JP2015226316A - Conversion quantization method, conversion quantization device and conversion quantization program - Google Patents

Abstract

PROBLEM TO BE SOLVED: To reduce the amount of computation sufficiently, when performing conversion quantization at high speed in an HEVC using the differential feature amount.SOLUTION: A conversion quantization method performing conversion quantization by calculating quantization conversion coefficients from an inter-prediction residue in an image coding and the sum of absolute difference thereof, includes the steps of: determining whether or not it is a TW block on the basis of comparison results of the sum of absolute difference thus obtained and a predetermined threshold; outputting the inter-prediction residue, and performing conversion quantization of full size full depth when it is determined that it is the TW block on the basis of the determination result, otherwise outputting the inter-prediction residue and the sum of absolute difference; and determining a TU size and determining whether it is an M block or an L block, and performing conversion quantization of the inter-prediction residue, on the basis of the TU size thus determined, when it is determined that it is the M block, otherwise skipping conversion quantization by making all quantization conversion coefficients zero.

Description

  The present invention relates to a transform quantization method, transform quantization apparatus, and transform quantization program for performing transform quantization at high speed in HEVC (High Efficiency Video Coding) using a residual feature.

  HEVC, which is attracting attention as a next-generation video compression standard, has excellent compression efficiency due to techniques such as optimization of the block size. In comparison with H.264 / AVC, it has been announced that the bit rate can be reduced by about 40%. However, while compression efficiency is excellent, compression is said to require several times the amount of calculation (see, for example, Non-Patent Document 1). This large amount of calculation results mainly from “tree structure prediction” and “multiple prediction modes”, which are features of HEVC.

  In order to solve the problem related to the large amount of calculation, an all-zero block (AZB) algorithm has been used in a conventional encoding technique. The AZB algorithm is a method in which an all-zero block, that is, a block in which all coefficients after quantization are all zero can be determined in advance without calculating transform quantization. Since it is not necessary to perform transform quantization on a block determined to be an all-zero block, the calculation amount is reduced as a result.

  Many of the proposals so far have been related to H.264, which is a conventional video compression standard. H.263 8 × 8 conversion and H.264. This is done for AZB detection in H.264 4 × 4 integer transform. In Non-Patent Document 2, the determination of all-zero transform coefficient blocks is not analyzed in detail, so that sufficient conditions are not derived and erroneous detection occurs.

  In Non-Patent Document 3, theoretical conditions are derived in order to improve the conventional algorithm. Non-Patent Document 4 proposes a low error detection determination method based on an analysis of necessary and sufficient conditions for early detection of all-zero coefficient blocks. In Non-Patent Document 5 and Non-Patent Document 6, H. A model applied to 4 × 4 integer conversion in H.264 / AVC is proposed. Recently, in Non-Patent Document 7, H.C. A more effective sufficient condition has been derived for the early determination of all-zero 4 × 4 blocks in H.264 / AVC.

  On the other hand, in Non-Patent Document 8, there is a method for determining the TU (conversion unit processing unit) size by comparing the threshold calculated based on the Gaussian model and the SAD (sum of absolute differences) of the block. Proposed. In Non-Patent Document 8, the threshold value is calculated using the Gaussian model, but the model used for calculating the threshold value is not limited to the Gaussian distribution (Gaussian model), and other distributions such as the Laplace distribution (Laplacian model). You may follow the model.

