JP6952513B2 - Mode information encoder, mode information decoder, and program - Google Patents

Description

The present invention relates to a mode information encoding device, a mode information decoding device, and a program.

In the intra-slice processing of image coding and video coding, the pixel value sequence of the target area to be encoded is predicted based on the information in the already coded area in the screen. Then, the difference between the pixel value sequence of the target region obtained by the prediction and the actual pixel value sequence of the target region is taken and entropy-coded. As a result, the coding efficiency is improved by utilizing the tendency that the difference is statistically unevenly distributed to the value near 0.

For example, in MPEG-H HEVC / H.265 (hereinafter referred to as "HEVC"), a total of 35 types of in-screen prediction modes are available: direction prediction mode (33 types), DC prediction, and plane prediction. be. Of these, in the direction prediction mode, the prediction block is obtained by extrapolating the reference pixel value sequence in the vicinity of the coded block in the predetermined direction. Further, in the DC prediction, all the pixels in the prediction block are set as the average value of the reference pixel value sequence. Further, in the plane prediction, a prediction block is obtained by applying approximate bilinear interpolation to the reference pixel value sequence.

In HEVC, in-screen prediction is executed in block units called TU (Transform Unit). At this time, it is necessary to notify the in-screen prediction mode information applied in a certain TU from the encoder side to the decoder side. In HEVC, the TU adjacent to the left of the TU of interest (hereinafter referred to as "target TU") (if there are multiple TUs adjacent to the left, the top TU among them), and above the target TU. The in-screen prediction mode number of the target TU is encoded according to the in-screen prediction mode number of the TU adjacent to (if there are a plurality of TUs adjacent to the top, the leftmost TU). As a result, the entropy of the transmitted in-screen prediction mode is reduced by utilizing the spatial correlation of the in-screen prediction mode.

Further, Patent Document 1 discloses a method of predicting an in-screen prediction mode (direction of direction prediction) of a target TU according to a pattern of pixel values of a TU adjacent to the target TU.

Japanese Patent No. 5514130

The in-screen prediction mode coding method in HEVC utilizes the tendency that the in-screen prediction mode of the target TU is likely to match the in-screen prediction mode of the adjacent TU. At this time, the in-screen prediction mode of the TU adjacent to the left or above is applied to the default case classification rule, and three candidates for the in-screen prediction mode of the target TU are uniformly determined. Then, if the actual in-screen prediction mode of the target TU is included in these three candidates, the index that identifies one of the three candidates is notified, and if it is not included, the in-screen screen when counting excluding the three candidates. It works to notify the prediction mode number.

For example, the left adjacent block is the in-screen prediction mode 16 (direction prediction referring to the direction of the slope 21/32 (about 33.3 degrees) diagonally upward to the left), and the upper adjacent block is the in-screen prediction mode 18 (left diagonally upward). In the case of (direction prediction referring to the gradient 32/32 (45.0 degrees) direction), mode 16, mode 18, and mode 0 (plane prediction) are set as the three candidates in HEVC. However, in this case, there is a high probability that mode 17 (direction prediction that refers to the direction of the gradient 26/32 (about 39.1 degrees) diagonally upward to the left) that refers to the intermediate direction between modes 16 and 18 will appear. Despite the possibility, mode 17 is not included in the above three candidates. That is, the method of notifying the mode information may be inefficient.

Further, in HEVC, if the in-screen prediction mode of the target TU is included in the above three candidates, a significant reduction in the number of bits can be expected. On the other hand, the number of bits for those leaked from the three candidates increases relatively, and a loss is unavoidable.

In the method of Patent Document 1, candidates for the intra prediction mode of the target TU are determined according to the spatial frequency of the pixel value pattern of the adjacent TU. In this method, an appropriate intra-prediction mode candidate can be presented for an image region having a regular pattern. However, when the regularity of the image pattern is poor or the image pattern is flat, even if the intra prediction mode has a strong statistical correlation between blocks, the amount of information is reduced in consideration of the correlation. There is also the problem that it does not contribute to.

Further, in both the methods of HEVC and Patent Document 1, the method of narrowing down the intra-prediction mode candidates of the target TU is uniform. That is, the mode candidates are not changed according to the image to be encoded. That is, there is still room for improving the efficiency in coding the mode information because the mode candidates are not generated according to the characteristics of the video every moment.

The present invention has been made in consideration of the above circumstances, and mode information for efficiently encoding and decoding the mode information of the coded target block when encoding an image / video. It is intended to provide a coding device, a mode information decoding device, and a program.

[1] In order to solve the above problem, the mode information coding apparatus according to one aspect of the present invention includes the mode information of the learning target area in the coded area, which is the coded area, and the coded area. A frequency information generator that generates frequency information indicating the frequency of appearance of a set with the mode information of the learning reference area corresponding to the learning target area in the area, and a reference corresponding to the target area that is the coding target area. The ordering unit obtains an ordering unit for ordering the mode information of the target area and the mode information of the target area based on at least a part of the appearance frequency held by the frequency information regarding the mode information of the area. It is characterized by comprising a coding unit for coding as a code based on at least a part of the order.

[2] Further, in one aspect of the present invention, in the mode information coding apparatus, the ordering unit orders all the appearance frequencies held by the frequency information with respect to the mode information of the reference region. The mode information of the target area is ordered, and the coding unit encodes the mode information of the target area as a code based on all the orders obtained by the ordering unit.

[3] Further, the mode information decoding device according to one aspect of the present invention corresponds to the mode information of the learning target area in the already decoded area, which is the decoded area, and the learning target area in the already decoded area. The frequency information holds the frequency information generation unit that generates frequency information indicating the frequency of appearance of a set with the mode information of the learning reference area to be used, and the mode information of the reference area corresponding to the target area that is the decoding target area. Select from the ordering unit that orders the mode information of the target area based on at least a part of the appearance frequency, and the mode information of the target area that is ordered based on the order corresponding to the input code value. It is characterized by including a decoding unit that decodes specific mode information.

