Transform coefficient block coding

By employing context-dependent entropy encoding and adaptive binarization for transform factor blocks, the coding efficiency of images and videos is enhanced, addressing inefficiencies in existing technologies and improving data compression and decoding.

KR1020260161674APending Publication Date: 2026-11-09프라운호퍼게젤샤프트쭈르푀르데룽데어안겐반텐포르슝에베
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
KR1020267036414
Authority / Receiving Office
KR · KR
Patent Type
Applications
Current Assignee / Owner
Priority Date
2018-03-29
Filing Date
2019-03-28
Publication Date
2026-11-09

AI Technical Summary

Technical Problem

Existing image and video coding technologies face inefficiencies in encoding and decoding processes, particularly in the context of transform factor blocks, which can be improved through enhanced entropy encoding and decoding methods.

Method used

The proposed solution involves using context-dependent entropy encoding and decoding for transform factor blocks, where the context for encoding/decoding the second transform factor block is based on the first, and employing separate sets of contexts for entropy-coding quantization levels, along with adaptive binarization and cutoff value adjustments to enhance coding efficiency.

Benefits of technology

This approach increases coding efficiency by improving probability estimation and reducing redundancy in transform coefficient blocks, leading to more effective data compression and decoding processes.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure P1020267036414_ABST
    Figure P1020267036414_ABST
Patent Text Reader

Abstract

Concepts for more efficiently coding images and / or videos into data streams, in particular concepts using efficient transform factor block coding, are explained.
Need to check novelty before this filing date? Find Prior Art

Description

Technology Field

[0001] This application relates to the coding of images and videos using transform factor block coding.

[0002] That is, the present application relates to the field of digital signal processing, and in particular to methods and devices for image and video decoders and encoders.

[0003] The object of the present invention is to provide concepts for more efficiently coding images and / or videos into data streams.

[0004] This objective is achieved by the subject matter of the independent claims of the present application.

[0005] According to a first aspect of the present application, an increase in the coding efficiency of image coding is achieved by using a context or context model for entropy encoding / decoding a specific transform factor block of a specific color component that depends on a transform factor block of a different color component. That is, a block of an image is transformed individually for the first color component and the second color component, respectively, to obtain a first transform factor block for the first color component and a second transform factor block for the second color component, and when the second transform factor block is context-adaptively entropy encoded / decoded, the context used is selected according to the first transform factor block. Both transformations may be related to the prediction residual determined for the block for each color component, but this is merely optional. In any case, making the context used to entropy encode / decode the second transform factor block depend on the first transform factor block improves the probability estimation underlying the entropy encoding / decoding and thereby increases coding efficiency. According to one embodiment, a context dependency for a first transformation coefficient block is used or applied to a mark coded into a data stream for the first transformation coefficient block and a second transformation coefficient block, such mark indicating whether the transformation coefficients within each transformation coefficient block are encoded into the data stream, or whether the encoding of the transformation coefficients of each transformation coefficient block is skipped and all transformation coefficients within each transformation coefficient block are zero. This mark may be a flag signaled for both transformation coefficient blocks of the data stream. Thus, according to this embodiment, the encoding / decoding of the mark for the second transformation coefficient block is performed using context-adaptive entropy encoding / decoding that uses a context dependent on the first transformation coefficient block, for example, according to the mark signaled into the data stream for the first transformation coefficient block or the quantization levels of the transformation coefficients of the first transformation coefficient block.

[0006] According to a second aspect of the present application, image coding using transform coefficient block entropy coding becomes more efficient by consuming a separate set of contexts to entropy-code the quantization levels of specific transform coefficients. In particular, it is assumed that a transform coefficient block representing a block of an image is coded into a data stream using a scan pattern that sequentially traverses the transform coefficients of the transform coefficient block. Accordingly, the encoding / decoding of data representing a coding set of transform coefficients, where the scan pattern traverses forward from a first end coefficient position to a second end coefficient position in a predetermined direction, is performed using entropy encoding / decoding. The data includes quantization levels of transform coefficients within the coded set of transform coefficients. The encoding / decoding of the quantization levels is performed context-adaptively using a first set of contexts for the quantization levels of the transform coefficients at the first end coefficient position or at the second end coefficient position, wherein this first set of contexts is disjoint with a second set of contexts used for any other transform coefficient within the coded set of transform coefficients. For example, a separate set of contexts is used for the DC transform coefficients of the transform coefficient block, or in other words, for the transform coefficients at the top left, but the predetermined direction leads to these DC transform coefficients representing the second end coefficient position. For example, the quantization levels of the transform coefficients can be coded using context-adaptive binary entropy coding / decoding by using the binarization of the quantization levels of the transform coefficients, and contexts of a disjoint set can be adapted or used for one or more bins of the binarization, such as the transpose of the binarization.

[0007] According to a third aspect of the present application, an improvement in coding efficiency when encoding an image using entropy coding of a transform factor block is achieved by determining a context for entropy encoding / decoding the quantization level of a currently encoded transform factor of a transform factor block based on the sum and / or number of significant transform factors among one or more previously encoded transform factors located at positions determined by a local template positioned on the currently encoded / decoded transform factor. That is, a local template is used to examine transform factors located nearby in the transform factor block, provided that the quantization level of one or more transform factors thus examined is revealed by the previous part of the data stream. For example, the entropy encoding / decoding of the quantization levels of the transform coefficients of a transform coefficient block may be performed in binary units by sequentially entropy encoding / decoding bins of the binarization of the quantization levels of the transform coefficients in multiple passes, and thus the context for at least one bin of the binarization of the currently encoded / traversed transform coefficient may be determined based on the aforementioned sum and / or the aforementioned number of valid transform coefficients among the transform coefficients located in the local template. For example, the sum may be the sum of the absolute values ​​of the coefficient levels of one or more previous transform coefficients located in the local template, wherein these coefficient levels of one or more previous transform coefficients have absolute values ​​that are minimized according to the previously encoded / decoded bins of the binarization of the coefficient levels of one or more previous transform coefficients. Additionally, or alternatively, the context may be determined based on the number of valid transform coefficients among one or more previous transform coefficients located in the local template, with significance determined based on the previously encoded / decoded bins.

[0008] Further aspects of the present application relate to coding the transform coefficients of a transform coefficient block using binarization of the absolute values ​​of the quantization levels of the transform coefficients. According to this aspect of the present application, increased coding efficiency is achieved by setting a binarization parameter for parameterizing the binarization of the currently encoded / decoded transform coefficients based on the sum and / or number of valid transform coefficients among one or more previously encoded / decoded transform coefficients located at positions determined by a local template positioned on the currently encoded / decoded transform coefficients. For example, a first part of the binarization of the transform coefficients may be coded using context-adaptive entropy encoding / decoding, while a second part of the binarization is encoded / decoded using an equi-probability bypass mode. That is, the second part is written to a data stream at a code rate of 1 and read from the data stream. The second part may include a prefix and a postfix, and the binarization parameter may determine the length of the prefix. For example, the length may be an Exp-Golomb order or a Rice parameter. Descriptions similar to those made above may be true in relation to the possibility of coding the transformation coefficients of a transformation coefficient block or their coefficient levels across multiple passes, and in relation to the consideration of such situations when forming the sum of these effective coefficients and / or determining the number of effective coefficients.

[0009] Further aspects of the present application also relate to binarization used to code the absolute values ​​of the quantization levels of the transform coefficients of a transform coefficient block. In particular, according to this aspect, image coding is performed more efficiently by adaptively changing a cutoff value associated with binarization. Binarization includes a first binarization code below the cutoff value and a second binarization code above the cutoff value, in which the codeword of the first binarization code is transposed with respect to the cutoff value. Adaptive changing of the cutoff value is performed according to previously encoded / decoded transform coefficients. For example, the adaptation may be performed by setting the cutoff value to 0 as well, wherein the binarization includes only the second binarization code. Coding may be performed in such a way that the bins of the first binarization code are coded context-adaptively, while the bins of the second binarization code are coded in bypass mode.

[0010] Further aspects of the present application also relate to cutoff values, aiming to improve coding efficiency by setting a cutoff value according to one or more of the size of the block, the color component of the block, the prediction mode on which the prediction signal is based—wherein the block represents the prediction residual of the prediction signal—the transformation on which the transformation coefficient block is based, the quantization parameter used to quantize the transformation coefficient block, and the energy measurement of previously encoded / decoded transformation coefficients, wherein the previously encoded / decoded transformation coefficients may be positioned at positions determined by a local template positioned at the currently encoded / decoded transformation coefficients, or at positions within a partition of the transformation coefficient block preceding the current partition according to the coding order used to encode / decode the absolute value of the quantization level of the transformation coefficients of the transformation coefficient block defined by traversing the transformation coefficients of the transformation coefficient block along a scan pattern in a predetermined direction.

[0011] Another additional aspect of the present application aims to increase the coding efficiency of image coding by using transform factor block coding in an intelligent way that disables a local template used to context-adaptively entropy-encode the quantization levels of the transform factors of the transform factor block, or sets the shape of the local template. In particular, the dependency may involve one or more of the block size, the block color component, the position of the terminal factor position where a non-zero last transform factor is generated when traversing the scan pattern along the forward direction, and the transform on which the transform factor block is based. Then, the context may be determined based on one or more previously encoded / decoded transform factors located at positions determined by the local template positioned on the currently encoded / decoded transform factor, or independently of these transform factors if the local template is disabled. In this way, context efficiency in terms of context management complexity, the number of contexts, and the accuracy of probability estimation can be better determined or adapted to actual requirements. In this regard, it should be noted that the use of too many contexts does not necessarily increase coding efficiency. Rather, the selection of contexts requires consideration of the fact that contexts should preferably be used sufficiently frequently to achieve good probability estimates through context updates. Therefore, this aspect is responsible for adapting context management features to the needs in order to improve coding efficiency.

[0012] According to a further aspect of the present application, a transform factor block is partitioned into patterns to use separate sets of contexts for various partitions. The goal is to increase coding efficiency by changing the shape of partitions so that the partitions within the transform factor block are not identical to each other, and / or by shaping the partitions according to the scan pattern in which the entropy encoding / decoding of the quantization levels of the transform factor block is performed, and / or by shaping the partitions according to the size of the block, and / or by shaping the partitions according to explicit partition shaping information. Furthermore, the idea according to this aspect is related to the need to adapt context management complexity to actual requirements. This thereby increases coding efficiency.

[0013] Further aspects of the present application relate to partitioning a conversion factor block into partitions in terms of signaling in the data stream for each partition an indication of whether all conversion factors within each partition are coded into the data stream, or whether the coding of all conversion factors within each partition is skipped and all conversion factors within each partition are zero. In particular, according to this aspect of the present application, the indication is interpreted in a manner that indicates whether all factors are coded. Each of these partitions may have a set of contexts associated with them, using which the conversion factors within each partition are encoded if so indicated by the indication. That is, the zeroness indication at the partition level, if so set, indicates that all conversion factors within a particular partition are coded and that this situation no longer needs to be questioned.

[0014] Further aspects of the present application also relate to aspects of consuming partition-specific context sets to encode quantization levels of transform coefficients within various partitions, and aim to improve coding efficiency by signaling to a data stream via a partition mode for a transform coefficient block whether partitioning is used or whether such partitioning is disabled and this set of contexts is instead used globally for the transform coefficient block.

[0015] Finally, additional aspects of the present application also address the use of partition-specific context sets and propose improving coding efficiency by separating a scan pattern in which the transformation coefficients of a transformation coefficient block are sequentially coded from partitioning, in that the scan pattern sequentially traverses the transformation coefficients of a transformation coefficient block such that at least one transformation coefficient of a first partition is traversed between two transformation coefficients of a second partition. In this way, it is feasible to traverse the transformation coefficients in such a way that "knowledge growth" during scanning the transformation coefficients increases more rapidly, thereby obtaining an improved coding history for context selection / adaptation, while at the same time appropriately associating individual context sets with various partitions of the transformation block. Brief explanation of the drawing

[0016] Advantageous implementations of the embodiments and aspects described above are the subject of the dependent claims. Preferred embodiments of the present application are described below with respect to the drawings in the following: FIG. 1 illustrates a schematic block diagram of a block-based predictive decoder using transform-based residual coding, which serves as an example of possible implementations of the embodiments of the present application described herein. FIG. 2 illustrates a schematic block diagram of a block-based predictive video decoder that fits the encoder of FIG. 1, which serves as an example of possible implementations of embodiments for the decoder described herein. FIG. 3 illustrates an example of partitioning an image into encoded coding blocks using intra-prediction and inter-prediction, respectively, and partitioning the same image into residual blocks for transformation-based residual coding of the prediction residuals, while schematically illustrating prediction correction based on residuals through the addition of the prediction signal and the residual signal, respectively. FIG. 4 illustrates a schematic diagram illustrating transformation-based residual coding of a residual block by decoding / encoding a transformation coefficient block (92) associated with the residual block through a specific transformation. FIG. 5 illustrates a schematic diagram illustrating the coding of a coded set by defining a coded set of coded coefficients in a coded set that includes all coefficients between the last non-zero coefficient and the first coefficient position, through a mark indicating the last non-zero coded coefficient when scanning coded coefficients from a specific first scanned coded coefficient, such as a DC coefficient, to the coefficient furthest from this first coefficient. FIG. 6 illustrates a schematic diagram illustrating the possibility of coding transformation coefficients using a binaryization composed of a first binaryization code and a second binaryization code, wherein the binaryization containing only the first binaryization code is converted into a state containing a codeword in which the second binaryization code follows the first binaryization code for a cutoff value. FIG. 7 illustrates a schematic diagram illustrating the possibility of having a binarization having a prefix and a postfix, where such binarization can be the basis of, for example, the second binarization code of FIG. 6. FIG. 8 illustrates the partitioning of a conversion factor block into partitions, wherein the regular partitioning into rows and columns of the partitions is exemplified, so that each partition is composed of an array of factors, wherein the partitioning can be used for context set associations for conversion factors and / or zeroing indications of conversion factors as described below in this specification. Figure 9 illustrates a schematic diagram illustrating the use of partitions for context set association. FIG. 10 schematically illustrates the use of a local template for this currently encoded / decoded transform factor block to select a context for the transform factor block and the currently entropy encoded / decoded transform factor. FIG. 11a illustrates a schematic diagram illustrating the coding of residual blocks for the decomposition of this block into more than one color component. FIG. 11b schematically illustrates data (96) coded into a data stream (14) by including a global zeroing indicator (190) indicating whether all transformation coefficients of the transformation coefficient block (92) are zero, in which case the quantization levels of the transformation coefficients do not need to be coded but can be inferred as zero at the decoder side. FIG. 11c schematically illustrates the use of partitioning a conversion factor block into partitions to represent the zeroing of conversion factors in units of partitions through coded sub-block flags (CBF). FIG. 11d illustrates that the data (96) to which the conversion coefficient block is coded may include a last position indicator (114) indicating the last coefficient position of the non-zero conversion coefficient as illustrated in FIG. 5. FIG. 11e exemplarily illustrates the entropy decoding / encoding of quantization levels of transform coefficients in a context-adaptive manner using a specific context from a context set. FIG. 11f schematically illustrates the fact that the data being coded in the transform coefficient block may include sequential coding of quantization levels of transform coefficients in a coding order (102) of a data stream corresponding to traversing the scan portion of FIG. 5 from the last coefficient position to the first coefficient position. FIG. 12 illustrates a schematic diagram illustrating a first embodiment of the present application in which context selection for one color component is performed according to the transformation coefficient block of the same residual block for another color component. FIG. 13 illustrates a schematic diagram illustrating the partitioning of a conversion factor block into partitions with mismatched shapes as shown in FIG. 8. FIG. 14 illustrates a schematic diagram illustrating the possibility of changing, or even disabling, the shapes of local templates used for context selection as taught in FIG. 10. FIG. 15 illustrates a schematic diagram illustrating a conversion factor block partitioned into partitions, wherein the scan portion for sequentially encoding the conversion factors traverses the conversion factors in such a way that the conversion factors of the partitions are interleaved with each other. FIG. 16 illustrates the coding of a transform factor block and an interleaving scan portion as illustrated in FIG. 15, and illustrates a schematic diagram exemplifying the currently entropy-encoded / decoded transform factor and previously encoded transform factor exceeding certain thresholds, and the count of these transform factors is used, for example, for context selection for the currently entropy-encoded / decoded transform factor. FIG. 17 illustrates a schematic diagram illustrating the result of coding a transform coefficient block having an interleaving scan portion exemplarily illustrated in FIG. 18, on the one hand, a coding sequence of the quantization levels of the coefficients and on the other hand, zeroing indications of the coded sub-block flags. FIG. 18 also illustrates a transformation coefficient block having a scan portion interleaving transformation coefficients of different partitions as shown in FIG. 15 and FIG. 16, exemplified by using hatching of non-zero coefficients. Figure 19 illustrates a partition-unit scan example of transformation examples of a transformation coefficient block. Figure 20 illustrates pseudocode for coding a transformation factor block. Figure 21 illustrates a pseudocode for binarizing the absolute value of the quantization level of a transformation coefficient. Figure 22 illustrates an example for binarizing the absolute value of the quantization level of a transformation coefficient. FIG. 23 illustrates pseudocode for coding the transformation coefficients of a transformation coefficient block. Figure 24 illustrates pseudocode for coding the transformation coefficients of a transformation coefficient block. Figure 25 illustrates other examples of partitioning a transformation factor block into partitions. Specific details for implementing the invention

[0017] The following description of the drawings begins with a description of a video encoder and a video decoder of a block-based predictive codec for coding images of video to form an example of a coding framework in which the embodiments described herein are constructed. The video encoder and the video decoder are described in relation to FIGS. 1 through 3. Below, descriptions of additional embodiments of the present application are presented in relation to the drawings. Likely numbered, references are made to which parts above refer to which embodiments described and claimed below. While all such embodiments may be inherent in the video encoder and decoder of FIGS. 1 and 2, the embodiments described herein, such as those described in the following drawings, may also be used to form video encoders and video decoders that do not operate according to the coding framework underlying the video encoder and video decoder of FIGS. 1 and 2.

[0018] FIG. 1 illustrates a device for predictive coding a video (11) composed of a series of images (12) into a data stream (14). For this purpose, block-based predictive coding is used. Additionally, transform-based residual coding is used exemplarily. The device or encoder is indicated by reference numeral 10. FIG. 2 illustrates a device (20) configured to predictive decode a video (11') composed of images (12') from a data stream (14) into blocks of images, exemplarily using transform-based residual decoding, wherein the apostrophe is used to indicate that the images (12') and video (11'), respectively reconstructed by the decoder (20), deviate from the images (12) originally encoded by the device (10) in terms of coding loss induced by the quantization of the predictive residual signal. FIGS. 1 and 2 use transformation-based predictive residual coding as an example, but the embodiments of the present application are not limited to this type of predictive residual coding. This also applies to other details described with respect to FIGS. 1 and 2, as outlined below.

[0019] The encoder (10) is configured to apply a spatial-spectral transformation to the predicted residual signal and to encode the obtained predicted residual signal into a data stream (14). Likewise, the decoder (20) is configured to decode the predicted residual signal from the data stream (14) and to apply a spectrum-space transformation to the obtained predicted residual signal.

[0020] Internally, the encoder (10) may include a prediction residual signal generator (22) that generates a prediction residual (24) to measure the deviation of the prediction signal (26) from the original signal, i.e., the video (11) or the current image (12). The prediction residual signal generator (22) may be, for example, a subtractor that subtracts the prediction signal from the original signal, i.e., the current image (12). The encoder (10) may further include a converter (28) that applies a spatial-spectral transformation to the prediction residual signal (24) to obtain a spectrum domain prediction residual signal (24'), the prediction residual signal (24') then undergoes quantization by a quantizer (32) which is also included in the encoder (10). The quantized prediction residual signal (24'') is then encoded into a bitstream (14). To this end, the encoder (10) may optionally include an entropy coder (34) that entropy codes a prediction residual signal converted and quantized into a data stream (14). A prediction residual (26) is generated by the prediction stage (36) of the encoder (10) based on a prediction residual signal (24'') that is decoded into the data stream (14) and decodingable therefrom. To this end, the prediction stage (36) may internally include an inverse quantizer (38) that inversely quantizes the prediction residual signal (24'') to obtain a spectrum domain prediction residual signal (24''') corresponding to the signal (24') excluding quantization loss, as illustrated in FIG. 1, and then an inverse converter (40) that applies an inverse converter, i.e., a spectrum-space converter, to the spectrum domain prediction residual signal (24''') to obtain a prediction residual signal (24'''') corresponding to the original prediction residual signal (24) excluding quantization loss. Then, a combiner (42) of the prediction stage (36) recombines the prediction signal (26) and the prediction residual signal (24'''') by, for example, addition, to obtain a reconstructed signal (46), i.e., a reconstruction of the original signal (12). The reconstructed signal (46) may correspond to the signal (12').

