CN105430418B - H.264/AVC, one kind arrives HEVC fast transcoding methods - Google Patents

Abstract

It is a kind of rapidly and efficiently H.264/AVC arrive HEVC fast transcoding methods, in code stream information H.264/AVC, if all MB use SKIP patterns in the region of one 64 × 64, and these are less than threshold value 0.01 using the MB of SKIP patterns MV variance, when HEVC is encoded, in decomposable process, CU Depth are selected LCU corresponding to 64 × 64 region from 0 or 1, and each PU predictive mode then only selects from SKIP, MERGE or 2N × 2N；If all MB use SKIP patterns in the region of one 32 × 32, and the MB of these SKIP patterns MV variance is both less than threshold value 0.01, when HEVC is encoded, CU is in decomposable process corresponding to 32 × 32 region, CU Depth only select from 1 or 2, and each PU predictive mode then only selects from SKIP, MERGE or 2N × 2N.Encoder complexity is this method reduce, improves transcoding efficiency.

Description

A fast transcoding method from H.264/AVC to HEVC

technical field

The invention relates to a transcoding method of different coding standards in video coding of video communication, and belongs to the technical field of transcoding between different coding standards of video communication.

Background technique

The new-generation video coding standard HEVC is developing rapidly, and the previous-generation video coding standard H.264/AVC is still widely used. In order to reduce the cost of installing multiple decoders on terminal equipment, the birth of transcoders has become inevitable.

The largest coding unit in H.264/AVC is a macro block (Macro Block, MB), the size is fixed at 16×16, and the intra prediction mode has 16×16, 8×8, 4×4 brightness prediction, and the inter prediction mode There are SKIP, 16×16, 16×8, 8×16, 8×8, 8×4, 4×8 and 4×4.

Compared with H.264/AVC, HEVC has more flexible coding units and more diverse prediction modes. Its largest coding unit (Largest Coding Unit, LCU) is , and its size is usually set to 64×64. Each LCU can be divided into 64×64, 32×32, 16×16, 8×8, 4×4 coding units (Coding Unit, CU). In an LCU with a size of 64×64, the coding unit depth (CodingUnit Depth, CU Depth) corresponding to each size (64×64, 32×32, 16×16, 8×8, 4×4) of the CU is respectively It is 0, 1, 2, 3, 4. For each CU, in the prediction process, it can be divided into several prediction units (Prediction Unit, PU). The maximum size of each PU is 32×32 and the minimum is 4×4. In intra prediction, HEVC defines 33 intra prediction directions and two special prediction modes of DC prediction and planar prediction; in inter prediction, HEVC defines SKIP, MERGE, 2N×2N, 2N×N, N Seven prediction modes including ×2N, AMP and N×N.

Generally speaking, transcoding from H.264/AVC to HEVC requires the cascading of H.264/AVC decoder and HEVC encoder. The transcoder first decodes the code stream encoded with H.264/AVC, then re-encodes the decoded video using the HEVC standard, and converts it into an encoded code stream of the HEVC standard. However, due to the extremely high encoding complexity of HEVC, the transcoding efficiency will naturally be low.

The existing H.264/AVC to HEVC transcoding uses the H.264/AVC to HEVC transcoding technology based on probability statistics analysis.

The main purpose of the transcoding algorithm using probability and statistical analysis is to find the correlation between two coding standards. After a lot of statistical analysis, it is found that in the video encoded by the H.264/AVC standard, if a part of the area is coded in skip mode (SKIP); then, when the video is encoded using the HEVC standard, the CU Depth of this area is usually Less than 1 (but not absolute), the PU division is simpler. Similarly, if the video encoded by the H.264/AVC standard has a small variance in the motion vector (Motion Vector, MV) in a part of the area; then, when the video is encoded using the HEVC standard, the CU Depth of the area It is also likely to be lower. Therefore, in the existing transcoding methods, some researchers use the SKIP mode flag or the MV variance in the H.264/AVC code stream to set the CU Depth termination conditions for HEVC coding, thereby reducing the coding complexity of HEVC and improving Transcoding efficiency.

The limitation of the existing transcoding technology is that, first, simply considering the SKIP mode or MV variance to set the CUDepth of HEVC will lead to misjudgment of the CU Depth of many CUs during HEVC encoding, and the CUs that should continue to be decomposed will be forced to fail. Then decompose, that is, use a CU with a smaller CU Depth for encoding. Second, in the determination of the PU mode, the prior art only considers the SKIP mode; in fact, in the H.264/AVC code stream, even when a certain area in the image adopts the SKIP mode, in HEVC, the Regions are also very likely not to use SKIP mode. It can be seen that although the existing transcoding technology can improve the encoding speed of HEVC, the compression performance of HEVC is also greatly reduced. Third, in the HEVC encoding process of the existing transcoding technology, the rate-distortion cost is still calculated in the order of CU Depth from small to large (from 0 to 4), and there is no code stream information based on H.264/AVC , to determine the most likely fine range of CU Depth in advance, so that the reduction of HEVC coding complexity is also limited.