F Bossen, B Bross, K Suhring, "HEVC Complexity and Implementation Analysis" IEEE Transactions on Circuits and Systems for Video Technology, vol. 22, no. 12, pp. 1685 --1696, 2012. Zhou Xuan, Yu Zhenghua, and Yu Songyu, "Method for detecting all-zero dct coefficients ahead of discrete cosine transformation and quantisation," Electronics Letters, vol. 34, no. 19, pp. 1839 --1840, sep 1998. LA Sousa, "General method for reducing redundant computations in video coding," Electronics Letters, vol. 36, no. 4, pp. 306--307, 2000. S. Jun and S. Yu, "Efficient method for early detection of all-zero DCT coefficients," Electronics letters, vol. 37, no. 3, pp. 160--161, 2001. Y. Wan, Y. Zhou, and H. Yang, "Early detection method of all-zero integer transform coefficients," Consumer Electronics, IEEE Transactions on, vol. 50, no. 3, pp. 923--928, 2004. YH Moon, GY Kim, and JH Kim, "An improved early detection algorithm for all-zero blocks in H. 264 video encoding," Circuits and Systems for Video Technology, IEEE Transactions on, vol. 15, no. 8, pp. 1053--1057, 2005. H. Wang, S. Kwong, and CW Kok, "Efficient prediction algorithm of integer DCT coefficients for H.264 / AVC optimization," Circuits and Systems for Video Technology, IEEE Transactions on, vol. 16, no.4, pp. 547--552, 2006. Jia Su, Koyo Nitta, Mitsuo Ikeda, Atsushi Shimizu, "Residue role assignment based transform partition predetermination on HEVC" IEEE ICIP pp. 2019--2023, Sept. 2013.

  However, the existing method of determining the TU size based on the threshold based on the Laplace distribution or the Gaussian distribution according to Non-Patent Document 8 has a problem that a sufficient amount of calculation reduction effect cannot be obtained mainly for two reasons. The first reason is that for a block determined to be a TW (Temporary Worker) block based on the procedure shown in Non-Patent Document 8, the TU size is determined to be unpredictable in advance. Then, the transform quantization operation is recursively performed for the current size and a size smaller than the current size (for example, 3 sizes of 2N × 2N, N × N, and (N / 2) × (N / 2)). Therefore, there are many redundant operations. The second reason is that the Laplace distribution or Gaussian distribution cannot provide sufficient accuracy for determining a TW block.

  The present invention has been made in view of such circumstances, and a transform quantization method capable of sufficiently reducing the amount of computation when performing transform quantization at high speed in HEVC using residual feature amounts, An object of the present invention is to provide a transform quantization apparatus and a transform quantization program.

  The present invention is a transform quantization method for performing transform quantization by calculating a quantized transform coefficient from an inter prediction residual in image coding and a sum of absolute differences thereof, and the obtained difference absolute value sum and a predetermined value A determination step for determining whether or not the transform quantization target block corresponds to a TW block for which the transform quantization operation cannot be simplified, and a determination result of the determination step. When it is determined that the block corresponds to the TW block, the inter prediction residual is output, transform quantization of all the depths of all sizes is performed, and when it is determined that the block does not correspond to the TW block, the inter prediction is performed. Output the residual and the sum of absolute differences, determine the TU size, determine whether it is an M block or an L block, and determine if it is determined to be the M block Transform quantization of the inter prediction residual based on the TU size, and when it is determined to be the L block, the quantized transform coefficient is all zero and has a transform quantization step for skipping transform quantization It is characterized by.

  The present invention is characterized in that the threshold value is determined by fitting with a beta distribution.

  The present invention is a transform quantization apparatus that performs transform quantization by calculating a quantized transform coefficient from an inter prediction residual in image coding and a sum of absolute differences thereof, and the obtained difference absolute value sum and a predetermined value Based on the result of the comparison with the threshold value, determination means for determining whether the transform quantization target block corresponds to a TW block that cannot simplify the transform quantization operation, and based on the determination result of the determination means When it is determined that the block corresponds to the TW block, the inter prediction residual is output, transform quantization of all the depths of all sizes is performed, and when it is determined that the block does not correspond to the TW block, the inter prediction is performed. The residual and the sum of absolute differences are output, the TU size is determined, whether the block is an M block or an L block, and when the block is determined to be the M block, the determined TU And transform quantization of the inter prediction residual based on noise, and when it is determined to be the L block, the quantized transform coefficients are all zero, and transform quantization means for skipping transform quantization is provided. And

  The present invention is a transform quantization program for causing a computer to execute the transform quantization method.