[4] Further, in one aspect of the present invention, in the mode information decoding apparatus, the ordering unit orders all the appearance frequencies held by the frequency information with respect to the mode information of the reference region. It is characterized in that the mode information of the target area is ordered.

[5] Further, in one aspect of the present invention, the computer is divided into the mode information of the learning target area in the coded area, which is the coded area, and the learning target area in the coded area. The frequency information is related to the frequency information generation unit that generates frequency information indicating the frequency of appearance of the pair with the mode information of the corresponding learning reference area, and the mode information of the reference area corresponding to the target area that is the area to be encoded. An ordering unit that orders the mode information of the target area based on at least a part of the frequency of occurrence to be held, and the mode information of the target area are encoded as codes based on the order obtained by the ordering unit. It is a program for functioning as a mode information coding device including a coding unit.

[6] Further, in one aspect of the present invention, the computer is subjected to learning corresponding to the mode information of the learning target area in the already decoded area, which is the decoded area, and the learning target area in the already decoded area. Frequency information generation unit that generates frequency information indicating the frequency of appearance of a set with the mode information of the reference area, and the appearance frequency held by the frequency information regarding the mode information of the reference area corresponding to the target area that is the decoding target area. An ordering unit that orders the mode information of the target area based on at least a part of the above, and a specification selected from the mode information of the target area that is ordered based on the order corresponding to the input code value. It is a program for functioning as a mode information decoding device including a decoding unit for decoding the mode information of the above.

According to the present invention, the mode information coding apparatus can improve the coding efficiency of mode information. Further, the mode information decoding device can decode the mode information encoded by the mode information coding device. Further, since the mode information decoding device can perform decoding without receiving frequency information (histogram data) from the outside (for example, from the mode information coding device), an overhead of information transmission that causes a decrease in coding efficiency occurs. No.

A block diagram showing a schematic functional configuration of a mode information coding device and a mode information decoding device according to an embodiment of the present invention, and information handled by the mode information coding device and the mode information decoding device (mode information of the coding block) are shown. It is a schematic diagram which shows. It is the schematic for demonstrating the main point of processing of the histogram totaling part 10 by the same embodiment. It is the schematic for demonstrating the main point of processing of the sort part 11 by the same embodiment. It is a schematic diagram for demonstrating the main point of processing of the coding unit 12 by the same embodiment. It is the schematic for demonstrating the main point of processing of the decoding unit 42 by the same embodiment.

[Embodiment]
Next, an embodiment of the present invention will be described with reference to the drawings.
FIG. 1 is a block diagram for explaining a schematic functional configuration of a mode information coding device and a mode information decoding device according to the present embodiment. In addition, FIG. 1 also shows information (mode information of the coding block) handled by the mode information coding device and the mode information decoding device according to the present embodiment.

The mode information coding device 1 encodes the mode value of the target block while referring to the mode value of one or more reference blocks in the vicinity of the target block. The mode information coding apparatus 1 according to the present embodiment determines the relationship between the mode value of the reference block and the mode value of the target block based on the mode information already encoded before the time when the target block is encoded. learn. The mode information already encoded before the time when the target block is encoded is, for example, the mode information of the block included in the frame before the previous frame in the coding order. Alternatively, in the mode information of a block that is contained in the same frame as the target block and is already encoded (usually a block that is above the target block or to the left of the target block on the same line as the target block). be. The mode information coding device 1 performs coding different from that of HEVC in that it realizes conversion of mode values adapted to the scene based on such learning.

As shown in FIG. 1, the mode information coding device 1 includes a histogram totaling unit 10 (frequency information generation unit), a sorting unit 11 (ordering unit), and a coding unit 12.
The histogram aggregation unit 10 includes mode information of the learning target area in the coded area, which is a coded area, and mode information of the learning reference area corresponding to the learning target area in the coded area. Frequency information (diaphragm data) indicating the frequency of appearance of the pair with is aggregated and generated.
The sort unit 11 obtains the mode information of the target area based on at least a part of the appearance frequency held by the frequency information (histogram data) with respect to the mode information of the reference area corresponding to the target area which is the area to be encoded. Order. Further, the sort unit 11 according to the present embodiment orders the mode information of the target area by ordering all the appearance frequencies held by the frequency information with respect to the mode information of the reference area by the sort process.
The coding unit 12 encodes the mode information of the target area as a code based on at least a part of the order obtained by the sorting unit 11. Further, the coding unit 12 according to the present embodiment encodes the mode information of the target area as a code based on all the orders obtained by the sorting unit 11.

Further, the entropy coding unit 2 entropy-codes the code value of the mode information output by the mode information coding device 1.
Further, the entropy decoding unit 3 decodes the code encoded by the entropy coding unit 2 into the code output by the mode information coding device 1.

Further, the mode information decoding device 4 includes a histogram aggregation unit 40 (frequency information generation unit), a sort unit 41 (ordering unit), and a decoding unit 42.
The histogram aggregation unit 40 is a set of the mode information of the learning target area in the already decoded area, which is the decoded area, and the mode information of the learning reference area corresponding to the learning target area in the already decoded area. Generates frequency information (histogram data) that represents the frequency of appearance of.
The sort unit 41 orders the mode information of the target area based on at least a part of the appearance frequency held by the frequency information with respect to the mode information of the reference area corresponding to the target area which is the area to be decoded. Further, the sort unit 41 of the present embodiment orders the mode information of the target area by ordering all the appearance frequencies held by the frequency information with respect to the mode information of the reference area by the sort process.
The decoding unit 42 decodes specific mode information selected from the ordered mode information of the target area based on the order corresponding to the input code value.

