HK1218595B - Rice parameter initialization for coefficient level coding in video coding process - Google Patents

Description

Initialization of Rice parameters for coefficient-level decoding during video decoding

This application claims the benefit of U.S. Provisional Patent Application No. 61/845,850, filed on July 12, 2013, U.S. Provisional Patent Application No. 61/846,512, filed on July 15, 2013, U.S. Provisional Patent Application No. 61/882,536, filed on September 25, 2013, U.S. Provisional Patent Application No. 61/898,968, filed on November 1, 2013, U.S. Provisional Patent Application No. 61/907,693, filed on November 22, 2013, and U.S. Provisional Patent Application No. 61/915,337, filed on December 12, 2013, each of which is incorporated herein by reference in its entirety.

Technical Field

This disclosure relates to video coding, and more particularly to techniques for coding transform coefficients.

Background Art

Digital video capabilities may be incorporated into a wide range of devices, including digital televisions, digital live broadcast systems, wireless broadcast systems, personal digital assistants (PDAs), laptop or desktop computers, digital cameras, digital recording devices, digital media players, video game devices, video game consoles, cellular or satellite radio telephones, video teleconferencing devices, and the like. Digital video devices implement video compression techniques (e.g., those described in the standards defined by MPEG-2, MPEG-4, ITU-T H.263, ITU-T H.264/MPEG-4 Part 10 (Advanced Video Coding (AVC)), the High Efficiency Video Coding (HEVC) standard currently under development, and extensions of these standards) to more efficiently transmit, receive, and store digital video information.

Video compression techniques include spatial prediction and/or temporal prediction to reduce or remove redundancy inherent in video sequences. For block-based video coding, a video slice (i.e., a video frame or portion of a video frame) may be partitioned into blocks. Video blocks in an intra-coded (I) slice of a picture are encoded using spatial prediction with respect to reference samples in neighboring blocks in the same picture. Video blocks in an inter-coded (P or B) slice may use spatial prediction with respect to reference samples in neighboring blocks in the same picture or temporal prediction with respect to reference samples in other reference pictures. Pictures may be referred to as frames, and reference pictures may be referred to as reference frames.

Spatial prediction or temporal prediction results in coding a predictive block for the block. Residual data represents the pixel differences between the original block to be coded and the predictive block. Inter-coded blocks are encoded based on a motion vector pointing to a block of reference samples forming the predictive block, and residual data indicating the difference between the coded block and the predictive block. Intra-coded blocks are encoded based on an intra-coding mode and the residual data. For further compression, the residual data can be transformed from the pixel domain to the transform domain, generating residual transform coefficients that can then be quantized. The quantized transform coefficients, initially arranged in a two-dimensional array, can be scanned in a particular order to generate a one-dimensional vector of transform coefficients, and entropy coding can be applied to achieve even greater compression.

Summary of the Invention

In general, this disclosure describes techniques for initializing Rice parameters used to define codes for coefficient-level coding in a video coding process. In particular, this disclosure describes techniques for determining initial values of the Rice parameters used to define codes (e.g., Colomb-Rice codes or Exponential-Golomb codes) used to code residual absolute values of coefficient levels of coefficients, where context-adaptive binary arithmetic coding (CABAC) is used to code significant coefficients, coefficient levels greater than 1, and indications of coefficient levels greater than 2. In some examples, the techniques may be applied to Rice parameter initialization for coefficient-level coding in a range extension of the High Efficiency Video Coding (HEVC) standard.

The techniques described in the present invention determine the initial values of the Rice parameters for the current coefficient group (CG) (i.e., coefficient block) in the transform block of the video data based on statistics of coefficient levels collected for previously decoded coefficients of the video data. In the case of lossy decoding, the CG may include transform coefficients, or in the case of lossless decoding or lossy decoding in transform skip mode, the CG may include coefficients to which the transform is not applied. The statistics may be statistics of the absolute values of the coefficient levels of the previously decoded coefficients or the residual absolute values of the coefficient levels. The values of the statistics may be initialized to zero at the beginning of each slice of the video data, and the statistics may be updated based on one or more coefficient levels decoded in each CG of the slice. In an example, when the first coefficient level is decoded in the CG, the statistics may be updated once for each CG. In some cases, statistics may be collected separately for each of multiple different categories of CGs defined based on the characteristics of the transform block containing the CG. According to the technology of the present invention, at the beginning of the current CG in the transform block, the value of the statistic is mapped to the initial value of the Rice parameter for the current CG.

In an example, the present invention relates to a method for decoding coefficients in a video decoding process, the method comprising determining statistics of coefficient levels of previously decoded coefficients of residual video data; determining initial values of Rice parameters for a current coefficient group in a transform block of the residual video data based on the statistics; and decoding the residual absolute value of the coefficient level of at least one coefficient in the current coefficient group using a code defined by the Rice parameters.

In another example, the present invention relates to a method for encoding coefficients in a video encoding process, the method comprising determining statistics of coefficient levels of previously encoded coefficients of residual video data; determining initial values of Rice parameters for a current coefficient group in a transform block of the residual video data based on the statistics; and encoding the residual absolute value of the coefficient level of at least one coefficient in the current coefficient group using a code defined by the Rice parameters.

In another example, the present invention relates to a video decoding device comprising a memory configured to store video data, and one or more processors configured to: determine statistics of coefficient levels of previously decoded coefficients of residual video data; determine initial values of Rice parameters for a current coefficient group in a transform block of the residual video data based on the statistics; and decode the residual absolute value of the coefficient level of at least one coefficient in the current coefficient group using a code defined by the Rice parameters.

In another example, the present invention relates to a video decoding device comprising: a device for determining statistics of coefficient levels of previously decoded coefficients of residual video data; a device for determining initial values of Rice parameters for a current coefficient group in a transform block of the residual video data based on the statistics; and a device for decoding residual absolute values of the coefficient levels of at least one coefficient in the current coefficient group using a code defined by the initial values of the Rice parameters.

In another example, the present invention relates to a computer-readable storage medium comprising instructions that, when executed by one or more processors of a video decoding device, cause the processors to perform the following operations: determine statistics of coefficient levels of previously decoded coefficients of residual video data; determine initial values of Rice parameters for a current coefficient group in a transform block of the residual video data based on the statistics; and decode the residual absolute value of the coefficient level of at least one coefficient in the current coefficient group using a code defined by the Rice parameters.

The details of one or more examples are set forth in the accompanying drawings and the description below. Other features, objects, and advantages will be apparent from the description and drawings, and from the claims.

BRIEF DESCRIPTION OF THE DRAWINGS

1 is a block diagram illustrating an example video encoding and decoding system that may utilize the techniques for coding coefficient levels described in this disclosure.

2 is a conceptual diagram showing an example inverse scan order for coefficient-level coding.

3 is a conceptual diagram showing an example of coefficient-level coding for coefficient groups (CGs) based on inverse diagonal scan order for sub-blocks.

Figure 4 is a conceptual diagram showing an example anti-diagonal scan order for decoding coefficient levels for CG.

5 is a block diagram illustrating an example video encoder that may implement the techniques described in this disclosure for encoding coefficient levels.

6 is a block diagram illustrating an example video decoder that may implement the techniques for decoding coefficient levels described in this disclosure.

7 is a flowchart illustrating an example operation of determining initial values of Rice parameters during entropy encoding at the coefficient level according to the techniques described in this disclosure.

8 is a flowchart illustrating an example operation of determining initial values of Rice parameters during entropy decoding at the coefficient level according to the techniques described in this disclosure.

9 is a flowchart illustrating an example operation of determining statistics for a coefficient level of previously coded coefficients during entropy coding of the coefficient level according to the techniques described in this disclosure.

10 is a flowchart illustrating example operations for determining initial values of Rice parameters for a current group of coefficients based on determined statistics according to the techniques described in this disclosure.

DETAILED DESCRIPTION

This disclosure describes techniques for coding coefficients associated with residual data in a video coding process. The techniques are configured for initializing Rice parameters used to define codes for coefficient-level coding in the video coding process. In particular, this disclosure describes techniques for determining initial values for Rice parameters used to define codes (e.g., Golomblais codes or Exponential Golomb codes) used to code residual absolute values of coefficient levels of a coefficient block, where context-adaptive binary arithmetic coding (CABAC) is used to code significant coefficients, coefficient levels greater than 1, and indications of coefficient levels greater than 2. In the case of lossy coding, the coefficient levels may be levels of transform coefficients, or in the case of lossless coding or lossy coding in transform skip mode, the coefficient levels may be levels of coefficients to which a transform is not applied (i.e., residual pixel values). In some examples, the techniques may be applied to Rice parameter initialization for coefficient-level coding in a range extension of the High Efficiency Video Coding (HEVC) standard.

The Rice parameter is an adjustable value for selecting a set of codewords from the same family of Golomb codes (e.g., Golombrite codes or exponential Golomb codes). The code defined by the Rice parameter can be used to decode the residual absolute value of the coefficient level of at least one coefficient in a coefficient group (CG) (i.e., a coefficient block). In the example of HEVC, each of the CGs may include a 4×4 transform block of video data, or a 4×4 sub-block of a transform block. In the case of lossy decoding, the CG may include transform coefficients, or in the case of lossless decoding or lossy decoding in transform skip mode, the CG may include coefficients to which the transform is not applied. In some processes, the initial value of the Rice parameter is set to zero at the beginning of each CG and is conditionally updated after the residual absolute value of the coefficient level in the decoding CG. In the case of decoding the coefficient level of screen content, or in the case of lossless decoding or lossy decoding in transform skip mode, it may not be optimal to initialize the value of the Rice parameter to zero for each CG.

The techniques described in this disclosure may adaptively set initial values of Rice parameters for coding (e.g., encoding or decoding) each CG, rather than setting the initial values of the Rice parameters to zero for all situations. In particular, this disclosure describes techniques for determining initial values of Rice parameters for the current CG based on statistics collected at the coefficient level for previously coded coefficients. This disclosure also describes techniques for determining statistics at the coefficient level for previously coded coefficients of video data.

The statistics may be statistics of the absolute values of a coefficient level or the remaining absolute values of a coefficient level for previously coded coefficients. The statistics may be initialized to zero at the beginning of each slice of video data and may be updated based on one or more coefficient levels coded in each CG of the slice. In some cases, the statistics may be updated once per CG when the first absolute value of a coefficient level or the first remaining absolute value of a coefficient level is being coded in the CG. In other cases, statistics may be collected more frequently or based on different coefficient levels, such as when the last absolute value of a coefficient level or the last remaining absolute value of a coefficient level is being coded in the CG.

As one example, the statistic may be determined by comparing the coefficient level of a given previously coded coefficient to a predefined function of the statistic, and then determining whether to increase or decrease the statistic value based on the comparison. The predefined function of the statistic used to update the statistic may be based on a first constant value of the statistic value that is left-shifted and divided by a second constant value. In other examples, the statistic may be determined according to different techniques.

In some cases, statistics may be collected separately for each of a plurality of different categories of CGs defined based on characteristics of the transform block that includes the CG. In this case, the category of the current CG in the transform block may be determined based on the characteristics of the transform block, and the Rice parameters for the current CG may be initialized based on the statistics for the determined category. In one example, separate statistics may be collected for each of four different categories based on whether the transform block is a luma block, and whether the transform block is a transform skip block. In other examples, the statistics may be split into different numbers of categories defined based on different types of transform block characteristics.

According to the technology of the present invention, at the beginning of the current CG, the statistical value is mapped to the initial value of the Rice parameter for the current CG. In some instances, the statistical value can be mapped to the initial Rice parameter value according to a statistical function. The statistical function for initializing the Rice parameter can be based on the selection of the minimum of the maximum value of the Rice parameter or the statistical value divided by a constant value. In other instances, the statistical value can be mapped to the initial value of the Rice parameter according to different functions or according to a storage table. The initial value of the Rice parameter for the current CG is used to define a code (e.g., a Golomblais code or an exponential Golomb code), which is used to decode the residual absolute value of the coefficient level of at least one coefficient in the current CG.

In some examples, a video encoder may encode coefficient levels of coefficients associated with the residual video data into a bitstream for transmission to a video decoder or storage device. After receiving the encoded bitstream, the video decoder may decode the coefficient levels of the residual video data in a manner reciprocal to that of the video encoder. In the case of lossy video coding, the coefficients may be quantized transform coefficients. In this case, the quantized transform coefficients may be generated, for example, by applying a transform (e.g., a discrete cosine transform (DCT)) to the residual video data and then applying quantization to the transform coefficients. In the case of lossless video coding or lossy video coding with transform skipping or bypassing, the coefficients may be pixel values of the residual video data and have coefficient levels (i.e., pixel values) with large absolute values. When the coefficients represent screen content (which may include graphics and text areas), the content may not be well predicted, resulting in large absolute values of the coefficient levels of the coefficients.

The Rice parameter initialization scheme described in the present invention allows the initial values of the Rice parameters to be set to non-zero values at the beginning of the current CG so that the Rice parameters can adapt to large coefficient values quickly and efficiently, which may occur if the current CG contains screen content or is decoded with a transform skip or bypass. According to the technology, the initial values of the Rice parameters can be determined based on the statistics of the coefficient levels of the previously decoded coefficients. In this way, the Rice parameters can be initialized to non-zero values in order to adapt to slices or decoding units of screen content and/or coefficients that have not yet been transformed or quantized, but in the case of slices or decoding units of natural content, they can still be initialized to zero. For example, when a large coefficient level of the previously decoded coefficients brings about a large statistical value, the initial value of the Rice parameter can be set to a non-zero value based on the large statistical value in order to adapt more quickly to the large coefficient values that are likely to occur in the current CG.

FIG1 is a block diagram illustrating an example video encoding and decoding system 10 that may utilize the techniques for coding coefficient levels described in this disclosure. As shown in FIG1 , system 10 includes a source device 12 that transmits encoded video to a destination device 14 via a communication channel 16. As desired, the encoded video data may also be stored on a storage medium 34 or a file server 36 and accessed by the destination device 14. When stored on the storage medium or file server, video encoder 20 may provide the coded video data to another device (e.g., a network interface, a compact disc (CD), Blu-ray or digital video disc (DVD) recorder or imprint facility device, or other device) for storing the coded video data to the storage medium. Similarly, a device separate from video decoder 30 (e.g., a network interface, a CD or DVD reader, or the like) may retrieve the coded video data from the storage medium and provide the retrieved data to video decoder 30.

Source device 12 and destination device 14 may comprise any of a wide range of devices, including desktop computers, notebook (i.e., laptop) computers, tablet computers, set-top boxes, telephone handsets such as so-called smartphones, televisions, cameras, display devices, digital media players, video game consoles, or the like. In many cases, these devices may be equipped for wireless communication. Thus, communication channel 16 may comprise a wireless channel, a wired channel, or a combination of wireless and wired channels suitable for transmitting encoded video data. Similarly, file server 36 may be accessed by destination device 14 via any standard data connection, including an Internet connection. This data connection may include a wireless channel (e.g., a Wi-Fi connection), a wired connection (e.g., DSL, cable modem, etc.), or a combination of the two suitable for accessing encoded video data stored on the file server.

According to examples of this disclosure, techniques for coding coefficient levels may be applied to video coding to support any of a variety of multimedia applications, such as over-the-air television broadcasting, cable television transmission, satellite television transmission, streaming video transmission, for example, over the Internet, encoding of digital video for storage on a data storage medium, decoding of digital video stored on a data storage medium, or other applications. In some examples, system 10 may be configured to support one-way or two-way video transmission to support applications such as video streaming, video playback, video broadcasting, and/or video telephony.

1 , source device 12 includes a video source 18, a video encoder 20, a modulator/demodulator 22, and a transmitter 24. In source device 12, video source 18 may include, for example, a video capture device (e.g., a video camera), a video archive disk containing previously captured video, a video feed interface for receiving video from a video content provider, and/or a computer graphics system for generating computer graphics data as source video, or a combination of these sources. As an example, if video source 18 is a video camera, source device 12 and destination device 14 may form so-called camera phones or video phones, which may be provided within, for example, a smartphone or tablet computer. However, the techniques described in this disclosure may be applicable to video coding in general and may be applied to wireless and/or wired applications, or applications in which the encoded video data is stored locally on disk.