Contents of the invention

Aiming at the deficiencies in the existing H.264/AVC to HEVC transcoding technology, the present invention proposes a fast and efficient H.264/AVC to HEVC fast transcoding method based on the effective information extracted from the H.264/AVC code stream , while greatly reducing the complexity of HEVC encoding, the method keeps the compression performance of HEVC basically unchanged.

The fast transcoding method from H.264/AVC to HEVC of the present invention is:

In the stream information of H.264/AVC, if all MBs (macroblocks) in a 64×64 area adopt SKIP mode (skip mode), and the MV (motion vector) of these MBs adopting SKIP mode If the variance is less than the threshold 0.01, then during HEVC encoding, during the decomposition process of the LCU (largest coding unit) corresponding to the 64×64 area, the CU Depth (coding unit depth) is selected from 0 or 1, and each PU (prediction The prediction mode of unit) is only selected from SKIP, MERGE or 2N×2N; if all MBs in a 32×32 area adopt SKIP mode, and the variance of the MV of these SKIP mode MBs is less than the threshold 0.01, then During HEVC encoding, during the decomposition process of the CU (coding unit) corresponding to the 32×32 area, the CU Depth is only selected from 1 or 2, and the prediction mode of each PU is only selected from SKIP, MERGE or 2N×2N to choose from.

The present invention focuses on H.264/AVC and HEVC transcoding technology, adopts the method of probability and statistics analysis, finds out the correlation between the coding mode of H.264/AVC video coding standard and HEVC video coding standard, and judges the possibility of HEVC encoder in advance The encoding mode reduces the complexity of the HEVC encoder and improves the transcoding efficiency.

Description of drawings

Fig. 1 is a flowchart of the fast transcoding method from H.264/AVC to HEVC according to the present invention.

Fig. 2 is a schematic diagram of the HEVC encoding process of a 64×64 area.

FIG. 3 is a schematic diagram of the HEVC encoding process for a 32×32 area.

Fig. 4 is a graph comparing the rate-distortion curves of HEVC encoding and original HEVC encoding after applying the present invention. Among them: Origin is the rate-distortion curve encoded under the original HEVC standard; Full_alg is the rate-distortion curve after applying this algorithm. Figure 4(a) is the rate-distortion curve comparison of the video sequence "Flowervase", Figure 4(b) is the rate-distortion curve comparison of the video sequence "SlideEdit", and Figure 4(c) is the rate-distortion curve of the video sequence "Cactus" Curve comparison chart.

Detailed ways

The fast transcoding method from H.264/AVC to HEVC of the present invention is that in the code stream information of H.264/AVC, if all MBs (macroblocks) in a 64×64 area adopt the SKIP mode (skip mode), and the variance of the MV (motion vector) of these MBs using SKIP mode is less than the threshold 0.01, then during HEVC encoding, the LCU (largest coding unit) corresponding to the 64×64 area is in the decomposition process, CU Depth( Coding unit depth) is selected from 0 or 1, and the prediction mode of each PU (prediction unit) is only selected from SKIP, MERGE or 2N×2N; if all MBs in a 32×32 area use SKIP mode , and the variance of the MV of these MBs in SKIP mode is less than the threshold 0.01, then during HEVC encoding, during the decomposition process of the CU corresponding to the 32×32 area, CUDepth is only selected from 1 or 2, and each PU’s The prediction mode can only be selected from SKIP, MERGE or 2N×2N. As shown in Figure 1, it specifically includes the following steps:

(1) Divide each frame of the video into several 64×64 areas. If it is found in the process of parsing the H.264/AVC code stream, all the MBs contained in the current 64×64 area are in SKIP mode, and The sum of the variances of all motion vectors in the 64×64 area in the horizontal and vertical directions is less than the threshold 0.01, then set the fast transcoding flag L ₁ in the 64×64 area, otherwise set it as L ₀ ;

(2) Divide each frame of the video into several 32×32 areas. If it is found in the process of parsing the H.264/AVC code stream, the 4 MBs contained in the current 32×32 area are all in SKIP mode, And the sum of the variances of all motion vectors in the 32×32 area in the horizontal and vertical directions is less than the threshold 0.01, then during the HEVC encoding process, set the fast transcoding flag to S ₁ for the current 32×32 area, otherwise set is S ₀ ;

(3) Referring to Figure 2, in the HEVC encoding process, if the transcoding flag of the 64×64 area corresponding to an LCU (64×64 in size) is found to be L ₁ , then during the decomposition process of the LCU, CU Depth is only selected from 0 or 1, and the prediction modes of all PUs in the LCU are only selected from SKIP, MERGE, or 2N×2N.