The functions of the respective parts constituting the mode information coding device 1 and the mode information decoding device 4 and the functions of the entropy coding unit 2 and the entropy decoding unit 3 are configured by using, for example, an electronic circuit. Further, in the function of each part, a storage means such as a magnetic hard disk device or a semiconductor memory is used as necessary to store information. Further, the functions of each of these parts may be realized by a computer and a program.

The mode information coding device 1 and the mode information decoding device 4 are devices that encode and decode mode information, respectively. Information other than the mode information (for example, information representing the content of the image or the video itself) is separately encoded, transmitted, recorded on a recording medium, or the like, and then decoded, if necessary.

Further, in the figure, 21 is a target area. The target area 21 includes one block (target block). Reference numeral 6 is a reference area. In the illustrated example, the reference region 6 includes two blocks (reference blocks). In this embodiment, two reference blocks correspond to one target block, and these reference blocks are adjacent to the left and above the target block. However, the number of reference blocks may be other than 2. And x is the mode (target mode) of the target block. Further, y ₁ and y ₂ are modes (reference modes) of two reference blocks, respectively. Reference numeral 5 denotes an area in which the learning target area and the learning reference area are combined. Two learning reference blocks (adjacent to the left and above the learning target block) correspond to the nth learning target block. The mode of the nth learning target block is x ⁽ⁿ⁾ . The modes of the two learning reference blocks corresponding to the learning target block are y ₁ ⁽ⁿ⁾ and y ₂ ⁽ⁿ⁾ , respectively.

Reference numeral 22 denotes a target area on the mode information decoding device 4 side. Reference numeral 8 is a reference area. Reference numeral 7 denotes an area in which the learning target area and the learning reference area are combined. The target area 22 corresponds to the above-mentioned target area 21. The reference area 8 corresponds to the above-mentioned reference area 6. The learning target area and the learning reference area 7 correspond to the learning target area and the learning reference area 5. _{Then, x, y 1} , y ₂ , x ⁽ⁿ⁾ , y ₁ ⁽ⁿ⁾ , y ₂ ⁽ⁿ⁾ (where n = 1, 2, ...) On the mode information decoding device 4 side are the mode information. It is the same as those on the encoding device 1 side, and description thereof will be omitted here.

In generalization, the mode value of the k-th (k is an integer greater than or equal to 1 and less than or equal to K. K is the number of reference blocks corresponding to one target block and is an integer of 1 or more) is referred to as the reference mode. Called y _k. Further, the mode value of the target block is called the target mode x. That is, the number K of reference blocks is 1 or more for one target block. Further, in both the target block and the reference block, the range of the mode value is 1 or more and M or less (M is an integer of 2 or more).

The relative position of each reference block with respect to the position of the target block has a common relative positional relationship between the mode information coding device 1 and the mode information decoding device 4. For example, in both the mode information coding device 1 and the mode information decoding device 4, the first reference block is adjacent to the left of the target block, and the second reference block is adjacent to the top of the target block.

Further, the block size may be different between the target block and the reference block. In that case, for example, the upper right pixel of the first reference block (adjacent to the left of the target block) is located to the left of one pixel of the upper left pixel of the target block. Further, in that case, for example, the lower left pixel of the second reference block is located one pixel above the upper left pixel of the target block.

Further, the number of learning reference blocks per learning target block is the same as the number of reference blocks per target block. Further, the relative positional relationship between the learning target block and the learning reference block is the same as the relative positional relationship between the target block and the reference block. That is, as shown by reference numerals 5 and 6 in FIG. Both the learning target block and the learning reference block are partial regions of the video region (or the video partial region (for example, only the video region to which the intra prediction is applied)) for which the mode information has already been encoded. It is set within (that is, the area shown by hatching in FIG. 1; hereinafter, referred to as “already coded area”). The set of the learning target block and the learning reference block is N sets (N is an integer of 1 or more. Even if N is a constant, N is a variable (for example, N is each time the block to be processed advances). It does not matter if it increases, etc.). The set of the learning target block and the learning reference block is used for the learning process in the histogram aggregation unit 10 described later.

Next, the processing of the histogram totaling unit 10 will be described in detail.
FIG. 2 is a schematic diagram for explaining the main points of processing of the histogram totaling unit 10. FIG. (A) shows the positional relationship between the learning target area and the learning reference area referred to by the histogram aggregation unit 10 for counting the frequency value, and the target area to be encoded and the associated reference area. It is a schematic diagram. FIG. (B) is ^{a schematic view showing the histogram H (n-1)} (histogram 50) at the time point (n-1) as a three-dimensional matrix. FIG. (C) is ^{a schematic diagram showing the histogram H (n)} (histogram 53) at the time point (n) as a three-dimensional matrix. That is, time is advancing while shifting from the state shown in FIG. (B) to the state shown in FIG.

Reference numeral 51 in FIG. 5A is a learning target mode value and a learning reference mode value. Reference numeral 52 in FIG. 5B is one frequency value included in the histogram 50, and is H ^(n-1) (1, 2, 3). That is, it is the frequency value of the set of _{x = 1, y 1} = 2, y ₂ = 3 at the time point (n-1). Reference numeral 54 in FIG. 5C is one frequency value included in the histogram 53, and is H ⁽ⁿ⁾ (1, 2, 3). That is, it is the frequency value of the set of x = 1, y ₁ = 2, y _{2 = 3 at the time point (n).}

The histogram aggregation unit 10 includes a memory for counting the frequency distribution of a set of a learning reference mode value and a learning target mode value. The histogram aggregation unit 10 uses this memory to sequentially count the appearance of a set of a learning reference mode value and a learning target mode value.

Specifically, the histogram aggregation unit 10 is a frequency value of the ^{learning target mode X (n)} and the learning reference mode Y ₁ ⁽ⁿ⁾ , Y ₂ ⁽ⁿ⁾ , ..., Y _K ^(n). H ⁽ⁿ⁾ (X, Y ₁ , Y ₂ , ..., Y _K ) is counted. Here, (n) in the upper right indicates an index value. The index value n is an integer. When n ≧ 1, H ⁽ⁿ⁾ represents a histogram constructed by using a total of n sets of mode information. When n = 0, H ⁽ⁿ⁾ represents an initial value when constructing a histogram.