The captured video, pre-captured video, or computer-generated video may be encoded by video encoder 20. The encoded video information may be modulated by modem 22 according to a communication standard (e.g., a wired or wireless communication protocol) and transmitted to destination device 14 via transmitter 24. Modem 22 may include various mixers, filters, amplifiers, or other components designed for signal modulation. Transmitter 24 may include circuits designed for transmitting data, including amplifiers, filters, and, in the case of wireless communication, one or more antennas.

The captured video, pre-captured video, or computer-generated video encoded by video encoder 20 may also be stored on storage medium 34 or file server 36 for later consumption. Storage medium 34 may include a Blu-ray disc, a DVD, a CD-ROM, a flash memory, or any other suitable digital storage medium for storing encoded video. The encoded video stored on storage medium 34 may then be accessed by destination device 14 for decoding and playback. Although not shown in FIG1 , in some examples, storage medium 34 and/or file server 36 may store the output of transmitter 24.

File server 36 may be any type of server capable of storing encoded video and transmitting it to destination device 14. Example file servers include a web server (e.g., for a website), an FTP server, a network attached storage (NAS) device, a local disk drive, or any other type of device capable of storing encoded video data and transmitting it to a destination device. The transmission of encoded video data from file server 36 may be a streaming transmission, a download transmission, or a combination of both. File server 36 may be accessed by destination device 14 via any standard data connection, including an Internet connection. This standard data connection may include a wireless channel (e.g., a Wi-Fi connection), a wired connection (e.g., DSL, cable modem, Ethernet, USB, etc.), or a combination of the two suitable for accessing encoded video data stored on the file server.

1 , destination device 14 includes a receiver 26, a modem 28, a video decoder 30, and a display device 32. Receiver 26 of destination device 14 receives information via channel 16, and modem 28 demodulates the information to produce a demodulated bitstream for video decoder 30. The information communicated via channel 16 may include a variety of syntax information generated by video encoder 20 for use by video decoder 30 in decoding the video data. This syntax may also be included in the encoded video data stored on storage medium 34 or file server 36. Each of video encoder 20 and video decoder 30 may form part of a respective encoder-decoder (CODEC) capable of encoding or decoding video data.

Display device 32 may be integrated with destination device 14 or external to the destination device. In some examples, destination device 14 may include an integrated display device and also be configured to interface with an external display device. In other examples, destination device 14 may be a display device. Generally, display device 32 displays decoded video data to a user and may include any of a variety of display devices, such as a liquid crystal display (LCD), a plasma display, an organic light emitting diode (OLED) display, or another type of display device.

1 , communication channel 16 may comprise any wireless or wired communication medium, such as a radio frequency (RF) spectrum or one or more physical transmission lines, or any combination of wireless and wired media. Communication channel 16 may form part of a packet-based network, such as a local area network, a wide area network, or a global network such as the Internet. Communication channel 16 generally represents any suitable communication medium, or collection of different communication media, for transmitting video data from source device 12 to destination device 14, including any suitable combination of wired or wireless media. Communication channel 16 may include routers, switches, base stations, or any other equipment that can be used to facilitate communication from source device 12 to destination device 14.

Video encoder 20 and video decoder 30 may operate according to a video compression standard, such as the High Efficiency Video Coding (HEVC) standard developed by the Joint Collaboration Team on Video Coding (JCT-VC) of the ITU-T Video Coding Experts Group (VCEG) and the ISO/IEC Moving Picture Experts Group (MPEG). A draft of the HEVC standard in document JCTVC-L1003v34 (Bross et al., “High Efficiency Video Coding (HEVC) Text Specification Draft 10,” Joint Collaboration Team on Video Coding (JCT-VC) of ITU-T SG16 WP3 and ISO/IEC JTC1/SC29/WG11, 12th Meeting, Geneva, Switzerland, January 14-23, 2013) is available at http://phenix.int-evry.fr/jct/doc_end_user/documents/12_Geneva/wg11/JCTVC-L1003-v34.zip.

Although not shown in FIG1 , in some aspects, video encoder 20 and video decoder 30 may each be integrated with an audio encoder and decoder and may include appropriate MUX-DEMUX units or other hardware and software to handle the encoding of both audio and video in a common data stream or in separate data streams. If applicable, in some examples, the MUX-DEMUX units may conform to the ITU H.223 multiplexer protocol, or other protocols such as the User Datagram Protocol (UDP).

Video encoder 20 and video decoder 30 may each be implemented as any of a variety of suitable encoder circuitry, such as one or more microprocessors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), discrete logic, software, hardware, firmware, or any combination thereof. When the techniques are partially implemented in software, a device may store instructions for the software in a suitable, non-transitory computer-readable medium and execute the instructions in hardware using one or more processors to perform the techniques of this disclosure. Each of video encoder 20 and video decoder 30 may be included in one or more encoders or decoders, either of which may be integrated as part of a combined encoder/decoder (CODEC) in the respective device.

Video encoder 20 may implement any or all of the techniques of this disclosure for encoding coefficient levels during video encoding. Similarly, video decoder 30 may implement any or all of these techniques for decoding coefficient levels during video decoding. As described in this disclosure, a video decoder may refer to a video encoder or a video decoder. Similarly, a video decoding unit may refer to a video encoder or a video decoder. Likewise, video decoding may refer to video encoding or video decoding.

Digital video devices implement video compression techniques to more efficiently encode and decode digital video information. Video compression may apply spatial (intra-frame) prediction and/or temporal (inter-frame) prediction techniques to reduce or remove redundancy inherent in video sequences. The HEVC standard described above is based on an evolved model of a video coding device, referred to as the HEVC Test Model (HM). The HM assumes several additional capabilities of video coding devices relative to existing devices according to, for example, ITU-T H.264/AVC. For example, H.264 provides nine intra-frame prediction coding modes, while the HEVC HM can provide up to thirty-three intra-frame prediction coding modes. The following sections discuss certain aspects of the HM in more detail.

For video coding according to the HEVC standard, video frames can be divided into coding units. A coding unit (CU) generally refers to an image region that serves as the basic unit for video compression to which various coding tools are applied. A CU typically has a luma component, denoted as Y, and two chroma components, denoted as U and V. Depending on the video sampling format, the sizes of the U and V components can be the same as or different from the size of the Y component, depending on the number of samples.

A CU is typically square and can be considered similar to, for example, so-called macroblocks under other video coding standards, such as ITU-T H.264. For purposes of illustration, coding according to some of the currently proposed aspects of the developing HEVC standard will be described in this application. However, the techniques described in this disclosure may be used for other video coding processes, such as those defined according to H.264 or other standards or proprietary video coding processes.

According to the HM, a CU may include one or more prediction units (PUs) and/or one or more transform units (TUs). Syntax data within the bitstream may define a largest coding unit (LCU), which is the largest CU in terms of the number of pixels. Generally speaking, a CU has a similar purpose to a macroblock in H.264, except that CUs do not have a size distinction. Thus, a CU may be split into sub-CUs. Generally speaking, references to a CU in this disclosure may refer to a largest coding unit of a picture or a sub-CU of an LCU. An LCU may be split into multiple sub-CUs, and each sub-CU may be further split into multiple sub-CUs. Syntax data for the bitstream may define the maximum number of times an LCU may be split (referred to as the CU depth). Thus, the bitstream may also define a smallest coding unit (SCU). This disclosure also uses the terms "block" or "portion" to refer to any of a CU, PU, or TU. Generally speaking, a "portion" may refer to any subset of a video frame.

An LCU may be associated with a quadtree data structure. Generally speaking, the quadtree data structure includes a node for each CU, where the root node corresponds to the LCU. If a CU is split into four sub-CUs, the node corresponding to the CU includes four leaf nodes, each of which corresponds to one of the sub-CUs. Each node in the quadtree data structure may provide syntax data for the corresponding CU. For example, a node in the quadtree may include a split flag, indicating whether the CU corresponding to the node is split into sub-CUs. Syntax elements for a CU may be defined recursively and may depend on whether the CU is split into sub-CUs. If a CU is not further split, the CU is referred to as a leaf-CU. In the present invention, the four sub-CUs of a leaf-CU are also referred to as leaf-CUs even though there is no explicit splitting of the original leaf-CU. For example, if a CU in size 16×16 is not further split, the four 8×8 sub-CUs will also be referred to as leaf-CUs even though the 16×16 CU has never been split.

A leaf-CU may include one or more prediction units (PUs). In general, a PU represents all or a portion of the corresponding CU and may include data used to retrieve reference samples for the PU. For example, when the PU is inter-mode encoded, the PU may include data defining a motion vector for the PU. The data defining the motion vector may describe, for example, the horizontal component of the motion vector, the vertical component of the motion vector, the resolution of the motion vector (e.g., quarter-pixel precision or eighth-pixel precision), the reference frame to which the motion vector points, and/or the reference list for the motion vector (e.g., list 0 or list 1). The data defining the leaf-CU of the PU may also describe, for example, the partitioning of the CU into one or more PUs. The partitioning mode may be different depending on whether the CU is uncoded, intra-prediction mode encoded, or inter-prediction mode encoded. For intra-coding, PUs may be treated the same as leaf transform units described below.

The emerging HEVC standard allows transforms to be performed in terms of transform units (TUs), which can be different for different CUs. The TU size is typically set based on the size of the PU within a given CU defined for a partitioned LCU, but this may not always be the case. The TU is typically the same size as the PU, or smaller than the PU. In some examples, a quadtree structure known as a "residual quadtree" (RQT) can be used to subdivide the residual samples corresponding to a CU into smaller units. The leaf nodes of the RQT can be referred to as transform units (TUs). Pixel difference values associated with a TU can be transformed to produce quantizable transform coefficients. A TU includes a luma transform block and two chroma transform blocks. Thus, any decoding process discussed below that applies to a TU can actually be applied to the luma and chroma transform blocks.

Generally speaking, a PU refers to data related to the prediction process. For example, when the PU is intra-coded, the PU may include data describing the intra-prediction mode of the PU. As another example, when the PU is inter-coded, the PU may include data defining the motion vector of the PU.

In general, TUs are used for the transform and quantization processes. A given CU with one or more PUs may also include one or more transform units (TUs). After prediction, the video encoder 20 may calculate residual values based on the PUs, the video blocks identified by the free coding nodes. The coding nodes are then updated to reference the residual values instead of the original video blocks. The residual values include pixel difference values, which can be transformed into transform coefficients using the transform and other transform information specified in the TU, quantized and scanned to produce serialized transform coefficients for entropy coding. The coding nodes can be updated again to refer to these serialized transform coefficients. This disclosure generally uses the term "video block" to refer to the coding nodes of a CU. In some specific cases, this disclosure may also use the term "video block" to refer to a tree block (i.e., LCU or CU) that includes the coding nodes, PUs, and TUs.

A video sequence typically includes a series of video frames or pictures. A group of pictures (GOP) typically includes one or more of a series of video pictures. A GOP may include a header for the GOP, a header for one or more of the pictures, or syntax data elsewhere that describes the number of pictures included in the GOP. Each slice of a picture may include slice syntax data that describes the coding mode of the corresponding slice. Video encoder 20 typically operates on video blocks within individual video slices to encode the video data. A video block may correspond to a coding node within a CU. A video block may have a fixed or varying size and may differ in size according to a specified coding standard.

To code a block (e.g., a prediction unit of video data), a predictor for the block is first derived. The predictor (also called a predictive block) can be derived via either intra (I) prediction (i.e., spatial prediction) or inter (P or B) prediction (i.e., temporal prediction). Thus, some prediction units may be intra-coded (I) using spatial prediction relative to reference samples in neighboring reference blocks in the same frame (or slice), and other prediction units may be unidirectionally inter-coded (P) or bidirectionally inter-coded (B) relative to blocks of reference samples in other, previously coded frames (or slices). In each case, the reference samples may be used to form a predictive block for the block to be coded.

After identifying the predictive block, the difference between the pixels in the original video data block and the pixels in the predictive block is determined. This difference may be referred to as prediction residual data and indicates the pixel differences between the pixel values in the block to be coded and the pixel values in the predictive block selected to represent the coded block. To achieve better compression, the prediction residual data may be transformed, for example, using a discrete cosine transform (DCT), a discrete sine transform (DST), an integer transform, a K-L transform, or another transform to produce transform coefficients.

The residual data in a transform block (e.g., a TU) can be arranged as a two-dimensional (2D) array of pixel difference values residing in the spatial pixel domain. The transform converts the residual pixel values into a two-dimensional array of transform coefficients in a transform domain (e.g., the frequency domain). For further compression, the transform coefficients can be quantized before entropy coding. In some examples (e.g., lossless coding or lossy coding with transform skipping or bypassing), both the transform process and the quantization process of the coefficients can be skipped.

Then, an entropy coder applies entropy coding to the coefficients, such as context-adaptive variable length coding (CAVLC), context-adaptive binary arithmetic coding (CABAC), probability interval partitioning entropy coding (PIPE), or the like. In some examples, video encoder 20 may scan the coefficients using a predefined scan order to generate a serialized vector that can be entropy encoded. In other examples, video encoder 20 may perform an adaptive scan. After scanning the coefficients to form a one-dimensional vector, video encoder 20 may entropy encode the one-dimensional vector. Video encoder 20 may also entropy encode syntax elements associated with the encoded video data used by video decoder 30 when decoding the video data.

This disclosure relates to techniques for bypassing coding in conjunction with a context-adaptive binary arithmetic coding (CABAC) entropy coder or other entropy coders, such as probability interval partitioning entropy coding (PIPE) or correlation coders. Arithmetic coding is a form of entropy coding used in many compression algorithms with higher coding efficiency because it can map symbols to codewords of non-integer length. An example of an arithmetic coding algorithm is context-based binary arithmetic coding (CABAC).

In general, entropy coding data symbols using CABAC involves one or more of the following steps:

(1) Binarization: If the symbol to be decoded is a non-binary value, it is mapped into a sequence of so-called "bits". Each bit can have a value of "0" or "1".

(2) Context assignment: Assign a context to each bit (in normal mode). The context model determines how to calculate the context for a given bit based on information available for the bit (e.g., the value of the previously encoded symbol or the number of bits).

(3) Bit encoding: Bits are encoded by an arithmetic encoder. To encode a bit, the arithmetic encoder requires as input the probability of the value of the bit, i.e., the probability that the value of the bit is equal to "0" and the probability that the value of the bit is equal to "1". The estimated probability for each context is represented by an integer value called "context state". Each context has a state, and therefore, for a bit assigned to a context, the state (i.e., the estimated probability) is the same and different between contexts.

(4) State update: Update the probability state of the selected context based on the actual coded value of the bit (eg, if the bit value is "1," increase the probability of "1's").

In the case of CABAC entropy coding of data symbols in bypass mode, the symbols to be coded are binarized into a sequence of bits and arithmetically coded using a fixed equal probability model (e.g., using an exponential Golomb or Golombrite code). Bypass mode does not require context assignment or probability state updates. For example, this disclosure describes techniques for bypassing the residual absolute values of coefficient levels of coded coefficients using a Golomb code defined by Rice parameters. It should be noted that probability interval partitioning entropy coding (PIPE) uses principles similar to those of arithmetic coding and, therefore, can also utilize the techniques of this disclosure.

CABAC in H.264/AVC and HEVC uses states, and each state implicitly involves a probability. There are variations of CABAC where the probability of a symbol ("0" or "1") is used directly, i.e., the probability or an integer version of the probability is the state. For example, these variations of CABAC are described in “Description of video coding technology proposal by France Telecom, NTT, NTT DOCOMO, Panasonic and Technicolor” (JCTVC-A114, 1st JCT-VC meeting, Dresden, Germany, April 2010, hereinafter referred to as “JCTVC-A114”) and in A. Alshin and E. Alshina, “Multi-parameter probability update for CABAC” (JCTVC-F254, 6th JCT-VC meeting, Torino, Italy, July 2011, hereinafter referred to as “JCTVC-F254”).

