Picture coding supporting block partitioning and block merging

By combining block partitioning with block merging and sharing encoding parameters across merged blocks, the solution addresses inefficiencies in existing codecs, reducing redundancy and bit rate while enhancing coding efficiency in image and video compression.

JP2024156727AInactive Publication Date: 2024-11-06GE VIDEO COMPRESSION LLC
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
JP2024117498
Authority / Receiving Office
JP · JP
Patent Type
Applications
Current Assignee / Owner
Priority Date
2010-10-08
Filing Date
2024-07-23
Publication Date
2024-11-06
Estimated Expiration
Not applicable · inactive patent

AI Technical Summary

Technical Problem

Existing image and video codecs face inefficiencies due to redundancy caused by block aggregation and subdivision, particularly in setting prediction parameters, leading to increased bit rates and suboptimal coding efficiency.

Method used

The solution involves avoiding redundancy by combining block partitioning with block merging, where encoding parameters are shared across merged blocks, reducing the set of coding parameter candidates, and signaling only the differences, thereby maintaining coding efficiency while allowing for more flexible partitioning patterns.

Benefits of technology

This approach reduces the auxiliary information required for signaling, decreases bit rate, and enhances coding efficiency by allowing for more flexible partitioning without increasing signal overhead, thus improving image and video compression.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure 2024156727000001_ABST
    Figure 2024156727000001_ABST
Patent Text Reader

Abstract

To provide a method for avoiding redundancy between partitioning coding and merging coding.SOLUTION: Further coding efficiency increase may be achieved if for a current block 40 of a picture 20, for which the bit stream 30 signals one of supported partitioning patterns, a reversal of the partitioning by block merging is avoided. In particular, if the signaled one of the supported partitioning patterns specifies a subdivision of the block into two or more further blocks 50 and 60, a removal of certain coding parameter candidates for all further blocks, except a first further block of the further blocks in a coding order, is performed. In particular, those coding parameter candidates are removed from the set of coding parameter candidates for the respective further block, the coding parameters of which are the same as coding parameters associated with any of the further blocks which, when being merged with the respective further block, would result in one of the supported partitioning pattern.SELECTED DRAWING: Figure 1
Need to check novelty before this filing date? Find Prior Art

Description

[Technical field]

[0001] This application relates to image and / or video coding, in particular to codecs that support block partitioning and block merging. [Background technology]

[0002] Many image and / or video codecs process images on a block-by-block basis. For example, predictive codecs use block granularity to obtain a good compromise between setting the prediction parameters very accurately with high spatial resolution, which would consume too much auxiliary information for the prediction parameters, and setting the prediction parameters very coarsely, which would lead to an increase in the amount of bits required to code the prediction residual due to the low spatial resolution of the prediction parameters. In short, the optimal setting for the prediction parameters lies somewhere between the two extremes.

[0003] Several attempts have been made to obtain an optimal solution to the above problem. For example, instead of using a regular subdivision of the image into blocks regularly arranged in rows and columns, multi-tree partitioning subdivision attempts to increase the freedom of subdividing the image into blocks with appropriate requirements for subdivision information. However, even multi-tree subdivision requires the signaling of a significant amount of data, and the freedom in subdividing the image is rather limited even when using such multi-tree subdivision.

[0004] To allow a better trade-off between the amount of auxiliary information required to signal the image subdivision and the degree of freedom in subdividing the image, block merging can be used to increase the number of possible image subdivisions, with a reasonable amount of additional data required to signal the merging information. For blocks that are merged, the coding parameters likewise need to be transmitted in their entirety only once in the bitstream, as if the resulting merged group of blocks were a directly subdivided part of the image.

[0005] However, there still remains a need to achieve better coding efficiency due to the redundancy introduced by the combination of block merging and block subdivision. Summary of the Invention [Problem to be solved by the invention]

[0006] It is thus an object of the present invention to provide a coding concept with increased coding efficiency. This object is achieved by the independent claims of the application. [Means for solving the problem]

[0007] The idea behind the present invention is that a further increase in coding efficiency can be achieved if the inversion of the partitioning by block merging is avoided for a current block of an image whose bitstream signals one of the supported partitioning patterns. In particular, if the signaled one of the supported partitioning patterns specifies a subdivision of the block into two or more further blocks, then for all further blocks except the first one of the further blocks in the coding order, a removal of certain coding parameter candidates is performed. In particular, those coding parameter candidates whose coding parameters are the same as the coding parameters associated with any of the further blocks that would result in one of the supported partitioning patterns when merged with the respective further block are removed from the set of coding parameter candidates for each further block. In this way, redundancy between partition coding and merger coding is avoided and the signaling overhead for signaling merger information can be further reduced by utilizing a set of coding parameter candidates of reduced size. Furthermore, the positive effect of combining block partitioning with block merging is maintained. That is, due to combining block partitioning with block merging, the variety of achievable partitioning patterns is increased compared to the case without block merging. The increase in signaling overhead is kept within reasonable limits. Finally, block consolidation allows additional blocks to be combined across boundaries with the current block, thereby providing granularity that would not be possible without block consolidation.

[0008] Applying a slightly different idea of ​​a set of merging candidates, the idea described above, according to a further aspect of the present invention, in a decoder configured to decode a bitstream signaling one of the supported partitioning patterns for a current block of an image which the decoder is configured to remove, if the signaled one of the supported partitioning patterns specifies a subdivision of the block into two or more further blocks, then for all further blocks except the first of the further blocks in the coding order, identify from the set of candidate blocks of the respective further block a candidate block which, when merged with the respective further block, results in one of the supported partitioning patterns.

[0009] Advantageous embodiments of the invention are the subject of the accompanying dependent claims.

[0010] Preferred embodiments of the present application are explained in more detail below with reference to the figures. [Brief description of the drawings]

[0011] [Figure 1] FIG. 1 shows a block diagram of an encoder according to an embodiment. [Diagram 2] FIG. 2 shows a block diagram of a decoder according to an embodiment. [Diagram 3] FIG. 3 shows a block diagram of a possible internal structure of the encoder of FIG. [Figure 4] FIG. 4 shows a block diagram of a possible internal structure of the decoder of FIG. [Figure 5A] FIG. 5A illustrates diagrammatically a possible subdivision of an image into tree root blocks, coding units (blocks) and prediction units (partitions). [Figure 5B] FIG. 5B illustrates a subdivision tree of the tree root block shown in FIG. 5A, down to the level of the partition as shown. [Figure 6] FIG. 6 illustrates an embodiment for a set of possible supported splitting patterns according to an embodiment. [Figure 7] FIG. 7 shows a possible partitioning pattern that effectively results from combining block consolidation and block partitioning when using block partitioning according to FIG. [Figure 8] FIG. 8 illustrates a schematic of candidate blocks for the SKIP / DIRECT mode according to an embodiment. [Figure 9] FIG. 9 illustrates a syntax portion of a syntax according to an embodiment. [Figure 10] FIG. 10 illustrates a syntax portion of a syntax according to an embodiment. [Figure 11A] FIG. 11A illustrates a syntax portion of a syntax according to an embodiment. [Figure 11B] FIG. 11B illustrates a syntax portion of a syntax according to an embodiment. [Figure 12] FIG. 12 illustrates a schematic of a definition of adjacent partitions for a partition according to an embodiment. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS

[0012] With respect to the following description, it should be noted that whenever the same reference designations are used in relation to different figures, the description associated with each element presented with respect to one of these figures applies to the other figure as well, provided that moving from one figure to the other figure does not contradict the remaining description of this other figure.

[0013] Figure 1 shows an encoder 10 according to one embodiment of the present invention. The encoder 10 is arranged to encode an image 20 into a bitstream 30. Of course, the image 20 can be part of a video if the encoder is a video encoder.

[0014] The image 20 comprises a block 40 that is currently to be coded by the encoder 10. As shown in Fig. 1, the image 20 may comprise two or more blocks 40. For example, the image 20 may be subdivided into a regular arrangement of blocks 40, such that the blocks 40 are arranged in rows and columns, as shown by way of example in Fig. 1. However, any other subdivision of the image 20 into blocks 40 may be possible. In particular, the subdivision of the image 20 into blocks 40 may be fixed, i.e. known to the decoder by default, or may be signaled to the decoder in the bitstream 30. In particular, the blocks 40 of the image 20 may differ in size. For example, a multi-tree subdivision, such as a quad-tree subdivision, may be applied to the image 20, in this case to obtain blocks 40 that form leaf blocks of the multi-tree subdivision, or the image 20 may be regularly pre-subdivided into regularly arranged tree root blocks.

[0015] In any case, the encoder 10 is configured to signal in the bitstream 30 one of the supported partitioning patterns for the current block 40. That is, the encoder 10 decides as to which of the supported partitioning patterns should be used for the current block 40, for example, whether it is better to further partition the partition block 40 in terms of rate-distortion optimization, or to adapt the granularity at which certain coding parameters are set in the current block 40 of the image 20. As will be outlined in more detail below, the coding parameters may indicate prediction parameters, such as inter-prediction parameters. Such inter-prediction parameters may include, for example, reference image indexes, motion vectors, etc. The supported partitioning patterns may include, for example, a non-partitioning mode, i.e., the current block 40 is not further subdivided, a horizontally partitioning mode, i.e., the current block 40 is subdivided along a horizontally extending line into an upper or top and a lower or bottom portion, and a vertically partitioning mode, i.e., the current block 40 is vertically subdivided along a vertically extending line into a left part and a right part. Besides this, the supported partitioning patterns may also include an option where the current block 40 is further regularly subdivided into four further blocks, each of which is considered as one-quarter of the current block 40. Furthermore, the partitioning may concern all blocks 40 of the image 20, or only a suitable subset thereof, such as those having a particular coding mode associated therewith, such as an inter-prediction mode. Furthermore, the set of possible blocks for which merging is to be applied to a partition of the block may be further limited by bitstream signaling for each block 40 for which merging may be performed, as to whether merging is available for the partition of the block or not. Of course, this kind of signaling can also be done individually for each possible merging candidate partition.Furthermore, various subsets of the supported partitioning modes may be available for a block 40, in combination or individually, depending on the subdivision level of the block 40, e.g., the block size, if it is a multi-tree subdivision leaf block.

[0016] That is to say, the subdivision of the image 20 into blocks, in particular to obtain the blocks 40, may be fixed or signalled in the bitstream, while the partitioning pattern to be used for the blocks 40 now is signalled in the bitstream 30 in the form of partitioning information, which thus can be considered as a kind of extension of the subdivision of the image 20 into blocks 40. On the other hand, the further relevance of the original granularity of the subdivision of the image 20 into blocks 40 may still be maintained. For example, the encoder 10 may be configured to signal in the bitstream 30 the coding modes used for each part or block 40 of the image 20 with the granularity defined by the blocks 40, while the encoder 10 is configured to vary the coding parameters of each coding mode within each block 40 with an increased (finer) granularity defined by the respective partitioning pattern selected for each block 40. For example, the coding modes signaled at the granularity of the block 40 may be distinguished between temporal inter-prediction modes, intra-prediction modes such as inter-view prediction modes, inter-prediction modes, etc. The type of coding parameters associated with one or more sub-blocks (partitions) resulting from the division of each block 40 depends on the coding mode assigned to each block 40. For example, for intra-coded blocks 40, the coding parameters may include spatial directions as to which image content of a previously decoded portion of the image 20 is used to fill each block 40. In the case of inter-coded blocks 40, the coding parameters may include, among other things, motion vectors for motion compensated prediction.

[0017] Fig. 1 shows the current block 40 as being subdivided into two further (smaller) blocks 50 and 60 by way of example. In particular, a vertical partitioning mode is shown by way of example. The smaller blocks 50 and 60 can also be called sub-blocks 50 and 60, or partitions 50 and 60, or prediction units 50 and 60. In particular, the encoder 10 is configured to, if the signaled one of the supported partitioning patterns specifies a subdivision of the current block 40 into two or more further blocks 50 and 60, exclude, for all further blocks other than the first of the further blocks 50 and 60 in the coding order, from the set of coding parameter candidates for each further block, coding parameter candidates having coding parameters that are the same as coding parameters associated with any of the further blocks that, when combined with each further block, would be one of the supported partitioning patterns. More precisely, for each of the supported partitioning patterns, a coding order is defined between the resulting partition or partitions 50 and 60. In the case of Fig. 1, the coding order is indicated by an arrow 70 which, by way of example, defines that the left partition 50 is coded before the right partition 60. In case of a horizontal partitioning mode, it may be defined that the upper partition is coded before the lower partition. In any case, the encoder 10 is configured, for the second partition 60 in the coding order 70, to exclude from the set of coding parameter candidates for each second partition 60 those coding parameter candidates which have coding parameters which are the same as the coding parameters associated with the first partition 50, in order to avoid a result of this merging, i.e. that both partitions 50 and 60 would in fact have the same coding parameters associated with them which could equally result from selecting a non-splitting mode for the current block 40 at a lower coding rate.