(4) Referring to Figure 3, during HEVC encoding, if it is found that the transcoding flag of a 64×64 area corresponding to an LCU (64×64 in size) is L ₀ , and the 64×64 area is The transcoding flags of the four 32×32 areas included are not all S ₁ (that is, the four 32×32 area transcoding flags include both S ₁ and S ₀ ), then during the decomposition process of the LCU, Skip the case where the CU depth is 0. For a 32×32 CU in the LCU, if its fast transcoding flag is S ₁ , then the CU is only decomposed within the range of CU Depth between 1 and 2, and the optimal prediction mode of all PUs in the CU is only It can be selected among SKIP, MERGE or 2N×2N; if its quick encoding flag is S ₀ , then the CU Depth of the CU can be selected in a range other than 0, and the optimal prediction mode of all PUs is from the HEVC specified Select from all prediction modes.

Experimental results show that using the coding information of the H.264/AVC standard can greatly reduce the coding complexity of HEVC and improve the efficiency of the transcoder while keeping the compression performance of HEVC basically unchanged. Figure 4 shows the comparison of rate-distortion curves between HEVC encoding and original HEVC encoding after applying the present invention, Figure 4(a) is a comparison of the rate-distortion curves of the video sequence "Flowervase", and Figure 4(b) is the video sequence "SlideEdit" The rate-distortion curve comparison chart of the video sequence "Cactus" is shown in Fig. 4(c). After adopting the method of the present invention, the reduction degree (ΔTime) of the encoding complexity of the HEVC encoder is as follows (ΔTime represents the time saved, and the larger the absolute value, the more the complexity is reduced, and the transcoder efficiency is higher):

video ΔTime Flower Vase -31.13% Slide Editing -63.13% Cactus -23.32%

Compared with the encoding platform of the HEVC standard, the method proposed by the present invention can shorten the encoding time by 23.32%-63.13%, and the average shortening is 39.19%, and under the same encoding bit rate condition, the quality of the decoded video is only reduced by 0.026dB.

Claims (1)

1. it is a kind of rapidly and efficiently H.264/AVC arrive HEVC fast transcoding methods, it is characterized in that：

(1) each two field picture of video is divided into the region of several 64 × 64 sizes, if parsing H.264/AVC code stream mistake Found in journey, all MB included in current 64 × 64 region are SKIP patterns, and all motions in 64 × 64 region The variance sum of vector in the vertical and horizontal directions is less than threshold value 0.01, then by 64 × 64 region setting fast transcoding mark L₁, otherwise it is set to L₀；

(2) each two field picture of video is divided into the region of several 32 × 32 sizes, if parsing H.264/AVC code stream mistake Found in journey, 4 MB included in current 32 × 32 region are SKIP patterns, and all motion arrows in 32 × 32 region The variance sum of amount in the vertical and horizontal directions is less than threshold value 0.01, then in HEVC cataloged procedures, to current 32nd × 32 area Domain sets fast transcoding flag bit as S₁, otherwise it is set to S₀；

(3) in HEVC cataloged procedure, if the transcoding flag bit for finding 64 × 64 region corresponding to a LCU is L₁, that In decomposable process, CU Depth are only selected the LCU from 0 or 1, and in the LCU all PU predictive mode all only from Selected in SKIP, MERGE or 2N × 2N；

(4) in HEVC cataloged procedure, if the transcoding flag bit for finding 64 × 64 region corresponding to a LCU is L₀, and And the transcoding flag bit in the region of 4 32 × 32 that 64 × 64 region is included is not all S₁, then the LCU was being decomposed Cheng Zhong, skip the situation that CU depth is 0；Some 32 × 32 CU in the LCU, if its fast transcoding flag bit is S₁, that The CU only CU Depth be 1 and 2 in the range of decomposed, in CU all PU optimal prediction modes can only SKIP, Selected in MERGE or 2N × 2N；If its fast coding flag bit is S₀, then the CU Depth of the CU are beyond 0 Selected in the range of other, all PU optimal prediction modes then select from whole predictive modes as defined in HEVC.