To entropy code a block of coefficients (whether transformed and quantized or neither transformed nor quantized), a scanning process is typically performed such that the two-dimensional (2D) array of coefficients in the block is rearranged into an ordered one-dimensional (1D) array of coefficients, i.e., a vector, according to a particular scan order. Entropy coding is then applied to the vector of coefficients. Scanning the coefficients in the transform unit serializes the 2D array of coefficients for the entropy coder. A significance map may be generated to indicate the locations of significant (i.e., non-zero) coefficients. Scanning may be applied to the scan level of significant (i.e., non-zero) coefficients and/or the code sign of the significant coefficients.

In the HEVC standard, position information for significant transform coefficients (e.g., a significance map) is first coded for a transform block to indicate the position of the non-zero coefficients and the last non-zero coefficient in the scan order. For each coefficient in the reverse scan order, the significance map and level information (i.e., the absolute value and sign of the coefficient) are coded.

FIG2 is a conceptual diagram showing an example inverse scan order for coefficient-level coding. The H.264 standard defines zigzag scanning. The HEVC standard defines three different scan types: sub-block diagonal scanning, sub-block horizontal scanning, and sub-block vertical scanning. FIG2 illustrates inverse zigzag scanning pattern 29, inverse vertical scanning pattern 31, inverse horizontal scanning pattern 33, and inverse diagonal scanning pattern 35, each applied to an 8×8 sub-block of a transform block. Note that each of inverse diagonal scanning pattern 35, inverse zigzag scanning pattern 29, inverse vertical scanning pattern 31, and inverse horizontal scanning pattern 33 proceeds from the higher-frequency coefficients at the lower-right corner of the transform block to the lower-frequency coefficients at the upper-left corner of the transform block.

In the HEVC standard, the sub-block diagonal scan pattern 35, the sub-block horizontal scan pattern 33, and the sub-block vertical scan pattern 31 are applicable to 4×4 and 8×8 transform blocks. In the HEVC standard, the sub-block diagonal scan pattern 35 is also applicable to 16×16 and 32×32 transform blocks. In some examples, the sub-block diagonal scan pattern 35 is also applicable to 8×8 TUs. In sub-block based scanning, a 4×4 sub-block of a larger transform block is scanned before proceeding to another 4×4 sub-block within the larger transform block. In other examples, a "sub-block" may consist of a number of consecutively scanned coefficients according to the scan order used. For example, a "sub-block" may consist of 16 consecutively scanned coefficients along a diagonal scan order.

FIG3 is a conceptual diagram showing an example of coefficient-level coding for coefficient groups (CGs) based on an anti-diagonal scan order for subblocks. FIG3 illustrates an 8×8 transform block 38 composed of four 4×4 subblocks (37A, 37B, 37C, 37D). As shown in FIG3, the coefficients in subblock 37D are scanned before the coefficients in subblock 37C. The scan then proceeds from subblock 37C to subblock 37B and finally to subblock 37A. FIG3 depicts an anti-diagonal scan order within each subblock. In other examples, any scan order (e.g., horizontal, vertical, zigzag, etc.) may be used. In some examples, a forward scan order may be used within each subblock.

In the HEVC standard, coefficients may be grouped into fragments or subsets. For each subset, the significance map and level information (i.e., absolute value and sign) of the coefficients are coded. In this disclosure, a subset of coefficients may be referred to as a coefficient group (CG). A CG may be defined as n (e.g., n=16) consecutive coefficients of a transform block in scan order, which may correspond to a 4x4 sub-block. In one example, a subset consists of 16 consecutive coefficients along a scan order (e.g., forward or anti-diagonal, horizontal or vertical scan order) for a 4x4 transform block and an 8x8 transform block. For 16x16 and 32x32 transform blocks, a 4x4 sub-block of coefficients within the larger transform block is treated as a subset. In the example of FIG. 3 , each of the sub-blocks 37 may be a CG.

The symbols described below are coded to represent coefficient level information within the CG. In one example, all symbols are coded in reverse scan order. Each of the symbols can be coded in a separate scan of the CG according to the reverse scan order. The following symbols may be referred to as "flags." It should be noted that any of the "flags" discussed in this disclosure need not be limited to binary symbols, but may also be implemented as multi-bit syntax elements.

The significant_coeff_flag (also referred to as sigMapFlag) indicates the significance of each coefficient in the subset. Coefficients with an absolute value greater than zero are considered significant. As an example, a sigMapFlag value of 0 (i.e., not greater than zero) indicates that the coefficient is not significant, while a value of 1 (i.e., greater than zero) indicates that the coefficient is significant. This flag can generally be referred to as a significance flag. The coeff_sign_flag (also referred to as signFlag) indicates the sign information of any non-zero coefficient (i.e., a coefficient with a sigMapFlag of 1). For example, a value of zero for this flag indicates a positive sign, while a value of 1 indicates a negative sign.

coeff_abs_level_greater1_flag (also referred to as gr1Flag) indicates, for any non-zero coefficient (i.e., a coefficient with a sigMapFlag of 1 or where sigMapFlag is implicitly derived to be 1), whether the absolute value of the coefficient exceeds 1. As an example, a gr1Flag value of 0 indicates that the coefficient does not have an absolute value greater than 1, while a gr1Flag value of 1 indicates that the coefficient does have an absolute value greater than 1. This flag may generally be referred to as a greater than 1 flag.

coeff_abs_level_greater2_flag (also referred to as gr2Flag) indicates whether the absolute value of any coefficient with an absolute value greater than 1 (i.e., a coefficient with a gr1Flag of 1) exceeds 2. As an example, a gr2Flag value of 0 indicates that the coefficient does not have an absolute value greater than 2, while a gr2Flag value of 1 indicates that the coefficient does have an absolute value greater than 2. This flag can generally be referred to as a greater than 2 flag. sigMapFlag, gr1Flag, and gr2Flag can each be coded using CABAC.

The coeff_abs_level_remaining syntax element (also referred to as the levelRem syntax element) indicates the remaining absolute values of the coefficient level for any coefficient whose absolute value is greater than the value coded by the previous flag. In general, for the levelRem syntax element, the absolute value of the coefficient level minus three (i.e., abs(level)-3) is coded for each coefficient with an absolute value greater than 2 (i.e., a coefficient with a gr2Flag of 1). In some examples, when the maximum number of gr1Flags and/or gr2Flags for the current CG is reached, the levelRem syntax element may be used to code coefficient levels with an absolute value less than or equal to two. The levelRem syntax element may be coded using a code defined by the value of the Rice parameter (e.g., a Golomblais code or an Exponential Golomb code).

Figure 4 is a conceptual diagram showing an example inverse diagonal scan order of coefficient levels for coding CG 39. CG 39 can be a 4x4 transform block or can be a 4x4 sub-block in an 8x8, 16x16, or 32x32 transform block. The encoded symbols for the coefficients shown in Figure 4 scanned in inverse scan order are summarized in Table 1. In Table 1, scan_pos refers to the position of the coefficient along the inverse diagonal scan pattern of CG 39 shown in Figure 4. Scan_pos 15 is the first coefficient scanned and is located at the lower right corner of CG 39. The coefficient at scan_pos 15 has an absolute value of 0. Scan_pos 0 is the last coefficient scanned and is located at the upper left corner of CG 39. The quantized coefficient at scan_pos 0 has an absolute value of 10. In the case of the last 4x4 sub-block in a 4x4 transform block or a larger transform block, the first four sigMapFlags do not need to be coded because the position of the last non-zero coefficient is known. That is, coding of sigMapFlag may begin at the last non-zero coefficient (in this example, the coefficient at scan_pos 11).

Table 1. Coded symbols for coefficients of a coefficient group

Among these symbols, the bits of sigMapFlag, gr1Flag, and gr2Flag are coded by an adaptive context model (e.g., using CABAC). The binarized bits of signFlag and levelRem are coded via a bypass mode with a fixed equal probability model (e.g., by exponential Golomb or Golomb-Blaise codes).

As discussed above, the syntax element coeff_abs_level_remaining (i.e., levelRem) in the HEVC standard indicates the remaining absolute value of a coefficient level for a coefficient if the value exceeds the value coded in the previous scan pass for coefficient coding. This syntax element is coded in bypass mode to increase throughput. For small values, the HEVC standard utilizes Golomb coding, and for larger values, it switches to Exponential Golomb (Exp-Golomb) coding. The transition point between Golomb and Exp-Golomb codes is when the unary code length is equal to 4. The Rice parameter is an adjustable value used to select a codeword set from the family of Golomb codes.

For example, Golomb codes are a subset of Golomb codes and, given an adjustable Rice parameter m, represent values n >= 0 as the quotient q = floor (n/m) and the remainder r = n-q × m, where m is a power of 2. The quotient q is a prefix and has a unary code representation. The remainder r is a suffix and has a fixed-length representation. In Exp-Golomb codes, the code structure is similarly formed by a unary prefix followed by a fixed-length suffix, but after each bit in the unary code, the number of codewords in the suffix portion is doubled. Therefore, the codeword length of Exp-Golomb codes grows more slowly. Generally speaking, larger values of the Rice parameter result in slower code growth, which allows for greater efficiency when decoding large coefficient values. Additional details on Ricean parameters can be found in J. Sole, R. Joshi, M. Karczewicz, N. Nguyen, T. Ji, G. Clare, F. Henry, A. Duenas, “Transform Coefficient Coding in HEVC,” IEEE Transactions on Circuits and Systems for Video Transmission (Special Issue on HEVC), December 2012.

In the HEVC standard, the Rice parameter is set to an initial value equal to zero at the beginning of each coefficient group (CG), and during coding of the CG, the parameter is conditionally updated depending on the value of the Rice parameter and the absolute value of the coefficient level of the current coefficient being coded as follows:

If absCoeffLevel>3*2 ^cRiceParam , then cRiceParam＝min(cRiceParam+1,4)

Otherwise, cRiceParam = cRiceParam,

Where cRiceParam is the Rice parameter, absCoeffLevel is the absolute value of the coefficient level of the current coefficient, and min() is a function that selects the minimum value. The HEVC Rice parameter update scheme allows the binarization process to gradually adapt to the coefficient statistics when large absolute values are observed in the distribution.

As mentioned above, in the HEVC standard, after decoding the residual absolute values of the coefficients in the previous CG, the Rice parameters are reset to initial values of zero for the current CG in the transform block of the video data. Initializing the values of the Rice parameters to zero may not be optimal in the case of coefficient levels of decoding screen content or in the case of lossless decoding or lossy decoding in transform skip mode. The technology of the present invention can adaptively set the initial values of the Rice parameters at the beginning of each CG instead of always resetting the Rice parameters to zero. According to the technology, the initial values of the Rice parameters for the current CG can be set to be equal to non-zero values. In some instances, the initial values of the Rice parameters can be determined based on the statistics of the video data in order to provide better decoding performance, especially for screen content and lossless decoding.

In one example of the Rice parameter initialization scheme, the Rice parameters may not be reset after decoding the previous CG. In fact, the initial values of the Rice parameters for the current CG may be set to the same values as the Rice parameters obtained at the end of decoding the previous CG. As in the HEVC initialization scheme, the initial values may be set to 0 at the beginning of the current CG. However, unlike the HEVC scheme, it is not required to set the initial values of the Rice parameters to 0.

In another example of the Rice parameter initialization scheme, the initial value of the Rice parameter for the current CG can be set to a value based on the value of the Rice parameter after decoding the previous CG. In a specific example, the Rice parameter at the beginning of each CG can be initialized as follows.

cRiceParam=max(0,cRiceParam-1)

In the above example, the value of the Rice parameter is initialized for the current CG based on the selection of the largest of zero or the decrease in the value of the Rice parameter after decoding the previous CG. In some other examples, the upper limit of the initial value of the Rice parameter can be set as in the following example.

cRiceParam＝min(2,max(0,cRiceParam-1))

In this example, the initial value of the Rice parameter is set to an upper limit of no more than 2.

In some examples, a value other than 1 may be used to reduce the previous value of the Rice parameter (e.g., value n), including subtracting 2 from the previous value of the Rice parameter (i.e., n is equal to 2) instead of 1. For example, the reduction value used to initialize the Rice parameter may vary based on whether the current CG is included in a transform block to which a transform is applied. An example formula may be as follows.

In the above example, if the transform block is a transform skip block (i.e., transform_skip_flag=1), the value of the Rice parameter is initialized for the current CG based on the selection of the maximum of zero or the value of the Rice parameter after decoding the previous CG is reduced by 1. On the other hand, if the transform block is a transform skip block (i.e., transform_skip_flag=0), the value of the Rice parameter is initialized for the current CG based on the selection of the maximum of zero or the value of the Rice parameter after decoding the previous CG is reduced by 2.

Based on whether the transform is applied to the transform block, the example case of the Rice parameter initialization scheme described in the present invention may be applied to CGs included in all transform blocks, or may be applied to CGs. For example, in the case of a transform block with transform skipping or transform bypassing, the Rice parameter value for the current CG in the transform block may not be reset to 0, but for the transform block to which the transform has been applied, the Rice parameter for the current CG in the transform block may be reset to 0.

For example, in the case of lossy decoding in transform skip mode, reducing the value of the Rice parameter initialized for the current CG can be applied only to the CG in the transform skip block. The following may be an example formula.

In the above example, if the transform block is a transform skip block (i.e., transform_skip_flag=1), the value of the Rice parameter is initialized for the current CG based on the selection of the maximum of zero or the value of the Rice parameter decreased by 1 after decoding the previous CG. On the other hand, if the transform block is a transform skip block (i.e., transform_skip_flag=0), the value of the Rice parameter is initialized to zero for the current CG, as in the HEVC initialization scheme.

In another example of a Rice parameter initialization scheme, initial values of Rice parameters for a current CG may be determined based on coefficient-level statistics of previously decoded coefficients. The coefficient-level statistics may include absolute value statistics of the coefficient-level of previously coded coefficients or residual absolute value statistics of the coefficient-level. The initialization scheme may depend on previously coded coefficients in CGs included in the same transform block as the current CG and/or previously coded coefficients in CGs in different transform blocks preceding the transform block including the current CG.

In some cases, the statistics-based Rice parameter initialization scheme may depend on one or more of the transform block type, the transform block size, the position of the CG in the transform block, whether the transform block has an intra-frame prediction or inter-frame prediction slice type, the color components of the transform block, and the bit depth of the transform block. In addition, the statistics-based Rice parameter initialization scheme may depend on the residual absolute levels of previously decoded coefficients in the previous and current transform blocks. For example, the Rice parameter initialization scheme may depend on the last decoded absolute value of the coefficient level in the previous CG, or the last decoded residual absolute value of the coefficient level, on the sum or other statistics of the coefficient levels of the previously decoded coefficients in the previous CG, or more simply on the coefficient level of the first coefficient decoded in the previous CG.

Several examples of statistics collection for a statistics-based Rice parameter initialization scheme are described below. In this disclosure, the terms "statCoeff" and "m_sumCoeff" are used interchangeably to represent statistics, and the term "uiLevel" is used to represent the absolute value or residual absolute value of the coefficient level of a previously coded coefficient.

In one example, a statistic may be determined by computing an average or running average, or similar statistic, of the absolute values or remaining absolute values of coefficient levels of previously coded coefficients across a slice or coding unit (CU) of video data. This average or running average may be initialized at the beginning of a slice of video data in a manner similar to a context-adaptive binary arithmetic coding (CABAC) context and updated at each CG of the slice based on the current coded coefficient level in the CG. It should be understood that the use of a code defined by Rice parameters (e.g., a Golomblais code or an Exponential Golomb code) bypasses the remaining absolute values of the coded coefficient levels (i.e., the coeff_abs_level_remaining value). The description of CABAC and comparison with the initialization of the CABAC context is provided to aid understanding only.