[0021] Next, the prediction module (44) of the prediction stage (36) generates a prediction signal (26) based on the signal (46) by using, for example, spatial prediction, i.e., intra prediction, and / or temporal prediction, i.e., inter prediction. Details regarding this point are described below.

[0022] Likewise, the decoder (20) is internally composed of components corresponding to the prediction stage (36) and can be interconnected in a manner corresponding to the prediction stage (36). In particular, the entropy decoder (50) of the decoder (20) can entropy decode a quantized spectrum domain prediction residual signal (24'') from a data stream, and accordingly, the inverse quantizer (52), inverse converter (54), combiner (56), and prediction module (58), which are interconnected and cooperate in the manner described above with respect to the modules of the prediction stage (36), recover a reconstructed signal based on the prediction residual signal (24''), so that, as shown in FIG. 2, the output of the combiner (56) becomes the reconstructed signal, i.e., the video (11') or its current image (12').

[0023] Although not specifically described above, it is easily apparent that the encoder (10) can set some coding parameters, such as prediction modes, motion parameters, etc., according to some optimization method, such as some rate and distortion-related criteria, that is, how to optimize coding costs and / or how to use some rate control. As described in more detail below, the encoder (10) and the decoder (20) and their respective corresponding modules (44, 58) support different prediction modes, such as intra-coding modes and inter-coding modes, or modes that form a kind of set or pool of primitive prediction modes on which the predictions of image blocks are constructed. The degree of granularity with which the encoder and decoder switch between these prediction configurations can correspond to subdividing the images (12, 12') into blocks, respectively. Note that some of these blocks may be merely intra-coded blocks, some may be merely inter-coded blocks, and optionally additional blocks may be obtained using both intra-coding and inter-coding, but details are presented below. According to the intra-coding mode, a prediction signal for a block is obtained based on the already spatially coded / decoded neighbors of each block. Several intra-coding submodes may be selected, among which "quasi" represents a type of intra-prediction parameter. There may be directional or angular intra-coding submodes; depending on these modes, the prediction signal for each block is filled by extrapolating the sample values ​​of the neighbors along a specific direction particular to each directional intra-coding submode into each block.Intra-coding sub-modes may also include one or more additional sub-modes, such as a DC coding mode in which a prediction signal for each block is assigned a DC value to all samples within each block, and / or a plane intra-coding mode in which a slope and offset of a plane defined by a two-dimensional linear function are derived based on neighboring samples so that a prediction signal for each block is estimated or determined as a spatial distribution of sample values ​​described by a two-dimensional linear function for the sample positions of each block. In contrast, according to the inter-prediction mode, a prediction signal for a block may be obtained, for example, by predicting the interior of the block in time. For parameterization of the inter-prediction mode, motion vectors may be signaled within the data stream, the motion vectors representing the spatial displacement of a previously coded portion of the video (11) in which the previously coded / decoded image is sampled to obtain a prediction signal for each block. This means that in addition to residual signal coding consisting of a data stream (14) such as entropy-coded transform coefficient levels representing a quantized spectral domain prediction residual signal (24''), the data stream (14) may be encoded with prediction-related parameters for assigning prediction modes to blocks, prediction parameters for assigned prediction modes such as motion parameters for inter-prediction modes, as outlined in more detail below, and optionally, additional parameters for controlling the composition of the final prediction signal for the blocks using the assigned prediction modes and prediction parameters. Additionally, the data stream may include parameters that control and signal the subdivision of images (12, 12') into blocks, respectively. The decoder (20) uses these parameters to subdivide the images in the same way as the encoder, assign the same prediction modes and parameters to the blocks, and perform the same prediction to produce the same prediction signal.

[0024] FIG. 3 illustrates the relationship between a combination of a reconstructed signal, i.e., a reconstructed image (12'), on the one hand, and a prediction residual signal (24'''') and a prediction signal (26) that are signaled as a data stream on the other. As previously mentioned above, the combination may be addition. Although the prediction signal (26) is illustrated in FIG. 3 as subdividing an image region into blocks (80) of various sizes, this is merely an example. The subdivision may be any subdivision, such as regular subdivision of the image region into rows and columns of blocks, or multi-tree subdivision of the image (12) into leaf blocks of various sizes, such as quadtree subdivision, and a mixture thereof is illustrated in FIG. 3, where the image region is first subdivided into rows and columns of tree root blocks, which are then further subdivided according to recursive multi-tree subdivision to become blocks (80).

[0025] The predicted residual signal (24'''') of FIG. 3 is also exemplified as a subdivision of the image region into blocks (84). These blocks may be called transformation or residual blocks to distinguish them from the coding blocks (80). In fact, FIG. 3 illustrates that the encoder (10) and the decoder (20) can use two different subdivisions of the image (12) and image (12'), respectively, into blocks: one subdivision into coding blocks (80) and the other subdivision into blocks (84). The two subdivisions may be identical, that is, each block (80) may simultaneously form a transformation block (84) and vice versa, but FIG. 3 illustrates, for example, a case where any boundary between two blocks (80) overlays the boundary between two blocks (84), or alternatively, a subdivision into transformation blocks (84) forms an extension of the subdivision into blocks (80) such that each block (80) corresponds to one of the transformation blocks (84) or corresponds to a cluster of transformation blocks (84). However, the subdivisions may also be determined or selected independently of each other so that the transformation blocks (84) may alternatively cross the block boundaries between blocks (80). As far as subdivision into transformation blocks (84) is concerned, descriptions similar to those raised in relation to subdivision into blocks (80) are true, namely, the blocks (84) may be the result of regular subdivision of an image region into blocks arranged in rows and columns, the result of recursive multitree subdivision of an image region, or a combination thereof, or any other type of segmentation. Furthermore, it is noted that the blocks (80, 84) are not limited to being square, rectangular, or any other shape. Also, subdividing the current image (12) into blocks (80) where prediction signals are formed, and subdividing the current image (12) into blocks (84) where prediction residuals are coded, may not be the only subdivisions used for coding / decoding.This subdivision is caused by the granularity at which prediction signal determination and residual coding are performed, but first, residual coding can alternatively be performed without subdivision, and second, at granularities other than these subdivisions, the encoder and decoder can set specific coding parameters that may include some of the previously described parameters, such as prediction parameters, prediction signal configuration control signals, etc.

[0026] FIG. 3 illustrates that a combination of a prediction signal (26) and a prediction residual signal (24'''') directly produces a reconstructed signal (12'). However, it should be noted that, according to alternative embodiments such as prediction signals obtained from other coding layers or other views that are coded / decoded in separate prediction loops using a separate DPB, more than one prediction signal (26) may be combined with the prediction residual signal (24'''') to produce an image (12').

[0027] In FIG. 3, the transformation blocks (84) will have the following meanings. The converter (28) and the inverse converter (54) perform their transformations in units of these transformation blocks (84). For example, many codecs use some kind of DST or DCT for all transformation blocks (84). Some codecs allow skipping of transformations for some of the transformation blocks (84) so ​​that the predicted residual signal is coded directly in the spatial domain. However, according to the embodiments described below, the encoder (10) and the decoder (20) are configured in such a way that they support multiple transformations. For example, the transformations supported by the encoder (10) and the decoder (20) may include the following:

[0028] o DCT-II (or DCT-III), where DCT stands for Discrete Cosine Transform

[0029] o DST-IV, where DST stands for Discrete Sine Transform

[0030] o DCT-IV

[0031] o DST-VII

[0032] o Identity transformation (IT)

[0033] Naturally, the converter (28) will support all forward versions of these conversions, while the decoder (20) or inverse converter (54) will support the corresponding inverse or inverse versions:

[0034] o Inverse DCT-II (or Inverse DCT-III)

[0035] o Station DST-IV

[0036] o Inverse DCT-IV

[0037] o Station DST-VII

[0038] o Identity transformation (IT)

[0039] In any case, it should be noted that the set of supported transformations may include only one transformation, such as a single spectral-space or space-spectral transformation.

[0040] As outlined above, FIGS. 1–3 are presented as examples in which concepts further described below may be implemented to form specific examples of video encoders and decoders according to the present application. So far, the video encoder and decoder of FIGS. 1 and FIGS. 2 represent possible implementations of the video encoders and decoders described below in this specification.

[0041] Embodiments of various aspects of the present application to be described later may occasionally use the common basic concepts presented earlier. That is, the encoders and decoders of the embodiments to be described later do not necessarily have to correspond to the details presented earlier in relation to FIGS. 1–3 as outlined above, but they may correspond to one or more of the following technical details, which are helpful or assist in understanding the embodiments to be described later. For example, starting with FIG. 4. The embodiments to be described later are consistent in that they relate to image coding. An image may be part of a video, and the decoders and encoders described below may be video encoders and video decoders, respectively, but image coding may also be applied. Accordingly, FIG. 4 schematically illustrates the general operation or task of the encoders and decoders described below.

[0042] FIG. 4 shows an image to be coded. To aid understanding of how the details set forth below in FIG. 1–3 can be used to implement the examples of FIG. 1–3 in a manner that leads to embodiments of the present application, the same reference numerals used in FIG. 1–3 are used. Additionally, FIG. 4 illustrates a block (84) of an image (12). This is a currently coded block or a currently decoded block. The use of the reference numeral (84) reveals that the transformation coding outlined below in the present specification may be related to the prediction residuals as outlined above in relation to FIG. 1–3, but this is a situation that may be excluded in principle in alternative embodiments. That is, for all embodiments set forth below in the present specification, the transformation factor block mentioned herein may represent a transformation of the prediction residuals within the block (84), such as the residuals for in-image prediction or inter-image prediction. The block (84) is related to the transformation factor block (92) through the transformation (90). Examples of transformations are presented above in relation to FIGS. 1–3 and will apply to all embodiments. One transformation is selected for block (84). Alternatively, the transformation may be fixed. In the case of selection, the selection may be signaled in the data stream. One of the transformations may be an identity transformation. The transformation of the transformation coefficient block (92) may have dimensions or sizes that match the size of block (84), that is, the number of coefficients of block (92) may match the number of samples of block (84), but this does not need to be applied to all supported transformations, or may not be applied when only one transformation (90) is used. Both the encoder and the decoder may use the inverse transformation to reconstruct the interior of block (84), such as the predicted residual in the spatial domain, the decoder may use the inverse transformation for presentation, and the encoder may use the inverse transformation to supply the decoded image buffer.The embodiments described below focus on the task (108) of decoding / encoding a conversion factor block (92) into / from a data stream (14).

[0043] As illustrated in FIG. 4, the transformation coefficient block (92) may be composed of transformation coefficients (91) arranged in a rectangular array. That is, the transformation coefficient block (92) may be a rectangular block.

[0044] The goal of the transformation (90) is to aim for redundancy reduction and to collect non-zero coefficients toward a smaller number of coefficients (91). For example, refer to FIG. 5, which shows a transformation coefficient block (92). The transformation coefficients at the transformation coefficient position (104), i.e., the upper left coefficient position, may be DC coefficients, and toward this coefficient position (104), the probability of a specific coefficient being quantized to a non-zero quantization level increases, for example. To take advantage of this situation, the coding of the transformation coefficient block (92) may involve signaling of an indication in the data stream representing the transformation coefficient position (98) furthest from the coefficient position (104) when sequentially scanning the transformation coefficients starting from the coefficient position (104) along the scan path (94) in a manner that generally proceeds away from the coefficient position (104). The indication of these end positions (98) serves as an indication that the data stream (14) is encoded only with the transformation coefficients (91), that is, the coded set (100) of the hatched ones in FIG. 5, and the coded set (100) includes all transformation coefficients, including the end coefficient position (98) and the end coefficient position (104). Thus, other coefficients not hatched in FIG. 5 are known to be 0 at the decoder side.

[0045] This may be useful when, as used according to some embodiments of the present application, a coded set of conversion coefficients (91) is coded into a data stream (14) in the opposite direction leading to the end coefficient position (104), rather than in the forward direction along the reverse direction (118), that is, along the scan path (94) that generally proceeds away from the end coefficient position (104).

[0046] Although not specifically discussed in relation to some embodiments described later, the quantization levels of the transformation coefficients (91) being coded into the data stream (14) using binarization may (optionally) also apply to the latter embodiments. That is, the quantization level of a specific transformation coefficient (91) is by binarization mapped to a bin string or codeword (161), as illustrated in FIG. 6. Binarization may consist of a first binarization code (160) and a second binarization code (162) that serve as a kind of transposition. More precisely, 106 in FIG. 6 is referred to as a quantization level. If the quantization level (106) exceeds a specific cutoff value (164), the quantization level (106) is binarized into a bin string (161) consisting of a portion (162) generated from the second binarization code in which the codeword of the first binarization code corresponding to the cutoff value (164) is transposed. However, if the quantization level (106) is not higher than the cutoff value, the quantization level (106) is mapped to an empty string consisting only of the first binary code (160). In some embodiments, the cutoff value (164) may be zero for certain conversion coefficients, in which case the binary (161) includes only the second binary code (162).

[0047] One or more bins forming the binarization (161) of a specific transformation coefficient may be encoded into a data stream and decoded coefficient by coefficient along the scan path (94) therefrom, but some embodiments described later use multiple passes traversing the coefficients. This ultimately means that whenever "previously encoded / decoded transformation coefficients" are mentioned below, the binarization of these coefficients may not have been fully encoded / decoded. Rather, at least one bin for these previously encoded / decoded transformation coefficients has been encoded / decoded. The binarization (161) may be formed such that the length of the first binarization code, the length of the second binarization code, as well as the total binarization length of the binarization (161), increases monotonically with respect to the quantization level (106) or, more precisely, its magnitude value. That is, for a specific quantization level (106), the more preceding bins of the binarization (161) are read, the higher the probability that the absolute value of the minimum quantization level will become larger, such that first, the binarization (161) is completely read, and second, the quantization level can drop down, or if there are bins of the binarization (161) following these preceding bins that have not yet been read.

[0048] As a specific example, for instance, the bins of the first binary code (160) may first be coded / decoded in one or more passes and then participate in the coding / decoding of the second binary code. The bins of the first binary code (160) may be coded using context-adaptive entropy coding, but the bins of the second binary code (162) may be coded using an equal-probability bypass mode, i.e., a code rate of 1.

[0049] As illustrated in FIG. 7, the second binary code itself may be composed of a binary code including a prefix (162a) and a postfix (162b). A binary parameter (163) may control the length of the prefix (162a). For example, the binary parameter (163) may be an Exp-Golomb order or a Rice parameter.

[0050] Additionally, the binaries (165) of the binarization (161) belonging to the first binarization code may be coded using context-adaptive binary coding. This is exemplified using hatching in FIG. 6. Other binaries (165) belonging to the second binarization code (162) may be coded using bypass mode and are exemplified without hatching in FIG. 6 and FIG. 7. Many embodiments described later relate to context selection. This context selection may then be used, for example as a specific implementation, to context-adaptively code one or more of the binaries (165). Further details are presented below.

[0051] Before beginning the description of the embodiments of the present application, additional concepts that are sometimes commonly used by some of the embodiments described later will be presented. In particular, according to some embodiments, the conversion factor block (92) is divided into partitions (120) as exemplified by dashed lines in FIG. 8 to signal that all conversion factors within a specific partition (120) are zero, and thus may be skipped when coding the conversion factors (91). If such partitioning for zeroing signaling is used, a partition-by-partition indication is signaled to the data stream (14) to signal whether, for a specific partition, all conversion factors (91) within that partition are zero, or whether this is not true and the conversion factors (91) within that partition are coded into the data stream. Unless otherwise taught in the description of the various embodiments of the present application, one applicable to all embodiments applied thereafter is that the scan portion (94) may be selected in such a way that all transformation coefficients (91) of a specific partition (120) are scanned or traversed by the scan path (94) before the transformation coefficients of any other partition. When the latter possibility is combined with the multiple path alternatives mentioned above in relation to FIGS. 6 and 7, the transformation coefficients (91) within one partition (120) may be fully coded before coding / decoding proceeds to the transformation coefficients of any other partition, that is, the bins of the empty string (161) may be fully coded. That is, before coding / decoding proceeds in one or more passes to encoding / decoding the transformation coefficients of the next partition along the scan path (94) in the same scan order as the scan order (118), one or more passes are used to traverse the encoding / decoding of the transformation coefficients (91) of a specific partition (120) along the scan path (94).

[0052] Another concept using partitions is described in relation to FIG. 9. Several embodiments described later use this concept. Here, the transformation coefficient block (92) is partitioned into partitions (120) for context use. Context sets (110a to 110c) are shown on the right side of FIG. 9. One set of contexts (110a-c) is "P # It includes one or more contexts indicated by ". Each "context" may alternatively be represented as a context model and represents a probability estimate that forms the basis of entropy coding using this context. For example, in the case of binary entropy coding, i.e., entropy coding of bins, the context represents, for example, a probability estimate for one of the bin values, and accordingly, automatically represents a probability estimate for other bin values ​​as well. Each partition (120) has one of the context sets (110a-c) associated with it. For example, the transformation coefficients (91) within a specific partition (120) are entropy coded / decoded using a set selected from the context sets (110) associated with that partition (120), the selection from this set being based, for example, on the evaluation of coefficients near the current coefficient (91).

[0053] The latter situation is illustrated, for example, in FIG. 10. The currently encoded / decoded transformation coefficients are highlighted here using a cross (130). "Currently encoded / decoded" may mean that one bin of these coefficients (130) is currently encoded / decoded when using the multi-pass approach mentioned above. Reference numeral (132) indicates a local template positioned on / around the coefficient (130), and coefficients within or positioned therein may be evaluated or used to determine the context to be used to encode the coefficients (130) or the currently encoded / decoded bin. Additionally, some embodiments directly mention the use of the local template (132). However, embodiments described later that do not specifically mention the use of the local template (132) to determine the context to be used to encode the currently encoded / decoded coefficient (130) may actually be modified to also use this method of selecting the context. The selection of the coefficient (130) may be made from a set associated with the partition (120a) where the current coefficient (130) is located, according to the embodiment of FIG. 9. However, the use of partitioning for a partition-specific design of context sets is merely an example and may not be used by all embodiments.

[0054] Another aspect, which is directly related to some of the embodiments described later and can also be used to modify or extend other embodiments, relates to the contents of a block (84) related to a number of components, namely a number of color components. Examples of color components are, for instance, a luma / coma representation, in which the contents of the block (84) are spatially sampled for the luma (L) and for two coma components (C1, C2), but other embodiments may also exist. Thus, the block (84) will consist of a sample array (841) for a first color component, a sample array (842) for a second color component, and a sample array (843) for a third color component. Here, the aspect is that each component can be processed individually. That is, for each component, each sample array of the block (84) can represent the prediction residual for each component of the predictor. Each sample array undergoes a transformation (90), preferably the same transformation, and yields a corresponding transformation factor block (921 to 923). However, it should be noted here that the sample resolutions of the various color components may differ from one another, and thus the sizes between sample arrays and the sizes between conversion factor blocks may likewise differ. Each conversion factor block (921 to 923) undergoes decoding / encoding (108) into / from the data stream (14), and each conversion factor block (921 to 923) undergoes the measurements, steps, and processes described in relation to FIGS. 3 through 9. It should be noted that the multi-component concept described in FIG. 11a may be applied to the embodiments described below, which do not further mention any color component aspect, but the embodiments described below may likewise be used in cases or implementations where only one component exists.

[0055] Another aspect not yet described above is the fact that the coding of a specific transformation factor block (92) and its decoding accordingly may begin with an indication (190) of whether any of the transformation factors (91) of the transformation factor block (92) are, in any way, not zero. If they are not zero, the transformation factor block is known to the decoder as being globally zero. Any combination of the concepts outlined above is used to code the transformation factor block only when this indication (190), exemplified in FIG. 11b, indicates that at least one non-zero factor (91) exists. In a similar manner, FIG. 11c exemplifies similar indications (150) for partitions (120), namely, these indications indicate zeroing within the associated partition (120). FIG. 11d exemplifies a final position indication (114) which may indicate a position (98). FIG. 11e illustrates individual contexts currently used to entropy encode / decode (116) a quantization level (106) of a current coefficient selected from an associated context set (110), and FIG. 11f illustrates a coding order (102) in which the quantization levels (106) are coded into a data stream, i.e., an order occurring from a path (94) traversing in a direction (118).