As this initial value (when n = 0), a predetermined value is appropriately given. When n = 0, for example, for all (X, Y ₁ , Y ₂ , ..., Y _K ) pairs, H ⁽ⁿ⁾ (X, Y ₁ , Y ₂ , ..., Y _K ) = 0 may be set. Alternatively, when n = 0, an appropriate value other than 0 may be given as the frequency value of each set of _{(X, Y 1} , Y ₂ , ..., Y _K). For example, a typical frequency value distribution measured in advance using various images may be given as a value of H ⁽ⁿ⁾ (X, Y ₁ , Y ₂ , ..., Y _K ). good. When a non-zero histogram is given as an initial value, it is possible to reduce the time required for learning the mode information (in other words, the amount of training data).

The histogram aggregation unit 10 generates the ^{histogram H (n)} by using one or more sets of the learning target mode and the reference mode in the coded region.
For example, when the mode information coding device 1 performs the coding process of the target mode at a certain time point, the histogram aggregation unit 10 sets the target block for learning at the positions of the target block and the reference block at the processing time point one block before. And put a reference block for learning. Then, the histogram totaling unit 10 updates the ^{histogram H (n)} based on the information of the set of the learning target block and the learning reference block. Then, as the coding process of the target mode progresses, the learning target block and the learning reference block are also moved, and the histogram H ⁽ⁿ⁾ is sequentially updated.

The histogram aggregation unit 10 is based on the set of the learning target mode x ⁽ⁿ⁾ and the learning reference mode (y ₁ ⁽ⁿ⁾ , y ₂ ⁽ⁿ⁾ , ..., Y _K ⁽ⁿ⁾ ) below. The histogram is updated from H ^{(n -1)} to H (n) by the equation (1). In the ^{upper right (n) of each of x (n)} , y ₁ ⁽ⁿ⁾ , y ₂ ⁽ⁿ⁾ , ..., Y _K (n), their mode information is the nth (n ≧ 1). ) Mode information (learning target mode and learning reference mode, respectively).

The histogram value is not updated for a set of modes other than ^{x (n)} , y ₁ ⁽ⁿ⁾ , y ₂ ⁽ⁿ⁾ , ..., Y _K ^(n).

In the specific example shown in FIG. 2, the operation of the histogram totaling unit 10 is as follows. As shown, the histogram ^{H (n-1)} and ^{H (n),} respectively, mode information _x, expressed as a three-dimensional matrix according to y 1, _{y 2.} Generally, a histogram of mode information, which is a set of one learning target mode and K learning reference mode, is represented as a (K + 1) dimensional matrix. Each element of the matrix is a frequency value of a set of learning target modes and learning reference modes K, respectively.

In the figure, reference numeral 50 indicates a histogram H ^(n-1) at the time point (n-1). Reference numeral 53 indicates a histogram H ⁽ⁿ⁾ at the time point (n). Each square (matrix element) in the figure represents a histogram bin, and the numbers in the square represent the frequency value. For example, in the region indicated by reference numeral 51 in FIG. 2A, the learning target mode x ⁽ⁿ⁾ is 1, and the learning reference modes y ₁ ⁽ⁿ⁾ and y ₂ ⁽ⁿ⁾ are as shown in. 2 and 3, respectively. The frequency value recorded in the bin of the target mode and the reference mode at the time point (n-1) was “2” as shown by reference numeral 52 in FIG. 2 (B). Therefore, the frequency values recorded in the bins of the target mode and the reference mode at the time point (n) are updated to "3" as indicated by reference numeral 54 in FIG. 2C. The frequency values other than the bin do not change between the histogram time point (n-1) and the time point (n).

Next, the processing of the sort unit 11 will be described in detail.
FIG. 3 is a schematic view for explaining the main points of the processing of the sort unit 11. FIG. 3A is a schematic diagram showing a selection condition (condition according to the mode value of the reference block) of the frequency value that the sort unit 11 intends to sort. FIG. (B) is an outline showing a set of frequency values ^{selected from the histogram H (n)} at the time point (n) according to the selection condition (the value of the reference block is y ₁ = 3 and y _{2 = 4).} It is a figure. FIG. 3C is a schematic diagram showing a result of sorting the frequency values selected in FIG. 3B (frequency value sorting result 58) by a mathematical formula. _{FIG. 3D is a schematic diagram showing a mode value X d} (target mode value 60) associated with a rank d determined based on a sort result.

The sort unit 11 performs the following processing on the premise that the histogram aggregation unit 10 has already aggregated the histogram. That is, the sort unit 11 has a histogram H ⁽ⁿ _{) based on the reference mode value sequence (y 1} , y ₂ , ..., Y _K ) of the reference area associated with the target area to be encoded at that time. ⁾ Sort some of the frequency values included in (X, Y ₁ , Y ₂ , ..., Y _K). Specifically, the sort unit 11 has _{Y 1} = y ₁ , Y ₂ = y ₂ among all the bins included ^{in the histogram H (n)} (X, Y ₁ , Y ₂ , ..., Y _K). , ..., Sort the bins _{with Y K} = y _K in descending order of frequency value. The sort unit 11 orders the target mode value X by this sort process. That is, the sort unit 11 compares the frequency values H ⁽ⁿ⁾ (X, y ₁ , y ₂ , ..., Y _K ) for all X ∈ {1, 2, ..., M}. Xs are arranged in order from the one with the highest frequency value.

The sort process by the sort unit 11 is expressed by a mathematical formula as shown in the following formula (2).