[0018] More precisely, the encoder 10 is configured to use block merging in an efficient manner together with the block partitioning. As far as the block merging is concerned, the encoder 10 determines a respective set of coding parameter candidates for each partition 50 and partition 60. The encoder may be configured to determine the set of coding parameter candidates for each partition 50 and partition 60 based on coding parameters associated with previously decoded blocks. In particular, at least some of the coding parameter candidates in the set of coding parameter candidates may be equal to the coding parameters of a previously decoded partition, i.e. may be adopted from the coding parameters of a previously decoded partition. Additionally or alternatively, at least some of the coding parameter candidates may be obtained from the coding parameter candidates associated with two or more previously coded partitions by a suitable combination of median, average, etc. However, since the encoder 10 is configured to perform the determination of a reduced set of coding parameter candidates and the selection of one of the remaining non-removed coding parameter candidates for each non-first partition 60 in order to set the coding parameters associated with each partition in dependence on the one non-removed or selected coding parameter candidate if one or more such coding parameter candidates remain after the removal, the encoder 10 is configured to perform the removal in such a way that coding parameter candidates that would efficiently lead to a recombination of partitions 50 and 60 are removed, i.e. a collection of syntaxes that would thereby make an efficient partitioning situation more complex to code than would be the case if this partitioning were directly signaled simply using the partitioning information alone are effectively avoided.

[0019] Furthermore, as the set of coding parameter candidates becomes smaller, the amount of auxiliary information required to code the integration information into the bitstream 30 may decrease due to the smaller number of elements in these candidate sets. In particular, since the decoder is able to determine and then reduce the set of coding parameter candidates in the same way as the encoder of FIG. 1 does, the encoder 10 of FIG. 1 may utilize a reduced set of coding parameter candidates, for example using fewer bits to insert syntax elements into the bitstream 30, defining which of the coding parameter candidates that were not removed are to be used for integration. Of course, the introduction of syntax elements into the bitstream 30 may be suppressed entirely in the case where the number of coding parameter candidates that were not removed for each partition is simply one. In any case, by virtue of the integration, i.e. by setting the coding parameters associated with each partition depending on the remaining or selected one of the coding parameter candidates that were not removed, the encoder 10 is able to suppress an entirely new insertion of coding parameters for each partition into the bitstream 30, thereby reducing the auxiliary information as well. According to some embodiments of the present application, the encoder 10 may be configured to signal in the bitstream 30 refinement information for refining a remaining or selected one of the coding parameter candidates for each partition.

[0020] As mentioned above, according to the description of Fig. 1, the encoder 10 is configured to determine merging candidates to be eliminated by a comparison of their coding parameters with the coding parameters of the partitions, whereby merging results in another supported partitioning pattern. If, for example, the coding parameters of the left partition 50 form one element of the set of coding parameter candidates for the right partition 60, then this method of processing coding parameter candidates effectively eliminates at least one coding parameter candidate in the case shown in Fig. 1. However, further coding parameter candidates may also be eliminated if they are equal to the coding parameters of the left partition 50. However, according to another embodiment of the present invention, the encoder 10 may be configured to determine a set of candidate blocks for each partition after the second in the coding order by removing that or those candidate blocks from this set of candidate blocks that, when merged into the respective partition, would result in one of the supported partitioning patterns. In a sense, this means that: The encoder 10 may be configured to determine a combined candidate for each partition 50 or partition 60 (i.e., the first or the next in coding order) such that each element of the candidate set has exactly one partition of either the current block 40 or the previously coded block 40 associated with it in that the candidate adopts the respective coding parameters of its associated partition. For example, each element of the candidate set may be equal to, i.e., may be adopted from, one of such coding parameters of previously coded partitions, or at least may be derived from the coding parameters of exactly one such previously coded partition, such as by additional scaling or refinement using additionally transmitted refinement information.However, the encoder 10 may also be configured to add further elements or candidates to this kind of candidate set, i.e. to add coding parameter candidates obtained from a combination of coding parameters of one or more previously coded partitions or, by modification, from the coding parameters of one previously coded partition, such as by taking only the coding parameters of one motion parameter list. For "combined" elements, there is no 1:1 association between the coding parameters of each candidate element and each partition. According to a first variant of the description of FIG. 1, the encoder 10 may be configured to remove all candidates from the total candidate set, whose coding parameters are equal to the coding parameters of the partition 50. According to a latter variant of the description of FIG. 1, the encoder 10 may be configured to remove only those elements of the candidate set that are associated with the partition 50. Reconciling both views, the encoder 10 may be configured to remove candidates from a part of the candidate set that shows a 1:1 association with some (e.g. adjacent) previously coded partitions, without extending the removal (and the search for candidates with equal coding parameters) to the remaining part of the candidate set with coding parameters obtained by the combination. But of course, if one combination also leads to a redundant representation, this can be resolved by removing the redundant coding parameters from the list, or by performing a redundancy check on the combined candidates as well.

[0021] After describing an encoder according to an embodiment of the present invention, a decoder 80 according to an embodiment will be described with reference to Fig. 2. The decoder 80 of Fig. 2 is configured to decode a bitstream 30 signaling one of the supported partitioning patterns for a current block 40 of an image 20 as described above. If the signaled one of the supported partitioning patterns specifies a subdivision of the current block 40 into two or more partitions 50 and 60, the decoder 80 is configured to remove from the set of coding parameter candidates, for all partitions other than the first one of the partitions 50 in the coding order 70, i.e. for partition 60 in the example shown in Figs. 1 and 2, each partition coding parameter candidate having the same or equal coding parameters as the coding parameters associated with any of the partitions, which, when combined with each partition, becomes one of the supported partitioning patterns, i.e. one of the partitioning patterns that is not signaled in the bitstream 30 but is nevertheless supported.

[0022] That is, the decoder functionality largely corresponds to that of the encoder described with respect to Fig. 1. For example, the decoder 80 may be configured to set the coding parameters associated with each partition 60 depending on one of the non-removed coding parameter candidates if the number of non-removed coding parameter candidates is non-zero. For example, the decoder 80 sets the coding parameters of each partition 60 to be equal to one of the non-removed coding parameter candidates, with or without additional refinement and / or with or without scaling the coding parameters by the respective associated temporal distance. For example, the integrating coding parameter candidates among the non-removed candidates may have a reference image index associated therewith that is different from the reference image index explicitly signaled in the bitstream 30 for the partition 60. In that case, the coding parameters of the coding parameter candidates may define a motion vector associated with the respective reference image index, and the decoder 80 may be configured to scale the motion vector of the last selected non-removed coding parameter candidate according to the ratio between both reference image indexes. Thus, according to the variant described above, the coding parameters following the integration include the motion parameters, while the reference image index is separated therefrom, but as mentioned above, according to another embodiment, the reference image index can also be part of the coding parameters following the integration.

[0023] That the merging behavior may be restricted to inter-predicted blocks 40 applies equally to the encoder of Figure 1 and the decoder of Figure 2. Thus, the decoder 80 and the encoder 10 may be configured to support intra- and inter-prediction modes for the current block 40 and to perform candidate merging and removal only if the current block 40 is coded in an inter-prediction mode. Thus, only coding / prediction parameters of such inter-predicted previously coded partitions can be used to determine / build the candidate list.

[0024] As already mentioned above, the coding parameters may be prediction parameters and the decoder 80 may be configured to use the prediction parameters of partitions 50 and 60 to obtain a prediction signal for each partition. Of course, the encoder 10 also performs the extraction of the prediction signal as well. However, the encoder 10 additionally sets the prediction parameters together with all other syntax elements in the bitstream 30 in order to obtain some optimization in the sense of an appropriate optimization.

[0025] Furthermore, as already explained above, the encoder can be configured to insert an index into the non-removed coding parameter candidates only if the number of non-removed coding parameter candidates for each partition is greater than one. Thus, the decoder 80 can be configured to simply expect that the bitstream 30 will include a syntax element specifying which of the non-removed coding parameter candidates are used for the consolidation, depending on the number of non-removed coding parameter candidates for the partition 60, if the number of non-removed coding parameter candidates is greater than one. However, in cases where the candidate set is smaller than two in total, the performance of the candidate set reduction can be prevented from generally occurring by restricting the candidates obtained by adopting or deriving from the coding parameters of exactly one previously coded partition, by extending the list / set of candidates using combined coding parameters, i.e. parameters obtained by combining coding parameters of one or more or more previously coded partitions, as described above. Similarly, it is also possible to do the opposite, i.e. generally remove all coding parameter candidates having the same value as that of a partition that would normally result in another supported partitioning pattern.

[0026] In terms of the determination, the decoder 80 operates as the encoder 10 does. That is, the decoder 80 can be configured to determine a set of coding parameter candidates for partitions after the first partition 50 in the coding order 70 based on the coding parameters associated with previously decoded partitions. That is, the coding order is defined not only in the partitions 50 and 60 of each block 40, but also in the blocks 40 of the image 20 itself. All partitions that were coded before the partition 60 can thus serve as a reference for the determination of a set of coding parameter candidates for any of the following partitions, such as the partition 60 in the case of FIG. 2. As also explained above, the encoder and decoder can restrict the determination of the set of coding parameter candidates to partitions that are in a certain spatial and / or temporal neighborhood. For example, the decoder 80 can be configured to determine a set of coding parameter candidates for the non-first partition 60 based on the coding parameters associated with previously decoded partitions that are respectively adjacent to the non-first partition, such partitions being located outside and inside the current block 40. Of course, the determination of the merged candidate can also be performed for the first partition in the coding order. Only removal is not performed.

[0027] Consistent with the description of FIG. 1, the decoder 80 can be configured to determine a set of candidate coding parameters for each non-first partition 60 from an initial set of previously decoded partitions, excluding those that are coded in intra-prediction mode.

[0028] Furthermore, if the encoder introduces subdivision information into the bitstream to subdivide the image 20 into blocks 40, the decoder 80 may be configured to restore this subdivision of the image 20 into such coded blocks 40 according to the subdivision information in the bitstream 30.

[0029] With regard to figures 1 and 2, it should be noted that the residual signal for the current block 40 can be transmitted via the bitstream 30 with a granularity that can be different from the granularity defined by the partitions with respect to the coding parameters. For example, the encoder 10 of figure 1 can be configured to subdivide the block 40 into one or more transform blocks in a parallel or independent manner to the division into partitions 50 and 60. The encoder can signal each transform block subdivision for the block 40 by means of further subdivision information. The decoder 80 can then be configured to undo this further subdivision of the block 40 into one or more transform blocks by means of the further subdivision information in the bitstream and to derive the residual signal of the current block 40 from the bitstream in units of these transform blocks. The significance of the transform block division is that a transform such as the DCT in the encoder and a corresponding inverse transform such as the IDCT in the decoder are performed individually within each transform block of the block 40. To reconstruct the image 20 as block 40, the encoder 10 combines, e.g. by adding, the prediction and residual signals obtained by applying the coding parameters in each partition 50 and partition 60, respectively. It should be noted, however, that the residual coding does not involve a transform and an inverse transform, respectively, and the prediction residual is, for example, coded in the spatial domain instead.

[0030] Before describing further possible details of further embodiments below, possible internal structures of the encoder and decoder of Fig. 1 and Fig. 2 are described with reference to Fig. 3 and Fig. 4. Fig. 3 shows, as an example, how the encoder 10 is internally constructed. As shown in Fig. 3, the encoder 10 may include a subtractor 108, a transformer 100, and a bitstream generator 102, which may perform entropy coding, as shown in Fig. 3. The elements 108, 100, and 102 are connected in series between an input 112 receiving an image 20 and an output 114 outputting said bitstream 30. In particular, the subtractor 108 has its non-inverting input connected to the input 112, and the transformer 100 is connected between the output of the subtractor 108 and a first input of the bitstream generator 102, which in turn has an output connected to the output 114. The encoder 10 of Figure 3 further includes an inverse transformer 104 and an adder 110, which are serially connected, in the order shown, to the output of the transformer 100. The encoder 10 further includes a predictor 106 connected between the output of the adder 110, a further input of the adder 110 and an inverting input of the subtractor 108.