[0056] The reader should now be ready to understand and follow the embodiments of the present application described below. An embodiment relating to the first aspect of the present application is described in relation to FIG. 12. Here, the encoder and decoder encode / decode the image (12) into a data stream (14) in a special way in terms of performing entropy coding of different transformation factor blocks of different color components of one block (84). FIG. 12 illustrates the block (84) as being composed of sample arrays (841, 842) by being composed of two color components, but it is evident that the configuration can encompass three color components as described in relation to FIG. 11a. As described with respect to FIG. 11a, the encoder obtains a first transformation factor block (921) and a second transformation factor block (922) by applying a transformation (90) to a block (84) of the image (12) individually for the first color component and the second color component, respectively, and likewise, the decoder derives a block (84) of the image by performing an inverse transformation individually on the first transformation factor block (921) and the second transformation factor block (922), respectively. With respect to the entropy encoding and entropy decoding of the second transformation factor block (922) to / from the data stream (14), the encoder and decoder operate context-adaptively using a context that depends on the first transformation factor block (921). The dependency is exemplified by the arrow (300) in FIG. 12.

[0057] For example, the context determined using the dependency (300) for the first transformation coefficient block may be related to the partition-specific indications described above in relation to FIG. 8, indicating whether the transformation coefficients (91) of a specific partition (120) are all zero or contain at least one non-zero coefficient—wherein the quantization levels of the transformation coefficients (91) within that partition are entropy encoded / decoded into / from the data stream (14), whereas this entropy encoding / decoding is skipped for other partitions that indicate all-zeroness. Such an indication for the partition (120) may be a flag and may be called a coded subblock flag. The context selected for such a coded subblock flag for a specific partition / subblock (120) may be selected, for example, according to a coded subblock flag (additional indication) coded for a subblock (partition) collocated within the first transformation coefficient block. Therefore, the selection can be made among two available contexts: if the flag for the partition collocated to the first transformation coefficient block (921) is 1, the first available context is used, and if the flag for the partition collocated to the first transformation coefficient block (921) is 2, the second available context is used for the flag of the specific partition of the second transformation coefficient block (922). However, the dependency (300) for the specific indication of the specific partition of the second transformation coefficient block (922) can be implemented in a different alternative way. For example, the context selected for the specific indication of zeroing for the specific partition (120) of the second transformation coefficient block (922) may depend on the quantization levels of the transformation coefficients of the first transformation coefficient block, such as the transformation coefficients within the collocated partition.

[0058] That is, without explicitly mentioning it above, it can be assumed that the two transformation factor blocks (921, 922) may have the same shape and size or the same aspect ratio for horizontal and vertical dimensions, and that the partitioning into partitions (120) may be the same for these blocks (921, 922) or may match when scaling one transformation factor block (921) to another transformation factor block.

[0059] Alternatively or additionally, the dependency (300) may be related to the global indication of the zeroing of the transformation coefficients described above. That is, the indication coded into the data stream (14) or decoded from the data stream (14) for the transformation coefficient block (922) will then depend on the first transformation coefficient block (921) for the context in which it is used, through the dependency (300). If this indication indicates that the transformation coefficients (91) of the second transformation coefficient block (922) contain at least one non-zero coefficient, the quantization levels of the transformation coefficients (91) of the block (922) are entropy-encoded into the data stream (14) and entropy-decoded from the data stream (14); however, if the indication indicates that all transformation coefficients within the second transformation coefficient block are zero, the coding / decoding of the block (922) is performed by the decoder by inferring that all transformation coefficients (91) of the second transformation coefficient block (922) are zero. For example, the dependency (300) may render a context used for the indication of zeroing for the second transformation factor block (922) according to the corresponding indication coded / decoded for the first transformation factor block (921) in that case. Alternatively, the dependency (300) may render a context used for the indication of zeroing for the transformation factor block (922) according to the quantization levels of the transformation factors of the first transformation factor block (921). For example, if the indication for the first transformation factor block (921) indicates non-zeroing of at least one transformation factor of the first transformation factor block (921), the first context may be selected for the indication for the second transformation factor block, and if the corresponding zeroing indication for the first transformation factor block (921) indicates zeroing of all transformation factors of the first transformation factor block (921), the second context may be selected for the indication of zeroing for the second transformation factor block (922).

[0060] Additionally or alternatively, the dependency (300) may be associated with the position indication of the end coefficient position (98). There are other possibilities for signaling this position indication in the data stream (14). For example, the position indication may be signaled in the data stream as a one-dimensional address or rank representing the position (98) as the address or rank of the position (98) among all coefficient positions when measured forward (116) along the scan path (94). Alternatively, the position indication may represent the position (98) within each transform coefficient block by two coordinates representing the position (98) in terms of horizontal and vertical positions, respectively, within the transform coefficient blocks (921, 922). Now, in addition to or alternatively to the possibilities for implementing the dependency (300) already discussed above, the dependency (300) may affect the final position indication of the position (98) for the second transform coefficient block (922). In particular, the context used to encode this mark may depend on the first transformation factor block, such as the last position mark of the position (98) for the first transformation factor block (921) or the evaluation of the position or distribution of the effective transformation factors (91) within the first transformation factor block (921).

[0061] The way the last position marks of the blocks (921, 922) are encoded may also involve binaryization into a binary code, and a context in which the dependency involves dependency (300) may be used for one or more bins such as the first bin of the binary code of the last position mark.

[0062] Additionally or alternatively, a dependency (300) may be applied to one or more bins such as the first bin of the binarization (161) of the conversion coefficients (91) within the second conversion coefficient block (922), that is, the context used to encode one or more bins of the binarization (161) may depend on the first conversion coefficient block (921), such as the corresponding bin of the binarization of the collected conversion coefficients of the first conversion coefficient block (921).

[0063] To conclude the description of the embodiment(s) of FIG. 12, the two color components shown in FIG. 12 may be coma components, and in particular, the color component of the second transformation coefficient block is C r Meanwhile, the color component of the first conversion coefficient block (921) may be C b It could be.

[0064] The following embodiment relating to an encoder for encoding an image into a data stream and a decoder for decoding an image from a data stream relates to a further aspect of the present application, namely, to increasing coding efficiency by providing a specific context set (110) for specific transform coefficient positions, namely, transform coefficient positions (104) which may be DC coefficient positions as outlined above. Although the embodiment is presented below in relation to this transform coefficient position (104), it should be noted that the provision of a specific coefficient individual context set may be applied alternatively or additionally to a coefficient position (98), namely, the position of a non-zero coefficient furthest from the coefficient position (104) measured along the scan path (94). "Context individual" will indicate that the context set is not used for any other coefficient (91) of the same transform coefficient block, or even that none of the contexts included in the context set of the corresponding transform coefficient position, here 104, are used to code any other transform coefficient within the same transform coefficient block.

[0065] An embodiment may be illustrated based on FIG. 5. An encoder and a decoder are configured to encode / decode a conversion coefficient block (92) into / from a data stream using a scan pattern (94) by encoding / decoding data (96) (compare FIG. 4) representing a coded set (100) of conversion coefficients (91) traversed by a scan pattern (94) from a first end coefficient position (98) to a second end coefficient (104) in a predetermined direction (102), wherein the data (96) includes quantization levels of non-zero conversion coefficients within the coded set (100) of conversion coefficients, i.e., quantization levels of non-zero conversion coefficients together with at least those positions or quantization levels of all conversion coefficients of the coded set (100). In relation to FIG. 5, the signaling of the position indication of the position (98) described above is to be understood merely as an example for the embodiment currently being described, and the signaling of this position indication may be alternatively excluded in this embodiment, for example, it should be understood that the coded set (100) necessarily covers all transformation coefficients (91) of the transformation coefficient block (92) or synchronizes the encoder and decoder when determining the coded set (100).

[0066] In any case, according to the present embodiment, the encoder and decoder are configured to context-adaptively entropy-encode / entropy-decode quantization levels using a specific (first) set of contexts (110) for a terminal coefficient position (104) which may be a DC coefficient position, and this set of coefficients (110) used for coefficients at position (104) is coprime to one or more sets of contexts used for any other transform coefficients (91) in the coded set (100) of transform coefficients. According to a specific example, for instance, one context set (110) is used to entropy-encode the quantization levels of the DC transformation coefficients at position (104), while another / different context set (110) is used to entropy-encode / decode the quantization levels of all other transformation coefficients (91) of block (92), where the "difference" between the context sets may involve that none of the contexts included in the context set of the coefficients at position (104) are included in the context set of the other transformation coefficients, and vice versa.

[0067] According to the above-mentioned alternative in which a position-specific context set is used for the quantization level of the transform coefficient at position (98), all other transform coefficients (91) may use the same context set, which is different from the context set used for the transform coefficient at position (98) next. And in the case of the additional alternative also mentioned above, position-specific context sets may be used for both the transform coefficients at position (104) and the transform coefficients at position (98), and these may be different from each other as well as from the context set to be used jointly later, for example, to entropy encode / decode the coefficients (91) between positions (104, 98) along the scan path (94) and included in the coded set (100).

[0068] Accordingly, a set of position-specific contexts provided individually for a single coefficient position, i.e., the coefficient position (104) here, may be used, for example, to encode one or more bins of the binarization (161) of the transformation coefficients as the position (104). These one or more bins (165) may be one or more bins (165) of the first binarization code (160). This may be, for example, the first bin in the binarization (161) order, which indicates, for example, whether the transformation coefficient at the corresponding position is valid, i.e., not zero. Likewise, contexts of a set of disjointly used for other transformation coefficients will be used for one or more corresponding bins of the binarization (161) of these other transformation coefficients.

[0069] In relation to FIG. 5, the aforementioned context sets may subsequently serve as a repository, and it should be recalled that a specific context from this repository is subsequently used for the currently encoded / decoded transform coefficient. That is, the position-specific context set (110) for the transform coefficient at position (104) will serve as the basis for the repository from which the context actually to be used for entropy encoding / decoding one or more bins of the coefficient level binarization (161) of the transform coefficient at position (104) is selected using the local template (132) illustrated in FIG. 10. The local template (132) illustrated in FIG. 10 includes five transform coefficient positions in the vicinity of the currently entropy encoded / decoded transform coefficient (130), namely, coefficient positions to the right and below, diagonally to the right and bottom of the currently entropy encoded / decoded transform coefficient (130), as well as the position closest to the right but to the side and the position closest to the side but below the coefficient (130). However, this is just an example, and the number and location of positions covered by the local template (132) may be selected differently.

[0070] For completeness, the embodiments just described, in which a context-specific set of contexts is used for one or two specific coefficient positions (104 and / or 98), can be combined with the possibilities discussed in relation to FIG. 8, so that, accordingly, for the remaining transformation coefficients, partition-specific sets of coefficients (110) can be used in partitions (120) in which the transformation coefficient block (92) is partitioned. For example, FIG. 13 illustrates a position (104) and shows that the transformation coefficient block (92) is partitioned into several partitions, namely, four partitions (120) in this case. According to the present embodiment, while a position-specific context set is used to entropy-encode / decode the quantization level of a coefficient at position (104), the context set to be used for all other transformation coefficients (91) of block (92) will be determined to be associated with the partition where each transformation coefficient (91) is located, and these context sets may be mutually exclusive, as well as mutually exclusive with respect to the context set specifically provided for position (104).

[0071] In the example of FIG. 13, the partitions extend diagonally along the diagonal (122) of the conversion factor block through the end factor position (104), such as the DC factor position, and the factor position at the opposite corner of the conversion factor block, i.e., the bottom right corner of the factor block (92), which corresponds to the highest frequency in both horizontal and vertical directions.

[0072] It is evident that the embodiment just described in relation to FIGS. 5 and FIGS. 13 can be easily combined with the embodiment described in relation to FIG. 12.

[0073] The embodiments described below relate to further aspects of the present application for an encoder for encoding an image into a data stream and a decoder for decoding an image from a data stream, which are specific to context selection or context determination for a currently encoded / decoded transform factor. The embodiments are best understood and described in relation to FIG. 10. Specifically, the encoder and decoder are configured to context-adaptively entropy-encode / decode the quantization level of the currently encoded / decoded transform factor (130) of a transform factor block (92) by using a context determined based on the sum and / or number of valid transform factors among one or more previously encoded / decoded transform factors located at positions determined by a local template (132) in which a template is positioned on the currently encoded / decoded transform factor (130). According to the present embodiment(s), it may be possible for the transformation coefficients, or more precisely their quantization levels, to be completely entropy-coded / decoded in sequence, that is, before proceeding to the entropy encoding / decoding of the next transformation coefficient (91); however, it should be noted that the concept discussed above may also be applied, so that the entropy encoding / decoding of the quantization levels of the transformation coefficients occurs using context-adaptive binary entropy coding of bins of the binarization of quantization levels in multiple passes. Context-adaptiveness may be limited to the first binarization code (160) as discussed above. That is, it is noted that, on the one hand, a context that is selected / determined using a local template (132) and forms a sum of effective transformation coefficients among one or more transformation coefficients within the local template (132) and / or counts a number may be applied to one or more bins of the first binarization code (160). Additionally, multi-pass coding may be used. That is, the current entropy encoded / decoded conversion coefficient (130) may be the current entropy encoded / decoded conversion coefficient of the bin.Additionally, since the bins (165) of the binarization (161) of the transformation coefficients (91) of the block (92) are coded sequentially in multiple passes, the bins (165) of the transformation coefficients (91) or the binarization (161) encompassed by the local template (132) may have been incompletely entropy encoded / decoded so far. Accordingly, with respect to the context of the coefficient (130) or, more precisely, the determination of its currently encoded / decoded bin, the sum can be calculated as follows. In particular, the sum of the coefficients (91) within the local template (132) can be calculated as the sum of the predetermined absolute values ​​of the coefficient levels of one or more transformation coefficients located within the local template (132), wherein the coefficient level of each coefficient within the local template (132) must have a predetermined absolute value at least according to the bins of the binarization of the corresponding coefficient level of the corresponding transformation coefficient that have been encoded / decoded so far in the multi-pass bin encoding / decoding process. As a specific example, one can imagine that the binarization (161) includes the following bins: the first bin of the first binarization code (160) may indicate whether the quantization level of the transform coefficient associated with the binarization (161) is valid, i.e., not zero, as already mentioned above. The second bin in the bin order of the binarization (161) may also belong to the first binarization code (160), i.e., may also be subject to context-adaptive entropy encoding / decoding, and may indicate whether the absolute value of the quantization level of the coefficient associated with the binarization (161) is greater than 1. The third bin in the bin order may also belong to the first binarization code (160) and may indicate whether the absolute value of the transform coefficient is odd or even, i.e., the parity of the absolute value. The bin much further down in the bin order, namely the fourth bin, may also belong to the first binarization code (160) and may indicate whether the current conversion factor is greater than 2 when the parity is even or greater than 3 when the parity is odd.Other bins may belong to the second binary code (162). Thus, depending on the number of bins that have already been entropy encoded / decoded for the coefficients within the local template (132), the absolute quantization level value may be fully known and form the sum addend, or the minimum absolute value may be used instead and derived based on the bins that have already been encoded / decoded for the quantization level of the corresponding coefficient.

[0074] As far as the number of valid coefficients in the local template is concerned, the first bin is already encoded for the coefficients in the local template and is available in the same way as the example just outlined, and at least such coefficients in the local template precede the currently encoded / decoded coefficients (130) in reverse (118) along the scan path.

[0075] In a specific example, the context determination may be based solely on the sum. In another example, the context may be determined / selected based solely on the number or count of valid coefficients within the template (132). Yet another example is that the difference between the sum and the number of valid coefficients is used. Additionally, alternatively or furthermore, the context determined in the manner just outlined may be used for one or more bins of the first binary code (160), such as any of the bins just specifically exemplified or more than one bin.

[0076] The embodiment(s) just described in relation to FIG. 10 can be combined with any of the embodiments described above.

[0077] The embodiments described below relate to further aspects of the present application and may be combined with any of the previously described embodiments of the aspects of the present application. Description may also be provided in relation to FIG. 10. As with the decoders and encoders related to the previously described embodiments described in relation to FIG. 10, the embodiments described now relate to an encoder for encoding an image into a data stream and a decoder for decoding an image from a data stream, and an encoder and decoder configured to encode / decode a transform factor block (92) into / from a data stream. However, according to the embodiments described now, local templates and, in particular, one or more previously encoded / decoded transform factors located at those positions are used to set binarization parameters. The absolute value of the quantization level of the currently encoded / decoded transform factor (130) of the transform factor block (92) is encoded / decoded in such a way that it is binarized using binarization parameterized using binarization parameters. The sum and / or number of valid transformation coefficients among one or more previously encoded / decoded transformation coefficients are used for determination.

[0078] That is, the description of the present embodiment is largely consistent with the previously provided description regarding the determination of context based on the evaluation of one or more previously encoded / decoded transform coefficients of the local template in the present embodiment(s), but focuses on the setting of the binarization parameter. It should be noted that, according to the present embodiment(s), the binarization in which the absolute value of the quantization level of the transform coefficients of the transform coefficient block is used for encoding / decoding and the binarization parameter is set using the local template (132) may be the binarization (161) described above in relation to FIG. 6, but is not necessarily so. That is, the binarization meant here may have a first part or a transpose part or a first binarization code in which the bins are context-adaptively coded, but this may also mean that the entire transform coefficient block is coded, for example, using a bypass mode, or that the binarization meant here consists only of a second binarization code (162). The binarization parameter determined based on the evaluation of the local template (132) may determine the length of the transpose part (162a). A lookup table may be used to map the sum of the absolute values ​​of the quantization levels of one or more previously encoded / decoded transform coefficients within a local template (132) to a binarization parameter. The binarization parameter may be determined in such a way that the binarization parameter continuously increases or decreases over the transform coefficients.

[0079] With regard to additional details, reference is made to the description of the previously described embodiment, namely regarding the determination of the context on which the context-adaptively entropy-coded bins are coded for binarization. These details provide information on how the sum can be determined when traversing the transformation coefficients using a multipath approach. In this regard, it should be noted that, as with the present embodiment as well as the previous embodiment, the multipath approach can be implemented in a manner where the multipath approach is applied per partition. However, in any case, since the scan path (94) generally points from the bottom right to the top left of the transformation coefficient block (92), the template positions illustrated in FIG. 10, namely the template positions to the right and / or bottom of the currently encoded / decoded transformation coefficient (130), tend to be already encoded / decoded when the currently encoded / decoded transformation coefficient or its current bin is to be encoded / decoded into / from the data stream.

[0080] The embodiment(s) described below relate to an encoder and a decoder for encoding / decoding an image into / from a specific data stream in relation to a cutoff value (164). Thus, the following description may be best understood based on a combination of FIG. 4 and FIG. 6. Here, the encoder and the decoder are configured to encode / decode the transform factor block (92) into / from a data stream using the scan pattern (94) shown in FIG. 5 by encoding / decoding the absolute values ​​of the quantization levels (106) of the transform factors (91) of the transform factor block (92) in such a manner that the first binary code (160) below the cutoff value (164) and the second binary code (162) above the cutoff value, in which the codeword of the first binary code is transposed with respect to the cutoff value, are used to binarize. As becomes apparent from the above description of possible general implementation examples or concepts, the bins of the first binary code (160) are coded using context-adaptive binary entropy coding, whereas the second binary code may be coded using equal probability bypass. According to the present embodiment, the cutoff values ​​for the transformation coefficients (91) of the transformation coefficient block (92) are adaptively changed according to the previously encoded / decoded transformation coefficients.