  According to the present invention, it is possible to sufficiently reduce the amount of calculation when transform quantization is performed at high speed in HEVC using residual feature quantities.

It is a block diagram which shows the structure of one Embodiment of this invention. It is a figure which shows the processing operation of the TW determination part 2 shown in FIG. It is a figure which shows the operation | movement of the TU size determination process in the AZB determination part 3 shown in FIG. It is the figure which showed the fitting by a probability distribution model with respect to the probability density function with respect to the value of SAD. It is a figure showing the difference of the fitting value by a probability distribution model, and the distribution actually observed.

  Hereinafter, a transform quantization apparatus according to an embodiment of the present invention will be described with reference to the drawings. FIG. 1 is a block diagram showing the configuration of the embodiment. FIG. 1 is a diagram showing a configuration of an HEVC inter coding processing apparatus including a transform quantization apparatus according to the present invention. In FIG. 1, reference numeral 1 denotes an inter prediction unit that performs inter prediction. Reference numeral 2 denotes a TW determination unit that determines a TW block. Reference numeral 3 denotes an AZB determination unit that determines role assignment.

  Reference numeral 4 denotes an integer transform quantization unit with a predicted TU size that performs integer transform quantization with a predicted TU size. Reference numeral 5 denotes a traversal integer transform quantization unit that performs integer transform quantization of all sizes and depths. Reference numeral 6 denotes an entropy encoding unit that performs entropy encoding and outputs an output bit stream. In this embodiment, the TW determination unit 2 is arranged before the AZB determination unit 3, which is a great feature.

  Next, the processing operation of the HEVC inter coding processing apparatus shown in FIG. 1 will be described. First, the inter prediction unit 1 that has input the input original image processes the inter prediction, and then outputs the residual between the encoding target image and the predicted image and the SAD (sum of absolute differences) as a result of the inter prediction. . The inter prediction unit 1 outputs the residual between the encoding target image and the prediction image and the value of the SAD (sum of absolute differences) to the TW determination unit 2 in units of HEVC coding units (CU).

  Here, the processing operation of the TW determination unit 2 shown in FIG. 1 will be described with reference to FIG. FIG. 2 is a diagram illustrating a processing operation of the TW determination unit 2 illustrated in FIG. The TW determination unit 2 compares the SAD value with a threshold value THbeta (QP) calculated based on a procedure to be described later (step S1). When the comparison result is false (N), the TW determination unit 2 determines that this CU is a TW block, that is, a block whose TU size cannot be determined in advance.

Then, the TW determination unit 2 outputs the residual r _ij to the traversal integer transform quantization unit 5. Then, the transform quantization operation is recursively performed for the current size and a size smaller than the current size (that is, 3 sizes of 2N × 2N, N × N, and (N / 2) × (N / 2)). . On the other hand, when the comparison result is true (Y), the TW determination unit 2 determines that this CU is not a TW block, and outputs the SAD and the residual r _ij to the AZB determination unit 3.

Next, the operation of the TU size determination process in the AZB determination unit 3 shown in FIG. 1 will be described with reference to FIG. FIG. 3 is a diagram illustrating the operation of the TU size determination process in the AZB determination unit 3 illustrated in FIG. As shown in Tables 1 and 2, the threshold tables for the transform quantization size (N), the QP value (QP), and the bit depth (Bit depth) are calculated and stored in advance. The threshold table is a table in which thresholds are defined for combinations of transform quantization size (N), bit depth (Bit depth), and QP value (QP).

First, the AZB determination unit 3 inputs the residuals r _ij , SAD, and QP (step S11). Then, the AZB determination unit 3 determines whether or not SAD _{2N × 2N} <TH _{2N × 2N} (QP) is satisfied (step S12). As a result of this determination, if SAD _{2N × 2N} <TH _{2N × 2N} (QP) is satisfied, the AZB determination unit 3 outputs 2N × 2N, L block (Leader: leader block), q _ij = 0 (step S17). ). The L block means a block that is determined to be an all-zero block, and is a block that can directly set the transform / quantization matrix q _ij to 0 and skip the transform quantization operation process.