[0031] The elements of FIG. 3 interact as follows: The predictor 106 predicts parts of the image 20 using the result of the prediction, i.e. the prediction signal, applied to the inverting input of the subtractor 108. The output of the subtractor 108 then represents the difference between the prediction signal and each part of the image 20, i.e. the residual signal. The residual signal is subjected to a transform coding by the transformer 100, i.e. the transformer 100 may perform a transform, such as a DCT, to obtain transform coefficient levels, and a subsequent quantization on the transformed residual signal, i.e. the transform coefficients. The inverse transformer 104 reconstructs the final residual signal output by the transformer 100 to obtain a reconstructed residual signal, which corresponds to the residual signal input to the transformer 100 except for information loss due to the quantization of the transformer 100. The combination of the reconstructed residual signal and the prediction signal as output by the predictor 106 results in a reconstruction of each part of the image 20, which is sent from the output of the adder 110 to the input of the predictor 106. The predictor 106 operates in various modes as mentioned above, such as an intra-prediction mode, an inter-prediction mode, etc. The prediction mode and corresponding coding or prediction parameters applied by the predictor 106 to obtain a prediction signal are sent by the predictor 106 to the entropy coder 102 for insertion into the bitstream.

[0032] A possible implementation of the internal structure of the decoder 80 of Fig. 2, corresponding to the possibilities shown in Fig. 3 for the encoder, is shown in Fig. 4. As shown in the figure, the decoder 80 may include a bitstream extractor 150, implemented as an entropy decoder as shown in Fig. 4, an inverse transformer 152, and an adder 154, which are connected, in the order mentioned, between an input 158 ​​and an output 160 of the decoder. Furthermore, the decoder of Fig. 4 includes a predictor 156 connected between the output of the adder 154 and its further input. The entropy decoder 150 is connected to a parameter input of the predictor 156.

[0033] Briefly, the function of the decoder of Fig. 4 is that the entropy decoder 150 is intended to extract all the information contained in the bitstream 30. The entropy coding scheme used may be variable length coding or arithmetic coding. Thereby, the entropy decoder 150 recovers the residual signal from the bitstream representing the transform coefficient levels and sends it to the inverse transformer 152. Furthermore, the entropy decoder 150 recovers all the coding modes and related coding parameters from the bitstream and sends it to the predictor 156. In addition, the segmentation and synthesis information is extracted from the bitstream by the extractor 150. The inverse transformed, i.e. reconstructed, residual signal and the prediction signal as obtained from the predictor 156 are combined, for example by being added by the adder 154, and then the thus recovered reconstructed signal is output at the output 160 and sent to the predictor 156.

[0034] As will be apparent from a comparison of FIGS. 3 and 4, elements 152, 154 and 156 correspond functionally to elements 104, 110 and 106 of FIG.

[0035] In the above description of figures 1 to 4, several different possibilities have been presented regarding possible subdivisions of the image 20 and the corresponding granularity in varying some of the parameters related to the encoded image 20. Such possibilities are also described with respect to figures 5a and 5b. Figure 5a shows a portion of the image 20. According to the embodiment of figure 5a, the encoder and decoder are configured to firstly subdivide the image 20 into tree root blocks 200. One such tree root block is shown in figure 5a. The subdivision of the image 20 into tree root blocks is made regularly in rows and columns as shown by the dotted lines. The size of the tree root blocks 200 can be selected by the encoder and signalled to the decoder by the bitstream 30. Alternatively, the size of these tree root blocks 200 can be fixed by default. The tree root blocks 200 are subdivided using a quad-tree partitioning to give rise to the above-mentioned differentiated blocks 40, which may be called coding blocks or coding units. These coding blocks or coding units are depicted by thin solid lines in figure 5a. Thereby, the encoder adds subdivision information to each tree root block 200 and inserts the subdivision information into the bitstream. This subdivision information indicates how the tree root block 200 is to be sub-divided into blocks 40. At the granularity of these blocks 40, and for which unit, the prediction mode changes in the image 20. As mentioned above, each block 40, or each block having a particular prediction mode, such as an inter prediction mode, is accompanied by partition information about which supported partitioning pattern is used for each block 40. In the case shown for FIG. 5a, for many coding blocks 40, a non-partitioned mode is selected such that the coding blocks 40 spatially match the corresponding partitions. In other words, the coding blocks 40 are simultaneously partitions with respective sets of prediction parameters associated with them. The type of prediction parameters then depends on the mode associated with each coding block 40. However, other coding blocks are shown as being further divided as an example.The coding block 40 at the top right corner of the tree root block 200 is shown to be divided into, for example, four partitions, while the coding block at the bottom right corner of the tree root block 200 is exemplarily shown to be subdivided vertically into two partitions. The subdivision for the division into partitions is indicated by dotted lines. FIG. 5a also shows the coding order within the partitions thus defined. As shown in the figure, a depth-first scan order is used. Across the tree root block boundary, the coding order may continue in a scan order in which the rows of the tree root block 200 are scanned row by row from the top to the bottom of the image 20. In this manner, it is possible to have an opportunity to maximize the likelihood that a particular partition has a previously coded partition adjacent to its upper and left boundaries. Each block 40, or each block having a particular prediction mode, such as an inter prediction mode, may have a merge switch indicator in the bitstream indicating whether merge is activated for the corresponding partition therein. It should be noted that the division of blocks into partitions / prediction units can be limited to a division into at most two partitions, with the exception that this rule is only made with respect to the smallest possible block size of blocks 40. This can avoid redundancy between the subdivision information for subdividing the image 20 into blocks 40 and the partitioning information for subdividing the blocks 40 into partitions, when using quadtree subdivision to obtain the blocks 40. Alternatively, only a division into one or two partitions, with or without asymmetric ones, can be allowed.

[0036] 5b shows a subdivision tree: for the solid lines a subdivision of the tree root block 200 is shown, while the dotted lines represent the division of the leaf blocks of the quadtree subdivision, which are the coding blocks 40. That is to say, the division of the coding blocks represents a kind of extension of the quad subdivision.

[0037] As already mentioned above, each coding block 40 may be subdivided in parallel into transform blocks, such that the transform blocks may represent different subdivisions of each coding block 40. A transformation for transforming the residual signal of the coding block into each of these transform blocks, which are not shown in Figures 5a and 5b, may be performed separately.

[0038] In the following, further embodiments of the present invention are described. While the above embodiments have focused on the relationship between block merging and block division, the following description also includes aspects of the present application related to other coding principles known in current codecs, such as SKIP / DIRECT modes. Nevertheless, the following description is not to be considered as merely describing another embodiment, i.e. an embodiment separated from the above. Rather, the following description also clarifies details of possible implementations related to the above embodiment. The following description therefore uses the reference coding of the figures already shown above, so that each possible implementation described below also defines possible variations of the above embodiment. Most of these variations can be transferred individually to the above embodiment.

[0039] In other words, the present embodiment describes a method for reducing the rate of auxiliary information in image and video coding applications by aggregating syntax elements associated with a particular set of samples, i.e. blocks, in order to transmit the associated coding parameters. The present embodiment may in particular consider the combination of the aggregated syntax elements with the division of the image parts into various division patterns, and with SKIP / DIRECT modes, in which the coding parameters are determined from the spatial and / or temporal neighbors of the current block. To that extent, the above embodiment may be modified to perform the aggregation for a set of samples, i.e. blocks, in combination with various division patterns and SKIP / DIRECT modes.

[0040] Furthermore, before describing these variations and details, a brief overview regarding image and video codecs is provided.

[0041] In image and video coding applications, a sample array associated with an image is usually partitioned into a specific set of samples (or sample sets), which may represent rectangular or square blocks or other collections of samples, including arbitrarily shaped regions, triangles, or other shapes. The subdivision of the sample array may be fixed by a syntax, or the subdivision is signaled (at least partially) in the bitstream. In order to keep the auxiliary information rate for signaling the subdivision information small, the syntax usually allows only a limited number of choices, which are simple subdivisions, such as subdivision of a block into smaller blocks. A commonly used partitioning scheme is the partitioning of a square block into four smaller square blocks, or into two rectangular blocks of the same size, or into two rectangular blocks of different sizes. Here, the actual partitioning used is signaled in the bitstream. The sample sets are associated with specific coding parameters, which may specify prediction information or residual coding modes, etc. In video coding applications, partitioning is often done for motion representation. All samples of a block (within a partitioning pattern) are associated with the same set of motion parameters, which may include parameters specifying the type of prediction (e.g., list 0, list 1, or bi-prediction; and / or translation or affine prediction or prediction using different motion models), parameters specifying the reference image to be used, parameters specifying the motion relative to the reference image that is usually transmitted as a difference to the predictor (e.g., a displacement vector, an affine motion parameter vector, a motion parameter vector for other motion models), the precision of the motion parameters (e.g., half-sample or quarter-sample precision), parameters specifying the weighting of the reference sample signals (e.g., for illumination compensation), or parameters specifying the interpolation filters used to obtain the motion-compensated prediction signal of the current block. It is assumed that for each sample set, individual coding parameters are transmitted (e.g., to specify the prediction and / or residual coding).To obtain improved coding efficiency, the present invention presents a method and specific embodiments for merging two or more sample sets into a so-called group of sample sets. All sample sets of such a group share the same coding parameters, which can then be transmitted together with one of the sample sets in the group. By doing this, the coding parameters do not have to be transmitted for each sample set of the group of sample sets individually, instead the coding parameters are transmitted only once for the whole group of sample sets. As a result, the auxiliary information rate for transmitting the coding parameters is reduced and the overall coding efficiency is improved. As an alternative approach, an additional refinement for one or more of the coding parameters can be transmitted for one or more of the sample sets of the group of sample sets. The refinement can be applied to all sample sets of the group or only to the sample set for which it is transmitted.

[0042] An embodiment of the present invention is particularly concerned with the combination of the merging process with the division of the block into various sub-blocks 50, 60 (as described above). Usually, an image or video coding system supports various partitioning patterns for the block 40. As an example, a square block may be not divided or may be divided into four square blocks of the same size, into two rectangular blocks of the same size (where the square block is divided vertically or horizontally), or into rectangular blocks of different sizes (vertically or horizontally). The described typical partition patterns are shown in FIG. 6. In addition to the above description, the division may even involve one or more levels of division. For example, a square sub-block may also be optionally further divided using the same division pattern. The problem that arises when this kind of division process is combined with a merging process that allows a (square or rectangular) block to be merged with, for example, one of its neighboring blocks, is that the division that occurs may be obtained by different combinations of the division pattern and the merging signal. Therefore, the same information may be transmitted in the bitstream using different codewords, which is obviously suboptimal in terms of coding efficiency. As a simple example, consider a square block that is not further divided (as shown in the upper left corner of FIG. 6). This division can be signaled directly by sending a syntax element that this block 40 is not to be subdivided. However, the same pattern can also be signaled by sending a syntax element that specifies that this block is to be subdivided, for example, into two vertically (or horizontally) arranged rectangular blocks 50, 60. Then, we can send merge information that specifies that the second of these rectangular blocks is to be merged with the first rectangular block, resulting in exactly the same division as when we signal that the block is not to be further divided. The same can be obtained by first specifying that the block is to be subdivided into four square sub-blocks and then sending merge information that effectively merges all these four blocks. This concept is clearly suboptimal (because we have different code words to signal the same thing).

[0043] The present invention describes the concept and possibility of reducing the auxiliary information rate and thus increasing the coding efficiency for the combination with the concept of merging with the concept of providing different partitioning patterns for blocks. Looking at the partitioning pattern example of FIG. 6, the "simulation" of the block that was not further divided by any of the partitioning patterns with two rectangular blocks can be avoided when the rectangular block is forbidden (i.e., excluded from the bitstream syntax specification) when it is merged with the first rectangular block. Looking at the problem more deeply, it is also possible to "simulate" the pattern that was not subdivided by merging the second rectangular block with other neighbors (i.e., non-first rectangular blocks) that are related with the same parameters (e.g., information for defining prediction) as the first rectangular block. The present invention conditions the transmission of the merged information in such a way that the transmission of certain merged parameters is excluded from the bitstream syntax when these merged parameters result in a pattern that can also be obtained by signaling one of the supported partitioning patterns. As an example, if the current partitioning pattern specifies a subdivision into two rectangular blocks, as shown in Fig. 1 and Fig. 2, before sending the merger information for the second block, i.e. block 60 in the case of Fig. 1 and Fig. 2, it can be checked which of the possible merge candidates has the same parameters (e.g. parameters for specifying a prediction signal) as the first rectangular block, i.e. block 50 in the case of Fig. 1 and Fig. 2. Then, all candidates with the same motion parameters (including the first rectangular block itself) are removed from the set of merger candidates. The codeword or flag sent to signal the merger information is adapted to the resulting candidate set. If the candidate set is empty by the parameter check, no merger information is sent. If the candidate set consists of just one entry, it is only signaled whether the block is merged or not, and the candidate does not need to be signaled, since it is obtained on the decoder side, etc.As for the above example, the same concept is also used for a partitioning pattern that partitions a square block into four smaller square blocks. Here, the transmission of the integration flag is adapted in such a way that neither a partitioning pattern that specifies no subdivision, nor two partitioning patterns that specify subdivision into two rectangular blocks of the same size, can be achieved by a combination of integration flags. Although most of the concepts have been described in the above example with specific partitioning patterns, it is clear that the same concept (avoiding the specification of a specific partitioning pattern by combination with other partitioning patterns and corresponding integration information) can be used for other sets of partitioning patterns.