That is, the sort unit 11 obtains a _{sequence (X m} ) _{m = 1, 2, ..., M} satisfying the equation (2) by the sort process. However, this sequence does not contain terms with the same value. Also, for all _i ∈ {1, 2, ..., M}, X i ∈ {1, 2, ..., M}.

It should be noted that there may be a plurality of Xs having the same ^{H (n)} (X, Y ₁ , Y ₂ , ..., Y _K). That is, in the inequality sign “≦” in the equation (2), there may be one or more places where the equal sign “=” holds. In such a case, as a result of the sorting process, the sequence (X _m ) _{m = 1, 2, ..., M} is not uniquely determined. In that case, the sort unit 11 determines the order of the sort results according to, for example, a predetermined rule or the like. An example of such a rule is a rule such as "among the mode values having the same rank in terms of frequency, the smaller (or larger) mode value is regarded as having the higher frequency". Of course, the sort result may be determined according to other rules or the like.

That is, the sort unit 11 obtains a _{sequence (X m} ) _{m = 1, 2, ..., M} that satisfies the equations (2) and (3).

In the example shown in FIG. 3A, the mode information (target mode value) of the block (target block) included in the target area 55 for which the mode information is to be encoded at the present time is x. Further, the mode information (reference mode value) of the two blocks (reference block) included in the reference area 56 associated with the target block is y ₁ = 3 and y ₂ = 4, respectively. The frequency value of each target mode value x with respect to this set of reference mode values is the projection 57 in the ^{histogram H (n) shown in FIG. 3 (B).} That is, the frequency values corresponding to x = 0, 1, 2, 3, and 4, respectively, are 1, 4, 3, 1, and 5. As a result of the sorting unit 11 sorting these frequency values in descending order, the sorting result 58 of the frequency values shown in FIG. 3C is obtained. That is, the frequency values included in the sort result 58 are as follows.
H ⁽ⁿ⁾ (4,3,4) = 5
H ⁽ⁿ⁾ (1,3,4) = 4
H ⁽ⁿ⁾ (2,3,4) = 3
H ⁽ⁿ⁾ (0,3,4) = 1
H ⁽ⁿ⁾ (3,3,4) = 1

That is, the mutual relationship between these frequency values is as follows.
H ⁽ⁿ⁾ (4,3,4)> H ⁽ⁿ⁾ (1,3,4)> H ⁽ⁿ⁾ (2,3,4)> H ⁽ⁿ⁾ (0,3,4) = H ^{( n)} (3,3,4)

As indicated by the equal sign 59 representing bins having the same frequency value, the frequency values in the case of x = 0 and the case of x = 3 are 1 respectively, which are the same values. Here, when the above equation (3) is applied, _{the value of the target mode X d} of the rank d (1 ≦ d ≦ 5) is as the target mode value 60 shown in FIG. 3 (D). That is, it is uniquely determined as follows.
x ₁ = 4, x ₂ = 1, x ₃ = 2, x ₄ = 0, x ₅ = 3

Next, the processing of the coding unit 12 will be described in detail.
FIG. 4 is a schematic diagram for explaining the main points of the processing of the coding unit 12. _{The figure shows an example of a series of processes for obtaining the code value c based on the numerical sequence X d} (1 ≦ d ≦ 5) obtained by the sort unit 11 and the mode value of the target block and the mode value of the reference block. Is shown.

The coding unit 12 encodes the target mode x based on _{the sequence (X m} ) _{m = 1, 2, ..., M} obtained by the sorting unit 11 and outputs the code c. Therefore, in the coding unit 12, the frequency value of the target mode x is H ⁽ⁿ⁾ (X = 1, y ₁ , y ₂ , ..., Y _K ), H ⁽ⁿ⁾ (X = 2, y _1). , Y ₂ , ..., y _K ), ..., H ⁽ⁿ⁾ (X = M, y ₁ , y ₂ , ..., y _K ) .. That is, the coding unit 12 obtains the rank of the target mode x in the frequency value. Then, the coding unit 12 determines the code c based on this order.

For example, the coding unit 12 determines the code c based on the order d that satisfies the following equation (4).

As an example, the coding unit 12 determines the reference numeral c by the following equation (5).

In the equation (5), the value obtained by subtracting 1 from d is obtained because the value of the rank d is offset so that the sign value c starts from 0.

In the example shown in FIG. 4, the mode value of the block (target block) included in the target area 61 is 4. Further, the point that the reference mode values are y ₁ = 3 and y ₂ = 4, respectively, is the same as the situation in FIG. The target mode value 62 for each rank obtained by the sort unit 11 is as shown in the figure. The coding unit 12 obtains the order of the target mode value “4” in light of the target mode value 62. That is, the coding unit 12 obtains d = 1 as the order. Then, when the coding represented by the equation (5) is performed, the coding unit 12 obtains the code value c = 0 in correspondence with d = 1.

When the above equation (5) is applied, the code value c is always a non-negative integer. Further, the smaller the code value c is closer to 0, the higher the probability of occurrence of the mode tends to be. That is, by using the mode information coding device 1, the tendency that the occurrence probability distribution of the code value c becomes a small value closer to 0 is maintained as the probability of occurrence of the target mode increases. This is because the correlation between the reference mode value and the target mode value is maintained between the image area in which the histogram totaling unit 10 aggregates the histogram and the image area in which the sorting unit 11 and the coding unit 12 perform each processing. It uses the premise that it is. This makes it possible to reduce the entropy of the code value c.

The entropy coding unit 2 shown in FIG. 1 entropy-codes the code value c output by the coding unit 12. This makes it possible to further compress the mode information. In other words, it is possible to reduce the number of bits expressing the mode information transmitted from the entropy coding unit 2 to the entropy decoding unit 3.