In another example, the statistic may be determined by directly comparing the coefficient level of a given previously coded coefficient to the statistic value, and then determining whether to increase/decrease or maintain the statistic value based on the comparison. For example, the statistic may be determined according to the following conditional equation.

statCoeff+=(uiLevel==statCoeff)? 0:((uiLevel<statCoeff)?-1:1);

In the above equation, if the current coefficient level (uiLevel) exceeds the previous statistic (statCoeff), the value of statCoeff is increased, and if the current coefficient level is less than statCoeff, the value of statCoeff is decreased, or if the current coefficient level is equal to the previous statCoeff, the value of statCoeff remains unchanged. The value of statCoeff may be initialized to 0 at the same point in the coding process where the CABAC context is initialized, i.e., at the beginning of each slice of video data being coded.

In another example, the statistic can be determined by comparing the coefficient level of a given previously coded coefficient to a predefined function of the statistic, and then determining whether to increase or decrease the statistic based on the comparison. Furthermore, in this example, the statistic value (m_sumCoeff) can be reset to zero at the beginning of each slice of the video data. The predefined function of the statistic can be based on a first constant value of the statistic that is left-shifted and divided by a second constant value. An example of a function of the statistic (m_sumCoeff) is given below.

In the above pseudocode, a, b, c, d, e, f, g, and h are parameters, and << represents a left shift operation.

Below are some examples of the above equation using example values for the parameters a, b, c, d, e, f, g, and h. With a=3, d=2, a first constant value (h) equal to 1, a second constant value (f) equal to 4, and the remaining parameters set to 0, the function for the statistic (m_sumCoeff) is given below.

Under the condition that a=1, d=1, the first constant value (h) is equal to 1 and the second constant value (f) is equal to 4, and the remaining parameters are set equal to 0, the function of statistics (m_sumCoeff) is given as follows.

In some examples, the function of coefficient-level statistics may include a variable related to the total number of updates applied to the statistics since the statistics were initialized to zero at the beginning of the slice. An example of a function of statistics (m_sumCoeff) including a total count variable (m_total_counter) is given below.

In the pseudocode above, a, b, c, d, e, f, g, and h are parameters, << represents a left shift operation, and m_total_counter is incremented after each statistic update (m_sumCoeff) regardless of whether the statistic is increasing or decreasing.

The statistics collection described above may be performed for a statistics-based Rice parameter initialization scheme according to a predefined frequency. In one example, the statistics may be updated after each absolute value of a coefficient level in the coded CG, or after each remaining absolute value of a coefficient level in the coded CG. In another example, to limit the increase in complexity, the statistics may be updated only once per CG or per transform block. With this approach, there may be no need to update the statistics for each coded coefficient level, and in fact, the statistics may be updated once per transform block or per CG (in HEVC, once per 16 coefficients).

In some cases, statistics may be updated only when the first absolute value of a coefficient level in the CG is coded. In other cases, statistics may be updated only when the first remaining absolute value of a coefficient level in the CG is coded. The following pseudo-code illustrates this restriction with respect to the statistics collection equations described above, where the value of the first coded coefficient level (uiLevel) is directly compared to the value of the statistic (statCoeff).

In the above equations, the term "firstGolombCoeffinCG" is used to indicate whether the current decoded coefficient level is the first residual absolute value in the CG, the term "statCoeff" is used to indicate statistics, and the term "uiLevel" is used to indicate the first absolute value or the first residual absolute value of the coefficient level of the previously decoded coefficient.

Frequency limiting may be similarly applied to the statistics collection equations described above, where the value of the first coded coefficient level (uiLevel) is compared to a predefined function of statistics (statCoeff), as shown in the following pseudo-code.

In some cases, the statistics collection described above may be performed separately for each of a plurality of different categories of CGs defined based on the characteristics of the transform block containing the CG. In this case, the category of the current CG in the transform block may be determined based on the characteristics of the transform block, and the Rice parameters for the current CG may be initialized based on the statistics for the determined category. Several examples of segmenting statistics for a statistics-based Rice parameter initialization scheme are described below.

The characteristics of the transform block used for segmentation or classification statistics may include one or more of whether the transform block is a luma block type or a chroma block type, whether the transform block has an intra-frame prediction or inter-frame prediction slice type, the size of the transform block, and the position of the CG within the transform block. The position characteristic of the CG within the transform block may indicate whether the current CG is the upper left 4×4 sub-block in the transform block. In addition, the characteristics may include whether the transform block is decoded as a transform skip block, or whether the transform block is decoded as a transform quantization bypass block. Therefore, depending on one or more of the above characteristics, statistics can be saved separately. Separate statistics can be determined for each type or category of CG or transform block. Separate statistics can provide more accurate estimates for the statistics-based Rice parameter initialization scheme, but also require more storage resources.

As a first example, the following function may be used to determine the partition or category of a statistic based on a variable TYPE that depends on whether the transform block is a luma block, and whether the CG is the top left sub-block in the transform block.

TYPE=2*isLuma+(iSubSet>0);

According to the above function, the variable TYPE can have 4 values depending on whether the transform block is a luma block (isLuma=1) or a chroma block (isLuma=0), and whether the CG is the upper left sub-block (iSubSet==0) or not (iSubSet>0).

In another example, the partitioning or classification of the statistics depends on whether the transform block is a luma block and whether the transform block is coded in transform skip mode.

TYPE=2*isLuma+(isTransformSkip? 0:1);

According to this function, the variable TYPE can have 4 values depending on whether the transform block is a luma block (isLuma=1) or a chroma block (isLuma=0), and whether the transform block is coded in transform skip mode (isTransformSkip=1) or not (isTransformSkip=0).

In another example, the partitioning or classification of the statistics depends on whether the transform block is a luma block, whether the CG is the top left sub-block in the transform block, and whether the transform block is decoded in transform skip mode.

TYPE＝4*isLuma+2*(isTransformSkip?0:1)+(iSubSet>0);

According to this function, the variable TYPE can have 8 values depending on whether the transform block is a luma block (isLuma=1) or a chroma block (isLuma=0), whether the transform block is coded in transform skip mode (isTransformSkip=1) or not coded in the said mode (isTransformSkip=0), and whether the CG is the upper left sub-block (iSubSet==0) or not the upper left sub-block (iSubSet>0).

In another example, the partitioning or category of the statistics depends on whether the transform block is coded as transform-quantization bypass (ie, is losslessly coded, bypassing both the transform and quantization processes).

Using the TYPE variable defined according to one of the several examples described above, and combining it with the statistics collection equation described above (where the value of the first coded coefficient level (uiLevel) is directly compared to the value of the statistics (statCoeff)), and the statistics collection frequency restriction described above (where statistics are only updated when the first coefficient level in the coded CG (firstGolombCoeffinCG) is coded), the combined method will be as follows.

In the pseudo-code above, the term "statCoeff[TYPE]" represents the statistics collected for the CG category indicated by the variable TYPE.

Another example of a combined approach using the statistics collection equation described above, where the value of the first coded coefficient level (uiLevel) is compared to a predefined function of statistics (statCoeff), is as follows.

Furthermore, in the pseudo-code above, the term "statCoeff[TYPE]" represents statistics collected for the CG category indicated by the variable TYPE.

In some examples, a codec (i.e., a video encoder/decoder) may support multiple methods for determining the TYPE variable for a CG. For example, the codec may support two or more of the example functions described above for the TYPE variable. In this case, the codec may select one of the methods to determine how to partition the statistics for the statistics-based Rice parameter initialization scheme based on a decoded or derived indication. Example pseudocode for multiple method cases is given below.

Method 1:

TYPE=2*isLuma+(isTransformSkip? 0:1);

Method 2:

TYPE=(isTransformSkip?0:1);

In one case, a syntax element may be signaled in the coded bitstream to indicate which characteristic of a transform block is used to define different categories of CGs. The syntax element may include a flag included in one of a sequence parameter set (SPS) or a picture parameter set (PPS) for the residual video data. As an example, video encoder 20 may signal a flag (e.g., method_flag) in high-level syntax in the coded bitstream, and video decoder 30 may parse the coded bitstream to receive the method_flag. In this example, method_flag=0 indicates that method 1 is used to determine the category of statistics based on whether the transform block is a luma block and whether the transform block is coded in transform skip mode. method_flag=1 indicates that method 2 is used to determine the category of statistics based only on whether the transform block is coded in transform skip mode. If there are more than two methods supported by the codec, the syntax element may include an index value to indicate the selected method instead of a binary flag.

In another case, the codec may select one of the methods for defining different categories of CG based on the color format of the video data, in which case no additional syntax elements need to be signaled in the bitstream to indicate the method. For example, a method for determining the TYPE variable for the CG of the video data may be selected based on whether the YUV color format or the RGB color format is used to decode the video data. As an example, when the decoded video data is in the YUV color format, method 1 is used, which determines the category of statistics based on whether the transform block is a luma block and whether the transform block is decoded in a transform skip mode. When the decoded video data is in the RGB color format, method 2 is used, which determines the category of statistics based only on whether the transform block is decoded in a transform skip mode. In the case where video encoder 20 and video decoder 30 are capable of detecting the color format of the video data, a syntax element (e.g., a method flag or a method index) indicating the selected method is not signaled in the bitstream.

According to the techniques described in the present invention, a statistics-based Rice parameter initialization scheme determines initial values of Rice parameters for the current CG based on coefficient-level statistics collected for previously coded coefficients according to any combination of the examples described above. Several example techniques for mapping the values of the collected statistics to initial values of Rice parameters for the current CG are described below. The mapping of statistics to initial values of Rice parameters can be performed for each CG (e.g., a 4×4 transform block or a 4×4 sub-block of a transform block), or can be performed only once for each TU (e.g., at the beginning of the TU).

In one example, the statistical values can be mapped to initial Rice parameters according to a storage table. The input of the mapping table can be the statistical values, and the output of the table can be the initial values of the Rice parameters. In some cases, a reduced version of the statistics can be used as input to the mapping table. For example, if the maximum Rice parameter is 5 and the statistics are reduced to between 0 and 31, the mapping table can be given as follows.

g_golombTab[32]={0,0,0,0,1,1,2,2,3,3,3,3,3,3,4,4,4,4,4,4,4,4,4,4,5,5,5,5,5,5,5,5};

At the beginning of each CG, the above table can be used to determine the initial values of the Rice parameters based on the collected statistics. In other instances, different mapping tables can be used to determine the initial values of the Rice parameters based on the collected statistics.

Three main example methods of statistics collection are described above, where the notations "statCoeff" and "m_sumCoeff" are used interchangeably to represent statistics: (1) statistics collection that compares the value of the decoded coefficient level (uiLevel) directly with the statistic value (statCoeff), (2) statistics collection that compares the value of the decoded coefficient level (uiLevel) with a predefined function of the statistics (m_sumCoeff), and (3) statistics collection that compares the value of the decoded coefficient level (uiLevel) with a predefined function of the statistics (m_sumCoeff) that includes a total count variable (m_total_counter).

Using one of the specific example methods of statistics collection and the mapping table described above, Rice parameter initialization based on statistics collected for the CG category indicated by the variable TYPE can be performed as follows.

cRiceParam=g_golombTab[min(statCoeff[TYPE],31)];

In other examples, the statistical values can be mapped to the initial values of the Rice parameters according to the function that executes the mapping. In some cases, the mapping function can be a function of the collected statistics. Using the mapping function can avoid the additional storage consumption of the mapping table.

As an example, the Rice parameters may be initialized based on collected statistics of right-shifted constant values, as given below.

cRiceParam=(statCoeff[TYPE]>>R);

In the above equation, ">>" represents a right shift operation and R is a parameter. In some cases, it may be necessary to limit the maximum value of the Rice parameter to MAX_RICE, which may be an integer value greater than or equal to 4. In this case, the function may be given as follows.

cRiceParam=min(statCoeff[TYPE]>>R,MAX_RICE)

As another example, Rice parameters can be initialized based on a linear function of the collected statistics, as given below.

cRiceParam＝a*(statCoeff[TYPE]+b)/c+d

In the above equations, a, b, c, and d are parameter values.

As another example, the Rice parameters can be initialized based on a piecewise linear function of the collected statistics. Two specific examples of piecewise linear functions are given below.

cRiceParam=statCoeff[TYPE]<16? (statCoeff[TYPE]+1)/4:

(4+(statCoeff[TYPE]+40)/64);

cRiceParam=statCoeff[TYPE]<16? (statCoeff[TYPE])/4:

(4+(statCoeff[TYPE]+40)/64);

In another example, the Rice parameters may be initialized based on the statistics collection method described above, where the value of the coded coefficient level (uiLevel) is compared to a predefined function of statistics (m_sumCoeff). In this case, the values of the collected statistics are mapped to initial values of the Rice parameters according to the function of the statistics. In one example, the function of the statistics used to initialize the Rice parameters may be based on the selection of the minimum of the maximum value of the Rice parameters or the statistical value divided by a constant value.

An example of a mapping function is given below.

uiGoRiceParam=min(m_sumCoeff/DELAY,MAX_RICE_PARAM_INIT);

In the equation above, DELAY is a constant value, and MAX_RICE_PARAM_INIT is the maximum value of the Rice parameter. In some cases, the maximum value of the Rice parameter may be greater than or equal to 4. The constant value may be a user-defined parameter. An example of a mapping function with DELAY=4 is given below.

cRiceParam=Min(maxRicePara,statCoeff/4).

In the above examples, the term "statCoeff" rather than "m_sumCoeff" is used to represent the statistical value, and maxRicePara rather than MAX_RICE_PARAM_INIT is used to represent the maximum value of the Rice parameter.

In another example, the Rice parameters can be initialized based on the statistics collection method described above, where the value of the decoded coefficient level (uiLevel) is compared with a predefined function of statistics (m_sumCoeff) including a total count variable (m_total_counter). In this case, the values of the collected statistics are mapped to the initial values of the Rice parameters according to the function of the statistics. An example of a mapping function is given below.

If m_sumCoeff/m_total_counter>thres0,cRiceParam+＝k0

Elseif m_sumCoeff/m_total_counter<thres1,cRiceParam+＝k1

In the above equation, the term "thres0" represents a first threshold value, and "thres1" represents a second threshold value, and k0 and k1 are parameters. According to the function, if the statistical value divided by the total number of statistical updates is greater than thres0, the initial value of the Rice parameter is set to be equal to the previous value of the Rice parameter incremented by k0. If the statistical value divided by the total number of statistical updates is less than thres1, the initial value of the Rice parameter is set to be equal to the previous value of the Rice parameter incremented by k1.

In some instances, the initial Rice parameters can be reduced to between a minimum and a maximum value. In this case, an additional reduction function is added to the mapping function for initializing the Rice parameters. An example of a reduced mapping function is given below.

cRiceParam＝Clip(MIN_RICE,cRiceParam,MAX_RICE)

In the above equation, MIN_RICE is the minimum value of the Rice parameter, and MAX_RICE is the maximum value of the Rice parameter. In one example, the value of MIN_RICE may be equal to 0, and the value of MAX_RICE may be an integer value greater than or equal to 4. The values of MIN_RICE and MAX_RICE may depend on side information, such as one or more of bit depth, layout file, color format, coding mode (i.e., lossless coding or lossy coding), and other types of side information.

As described above, in HEVC, the value of the Rice parameter may be updated after coding each residual absolute value of the coefficient level in the CG. In some examples, similar clipping may be applied to the updated value of the Rice parameter. The range (i.e., minimum and maximum values) used for the update process may be different than that used for the initialization process, or the range may be the same as that used for the initialization process. Similarly, the range may depend on one or more of the bit depth, layout file, color format, coding mode (i.e., lossless coding or lossy coding), and other side information.

Four additional examples of mapping statistics to initial values of Ricean parameters are given below.

In a first example, two statistics called m_sumCoeff and m_sumCoeff2 are used.The value of m_sumCoeff may be derived using the statistics collection method described above as follows, where the value of a coded coefficient level (uiLevel) is compared to a predefined function of the statistic (m_sumCoeff).