[0044] The advantage of the described invention over schemes where only partitioning is allowed is that a much greater degree of freedom is provided with regard to signaling the partitioning of an image into parts related to the same parameters (e.g. for defining a prediction signal). As an example, an additional partitioning pattern resulting from the merging of a square block of a subdivided larger block is represented in FIG. 7. However, it should be noted that a great many resulting patterns can be achieved by merging with further adjacent blocks (outside the previously subdivided block). With only a few codewords for signaling partitioning and merging information, various partitioning possibilities are provided and the encoder can select the best option (for a given encoder complexity) in terms of rate distortion (e.g. by minimizing a certain rate-distortion measure). The advantage compared to approaches where only one partitioning pattern (e.g. subdivision into four blocks of the same size) is provided in combination with a merging approach is that commonly used patterns (e.g. rectangles of different sizes) can be signaled by short codewords instead of several partitioning and merging flags.

[0045] Another aspect that needs to be considered is that the integration concept is in some sense similar to the SKIP or DIRECT modes found in video coding designs. In the SKIP / DIRECT modes, basically, motion parameters are not transmitted for the current block, but are determined from spatial and / or temporal neighbors. In a particular efficient concept of the SKIP / DIRECT mode, a list of motion parameter candidates (reference frame indexes, displacement vectors, etc.) is generated from spatial and / or temporal neighbors, and an index into this list is transmitted that specifies which of the candidate parameters is selected. For bi-predicted blocks (or multiple hypothesis frames), another candidate can be signaled for each reference list. Possible candidates include the block above the current block, the block to the left of the current block, the block to the top left of the current block, the block to the top right of the current block, the median predictor of various of these candidates, the co-located block in one or more previous reference frames (or other already coded blocks, or a combination obtained from already coded blocks). When combining the integration scheme with the SKIP / DIRECT mode, it must be ensured that both the SKIP / DIRECT mode and the integration mode should not contain the same candidates. This can be achieved by different configurations. It is possible to enable the SKIP / DIRECT mode (e.g. with more candidates than the integration mode) only for certain blocks (e.g. with sizes larger than a specified size, or only for square blocks, etc.) and not support the integration mode for these blocks. Or the SKIP / DIRECT mode can be deleted and all candidates (including parameters indicating the combination of parameters for spatially / temporally adjacent blocks) are added as candidates to the integration mode. This option has already been mentioned above with reference to Figs. 1 to 5. The increased candidate set can only be used for certain blocks (such as those with sizes larger than a certain minimum size, or square blocks), while for other blocks a reduced candidate set is used.Or, as a further variant, the integration mode is used for a reduced candidate set (e.g. only the top and left neighbors) and the additional candidates (e.g. top left, co-located blocks, etc.) are used for the SKIP / DIRECT mode. Also, in this kind of configuration, the SKIP / DIRECT mode can only be enabled for certain blocks (such as those with a size larger than a certain minimum size, or square blocks), while the integration mode is enabled for a larger set of blocks. The advantage of this kind of combination is that multiple options for signaling the reuse of already transmitted parameters (e.g. to define prediction) are provided for different block sizes. As an example, for larger square blocks, more options are provided, since here the additional bit rate spent provides an increase in the rate-distortion efficiency. For smaller blocks, a smaller set of options is given. An increase in the candidate set would not result in a gain in the rate-distortion efficiency, due to the small ratio of samples per bit required to signal the selected candidate here.

[0046] As mentioned above, the embodiment of the present invention also provides the encoder with more freedom to generate a bitstream, since the merging approach significantly increases the number of possibilities for selecting a partition for the sample array of an image. The coding efficiency can be improved, since the encoder can choose among more options, for example, to minimize a certain rate-distortion measure. As an example, some of the additional patterns that can be exhibited by the combination of subdivision and merging (e.g., the patterns in FIG. 7) can be additionally tested (using the corresponding block sizes for motion estimation and mode decision), and the best pattern provided by purely partitioning (FIG. 6) and by partitioning and merging (FIG. 7) can be selected based on a certain rate-distortion measure. In addition, for each block, it can be tested whether merging with any of the already coded candidate sets results in a reduction of a certain rate-distortion measure, and the corresponding merging flag is set during the coding process. In summary, there are several possibilities to operate the encoder. In a simple approach, the encoder can first determine the maximum subdivision of the sample array (as the highest-level coding scheme). Then, for each sample set, it can check whether merging with other sample sets or other groups of sample sets reduces the particular rate-distortion cost measure. Here, the prediction parameters associated with the merged group of sample sets can be re-estimated (e.g., by performing a new motion search). Or, the prediction parameters already determined for the current sample set and the candidate sample sets (or groups of sample sets) for merging can be evaluated for the considered group of sample sets. In a more extensive approach, the particular rate-distortion cost measure can be evaluated for additional candidate groups of sample sets. As a special case, when testing various possible splitting patterns (e.g., see FIG. 6), some or all of the patterns that can be exhibited by a combination of splitting and merging (e.g., see FIG. 7) can be tested in addition.That is, for all of the patterns, a specific motion estimation and mode decision process is performed, and the pattern that obtains the smallest rate-distortion measure is selected. This process can also be combined with the low-complexity process described above. As a result, for the resulting block, it is additionally tested whether integration with already coded blocks (e.g., outside the patterns of Figures 6 and 7) results in a decrease in the rate-distortion measure.

[0047] In the following, some possible detailed implementations for the above outlined embodiment are described, for example with respect to the encoders of Figs. 1 and 3 and the decoders of Figs. 2 and 4. As already mentioned above, it can be used in image and video coding. As mentioned above, an image or a specific set of sample arrays for an image can be decomposed into blocks, which are associated with specific coding parameters. An image usually consists of several sample arrays. In addition, an image can be associated with additional auxiliary sample arrays, which can define, for example, transparency information or a depth map. The sample arrays of an image (including the auxiliary sample arrays) can be classified into one or several so-called plane groups, where each plane group consists of one or several sample arrays. The plane groups of an image can be coded independently or, in case an image is associated with several plane groups, by prediction from other plane groups of the same image. Each plane group is usually decomposed into blocks. The blocks (or the corresponding blocks of the sample arrays) are predicted by inter-image prediction or intra-image prediction. The blocks can have different sizes and can be square or rectangular. The division of the image into blocks can be fixed by a syntax or it can be (at least partially) signaled in the bitstream. Often syntax elements are transmitted that signal the subdivision for blocks of a defined size. This kind of syntax element can specify if and how a block is subdivided into smaller blocks and is associated with coding parameters, for example for prediction purposes. An example of a possible division pattern is shown in Figure 6. For all samples of a block (or a corresponding block of a sample array), the decoding of the associated coding parameters is specified in a specific way.In an example, all samples of a block are predicted using the same set of prediction parameters, such as a reference index (identifying a reference image in a set of already coded images), motion parameters (defining a measure for the motion of the block between the reference image and the current image), parameters for defining an interpolation filter, intra prediction mode, etc. The motion parameters can be indicated by a displacement vector with horizontal and vertical components, or by higher-order motion parameters, such as affine motion parameters consisting of six components. It is also possible that multiple sets of specific prediction parameters (e.g., reference indexes and motion parameters) are associated with one block. In that case, for each set of these specific prediction parameters, one intermediate prediction signal for the block (or the corresponding block of the sample array) is generated, and a final prediction signal is constructed by a combination that includes superposing the intermediate prediction signals. The corresponding weighting parameters and possibly a constant offset (added to the weighted sum) may be fixed with respect to the image, or with respect to the reference image, or with respect to the reference image set, or they may be included in the set of prediction parameters for the corresponding block. The difference between the original blocks (or the corresponding blocks of the sample array) and their prediction signal (also called the residual signal) is usually transformed and quantized. Often, a two-dimensional transform is applied to the residual signal (or the corresponding sample array for the residual block). For transform coding, the block (or the corresponding block of the sample array) for which a particular set of prediction parameters was used can be further divided before applying the transform. The transform block can be equal to or smaller than the block used for prediction. It is also possible that the transform block contains two or more of the blocks used for prediction. Different transform blocks can have different sizes and the transform blocks can represent square or rectangular blocks.It is noted that in the above examples with respect to figures 1 to 5, the leaf nodes of the first subdivision, i.e. the coding blocks 40, can be further divided in parallel, on the one hand, into partitions defining the granularity of the coding parameters, and on the other hand, into transform blocks, to which a two-dimensional transform is applied individually. After transformation, the resulting transform coefficients are quantized and so-called transform coefficient levels are obtained. The transform coefficient levels as well as the prediction parameters and, if any, the subdivision information are entropy coded.

[0048] In state-of-the-art image and video coding standards, the possibilities of subdividing an image (or planes) into blocks provided by the syntax are very limited. Usually, it can only be specified whether (and possibly how) a block of a predefined size can be subdivided into smaller blocks. For example, the maximum block size in H.264 is 16x16. A 16x16 block is also called a macroblock, and each image is divided into macroblocks in a first step. For each 16x16 macroblock, it can be signaled whether it is coded as a 16x16 block, or as two 16x8 blocks, or as two 8x16 blocks, or as four 8x8 blocks. If a 16x16 block is subdivided into four 8x8 blocks, each of these 8x8 blocks can be coded as one 8x8 block, or as two 8x4 blocks, or as two 4x8 blocks, or as four 4x4 blocks. The small set of possibilities for defining the partition into blocks of state-of-the-art image and video coding standards has the advantage that the auxiliary information rate for signaling the subdivision information can be kept small, but the disadvantage that the bit rate required for transmitting the prediction parameters for a block can become significant, as explained below. The auxiliary information rate for signaling the prediction information usually represents a significant amount of the overall bit rate for a block. And when this auxiliary information is reduced, the coding efficiency can be increased, which can be achieved, for example, by using a larger block size. It is also possible to increase the set of supported partition patterns compared to H.264. For example, the partition patterns shown in FIG. 6 can be provided for square blocks of all sizes (or a selected size). Real images or pictures of video sequences consist of arbitrarily shaped objects with certain properties. For example, objects or parts of objects of this kind are characterized by a unique texture or a unique motion.And usually the same set of prediction parameters can be used for this kind of object or part of the object. But object boundaries usually do not coincide with possible block boundaries for larger prediction blocks (e.g. 16x16 macroblocks in H.264). The encoder usually decides the subdivision (among a limited set of possibilities) that results in the minimum of a certain rate-distortion cost measure. For arbitrarily shaped objects, this results in a large number of small blocks. This statement also holds when many partitioning patterns (as mentioned above) are provided. It should be noted that the amount of partitioning patterns should not become too large, since a lot of side information and / or coder / decoder computational effort would be required to signal and process these patterns. So arbitrarily shaped objects often result in a large number of small blocks due to the partitioning. And since each of these small blocks is associated with a set of prediction parameters that needs to be transmitted, the side information rate can become a significant part of the overall bitrate. But since some of the small blocks still represent the same object area or part of the object, the prediction parameters for many resulting blocks are the same or very similar. Intuitively, the coding efficiency can be increased when the syntax is extended in a way that allows not only to subdivide a block, but also to merge two or more blocks obtained after the subdivision. As a result, one obtains a group of blocks that are coded with the same prediction parameters. The prediction parameters for such a group of blocks need to be coded only once. In the above example of Figs. 1-5, for example, if a merge occurs, i.e., the reduced set of candidates does not become zero, no coding parameters for the current block 40 are transmitted. That is, the encoder does not transmit coding parameters associated with the current block, and the decoder does not expect the bitstream 30 to include coding parameters for the current block 40. Rather, according to the particular embodiment, only refinement information can be transmitted for the merged current block 40.The determination of the candidate sets and their reduction and merging etc. are performed for other coding blocks 40 of the image 20. The coding blocks somehow form groups of coding blocks along the coding chain, where the coding parameters for these groups are transmitted in the bitstream only once in their entirety.