[0081] According to a specific embodiment, the adaptive change may be implemented as follows. The transformation factor block (92) may be partitioned into partitions (120) having a scan path (94) that follows through the transformation factor block (92) to scan the transformation factors (91) by partition. That is, all transformation factor positions within the partitioned block will be scanned / traversed by the scan path (94) before the next partition. Furthermore, the encoding / decoding of the transformation factors (91) will be performed in such a way that the bins (165) are coded in the passes. The bins (165) of the first binary code (160) will be subject to context entropy encoding, while the bins of the second binary code (162) will be subject to equal probability bypass mode. The bins of the first binary code (160) will be encoded / decoded before the bins of the second binary code (162). In this coding environment, adaptive changes to the cutoff value can be implemented as follows: bins (165) of the transform coefficients (91) within the current partition are scanned in one or more leading passes. However, the number of bins (165) coded using context-adaptive binary entropy coding / decoding may be limited. That is, in that case, it is assumed that one pass is used to code the bins (165) of the first binary code (160). Then, the bins (165) of the first binary code (160) of the absolute value of the quantization level of the coefficients (91) within the current partition (120) will be coded by coefficient. Then, the number of bins coded in this way for individual coefficients (91) will vary according to the value of the quantization level of each transform coefficient. At a specific coefficient, the maximum number of context-adaptive binary entropy coded bins (165) of the first binary code (160) will be reached. In this way, the first pass will be completed. The next pass will scan the counts (91) of the current partition again from the beginning, that is, from the count position within the corresponding partition (120) furthest from the position (104) of the block (92).For all coefficients prior to reaching the maximum number of context-adaptive binary entropy-coded bins, the cutoff value (164) will be the same as the initial value for the cutoff value. For coefficients where the bins (165) of the first binary code (160) cannot be perfectly context-adaptive binary entropy-coded / decoded due to reaching the maximum number of context-adaptive entropy-coded bins, the cutoff value is reduced in a manner corresponding to the number of coded bins (165) of the first binary code (160) for that coefficient. In particular, the cutoff value is reduced so that this coefficient corresponds to the number of bins coded for that coefficient, and the second binary code (162) for this coefficient simply codes the "remainder" above the maximum absolute value associated with the bins (165) of the first binary code. For subsequent coefficients within that partition (120), the cutoff value will be even 0 and the binarization (161) will consist only of a second binarization code (162).

[0082] In the example just outlined, the adaptive change of the cutoff value played a central role. In the embodiment(s) outlined thereafter, the setting of the cutoff value is used to increase coding efficiency; however, according to this embodiment, the setting may be an initial setting of the cutoff value when the conversion factor block is coded and when it begins to be decoded, respectively, rather than an adaptive setting of the cutoff value during the coding / decoding of the conversion factor block; according to this embodiment, an option for an adaptive setting of the cutoff value also exists. That is, according to these embodiment(s), an encoder and a decoder for encoding / decoding an image from / to a data stream encode / decode a conversion coefficient block (92) by encoding / decoding the absolute value of the quantization level (106) of the conversion coefficients (91) of the conversion coefficient block (92) in such a manner that the conversion is performed using a first binary code below the cutoff value (164) and a second binary code (162) above the cutoff value, in which the codeword of the first binary code (160) is transposed for the cutoff value (164). The cutoff value (164) is set according to one or more of the following:

[0083] Size of block (84),

[0084] Color component of block (84),

[0085] A prediction mode that forms the basis of a prediction signal having a prediction residual represented by block (84), such as an intra or inter coding mode,

[0086] The transformation that forms the basis of the transformation coefficient block (92),

[0087] Quantization parameters used to quantize the conversion coefficient block (92),

[0088] Energy measurements of previously encoded / decoded transformation coefficients, and

[0089] Evaluation of previously encoded / decoded transformation coefficients.

[0090] Additionally, the configuration described herein may be made for each coefficient (91) or alternatively, only for the first coefficient coded, such as the coefficient at position (98), or alternatively, only for the coefficients first traversed by the scan path (94) within each partition, i.e., the coefficient as position (98) within the first traversed partition (120), and likewise, only for the coefficients first traversed in any subsequent partition (120).

[0091] The aforementioned alternatives for setting the cutoff value (164) can be implemented in various ways. However, in principle, the setting of the cutoff value can also be implemented as an example, in the manner outlined in relation to the previous example: that is, for each coefficient, the cutoff value (164) can be set according to the number of bins of the first binary code that have already been context-adaptively coded / decoded. This number can be used as a measure of energy, or this number can be the result of evaluating previously decoded / encoded transformation coefficients. If the number or count of already encoded / decoded bins of the first binary code (160) is performed per partition (120), this number or count will represent a measure of energy of the previously encoded / decoded transformation coefficients within the current partition (120). However, regarding the present embodiment for setting the cutoff value (164), as well as the previous embodiment(s) for adaptively changing the cutoff value (164), there is also a change in the maximum number of context-adaptive binary entropy-coded bins: it should be noted that instead of defining this individually for each partition (120), the maximum number can be determined globally. This would mean reaching the maximum number for a single specific coefficient (91) positioned in only one partition (120) and for the entire transformation coefficient block (92). For all previously scanned partitions (120), the cutoff value corresponds to the initial value, while the reduction of the cutoff value (164) to zero occurs only within that partition. The next partitions will have a cutoff value (164) that is zero for all of their coefficients (91).

[0092] There are many additional embodiments of the embodiment just outlined for setting the cutoff value (164) according to one or more of the alternatives outlined / listed above. For example, the evaluation of previously encoded / decoded conversion coefficients may be limited to coefficients (91) located in positions determined by the local template (132). That is, the setting will be made for the currently encoded / decoded conversion coefficient (130), and the setting will effectively be an adaptive change. Even alternatively, the evaluation of previously encoded / decoded conversion coefficients may be made for coefficients located in positions within the partition of the conversion coefficient block that are offset from the current position where the currently encoded / decoded conversion coefficient (130) is located and precede the current partition according to the coding order (118). In the latter alternative, the evaluation may be made for each partition (120), that is, at the start of coding the conversion coefficients (91) within each partition (120), in a kind of initial setting of the cutoff value. Further alternatives will be discussed later. For instance, an initial setting may be made according to any of the one or more alternatives listed above. Depending on the initial setting value of the cutoff value (164) that follows a predetermined threshold or exceeds a predetermined threshold, an adaptive change of the cutoff value may be enabled or the cutoff value may be kept constant. An example is, for instance, that the cutoff value is initially set according to a quantization parameter. If the quantization determined by the quantization parameter is finer than the predetermined threshold, an adaptive change may be enabled and the cutoff value may be adaptedly changed and thus modified compared to the initial setting according to previously encoded / decoded transform coefficients, but if the quantization parameter corresponds to a more approximate quantization, the cutoff value is kept constant.

[0093] In the following, the embodiments of the present application relate to aspects of the present application that specifically deal with the shape of a local template (132). That is, according to the embodiments, the encoder is for encoding an image into a data stream (14), and the decoder is for decoding an image from the data stream. The encoder and the decoder are configured to entropy encode / decode the quantization level (106) of the currently encoded transform coefficient (130) of the transform coefficient block (92) by using a context determined independently of the previously encoded transform coefficients (91), based on one or more previously encoded transform coefficients located at positions determined by the local template (132) positioned at the currently encoded / decoded transform coefficient (130), or if the local template (132) is deactivated. In particular, according to the embodiment of FIG. 14, the shape of the local template (132) is set or the local template (132) is deactivated accordingly, depending on 1) previously encoded / decoded conversion coefficients (91), and / or 2) the size of the block (84), the color component of the block (84), the position of the terminal coefficient position where a non-zero last conversion coefficient is caused when traversing the scan pattern (94) along the forward direction (116), i.e., 98, and one or more of the conversion (90) that forms the basis of the conversion coefficient block (92), which means that the local template (132) disappears and does not contain any coefficient positions at all. FIG. 14 illustrates two shapes (170, 172) for the local template (132) as an example. One, i.e., 172, corresponds to the example illustrated and described above in relation to FIG. 10. The other one, namely 170, includes only the immediately right neighboring coefficient position and the immediately neighboring position below the current entropy encoded / decoded coefficient (130).

[0094] According to the first alternative in which the dependency is related to previously encoded / decoded transformation coefficients, their dependency on previously encoded / decoded transformation coefficients may be limited to one or more such previously encoded / decoded transformation coefficients located at positions determined by a smaller shape for a local template, such as a local raw template (170). Depending on the evaluation of the coefficients positioned in the raw template (170), for example, it is determined whether this raw template (170) is used for context determination, or whether a larger raw template (172) is used. The larger raw template (172) extends further from the currently encoded transformation coefficient (130) than the raw template (170). The second raw template (172) may or may not include positions also included by the first raw template (170).

[0095] The determination of the shape to be used based on the evaluation of coefficients within the smaller template (170) may be made according to the sum or number of valid transformation coefficients among one or more previously encoded transformation coefficients within the original template (170). A decision to disable may even be made based on the evaluation of one or more coefficients within the original template (170). Furthermore, a decision to disable may also be made depending on whether the count of previously encoded / decoded transformation coefficients within the transformation coefficient block, or the accumulator value derived from previously encoded / decoded transformation coefficients within the transformation coefficient block, exceeds a predetermined amount, which is greater than a certain threshold. Additionally, or alternatively, if the count of previously encoded / decoded transformation coefficients (91) within the insignificant transformation coefficient block exceeds a predetermined amount, the local template may be disabled.

[0096] Regarding context dependency for one or more coefficients within the local template (132), the details presented above may be valid, namely, the sum and / or number of valid coefficients within the local template may be used. Additionally, the context derived based on the local template (132) may be restricted to use in the bins (165) of the transposition part (160). Furthermore, the above remarks regarding the use of multiple passes and the relevance to the determination of consensus within the local template (132) may also apply here.

[0097] Referring to a specific embodiment, the local template (132) is disabled for the chroma component, whereas the local template (132) is not disabled for the lumina component.

[0098] Further embodiment(s) of the present application relate to the sizes of partitions (120) when using partitions (120) as a granularity for determining the use of a context set. For example, FIGS. 8 and FIGS. 13 are referenced. Embodiments(s) of the present application related to this aspect describe an encoder and a decoder for encoding / decoding features into / from a data stream (14). For each partition (120), they encode / decode the transform coefficients (91) of a transform coefficient block (92) by context-adaptive entropy encoding / decoding the quantization levels (106) of the transform coefficients (91) of the transform coefficient block (92) using a context set (110) associated with each partition. However, specifically, the partitions (120) into which the transformation coefficient blocks (92) are subdivided are shaped such that 1) they do not coincide with each other as shown in FIG. 13, for example, but coincide with each other in FIG. 8, and / or 2) the entropy encoding / decoding of the quantization levels of the transformation coefficients (91) is performed and can be signaled in, for example, a data stream (14), and are shaped according to a scan pattern that is essentially determined for each current block (84 / 92), for example, and / or 3) the size of the block (84) and / or 4) the explicit partition shape information signaled in the data stream (14). In this embodiment, for each coefficient (91) within a specific partition (120), it may be possible to determine which of the sets of contexts associated with each partition is actually used to entropy encode each coefficient or its current bin, for each coefficient (91). The partitions (120) of the current embodiment may additionally serve as indicators of whether the coded block flags are zero, i.e., whether all transformation coefficients (91) within each partition are zero, as described above, in the partitions transmitted in the data stream.

[0099] Further embodiment(s) of the present application relate to a further aspect of the present application, namely, an aspect in which a partition is used for zeroing indication, but the indication (150) indicates, for a partition, whether all transformation coefficients within each partition are coded into a data stream or whether the coding of all transformation coefficients within each partition is skipped and all transformation coefficients within each partition are zero. In particular, according to this aspect of the present application, the indication is interpreted in a manner indicating whether all coefficients are coded. Each of these partitions may have a set of contexts associated with them, using which the transformation coefficients within each partition are encoded if so indicated by the indication. That is, the zeroing indication at the partition level here indicates that, if so set, all transformation coefficients within a specific partition are coded and this situation no longer needs to be questioned. Even the valid flag of the last coded coefficient of the partition where the indication (150) signals the coding of coefficients cannot be inferred based on all previous coefficients being zero.

[0100] Further embodiment(s) of the present application relate to further aspects of the present application, namely, aspects in which a partition mode is signaled in a data stream for a conversion factor block (92), which indicates whether partitioning into partitions (120) to be used to associate partition individual context sets (110) is applied. More precisely, the embodiment(s) describe an encoder and a decoder for encoding / decoding an image into / from a data stream by encoding / decoding the partitioning mode of a transform factor block (92) into / from a data stream (14) and encoding / decoding the transform factor block (92) by context-adaptive entropy encoding / decoding the quantization levels (106) of the transform factor block (92) using a context set (110) associated with each partition for each partition (120) of the subdivided partitions, and by context-adaptive entropy encoding / decoding the quantization levels (106) of the transform factor block (92) using a global set of contexts, if the partition mode is a first mode indicating that the partitioning into partitions (120) is switched on, for example, and the partitioning mode is a second mode indicating that the partitioning into partitions (120) is switched off, for example, by context-adaptive entropy encoding / decoding the quantization levels (106) of the transform factor block (91) using a global set of contexts. However, it should be noted that globality may be limited in relation to a special context set used for coefficients in position (104) and / or coefficients positioned in position (98). Additionally, in this embodiment as well, partitions (120) may additionally serve to form units transmitted in the data stream that indicate the zeroing of conversion coefficients: that is, for each partition (120), a coded subblock flag may be a signal in the data stream indicating whether all conversion coefficients (91) within each partition are zero.

[0101] Embodiments covering the last aspect of the present application may be best understood by referring to FIG. 15. Additionally, this aspect relates to the use of partitioning into partitions (120) for the association of partition-specific context sets (110) for partitions (120). An encoder and a decoder are configured to encode / decode an image from / to a data stream and, for each partition (120) of the partitions into which the transform factor block (92) is subdivided, to encode / decode the transform factor block (92) using a scan pattern (94) that sequentially traverses the transform factor blocks (91) of the transform factor block (92) by context-adaptive entropy encoding / decoding the quantization levels (106) of the transform factor blocks (91) of the transform factor block (92) using a context set (110) associated (126) with each partition (120). However, unlike previous descriptions in which the scan pattern (94) is described as scanning the transformation coefficients (91) by partition as an example, the scan pattern (94) of FIG. 15 scans the transformation coefficients (91) differently from FIG. 15, which indicates the situation by using an apostrophe, i.e., 94', for the scan pattern. The scan pattern (94') sequentially traverses the transformation coefficients (91) of the transformation coefficient block (92) in such a way that at least one transformation coefficient of the first partition traverses between two transformation coefficients of the second partition. For example, the coefficient (91') of FIG. 15 is such a coefficient: it is scanned between two coefficients of partition (120). However, this is a member of partition (1202).

[0102] However, the partition-specific association of context sets for partitions (120) does not mean that the context sets (110) associated with partitions (120) are mutually distinct or mutually disjoint for each pair of partitions (120). That is, a set of contexts can generally be associated with two partitions. For example, the interleaving of transformation coefficients (91) following a scan pattern (94) can be associated with partitions (120) having a common context set (110) associated with it. To understand this, reference is made to FIG. 16, which shows another example of a scan portion (94') traversing transformation coefficients (91) in such a way that the transformation coefficient block (92) is interleaved between the transformation coefficients of partitions (120a, 120b), where at least two partitions (120), in the example of FIG. 16, are partitioned. Here, a common context set may be associated with partitions (120a, 120b). However, in both examples of FIG. 15 and FIG. 16, the partitions may additionally be used as units, to which zeroing coding or zeroing indication, i.e., coded sound block flags, are signaled in the data stream. In the embodiments of FIG. 15 and FIG. 16, context selection from the context set (110) associated with each partition (120) where the currently encoded / decoded transformation factor (130) is located may be performed using a local template (132) as discussed above. Additionally, or alternatively, context selection from the associated context set (110) may be based on the count of previously decoded / encoded transformation factors within the current partition. The currently encoded / decoded transformation factor is located, for example, depending on whether this count exceeds one or more specific thresholds.In the example of FIG. 16, for instance, if the transformation coefficient (130) is the currently encoded / decoded transformation coefficient, the four hatched transformation coefficients may represent the previously encoded / decoded transformation coefficients of the block (92) that exceed one or more specific thresholds.

[0103] When zeroing marks or coded subblock flags are used, they may be coded into the data stream or decoded from the data stream, respectively, when the transformation factor of a specific partition (120) is first scanned. That is, the zeroing marks or coded subblock flags for a specific partition (120) may be scanned as soon as the first transformation factor (91) of that partition is scanned along the scan portion (94'). FIGS. 17 and 18 provide an example of this. FIG. 18 again illustrates an example of a transformation factor block (92) partitioned into three partitions (120a, 120b, 120c). One transformation factor (91) (interleaved) within the scan portion (94') between two transformation factors of another partition, namely partition (120b), is illustrated by reference numeral (91'). FIG. 17 assumes that this transformation coefficient (91') happens to also be the first coded transformation coefficient, that is, positioned at position (98) when using the last position indication also discussed above in relation to FIG. 5. Its quantization level (106) is coded, and then the coded subblock flag (CBF2) for partition (120b) is coded, because all transformation coefficients (91) within partition (120b) are zero. FIG. 18 assumes that all non-zero transformation coefficients (91) are indicated using hatching. The coefficient immediately below coefficient position (104) is also one of the coefficients (91') arranged in scan part order between two coefficients of another partition. This is simultaneously scanned partition (120 a It is the first coefficient of ), and accordingly, partition(120 b This partition (120) after CBF2 for )a As indicated for ), there is a zeroing indicator (CBF3). Also, since the coefficient (91') immediately below the coefficient position (104) is assumed not to be zero, CBF3 is signaled as 1, and then partition (120 a The coefficients (91) within ) are coded.

[0104] It is noted again that, as has been frequently mentioned above, the above-mentioned embodiments belonging to different aspects of the present application may be combined, such as, for instance, embodiments related to cross-color component context derivation and all other embodiments or combinations thereof, embodiments related to binarization parameter setting and embodiments related to context derivation, embodiments related to context set attributes for specific coefficients of partitions and embodiments related to context selection from such sets using local templates, etc.

[0105] Accordingly, the above embodiments relate to transform factor block coding for entropy coding in image and video compression, for example.

[0106] The above embodiments relate to the field of digital signal processing, and in particular to methods and apparatus for image and video decoders and encoders. In particular, the above embodiments produce a set of strategies applied to the entropy coding stage in image and video compression architectures utilizing partitioning techniques. Improved context modeling strategies were achieved by modifying the processing order of syntax elements. A set of techniques included, in particular: analysis of local activity for the current level, use of dedicated context models, modification of syntax processing order, evaluation of cross-component syntax elements, and exclusion or dedicated processing of coding states with very high probabilities.

[0107] In particular, the above embodiments present a set of novel techniques for entropy coding that are particularly suitable for syntactic elements associated with the residual compression stage in image and video compression architectures utilizing partitioning techniques. Such designs typically divide images into general shapes, such as rectangles or blocks. Without loss of generality, the term "block" refers to any form describing the partitioning technique applied in the description of this document. The starting point may be the maximum allowable size as described above in relation to Fig. 3, and the partitioning algorithm may further divide each block into smaller units. Such partitioning can be implemented, for example, by using a quadtree structure of the state-of-the-art High Efficiency Video Coding (HEVC) standard. For example, other strategies for partitioning using binary trees instead of quadtrees may also be possible. In addition to partitioning, new coding tools prior to the residual compression stage necessitate further advancements in existing approaches. To achieve improved compression efficiency, the following aspects are addressed:

[0108] Definition of syntax elements to alternate coding order (e.g., FIG. 15, FIG. 16, FIG. 18)

[0109] Along with the above modifications, binarization is adjusted (e.g., for adaptive changes in cutoff values).

[0110] Context modeling for each bean generated by the binarization process (e.g., refer to the concept of shape adaptation or disabling of local context templates for specific features)

[0111] Evaluation of local activity in the spatial (e.g., using local templates or setting cutoff values ​​according to specific features related, e.g., energy), frequency (e.g., shape adaptation or deactivation of local context templates for specific features and the concept of setting cutoff values ​​according to specific features related, e.g., energy), and mutual component (e.g., see FIG. 11a) domains

[0112] Exclusion of coding modes with very high probability

[0113] Dedicated processing for coding modes with very high probabilities

[0114] Individual processing of dedicated coding states

[0115] The entropy coding stage used in HEVC is Context-based Adaptive Binary Arithmetic Coding (CABAC), which consists of three internal processing stages. The first stage is the binarization stage, the second stage is the context modeling stage, and the final processing unit is the binary arithmetic coding engine. All input symbols pass through the binarization stage. The output is either the same symbol if the input is already binary, or a codeword representing the input symbol referred to as an empty string. The context modeling stage selects a context model and passes these selections to the entropy coding engine. Note that binary arithmetic coding engines implemented in CABAC, such as M-Coder, typically support a bypass mode for symbols with uniform probability. For the remainder of this document, coding using context models is referred to as normal mode, and coding without a context model is referred to as bypass mode.

[0116] In the residual compression stage, that is, for the transmission of transform coefficient levels, the input is a block composed primarily of integer coefficient values. These values ​​are typically prediction residuals transformed into the frequency domain, and the levels often represent only a fraction of the original information due to the quantization process. Since the processing in the entropy coding stage is sequential in nature, it is necessary to decompose the two-dimensional shape into one-dimensional vectors. The scan pattern specifies the mapping between the two-dimensional block and the one-dimensional vector. This scan pattern can be applied multiple times whenever only a fraction of the transform coefficient levels are transmitted in the bitstream, as in HEVC. This document describes the residual compression stage from various perspectives in the following paragraphs.