The value of m_sumCoeff2 may be derived using the statistics collection method described above as follows, where the value of the coded coefficient level (uiLevel) is directly compared to the statistical value (m_sumCoeff2).

m_sumCoeff2+=(uiLevel==m_sumCoeff2)? 0:

(uiLevel<m_sumCoeff2?-1:1);

In this first example, the mapping of statistics for transform coded blocks to initial values of Rice parameters may be different compared to transform skip blocks. In the case of transform skip blocks, the initialization function may be given as follows.

uiGoRiceParam=min(m_sumCoeff/DELAY,MAX_RICE_PARAM_INIT);

In the case of transforming a coded block, the initialization function may be given as follows.

In the pseudo code above, "Th" is a threshold value. An example value of Th may be (1<<uiGoRiceParam).

In a second example, a statistic called m_sumCoeff can be derived using the statistics collection method described above, where the value of the coded coefficient level (uiLevel) is compared to a predefined function of the statistic (m_sumCoeff). In this second example, the mapping of statistics to initial values of Rice parameters may be different for transform-coded blocks compared to transform-skip blocks. In the case of transform-skip blocks, the initialization function may be given as follows.

uiGoRiceParam=min(m_sumCoeff/DELAY,MAX_RICE_PARAM_INIT);

In the case of transforming a coded block, the initialization function may be given as follows.

In the pseudo code above, "Th" is a threshold value. An example value of Th may be MAX_RICE_PARAM_INIT/2-2.

In a third example, a statistic called m_sumCoeff can be derived using the statistics collection method described above, where the value of the coded coefficient level (uiLevel) is compared with a predefined function of the statistic (m_sumCoeff). In this third example, the mapping of the statistics to the initial values of the Rice parameters can be given as follows.

In the pseudo code above, "Th" is a threshold value and "d" is a parameter.

In a fourth example, a statistic called m_sumCoeff can be derived using the statistics collection method described above, where the value of the coded coefficient level (uiLevel) is compared with a predefined function of the statistic (m_sumCoeff). In this fourth example, the mapping of the statistics to the initial values of the Rice parameters can be given as follows.

In the pseudo code above, "Th" is the threshold value and "d", DELAY0 and DELAY1 are parameters. In this fourth example, when the statistical value is equal to the threshold value (Th), the initial value of the Rice parameter (uiGoRiceParam) derived using any of the above equations is the same.

In some examples, the mapping functions for determining the initial values and threshold values of the Rice parameters may be fixed and known to both the video encoder 20 and the video decoder 30. In other examples, the mapping functions for determining the initial values and threshold values of the Rice parameters may be adaptively determined based on side information. The side information may be independently derived by each of the video encoder 20 and the video decoder 30, or the side information may be signaled from the video encoder 20 to the video decoder 30 in an SPS or PPS using high-level syntax, or some combination of derivation and signaling may be used to determine the side information. For example, the side information may include one or more of the following: frame size, frame type, CU size, TU size, TU type (e.g., transform skip mode or transform mode), color component, intra or inter prediction mode, quantization parameter (QP), bit depth, color format (e.g., 444/422/420), significant coefficient flag (both number and distribution), flag greater than 1 (i.e., more than 1) (both number and distribution), and flag greater than 2 (i.e., more than 2) (both number and distribution). The information may also be explicitly signaled.

After initializing the Rice parameters for the current CG, the initial values of the Rice parameters may be updated using a code (e.g., a Golomblais code or an exponential Golomb code) defined by the Rice parameters after coding the remaining absolute values of the coefficient levels of at least one coefficient in the current CG. Statistics for a statistics-based Rice parameter initialization scheme may be determined based on one or more coefficient levels in the current CG before or after updating the values of the Rice parameters.

In some examples, statistics may be based on a comparison or calculation of the coded coefficient levels in the current CG and the current values of the Rice parameters used for the current CG. An example of statistics based on coefficient levels and Rice parameter values is given below.

In the pseudo code above, "UndershootCnt", "OvershootCnt" and "TotalCnt" represent the statistics that will be used to initialize the Rice parameters. In addition, "uiGoRiceParam" represents the value of the Rice parameter, "s" is the parameter, and absCoeff[idx] is the absolute value of the coefficient level of the coefficient at index idx.

In some examples, statistics can be determined before the value of the Rice parameter is updated. In this case, the unupdated value of the Rice parameter is used in the comparison or calculation for determining the statistics. An example combination of determining statistics based on the coefficient level and the unupdated value of the Rice parameter, and then updating the value of the Rice parameter is given below.

In the pseudo code above, "MAX_RICE_PARAM" represents the maximum value of the Rice parameter, and min<UInt>() is a function that selects the minimum value. In some instances, the maximum value of the Rice parameter may be equal to an integer value of at least 4. In other instances, statistics may be determined after updating the value of the Rice parameter. In this case, the updated value of the Rice parameter is used in the comparison or calculation for determining the statistics.

In some examples, the update of the statistics used for the statistics-based Rice parameter initialization scheme described above can be integrated with the update of the Rice parameters. As discussed above, in HEVC, the initial values of the Rice parameters can be conditionally updated based on the initial values of the Rice parameters and the absolute values of the coefficient levels of the coefficients being decoded in the current CG. After decoding each remaining absolute level of the coefficients in the current CG, the values of the Rice parameters can continue to be conditionally updated. The HEVC conditional update scheme is given as follows.

In the above pseudocode, "uiGoRiceParam" represents the value of the Rice parameter, absCoeff[idx] is the absolute value of the coefficient level of the coefficient at index idx, "MAX_RICE_PARAM" represents the maximum value of the Rice parameter, and min<UInt>() is a function that selects the minimum value.

In some examples, the update of the statistics used to initialize the values of Rice parameters at the beginning of the subsequent CG can be integrated with the update of the Rice parameters based on the coefficient level of the current CG. The following is an example of the integrated update of statistics and Rice parameter values.

In the pseudo code above, "UndershootCnt++" indicates the update of statistics.

In some examples, based on some or all of the parameters used for statistics, an updated value of a Rice parameter for the next decoding unit (e.g., a 4×4 CG, TU, or CU) (or a delta value compared to a predicted value (e.g., a current value of a Rice parameter)) is calculated. For example, in one example, the higher the value of OvershootCnt, the smaller the value of the updated Rice parameter. In another example, the higher the value of UndershootCnt, the larger the value of the updated Rice parameter.

The following describes several examples of initializing the Rice parameters for each CG based on the bit depth or another characteristic of the video data. In these examples, the initialization of the Rice parameters for each CG does not depend on the collected statistics. As in the statistics-based Rice parameter initialization scheme described above, Rice parameter initialization can be performed for each CG (e.g., a 4×4 transform block or a 4×4 sub-block of a transform block), or can be performed only once for each TU (e.g., at the beginning of the TU).

In one example, the initial value of the Rice parameter may be based on the bit depth of the video data instead. The following is an example where the initial value cRiceParam of the Rice parameter for each CG is a function of the bit depth of the current component of the CG (e.g., one of the luminance component or the chrominance component).

cRiceParam=max(0,bitDepth-n)

In the above equation, bitDepth is the bit depth of the current component, and n is a parameter. In one example, the parameter n can be set to 9.

In another example, the initial value cRiceParam of the Rice parameter for each CG may depend on the component type of the CG (e.g., luma or chroma). In another example, the initial value cRiceParam of the Rice parameter for each CG may depend on whether the current block is coded in transform quantization bypass mode (in which both transform and quantization are skipped). In another example, the initial value cRiceParam of the Rice parameter for each CG may depend on whether the current block is coded in transform skip mode (in which transform is skipped but quantization may be applied). In another example, the initial value cRiceParam of the Rice parameter for each CG may depend on the quantization parameter (QP) used for the TU. In another example, the initial value cRiceParam of the Rice parameter for each CG may depend on the color component and/or color space of the video data.

In another example, the initial value cRiceParam of the Rice parameter for each CG may depend on one or more of the CU size, the TU size, the frame resolution, the frame rate, the prediction mode (e.g., intra, inter, intra BC (intra block copy)), the frame type, and the mode and weight of the horizontal component residual prediction. Examples of the mode and weight of the horizontal component residual prediction are described in U.S. Provisional Application No. 61/846,581, filed on July 15, 2013, U.S. Provisional Application No. 61/847,839, filed on July 18, 2013, U.S. Provisional Application No. 61/826,396, filed on May 22, 2013, and U.S. Provisional Application No. 61/838,152, filed on June 21, 2013.

The following uses HEVC as an example to provide some background information about the coeff_abs_level_greater1 flag and the coeff_abs_level_greater2 flag. In HEVC, for a CG (e.g., a 4x4 sub-block), a significance map is first coded to indicate the positions of coefficients with non-zero coefficient levels. Then, for positions with significant coefficients, the coeff_abs_level_greater1 flag may be coded to indicate whether the absolute value of the coefficient exceeds 1. For positions where coeff_abs_level_greater1=1, the coeff_abs_level_greater2 flag may be coded to indicate whether the absolute value of the coefficient exceeds 2.

In some examples, the initial value cRiceParam of the Rice parameter for each CG may depend on the number of previously coded coefficient levels with valid (i.e., non-zero) values (the number may be indicated by coding the coeff_abs_level_greater1 flag equal to 0 or 1), and/or the number of previously coded coefficient levels with values greater than 1 (the number may be indicated by coding the coeff_abs_level_greater2 flag equal to 0 or 1). In other examples, the initial value cRiceParam of the Rice parameter for each CG may depend on the number of previously coded coefficient levels with values greater than 1 (the number may be indicated by coding coeff_abs_level_greater1=1), and/or the number of previously coded coefficient levels with values greater than 2 (the number may be indicated by coding coeff_abs_level_greater2=1).

In another example, the initial value cRiceParam of the Rice parameter for each CG may depend on any combination of the above examples. Two detailed examples of determining the initial value of the Rice parameter based on the bit depth of the video data are given below.

Example 1:

offset＝(iQP+6*(BD-12)+12)/6

In the pseudo code above, iQP is the QP value, and BD is the bit depth of the video data. In this first example, when the CG is included in a transform block coded in either transform skip mode or transform quantization bypass mode, the initial value of the Rice parameter is set based on the first equation, and when the CG is included in a transform-coded block, the initial value of the Rice parameter is set based on the second equation. In either case, the initial value is determined based on the bit depth (BD) of the video data.

Example 2:

offset=8-Gr1

In the pseudo code above, Gr1 is the target number of coded coeff_abs_level_greater1 flags, and BD is the bit depth of the video data. In this second example, when CG is included in a transform block coded in either transform skip mode or transform quantization bypass mode, the initial value of the Rice parameter is set based on the first equation, and when CG is included in a transform-coded block, the initial value of the Rice parameter is set based on the second equation. In either case, the initial value is determined based on the bit depth (BD) of the video data.

The following describes an example of a hybrid scheme that combines Rice parameter initialization for each CG according to any combination of the examples described above with explicit signaling of offset values. For example, the offset used to initialize the Rice parameters at the beginning of the CG can be decomposed into the sum of two parts: (1) a constant offset and (2) an adaptive offset. The adaptive offset can be derived using one or more of the example techniques described in this disclosure. The constant offset can be signaled from the video encoder 20 to the video decoder 30 in the bitstream.

As another example, to combine explicit Rice parameter signaling with the initial Rice parameter derivation process described in this disclosure, the determination of whether to signal a Rice parameter may depend on the initial Rice parameter value derived at the video encoder 20. For example, if the derived initial Rice parameter value is similar to a predefined value (e.g., the difference between the derived value and the threshold value is less than the threshold value), the Rice parameter may not be signaled to the video decoder 30, but the Rice parameter may actually be derived as described above. In this case, either the derived value or the predefined value may be used as the initial value of the Rice parameter. Otherwise, the initial value of the Rice parameter may be signaled, or the difference between the initial value and the predefined value of the Rice parameter may be signaled. Some additional examples of signaling techniques can be found in U.S. Provisional Application No. 61/870,120, filed August 26, 2013, U.S. Provisional Application No. 61/880,616, filed September 20, 2013, and U.S. Provisional Application No. 61/889,654, filed October 11, 2013.

The following describes several additional examples and considerations for the above-described examples of Rice parameter initialization schemes. Although the techniques of this disclosure may be described essentially individually and/or as part of a specific combination with other techniques, any two or more of the techniques described in this disclosure may describe such a combination. Additionally, it may be possible to implement any of the techniques described in this disclosure separately.

For lossy decoding conditions, the statistics-based Rice parameter initialization scheme can be applied only to transform skip blocks, as shown in the following conditional equation. On the other hand, when the transform has been applied to the transform block, the Rice parameters are initialized to zero according to the HEVC initialization scheme.

cRiceParam=isTransformSkip? 0:g_golombTab[min(statCoeff[TYPE],31)];

In other examples, the statistics-based Rice parameter initialization scheme may be applied only to transform skip blocks, but when the transform has been applied to the block, it may not be necessary to automatically initialize the Rice parameters to zero, as shown in the following equation.

cRiceParam=isTransformSkip? max(cRiceParam-1,0)：

g_golombTab[min(statCoeff[TYPE],31)];

In some examples, the HEVC Rice parameter update scheme executed in the CG may be disabled. In fact, the initial value of the Rice parameter may be determined according to any combination of the example techniques described above, and then the initial value of the Rice parameter may be used in the entire block. In other examples, different Rice parameter update schemes may be executed in the CG. In some cases, one or more of the example techniques described above may be used in the entire block in a given order or as an integrated method to perform Rice parameter updates and statistical updates. Again, in some cases, one or more of the example techniques described above may be used in the entire block to perform bit depth dependency initialization of Rice parameters.

Although the example methods for determining the initial values of the Rice parameters are described above separately, the technology of the present invention is not limited to this case. In general, various combinations of the example techniques described above may be possible. The example methods described above may also be implemented separately. In addition, all or a combination of the methods described above may be applied to CGs within all transform blocks, or only to transform blocks that skip or bypass transforms.

FIG5 is a block diagram illustrating an example of a video encoder 20 that may implement the techniques for encoding coefficient levels described in this disclosure. For illustrative purposes, the video encoder 20 will be described in the context of HEVC coding, but this disclosure is not limited with respect to other coding standards or methods that may require scan transform coefficients. The video encoder 20 may perform intra-coding and inter-coding of CUs within a video frame. Intra-coding relies on spatial prediction to reduce or remove spatial redundancy in video data within a given video frame. Inter-coding relies on temporal prediction to reduce or remove temporal redundancy between a current frame and a previously coded frame of a video sequence. Intra-mode (I-mode) may refer to any of several spatially-based video compression modes. Inter-mode, such as unidirectional prediction (P-mode) or bidirectional prediction (B-mode), may refer to any of several temporally-based video compression modes.

As shown in FIG5 , video encoder 20 receives a current video block within a video frame to be encoded. In the example of FIG5 , video encoder 20 includes a mode select unit 40, a video data memory 41, a motion compensation unit 44, a motion estimation unit 42, an intra-prediction processing unit 46, a decoded picture buffer (DPB) 64, a summer 50, a transform processing unit 52, a quantization unit 54, and an entropy encoding unit 56. The transform processing unit 52 illustrated in FIG5 is a unit that applies an actual transform or a combination of transforms to a block of residual data and should not be confused with a block of transform coefficients (which may also be referred to as a transform unit (TU) for a CU). To perform video block reconstruction, video encoder 20 also includes an inverse quantization unit 58, an inverse transform processing unit 60, and a summer 62. A deblocking filter (not shown in FIG5 ) may also be included to filter block boundaries to remove blockiness artifacts from the reconstructed video. If necessary, the deblocking filter will typically filter the output of summer 62.

Video data memory 41 may store video data to be encoded by components of video encoder 20. The video data stored in video data memory 41 may be obtained, for example, from video source 18. Decoded picture buffer 64 may be a reference picture memory that stores reference video data for use by video encoder 20 when encoding video data (e.g., in intra- or inter-coding modes). Video data memory 41 and decoded picture buffer 64 may be formed from any of a variety of memory devices, such as dynamic random access memory (DRAM), including synchronous DRAM (SDRAM), magnetoresistive RAM (MRAM), resistive RAM (RRAM), or other types of memory devices. Video data memory 41 and decoded picture buffer 64 may be provided by the same memory device or separate memory devices. In various examples, video data memory 41 may be on-chip with other components of video encoder 20, or off-chip relative to those components.