[0049] The described merging results in increased coding efficiency if the bitrate saved by reducing the number of coding prediction parameters is larger than the bitrate additionally spent to code the merging information. It should be further mentioned that the described syntax extension (for merging) provides the encoder with additional freedom in choosing the division of the image or plane group into blocks. The encoder is not restricted to first make a subdivision and then check whether some of the resulting blocks have the same set of prediction parameters. As one simple variant, the encoder can first determine the subdivision as the state-of-the-art coding technique. Then, for each block, it can check whether merging with one of its neighboring blocks (or related already determined groups of blocks) reduces the rate-distortion cost measure. In this, the prediction parameters associated with the new group of blocks can be estimated again (for example by performing a new motion search) or the prediction parameters already determined for the current block and the neighboring blocks or groups of blocks can be evaluated for the new group of blocks. The encoder can also directly check (a subset of) the patterns provided by the combination of partitioning and merging. That is, motion estimation and mode decision can be done on the resulting shape as already mentioned above. The integration information can be signaled on a block basis. In fact, the integration can also be interpreted as the result obtained in estimating the prediction parameters for the current block, where the estimated prediction parameters are set equal to the prediction parameters of one of the neighboring blocks.

[0050] In this, it should be noted that combinations of different partitioning patterns and merger information can result in the same shape (associated with the same parameters). This is clearly suboptimal, since the same message can be transmitted by different combinations of coded words. To avoid (or reduce) this drawback, an embodiment of the present invention presents a scheme that prohibits the same shape (associated with a particular set of parameters) from being signaled by different partitioning and merge syntax elements. Thus, for all blocks of a previously subdivided block other than the first one in the coding order, 10, 50, etc., encoder and decoder, it is checked for all merge candidates whether the merger would result in a pattern that could be signaled by partitioning without merger information. All candidate blocks for which this is true are removed from the set of merge candidates and the transmitted merger information is applied to the resulting candidate set. If no candidates remain, no merger information is transmitted. If only one candidate remains, a flag is transmitted that specifies whether the block is merged or not, etc. As a further example of this scheme, a preferred embodiment is described below. The advantage of the described embodiment with respect to the concept where only partitioning is allowed is that much more freedom is given to signal the partitioning of the image into parts that are associated with the same parameters (e.g. to define a prediction signal).The advantage compared to an approach where only one partitioning pattern (e.g. subdivision into 4 blocks of the same size) is provided in combination with a synthesis approach is that commonly used patterns (e.g. rectangles of different sizes) can be signaled by a short coding word instead of several subdivision and synthesis flags.

[0051] State-of-the-art video coding standards such as H.264 also include specific inter-coding modes called SKIP and DIRECT modes, in which the parameters defining the prediction are determined entirely from spatially and / or temporally adjacent blocks. The difference between SKIP and DIRECT is that the SKIP mode further signals that no residual signal is transmitted. In various proposed improvements of the SKIP / DIRECT modes, instead of one candidate (as in H.264), a list of possible candidates is determined from the spatial and / or temporal neighborhood of the current block. The possible candidates may include a block to the top of the current block, a block to the left of the current block, a block to the top left of the current block, a block to the top right of the current block, a median predictor of various of these candidates, a co-located block of one or more previous reference frames (or other already coded blocks, or a combination derived from already coded blocks). With regard to combination with the unified mode, it must be ensured that both the SKIP / DIRECT mode and the unified mode must not contain the same candidates. This can be achieved by various configurations, as described above. An advantage of the described combination is that multiple options for signaling the reuse of already transmitted parameters (eg, to define prediction) are provided for different block sizes.

[0052] One advantage of the present invention is to reduce the bit rate required to transmit prediction parameters by merging adjacent blocks into groups of blocks, where each group of blocks is associated with a unique set of coding parameters, e.g. prediction parameters and residual coding parameters. The merger information is signaled in the bitstream (in addition to the subdivision information, if any). By combining different partitioning patterns with SKIP / DIRECT modes and sending corresponding merger information, it can be ensured that the SKIP / DIRECT modes are "simulated" and that none of the provided patterns are "simulated". An advantage of the present invention is an increased coding efficiency resulting from a reduced auxiliary information rate for the coding parameters. The present invention can be applied in image and video coding applications, where a set of samples is associated with a particular coding or prediction parameter. The merger process described here also extends to three dimensions or higher dimensions. For example, blocks of several video image groups can be merged into one group of blocks. It can also be applied to 4D compression of light field coding. On the other hand, it can also be used for compression of 1D signals. Here, the 1D signal is split and the given splits are merged.

[0053] An embodiment of the present invention also relates to a method for reducing the auxiliary information rate in image and video coding applications. In image and video coding applications, a specific set of samples (which may represent a rectangular or square block or an arbitrarily shaped region or any other collection of samples) is usually associated with a specific set of coding parameters. For each of these sample sets, coding parameters are included in the bitstream. The coding parameters may represent prediction parameters that define how the corresponding set of samples is predicted using already coded samples. The division of the image sample array into sample sets can be fixed by syntax or can be signaled by corresponding subdivision information in the bitstream. Multiple division patterns for blocks can be allowed. The coding parameters for the sample sets are transmitted in a predefined order given by syntax. An embodiment of the present invention also shows how it can be signaled for a current set of samples to be combined (e.g. for prediction purposes) with one or more other sample sets into a group of sample sets. Thus, a possible set of values ​​for the corresponding combination information is applied to the division pattern used in such a way that a specific division pattern cannot be indicated by a combination of other division patterns and corresponding combination data. The coding parameters for a group of sample sets need only be transmitted once. In a particular embodiment, if the current sample set is merged with a sample set (or group of sample sets) for which coding parameters have already been transmitted, the coding parameters of the current sample set are not transmitted. Instead, the coding parameters for the current set of samples are set equal to the coding parameters of the sample set (or group of sample sets) with which the current set of samples is merged. As an alternative approach, additional refinements for one or more of the coding parameters can be transmitted for the current sample set.The refinement can be applied to all sample sets of the group or only to the sample set for which it is transmitted.

[0054] In a preferred embodiment, for each set of samples, the set of all previously coded sample sets is called the "set of causal sample sets". The set of samples that can be used for merging with the current set of samples is called the "set of candidate sample sets" and is always a subset of the "set of causal sample sets". The way in which this subset is formed can be known to the decoder or it can be specified in the bitstream. In any case, the encoder 10 and the decoder 80 determine the candidate set to be reduced. If a particular current set of samples is coded and that set of candidate sample sets is not empty, the current set of samples is merged with one sample set of this set of candidate sample sets or (if so / if there are multiple candidates) which one of them is signaled. Otherwise, merging cannot be used for this block. Candidate blocks for which merging results in a shape that is also directly specified by the partitioning pattern are excluded from the candidate set to avoid that the same shape can be indicated by different combinations of partitioning information and merging data. That is, the candidate set is reduced by removing candidates as described above with respect to FIGS.

[0055] In a preferred embodiment, the set of some candidate sample sets is zero or more sample sets that contain at least a certain non-zero number of samples (which can be one, two or more) that represent direct spatial neighbors of every sample in the current set of samples. In another preferred embodiment of the present invention, the set of candidate sample sets may additionally (or exclusively) contain samples that have the same spatial location, i.e., are contained by both the candidate sample set and the current sample set currently affected by the integration, but are contained in different images (which can be one, two or more). In another preferred embodiment of the present invention, the set of candidate sample sets may be extracted from previously processed data in the current image or in another image. The extraction method may include spatial direction information, such as a transformation coefficient related to a certain direction or image gradient of the current image, or it may include temporal direction information, such as neighboring motion indications. From this type of data and other data and auxiliary information (if any) available at the receiver, the set of candidate sample sets can be extracted. Removal of candidates (from the original candidate set) that result in the same shape as can be represented by a particular partitioning pattern is extracted in the same way in the encoder and decoder, so that the encoder and decoder extract the final candidate set for combination in exactly the same way.

[0056] In a preferred embodiment, the considered set of samples is a rectangular or square block. Then, the aggregated set of samples represents a collection of rectangular and / or square blocks. In another preferred embodiment of the present invention, the considered set of samples is an arbitrarily shaped image area and the aggregated set of samples represents a collection of arbitrarily shaped image areas.

[0057] In a preferred embodiment, one or more syntax elements are transmitted for each set of samples, which specify whether the set of samples is to be aggregated with other sample sets (which may be part of an already aggregated group of sample sets) and which of the set of candidate sample sets is to be used for aggregation. However, if the candidate set is empty (e.g. due to removal of a candidate that results in a split that may be signaled by a different splitting pattern without aggregation), no syntax elements are transmitted.

[0058] In a preferred embodiment, one or two syntax elements are transmitted to define the integration information: a first syntax element defines whether the current set of samples is integrated with another sample set; a second syntax element, transmitted only if the first syntax element defines that the current set of samples is integrated with another set of samples, defines which of the set of candidate sample sets is used for integration. In a preferred embodiment, the first syntax element is transmitted only if the extracted set of candidate sample sets is not empty (after possible removal of candidates that would result in a split that could be signaled by a different splitting pattern without integration). In another preferred embodiment, the second syntax element is transmitted only if the extracted set of candidate sample sets contains one or more sample sets. In a further preferred embodiment of the present invention, the second syntax element is transmitted only if at least two sample sets of the extracted set of candidate sample sets are associated with different coding parameters.

[0059] In a preferred embodiment of the present invention, the integration information for a set of samples is coded before the prediction parameters (or, more generally, the particular coding parameters associated with the sample set). The prediction or coding parameters are transmitted only if the integration information signal signals that the current set of samples is not to be integrated with other sets of samples.

[0060] In another preferred embodiment, the joint information for the set of samples is coded after a subset of prediction parameters (or, more generally, specific coding parameters associated with the sample set) has been transmitted. The subset of prediction parameters may consist of one or more reference image indexes, or one or more components of a motion parameter vector, or reference indexes, and one or more components of a motion parameter vector, etc. The already transmitted subset of prediction or coding parameters may be used to extract a (reduced) set of candidate sample sets. As an example, a difference measure between the already coded prediction or coding parameters and the corresponding prediction or coding parameters of the original set of candidate sample sets may be calculated. Then, only those sample sets for which the calculated difference measure is smaller than or equal to a predefined or extracted threshold are included in the final (reduced) set of candidate sample sets. The threshold may be obtained based on the calculated difference measure. Or, as another example, only those sets of samples for which the difference measure is minimized are selected. Or only one set of samples is selected based on the difference measure. In the latter case, the merger information may be reduced in such a way that it only defines whether the current set of samples is merged into one candidate set of samples.

[0061] The preferred embodiment below is described for a set of samples representing rectangular and square blocks, but it can be extended in a straightforward manner to arbitrarily shaped regions or other collections of samples.

[0062] 1. Extract the first set of candidate blocks The extraction of the initial set of samples described in this section concerns the extraction of the initial candidate set. Some of all candidate blocks can be subsequently removed by analyzing the relevant parameters (e.g., predictive information) and removing those candidate blocks whose consolidation results in a final partition that can also be obtained by using other partitioning patterns. This process is described in the next subsection.

[0063] In a preferred embodiment, the set of initial candidate blocks is formed as follows: starting from the top-left sample position of the current block, its left adjacent sample position and its top-most adjacent sample position are extracted. The set of initial candidate blocks has at most only two elements, i.e. those blocks from the set of causal blocks that contain one of the two sample positions. Thus, the set of initial candidate blocks can only have as its elements the two direct adjacent blocks of the top-left sample position of the current block.

[0064] In another preferred embodiment of the present invention, the set of initial candidate blocks is given by all blocks coded before the current block and containing one or more samples that represent direct spatial neighbors of any sample of the current block (direct spatial neighbors can be limited to the direct left and / or the direct upper and / or the direct right and / or the direct lower neighbors). In another preferred embodiment of the present invention, the set of initial candidate blocks additionally (or exclusively) includes blocks that are co-located with any of the samples of the current block but contain one or more samples that are included in a different (already coded) image. In another preferred embodiment of the present invention, the initial candidate set of blocks represents a subset of the above-mentioned set of (adjacent) blocks. The subset of candidate blocks can be fixed, signaled or derived. The extraction of the subset of candidate blocks can take into account decisions made for other blocks of the image or of other images. As an example, blocks that are associated with the same (or very similar) coding parameters as other candidate blocks may not be included in the initial candidate set of blocks.

[0065] In a preferred embodiment of the present invention, the set of initial candidate blocks is extracted as for one of the above embodiments, but with the following restrictions: Only blocks using motion compensated prediction (inter prediction) can be members of the set of candidate blocks, i.e. intra-coded blocks are not included in the (initial) candidate set.

[0066] As already mentioned above, it is possible to extend the list of candidates with combined bi-predictive merging candidates, unscaled bi-predictive merging candidates, and additional candidates for block merging with zero motion vectors, etc.

[0067] The extraction of the initial set of candidate blocks is similarly performed by both the encoder and the decoder.