On the other hand, if SAD _{2N × 2N} <TH _{2N × 2N} (QP) is not satisfied, the AZB determination unit 3 determines whether or not SAD _{2N × 2N} <TH ′ _{2N × 2N} (QP) is satisfied (step S13). . If SAD _{2N × 2N} <TH ′ _{2N × 2N} (QP) is satisfied as a result of this determination, the AZB determination unit 3 outputs 2N × 2N M blocks (member blocks) (step S18). The M block means a block that performs transformation / quantization calculation in a state where the transformation / quantization size is determined. In the M block (2N × 2N), the transformation quantization size is determined to be 2N × 2N. The conversion / quantization operation can be performed in step S1, and the operation of the N × N or (N / 2) × (N / 2) conversion / quantization block can be skipped.

Next, if SAD _{2N × 2N} <TH ′ _{2N × 2N} (QP) is not satisfied, the AZB determination unit 3 determines whether or not SAD _{N × N} <TH _{N × N} (QP) is satisfied (step S14). ). As a result of this determination, if SAD _{N × N} <TH _{N × N} (QP) is satisfied, the AZB determination unit 3 outputs N × N, L block, q _ij = 0 (step S19).

On the other hand, if SAD _{N × N} <TH _{N × N} (QP) is not satisfied, the AZB determination unit 3 determines whether or not SAD _{N × N} <TH ′ _{N × N} (QP) is satisfied (step S15). . If SAD _{N × N} <TH ′ _{N × N} (QP) is satisfied as a result of this determination, the AZB determination unit 3 outputs N × N, M blocks (step S20).

Next, if SAD _{N × N} <TH ′ _{N × N} (QP) is not satisfied, the AZB determination unit 3 satisfies SAD _{N / 2 × N / 2} <TH _{N / 2 × N / 2} (QP). It is determined whether or not (step S16). If SAD _{N / 2 × N / 2} <TH _{N / 2 × N / 2} (QP) is satisfied as a result of this determination, the AZB determination unit 3 determines that N / 2 × N / 2, L block, q _ij = 0 Is output (step S21).

On the other hand, if SAD _{N / 2 × N / 2} <TH _{N / 2 × N / 2} (QP) is not satisfied, the AZB determination unit 3 outputs N / 2 × N / 2 and M blocks (step S22). .

  In the inter prediction process by the inter prediction unit 1, the SAD is calculated by motion estimation for one coding unit (CU). When the SAD value in the block size of the currently input residual is compared with the corresponding threshold value and the SAD is smaller than the threshold value, the current residual block (for example, 2N × 2N) is Recognized as an all zero block (2N × 2N). If the condition is not satisfied, the residual block is divided into smaller sizes and repeated determination is performed.

  In the present embodiment, whether or not the block is a TW block has already been determined by the TW determination unit 2, and therefore the branching condition for determining whether or not the block is a TW block is omitted in the processing operation illustrated in FIG. 3.

  As described above, according to the processing operation shown in FIG. 3, the TU block size and the determination of the M block or the L block are performed. Since the block determined to be the L block is a block for which the coefficient after transform quantization is determined to be zero, all the coefficients are set to zero based on the determined TU size, and the transform quantization is skipped. For a block determined to be an M block, the residual is subjected to transform quantization in the integer transform quantization unit 4 having the predicted TU size based on the determined TU size. This processing result is entropy-encoded in the entropy encoding unit 5 and output as an output bit stream.

  Next, a method for determining the threshold value THbeta (QP) in the TW determination unit 2 shown in FIG. 1 will be described. In the following description, a calculation method based on the beta distribution will be described as an example of threshold setting in the TW determination unit 2. FIG. 4 shows fitting by three probability distribution models (beta distribution, Laplace distribution, and normal (Gaussian) distribution) for the probability density function (f (x)) with respect to the actually observed SAD value (x). FIG.