During the encoding process, video encoder 20 receives a video frame or slice to be coded. The frame or slice may be divided into multiple video blocks, such as largest coding units (LCUs). Motion estimation unit 42 and motion compensation unit 44 perform inter-predictive coding of the received video block relative to one or more blocks in one or more reference frames to provide temporal compression. Intra-prediction processing unit 46 may perform intra-predictive coding of the received video block relative to one or more neighboring blocks in the same frame or slice as the block to be coded to provide spatial compression.

Mode select unit 40 may select one of the coding modes (intra or inter) (e.g., based on the error (i.e., distortion) results of each mode) and provide the resulting intra- or inter-predicted block (e.g., prediction unit (PU)) to summer 50 to generate residual block data and to summer 62 to reconstruct the encoded block for the reference picture. Summer 62 combines the predicted block with the inverse quantized, inverse transformed data for that block from inverse transform unit 60 to reconstruct the encoded block, as described in more detail below. Some video frames may be designated as I-frames, in which all blocks in the I-frame are encoded in intra-prediction mode. In some cases, intra-prediction processing unit 46 may perform intra-prediction encoding of a block in a P or B frame, e.g., when a motion search performed by motion estimation unit 42 does not result in a sufficient prediction for the block.

Motion estimation unit 42 and motion compensation unit 44 may be highly integrated but are described separately for conceptual purposes. Motion estimation (or motion search) is the process of generating motion vectors, which estimate the motion of video blocks. For example, a motion vector may indicate the displacement of a prediction unit in a current frame relative to a reference sample of a reference frame. Motion estimation unit 42 calculates the motion vector for a prediction unit in an inter-coded frame by comparing the prediction unit with reference samples of a reference picture stored in decoded picture buffer 64. The reference sample may be a block found to closely match a portion of a CU including a PU coded based on pixel differences, which may be determined by sum of absolute differences (SAD), sum of squared differences (SSD), or other disparity metrics. The reference sample may occur anywhere within a reference frame or reference slice and does not necessarily occur at block (e.g., coding unit) boundaries of the reference frame or slice. In some examples, the reference sample may occur at a fractional pixel location.

Motion estimation unit 42 sends the calculated motion vector to entropy encoding unit 56 and motion compensation unit 44. The portion of the reference frame identified by the motion vector may be referred to as a reference sample. Motion compensation unit 44 may calculate prediction values for the prediction unit of the current CU, for example, by retrieving the reference samples identified by the motion vector of the PU.

As an alternative to the inter-prediction performed by motion estimation unit 42 and motion compensation unit 44, intra-prediction processing unit 46 may perform intra-prediction on the received block. Intra-prediction processing unit 46 may predict the received block relative to neighboring, previously coded blocks (e.g., blocks above, to the right, above, to the left, or to the left of the current block), assuming a left-to-right, top-to-bottom coding order for the blocks. Intra-prediction processing unit 46 may be configured with a variety of different intra-prediction modes. For example, based on the size of the CU being encoded, intra-prediction processing unit 46 may be configured with a number of directional prediction modes, e.g., thirty-three directional prediction modes.

Intra-prediction processing unit 46 may select an intra-prediction mode by, for example, calculating error values for various intra-prediction modes and selecting the mode that produces the lowest error value. Directional prediction modes may include functionality for combining values of spatially neighboring pixels and applying the combined value to one or more pixel positions in a PU. Once values for all pixel positions in a PU have been calculated, intra-prediction processing unit 46 may calculate an error value for the prediction mode based on pixel differences between the PU and the received block to be encoded. Intra-prediction processing unit 46 may continue testing intra-prediction modes until an intra-prediction mode that produces an acceptable error value is found. Intra-prediction processing unit 46 may then send the PU to summer 50.

Video encoder 20 forms a residual block by subtracting the prediction data calculated by motion compensation unit 44 or intra-prediction processing unit 46 from the original video block being coded. Summer 50 represents the component(s) that perform this subtraction operation. The residual block may correspond to a two-dimensional matrix of pixel difference values, where the number of values in the residual block is the same as the number of pixels in the PU corresponding to the residual block. The values in the residual block may correspond to the differences (i.e., errors) between the values of the co-located pixels in the PU and the original block to be coded. The differences may be chrominance differences or luma differences, depending on the type of block being coded.

Transform processing unit 52 may form one or more transform units (TUs) from the residual block. Transform processing unit 52 selects a transform from a plurality of transforms. For example, transform processing unit 52 may select one of a discrete cosine transform (DCT), a discrete sine transform (DST), an integer transform, a K-L transform, or another transform to generate transform coefficients. The transform may be selected based on one or more coding characteristics, such as block size, coding mode, or the like. Transform processing unit 52 then applies the selected transform to the TU, generating a video block comprising a two-dimensional array of transform coefficients. Transform processing unit 52 may send the resulting transform coefficients to quantization unit 54. Quantization unit 54 then quantizes the transform coefficients.

Next, entropy encoding unit 56 may perform a scan on the coefficients in the matrix according to a scan pattern. In the case of lossy coding, the coefficients may be quantized transform coefficients. In the case of lossless coding or lossy coding with transform skipping or bypassing, the coefficients may be coefficients that have not yet been transformed or quantized. This disclosure describes entropy encoding unit 56 as performing a scan. However, it should be understood that in other examples, other processing units, such as quantization unit 54, may perform the scan.

Once the transform coefficients are scanned into a one-dimensional array, entropy encoding unit 56 may apply entropy coding (e.g., CABAC, syntax-based context-adaptive binary arithmetic coding (SBAC), probability interval partitioning entropy (PIPE), or another entropy coding method) to the coefficients. Entropy encoding unit 56 may be configured to code the coefficients according to the techniques of this disclosure. In the example of CABAC, entropy encoding unit 56 may encode the coefficients using either a normal coding mode or a bypass mode. To perform CABAC, entropy encoding unit 56 may select a context model to apply a certain context to encode the symbols to be transmitted. For example, the context may relate to whether neighboring values are non-zero. Entropy encoding unit 56 may select a context model for encoding these symbols based on, for example, the intra-prediction direction for the intra-prediction mode, the scan position of the coefficients corresponding to the syntax element, the block type and/or transform type, and other factors used for context model selection.

Entropy encoding unit 56 encodes the coefficient levels of the residual video data into a bitstream for transmission to a video decoder or storage device. In the case of lossless video coding or lossy video coding with transform skip or bypass, the coefficients to be encoded may have coefficient levels with large absolute values. When the coefficients represent screen content (which may include graphics and text areas), the content may not be well predicted, resulting in large absolute values of the coefficient levels of the coefficients to be encoded.

The entropy coding unit 56 encodes the residual absolute value of the coefficient level (e.g., coeff_abs_level_remaining or levelRem) of at least one coefficient in the current coefficient group (CG) in a bypass mode of CABAC or another entropy decoding engine using a code defined by the Rice parameter. According to the technology of the present invention, the entropy coding unit 56 is configured to determine the initial value of the Rice parameter for the current CG based on the statistics of the coefficient level of the previously encoded coefficient. The statistics may be the statistics of the absolute value of the coefficient level or the residual absolute value of the coefficient level of the previously decoded coefficient. The statistics-based Rice parameter initialization scheme described in this invention allows the Rice parameter to adapt quickly and efficiently to large coefficient values, which may occur in blocks of screen content and blocks with transform skipping or bypassing.

In one example, entropy encoding unit 56 may be configured to determine the statistic by comparing the coefficient level for a given previously encoded coefficient to a function of the statistic, and then determining whether to increase or decrease the statistic value based on the comparison. The statistic value may be initialized to zero at the beginning of each slice of video data, and entropy encoding unit 56 may update the statistic once for each CG of the slice. In some examples, entropy encoding unit 56 may be configured to determine a separate statistic for each of a plurality of different categories of CGs. The categories may be defined based on characteristics of the transform block that includes the CG, such as whether the transform block is a luma block and whether the transform block is a transform skip block.

At the beginning of the current CG, the entropy coding unit 56 is configured to map the statistical value to the initial value of the Rice parameter for the current CG. In one example, the entropy coding unit 56 may map the statistical value to the initial Rice parameter value according to a function of a statistic selected based on the minimum of the maximum value of the Rice parameter or the statistical value divided by a constant value. Example equations representing the statistical collection, statistical segmentation, and statistical mapping processes for the Rice parameter initialization scheme are described in more detail above.

Following entropy coding by entropy encoding unit 56, the resulting encoded video may be transmitted to another device, such as video decoder 30, or archived for later transmission or retrieval. In some cases, entropy encoding unit 56 or another unit of video encoder 20 may be configured to perform other coding functions in addition to entropy coding.

Inverse quantization unit 58 and inverse transform processing unit 60 apply inverse quantization and inverse transform, respectively, to reconstruct the residual block in the pixel domain, e.g., for later use as a reference block. Motion compensation unit 44 may calculate a reference block by adding the residual block to a predictive block of one of the reference pictures in decoded picture buffer 64. Motion compensation unit 44 may also apply one or more interpolation filters to the reconstructed residual block to calculate sub-integer pixel values for motion estimation. Summer 62 adds the reconstructed residual block to the motion compensated prediction block produced by motion compensation unit 44 to produce a reconstructed video block for storage in decoded picture buffer 64. The reconstructed video block may be used by motion estimation unit 42 and motion compensation unit 44 as a reference block for inter-coding a block in a subsequent video frame.

FIG6 is a block diagram illustrating an example of a video decoder 30 that may implement the techniques for decoding coefficient layers described in this disclosure. In the example of FIG6 , video decoder 30 includes an entropy decoding unit 70, a video data memory 71, a motion compensation unit 72, an intra-prediction processing unit 74, an inverse quantization unit 76, an inverse transform processing unit 78, a decoded picture buffer (DPB) 82, and a summer 80. In some examples, video decoder 30 may perform a decoding pass that is generally reciprocal to the encoding pass described with respect to video encoder 20 (see FIG5 ).

Video data memory 71 may store video data, such as an encoded video bitstream, to be decoded by components of video decoder 30. The video data stored in video data memory 71 may be obtained, for example, from computer-readable medium 16 (e.g., from a local video source, such as a camera) via wired or wireless network communication of video data, or by accessing physical data storage media. Video data memory 71 may form a coded picture buffer (CPB) that stores encoded video data from the encoded video bitstream. Decoded picture buffer 82 may be a reference picture memory that stores reference video data for use by video decoder 30 when decoding video data (e.g., in intra- or inter-coding modes). Video data memory 71 and decoded picture buffer 82 may be formed from any of a variety of memory devices, such as dynamic random access memory (DRAM), including synchronous DRAM (SDRAM), magnetoresistive RAM (MRAM), resistive RAM (RRAM), or other types of memory devices. Video data memory 71 and decoded picture buffer 82 may be provided by the same memory device or separate memory devices. In various examples, video data memory 71 may be on-chip with other components of video decoder 30 , or off-chip relative to those components.

Entropy decoding unit 70 performs an entropy decoding process on the encoded bitstream to retrieve a one-dimensional array of residual video data coefficients. The entropy decoding process used depends on the entropy coding (e.g., CABAC) used by video encoder 20. The entropy coding process used by the encoder may be signaled in the encoded bitstream, or the entropy coding process may be a predetermined process. Entropy decoding unit 70 may be configured to decode the coefficients according to the techniques of this disclosure. In the example of CABAC, entropy decoding unit 70 may decode the coefficients using either a normal coding mode or a bypass mode.

In some examples, entropy decoding unit 70 may scan the received values using a scan that mirrors the scan pattern used by entropy encoding unit 56 of video encoder 20. Although the coefficient scanning may be performed in inverse quantization unit 76, for purposes of illustration, the scanning will be described as being performed by entropy decoding unit 70. Additionally, although shown as separate functional units for ease of illustration, the structure and functionality of entropy decoding unit 70, inverse quantization unit 76, and other units of video decoder 30 may depict such a high degree of integration.

Entropy decoding unit 70 decodes coefficient levels of residual video data coefficients from the bitstream in a reciprocal manner to video encoder 20. In the case of lossy video coding, the coefficients to be decoded may be quantized transform coefficients. In the case of lossless video coding or lossy video coding with transform skipping or bypassing, the coefficients to be decoded may be encoded pixel values and have coefficient levels (i.e., pixel values) with large absolute values. When the coefficients represent screen content (which may include graphics and text areas), the content may not be well predicted, resulting in large absolute values of the coefficient levels of the coefficients to be decoded.

The entropy decoding unit 70 decodes the residual absolute value of the coefficient level of at least one coefficient in a coefficient group (CG) using a code defined by a Rice parameter. According to the technology of the present invention, the entropy decoding unit 70 is configured to determine the initial value of the Rice parameter for the current CG based on the statistics of the coefficient level of the previously decoded coefficients. The statistics can be the absolute value of the coefficient level of the previously decoded coefficients or the statistics of the residual absolute value of the coefficient level. The statistics-based Rice parameter initialization scheme described in this invention allows the Rice parameters to adapt quickly and efficiently to large coefficient values, which may occur in blocks of screen content and blocks with transform skipping or bypassing.

In one example, entropy decoding unit 70 may be configured to determine the statistic by comparing the coefficient level for a given previously decoded coefficient to a function of the statistic, and then determining whether to increase or decrease the statistic value based on the comparison. The statistic value may be initialized to zero at the beginning of each slice of video data, and entropy decoding unit 70 may update the statistic once per CG of each slice. In some examples, entropy decoding unit 70 may be configured to determine a separate statistic for each of a plurality of different categories of CGs. The categories may be defined based on characteristics of the transform block that includes the CG, such as whether the transform block is a luma block and whether the transform block is a transform skip block.

At the beginning of the current CG, the entropy decoding unit 70 is configured to map the statistical value to the initial value of the Rice parameter for the current CG. In one example, the entropy decoding unit 70 may map the statistical value to the initial Rice parameter value according to a function of statistics based on the minimum of the selected maximum value of the Rice parameter or the statistical value divided by a constant value. Example equations representing the statistical collection, statistical segmentation, and statistical mapping processes for the Rice parameter initialization scheme are described in more detail above.

Inverse quantization unit 76 inverse quantizes (i.e., dequantizes) the quantized transform coefficients provided in the bitstream and decoded by entropy decoding unit 70. The inverse quantization process may include a conventional process, such as a process similar to that proposed for HEVC or a process defined by the H.264 decoding standard. The inverse quantization process may include using a quantization parameter QP calculated by video encoder 20 for a CU to determine the degree of quantization and, likewise, the degree of inverse quantization that should be applied. Inverse quantization unit 76 may inverse quantize the transform coefficients either before or after converting the coefficients from a one-dimensional array to a two-dimensional array.

Inverse transform processing unit 78 applies an inverse transform to the inverse quantized transform coefficients. For example, inverse transform processing unit 78 may apply one of a discrete cosine transform (DCT), a discrete sine transform (DST), an integer transform, a K-L transform, or another transform to produce residual data. In some examples, inverse transform processing unit 78 may determine the inverse transform based on signaling from video encoder 20, or by inferring the transform from one or more coding characteristics, such as block size, coding mode, or the like. In some examples, inverse transform processing unit 78 may determine the transform to apply to the current block based on a signaled transform at the root node of the quadtree for the LCU that includes the current block. Alternatively, the transform may be signaled at the root of the TU quadtree for a leaf node CU in the LCU quadtree. In some examples, inverse transform processing unit 78 may apply a cascaded inverse transform, in which inverse transform processing unit 78 applies two or more inverse transforms to the transform coefficients of the current block being decoded.

Intra-prediction processing unit 74 may generate prediction data for a current block of a current frame based on a signaled intra-prediction mode and data from previously decoded blocks of the current frame.

Motion compensation unit 72 may retrieve a motion vector, a motion prediction direction, and a reference index from the encoded bitstream. The reference prediction direction indicates whether the inter-prediction mode is unidirectional (e.g., P-frames) or bidirectional (B-frames). The reference index indicates which reference frame the candidate motion vector is based on. Based on the retrieved motion prediction direction, reference frame index, and motion vector, motion compensation unit 72 generates motion compensated blocks for the current portion. These motion compensated blocks essentially recreate the predictive blocks used to generate the residual data.