[0068] 2. Extracting the final set of candidate blocks After extracting the initial candidate set, the relevant parameters of the candidate blocks in the initial candidate set are analyzed to remove the merging candidates whose merging results in a division that can be represented by using a different division pattern. If the sample sequences that can be merged are of different shapes and / or sizes, there may be an identical division that can be represented by at least two different code words. For example, if the encoder decides to split a sample sequence into two sample sequences, this division will be reversed by merging the two sample sequences. To avoid this kind of redundant description, the set of candidate blocks for merging is constrained according to the specific block shapes and divisions allowed. On the other hand, the allowed shapes of the sample sequences can be constrained according to the specific candidate list used for merging. The two functions of division and merging need to be designed together so that the redundant description is avoided by combining the two.

[0069] In a preferred embodiment of the invention, the set of splitting modes (or division modes) shown in Figure 6 is supported for square blocks. If a square block of a certain size is divided into four smaller square blocks of the same size (lower left pattern in Figure 6), the same set of division patterns can be applied to the resulting four square blocks so that a hierarchical division can be defined.

[0070] After extracting the initial set of candidate blocks, the reduction of the candidate list is done as follows. If the current block is not further split (top left pattern in Fig. 6), the initial candidate list is not reduced, i.e. all initial candidates represent final candidates for consolidation. - If the current block is divided into exactly two blocks of any size, one of these two blocks is coded before the other, which is determined by the syntax. For the first coding block, the first candidate set is not reduced. However, for the second coding block, all candidate blocks that have the same associated parameters as the first block are removed from the candidate set (which includes the first coding block). - If a block is divided into 4 square blocks of equal size, the initial candidate list of the first 3 blocks (in coding order) is not reduced: all blocks in the initial candidate list are also present in the final candidate list. However, for the 4th (last) block in coding order, the following applies: - If a block in a different row (in the partitioning scheme as illustrated in the lower left of Figure 6) from the current block has the same associated parameters (e.g. motion parameters), then all candidates that already have the same motion parameters as a coding block in the same row as the current block are removed from the candidate set (which includes blocks in the same row). - If a block in a different column (in the partitioning scheme as illustrated in the lower left of Figure 6) from the current block has the same associated parameters (e.g. motion parameters), then all candidates that have the same motion parameters as an already coded block in the same column as the current block are removed from the candidate set (which includes blocks in the same column).

[0071] In a low-complexity variation of the embodiment (using the splitting pattern of FIG. 6), the reduction of the candidate list is done as follows. If the current block is not further split (top left pattern in Fig. 6), the initial candidate list is not reduced, i.e. all initial candidates represent final candidates for consolidation. - If the current block is split into exactly two blocks of any size, one of these two blocks is coded before the other, which is determined by the syntax. For the first coding block, the first candidate set is not reduced. However, for the second coding block, the first coding block of the splitting pattern is removed from the candidate set. - If a block is divided into 4 square blocks of equal size, the initial candidate list of the first 3 blocks (in coding order) is not reduced: all blocks in the initial candidate list are also present in the final candidate list. However, for the 4th (last) block in coding order, the following applies: - If the merged information signals that for a block in another row (than the current block) that will be coded later, it is merged into the first coded block of that row, the block in the same row as the current block is removed from the candidate set. - If the integration information signals that for a block in another column (than the current block) to be coded later, the block in the same column as the current block is integrated into the first coded block of that column, the block in the same column as the current block is removed from the candidate set.

[0072] In another preferred embodiment, the same splitting pattern as shown in Fig. 6 is supported, without the pattern of splitting a square block into two rectangular blocks of the same size. Except for the pattern of splitting a block into four square blocks, the reduction of the candidate list proceeds as described by any of the above embodiments. Here, all initial candidates are allowed for all sub-blocks, or only the candidate list of the last coded sub-block is constrained as follows: if three previously coded blocks are associated with the same parameters, all candidates associated with these parameters are removed from the candidate list. In a low-complexity version, the last coded sub-block cannot be merged with any of the three previously coded sub-blocks if these three sub-blocks are combined together.

[0073] In other preferred embodiments, different sets of splitting patterns for blocks (or other forms of sample sequence sets) are supported. For sample sequence sets that are not split, all candidates in the initial candidate list can be used for integration. If a sample sequence is split into exactly two sample sequences, for the sample sequence that is first in coding order, all candidates in the initial candidate set are inserted into the final candidate set. For the sample sequence that is second in coding order, all candidates that have the same associated parameters as the first sample sequence are removed. Or, in a low-complexity variation, only the first sample sequence is removed from the candidate set. For splitting patterns that split a sample sequence into two or more sample sequences, the removal of candidates depends on whether other splitting patterns can be simulated by the current splitting pattern and the corresponding integration information. The process of candidate removal follows the concept explicitly shown above, but takes into account the actual supported candidate patterns.

[0074] In another preferred embodiment, if SKIP / DIRECT mode is supported for a particular block, any merging candidate that is also the current candidate for SKIP / DIRECT mode is removed from the candidate list. This removal can replace the candidate block removal described above, or can be used in conjunction with the candidate block removal described above.

[0075] 3. Combination with SKIP / DIRECT mode SKIP / DIRECT modes can be supported for all or only once a particular block size and / or block shape. A set of candidate blocks is used for SKIP / DIRECT modes. The difference between SKIP and DIRECT is whether residual information is sent or not. The parameters (e.g. for prediction) of SKIP and DIRECT are assumed to be equal to one of the corresponding candidates. A candidate is selected by sending an index into the candidate list.

[0076] In a preferred embodiment, the candidate list for SKIP / DIRECT can contain different candidates. An example is shown in Figure 8. The candidate list may contain the following candidates (the current block is denoted by Xi): - Median (left, top, between corners) - Left Block (Li) - Above block (Ai) - Corner blocks (in order: top right (Ci1), bottom left (Ci2), top left (Ci3)) - Collocated blocks of a different but already coded image

[0077] In a preferred embodiment, the candidates for merging include Li (left block) and Ai (top block). Selecting these candidates for merging requires a small amount of side information to signal which block the current block is to be merged into.

[0078] The following symbols are used to describe the following embodiments: - set_mvp_ori is the set of candidates used for SKIP / DIRECT modes. This set consists of {Median, Left, Above, Corner, Collocated}, where Median is the median (the median in the ordered set of Left, Above and Corner) and collocated is scaled according to the temporal distance given by the closest reference frame. - set_mvp_comb is the set of candidates used for SKIP / DIRECT mode in combination with the block consolidation process.

[0079] For the preferred embodiment, the combination between SKIP / DIRECT and block merging modes can be processed by the original set of candidates. This means that the SKIP / DIRECT mode has the same set of candidates as when it works alone. The interest of combining the two modes comes from their complementarity in signaling auxiliary information between frames. Even though both of these modes currently use neighboring information to improve the signaling of blocks, block merging only processes the left and top neighbors, while the SKIP / DIRECT mode processes up to five candidates. The main complementarity lies in the different approaches to processing the neighboring information. The block merging process keeps for every reference list the entire set of information of its neighbors. This means that the block merging keeps for each reference list the entire auxiliary information from these neighbors as well as its motion vector, while the SKIP / DIRECT mode processes the prediction parameters separately for each reference list and sends an index to the candidate list for each reference list. That is, for a bi-predicted image, two indices are transmitted to signal candidates for reference list 0 and candidates for reference list 1.

[0080] In another preferred embodiment, a combined set of candidates called set_mvp_comb can be found for SKIP / DIRECT mode in combination with the block consolidation mode. This combined set is part of the original set (set_mvp_ori) and allows for a reduction in signaling for SKIP / DIRECT mode, as the list of candidates set_mvp_comb reduces. Candidates that have to be removed from the original list (set_mvp_ori) could be redundant or are used less frequently by the block consolidation process.

[0081] In another preferred embodiment, the combination between SKIP / DIRECT modes and block merging operations can be handled by a combined set of candidates (set_mvp_comb) that is the original set (set_mvp_ori) without the Median. Due to the low efficiency observed for the Median for SKIP / DIRECT modes, this reduction of the original list results in an improvement in coding efficiency.

[0082] In another preferred embodiment, the combination of SKIP / DIRECT modes and block consolidation can be processed by a combined set of candidates (set_mvp_comb) which is the original set (set_mvp_ori) that has only Corner and / or Collocated as candidates.

[0083] In another preferred embodiment, the combination of SKIP / DIRECT mode and block merging process can be handled by a combined set of candidates, set_mvp_ori, which has only Corner and Collocated as candidates. As already mentioned, despite the complementarity between SKIP / DIRECT mode and block merging, the candidates that need to be removed from the list are those that may overlap with the candidates of block merging process. These candidates are Left and Above. The combined set of candidates (set_mvp_comb) is reduced to only two candidates Corner and Collocated. SKIP / DIRECT mode using this candidate set set_mvp_comb combined with block merging process gives a high increase in efficiency of signaling auxiliary information between frames. In this embodiment, SKIP / DIRECT mode and merging mode do not share any candidate blocks.

[0084] In further embodiments, slightly different combinations of SKIP / DIRECT and merged modes can be used. It is possible to enable SKIP / DIRECT mode (e.g. for more candidates than merged mode) for certain blocks only (e.g. for blocks with a size larger than a specified size, or for square blocks only), and not support merged mode for these blocks. Or, SKIP / DIRECT mode can be turned off, and all candidates (including parameters indicating the combination of parameters for spatially / temporally adjacent blocks) are added as candidates to merged mode. This option was described in Fig. 1 to Fig. 5. An increased candidate set can be used for certain blocks only (e.g. for blocks with a size larger than a specified minimum size, or for square blocks only), while for other blocks a reduced candidate set is used. Or, as a further variation, merged mode is used with a reduced candidate set (e.g. only the top and left neighbors), and more candidates (e.g. top left neighbors, co-located blocks, etc.) are used for SKIP / DIRECT mode. Also, in this type of configuration, SKIP / DIRECT mode may be allowed only for certain blocks (e.g., blocks having a size larger than a specified minimum size, or only for square blocks), while integration mode is allowed for a larger set of blocks.

[0085] 4. Transmission of Integrated Information With respect to the preferred embodiment, and in particular with respect to the embodiment of figures 1 to 5, the following may apply: Assume that only two blocks, including the left and upper neighbor samples of the top left sample of the current block, are considered as candidates. If the final set of candidate blocks is not empty (after candidate removal as described above), one flag called merge_flag is signaled that specifies whether the current block is merged into any of the candidate blocks. If merge_flag is equal to 0 (for "false"), this block is not merged into one of the candidate blocks and all coding parameters are transmitted normally. If merge_flag is equal to 1 (for "true"), the following applies: If the set of candidate blocks contains only one block, this candidate block is used for merging. Otherwise, the set of candidate blocks contains exactly two blocks. If the prediction parameters of these two blocks are identical, these prediction parameters are used for the current block. Otherwise (if the two blocks have different prediction parameters), a flag called merge_left_flag is signaled. If merge_left_flag is equal to 1 (for "true"), a block containing the adjacent sample position to the left of the top-left sample position of the current block is selected from the set of candidate blocks. If merge_left_flag is equal to 0 (for "false"), the other (i.e., the adjacent above) block from the set of candidate blocks is selected. The prediction parameters of the selected block are used for the current block. In another embodiment, a combined syntax element is sent to signal the merging process. In another embodiment, merge_left_flag is sent regardless of whether two candidate blocks have the same prediction parameters.

[0086] It should be noted that the syntax element merge_left_flag can be named merge_index because its function is to index the selected one among the candidates that have not been eliminated.

[0087] In other preferred embodiments, more than one block may be included in the set of candidate blocks. The merge information (i.e. whether the block is merged and, if so, with which candidate block) is signaled by one or more syntax elements. In this, the set of coded words depends on the number of candidates in the final candidate set and is selected in the same way at the encoder and decoder. In one embodiment, the merge information is transmitted using one syntax element. In another embodiment, one syntax element specifies whether the block is merged with any of the candidate blocks (compare merge_flag above). This flag is transmitted only if the set of candidate blocks is not empty. A second syntax element signals which of the candidate blocks is used for the merge. It is transmitted only if the first syntax element signals that the current block is merged with one of the candidate blocks. In a preferred embodiment of the present invention, the second syntax element is transmitted only if the set of candidate blocks includes one or more candidate blocks and / or if any of the candidate blocks has prediction parameters different from the others of the candidate blocks. The syntax can depend on how many candidate blocks are given and / or how different prediction parameters are associated with the candidate blocks.

[0088] It is possible to add a set of candidates for block merging, as was done for the DIRECT mode.

[0089] As described in other preferred embodiments, the second syntax element, the unified index, can be transmitted only if the list of candidates contains one or more candidates. This requires extracting the list before parsing the unified index, which prevents these two processes from running in parallel. To allow increased parsing throughput and to make the parsing process more robust with respect to transmission errors, it is possible to remove this dependency by using a constant codeword and a constant number of candidates per index value. If this number is not reached by candidate selection, it is possible to extract auxiliary candidates to complete the list. These additional candidates can include so-called combined candidates, which are constructed from the motion parameters of possible different candidates already present in the list, and from the zero motion vector.