  Specifically, it is the probability density distribution of SAD values and its fitting observed for the residual of an inter-CU block of 32 × 32 pixels in an HDTV video composed of 1920 × 1080 pixels. The horizontal axis represents the SAD value, and the vertical axis represents the probability density. A line denoted by reference numeral C1 and a line denoted by reference numeral C2 represent fitting based on a Laplace distribution and a normal (Gaussian) distribution. The line indicated by the symbol C3 represents the fitting based on the beta distribution.

  Here, the fitting means modeling with respect to the residual distribution of the inter-CU block, and it is preferable to use a conventional method such as residual analysis or residual modeling. From this figure, it can be seen that the beta distribution can be fitted more closely than the fitting based on the previous Laplace or normal (Gaussian) distribution. This is because 74% of the samples in the residual SAD block are concentrated in the range of SAD from 0 to 10,000. This indicates that the TU size determination based on a threshold obtained only by fitting with a normal (Gaussian) distribution based on Non-Patent Document 8 or a Laplace distribution is inefficient.

  FIG. 5 shows the difference between the fitting value by each probability distribution model and the actually observed distribution. The horizontal axis represents the SAD value of 32 × 32 blocks, as in FIG. The vertical axis represents the difference between the fitting value by each probability distribution model and the actually observed distribution. It shows that the value by fitting and an actual distribution differ, so that a vertical axis | shaft leaves | separates from 0.

  The determination performed by the AZB determination unit 3 in the present embodiment shown in Non-Patent Document 8 is an all-zero block (AZB), that is, a property that determines whether or not an integer transform quantization result is all-zero in advance. belongs to. Therefore, a block having a residual value close to zero, that is, a block having a small SAD value exhibits high determination accuracy, but has a feature that it is difficult to determine a block having a large residual value, that is, a block having a large SAD value.

  Therefore, in this embodiment, SAD = 10000, which is the starting point at which the difference between the fitting based on the beta distribution and the actual distribution deviates in the positive direction in a large SAD region, is set as the threshold value. This setting of the threshold value is an example, and the threshold value may be obtained experimentally based on the actual encoding efficiency and the amount of calculation time reduction.

Next, effects produced by this embodiment will be described. First, encoding efficiency will be described. In order to compare the coding efficiency, generally, an RD curve in which a generated bit amount (bitRate) and distortion (distortion) of an original image are two-dimensionally graphed is used. The BDBR described in Reference 1 is used for video compression performance comparison in order to avoid the difference being slightly confusing.
Reference 1 “G. Bjontegard,“ Calculation of average PSNR differences between RD-curves, ”ITU-T VCEG-M33, 2001.”

This is an average bit rate difference between the processing based on this embodiment and the performance of the conventional HM (hereinafter referred to as default HM), and measures the quality of video as shown in Table 3. It is. + In BDBR represents a bit rate gain.

  The HEVC's new coding tool rate distortion optimization based on quantization (RDOQ) and TU = CU are combined and evaluated in a random access setup. The results are classified as first method I with RDO and TU = CU turned off, second method II with RDOQ turned on, and third method III with TU = CU turned on.

Table 4 shows the reduction in transform quantization time. Here, the time saving of transform quantization is calculated based on the equation (1).

TIME _HM is the transform quantization time consumed by HM12.1. TIME _proposal is the transform quantization time of the processing described above. It achieved a time reduction of 51% to 42% with RDOQ and NSQT. The processing method described above is described in H.C. Compared to H.264 conventional methods, it plays a more important role not only in transform quantization but also in saving total encoding time. By adding a quadtree structure for traverse calculation and large size transform quantization, the processing method described above has the effect of accelerating the entire encoder, using the AZB method.

  As described above, in the TU size determination method according to the present embodiment, it is determined in advance whether or not the block is a TW block. The process of determining whether or not the block is a TW block is executed before the TU size determination process based on Non-Patent Document 8. The threshold for this determination is set according to the probability density based on the beta distribution, for example. By determining the TW block in advance, the calculation redundancy is reduced and more time can be reduced.