As the coding method used by the entropy coding unit 2, for example, Huffman code, arithmetic code, CAVLC (context-adaptive variable-length coding method) and CABAC (context-adaptive arithmetic coding method), which are derivatives of these, are used. Can be used. In addition, when the probability distribution of the code value c is set to be high near 0 as described above, the consistency with the Golomb coding and its derivative, the exponential-Golomb coding, is also high. , The entropy coding unit 2 may use these.
The entropy decoding unit 3 decodes the encoded mode information by using a method of pairing with the entropy coding unit 2.

When the processing by the entropy coding unit 2 and the processing by the entropy decoding unit 3 are sequentially applied, it is desired that there is no loss of information (lossless).

Next, the processing of the histogram totaling unit 40, the sorting unit 41, and the decoding unit 42 will be described in detail.
FIG. 5 is a schematic diagram for explaining the main points of processing of the decoding unit 42. The figure shows an example of a series of processes for obtaining the mode value (x = 4) of the target block based on _{the numerical string X d} (1 ≦ d ≦ 5) obtained by the sort unit 41 and the code value c. Shown.

The histogram aggregation unit 40 uses one or more sets of mode values of the learning target area (learning target mode) and mode values of the learning reference area (learning reference mode) in the already decoded area, and uses the histogram H ^{(n). )} Is generated. The already decoded area is an area shown by hatching in FIG. 1, including a learning target area and a learning reference area 7. The positions of the learning target area and the learning reference area are commonly set between the histogram totaling unit 10 and the histogram totaling unit 40. That is, the histogram aggregation unit 40 learns in the learning target area and the learning reference area 5 in the already coded region, which is the target for which the histogram aggregation unit 10 obtains the frequency value, and in the already decoded region in the same frame and at the same position. ^{A histogram H (n)} is constructed using the target area for use and the reference area for learning 7.
Since the calculation method in which the histogram totaling unit 40 ^{generates the histogram H (n)} is the same as the calculation method in which the histogram totaling unit 10 ^{generates the histogram H (n)} , the description thereof will be omitted here.

The sort unit 41 orders the target mode values X based on _{the reference mode value strings (y 1} , y ₂ , ..., Y _K ) of the reference area associated with the target area to be decoded. Specifically, the sort unit 41 has _{Y 1} = y ₁ , among all the bins of ^{the histogram H (n)} (X, Y ₁ , Y ₂ , ..., Y _{K) generated by the histogram aggregation unit 40.} For Y ₂ = y ₂ , ..., Y _K = y _K bins, the frequency values are sorted in descending order. Then, the sort unit 41 orders the target mode values X according to the sort result. The sort unit 41 obtains the _{sequence (X m} ) _{m = 1, 2, ..., M} by the same arithmetic method as the processing by the sort unit 11.

The decoding unit 42 decodes the received code c into the target mode value x based on _{the sequence (X m} ) _{m = 1, 2, ..., M} obtained by the sort unit 41, and outputs the target mode value x. do.
Specifically, the rank d is obtained by a method corresponding to the coding unit 12 in the mode information coding device 1. When the coding unit 12 obtains the code c by the equation (5), the decoding unit 42 obtains the rank d by the following equation (6).

When the rank d is obtained, the decoding unit 42 subsequently obtains the target mode value x by the following equation (7) expressing the same relationship as the equation (4).

A series of processing flows of the histogram totaling unit 40, the sorting unit 41, and the decoding unit 42 described above will be described with reference to the example shown in FIG.
In this example, the code value c received by the mode information decoding device 4 is 0. This code value is obtained by the coding unit 12 in the mode information coding device 1. Since c = 0, the decoding unit 42 obtains d = 1 according to the equation (6). On the other hand, the sort unit 41 outputs a sequence of target modes arranged in descending order of frequency value. In the illustrated example, X ₁ = 4, X ₂ = 1, X ₃ = 2, X ₄ = 0, X ₅ = 4. This sequence is based on the assumption that _{y 1} = 3, y ₂ = 4 as the reference mode value. _{Then, the decoding unit 42 obtains X 1} = 4, which is the target mode value when d = 1, in light of the sequence of numbers output from the sort unit 41. That is, the target mode value x = 4 is obtained as the decoding result.

The functions and processing examples of the mode information coding device 1, the entropy coding unit 2, the entropy decoding unit 3, the mode information decoding device 4, and the like have been described above. The histogram aggregation unit 10 on the mode information coding device 1 side and the histogram aggregation unit 40 on the mode information decoding device 4 side update the histogram data while synchronizing with each other by some means. In other words, the series of processing by the sorting unit 11 and the coding unit 12 and the series of processing by the sorting unit 41 and the decoding unit 42 are performed based on the same histogram data. Therefore, the histogram aggregation unit 10 and the histogram aggregation unit 40 reset the histogram data to a predetermined value at a predetermined timing based on the common recognition. As an example, the histogram data is reset at the timing of the GoP (group of pictures) delimiter.

Further, for example, the histogram totaling unit 10 and the histogram totaling unit 40 reset the ^{histogram data to the initial value H (0).} As described above, the initial value H ⁽⁰⁾ of the histogram may be a frequency value of 0, or may use a non-zero value predetermined based on a predetermined probability distribution. ^{For example, the initial value H (0)} is set according to a typical distribution of frequency values measured or modeled in various images in advance.

When the frequency of resetting is relatively low, the mode information coding device 1 and the mode information decoding device 4 obtain the mode information of the learning target area and the mode information of the learning reference area obtained from a sufficient amount of images. You can build a reflected histogram. That is, there is a high possibility that more learning effects can be obtained. On the contrary, if the frequency of resetting is relatively high, even if the processing is started from the state where the histogram data does not match between the mode information coding device 1 side and the mode information decoding device 4 side, both of them are performed relatively early. The histograms can be matched to convey mode information. The state in which the histogram data does not match between the mode information coding device 1 side and the mode information decoding device 4 side is, for example, when the mode information decoding device 4 side starts decoding from the middle of the content (such as a video). Can occur in.