[0090] In another preferred embodiment, the syntax for signaling which of the candidate blocks is set is applied simultaneously at the encoder and decoder. For example, if three choices of blocks for synthesis are given, then those three choices are only present in the syntax and are considered for entropy coding. The probabilities for all other choices are considered to be 0, and the entropy codec is adjusted simultaneously at the encoder and decoder.

[0091] The prediction parameters determined as a result of the integration process may represent the entire set of prediction parameters associated with the block, or they may represent within a subset of these prediction parameters (e.g. prediction parameters for one hypothesis of a block where multiple hypothesis prediction is used).

[0092] In a preferred embodiment, the syntax elements related to the merge information are entropy coded using context modeling. The syntax elements may consist of the merge_flag and merge_left_flag described above.

[0093] In a preferred embodiment, one of three context models is used to code merge_flag. The used context model merge_flag_ctx is obtained as follows: If the set of candidate blocks contains two elements, the value of merge_flag_ctx is equal to the sum of the values ​​of merge_flag of the two candidate blocks. If the set of candidate blocks contains one element, the value of merge_flag_ctx is equal to twice the value of merge_flag of this one candidate block.

[0094] In a preferred embodiment, the merge_left_flag is coded using one probability model.

[0095] A different context model encoding for merge_idx(merge_left_flag) can be used.

[0096] In other embodiments, different context models may be used. Non-binary syntax elements may be mapped to a sequence of binary symbols (bins). The context model for some syntax elements or bins of syntax elements may be derived based on the number of already transmitted syntax elements of adjacent blocks or candidate blocks or other measures, while other syntax elements or bins of syntax elements may be coded with a fixed context model.

[0097] 5. Encoder operation The inclusion of the merging concept provides the encoder with more freedom for the generation of the bitstream, since the merging approach significantly increases the number of possibilities for selecting a partition for the sample array of the image, naturally with increased signaling overhead. Some or all of the additional patterns that can be exhibited by the combination of subdivision and merging (e.g., the patterns of FIG. 7 when the partitioning patterns of FIG. 6 are supported) can be additionally tested (using the corresponding block sizes for motion estimation and mode decision), and the best of the patterns provided by simply partitioning (FIG. 6) and by partitioning and merging (FIG. 7) can be selected based on a certain rate-distortion metric. In addition, for each block, it can be tested whether merging with any of the already coded candidate sets results in a reduction of a certain rate-distortion metric, and then the corresponding merging flag is set during the coding process.

[0098] In another preferred embodiment, the encoder can first determine the best subdivision of the sample sequence (as in the highest level coding scheme). Then, for each sample set, it can check whether merging with another sample set or another group of sample sets reduces a particular rate-distortion cost measure. In this case, prediction parameters associated with the merged group of sample sets can be re-estimated (e.g., by performing a new motion search), or prediction parameters already determined for the current sample set and the candidate sample set (or group of sample sets) for merging can be evaluated for the considered group of sample sets.

[0099] In another preferred embodiment, a specific rate-distortion cost measure can be evaluated for additional candidate groups of sample sets. As a specific example, when testing various possible splitting patterns (e.g., see FIG. 6), some or all of the patterns that can be indicated by the combination of splitting and merging (e.g., see FIG. 7) can be further tested. That is, for all of the patterns, a specific motion estimation and mode decision process is performed, and the pattern that follows the smallest rate-distortion measure is selected. This process can also be combined with the above low-complexity process, so that for the resulting blocks, it is further tested whether merging with already coded blocks (e.g., other than the patterns in FIG. 6 and FIG. 7) causes a decrease in the rate-distortion measure.

[0100] In another preferred embodiment, the encoder tests the various patterns that can be represented by splitting and merging in a priority order, so that it tests as many patterns as possible with certain real-time requirements. The priority order can also be changed based on the blocks already coded and the splitting pattern selected.

[0101] One way of transferring the above outlined embodiment to a specific syntax is described below with reference to the following figures. In particular, figures 9 to 11 show different parts of a syntax utilizing the above outlined embodiment. In particular, according to the below outlined embodiment, an image 20 is first up-divided into coding tree blocks whose image content is to be coded using the syntax coding_tree shown in figure 9. As shown there, for example for entropy_coding_mode_flag=1 for context-adaptive binary arithmetic coding or other specific entropy coding modes, the quadtree subdivision of the current coding tree block is signaled in the syntax part coding_tree by a flag called split_coding_unit_flag at symbol 400. As shown in figure 9, according to the below described embodiment, the tree root block is subdivided as signaled by split_coding_unit_flag in a depth-first traversal order as shown in figure 9a. Whenever a leaf node is reached, it indicates the coding unit that is to be coded immediately using the syntax function coding_unit. This can be seen from Fig. 9 when looking at the conditional clause in 402 which checks as to whether the current split_coding_unit_flag is set. If so, the function coding_tree is called recursively, leading to further transmission / extraction of further split_coding_unit_flag at the encoder and decoder, respectively. Otherwise, i.e., split_coding_unit_flag=0, the current sub-block of the tree root block 200 in Fig. 5a is a leaf block and the function coding_unit in Fig. 10 is called in 404 to code this coding unit.

[0102] In the currently described embodiment, the above mentioned options are used according to which merging is only available for images for which an inter prediction mode is available. That is, intra-coded slices / images do not use merging anyway. This can be seen from FIG. 10, where the flag merge_flag is sent at 406 only if the slice type is not equal to an intra image slice type. According to this embodiment, merging only concerns prediction parameters related to inter prediction. According to this embodiment, merge_flag is signaled for the whole coding unit 40 and also signals to the decoder a specific split mode for the current coding unit, i.e., non-split mode. Thus, the function prediction_unit is called at 408 for referring to the current coding unit as being a prediction unit. However, this is not the only possibility for switching to a merge option. Rather, if the merge_flag associated with the whole coding unit is not set in 406, the prediction type of the coding unit of the non-intra-picture slice is signaled in 410 by the syntax element pred_type depending on it, e.g. by calling the function prediction_unit in 412 for any partition of the current coding unit if the current coding unit is not further partitioned. In FIG. 10, only four different partitioning options are shown, but other partitioning options shown in FIG. 6 can be available as well. Another possibility is that the partitioning option PART_NxN is not available, but others. The association between the names from the partitioning modes used in FIG. 10 to the partitioning options shown in FIG. 6 is shown in FIG. 6 with each subscript under the individual partitioning option. The function prediction_unit is called for each partition, e.g. for each partition 50 and 60 in the coding order mentioned above. The function prediction_unit starts by checking the merge_flag in 414. If the merge_flag is set, the merge_index follows necessarily in 416.The check at step 414 is to check as to whether the merge_flag associated with the entire coding unit as signaled at 406 is set. If not, merge_flag is further signaled at 418, and if the latter is set, merge_index follows at 420 indicating a merging candidate for the current partition. Furthermore, merge_flag is only signaled for the current partition at 418 if the current prediction mode of the current coding unit is an inter prediction mode (see 422).

[0103] As can be seen from FIG. 11, the transmission of the prediction parameters used for the current prediction unit in 424 is performed according to the present embodiment only if no merging has been used for the current prediction unit.

[0104] The above description of the embodiment of Figures 9-11 already indicates most of the functionality and meaning, but some further information is given below.

[0105] merge_flag[x0][y0] specifies whether the inter prediction parameters for the current prediction unit (see 50 and 60 in the figure) are derived from the adjacent inter predicted partition. The array indexes x0, y0 specify the position (x0, y0) of the top left luma sample of the considered prediction block (see 50 and 60 in the figure) relative to the top left luma sample of the image (see 20 in the figure).

[0106] merge_idx[x0][y0] specifies the merge candidate index in the merge candidate list, where x0, y0 specifies the position (x0, y0) of the top-left luma sample of the considered prediction block relative to the top-left luma sample of the image.

[0107] Although not specifically shown in the above description of Figures 9 to 11, the merging candidate or list of merging candidates is not only determined in this embodiment using the coding parameters or prediction parameters of spatially adjacent prediction units / partitions as an example, but rather the list of candidates is also formed by using the prediction parameters of temporally adjacent partitions of temporally adjacent previously coded images. Furthermore, a combination of prediction parameters of spatially and / or temporally adjacent prediction units / partitions may be used and included in the list of merging candidates. Of course, only a subset thereof may be used. In particular, Figure 12 shows one possibility of determining spatial neighbor relationships, i.e. spatially adjacent partitions or prediction units. Figure 12 shows a prediction unit or partition 60 and pixels B0 to B2 and A0, A1 as an example. They are located directly adjacent to the boundary 500 of partition 60, i.e., B2 is diagonally adjacent to the top left pixel of partition 60, B1 is located vertically above and adjacent to the top right pixel of partition 60, B0 is diagonally located to the top right pixel of partition 60, A1 is located horizontally to the left and adjacent to the bottom left pixel of partition 60, and A0 is diagonally located to the bottom left pixel of partition 60. Partitions containing pixels B0-B2 and at least one of A0, A1 form a spatial neighborhood, and their prediction parameters form a merged candidate.

[0108] To perform the above mentioned elimination of those candidates that lead to other splitting modes that were also available, the following function can be used:

[0109] In particular, candidate N, i.e., the coding / prediction parameters resulting from the prediction unit / partition transforming pixel N=(B0, B1, B2, A0, A1), i.e., position (xN, yN), is removed from the candidate list if any of the following conditions are true (see Figure 6 for the partition mode PartMode and the corresponding partition index PartIdx, which indexes each partition within the coding unit):

[0110] - The PartMode of the current prediction unit is PART_2NxN, PartIdx is equal to 1, and the prediction units covering luma position (xP, yP-1) (PartIdx=0) and luma position (xN, yN) (Cand.N) have identical motion parameters. mvLX[xP,yP-1]==mvLX[xN,yN] refIdxLX[xP,yP-1]==refIdxLX[xN,yN] predFlagLX[xP,yP-1]==predFlagLX[xN,yN]

[0111] - The PartMode of the current prediction unit is PART_Nx2N, PartIdx is equal to 1, and the prediction units covering luma position (xP-1,yP) (PartIdx=0) and luma position (xN,yN) (Cand.N) have identical motion parameters. mvLX[xP-1,yP]==mvLX[xN,yN] refIdxLX[xP-1,yP]==refIdxLX[xN,yN] predFlagLX[xP-1,yP]==predFlagLX[xN,yN]

[0112] - The PartMode of the current prediction unit is PART_NxN, PartIdx is equal to 3, and the prediction units covering luma position (xP-1, yP) (PartIdx=2) and luma position (xP-1, yP-1) (PartIdx=0) have identical motion parameters. mvLX[xP-1,yP]==mvLX[xP-1,yP-1] refIdxLX[xP-1,yP]==refIdxLX[xP-1,yP-1] predFlagLX[xP-1,yP]==predFlagLX[xP-1,yP-1] And the prediction units covering the luma position (xP, yP-1) (PartIdx=1) and the luma position (xN, yN) (Cand.N) have the same motion parameters. mvLX[xP,yP-1]==mvLX[xN,yN] refIdxLX[xP,yP-1]==refIdxLX[xN,yN] predFlagLX[xP,yP-1]==predFlagLX[xN,yN]

[0113] - The PartMode of the current prediction unit is PART_NxN, PartIdx is equal to 3, and the prediction units covering luma position (xP, yP-1) (PartIdx=1) and luma position (xP-1, yP-1) (PartIdx=0) have identical motion parameters. mvLX[xP,yP-1]==mvLX[xP-1,yP-1] refIdxLX[xP,yP-1]==refIdxLX[xP-1,yP-1] predFlagLX[xP,yP-1]==predFlagLX[xP-1,yP-1] And the prediction units covering luma position (xP-1, yP) (PartIdx=2) and luma position (xN, yN) (Cand.N) have identical motion parameters. mvLX[xP-1,yP]==mvLX[xN,yN] refIdxLX[xP-1,yP]==refIdxLX[xN,yN]

[0114] In this regard, it should be noted that the position or location (xP, yP) means the maximum pixel of the current partition / prediction unit. That is, according to the first item, all coding parameter candidates are checked which are extracted by directly adopting the respective coding parameters of the adjacent prediction unit, i.e., prediction unit N. However, other additional coding parameter candidates can be checked as well for whether they are equal to the coding parameters of each prediction unit appearing for which other partitioning patterns supported by the syntax would be obtained. According to the embodiment just described, the equality of the coding parameters includes checking the equality of the motion vector, i.e., mvLX, the reference index, i.e., refIxLX, and the prediction flag predFlagLX indicating that the motion vector and reference index associated with the reference list X of its parameters, i.e., X being 0 or 1, are used for inter prediction.