  As described above, whether or not the transform quantization target block corresponds to a block (TW block) that cannot simplify the transform quantization operation is first determined using the sum of absolute differences of inter prediction residuals. By determining, the calculation redundancy can be reduced and the calculation efficiency can be improved. This is because the TW block determination means of various block sizes are realized by comparing the threshold value calculated by a predetermined method with the value of the residual, so that the encoding calculation time can be reduced.

  The HEVC inter coding processing apparatus in the above-described embodiment may be realized by a computer. In that case, a program for realizing this function may be recorded on a computer-readable recording medium, and the program recorded on this recording medium may be read into a computer system and executed. Here, the “computer system” includes an OS and hardware such as peripheral devices. The “computer-readable recording medium” refers to a storage device such as a flexible medium, a magneto-optical disk, a portable medium such as a ROM and a CD-ROM, and a hard disk incorporated in a computer system. Furthermore, the “computer-readable recording medium” dynamically holds a program for a short time like a communication line when transmitting a program via a network such as the Internet or a communication line such as a telephone line. In this case, a volatile memory inside a computer system serving as a server or a client in that case may be included and a program held for a certain period of time. Further, the program may be for realizing a part of the functions described above, and may be a program capable of realizing the functions described above in combination with a program already recorded in the computer system. It may be realized using hardware such as PLD (Programmable Logic Device) or FPGA (Field Programmable Gate Array).

  As mentioned above, although embodiment of this invention has been described with reference to drawings, the said embodiment is only the illustration of this invention, and it is clear that this invention is not limited to the said embodiment. is there. Therefore, additions, omissions, substitutions, and other modifications of the components may be made without departing from the technical idea and scope of the present invention.

  When transform quantization is performed at high speed in HEVC using the residual feature amount, it can be applied to an application in which it is indispensable to sufficiently reduce the calculation amount.

  DESCRIPTION OF SYMBOLS 1 ... Inter process part, 2 ... TW determination part, 3 ... AZB determination part, 4 ... Integer conversion quantization part of prediction TU size, 5 ... Traversal integer conversion quantization part, 6 ... Entry pea coding unit

Claims (4)

A transform quantization method that performs transform quantization by calculating a quantized transform coefficient from an inter prediction residual in image coding and a sum of absolute differences thereof,
A determination step of determining whether or not the transform quantization target block corresponds to a TW block that cannot simplify the transform quantization operation, based on a result of comparing the obtained difference absolute value sum and a predetermined threshold. When,
Based on the determination result of the determination step, when it is determined that the block corresponds to the TW block, the inter prediction residual is output, transform quantization of all sizes and depths is performed, and it is determined that the block does not correspond to the TW block The inter prediction residual and its sum of absolute differences are output, the TU size is determined and whether the block is an M block or an L block, and when the block is determined to be the M block, Based on the determined TU size, transform quantization of the inter prediction residual is performed, and when it is determined as the L block, a transform quantization step of skipping transform quantization with all quantized transform coefficients being zero A transform quantization method comprising:   The transform quantization method according to claim 1, wherein the threshold value is determined by fitting with a beta distribution. A transform quantization apparatus that performs transform quantization by calculating a quantized transform coefficient from an inter prediction residual in image coding and a sum of absolute differences thereof,
Determining means for determining whether or not the transform quantization target block corresponds to a TW block that cannot simplify the transform quantization operation, based on a result of comparing the obtained difference absolute value sum and a predetermined threshold value When,
Based on the determination result of the determination unit, when it is determined that the block corresponds to the TW block, the inter prediction residual is output, transform quantization of all sizes and depths is performed, and it is determined that the block does not correspond to the TW block The inter prediction residual and its sum of absolute differences are output, the TU size is determined and whether the block is an M block or an L block, and when the block is determined to be the M block, Based on the determined TU size, transform quantization of the inter prediction residual is performed, and when it is determined as the L block, the transform transform means for skipping transform quantization with all quantized transform coefficients set to zero, and A transform quantization apparatus comprising:   A transform quantization program for causing a computer to execute the transform quantization method according to claim 1.

Images