Further, the histogram may be reset on the mode information coding device 1 side and the mode information decoding device 4 side at the timing when the frequency value included ^{in the histogram H (n) reaches a predetermined state.} As an example, when ^{the maximum value of the frequency values (non-negative values) included in the histogram H (n)} exceeds a predetermined threshold value, a histogram is created between the mode information encoding device 1 side and the mode information decoding device 4 side. You may want to reset it. This makes it possible to prevent the frequency value from overflowing when the size of the memory area of each frequency value is fixed (for example, the maximum number of digits in the decimal representation is fixed, or the number of bits in the binary representation is fixed). .. In this case, as an example of the initialization method, each frequency value may be divided by a predetermined constant (for example, 2, 4, etc.) (the fractional part of the division result is rounded down or rounded up, or determined in advance). ..

Even if at least a part of the functions of the mode information coding device 1, the entropy coding unit 2, the entropy decoding unit 3, and the mode information decoding device 4 in the above-described embodiment are realized by a computer. good. In that case, the program for realizing this function may be recorded on a computer-readable recording medium, and the program recorded on the recording medium may be read by the computer system and executed. The term "computer system" as used herein includes hardware such as an OS and peripheral devices. The "computer-readable recording medium" is a portable medium such as a flexible disk, a magneto-optical disk, a ROM, a CD-ROM, a DVD-ROM, or a USB memory, or a storage device such as a hard disk built in a computer system. Say that. Further, a "computer-readable recording medium" is a communication line for transmitting a program via a network such as the Internet or a communication line such as a telephone line, and dynamically holds the program for a short period of time. It may also include a program that holds a program for a certain period of time, such as a volatile memory inside a computer system that serves as a server or a client in that case. Further, the above-mentioned program may be a program for realizing a part of the above-mentioned functions, and may be a program for realizing the above-mentioned functions in combination with a program already recorded in the computer system.

According to the mode information coding apparatus of the embodiment described above, the spatial correlation of the mode values is learned based on the mode information of the image region encoded before the target block. Based on the premise that the statistical properties of the image are similar between the image area used for learning and the image area including the target block, by using the learning results, the higher the frequency of appearance, the smaller the code value (c). Can be converted in the case of. The more similar the above statistical properties are, the more successful it is to convert a mode value with a high frequency of occurrence to a small code value. The frequency of appearance of codewords obtained as a result of conversion is higher for codewords with smaller values. As a result of concentrating energy on a small value, the entropy of the codeword is also reduced, and the coding efficiency of mode coding can be improved.

Further, according to the mode information decoding device of the embodiment, the mode information decoding device has already received the order information of the appearance frequency required for decoding the encoded mode information from the mode information decoding device. The ranking information can be obtained from the decoded mode information. This makes it possible to decode the mode information without an overhead that lowers the coding efficiency.

Although the embodiments have been described above, the present invention can be further implemented in the following modifications.
[First modification]
In the above embodiment, the relationship between the rank (order) d based on the frequency value and the code value c is defined to be represented by the equations (5) and (6). However, it is not always necessary to determine the relationship between the order (rank) d and the code value c according to these equations, and the order d and the code value c may be used as other correspondence relationships. However, the correspondence is determined so that the code value c is uniquely determined based on the order d and the order d is uniquely determined based on the code value c.

As an example, the relationship between the rank d and the code value c is determined by the following formula. Here, d _max is the maximum value of the rank d. By this equation, the rank d and the code value c can be converted to each other.
c = d _max −d

Further, the method of determining the code value c may be changed for each hierarchy of the frequency value ranking. As an example, when there are 35 types of modes, the layers are layered as follows according to the order d (1 ≦ d ≦ 35).
In the case of the first layer (1 ≦ d ≦ 8), the code value c is determined by the above equation (5).
In the case of the second layer (9 ≦ d ≦ 16), the code value c (however, 8 ≦ c ≦ 15) is set to each of the mode value candidates belonging to the second layer according to another predetermined rule independent of the rank d. Give to.
In the case of the third layer (17 ≦ d ≦ 24), the code value c (however, 16 ≦ c ≦ 23) is set to each of the mode value candidates belonging to the third layer according to another predetermined rule independent of the rank d. Give to.
In the case of the fourth layer (25 ≦ d ≦ 35), the code value c (however, 24 ≦ c ≦ 34) is set to each of the mode value candidates belonging to the fourth layer according to another predetermined rule independent of the rank d. Give to.
In this case, the code value c output from the mode information encoding device 1 does not represent the total order of the frequency values of the mode value candidates, but the order of the upper order (first layer in the above example) and the lower order (first layer in the above example). In the above example, it represents a half order (from the second layer to the fourth layer). That is, the code value c in the case of this example reduces the entropy of the mode information.
The mode information decoding device 4 side performs the ordering corresponding to the above ordering.

That is, the coding unit 12 on the mode information coding device 1 side encodes the mode information of the target region as a code based on at least a part of the order obtained by the sorting unit 11.
Further, the decoding unit 42 on the mode information decoding device 4 side decodes the mode information of the target region by using the method corresponding to the coding unit 12.

The relationship between the rank d and the code value c is arbitrary, not limited to the method of determining the code value c illustrated in the first modification.

[Second modification]
In the above embodiment, the sort unit 11 on the mode information coding device 1 side and the sort unit 41 on the mode information decoding device 4 side each perform the sorting process of the frequency value as follows. That is, the sort unit 11 and the sort unit 41 are reference mode value strings (y ₁ , y ₂ , ..., Y _K ) of the reference area associated with the target area to be encoded or decoded at that time. The frequency values of all mode value candidates X were sorted in descending order. On the other hand, in the modification described here, the sort unit 11 and the sort unit 41 order the mode value candidates X by other methods.