[0117] Residual compression in HEVC operates as follows: A coded block flag (CBF) syntax element is coded first to signal the insignificance of the current transform block, i.e., whether all transform factor levels are equal to zero. If a valid level exists, the last valid scan position is designated. This depends on the selected scan pattern, and all levels from the last signaled position to the first position defined by the scan pattern are coded. Implicitly, the scan order is equivalent to using the scan pattern in reverse order. Dividing the entire process into subtasks enables improvements in terms of implementation complexity and compression efficiency. Transform blocks larger than 4×4 are processed into 4x4 sub-units referred to as subblocks. The processing order of the subblocks is the same as that of the scan positions; that is, for an 8x8 block using a vertical scan pattern, as exemplified in FIG. 19, each subblock represents a normal frequency position and the transform block has a size of N / 4xM / 4 instead of N×M.

[0118] In the first step for the valid transformation block, the x and y coordinates are coded using the absolute coordinates of the last position, namely the top-left corner defined as the coordinate origin. One of three predefined scan patterns—diagonal, horizontal, or vertical—is selected. If the prediction mode is intra, the transformation block is less than or equal to 8x8, and the predictor's orientation is within a predefined range, a horizontal or vertical scan pattern is used. Next, starting from the subblock containing the last valid level, the remaining information regarding the level is coded until the level is reconstructed. The described process is repeated for the remaining subblocks up to the last subblock in the processing order, where the last subblock is always the one covering the DC frequency position. There are additional techniques embedded in the processing, such as the coding of CBFs for the subblocks. This is skipped for the last subblock because the signaling of the last scan position has already inferred validity for that subblock. Inference for the subblock covering the DC frequency position is also a special case. Additionally, sign bit hiding (SBH) techniques can be applied to each subblock. An overview is provided in Fig. 20.

[0119] From the perspective of the binarization stage, each integer value χ represents a point in the interval, and a codeword is assigned to each point in the interval. A unary code is the generally used method, where |χ| = n is a binary value b 0 to n Terminal binary value after multiplying bIt is represented as a sequence of 1s. If the set of integer value symbols is limited, the terminal binary value for the maximum input symbol is omitted, which can lead to a so-called truncated unary code. In HEVC, the binarization process for levels consists of three different methods. The first three absolute level values ​​belong to the truncated unary part. If the current absolute level is greater than the limit, the remainder of the value is binarized using Rice codes. The actual Rice code is of the order ( k It is selected by specifying ). However, the interval covered by Rice codes is also limited, and Exponential-Golomb codes are used for binary conversion for the remainder of the integer value. In HEVC, the order of the Exponential-Golomb code is always k It is equal to + 1. Transitions between codes are referred to as binarization boundaries, and both boundaries, namely the transition from truncation unary to Rice and the transition from Rice to Exponential-Golomb, are equal to the number of coded syntactic elements for the first boundary and the Rice parameter for the second boundary ( k It is adaptive depending on ). The first boundary is switched within each subblock and is equal to 3 at the start of processing for each subblock. Whenever an absolute level greater than 2 is coded, the boundary is lowered to 2. And whenever 8 absolute levels greater than 1 are coded, the boundary is lowered to 2. The result is that for all frequency positions covered by the subblock, validity is coded in normal mode, but the information on whether the level is greater than 1 is positioned only once up to 8, and the information on whether the level is greater than 2 is positioned once. Rice parameter( k The selection of ) is as follows. The parameter is equal to 0 at the start of each subblock. After the entire level is reconstructed, the absolute value is compared with a threshold, and if the absolute level is greater than the threshold, kIt is incremented by 1. However, given the main HEVC profile, k is limited to 4 and does not exceed that value. Binarization of the levels is summarized in FIGS. 21 and FIGS. 22.

[0120] Coding for each subblock is divided into multiple scan passes so that syntax elements are aligned into groups. In the first processing pass, validity information is transmitted. Next, syntax elements all greater than 1, followed by syntax elements all greater than 2, the remaining absolute levels, and finally the codes are transmitted. Binarization for the last valid position of the integer value operates as follows. Depending on the block size, the maximum value N For block sizes such as N - Limited to 1. Size( N According to ), the output of binarization is a combination of truncation unary and fixed-length binary codes.

[0121] Context modeling operates in detail as follows. Generally, binary symbols (bins) arising from fixed-length binarization, Golomb-Rice, or Exponential-Golomb codes are coded in bypass mode, while bins associated with unary codes or truncation-type unary codes are coded using normal mode. Additionally, input symbols that are already binary are usually coded in the normal mode of an arithmetic binary coder. For the CBF of the luminance component, a dedicated context model is used for residual quadtree (RQT) depths such as zero, and a different dedicated context model is used for the remaining depths. Since the CBF signaling of the chroma follows each depth instead of the leaf levels as in the luminance, a dedicated context model is used for each depth of the RQT. The last position, specified by x and y coordinates, uses the same context model for a specific number of consecutive bins depending on the block size. More specifically, instead of using a single context model for each bin, a uniform context quantizer is used to reduce the number of context models used. The coding of the levels utilizes four syntax element types that use different sets of context models. The CBF for subblocks uses three different context models depending on the validity of the adjacent right and bottom subblocks: both neighbors are valid, both neighbors are invalid, and one of them is valid. The validity flag uses four different patterns where the context model is fixedly assigned to the scan position within the subblock, and these patterns depend on the adjacent right and bottom subblocks. The remaining level information using normal modes is greater than 1 and 2, i.e., two bins of truncation unary code when the first binarization boundary is equal to 3. Since information greater than 2 is coded once, a single context model is used. For information greater than 1, the first non-terminal binary value ( bThe next context model in the set is used until 0) occurs. If this occurs, a dedicated context model is used until the end of the subblock. Additionally, a context set is selected based on the existence of levels greater than 1 in the previous subblock.

[0122] Context modeling plays an important role in the entropy coding stage. Higher compression efficiency results can be achieved by using more advanced techniques for context modeling. The embodiments presented herein improve compression efficiency for level coding by modifying the coding order so that existing information is available earlier in the context modeling stage. Possible coding structures are illustrated in FIG. 23.

[0123] According to one embodiment, the processing order is modified so that each level is fully transmitted before proceeding to the next scan position. By doing so, conditions for the context modeling stage are activated, enabling higher compression efficiency. In particular, instead of processing subblocks in multiple passes, the absolute level for each scan position is fully transmitted, requiring only a single processing pass. While the multi-scan pass approach has advantages for some hardware architectures, it also has disadvantages for other architecture types, such as in terms of memory access. An overview is shown in FIG. 24.

[0124] Since the modified processing order transmits the absolute levels completely, the entire absolute levels are available instead of just a fraction when evaluating adjacent frequency positions. Note that the modification is not necessarily fixed as described. Often, the residual energy is flat, so the absolute levels are small. The compromise operation mode enables template evaluation only when a specific amount of absolute levels greater than a specified threshold is detected (e.g., see the embodiment described in connection with FIG. 14). Additionally, for throughput reasons, the number of flags indicating that the absolute level is greater than 1 and greater than 2 may be limited to reduce the number of bins in normal mode (e.g., see the embodiment described in connection with FIG. 6 which aims to adjust or set the cutoff value). When using absolute level coding, the concept of subblocks no longer necessarily needs to rely on 4x4 shapes (e.g., the embodiment regarding the configuration of partitions in FIG. 13), but instead can be a group of consecutive scan positions. That is, interleaving of coefficients belonging to different partitions may not occur. Note that the primary reason for subblocks may be context modeling. The concept of 4x4 subblocks can be easily extended to groups of scan positions. For example, a fixed number of scan positions equal to 16 can be used to form a configuration similar to existing 4x4 subblocks. Figure 25 illustrates an example of converting a standard partition into four partitions of size 16 (left) and four partitions of the same size but following a zigzag scan (right).

[0125] Grouping into consecutive scan positions often consists only of scan positions according to a selected size. To improve compression efficiency, the size may be variable (see examples regarding the shape of partitions) but may be configured to be fixed for a given transformation block size. Another example is as follows: DC frequency positions can form a dedicated shape, and the number of scan positions formed is equal to the diagonal. Note that the size must, in some respects, be configured according to the block size. Forming a specific number of consecutive scan patterns for a group omits the neighbor relationships between frequency positions. Therefore, it is reasonable to group scan positions that are spatially close to each other together. To combine with existing scan patterns, it is not necessary to process a group of scan positions consecutively (see FIGS. 15, 16, and 18). Instead, the CBF of the group is coded when the first scan position of the group is processed. Refer to FIGS. 17 and 18. Furthermore, grouping can be pre-specified, for example, using forward drive (comparison of the aforementioned explicit partitioning information) or reverse drive approaches (comparison of the aforementioned partitioning information based on the scan pattern or block size), or a combination of both. In this configuration, an additional flag is coded to signal whether grouping is used depending on the last valid position, causing the presence or absence of CBFs for the groups. (See example using partition mode signaling.) Adapting the number of scan positions enables a reduction in the cost of levels at high-frequency positions as follows. Generally, levels at higher frequency positions have lower energy, and the distance between valid levels also tends to be greater than at lower frequency positions. Therefore, rate distortion-optimized quantization approaches quantize these levels into meaningless ones. When adaptively shaping the number of scan positions, the cost of the distance between these two levels can be further reduced.This concept can also be applied to the other side. At higher operating points with high residual energy, the cost for a group of scan positions having only effective levels can be further reduced. Additionally, the definition of the CBF for a group is changed, so that when the CBF is equal to 1, the group is coded instead of the existence of at least one effective coefficient within the group. The change in semantics reduces logic.

[0126] From the perspective of empirical data regarding 4:2:0 chroma subsampled signals, luminal residual signals are generally composed of higher energy than spatially adjacent (e.g., spatially corresponding) chroma residual signals. This fact is incorporated into the coding of levels when processing is reordered, so that levels are coded immediately after the signaling of the CBF. Then, more information can be evaluated for the context modeling stage. Context modeling for the luminal CBF and residuals can be adjusted based on the results of the chroma CBFs or chroma residuals. (Compare the explanation in Fig. 12) Furthermore, the coding of the last valid position can be separated from the level coding, that is, when all three transform blocks are valid. All three last valid positions are coded consecutively. Similar to the coding of the CBF, the coding of the last valid position can be adjusted based on the results of the last valid positions of adjacent (e.g., collocated) transform blocks. (Compare with the description in FIG. 12) An additional technique is the co-coding of the last effective position, that is, for two chroma conversion blocks, only a single last scan position is coded instead of two last effective positions, that is, the position indicator (114) will be co-coded only once for, for example, the luma and the chroma.

[0127] The first binarization boundary, i.e., the cutoff value (164), is adaptive and features a trade-off between complexity and compression efficiency. If the fixed boundary is chosen too low, lower compression efficiency is achieved, and if the fixed boundary is chosen too high, high complexity results. Note that a high boundary does not lead to a proportional increase in improved compression efficiency, but rather the improvement is very limited. However, operating points have been changed due to other partitioning strategies and additional or new coding techniques prior to the residual compression stage. For example, setting the first binarization boundary to 5 results in improved compression efficiency when additional transformation modes, such as adaptive transformations controlled forward adaptively or backward adaptively, are activated. The first binarization boundary can be used to control different coding modes. For example, the first binarization boundary can be set to 0 when high energy of levels is detected, so that no bins are coded in normal mode. (Compared to the above embodiments regarding the change and setting of cutoff values) Detection of high signal energy can be realized by analyzing the values ​​of syntax elements (or binary decisions) already transmitted for a transform block or a sub-block of a transform block. For example, detection of high signal energy can be realized by comparing the sum of absolute values ​​(of the transform block or a sub-block of a transform block) with a specific threshold. Alternatively, detection can be realized by comparing the number of bins already coded in the first scan pass for scan positions (of the transform block or a sub-block of a transform block) with a specific threshold (compared to the above embodiments regarding the change and setting of cutoff values ​​in combination with the multi-pass concept).

[0128] The configuration for common operating points is as follows. The initial boundary, i.e., the cutoff value, depends on the prediction and transform modes used and the quantization parameters, and is higher than 3, for example, equal to 5. (Refer to the examples above for the setting of the cutoff value.) For a combination of modes identical to the configuration of HEVC, the initial first boundary is set to 3. In this mode, the boundary decreases as level values ​​occur. (Refer to the examples above for changes in the cutoff value.) However, if the initial first boundary is equal to 5, it remains fixed for the group or even for the entire transform block.

[0129] Local templates for deriving context were used as part of existing techniques, but with limited information, that is, only a fraction of the absolute levels were evaluated. By applying changes to the coding structure, existing information becomes available earlier in the decoding process and thus can be used in the context modeling stage. Because absolute levels are available due to the aforementioned techniques applied to the coding structure, the frequency positions covered by the local templates consist of more information. Some of the presented embodiments utilize this fact as follows. Since the coding is aligned so that each level is fully coded, the local template evaluates neighboring spatial positions that are already coded. The absolute sum of neighboring levels and the number of valid levels are used to derive context models for bins that are part of a first binarization code, which may be a unary binarization or a truncated unary binarization. Binarization parameters (163), such as Rice parameters, can also be derived from the evaluation. An exemplary configuration is as follows.

[0130] The shape of the template can be fixed or adaptive. Adaptive shape embodiments have been described above. For example, if the template size is equal to 5 and the evaluation of the first two neighboring positions already leads to a very high absolute sum, the evaluation of the remaining neighboring positions can be skipped, resulting in an adaptive shape for the local template. (Compare Fig. 14) The shape can be configured according to the block size, color components, the last valid position, and the selected transformation. In a special case, the template size can be set to 0, so that none of the quantization levels already coded within the local periphery are used for context derivation. The local template is then deactivated. When the residual energy is distributed and flat between frequency positions—that is, when the absolute levels are low—a template that covers more neighboring frequency positions is more efficient. These situations often occur in larger transformation blocks with the last valid position at high frequency positions.

[0131] When bins of the binarization of quantization indices are coded in multiple scan passes for scan positions of a transform block or a sub-block of a transform block, only the bins transmitted in the current and previous scan passes may be used to derive the context model. In that case, partially reconstructed absolute levels (derived using the bins transmitted in the current and previous scan passes) are used. In particular, for context-coded bins transmitted in the first pass (for scan positions of the transform block or sub-block), a template is evaluated using the partially reconstructed transform coefficient levels provided by the bins coded in the first scan pass.

[0132] Different context model sets may be used depending on information such as the current diagonal, block size, applied transformation, selected predictor or prediction mode, and position within the truncation unary codeword. To prevent context dilution, context sets may be merged. For example, adjacent diagonals may be associated with a single partition, as illustrated on the right side of FIG. 25. For instance, a specific context set may be used for multiple diagonal positions (i.e., a block of transformation coefficients is partitioned into multiple regions according to the diagonal (x+y) position, and a separate context set is used for each of these partitions). Similarly, different context sets may be obtained by merging multiple values ​​of different parameters or combinations of different parameters. A dedicated context model set for a DC position is also advantageous in certain situations. (See example providing sample-specific context sets for positions (104 and / or 98), e.g.) This means that for the DC coefficient (located at the x = 0 and y = 0 positions, i.e., position (104) in the block of transformation coefficients), a different set of context models may be used than for all other coefficient positions. This distinction of context sets may be applied to all bins that are part of the truncation unary code. This means that any of the previously mentioned methods for specifying context sets may be applied to individual bins of the binarization of transformation coefficient levels, or alternatively to a subset of bins of the binarization of transformation coefficient levels, or alternatively to all context-coded bins of the binarization of transformation coefficient levels.

[0133] The selection of a context model depends on the results of the template evaluation. For example, the absolute sum of the levels and the number of valid positions within the template are derived. Based on these two numbers, the truncation unary code and Rice parameters ( kContext models can be derived for parts of all bins of the first binary code (160), such as ). For example, the value obtained by subtracting the number of valid positions from the absolute sum returns directly to the context model index within the context model set. Since the context model sets are limited, the value may be truncated by the number of context models within the context model set. In the case of the Rice parameter, a mapping table can be derived so that a direct mapping between the absolute sum and the Rice parameter can be established. However, to limit the table size, the absolute sum must also be limited.

[0134] The selection of context models can also be derived from the selected transformation. In particular, if a second transformation applied only to a subset of first-order coefficients is used, a dedicated set of context models can be used for both subsets.

[0135] There may be known and usable coding states to improve compression efficiency. One case involves level coding for the last valid scan position. For this position, a dedicated context model is used for coding the bins, which are part of a first binarization code that may be a truncation unary binarization. This can be combined with the current position in the block, for example, by evaluating the diagonal, so different context sets may be used. This can be further combined with the method described above, where different sets of context models are used for the bins at the DC transformation coefficient level (i.e., the transformation coefficient level at the x = 0 and y = 0 positions within the transformation block). When the last valid scan position is the same as the DC position, there are several possibilities: the context set for the last position is used (i.e., the last position has higher priority), the context set for the DC position is used (i.e., the last position has higher priority), or a separate context set is used for the DC positions, which are also the last scan positions. As previously mentioned, different context sets can be applied to individual bins of the binarization of transformation coefficient levels, or alternatively to a subset of bins of the binarization of transformation coefficient levels, or alternatively to all context-coded bins of the binarization of transformation coefficient levels.

[0136] Another case is when all coefficients within the local template are valid and their absolute levels all exceed a specific threshold. In this case, context modeling switches to a mode where local template evaluation is disabled until the end of the transformation block. Additionally, different sets of context models may be used, for example, depending on the current diagonal. Conversely, coding modes that occur at low frequencies but require signaling may be excluded. These coding modes are, for example, the coding of CBFs for groups when the previous group consists of a large number of absolute levels. In this example, the configuration is that the scan positions covered by the group are processed sequentially. Another example is the first binarization boundary (160). Instead of starting with a fixed boundary value at the beginning of each group, the absolute levels of the previous group are analyzed, and the boundary varies from 1 to 5. To mitigate throughput patterns, the boundary may be set to 0 if the previous group consists of very high residual energy. Then the first binarization code (160) will disappear. The use of local templates is efficient when statistics are not extreme, that is, when conditional probabilities do not approach minimum entropy. In the latter case, template evaluation is disabled, which can lead to lower complexity while maintaining compression efficiency. This is particularly useful at higher operating points where the quantity of levels plays a critical role in processing performance. A different case is given at the start of the transformation block when the levels of major quantities are invalid. In this case, template evaluation may be disabled because uncertainty is high and the amount of additional information available for context modeling is small.

[0137] There exist different planes, also referred to as color components, each of which conveys different color information, such as luminance and chrominance information. All planes cover the same area of ​​the image, and accordingly, the transformation blocks within a color plane have corresponding parts in all other color planes. These transformation blocks are each coded sequentially using the same syntax. Accordingly, the same syntax elements exist for each transformation block. Thus, information from the transformation blocks of the color components is available to the next transformation blocks. This is utilized to derive context models / sets for the syntax elements of the currently processed block by evaluating their corresponding parts in other color components. (Compare Fig. 12) There is one case where the coded block flag (CBF) of the second chroma component is coded. One of two different contexts is selected based on the preceding CBF values ​​of the first chroma components. Another example is the first bin of the truncation unary code for absolute count level coding. Additionally, the value of the first bin of the truncation unary code at the same position in the previous color plane is evaluated to select different context sets.

[0138] In a preferred embodiment related to the coding structure, the coding is configured so that each conversion factor level is fully transmitted for each scan position of the conversion block. In another preferred embodiment, the levels are coded in two passes so that all general coded bins (165) of part (160) are coded, and then the bypass bins (165) of part (162) are coded. In another preferred embodiment of the invention, the number of passes is maintained, but the processing is for the entire conversion block instead of processing per subblock.

[0139] In a preferred embodiment, the grouping shape is fixed to a fixed shape size smaller than the existing 4x4. In another preferred embodiment, the grouping shape is fixed to various shape sizes. The shape depends on the transformation block size and / or color component and / or prediction mode and / or selected transformations and / or the last effective position. In a further preferred embodiment, the shape size is adaptively derived using forward adaptive determination. In this embodiment, the shape size starts as a standard 4x4 shape, and the size can be, for example, 8x8.

[0140] In a preferred embodiment, the scan pattern is separated from the grouping as exemplarily illustrated in FIG. 13 so that the CBF for the groups is coded when the first frequency position of the group is coded. In this configuration, the scan pattern may be horizontal, for example, so that each row of the transformation block is fully processed, while the shape is still 4x4 or vertical.