Motion compensation unit 72 may generate motion compensated blocks, possibly performing interpolation based on interpolation filters. Identifiers of interpolation filters used for motion estimation with sub-pixel precision may be included in syntax elements. Motion compensation unit 72 may use interpolation filters, such as those used by video encoder 20, during encoding of a video block to calculate interpolated values for sub-integer pixels of a reference block. Motion compensation unit 72 may determine the interpolation filters used by video encoder 20 based on received syntax information and use the interpolation filters to generate predictive blocks.

In addition, in the HEVC example, motion compensation unit 72 and intra-prediction processing unit 74 may use some of the syntax information (e.g., provided by the quadtree) to determine the size of the LCU used to encode the frames of the encoded video sequence. Motion compensation unit 72 and intra-prediction processing unit 74 may also use the syntax information to determine split information that describes how each CU of the frames of the encoded video sequence is split (and similarly, how sub-CUs are split). The syntax information may also include a mode indicating how each split is encoded (e.g., intra-prediction or inter-prediction, and encoding modes for intra-prediction and intra-prediction), one or more reference frames for each inter-coded PU (and/or a reference list containing identifiers for reference frames), and other information used to decode the encoded video sequence.

Summer 80 combines the residual block with the corresponding prediction block produced by motion compensation unit 72 or intra-prediction processing unit 74 to form a decoded block. If desired, a deblocking filter may also be applied to the decoded block to remove blockiness artifacts. The decoded video block is then stored in decoded picture buffer 82, which provides reference blocks for subsequent motion compensation and also produces decoded video for presentation on a display device, such as display device 32 of FIG. 1 .

7 is a flowchart illustrating an example operation of determining initial values of Rice parameters during entropy encoding of coefficient levels according to the techniques described in this disclosure. The example operation is described with respect to video encoder 20 including entropy encoding unit 56 from FIG.

Entropy encoding unit 56 receives coefficients of residual video data to be encoded into a bitstream for transmission to video decoder 30 or for storage on storage medium 34 or file server 36. The residual video data coefficients may be included in coefficient groups (CGs), each of which is a subblock of a transform block (e.g., a 4x4 subblock as illustrated in FIG. 4). During encoding of the coefficients in the CGs, entropy encoding unit 56 determines statistics of coefficient levels of previously encoded coefficients (100). In the example described in more detail below with respect to FIG. 9, determining statistics of coefficient levels of previously encoded coefficients may include determining whether to increase or decrease a statistical value based on a function that compares the coefficient levels of one or more of the previously encoded coefficients to the statistics. The statistical value may be initialized to zero at the beginning of each slice of the residual video data.

In some examples, determining statistics for the coefficient levels of previously encoded coefficients may include calculating an average or a running average of the coefficient levels across a slice or coding unit (CU) of the residual video data. In other examples, determining statistics for the coefficient levels of previously encoded coefficients may include determining whether to increase, decrease, or maintain the statistical value based on a comparison of the coefficient level of one of the previously encoded coefficients with the statistical value.

The coefficient-level statistics may include coefficient-level absolute value statistics of previously coded coefficients or coefficient-level residual absolute value statistics. The coefficient-level statistics may be collected for previously coded coefficients in CGs that are all included in the same transform block, or for previously coded coefficients in CGs that are included in two or more different transform blocks.

In some examples, entropy encoding unit 56 may determine statistics for the coefficient levels of previously encoded coefficients once per coefficient group. The frequency of statistics collection is described in more detail below with respect to FIG. 9. For example, entropy encoding unit 56 may determine statistics when the first absolute value of the coefficient level is encoded in each of the previous CGs. Alternatively, entropy encoding unit 56 may determine statistics when the first residual absolute value of the coefficient level is encoded in each of the previous CGs. In other examples, entropy encoding unit 56 may determine statistics for the coefficient levels of previously encoded coefficients after encoding each of the coefficients.

9 , entropy encoding unit 56 may determine separate statistics for each of a plurality of different categories of CGs. The categories may be defined based on characteristics of the transform block that includes the CGs. For example, the characteristics of the transform block may include whether the transform block is a luma or chroma block type, whether the transform block is encoded as a transform skip block or a transform quantization bypass block, whether the transform block has an intra-prediction or inter-prediction slice type, the size of the transform block, and/or the location of the CG within the transform block.

For the current CG in the transform block of the residual video data to be encoded, the entropy encoding unit 56 determines the initial value of the Rice parameter based on the collected statistics of the coefficient level of the previously encoded coefficient (102). Conventionally, the value of the Rice parameter is initialized to zero at the beginning of each CG and is conditionally updated after the remaining absolute value of the coefficient level in the encoded CG. In the case of coefficient levels of decoded screen content, or in the case of lossless decoding or lossy decoding in transform skip mode, it may not be optimal to initialize the value of the Rice parameter to zero for each CG. The technical description of the present invention sets the value of the Rice parameter based on the collected statistics at the beginning of each CG, rather than automatically resetting the value of the Rice parameter to zero.

In one example described in more detail below with respect to FIG10 , the entropy coding unit 56 determines initial values of Rice parameters for the current CG by mapping the collected statistical values to initial values of the Rice parameters according to a statistical function. In other examples, the entropy coding unit 56 may map the collected statistical values to initial values of the Rice parameters according to a stored table.

As described in more detail above, entropy encoding unit 56 may perform one or more scans on the current CG according to an inverse scan order to encode coefficient levels of coefficients in the current CG. For example, entropy encoding unit 56 may encode a flag or syntax element in a conventional coding mode using a context model to indicate whether the coefficient level of the coefficient has an absolute value greater than 1 or greater than 2. Next, entropy encoding unit 56 encodes a flag or syntax element using a code defined by Rice parameters to indicate the remaining absolute value of the coefficient level of at least one of the coefficients in the current CG (104). For example, entropy encoding unit 56 may bypass mode encoding syntax elements that indicate the remaining absolute value of the coefficient level of any of the coefficients having a coefficient level greater than 2. In some examples, after encoding the first remaining absolute value of the coefficient level in the current CG, entropy encoding unit 56 may update statistics based on the coefficient level of the encoded coefficient.

In some cases, the entropy encoding unit 56 may conditionally update the initial value of the Rice parameter based on the initial value of the Rice parameter and the absolute value of the coefficient level of the coefficient being encoded in the current CG. After encoding each remaining absolute level of the coefficient in the current CG, the entropy encoding unit 56 may continue to conditionally update the value of the Rice parameter. At the end of the current CG, the entropy encoding unit 56 may determine the initial value of the Rice parameter for the subsequent CG based on statistics of the coefficient level of the previously encoded coefficients (including the most recently encoded coefficient included in the current CG).

8 is a flowchart illustrating an example operation of determining initial values of Rice parameters during entropy decoding of a coefficient level according to the techniques described in this disclosure. The example operation is described with respect to video decoder 30 including entropy decoding unit 70 from FIG.

Video decoder 30 receives an encoded video bitstream from video encoder 20, or from a storage device such as storage medium 34 or file server 36. The received video bitstream represents coefficients of residual video data to be decoded. The residual video data coefficients may be included in coefficient groups (CGs), each of which is a transform block subblock (e.g., a 4x4 subblock as illustrated in FIG. 4). During decoding of the coefficients in the CGs, entropy decoding unit 70 determines statistics of coefficient levels of previously decoded coefficients (110). In the example described in more detail below with respect to FIG. 9, determining statistics of coefficient levels of previously decoded coefficients may include determining whether to increase or decrease a statistical value based on a function that compares the coefficient levels of one or more of the previously decoded coefficients to the statistics. The statistical value may be initialized to zero at the beginning of each slice of the residual video data.

In some examples, determining statistics for the coefficient level of previously decoded coefficients may include computing an average or a running average of the coefficient level across a slice, coding unit (CU), or transform unit (TU) of the residual video data. In other examples, determining statistics for the coefficient level of previously decoded coefficients may include determining whether to increase, decrease, or maintain the statistical value based on a comparison of the coefficient level of one of the previously encoded coefficients with the statistical value.

The coefficient-level statistics may include coefficient-level absolute value statistics of previously decoded coefficients or coefficient-level residual absolute value statistics. The coefficient-level statistics may be collected for previously decoded coefficients in CGs that are all included in the same transform block, or for previously decoded coefficients in CGs that are included in two or more different transform blocks.

In some examples, entropy decoding unit 70 may determine statistics for the coefficient levels of previously decoded coefficients once per coefficient group. The frequency of statistics collection is described in more detail below with respect to FIG. 9. For example, entropy decoding unit 70 may determine statistics upon decoding the first absolute value of the coefficient level in each of the previous CGs. Alternatively, entropy decoding unit 70 may determine statistics upon decoding the first remaining absolute value of the coefficient level in each of the previous CGs. In other examples, entropy decoding unit 70 may determine statistics for the coefficient levels of previously decoded coefficients after decoding each of the coefficients.

9 , entropy decoding unit 70 may determine separate statistics for each of a plurality of different categories of CGs. The categories may be defined based on characteristics of the transform block that includes the CGs. For example, the characteristics of the transform block may include whether the transform block is a luma or chroma block type, whether the transform block is decoded as a transform skip block or a transform quantization bypass block, whether the transform block has an intra-prediction or inter-prediction slice type, the size of the transform block, and/or the location of the CG within the transform block.

For the current CG in the transform block of the residual video data to be decoded, the entropy decoding unit 70 determines the initial value of the Rice parameter based on the collected statistics of the coefficient level of the previously decoded coefficients (112). In some processes, the value of the Rice parameter is initialized to zero at the beginning of each CG and is conditionally updated after the remaining absolute value of the coefficient level in the decoded CG. In the case of coefficient levels of decoded screen content, or in the case of lossless decoding or lossy decoding in transform skip mode, it may not be optimal to initialize the value of the Rice parameter to zero for each CG. The technical description of the present invention sets the value of the Rice parameter based on the collected statistics at the beginning of each CG, rather than automatically resetting the value of the Rice parameter to zero.

In one example described in more detail below with respect to FIG10 , the entropy decoding unit 70 determines the initial values of the Rice parameters for the current CG by mapping the collected statistical values to the initial values of the Rice parameters according to a statistical function. In other examples, the entropy decoding unit 70 may map the collected statistical values to the initial values of the Rice parameters according to a stored table.

As described in more detail above, the entropy decoding unit 70 may perform one or more scans on the current CG according to an inverse scan order to decode the coefficient levels of the coefficients in the current CG. For example, the entropy decoding unit 70 may decode a flag or syntax element in a conventional coding mode using a context model to indicate whether the coefficient level of the coefficient has an absolute value greater than 1 or greater than 2. Next, the entropy decoding unit 70 decodes a flag or syntax element using a code defined by the Rice parameter to indicate the remaining absolute value of the coefficient level of at least one of the coefficients in the current CG (114). For example, the entropy decoding unit 70 may bypass the mode decoding syntax element that indicates the remaining absolute value of the coefficient level of any of the coefficients having a coefficient level greater than 2. In some examples, after decoding the first remaining absolute value of the coefficient level in the current CG, the entropy decoding unit 70 may update statistics based on the coefficient level of the decoded coefficient.

In some cases, the entropy decoding unit 70 may conditionally update the initial value of the Rice parameter based on the initial value of the Rice parameter and the absolute value of the coefficient level of the coefficient being decoded in the current CG. After decoding each remaining absolute level of the coefficient in the current CG, the entropy decoding unit 70 may continue to conditionally update the value of the Rice parameter. At the end of the current CG, the entropy decoding unit 70 may determine the initial value of the Rice parameter for the subsequent CG based on statistics of the coefficient level of the previously decoded coefficients (including the most recently decoded coefficient included in the current CG).

FIG9 is a flowchart illustrating an example operation of determining statistics for a coefficient level of previously coded coefficients during entropy coding of the coefficient level according to the techniques described in this disclosure. The illustrated operation may be an example of step 100 from FIG7 or step 110 from FIG8. The example operation is described with respect to video decoder 30 including entropy decoding unit 70 from FIG6. In other examples, the operation may be performed by entropy encoding unit 56 of video encoder 20 from FIG5.

Entropy decoding unit 70 decodes coefficient levels for coefficients in the CGs included in the transform block (120). The coefficient levels may include either absolute values of the coefficient levels or residual absolute values of the coefficient levels. If the decoded coefficient level is not the first coefficient level in the CG (NO branch of 122), entropy decoding unit 70 does not update statistics based on the decoded coefficient level (124). In this example, statistics are updated only once per CG when the first absolute value of the coefficient level or the first residual absolute value of the coefficient level is coded in each of the CGs. In other examples, statistics may be updated more frequently or based on coding of different coefficient levels.

If the decoded coefficient level is the first coefficient level in the CG (yes branch of 122), entropy decoding unit 70 determines the category of the CG based on the characteristics of the transform block (126). In one example, the characteristics of the transform block used to determine the category of the CG include whether the transform block is a luma block, and whether the transform block is a transform skip block. In other examples, the characteristics used to determine the category of the CG may be different, such as whether the transform block is a transform quantization bypass block, whether the transform block has an intra-frame prediction or inter-frame prediction slice type, the size of the transform block, and/or the position of the current CG within the transform block. In some examples, video decoder 30 may receive a syntax element indicating which characteristic of the transform block is used to define different categories of CGs.

Next, entropy decoding unit 70 compares the first coefficient level in the CG to a function of the statistic for the determined category (128). In one example, the function of the statistic used to update the statistics is a first constant value based on the statistic value that is left-shifted and divided by a second constant value. In other examples, the statistics may be determined according to different techniques, such as calculating an average or running average of the coefficient level across a slice or coding unit (CU) of the video data, or comparing the coefficient level in the CG directly to the statistic value.

If the first coefficient level in the CG is greater than or equal to the result of the function of the statistics (the yes branch of 130), entropy decoding unit 70 increases the statistic for the determined class (132). If the first coefficient level in the CG is less than the result of the function of the statistics (the no branch of 130), entropy decoding unit 70 decreases the statistic for the determined class (134). In either case, entropy decoding unit 70 may use the updated statistics to determine initial values of Rice parameters for subsequent CGs of the determined class, as described in more detail below with respect to FIG. 10.

In the example of FIG9 , the operation for determining statistics updates the statistics only once per CG and collects separate statistics for multiple different categories of CGs. In other examples, the operation for determining statistics may update the statistics after coding the remaining absolute values of the coefficient levels in the CGs. In additional examples, the operation for determining statistics may collect statistics for only one category.

FIG10 is a flowchart illustrating an example operation of determining initial values of Rice parameters for a current coefficient group based on determined statistics according to the techniques described in this disclosure. The illustrated operation may be an example of step 102 from FIG7 or step 112 from FIG8. The example operation is described with respect to a video decoder 30 including the entropy decoding unit 70 from FIG6. In other examples, the operation may be performed by the entropy encoding unit 56 of the video encoder 20 from FIG5.

At the beginning of the current CG in the transform block (140), the entropy decoding unit 70 determines the category of the current CG (142) based on the characteristics of the transform block that includes the current CG. In one example, the characteristics of the transform block used to determine the category of the current CG include whether the transform block is a luma block, and whether the transform block is a transform skip block. In other examples, the characteristics used to determine the category of the current CG may be different, for example, whether the transform block is a transform quantization bypass block, whether the transform block has an intra-frame prediction or inter-frame prediction slice type, the size of the transform block, and/or the position of the current CG within the transform block. In some examples, the video decoder 30 may receive a syntax element indicating which characteristic of the transform block is used to define different categories of CGs.

Then, the entropy decoding unit 70 maps the statistical values of the determined categories to the initial values of the Rice parameters for the current CG according to the statistical function (144). In one instance, the statistical function for initializing the Rice parameters is based on the selection of the minimum of the maximum value of the Rice parameters or the statistical value divided by the constant value. In other instances, the initial values of the Rice parameters can be determined from the statistics according to different techniques, such as right-shifting the statistical values by a constant value, or applying a linear or piecewise linear function of the statistics. In yet other instances, the entropy decoding unit 70 can map the statistical values to the initial values of the Rice parameters according to a storage table, wherein each statistical value within a predefined range is mapped to the value of the Rice parameter up to the maximum value of the Rice parameter.