For example, the sort unit 11 and the sort unit 41 sort the mode value candidate X in ascending order of the frequency value. Even when the frequency values are sorted in ascending order in this way, the information on the frequency order of the mode value X can be obtained in the same manner, and the mode information can be efficiently encoded.
Further, for example, the sort unit 11 and the sort unit 41 sort in order of frequency value (descending or ascending order) for all mode value candidates X, but sort in order of frequency value only for mode value candidate X belonging to a predetermined range. Perform processing. As an example, the sort unit 11 and the sort unit 41 sort only the mode value candidate X whose frequency value is higher than a predetermined threshold value, and do not perform the sort process for the other mode value candidates. Alternatively, as another example, the sort unit 11 and the sort unit 41 sort only the mode value candidate X of the upper N frequency values (N is a positive integer), and do not perform the sort process for the other mode value candidates. .. In these cases, the sort unit 11 and the sort unit 41 give an order other than the frequency value order to the mode value candidate X that is not the target of the sort process. In these cases as well, it is possible to reduce the entropy of the mode information for the mode value candidate X having a relatively high frequency value.
In any of the cases shown here, the sort unit 11 and the sort unit 41 perform common processing to order the mode value candidates X.

That is, the sort unit 11 and the sort unit 41 use the mode of the target area based on at least a part of the appearance frequency held by the histogram data with respect to the mode information of the reference area corresponding to the target area which is the region to be encoded. Order the information.

Although the embodiments and modifications of the present invention have been described in detail with reference to the drawings, the specific configuration is not limited to the embodiments and the like, and the design and the like within a range not deviating from the gist of the present invention are also included. included.

INDUSTRIAL APPLICABILITY The present invention can be used for encoding a still image or a moving image and recording it on a recording medium or transmitting it on a transmission medium. For example, it can be used in a broadcasting business, a communication business, a content distribution business, a business of packaging and selling content, and the like. It can also be used in various businesses that apply these technologies. It can also be used in the business of manufacturing and selling equipment or programs for their coding or after-wear. The availability of the present invention is not limited to the businesses (industries) exemplified here.

1 Mode information coding device 2 Entropy coding unit 3 Entropy decoding unit 4 Mode information decoding device 5 Learning target area and learning reference area 6 Reference area 7 Learning target area and learning reference area 8 Reference area 10 histogram aggregation unit (Frequency information generator)
11 Sorting section (ordering section)
12 Coding unit 21 and 22 Target area 40 Histogram aggregation unit (frequency information generation unit)
41 Sorting section (ordering section)
42 Decoding unit 50 Histogram (H ^(n-1) )
51 Target mode value for learning and reference mode value for learning 52 Mode value (H ^(n-1) (1, 2, 3))
53 Histogram (H ⁽ⁿ⁾ )
54 Frequency value (H ⁽ⁿ⁾ (1, 2, 3))
55 Target area 56 Reference area 57 Projection (corresponds to the case where the _{reference mode values y 1} = 3 and y _{2 = 4)}
58 Mode value sort result 59 Equal sign representing bins with equal frequency value 60 Target mode value 61 Target area 62 Target mode value

Claims (4)

Frequency of appearance of a set of mode information of a learning target area in a coded area which is a coded area and mode information of a learning reference area corresponding to the learning target area in the coded area. Frequency information generator that generates frequency information that represents
An ordering unit that orders the mode information of the target area based on a part of the appearance frequency held by the frequency information with respect to the mode information of the reference area corresponding to the target area which is the target area to be encoded.
A coding unit that encodes the mode information of the target region as a code based on at least a part of the order obtained by the ordering unit.
Equipped with
The ordering unit determines an order value in the order of the appearance frequency based on the frequency information, and divides the range of the order of the appearance frequency into a plurality of layers.
The coding unit assigns a code corresponding to the hierarchy to the mode information of the target area by offsetting a predetermined value to the sequence value with respect to the hierarchy to which the sequence value 1 belongs, and the sequence of the hierarchy to which the sequence value 1 does not belong. from said sequence value for each a shall be given a code in accordance with independent another predetermined rule mode information of the target region,
The learning reference area includes K learning reference blocks (where K is an integer of 2 or more).
The frequency information is represented as a (K + 1) dimensional matrix including the dimension of the mode information of the target area for learning and the dimension of the mode information for each reference block for learning.
The ordering unit orders the mode information of the target area by performing sorting processing according to the occurrence frequency value corresponding to the mode information for each learning reference block based on the (K + 1) dimensional matrix. do,
A mode information encoding device characterized in that. Frequency representing the appearance frequency of a set of the mode information of the learning target area in the already decoded area, which is the decoded area, and the mode information of the learning reference area corresponding to the learning target area in the already decoded area. Frequency to generate information Information generation unit and
An ordering unit that orders the mode information of the target area based on a part of the appearance frequency held by the frequency information with respect to the mode information of the reference area corresponding to the target area that is the area to be decoded.
A decoding unit that decodes specific mode information selected from the ordered mode information of the target area based on the input code value, and
Equipped with
The ordering unit determines an order value in the order of the appearance frequency based on the frequency information, and divides the range of the order of the appearance frequency into a plurality of layers.
The decoding unit decodes the mode information of the target area corresponding to the order value obtained by offsetting a predetermined value to the code value with respect to the layer to which the order value 1 belongs, and the layer to which the order value 1 does not belong. are those for respectively for decoding the mode information of the target region corresponding to the code values according to independent another predetermined rule and the order value,
The learning reference area includes K learning reference blocks (where K is an integer of 2 or more).
The frequency information is represented as a (K + 1) dimensional matrix including the dimension of the mode information of the target area for learning and the dimension of the mode information for each reference block for learning.
The ordering unit orders the mode information of the target area by performing sorting processing according to the occurrence frequency value corresponding to the mode information for each learning reference block based on the (K + 1) dimensional matrix. do,
A mode information decoding device characterized by the above. Computer,
The mode information encoding device according to claim 1.
A program to function as. Computer,
The mode information decoding device according to claim 2.
A program to function as.

Images