[0141] In a preferred embodiment, the coding structure is arranged so that levels are coded immediately after the coding of the CBF flag for the corresponding transformation block. The reordering enables the possibility of evaluating already coded spatially adjacent transformation blocks for context modeling. In a further preferred embodiment, the coding of the chroma blocks is interleaved. In this configuration, the last position is coded once, and the coding is switched between two chroma transformation blocks. In a further embodiment, the last position is coded regularly, and the switch between the two chroma blocks is activated when the scan positions are aligned. In another preferred embodiment, two chroma blocks are interleaved to create a transformation block having the same size as the chroma for 4:2:0 content. Then, the combined chroma block is coded regularly.

[0142] In a preferred embodiment, the last valid position is coded jointly for the chroma transformation blocks. In this configuration, since the last position is not the last valid position but rather the last position where coding begins, validity information must be coded for that specific position. In an additional embodiment, the last position is coded once for two chroma blocks but is specified for only one of the chroma blocks. For the other chroma block, validity must be coded at the last valid position of the other chroma block.

[0143] In a preferred embodiment, CBF implies that the transformation block must be coded. The result is that the transformation block may be invalid. In this particular configuration, the last valid position is simply the same as the last position. This configuration can be combined with specific conditions such as block size, prediction mode, transformation mode, etc. For example, if 1-dimensional partitions are used, the resulting stripes are only 1x4 for a 4x4 block. In this situation, if the first 1x4 stripe is valid, CBF can be inferred to be equal to 1.

[0144] In a preferred embodiment, CBF means that the frequency positions of the group must be coded. In this configuration, all scan positions are coded and the last scan position within the group is not inferred.

[0145] In a preferred embodiment regarding binarization, the first binarization boundary (164) is variable for each transformation block depending on the block size or color component or prediction mode or selected transformations. In this configuration, the first boundary may be 5 in the case of other transformation modes requiring signaling in the bitstream. However, in the case of a general transformation, the first boundary may be maintained at 3.

[0146] In a preferred embodiment, the first boundary (164) never decreases when the boundary is greater than 3. In this preferred embodiment, the assumption is that the levels are not fixed, and therefore it is meaningful to code more bins than usual in normal mode. In another preferred embodiment, the decrease of the first boundary depends on the number of coded symbols within the block (which may represent a transformation block or a subblock of a transformation block). In a further preferred embodiment, the first boundary is adaptive according to the template evaluation. For example, when the number of valid positions is 0, the first boundary is 5. However, if a given amount, for example, 3 valid levels are in the template, the boundary is set to 2, and if 5 valid levels are in the template, the boundary is set to 1.

[0147] In a preferred embodiment, the Rice parameter is derived using a template. In this configuration, the absolute sum of the levels is calculated and mapped to the Rice parameter. The table size may have a limit of 8, 16, or 32. Values ​​within the lookup table may be derived empirically. In another preferred embodiment of the invention, the Rice parameter is never reduced for a group even when using a template. In a further preferred embodiment of the invention, the Rice parameter depends on the optimal Rice parameter for the levels covered by the template. In this configuration, it is noted that the first binarization boundary of the already coded levels may be taken into account in the calculation.

[0148] In a preferred embodiment, the interval covered by the Rice codes depends on the Rice parameter and becomes larger with higher Rice parameters. In a further preferred embodiment, the interval covered by the Rice codes depends on the block size and / or prediction mode and / or transform size and / or quantization parameter. It should be noted that in the preferred embodiments of this paragraph, a second binarization boundary can be inferred from the interval covered by the Rice codes.

[0149] In a preferred embodiment, the order of Exp-Golomb is fixed and equal to 0. In a further preferred embodiment, the order of Exp-Golomb is initially the Rice parameter plus 1. In another preferred embodiment, the order of Exp-Golomb is initially the Rice parameter. In both preferred embodiments, the Exp-Golomb code covers only a limited interval, and when the value is greater than the end of the interval, Exp-Golomb is applied again for the remaining values, but in a different order. In a preferred embodiment, the order of Exp-Golomb decreases until it becomes 0, with a step size equal to 1. In a further preferred embodiment, the order of Exp-Golomb is exactly 0.

[0150] With respect to context modeling, in a preferred embodiment, the template geometry is fixed and covers all adjacent spatial neighbors. In a further preferred embodiment, the template geometry is fixed and also covers spatially adjacent neighbors in other color channels. This configuration is designed for coding chroma transformation blocks.

[0151] In a preferred embodiment, the template shape is adaptive and is derived at the start of the transform block. In particular, the shape is larger or smaller depending on the transform block size and / or the prediction mode and / or the transforms and / or the quantization parameters. In a further preferred embodiment of the present invention, the template shape is adjusted within the transform block. In this configuration, the shape becomes progressively larger and smaller. For example, if the shape size is equal to 5 and three or four neighboring positions are valid, but the absolute level sum is below a certain threshold, the template is increased to incorporate more neighboring positions. On the other hand, if all frequency positions are valid and the absolute level sum exceeds a certain threshold, the shape becomes smaller, for example, set to equal to 3.

[0152] In a preferred embodiment, the context model sets for the first bin of the truncation unary portion depend on the diagonal. In this configuration, the luminance transform blocks may use three different context sets, whereas the chroma transform blocks use only two different context sets. The diagonal defines the transition points and may differ between luminance and chroma. In a further preferred embodiment of the invention, different context sets are used when a transform to be signaled in the bitstream is selected.

[0153] In a preferred embodiment, the beans starting at the second position, i.e., all beans that are truncation unary codes excluding the first bean, use the same set of contexts. In a further preferred embodiment of the invention, the beans use the same set of context models, but the sets of contexts may differ along the diagonal. In another preferred embodiment, the beans use the same set of context models, but the sets of contexts may differ depending on the selected transformations. In another preferred embodiment, the first diagonal uses a dedicated set of context models. In a further preferred embodiment, the beans starting at the fourth position, i.e., all beans that are truncation unary codes excluding the first, second, and third beans, use different sets of contexts depending on the selected transformations.

[0154] In a preferred embodiment, the context model for the first bin is derived using the absolute sum of templates restricted to a specific threshold having the same size as the context set used. In a further preferred embodiment, the evaluation subtracts the number of valid positions from the absolute level sum before truncation. In another preferred embodiment, the evaluation subtracts the number of valid positions from the absolute level sum after truncation. In a further preferred embodiment, the variance of the absolute levels is calculated and normalized by the number of context models within the context model set.

[0155] In a preferred embodiment, a context model for the second bin, which is a context model for the end of the truncation unary code, is selected by subtracting the number of valid positions from the absolute level sum. In a further preferred embodiment of the invention, an existing approach is used but modified to be integrated with a template. In this configuration, the transition to a dedicated context model may occur earlier, for example, when the absolute level sum exceeds a certain threshold.

[0156] In a preferred embodiment, the selected transformation is decoded first. If this transformation consists of a first and second transformation stage and only a subset of the first transformations is further transformed by the second stage, two different templates are used for the coefficient positions of the two sets.

[0157] In a preferred embodiment, a dedicated context model is used for the last valid position for the bins of the truncation unary code. In this configuration, context models are used exclusively for the last valid position. Details have been discussed in relation to FIG. 5. In another preferred embodiment, context modeling for the bins of the truncation unary code for the last valid position depends on the already coded last valid position of spatially adjacent transformation blocks, including transformation blocks in the same position on a different color plane. In particular, the set of context models selected for the current bin of the truncation unary code depends on the last valid position on a different color plane. Whenever the current position, indicated by the current bin, is larger, a different set of context is selected.

[0158] In a preferred embodiment, context modeling for a luminance CBF relies on already coded chroma CBFs. In a further preferred embodiment, context modeling for a second chroma CBF relies on an already coded first CBF. In another preferred embodiment of the invention, the context model for a first chroma CBF relies on an already coded luminance CBF. In a further preferred embodiment of the invention, when the coding structure is changed, context modeling for a CBF uses energy information from spatially adjacent transform blocks, which also include transform blocks at the same position in a different color plane. In particular, the absolute sum is evaluated and compared with a list of thresholds. Whenever the value is greater than a threshold, the context index is incremented by 1. In a further preferred embodiment, context modeling for a CBF uses the same energy information, but the set of thresholds depends on the prediction mode of the current and / or adjacent blocks, e.g., whether the prediction mode is one of the LMChroma modes. In another preferred embodiment, context modeling for CBF uses the same energy information, but the context sets may differ depending on the prediction mode of the current and / or adjacent blocks, e.g., whether the prediction mode is one of the LMChroma modes.

[0159] In a preferred embodiment, context modeling for the first bin of a truncation unary code relies on already coded bins of spatially adjacent transformation blocks, including transformation blocks at the same position in a different color plane. In particular, a set of context models is selected based on the value of the first bin of the truncation unary code at the same position in a different color plane.

[0160] In a preferred embodiment, context modeling for the bins of a truncation unary code relies on already coded bins of spatially adjacent transformation blocks, including transformation blocks at the same position in a different color plane. Specifically, a set of context models is selected based on the value of the corresponding bin position of the truncation unary code at the same location in a different color plane. In a further preferred embodiment, context modeling for the bins of a truncation unary code uses energy information from spatially adjacent transformation blocks, including transformation blocks at the same position in a different color plane. Specifically, the absolute sum in individual regions of the transformation block is evaluated and compared with a list of thresholds. Whenever the value is greater than a threshold, a different set of contexts is selected.

[0161] In a preferred embodiment, the template is enabled / disabled when a sufficient number of absolute levels are coded. In a further preferred embodiment, the template is enabled / disabled when a sufficient number of valid positions are coded.

[0162] The embodiments presented above are summarized below in different terms.

[0163] According to one embodiment, an encoder for encoding an image into a data stream may be configured to convert blocks of the image individually for a first color component and a second color component to obtain a first conversion factor block and a second conversion factor block, respectively, and to context-adaptively entropy-encode the second conversion factor block using contexts dependent on the first conversion factor block.

[0164] According to one implementation, the encoder may be configured to encode, for each partition of a set of partitions in which a second transform factor block is subdivided, an indication (150) of whether the transform factors within each partition are encoded into a data stream or whether the encoding of the transform factors within each partition is skipped and all transform factors within each partition are zero, and for partitions where the indication indicates that the transform factors within each partition are encoded into a data stream, entropy-encode the quantization levels of the transform factors of the second transform factor block, and for partitions where the indication indicates that the encoding of the transform factors within each partition is skipped and all transform factors within the partitions are zero, wherein the encoding of the indication is performed using context-adaptive entropy coding that uses contexts dependent on the first transform factor block.

[0165] According to an additional implementation, the encoder may be configured to encode, for each additional partition of a set of additional partitions in which a first transformation coefficient block is subdivided, an additional indication of whether the transformation coefficients within each additional partition are encoded into a data stream or whether the encoding of the transformation coefficients within each additional partition is skipped and all transformation coefficients within each additional partition are zero, and for partitions where the additional indication indicates that the transformation coefficients within each additional partition are encoded into a data stream, entropy-encode the quantization levels of the transformation coefficients of the first transformation coefficient block, and for partitions where the additional indication indicates that the encoding of the transformation coefficients within each additional partition is skipped and all transformation coefficients within the additional partitions are zero, wherein the encoding of the indication is performed using context-adaptive entropy-coding that uses contexts dependent on the additional indication of the additional partitions and / or the quantization levels of the transformation coefficients of the first transformation coefficient block.

[0166] According to an additional implementation, the encoder may be configured to encode an indication (190) of whether the transform coefficients in the second transform coefficient block are encoded into a data stream or whether the encoding of the transform coefficients in each partition is skipped and all transform coefficients in the second transform coefficient block become zero, and if the indication indicates that the transform coefficients of the second transform coefficient block are encoded into a data stream, entropy-encode the quantization levels of the transform coefficients of the second transform coefficient block, and if the indication indicates that the encoding of the transform coefficients of the second transform coefficient block is skipped and all transform coefficients in the partitions are zero, wherein the encoding of the indication (190) is performed using context-adaptive entropy coding that uses contexts dependent on the first transform coefficient block.

[0167] According to an additional implementation, the encoder may be configured to encode an additional indication (190) of whether the transform coefficients of the first transform coefficient block are encoded into a data stream or whether the encoding of the transform coefficients of the first transform coefficient block is skipped and all transform coefficients within each additional partition are zero, and if the additional indication indicates that the transform coefficients of the first transform coefficient block are encoded into a data stream, entropy-encode the quantization levels of the transform coefficients of the first transform coefficient block, and if the additional indication indicates that the encoding of the transform coefficients of the first transform coefficient block within each additional partition is skipped and all transform coefficients of the first transform coefficient block are zero, wherein the encoding of the indication (190) is performed using context-adaptive entropy coding that uses contexts dependent on the additional indication and / or the quantization levels of the transform coefficients of the first transform coefficient block.

[0168] According to an additional implementation, the encoder may be configured to encode a first indication (114) of a first end coefficient position where the last non-zero first transformation coefficient exists into a data stream when traversing a first scan pattern that sequentially traverses the transformation coefficients of a first transformation coefficient block along the forward direction, and to encode a first data representing a first coded set of transformation coefficients traversed by the first scan pattern in the reverse direction from the first end coefficient position into a data stream — the first data includes quantization levels of non-zero transformation coefficients within the first coded set of transformation coefficients — and, when traversing a second scan pattern that sequentially traverses the transformation coefficients of a second transformation coefficient block along the forward direction, to encode a second indication (114) of a second end coefficient position where the last non-zero second transformation coefficient exists into a data stream, and to encode a second data representing a second coded set of transformation coefficients traversed by the second scan pattern in the reverse direction from the second end coefficient position into a data stream. The second data includes quantization levels of non-zero transformation coefficients within a second coded set of transformation coefficients, wherein the encoding of the second representation is performed using context-adaptive entropy coding that uses contexts dependent on the first end coefficient position.

[0169] According to another embodiment, a decoder for decoding an image from a data stream may be configured to convert blocks of the image individually for a first color component and a second color component to obtain a first conversion factor block and a second conversion factor block, respectively, and to context-adaptively entropy decode the second conversion factor block using contexts dependent on the first conversion factor block.

[0170] According to one implementation, the decoder may be configured to decode an indication (150) for each partition of a set of partitions in which a second transform factor block is subdivided, whether the transform factors within each partition are encoded into a data stream or whether the encoding of the transform factors within each partition is skipped and all transform factors within each partition are zero, and for partitions where the indication indicates that the transform factors within each partition are encoded into a data stream, to entropy decode the quantization levels of the transform factors of the second transform factor block, and to skip decoding the transform factors within partitions where the indication indicates that the encoding of the transform factors within each partition is skipped and all transform factors within the partitions are zero, wherein the decoding of the indication is performed using context-adaptive entropy coding that uses contexts dependent on the first transform factor block.

[0171] According to an additional implementation, the decoder may be configured to decode an additional indication for each additional partition of a set of additional partitions in which a first transformation coefficient block is subdivided, whether the transformation coefficients within each additional partition are encoded into a data stream or whether the encoding of the transformation coefficients within each additional partition is skipped and all transformation coefficients within each additional partition are zero, and for partitions where the additional indication indicates that the transformation coefficients within each additional partition are encoded into a data stream, entropy decode the quantization levels of the transformation coefficients of the first transformation coefficient block, and skip the decoding of the transformation coefficients within the additional partitions where the additional indication indicates that the encoding of the transformation coefficients within each additional partition is skipped and all transformation coefficients within the additional partitions are zero, wherein the decoding of the indication is performed using context-adaptive entropy coding that uses contexts dependent on the additional indication of the additional partitions and / or the quantization levels of the transformation coefficients of the first transformation coefficient block.

[0172] According to an additional implementation, the decoder may be configured to decode an indication (190) of whether the transform coefficients in the second transform coefficient block are encoded into a data stream or whether the encoding of the transform coefficients in each partition is skipped and all transform coefficients in the second transform coefficient block become zero, and if the indication indicates that the transform coefficients of the second transform coefficient block are encoded into a data stream, to entropy decode the quantization levels of the transform coefficients of the second transform coefficient block, and if the indication indicates that the encoding of the transform coefficients of the second transform coefficient block is skipped and all transform coefficients in the partitions are zero, wherein the decoding of the indication (190) is performed using context-adaptive entropy coding that uses contexts dependent on the first transform coefficient block.

[0173] According to an additional implementation, the decoder may be configured to decode an additional indication (190) of whether the transform coefficients of the first transform coefficient block are encoded into a data stream or whether the encoding of the transform coefficients of the first transform coefficient block is skipped and all transform coefficients in each additional partition are zero, and if the additional indication indicates that the transform coefficients of the first transform coefficient block are encoded into a data stream, to entropy decode the quantization levels of the transform coefficients of the first transform coefficient block, and if the additional indication indicates that the encoding of the transform coefficients of the first transform coefficient block in each additional partition is skipped and all transform coefficients of the first transform coefficient block are zero, wherein the decoding of the indication (190) is performed using context-adaptive entropy coding that uses contexts dependent on the additional indication and / or the quantization levels of the transform coefficients of the first transform coefficient block.

[0174] According to an additional implementation, the decoder decodes from the data stream a first indication (114) of a first end coefficient position where the last non-zero first conversion coefficient exists when traversing a first scan pattern that sequentially traverses the conversion coefficients of a first conversion coefficient block along the forward direction, and decodes from the data stream first data representing a first coded set of conversion coefficients traversed by the first scan pattern in the reverse direction from the first end coefficient position — the first data includes quantization levels of non-zero conversion coefficients within the first coded set of conversion coefficients —, decodes from the data stream a second indication (114) of a second end coefficient position where the last non-zero second conversion coefficient exists when traversing a second scan pattern that sequentially traverses the conversion coefficients of a second conversion coefficient block along the forward direction, and decodes from the data stream second data representing a second coded set of conversion coefficients traversed by the second scan pattern in the reverse direction from the second end coefficient position It can be configured to decode, and the second data includes quantization levels of non-zero transform coefficients within a second coded set of transform coefficients, wherein the decoding of the second representation is performed using context-adaptive entropy coding that uses contexts dependent on the first end coefficient position.

[0175] According to a further embodiment, an encoder for encoding an image (12) into a data stream (11) may be configured to encode (108) the conversion coefficient block (92) into a data stream (14) using a scan pattern (94) [or scan path] that sequentially traverses the conversion coefficients of a conversion coefficient block (92) representing a block (84) of the image (12) to a second termination coefficient (104) in a predetermined direction (102) from a first termination coefficient position (98), by encoding data (96) representing a coded set (100) illustrated by hatching, wherein the data (96) includes quantization levels (106) of non-zero conversion coefficients within the coded set (100) of conversion coefficients [at least], and wherein the encoder is at the first termination coefficient position (98) or the second termination coefficient The quantization levels (106) are configured to context-adaptively entropy-encode quantization levels of non-zero transform coefficients in position (104) using contexts (112) of the first set (110), and the contexts (112) of the first set (110) are coprime to the contexts (112) of the second set (110) used for any other non-zero transform coefficient in the coded set (100) of transform coefficients.

[0176] According to one implementation, the encoder may be configured to encode the indication (114) of the end factor position where the last non-zero transformation factor exists as a data stream (14) when traversing the scan pattern (94) along the forward direction (116), wherein the predetermined direction (102) is the reverse direction (118), the first end factor position is the end factor position indicated by the indication (114), and the second end factor position (104) is the factor position traversed last along the scan pattern (94) in the reverse direction (118).

[0177] According to an additional implementation, the encoder may be configured to use a first set of contexts for non-zero transformation coefficients at a first end coefficient position (98).

[0178] According to an additional implementation, the encoder may be configured to use a first set of contexts for non-zero transformation coefficients at a second end coefficient position.

[0179] According to an additional implementation, the encoder may be configured to use a first set of contexts for non-zero transformation coefficients at a first end coefficient position (98) and a third set of contexts that are coprime to the first set for non-zero transformation coefficients at a second end coefficient position (98), wherein the first set and the third set are coprime to the second set of contexts used for any non-zero transformation coefficient in a coded set of transformation coefficients between the first end coefficient position and the second end coefficient position along the scan pattern.

[0180] According to an additional implementation, the first set may be coprime to the contexts of the second set used for non-zero transformation coefficients in coefficient positions traversed along the scan pattern (94) immediately before the first end coefficient position (98) and immediately after the second end coefficient position (104) in the reverse direction (118).

[0181] According to an additional implementation, the mutually exclusive positions may be independent of the first end coefficient position (98).