[0115] It should also be noted that the aforementioned possibilities for removal of coding parameter candidates of adjacent prediction units / partitions are also applicable in case of supporting the asymmetric partitioning mode shown in the right half of Fig. 6. In that case, the mode PART_2NxN can indicate an all horizontal subdivision mode and PART_Nx2N can correspond to an all vertical subdivision mode. Furthermore, the mode PART_NxN can be excluded from the supported partitioning modes or partitioning patterns, in which case only the top two removal checks have to be performed.

[0116] With regard to the embodiments of Figures 9 to 12, it should also be noted that it is possible to exclude intra-predicted partitions from the list of candidates, i.e. their coding parameters are not naturally included in the list of candidates.

[0117] Furthermore, note that three contexts can be used for merge_flag and merge_index.

[0118] Although some aspects have been described in relation to an apparatus, it will be apparent that these aspects also refer to a description of a corresponding method, where a block or device corresponds to a method step or a function of a method step. Similarly, aspects described in relation to a method step also refer to a description of a corresponding block or item or function of a corresponding apparatus. Some or all of the method steps can be performed by (or using) a hardware apparatus, such as, for example, a microprocessor, a programmable computer, or an electronic circuit. In some embodiments, some or more of the most important method steps can be performed by such an apparatus.

[0119] Depending on specific implementation requirements, the embodiments of the present invention can be implemented in hardware or in software. The embodiments can be implemented using a digital storage medium, such as a floppy disk, DVD, Blue-Ray, CD, ROM, PROM, EPROM, EEPROM or FLASH memory, having electronically readable control signals stored thereon that cooperate (or can cooperate) with a programmable computer system so that the methods are executed. Thus, the digital storage medium can be computer readable.

[0120] Some embodiments according to the present invention include a data carrier having electronically readable control signals capable of cooperating with a programmable computing system to perform one of the methods described herein.

[0121] Typically, embodiments of the present invention can be implemented as a computer program product having program code which, when run on a computer, operates to perform one of the methods. The program code can for example be stored on a machine-readable carrier.

[0122] Other embodiments comprise the computer program for performing one of the methods described herein, stored on a machine readable carrier.

[0123] In other words, an embodiment of the inventive method is, therefore, a computer program having a program code for performing one of the methods described herein, when the computer program runs on a computer.

[0124] A further embodiment of the inventive method is therefore a data carrier (or digital storage medium or computer readable medium) comprising the computer program for performing one of the methods described herein, recorded thereon. The data carrier, digital storage medium or recording medium is generally tangible and / or non-transitory.

[0125] A further embodiment of the inventive method is, therefore, a data stream or a sequence of signals representing the computer program for performing one of the methods described herein, The data stream or the sequence of signals can for example be arranged to be transmitted via a data communication connection, for example via the Internet.

[0126] A further embodiment comprises a processing means, for example a computer, or a programmable logic device, configured to or adapted to perform one of the methods described herein.

[0127] A further embodiment comprises a computer having installed thereon the computer program for performing one of the methods described herein.

[0128] Further embodiments according to the present invention include a device or system configured to transfer (e.g. electronically or optically) a computer program for performing one of the methods described herein to a receiver. The receiver may be, for example, a computer, a mobile device, a storage device, etc. The device or system may, for example, include a file server for transferring the computer program to the receiver.

[0129] In some embodiments, a programmable logic device (e.g., a field programmable gate array) can be used to perform some or all of the functions of the methods described herein. In some embodiments, a field programmable gate array can cooperate with a microprocessor to perform one of the methods described herein. In general, the methods are preferably performed by any hardware apparatus.

[0130] The above-described embodiments are merely illustrative for the principles of the present invention. It is understood that modifications and variations of the devices and details described herein will be apparent to others skilled in the art. It is therefore the intention to be limited only by the scope of the impending claims and not by the specific details shown by the description and illustration of the embodiments herein.

Claims

1. 1. A decoder (50) configured to decode a bitstream (30) signaling one of supported partitioning patterns for a current block (40) of an image (20), the decoder comprising: if the signaled one of the supported partitioning patterns specifies a subdivision of the current block (40) into two or more further blocks (50, 60), For all further blocks other than a first one of said further blocks in coding order (70), From the set of coding parameter candidates for each further block (60), a decoder configured to eliminate coding parameter candidates having coding parameters that are the same as coding parameters associated with any of the further blocks that, when combined with each further block (60), would result in one of the supported partitioning patterns.

2. 2. The decoder of claim 1, wherein the decoder is configured to set a coding parameter associated with each further block in dependence on one of the non-removed coding parameter candidates if the number of non-removed coding parameter candidates is non-zero.

3. 2. A decoder according to claim 1, characterized in that the decoder (50) is arranged to set a coding parameter associated with each further block equal to one of the non-removed coding parameter candidates, with or without additional refinement and / or with or without scaling according to temporal distance, if the number of non-removed coding parameter candidates is non-zero.

4. 4. A decoder as claimed in claim 2 or claim 3, characterized in that the decoder is configured to support intra- and inter-prediction modes for the current block and to perform the merging and removal only if the current block (40) is coded in an inter-prediction mode.

5. A decoder according to any one of claims 2 to 4, characterised in that the coding parameters are prediction parameters, and the decoder is arranged to use the prediction parameters of each further block (60) to obtain a prediction signal for each further block (60).

6. 6. A decoder according to claim 1, characterized in that the decoder is configured to only expect that the bitstream (30) contains syntax elements specifying which of the non-removed coding parameter candidates are used for aggregation depending on the number of non-removed coding parameter candidates for each further block if the number of non-removed coding parameter candidates for each further block is greater than one.

7. A decoder according to any one of claims 1 to 6, characterized in that the decoder (50) is arranged to determine the set of coding parameter candidates for each further block (60) on the basis of coding parameters associated with a previously decoded block.

8. A decoder according to any one of claims 1 to 7, characterized in that the decoder (50) is arranged to determine the set of coding parameter candidates for each further block based, at least in part, on the coding parameters associated with previously decoded blocks adjacent to each further block and located outside and inside the current block, respectively.

9. A decoder according to any one of claims 1 to 8, characterized in that the decoder (50) is arranged to determine the set of coding parameter candidates for each further block from an initial set of previously decoded blocks other than those coded in an intra prediction mode.

10. A decoder according to any one of claims 1 to 9, characterized in that the decoder (50) is configured to subdivide the image (20) into coding blocks according to subdivision information contained in the bitstream (30), the coding blocks including the current block.

11. The decoder of claim 10, characterized in that the decoder (50) is configured to further subdivide the current block (40) into one or more transform blocks according to further subdivision information included in the bitstream, and to obtain a residual signal of the current block (40) from the bitstream (30) in units of the transform blocks.

12. A decoder according to any one of claims 1 to 11, characterized in that if the signalled one of the supported partitioning patterns specifies a subdivision of the block into two further blocks, the decoder is configured to exclude, for a second one of the further blocks in coding order, from the set of coding parameter candidates for the second further block, a coding parameter candidate having the same coding parameter as a coding parameter associated with the first one of the further blocks in coding order.

13. the supported partitioning patterns include a no-partition mode, a horizontal partitioning mode, and a vertical partitioning mode, and the decoder, if the signaled one of the supported partitioning patterns specifies a subdivision of the block into four further blocks: for a fourth one of the further blocks in coding order, from a set of coding parameter candidates for the fourth further block, excluding a coding parameter candidate having a coding parameter that is the same as a coding parameter associated with one of the further blocks in the same row as the fourth further block when two other of the further blocks in different rows have coding parameters associated therewith that are equal to each other, A decoder according to any of claims 1 to 12, characterized in that, with respect to the fourth block, it is arranged to exclude from the set of coding parameter candidates for the fourth further block a coding parameter candidate which has a coding parameter which is the same as a coding parameter associated with one of the further blocks in the same column as the fourth further block, when two other of the further blocks in different columns have coding parameters associated therewith which are equal to each other.

14. An encoder (10) configured to encode an image (20) into a bitstream (30), the encoder comprising: signaling in the bitstream (30) one of the supported partitioning patterns for the current block (40); if the signaled one of the supported partitioning patterns specifies a subdivision of the current block (40) into two or more further blocks (50, 60), For all further blocks (60) other than the first further block (50) of said further blocks in coding order (70), From the set of coding parameter candidates for each further block (60), an encoder configured to eliminate encoding parameter candidates having encoding parameters that are the same as encoding parameters associated with any of the further blocks that, when combined with each further block, would result in one of the supported partitioning patterns.

15. 1. A decoder configured to decode a bitstream signaling one of supported partitioning patterns for a current block of an image, the decoder comprising: if the signaled one of the supported partitioning patterns specifies a subdivision of the block into two or more further blocks, for all further blocks other than a first one of said further blocks in coding order, From the set of candidate blocks for each said further block:

20. A decoder configured to remove candidate blocks which, when combined with each further block, would result in one of the supported partitioning patterns.

16. 16. A decoder according to claim 15, characterized in that, if the number of non-removed candidate blocks is non-zero, the decoder is configured to merge each of the further blocks with one of the non-removed candidate blocks by setting an encoding parameter of the each further block in dependence on an encoding parameter associated with the one candidate block.

17. 16. A decoder according to claim 15, characterized in that, if the number of non-removed candidate blocks is non-zero, the decoder is configured to merge each further block with one of the non-removed candidate blocks by setting coding parameters of the each further block equal to coding parameters associated with the one candidate block, with or without additional refinement and / or scaling according to temporal distance.

18. 18. A decoder according to claim 16 or 17, characterized in that the decoder is configured to support intra- and inter-prediction modes for the current block and to perform the merging and removal only if the current block is coded in an inter-prediction mode.

19. A decoder according to any of claims 16 to 18, characterised in that the coding parameters are prediction parameters, and the decoder is arranged to use the prediction parameters for each further block to obtain a prediction signal for that block.

20. 20. A decoder according to any of claims 15 to 19, characterized in that the decoder is configured to only predict that the bitstream contains syntax elements specifying which of the non-removed candidate blocks are used for combining depending on the number of non-removed candidate blocks for each further block if the number of non-removed candidate blocks for each further block is greater than one.

21. A decoder according to any of claims 15 to 20, characterised in that the decoder is arranged to determine the set of candidate blocks for each further block from previously decoded blocks.

22. A decoder according to any one of claims 15 to 21, characterized in that the decoder is configured to determine the set of candidate blocks for each further block from an initial set of previously decoded blocks neighbouring each further block, the initial sets comprising, respectively, outer and inner neighbouring blocks of the current block.

23. A decoder according to any one of claims 15 to 22, characterized in that the decoder is arranged to determine the set of candidate blocks for each further block from an initial set of previously decoded blocks other than those coded in intra prediction mode.

24. A decoder according to any one of claims 15 to 23, characterized in that the decoder is configured to subdivide the image into coding blocks according to subdivision information contained in the bitstream, the coding blocks including the current block.

25. The decoder of claim 24, characterized in that the decoder is configured to further subdivide the current block into one or more transform blocks according to further subdivision information, and obtain a residual signal of the current block from a bitstream in units of the transform blocks.

26. 1. An encoder configured to encode an image (20) into a bitstream (30), the encoder comprising: signaling in the bitstream (30) one of the supported partitioning patterns for the current block (40); if the signaled one of the supported partitioning patterns specifies a subdivision of the block into two or more further blocks, for all further blocks other than a first one of said further blocks in coding order, From the set of candidate blocks for each said further block:

13. An encoder configured to eliminate candidate blocks which, when combined with each further block, would result in one of the supported partitioning patterns.

27. 1. A method for decoding a bitstream (30) signaling one of supported partitioning patterns for a current block (40) of an image (20), the method comprising: if the signaled one of the supported partitioning patterns specifies a subdivision of the current block (40) into two or more further blocks (50, 60), For all further blocks other than a first one of said further blocks in coding order (70), From the set of coding parameter candidates for each further block (60), The method includes a step of eliminating coding parameter candidates that have coding parameters that are the same as coding parameters associated with any of the further blocks that, when combined with each further block (60), would result in one of the supported partitioning patterns.

28. A method for encoding an image (20) into a bitstream (30), the method comprising the steps of: signaling in the bitstream (30) one of the supported partitioning patterns for the current block (40); if the signaled one of the supported partitioning patterns specifies a subdivision of the current block (40) into two or more further blocks (50, 60), For all further blocks (60) other than a first one of said further blocks in coding order (70), From the set of coding parameter candidates for each further block (60), A method comprising the step of eliminating coding parameter candidates having coding parameters that are the same as coding parameters associated with any of the further blocks that, when combined with each further block, would result in one of the supported partitioning patterns.

29. 29. Computer program having a program code for performing the method according to claim 27 or 28, when said computer program runs on a computer.