtain motion vector; Otherwise utilize the motion vector difference of decoding and the motion vector predictor of having judged to obtain motion vector;

Wherein said matching process is the adaptive template matching process, described adaptive template matching process is, be identified for the shape of the adaptive template in current data space based on correlation possible between adaptive template and current data, if the correlation between the different subregions in current data space is very low, the template of this subregion may not comprise the pixel in another subregion; Perhaps based on the consistency in adaptive template, be identified for the size of the adaptive template in current data space, wherein said consistency refers to the difference of the interior pixels of described adaptive template.

Images