[0182] According to an additional implementation, the encoder may be configured to context-adaptively entropy-encode (116) the quantization levels of non-zero transform coefficients of the transform coefficient block (92), and for each partition sub-block of the partitions (120) into which the transform coefficient block (92) is subdivided, use a set of contexts that are coprime to the first set.

[0183] According to an additional implementation, the partitions (120) into which the conversion factor block is subdivided may be extended diagonally along a direction oblique to the diagonal (122) of the conversion factor block (92) penetrating the second end factor position, e.g., DC position.

[0184] According to a further embodiment, a decoder for decoding an image (12) from a data stream (11) may be configured to decode a conversion coefficient block (92) from a data stream (14) using a scan pattern (94) [or scan path] that sequentially traverses the conversion coefficients of a conversion coefficient block (92) representing a block (84) of the image (12) by decoding data (96) from the data stream (14) representing a coded set (100) illustrated by hatching, wherein the data (96) includes quantization levels (106) of [at least] non-zero conversion coefficients within the coded set (100) of conversion coefficients, wherein the decoder is at the first termination coefficient position (98) or the second termination coefficient The quantization levels (106) are configured to context-adaptively decode the quantization levels of a non-zero transform coefficient in position (104) using contexts (112) of a first set (110), and the contexts (112) of the first set (110) are coprime to the contexts (112) of a second set (110) used for any other non-zero transform coefficient in the coded set (100) of transform coefficients.

[0185] According to one implementation, the decoder may be configured to decode from the data stream (14) an indication (114) of a terminal coefficient position where a non-zero last conversion coefficient exists when traversing the scan pattern (94) along the forward direction (116), wherein the predetermined direction (102) is the reverse direction (118), the first terminal coefficient position is the terminal coefficient position indicated by the indication (114), and the second terminal coefficient position (104) is the coefficient position traversed last along the scan pattern (94) in the reverse direction (118).

[0186] According to an additional implementation, the decoder may be configured to use a first set of contexts for non-zero conversion coefficients at a first end coefficient position (98).

[0187] According to an additional implementation, the decoder may be configured to use a first set of contexts for non-zero transform coefficients at a second end coefficient position.

[0188] According to an additional implementation, the decoder may be configured to use a first set of contexts for non-zero transformation coefficients at a first end coefficient position (98) and a third set of contexts that are coprime to the first set for non-zero transformation coefficients at a second end coefficient position (98), wherein the first set and the third set are coprime to the second set of contexts used for any non-zero transformation coefficient in a coded set of transformation coefficients between the first end coefficient position and the second end coefficient position along the scan pattern.

[0189] According to an additional implementation, the first set may be coprime to the contexts of the second set used for non-zero transformation coefficients in coefficient positions traversed along the scan pattern (94) immediately before the first end coefficient position (98) and immediately after the second end coefficient position (104) in the reverse direction (118).

[0190] According to an additional implementation, the mutually exclusive positions may be independent of the first end coefficient position (98).

[0191] According to an additional implementation, the decoder may be configured to context-adaptively entropy decode (116) the quantization levels of non-zero transform coefficients of the transform coefficient block (92), and for each partition sub-block of the partitions (120) into which the transform coefficient block (92) is subdivided, use a set of contexts that are coprime to the first set.

[0192] According to an additional implementation, the partitions (120) into which the conversion factor block is subdivided may be extended diagonally along a direction oblique to the diagonal (122) of the conversion factor block (92) that penetrates the second end factor position (e.g., DC position).

[0193] According to another embodiment, an encoder for encoding an image into a data stream may be configured to context-adaptively entropy-encode the quantization level of a currently encoded transform factor by using a context determined based on the sum and number of valid transform factors among one or more previously encoded transform factors located at positions determined by a local template (132) positioned at the currently encoded transform factor of a transform factor block (92) representing a block (84) of the image (12).

[0194] According to one implementation, the encoder may be configured to determine the context by forming a difference between a first value determined based on a sum and a second value determined based on a number.

[0195] According to an additional implementation, the encoder may be configured to entropy-encode quantization levels using context-adaptive binary arithmetic coding of the binarization of the absolute values ​​of quantization levels accompanied by unary codes, where the context is used for the bins of the unary codes.

[0196] According to another embodiment, a decoder for decoding an image from a data stream may be configured to context-adaptively entropy decode the quantization level of a currently decoded transform coefficient by using a context determined based on the sum and number of valid transform coefficients among one or more previously decoded transform coefficients located at positions determined by a local template (132) positioned at the currently decoded transform coefficient of a transform coefficient block (92) representing a block (84) of the image (12).

[0197] According to one implementation, the decoder may be configured to determine the context by forming a difference between a first value determined based on a sum and a second value determined based on a number.

[0198] According to an additional implementation, the decoder may be configured to entropy decode quantization levels using context-adaptive binary arithmetic coding of the binarization of the absolute values ​​of quantization levels involving unary codes, where the context is used for the bins of the unary codes.

[0199] According to another embodiment, an encoder for encoding an image into a data stream may be configured to encode a conversion factor block into a data stream by encoding the absolute value of the quantization level of a currently encoded conversion factor of a conversion factor block (92) representing a block (84) of an image (12) in a manner that is binaryized using a binarization (161) parameterized using a binarization parameter, and by setting a binarization parameter [e.g., cutoff value; Exp Golomb order or Rice parameter] based on the sum and number of valid conversion factors among one or more previously encoded conversion factors located at positions determined by a local template (132) positioned on the currently encoded conversion factor.

[0200] According to another embodiment, a decoder for decoding an image from a data stream may be configured to decode a conversion factor block from a data stream by decoding the absolute value of the quantization level of a currently decoded conversion factor of a conversion factor block (92) representing a block (84) of an image (12) in a manner that is binaryized using a binarization (161) parameterized using a binarization parameter, and by setting a binarization parameter [e.g., cutoff value; Exp Golomb order or Rice parameter] based on the sum and number of valid conversion factors among one or more previously decoded conversion factors located at positions determined by a local template (132) positioned on the currently decoded conversion factor.

[0201] According to another embodiment, an encoder for encoding an image into a data stream may be configured to encode the transformation coefficient block into a data stream using a scan pattern (94) that sequentially traverses the transformation coefficients of the transformation coefficient block, by encoding the absolute values ​​of the quantization levels (106) of the transformation coefficients of the transformation coefficient block (92) representing the image (12) in a manner that is binarized using a unary code (160) below the cutoff value (164) and a non-unary code (162) above the cutoff value to which the unary code (160) is transposed for the cutoff value, thereby adaptively changing the cutoff value for the transformation coefficients of the transformation coefficient block according to the previously encoded transformation coefficients (compared to claims 41 through 60 without an initial cutoff setting).

[0202] According to another embodiment, a decoder for decoding an image from a data stream may be configured to decode the transform factor block from a data stream using a scan pattern (94) that sequentially traverses the transform factors of the transform factor block, by decoding the absolute value of the quantization levels (106) of the transform factor block (92) representing a block (84) of an image (12) in a manner that is binarized using a unary code (160) below the cutoff value and a non-unary code (162) above the cutoff value to which the unary code (160) is transposed for the cutoff value, thereby adaptively changing the cutoff value for the transform factors of the transform factor block according to the previously decoded transform factors (compared to claims 41 through 60 without an initial cutoff setting).

[0203] According to another embodiment, an encoder for encoding an image into a data stream encodes the absolute values ​​of the quantization levels (106) of the transform coefficients of a transform coefficient block (92) representing a block (84) of an image (12) in a manner that is binarized using a unary code (160) below the cutoff value (164) [i.e., for absolute values ​​below the cutoff value] and a non-unary code (162) above the cutoff value to which the unary code (160) is transposed for the cutoff value, thereby representing the size of the block (84), the color components of the block (84), the prediction mode on which the prediction signal is based—the prediction residual of the prediction signal is represented by the block (84)—the transform on which the transform coefficient block (92) is based, the quantization parameters used to quantize the transform coefficient block (92), the energy measurements of previously encoded transform coefficients, and the current positions determined by the local template (132) positioned on the currently encoded transform coefficients, or the current position where the currently encoded transform coefficients are located. The transformation coefficient block can be configured to encode the transformation coefficient block into a data stream by setting a cutoff value (164) [e.g., initially] according to one or more evaluations of previously encoded transformation coefficients located in positions within the transformation coefficient block partition that precede the current partition, according to the coding order used to encode the absolute value of the quantization level (106) of the transformation coefficients of the transformation coefficient block defined by traversing the transformation coefficients of the transformation coefficient block (92) along a scan pattern in a predetermined direction (102) that is offset with respect to the partition.

[0204] According to one implementation, the encoder may be configured to adaptively change the cutoff value for the transform coefficients of the transform coefficient block according to the previously encoded transform coefficients if the cutoff value is initially set to a value following a predetermined threshold [meaning finer quantization], and to keep the cutoff value (164) constant at least provisionally [e.g., until the last partition is reached in the order direction (102)] if the cutoff value (164) is initially set to a value exceeding a predetermined threshold [meaning more approximate quantization].

[0205] According to an additional implementation, the encoder may adaptively change the cutoff value for the currently encoded conversion factor based on the sum and number of valid conversion factors among one or more previously encoded conversion factors located at positions determined by a local template (132) positioned at the currently encoded conversion factor.

[0206] According to an additional implementation, the encoder may be configured to context-adaptively entropy-encode one or more bins of unary code using a set of contexts associated with each partition for each partition of partitions in which the transform factor blocks are subdivided, wherein the encoder is configured to perform a configuration for a predetermined partition in a manner based on the evaluation of previously encoded transform factors located in a partition traversed by a coding order prior to the predetermined partition.

[0207] According to an additional implementation, the encoder may be configured to adaptively change the cutoff value, including setting the cutoff value to 0 so that no unary code exists if the energy measurement of previously encoded transform coefficients increases to a specific threshold.

[0208] According to another embodiment, a decoder for decoding an image from a data stream decodes the absolute value of the quantization level (106) of the transform coefficients of a transform coefficient block (92) representing a block (84) of an image (12) in a manner that is binarized using a unary code (160) below the cutoff value (164) [i.e., for absolute values ​​below the cutoff value] and a non-unary code (162) above the cutoff value to which the unary code (160) is transposed for the cutoff value, thereby determining the size of the block (84), the color component of the block (84), the prediction mode on which the prediction signal is based—the prediction residual of the prediction signal is represented by the block (84)—the transform on which the transform coefficient block (92) is based, the quantization parameter used to quantize the transform coefficient block (92), the energy measurement of the previously decoded transform coefficients, and the positions determined by the local template (132) positioned on the currently decoded transform coefficients, or the current The conversion factor block can be configured to decode from the data stream by setting a cutoff value (164) [e.g., initially] according to one or more evaluations of previously decoded conversion factors located in positions within the partition of the conversion factor block that precede the current partition, according to the coding order used to decode the absolute value of the quantization level (106) of the conversion factors of the conversion factor block defined by traversing the conversion factors of the conversion factor block (92) along a scan pattern in a predetermined direction (102) that is offset with respect to the current partition where the decoded conversion factor is located.

[0209] According to one implementation, the decoder may be configured to adaptively change the cutoff value for the transform coefficients of the transform coefficient block according to previously decoded transform coefficients if the cutoff value is initially set to a value following a predetermined threshold [meaning finer quantization], and to keep the cutoff value (164) constant at least provisionally [e.g., until the last partition is reached in the order direction (102)] if the cutoff value (164) is initially set to a value exceeding a predetermined threshold [meaning more approximate quantization].

[0210] According to an additional implementation, the decoder may be configured to adaptively change the cutoff value for the currently encoded conversion coefficient based on the sum and number of valid conversion coefficients among one or more previously encoded conversion coefficients located at positions determined by a local template (132) positioned at the currently decoded conversion coefficient.

[0211] According to an additional implementation, the decoder may be configured to context-adaptively entropy decode one or more bins of unary code using a set of contexts associated with each partition for each partition of partitions in which the transform factor blocks are subdivided, wherein the decoder is configured to perform a configuration for a predetermined partition in a manner based on the evaluation of previously decoded transform factors located in a partition traversed by coding order prior to the predetermined partition.

[0212] According to an additional implementation, the decoder may be configured to adaptively change the cutoff value, including setting the cutoff value to 0 so that no unary code exists if the measurement of the energy of previously decoded conversion coefficients increases a specific threshold.

[0213] According to another embodiment, an encoder for encoding an image into a data stream may be configured to context-adaptively entropy-encode the quantization level of the current encoded transform coefficient of the transform coefficient block (92) by setting the shape of the local template (132) or deactivating the local template (132) [so that template-specific context dependency is disabled], based on one or more previously encoded transform coefficients located at positions determined by a local template (130) positioned on the currently encoded transform coefficient, and / or based on the size of the block (84), the position of the terminal coefficient position where the last non-zero transform coefficient exists when traversing the scan pattern along the forward direction (116), the transform that forms the basis of the transform coefficient block (92) representing the block (84) of the image (12), or based on one or more previously encoded transform coefficients, or based on the use of a context determined independently of the previously encoded transform coefficients if the local template is disabled.

[0214] According to one implementation, when setting the shape of a local template according to previously encoded transformation coefficients, the encoder may be configured to set the shape of a local template according to one or more previously encoded transformation coefficients located at positions determined by a first local raw template (170) positioned at the currently encoded transformation coefficient.

[0215] According to an additional implementation, when setting the shape of a local template according to previously encoded transformation coefficients, the encoder may be configured to determine whether the shape of the local template (132) is the first local raw template (170) or the second raw template (172) according to one or more previously encoded transformation coefficients located at positions determined by the first local raw template (170) positioned at the currently encoded transformation coefficients, wherein the second raw template extends further from the currently encoded transformation coefficients than the first raw template and includes or does not include the positions determined by the first local raw template.

[0216] According to an additional implementation, the encoder may be configured to make a determination based on the sum or number of valid transformation coefficients among one or more previously encoded transformation coefficients located at positions determined by a first local raw template (170) positioned at the currently encoded transformation coefficient.

[0217] According to an additional implementation, the encoder may be configured to disable the local template if the count of previously encoded conversion coefficients within the conversion coefficient block, or the accumulated value derived based on the previously encoded conversion coefficients within the conversion coefficient block, exceeds a predetermined amount which is greater than a certain threshold, and / or disable the local template if the count of previously encoded conversion coefficients within the invalid conversion coefficient block exceeds a predetermined amount.

[0218] According to an additional implementation, the encoder may be configured to determine the context based on the sum or number of valid transformation coefficients among one or more previously encoded transformation coefficients located at positions determined by a local template positioned at the currently encoded transformation coefficient.

[0219] According to another embodiment, a decoder for decoding an image from a data stream may be configured to context-adaptively entropy decode the quantization level of the currently decoded transform coefficient of the transform coefficient block (92) by setting the shape of the local template (132) or deactivating the local template (132) [so that template-specific context dependency is disabled], based on one or more of the previously decoded transform coefficients located at positions determined by the local template (130) positioned at the currently decoded transform coefficients according to the size of the block (84), the position of the terminal coefficient position where the last non-zero transform coefficient exists when traversing the scan pattern along the forward direction (116), the transform that forms the basis of the transform coefficient block (92) representing the block (84) of the image (12), or based on the use of a context determined independently of the previously decoded transform coefficients if the local template is disabled.

[0220] According to one implementation, the decoder may be configured to set the shape of a local template according to one or more previously decoded transformation coefficients located at positions determined by a first local raw template (170) positioned at a currently decoded transformation coefficient (130) when setting the shape of a local template according to previously decoded transformation coefficients.

[0221] According to an additional implementation, when setting the shape of a local template according to previously decoded transformation coefficients, the decoder may be configured to determine whether the shape of the local template (132) is the first local raw template (170) or the second raw template (172) according to one or more previously decoded transformation coefficients located at positions determined by the first local raw template (170) positioned at the currently decoded transformation coefficients, wherein the second raw template extends further from the currently decoded transformation coefficients than the first raw template and includes or does not include the positions determined by the first local raw template.

[0222] According to an additional implementation, the decoder may be configured to make a determination based on the sum or number of valid transformation coefficients among one or more previously decoded transformation coefficients located at positions determined by a first local raw template (170) positioned at the currently decoded transformation coefficient.

[0223] According to an additional implementation, the decoder may be configured to disable the local template if the count of previously decoded conversion coefficients within the conversion coefficient block, or the accumulated value derived based on the previously decoded conversion coefficients within the conversion coefficient block, which is greater than a certain threshold, exceeds a predetermined amount, and / or disable the local template if the count of previously decoded conversion coefficients within the invalid conversion coefficient block exceeds a predetermined amount.

[0224] According to an additional implementation, the decoder may be configured to determine the context based on the sum or number of valid transformation coefficients among one or more previously decoded transformation coefficients located at positions determined by a local template positioned at the currently decoded transformation coefficient.

[0225] According to another embodiment, an encoder for encoding an image into a data stream may be configured to encode the transform coefficients (91) of a transform coefficient block by context-adaptive entropy encoding the quantization levels (106) of the transform coefficient blocks using a set of contexts associated with each partition for each partition of a transform coefficient block (92) representing a block (84) of the image (12), wherein the partitions of the transform coefficient blocks are shaped differently [i.e., not coincident with each other] and / or shaped according to a scan pattern in which the entropy encoding of the quantization levels of the transform coefficient blocks is performed [and, for example, may be signaled in the data stream or essentially determined], and / or shaped according to the size of the block (84) and / or shaped according to explicit partition shape information.

[0226] According to one implementation, the encoder may be configured to sequentially entropy-encode the quantization levels of the transform coefficients of the transform coefficient block in a context-adaptive manner by following a scan pattern that sequentially traverses the transform coefficients of the transform coefficient block in a manner that traverses the partitions sequentially without interleaving the transform coefficients of the other partitions.

[0227] According to the additional implementation, the partitions into which the conversion factor block is subdivided are extended diagonally along a direction oblique to the vertical direction through the DC conversion factor of the conversion factor block.

[0228] According to another embodiment, a decoder for decoding an image from a data stream may be configured to decode the transform coefficients (91) of a transform coefficient block by context-adaptive entropy decoding the quantization levels (106) of the transform coefficient blocks using a set of contexts associated with each partition for each partition of the subdivided partitions of the transform coefficient block (92) representing a block (84) of the image (12), wherein the partitions of the subdivided partitions of the transform coefficient block are shaped differently [i.e., not in agreement with each other] and / or shaped according to a scan pattern in which the entropy decoding of the quantization levels of the transform coefficient blocks of the transform coefficient block is performed [and, for example, may be signaled in the data stream or essentially determined], and / or shaped according to the size of the block (84) and / or shaped according to explicit partition shape information.

[0229] According to one implementation, the decoder can be configured to sequentially entropy decode the quantization levels of the transform coefficients of the transform coefficient block in a context-adaptive manner by following a scan pattern that sequentially traverses the transform coefficients of the transform coefficient block in a manner that sequentially traverses the partitions without interleaving the transform coefficients of the other partitions.

[0230] According to the additional implementation, the partitions into which the conversion factor block is subdivided are extended diagonally along a direction oblique to the vertical direction through the DC conversion factor of the conversion factor block.

[0231] According to another embodiment, an encoder for encoding an image into a data stream may be configured to encode the transform coefficients of a transform coefficient block by context-adaptive entropy encoding the quantization levels of all transform coefficients in each partition using a set of contexts associated with each partition, for each partition of a set of partitions in which a transform coefficient block (92) representing a block (84) of an image (12) is subdivided, and for each partition of a set of partitions in which a transform coefficient block (92) representing a block (84) of an image (12) is subdivided, an indication (150) of whether all transform coefficients in each partition are encoded into a data stream, or whether the encoding of all transform coefficients in each partition is skipped and all transform coefficients in each partition are zero, and for each partition in which the indication indicates that all transform coefficients in each partition are encoded into a data stream, the quantization levels of all transform coefficients in each partition are context-adaptive entropy encoded.

[0232] According to one implementation, the encoder may be configured to entropy encode the quantization levels of all transformation coefficients of each partition for each partition, such that for the last transformation coefficient of each partition, which is the last to be encoded among the transformation coefficients of each partition, a flag indicating whether the last transformation coefficient is zero is entropy encoded regardless of whether any of the previously encoded transformation coefficients within each partition are all zero.