In one or more examples, the functions described may be implemented in hardware, software, firmware, or any combination thereof. If implemented in software, the functions may be stored on or transmitted via a computer-readable medium as one or more instructions or program codes and executed by a hardware-based processing unit. Computer-readable media may include computer-readable storage media (which corresponds to tangible media such as data storage media) or communication media, which includes, for example, any media that facilitates the transfer of a computer program from one place to another according to a communication protocol. In this manner, computer-readable media may generally correspond to (1) tangible computer-readable storage media that is non-transitory, or (2) communication media such as a signal or carrier wave. Data storage media may be any available media that can be accessed by one or more computers or one or more processors to retrieve instructions, program codes, and/or data structures for implementing the techniques described in this disclosure. A computer program product may include a computer-readable medium.

By way of example, and not limitation, these computer-readable storage media may include RAM, ROM, EEPROM, CD-ROM or other optical disk storage, magnetic disk storage or other magnetic storage, flash memory, or any other medium that can be used to store desired program code in the form of instructions or data structures and that can be accessed by a computer. Also, any connection is properly termed a computer-readable medium. For example, if instructions are transmitted from a website, server, or other remote source using a coaxial cable, fiber optic cable, twisted pair, digital subscriber line (DSL), or wireless technologies (such as infrared, radio, and microwave), the coaxial cable, fiber optic cable, twisted pair, DSL, or wireless technologies (such as infrared, radio, and microwave) are included in the definition of medium. However, it should be understood that computer-readable storage media and data storage media do not include connections, carrier waves, signals, or other transitory media, but rather refer to non-transitory, tangible storage media. As used herein, disk and disc include compact disc (CD), laser disc, optical disc, digital laser video disc (DVD), floppy disk, and Blu-ray disc. Disks typically reproduce data magnetically, while discs reproduce data optically using lasers. Combinations of the above should also be included within the scope of computer-readable media.

Instructions may be executed by one or more processors, such as one or more digital signal processors (DSPs), general-purpose microprocessors, application-specific integrated circuits (ASICs), field-programmable logic arrays (FPGAs), or other equivalent integrated or discrete logic circuitry. Thus, the term "processor," as used herein, may refer to any of the aforementioned structures or any other structure suitable for implementing the techniques described herein. Additionally, in some aspects, the functionality described herein may be provided within dedicated hardware and/or software modules configured for encoding and decoding, or incorporated into a combined codec. Furthermore, the techniques may be fully implemented in one or more circuits or logic components.

The techniques of this disclosure can be implemented in a wide variety of devices or apparatuses, including wireless handsets, integrated circuits (ICs), or sets of ICs (e.g., chipsets). Various components, modules, or units are described in this disclosure to emphasize functional aspects of devices configured to perform the disclosed techniques, but do not necessarily require implementation by different hardware units. In fact, as described above, the various units can be combined in a codec hardware unit, or provided by a collection of interoperable hardware units (including one or more processors as described above) in conjunction with appropriate software and/or firmware.

Various embodiments have been described. These and other embodiments are within the scope of the following claims.

Claims (59)

1. A method for decoding coefficients in a video decoding process, the method comprising: determining statistics of a coefficient level of previously decoded coefficients of residual video data, the previously decoded coefficients being decoded prior to a current coefficient group in a transform block of the residual video data; determining, based on the statistics, initial values of Rice parameters at the beginning of the current coefficient group before decoding any coefficients in the current coefficient group; and Using a code defined by the initial value of the Rice parameter, the residual absolute value of the coefficient level of at least one of the coefficients in the current coefficient group is decoded. 2 . The method of claim 1 , wherein the statistics comprise statistics of one of absolute values of the coefficient level or residual absolute values of the coefficient level of the previously decoded coefficients. 3. The method of claim 1 , wherein the statistics comprise statistics of the coefficient levels of the previously decoded coefficients in a previous coefficient group included in one or more of the same transform block as the current coefficient group or a transform block different from the current coefficient group. 4. The method of claim 1 , wherein determining the statistics comprises: comparing a coefficient level of at least one of the previously decoded coefficients to a function of the statistic; and Based on the comparison, a determination is made as to whether to increase or decrease the value of the statistic. 5 . The method of claim 4 , wherein the function of the statistic comprises a first constant value that is left-shifted and divided by a second constant value. 6 . The method of claim 1 , further comprising initializing a value of the statistic to zero at the beginning of each slice of the residual video data. The method of claim 1 , wherein determining the statistic comprises determining the statistic once per group of coefficients. 8. The method of claim 7, wherein determining the statistic once per coefficient group comprises determining the statistic upon decoding one of an absolute value of a coefficient level or a residual absolute value of a coefficient level in each of a plurality of previous coefficient groups. 9. The method of claim 1, wherein determining the statistics comprises determining a separate statistic for each of a plurality of different categories of coefficient groups, wherein the different categories are defined based on characteristics of a transform block that includes the coefficient groups. 10. The method of claim 9, wherein the characteristics of the transform block include at least whether the transform block is a luma block, and whether the transform block is a transform skip block. 11. The method of claim 9, further comprising receiving a syntax element indicating which characteristics of the transform block are used to define the different categories of coefficient groups, wherein the syntax element is received in one of a sequence parameter set (SPS) or a picture parameter set (PPS) of the residual video data. 12. The method according to claim 9 further includes determining a category of the current coefficient group based on characteristics of the transform block including the current coefficient group, wherein determining the initial value of the Rice parameter includes determining the initial value of the Rice parameter for the current coefficient group in the transform block based on the statistics for the determined category. 13. The method according to claim 1, wherein determining the initial value of the Rice parameter for the current coefficient group includes mapping the value of the statistic to the initial value of the Rice parameter according to a function of the statistic. 14. A method according to claim 13, wherein the function of the statistic includes selection of the minimum of the maximum value of the Rice parameter or the value of the statistic divided by a constant value. 15. The method of claim 1, further comprising: After decoding the remaining absolute value of the coefficient level of the at least one of the coefficients in the current coefficient group, updating the initial value of the Rice parameter based on the initial value of the Rice parameter and the absolute value of the coefficient level of the at least one of the coefficients in the current coefficient group; and Using a code defined by the updated value of the Rice parameter, the remaining absolute value of the coefficient level of another one of the coefficients in the current coefficient group is decoded. 16. The method of claim 1, wherein the code defined by the initial value of the Rice parameter comprises one of a Golomblais code or an Exponential Golomb code. 17. The method of claim 1, wherein the current coefficient group comprises transform coefficients or coefficients to which no transform is applied. 18. A method for encoding coefficients in a video encoding process, the method comprising: determining statistics of coefficient levels of previously encoded coefficients of residual video data that were encoded before a current group of coefficients in a transform block of the residual video data; determining, based on the statistics, initial values of Rice parameters at the beginning of the current coefficient group before encoding any coefficients in the current coefficient group; and The residual absolute value of the coefficient level of at least one of the coefficients in the current coefficient group is encoded using a code defined by the initial value of the Rice parameter. 19. The method of claim 18, wherein the statistics comprise statistics of one of absolute values of the coefficient level or residual absolute values of the coefficient level of the previously encoded coefficients. 20. The method of claim 18, wherein the statistics comprise statistics of the coefficient levels of the previously encoded coefficients in a previous coefficient group included in one or more of the same transform block as the current coefficient group or a different transform block than the current coefficient group. 21. The method of claim 18, wherein determining the statistics comprises: comparing a coefficient level of at least one of the previously encoded coefficients to a function of the statistic; and Based on the comparison, a determination is made as to whether to increase or decrease the value of the statistic. 22. The method of claim 21, wherein the function of the statistic comprises a first constant value left-shifted divided by a second constant value. 23. The method of claim 18, further comprising initializing a value of the statistic to zero at the beginning of each slice of the residual video data. 24. The method of claim 18, wherein determining the statistics comprises determining the statistics once per group of coefficients. 25. The method of claim 24, wherein determining the statistic once per coefficient group comprises determining the statistic when one of an absolute value of a coefficient level or a residual absolute value of a coefficient level is encoded in each of a plurality of previous coefficient groups. 26. The method of claim 18, wherein determining the statistics comprises determining a separate statistic for each of a plurality of different categories of coefficient groups, wherein the different categories are defined based on characteristics of a transform block that includes the coefficient groups. 27. The method of claim 26, wherein the characteristics of the transform block include at least whether the transform block is a luma block, and whether the transform block is a transform skip block. 28. The method of claim 26, further comprising generating a syntax element that indicates which characteristics of the transform block are used to define the different categories of coefficient groups, wherein the syntax element is generated in one of a sequence parameter set (SPS) or a picture parameter set (PPS) of the residual video data. 29. The method according to claim 26 further includes determining a category of the current coefficient group based on characteristics of the transform block including the current coefficient group, wherein determining the initial value of the Rice parameter includes determining the initial value of the Rice parameter for the current coefficient group in the transform block based on the statistics for the determined category. 30. The method of claim 18, wherein determining the initial value of the Rice parameter for the current coefficient group comprises mapping the value of the statistic to the initial value of the Rice parameter according to a function of the statistic. 31. A method according to claim 30, wherein the function of the statistic includes a selection of the minimum of the maximum value of the Rice parameter or the value of the statistic divided by a constant value. 32. The method of claim 18, further comprising: After encoding the remaining absolute value of the coefficient level of the at least one of the coefficients in the current coefficient group, updating the initial value of the Rice parameter based on the initial value of the Rice parameter and the absolute value of the coefficient level of the at least one of the coefficients in the current coefficient group; and The remaining absolute value of the coefficient level of another one of the coefficients in the current coefficient group is encoded using a code defined by the updated value of the Rice parameter. 33. The method of claim 18, wherein the code defined by the initial value of the Rice parameter comprises one of a Golomblais code or an Exponential Golomb code. 34. The method of claim 18, wherein the current coefficient group comprises transform coefficients or coefficients to which no transform is applied. 35. A video decoding device, comprising: a memory configured to store video data; and One or more processors configured to: determining statistics of a coefficient level of previously coded coefficients of residual video data that were coded before a current coefficient group in a transform block of the residual video data; determining, based on the statistics, initial values of Rice parameters at the beginning of the current coefficient group before decoding any coefficients in the current coefficient group; and Using a code defined by the initial value of the Rice parameter, the residual absolute value of the coefficient level of at least one of the coefficients in the current coefficient group is decoded. 36. The device of claim 35, wherein the statistics comprise statistics of one of absolute values of the coefficient level or residual absolute values of the coefficient level of the previously coded coefficients. 37. The device of claim 35, wherein the statistics comprise statistics of the coefficient levels of the previously coded coefficients in a previous coefficient group included in one or more of the same transform block as the current coefficient group or a transform block different from the current coefficient group. 38. The device of claim 35, wherein the processor is configured to: comparing a coefficient level of at least one of the previously coded coefficients to a function of the statistic; and Based on the comparison, a determination is made as to whether to increase or decrease the value of the statistic. 39. The apparatus of claim 38, wherein the function of the statistic comprises a first constant value that is left-shifted and divided by a second constant value. 40. The device of claim 35, wherein the value of the statistic is initialized to zero at the beginning of each slice of the residual video data. 41. The device of claim 35, wherein determining the statistic comprises determining the statistic once per group of coefficients. 42. The device of claim 41, wherein the processor is configured to determine the statistic when coding one of an absolute value of a coefficient level or a residual absolute value of a coefficient level in each of a plurality of previous coefficient groups. 43. The device of claim 35, wherein the processor is configured to determine separate statistics for each of a plurality of different categories of coefficient groups, wherein the different categories are defined based on characteristics of a transform block that includes the coefficient groups. 44. The device of claim 43, wherein the characteristics of the transform block include at least whether the transform block is a luma block, and whether the transform block is a transform skip block. 45. The device of claim 43, wherein the processor is configured to determine a syntax element indicating which characteristics of the transform block are used to define the different categories of coefficient groups, wherein the syntax element is included in one of a sequence parameter set (SPS) or a picture parameter set (PPS) of the residual video data. 46. A device according to claim 43, wherein the processor is configured to determine the category of the current coefficient group based on characteristics of the transform block that includes the current coefficient group, and determine the initial value of the Rice parameter for the current coefficient group in the transform block based on the statistics for the determined category. 47. A device according to claim 35, wherein the processor is configured to map the value of the statistic to the initial value of the Rice parameter for the current coefficient group according to a function of the statistic. 48. An apparatus according to claim 47, wherein the function of the statistic comprises a selection of a maximum of the Rice parameter or a minimum of the value of the statistic divided by a constant value. 49. The device of claim 35, wherein the processor is configured to: After coding the remaining absolute value of the coefficient level of the at least one of the coefficients in the current coefficient group, updating the initial value of the Rice parameter based on the initial value of the Rice parameter and the absolute value of the coefficient level of the at least one of the coefficients in the current coefficient group; and Using a code defined by the updated value of the Rice parameter, the remaining absolute value of the coefficient level of another one of the coefficients in the current coefficient group is decoded. 50. A device according to claim 35, wherein the video decoding device includes a video decoding device, and wherein the processor of the video decoding device is configured to decode the residual absolute value of the coefficient level of at least one of the coefficients in the current coefficient group using a code defined by the Rice parameters. 51. A device according to claim 35, wherein the video decoding device comprises a video encoding device, and wherein the processor of the video encoding device is configured to encode the residual absolute value of the coefficient level of at least one of the coefficients in the current coefficient group using a code defined by the Rice parameters. 52. The apparatus of claim 35, wherein the code defined by the initial value of the Rice parameter comprises one of a Golomblais code or an Exponential Golomb code. 53. The device of claim 35, wherein the current coefficient group comprises transform coefficients or coefficients to which no transform is applied. 54. A video decoding device comprising: means for determining statistics of coefficient levels of previously coded coefficients of residual video data, the previously coded coefficients being coded before a current coefficient group in a transform block of the residual video data; means for determining, based on the statistics, initial values of Rice parameters at the beginning of the current group of coefficients before decoding any coefficients in the current group of coefficients; and A device for decoding the residual absolute value of the coefficient level of at least one of the coefficients in the current coefficient group using a code defined by the initial value of the Rice parameter. 55. The apparatus of claim 54, wherein the means for determining the statistic comprises: means for comparing a coefficient level of at least one of the previously coded coefficients to a function of the statistic, wherein the function of the statistic comprises a first constant value left-shifted from a value of the statistic divided by a second constant value; and Means for determining whether to increase or decrease the value of the statistic based on the comparison. 56. The device of claim 54, wherein the means for determining the statistic comprises means for determining the statistic once per coefficient group when coding one of an absolute value of a coefficient level or a residual absolute value of a coefficient level in each of a plurality of previous coefficient groups. 57. The device of claim 54, wherein the means for determining the statistics comprises means for determining a separate statistic for each of a plurality of different categories of coefficient groups, wherein the different categories are defined based on characteristics of a transform block that includes the coefficient groups, and wherein the characteristics of the transform block include at least whether the transform block is a luma block and whether the transform block is a transform skip block. 58. A device according to claim 54, wherein the device for determining the initial value of the Rice parameter for the current coefficient group includes a device for mapping the value of the statistic to the initial value of the Rice parameter according to a function of the statistic, wherein the function of the statistic includes a selection of the minimum of the maximum value of the Rice parameter or the value of the statistic divided by a constant value. 59. A computer-readable storage medium comprising instructions that, when executed by one or more processors of a video coding device, cause the processors to: determining statistics of a coefficient level of previously coded coefficients of residual video data that were coded before a current coefficient group in a transform block of the residual video data; determining, based on the statistics, initial values of Rice parameters at the beginning of the current coefficient group before decoding any coefficients in the current coefficient group; and Using a code defined by the initial value of the Rice parameter, the residual absolute value of the coefficient level of at least one of the coefficients in the current coefficient group is decoded.