[0233] According to an additional implementation, a decoder for decoding an image from a data stream may be configured to decode the transform coefficients of a transform coefficient block by context-adaptive entropy decoding the quantization levels of all transform coefficients in each partition using a set of contexts associated with each partition, for each partition of a set of partitions in which a transform coefficient block (92) representing a block (84) of an image (12) is subdivided, and for each partition of a set of partitions in which a transform coefficient block (92) representing a block (84) of an image (12) is subdivided, a mark (150) indicating whether all transform coefficients in each partition are decoded into a data stream, or whether the decoding of all transform coefficients in each partition is skipped and all transform coefficients in each partition are zero, and for each partition in which the mark indicates that all transform coefficients in each partition are decoded into a data stream, the quantization levels of all transform coefficients in each partition are context-adaptive entropy decoded.

[0234] According to an additional implementation, the decoder may be configured to entropy decode the quantization levels of all transformation coefficients of each partition for each partition, where the indication indicates that all transformation coefficients within each partition are decoded into a data stream, by entropy decoding a flag indicating whether the last transformation coefficient of each partition, which is the last transformation coefficient of each partition to be decoded, is zero regardless of whether any of the previously decoded transformation coefficients within each partition are zero.

[0235] According to another embodiment, an encoder for encoding an image into a data stream may be configured to encode the transformation coefficients by encoding the partitioning mode of a transformation coefficient block (92) representing a block (84) of the image (12) into the data stream, and if the partition mode is a first mode [e.g., the partition to partitions (120) is switched on], the transformation coefficient block entropily encodes the quantization levels of the transformation coefficients of the transformation coefficient block using a set of contexts associated with each partition for each partition of the partitions to which the transformation coefficient block is subdivided, and if the partition mode is a second mode [e.g., the partition to partitions (120) is switched off], the transformation coefficients are encoded by entropily encoding the quantization levels of the transformation coefficients of the transformation coefficient block using a global set of contexts (→ as in claims 21 through 40, but depending on the mode and mutually exclusive is merely optional).

[0236] According to one implementation, if the partition mode is a first mode, the encoder encodes, for each partition of a set of partitions, an indication of whether the quantization levels of the transform coefficients within each partition are encoded into a data stream, or whether the encoding of the quantization levels of the transform coefficients within each partition is skipped and all transform coefficients within each partition are zero, and when entropy encoding the quantization levels of the transform coefficients, the encoder may be configured to skip entropy encoding for partitions indicating that the encoding of the quantization levels of the transform coefficients within the partitions is skipped and all transform coefficients within the partitions are zero.

[0237] According to one embodiment, a decoder for decoding an image from a data stream decodes the partitioning mode of a transform factor block (92) representing a block (84) of an image (12) from the data stream, and if the partition mode is a first mode [e.g., the partition to partitions (120) is switched on], the quantization levels of the transform factor blocks are context-adaptive entropy decoded for each partition of the partitions to which the transform factor block is subdivided using a set of contexts associated with each partition, and if the partition mode is a second mode [e.g., the partition to partitions (120) is switched off], the quantization levels of the transform factor blocks are context-adaptive entropy decoded using a global set of contexts (→ as in claims 21 through 40, but depending on the mode and mutually exclusive is merely optional; e.g., one for the entire block (920)).

[0238] According to one implementation, if the partition mode is a first mode, the decoder decodes, for each partition of a set of partitions, an indication of whether the quantization levels of the transform coefficients within each partition are decoded into a data stream, or whether the decoding of the quantization levels of the transform coefficients within each partition is skipped and all transform coefficients within each partition are zero, and when entropy decoding the quantization levels of the transform coefficients, the decoder may be configured to skip entropy decoding for partitions where this indication indicates that the decoding of the quantization levels of the transform coefficients within the partitions is skipped and all transform coefficients within the partitions are zero.

[0239] According to another embodiment, an encoder for encoding an image into a data stream may be configured to encode the transform coefficient block into a data stream (14) using a scan pattern (94') that sequentially traverses the transform coefficients (91) of the transform coefficient block by context-adaptive entropy encoding the quantization levels (106) of the transform coefficients of the transform coefficient block using a set of contexts (110a, b, c) associated (126) with each partition (120) for each partition (120) in which the transform coefficient block (92) representing a block (84) of the image (12) is subdivided, wherein the scan pattern (94') is configured such that at least one transform coefficient (such as the hatched transform coefficients (91')) of the first partition, e.g. 120a or 120c, traverses between two transform coefficients of the second partition, e.g. 120b, in such a manner that the transform of the transform coefficient block (92) Traverse the coefficients (91) sequentially.

[0240] According to one implementation, the encoder may be configured so that a set of contexts are commonly associated with the first partition and the second partition, e.g., 120a or 120c.

[0241] According to an additional implementation, when the encoder context-adaptively entropy-encodes the quantization levels of the transform coefficients of the transform coefficient block (92), for the currently encoded transform coefficient (130), it may be configured to determine the context associated with the partition (112) in which the currently encoded transform coefficient is located among a set (110) of contexts (112), such as 120a, based on one or more previously encoded transform coefficients located in positions determined by a local template (132) positioned on the currently encoded transform coefficient.

[0242] According to an additional implementation, when the encoder context-adaptively entropy-encodes the quantization levels of the transform coefficients of the transform coefficient block, the encoder may be configured to determine the context associated with the partition where the currently encoded transform coefficient is located among a set of contexts based on one or more previously encoded transform coefficients located in positions determined by a local template (132) positioned on the currently encoded transform coefficient, and / or a count (e.g., 4 in the exemplary drawing) of previously encoded transform coefficients, e.g., hatched transform coefficients (140), located in the partition where the currently encoded transform coefficient (130) is located, e.g., 120a, and exceeding one or more specific thresholds.

[0243] According to an additional implementation, the encoder may be configured to encode, for each partition (120a, b, c) of a set of partitions in which the transform coefficient block (92) is subdivided, an indication (150) of whether the transform coefficients within each partition are encoded into a data stream, or whether the encoding of the transform coefficients within each partition is skipped and all transform coefficients within each partition are zero, and when entropy encoding the quantization levels of the transform coefficients, the encoder may be configured to skip entropy encoding for partitions in which the indication indicates that the encoding of the transform coefficients within the partitions is skipped and all transform coefficients within the partitions are zero.

[0244] According to an additional implementation, the encoder may be configured to encode the indication into a data stream between the quantization levels of the transformation coefficients within the partitions, e.g. 1201 and 1203, instead of the first transformation coefficient encountered within the partition to which the indication (150) is associated [where the CBF is 0, such as CBF2 in this example], or before it [where the CBF is 1, such as CBF3 in this example], so that the indication indicates that the transformation coefficients within each partition are encoded into a data stream.

[0245] According to another embodiment, a decoder for decoding an image from a data stream may be configured to decode a transform factor block from a data stream (14) using a scan pattern (94') that sequentially traverses the transform factor blocks (91) of a transform factor block by context-adaptive entropy decoding the quantization levels (106) of the transform factor blocks (92) representing a block (84) of an image (12) using a set of contexts (110a, b, c) associated with each partition (120), for each partition of partitions in which a transform factor block (92) is subdivided, wherein the scan pattern (94') sequentially traverses the transform factor blocks (91) of the transform factor block (92) in such a manner that at least one transform factor (such as the hatched transform factor (91')) of the first partition is traversed between two transform factors of the second partition.

[0246] According to one implementation, the decoder may be configured such that a set of contexts are commonly associated with the first partition and the second partition, e.g., 120a or 120c.

[0247] According to an additional implementation, when the decoder context-adaptively entropy-decoding the quantization levels of the transform coefficients of the transform coefficient block (92), for the currently encoded transform coefficient (130), it may be configured to determine the context associated with the partition (112) in which the currently encoded transform coefficient is located among a set (110) of contexts (112), such as 120a, based on one or more previously decoded transform coefficients located in positions determined by a local template (132) positioned on the currently decoded transform coefficient.

[0248] According to an additional implementation, when the decoder context-adaptively entropy-decoding the quantization levels of the transform coefficients of the transform coefficient block, the decoder may be configured to determine the context associated with the partition where the currently decoded transform coefficient is located among a set of contexts, based on the positions determined by the local template (132) positioned on the currently decoded transform coefficient, and / or one or more previously decoded transform coefficients located in the count (exemplarily 4 in the exemplary drawing) of previously decoded transform coefficients, e.g., hatched transform coefficients (140), which are located within the partition where the currently decoded transform coefficient is located, e.g., 120a, where the currently decoded transform coefficient (130) is located, and which exceed one or more specific thresholds.

[0249] According to an additional implementation, the decoder decodes, for each partition (120a, b, c) of a set of partitions in which the transform coefficient block (92) is subdivided, an indication (150) of whether the transform coefficients within each partition are decoded into a data stream, or whether the decoding of the transform coefficients within each partition is skipped and all transform coefficients within each partition are zero, and when entropy decoding the quantization levels of the transform coefficients, the decoder may be configured to skip entropy decoding for partitions in which the indication indicates that the decoding of the transform coefficients within the partitions is skipped and all transform coefficients within the partitions are zero.

[0250] According to an additional implementation, the decoder may be configured to decode the indication from the data stream between the quantization levels of the transform coefficients within the partitions, e.g. 1201 and 1203, where the indication indicates that the transform coefficients within each partition are decoded into the data stream, instead of the first transform coefficient encountered within the partition where the indication (150) is involved [where the CBF is 0, such as CBF2 in this example], or before it [where the CBF is 1, such as CBF3 in this example].

[0251] While some aspects have been described in relation to devices, these aspects also represent descriptions of methods for corresponding steps, where it is evident that a block or device corresponds to a method step or a feature of a method step. Similarly, aspects described in relation to method steps also represent descriptions of corresponding blocks, items, or features of a corresponding device. Some or all of the method steps may be executed by (or using) a hardware device, such as, for example, a microprocessor, a programmable computer, or an electronic circuit. In some embodiments, one or more of the most important method steps may be executed by such a device.

[0252] The encoded data stream of the present invention may be stored on a digital storage medium or transmitted through a transmission medium, such as a wireless transmission medium or a wired transmission medium, such as the Internet.

[0253] Depending on specific implementation requirements, embodiments of the present invention may be implemented in hardware or software. Implementation may be performed using a digital storage medium, such as a floppy disk, DVD, Blu-ray, CD, ROM, PROM, EPROM, EEPROM, or flash memory, which stores electronically readable control signals that cooperate (or can cooperate) with a programmable computer system to perform each method. Thus, the digital storage medium may be computer-readable.

[0254] Some embodiments according to the present invention include a data carrier having electronically readable control signals that can cooperate with a programmable computer system so that one of the methods described herein is performed.

[0255] Generally, embodiments of the present invention may be implemented as a computer program product having program code that operates to perform one of the methods when the computer program product is executed on a computer. The program code may be stored, for example, on a machine-readable carrier.

[0256] Other embodiments include a computer program for performing one of the methods described herein, stored on a machine-readable carrier.

[0257] That is, one embodiment of the method of the present invention is, accordingly, a computer program having program code for performing one of the methods described in this specification when the computer program is executed on a computer.

[0258] Accordingly, additional embodiments of the methods of the present invention include a data carrier (or digital storage medium, or computer-readable medium) recorded thereon, comprising a computer program for performing one of the methods described herein. The data carrier, digital storage medium, or recorded medium is typically tangible and / or non-transient.

[0259] Accordingly, a further embodiment of the method of the present invention is a data stream or sequence of signals representing a computer program for performing one of the methods described herein. The data stream or sequence of signals may be configured to be transmitted, for example, through a data communication connection, for example, over the Internet.

[0260] Additional embodiments include processing means, for example, a computer or programmable logic device configured or adapted to perform one of the methods described herein.

[0261] Additional embodiments include a computer installed with a computer program for performing one of the methods described herein.

[0262] Further embodiments according to the present invention include a device or system configured to transmit a computer program for performing one of the methods described herein to a receiver (e.g., electronically or optically). The receiver may be, for example, a computer, a mobile device, a memory device, etc. The device or system may include, for example, a file server for transmitting the computer program to the receiver.

[0263] In some embodiments, a programmable logic device (e.g., a field-programmable gate array) may be used to perform some or all of the functions of the methods described herein. In some embodiments, a field-programmable gate array may cooperate with a microprocessor to perform one of the methods described herein. Generally, the methods are preferably performed by any hardware device.

[0264] The device described in this specification may be implemented using a hardware device, or using a computer, or using a combination of a hardware device and a computer.

[0265] Any of the devices described in this specification or any components of the devices described in this specification may be implemented at least partially in hardware and / or software.

[0266] The methods described in this specification may be performed using a hardware device, or using a computer, or using a combination of a hardware device and a computer.

[0267] Any of the methods described in this specification or any components of the device described in this specification may be performed at least partially by hardware and / or software.

[0268] The embodiments described above are merely examples of the principles of the invention. It is understood that modifications and variations of the arrangements and details described herein will be apparent to those skilled in the art. Accordingly, this is intended to be limited only to the appended claims, rather than to the specific details presented by the description and explanation of the embodiments in this specification.

Claims

Claim 1 A video encoder comprising at least one processor, wherein the at least one processor: encodes the absolute value of the quantization level of a current transform coefficient at a position (xC, yC)—the absolute value being encoded based on a binarization parameterized using a binarization parameter, wherein a portion of the binarization comprises a prefix and a postfix, and the length of the prefix is ​​based on the binarization parameter—; determines the binarization parameter based on the sum of the absolute values ​​of the quantization levels of transform coefficients at one or more positions among (xC+1, yC), (xC+2, yC), (xC+1, yC+1), (xC, yC+1), and (xC, yC+2); and determines whether the number of context-coded bins for previous transform coefficients in the scan order of a transform block identical to the current transform coefficient has reached a limit; A video encoder configured to determine an initial portion of the binarization and encode the initial portion using context-adaptive entropy encoding in response to determining that the number of context-coded bins has not reached the limit, and to encode the portion of the binarization including the prefix and the postfix using equi-probability bypass coding; and also to determine that the binarization includes only the prefix and the postfix in response to determining that the number of context-coded bins has reached the limit, and to encode all bins of the binarization using equi-probability bypass coding. Claim 2 A video encoder according to claim 1, wherein the binarization parameter is an Exponential Golomb order or a Rice parameter. Claim 3 A video encoder according to claim 1, wherein the binarization parameter is determined by mapping the sum of the absolute values ​​of the quantization levels of the transformation coefficients at one or more positions to the binarization parameter using a lookup table. Claim 4 As a video encoding method, the method comprises: a step of encoding the absolute value of the quantization level of a current transform coefficient at a position (xC, yC) — said absolute value is encoded based on a binarization parameterized using a binarization parameter, said binarization part including a prefix and a postfix, said length of said prefix part based on said binarization parameter —; a step of determining said binarization parameter based on the sum of the absolute values ​​of the quantization levels of transform coefficients at one or more positions among (xC+1, yC), (xC+2, yC), (xC+1, yC+1), (xC, yC+1), and (xC, yC+2); and a step of determining whether the number of context-coded bins for previous transform coefficients in the scan order of a transform block identical to the current transform coefficient has reached a limit. A video encoding method comprising: determining an initial portion of the binarization and encoding the initial portion using context-adaptive entropy encoding in response to determining that the number of context-coded bins has not reached the limit, and encoding the portion of the binarization including the prefix and the postfix using equal probability bypass coding; and determining that the binarization includes only the prefix and the postfix in response to determining that the number of context-coded bins has reached the limit, and encoding all bins of the binarization using equal probability bypass coding. Claim 5 A video encoding method according to claim 4, wherein the binarization parameter is an Exponential Golomb order or a Rice parameter. Claim 6 A video encoding method according to claim 4, wherein the binarization parameter is determined by mapping the sum of the absolute values ​​of the quantization levels of the transformation coefficients at one or more positions to the binarization parameter using a lookup table. Claim 7 A video decoder comprising at least one processor, wherein the at least one processor: decodes the absolute value of the quantization level of a current transform coefficient at a position (xC, yC), said absolute value is encoded based on a binarization parameterized using a binarization parameter, said binarization part comprising a prefix and a postfix, said prefix length is based on said binarization parameter; determines said binarization parameter based on the sum of the absolute values ​​of the quantization levels of transform coefficients at one or more positions among (xC+1, yC), (xC+2, yC), (xC+1, yC+1), (xC, yC+1), and (xC, yC+2); and determines whether the number of context-coded bins for previous transform coefficients in the scan order of the same transform block as the current transform coefficient has reached a limit; A video decoder configured to: decode an initial portion of the binarization using context-adaptive entropy decoding in response to a determination that the number of context-coded bins has not reached the limit, and decode a portion of the binarization including the prefix and the postfix using equal-probability bypass coding; and also, in response to a determination that the number of context-coded bins has reached the limit, determine that the binarization includes only the prefix and the postfix and decode all bins of the binarization using equal-probability bypass coding. Claim 8 A video decoder according to claim 7, wherein the binarization parameter is an Exponential Golomb order or a Rice parameter. Claim 9 A video decoder according to claim 7, wherein the binarization parameter is determined by mapping the sum of the absolute values ​​of the quantization levels of the transformation coefficients at one or more positions to the binarization parameter using a lookup table. Claim 10 As a video decoding method, the method comprises: a step of decoding the absolute value of the quantization level of a current transform coefficient at a position (xC, yC) — said absolute value is determined based on a binarization parameterized using a binarization parameter, said binarization part including a prefix and a postfix, said prefix length is based on said binarization parameter —; a step of determining said binarization parameter based on the sum of the absolute values ​​of the quantization levels of transform coefficients at one or more positions among (xC+1, yC), (xC+2, yC), (xC+1, yC+1), (xC, yC+1), and (xC, yC+2); and a step of determining whether the number of context-coded bins for previous transform coefficients in the scan order of a transform block identical to the current transform coefficient has reached a limit. A video decoding method comprising: a step of decoding an initial portion of the binarization using context-adaptive entropy decoding in response to determining that the number of context-coded bins has not reached the limit, and decoding a portion of the binarization including the prefix and the postfix using equal-probability bypass coding; and a step of determining that the binarization includes only the prefix and the postfix in response to determining that the number of context-coded bins has reached the limit, and decoding all bins of the binarization using equal-probability bypass coding. Claim 11 A video decoding method according to claim 10, wherein the binarization parameter is an Exponential Golomb order or a Rice parameter. Claim 12 A non-transient digital storage medium storing a computer program for performing a video encoding method when the computer program is executed by a computer, wherein the method comprises: a step of encoding the absolute value of the quantization level of a current transform coefficient at a position (xC, yC) ― said absolute value is encoded based on a binarization parameterized using a binarization parameter, said binarization part including a prefix and a postfix, said length of said prefix part based on said binarization parameter ―; a step of determining said binarization parameter based on the sum of the absolute values ​​of the quantization levels of transform coefficients at one or more positions among (xC+1, yC), (xC+2, yC), (xC+1, yC+1), (xC, yC+1), and (xC, yC+2); a step of determining whether the number of context-coded bins for previous transform coefficients in the scan order of a transform block identical to the current transform coefficient has reached a limit; A non-transient digital storage medium comprising: a step of determining an initial portion of the binarization and encoding the initial portion using context-adaptive entropy encoding in response to determining that the number of context-coded bins has not reached the limit, and encoding the portion of the binarization including the prefix and the postfix using equal probability bypass coding; and a step of determining that the binarization includes only the prefix and the postfix in response to determining that the number of context-coded bins has reached the limit, and encoding all bins of the binarization using equal probability bypass coding. Claim 13 In claim 12, the non-transient digital storage medium wherein the binarization parameter is an Exponential Golomb order or a Rice parameter. Claim 14 A non-transient digital storage medium storing a computer program for performing a video decoding method when the computer program is executed by a computer, wherein the method comprises: a step of decoding the absolute value of the quantization level of a current transform coefficient at a position (xC, yC) — said absolute value is determined based on a binarization parameterized using a binarization parameter, said binarization part including a prefix and a postfix, said length of said prefix part based on said binarization parameter —; a step of determining said binarization parameter based on the sum of the absolute values ​​of the quantization levels of transform coefficients at one or more positions among (xC+1, yC), (xC+2, yC), (xC+1, yC+1), (xC, yC+1), and (xC, yC+2); a step of determining whether the number of context-coded bins for previous transform coefficients in the scan order of a transform block identical to the current transform coefficient has reached a limit; A non-transient digital storage medium comprising: a step of decoding an initial portion of the binarization using context-adaptive entropy decoding in response to determining that the number of context-coded bins has not reached the limit, and decoding a portion of the binarization including the prefix and the postfix using equal-probability bypass coding; and a step of determining that the binarization includes only the prefix and the postfix in response to determining that the number of context-coded bins has reached the limit, and decoding all bins of the binarization using equal-probability bypass coding. Claim 15 In claim 14, a non-transient digital storage medium wherein the binarization parameter is an Exponential Golomb order or a Rice parameter.