CN102984521B - High-efficiency video coding inter-frame mode judging method based on temporal relativity - Google Patents

Abstract

The invention discloses a high-efficiency video coding (HEVC) inter-frame mode judging method based on temporal relativity, which comprises the steps of forecast method configuration and forecast mode selection. The forecast mode selection adopts the temporal relativity between two adjacent frames; the similarity of PU (Physical Unit) modes of a large size CU (Control Unit) of a corresponding piece in the previous frame and a small size CU in the current piece is analyzed according to the relativity; and finally, a PU mode selection method of the current CUs in various dimension is designed for the corresponding pieces in various dimensions according to the similarity. Compared with the HEVC standard in the prior art, with the adoption of the HEVC inter-frame mode judging method, the complexity of codding computation is reduced to a greater extent on the premise that the bit rate and video quality are almost unchanged.

Description

Inter-mode decision method for high-performance video coding based on temporal correlation

technical field

The invention relates to the technical problem of video coding inter-frame mode in the field of image communication, in particular to a high-performance video coding inter-frame mode judgment method.

Background technique

The current international video coding standard is Advanced Video Coding (H.264/AVC). Compared with previous video coding standards, this video coding standard has greatly improved video coding performance. With the wide application of high-definition video technology, the maximum block size of 16×16 in H.264/AVC is no longer suitable for encoding high-definition video. To this end, the International Organization for Standardization-International Electrotechnical Commission/Moving Picture Experts Group (ISO-IEC/MPEG) and the International Telecommunication Union Telecommunication Standardization Organization/Video Coding Experts Group (ITU-T/VCEG) have been established. Video Coding Joint Development Team (JCT-VC), and is developing a new generation of international video standards, the High Performance Video Coding (HEVC) standard. The goal of the HEVC standard is to reduce the bit rate by half while maintaining the video quality of the H.264/AVC standard, that is, to double the compression rate. With the efforts of many scholars at home and abroad, the coding performance of the HEVC video coding standard has been greatly improved compared with H.264/AVC.

The H.264/AVC standard uses 16×16, 16×8, 8×16, 8×8, 8×4, 4×8 and 4×4 blocks for inter-frame prediction, while the HEVC standard uses 8×8 , 16×16, 32×32, and 64×64 size coding units (CUs) are first divided into blocks, and then each CU uses up to 8 different prediction unit (PU) modes for inter-frame prediction, 8 different The PU modes are PART_2N×2N, PART_N×2N, PART_2N×N, PART_N×N, PART_nL×2N, PART_nR×2N, PART_2N×nD, and PART_2N×nU. Calculate the rate-distortion cost for each size of CU and its internal PU prediction mode, and finally select an optimal CU block mode and its internal PU mode for video encoding. For 64×64 size CU, there are only 3 PU modes to choose from, for 32×32 and 16×16 size CU, there are 7 kinds of PU modes to choose, for 8×8 size CU, there are 4 PU modes to choose choose. For each PU mode, the encoder needs to calculate the rate-distortion cost. Therefore, the HEVC standard brings huge computational complexity for video encoding. For H.264/AVC, many scholars have proposed some fast inter-frame judgment methods, such as the article "Fast mode decision algorithm for H.264 using statistics of rate-distortion cost" published in the magazine "IET Electronics Letters" in the analysis On the basis of the statistical distribution of macroblock rate-distortion cost in the previous frame, a H.264/AVC fast inter-mode decision method based on adaptive threshold is proposed. The article "Fast mode decision based on mode adaptation" published in the journal "IEEE Transactions on Circuits and Systems for Video Technology" proposes a fast mode decision method based on mode adaptation. According to the video coding characteristics of adjacent blocks in space and time domain, the algorithm constructs a priority-based candidate mode list, and effectively selects the best inter-frame prediction mode according to the list. The article "Direct inter-mode selection for H.264 video coding using phase correlation" published in the magazine "IEEE transactions on imageprocessing" captures its motion vector by using the phase correlation between the current block and the reference block, according to the motion vector information Choose the best inter prediction mode from the list of candidate modes.

In order to reduce the computational complexity of HEVC video coding, scholars have proposed some methods, such as the article "Encoding complexity reduction by removal of N×N partition type" in the HEVC proposal "JCTVC-D087" from 16×16, 32×32 The PU mode of PART_N×N is canceled in the CU of 64×64 size and is only reserved in the CU of 8×8 size, thereby reducing the computational complexity of video encoding. The article "Early termination of CU encoding to reduce HEVC complexity" in the proposal "JCTVC-F045" proposes that when the parameter cbf=0, that is, when the AC coefficients after the discrete cosine transform (DCT) are all zero, skip all but PART_2N×2N other PU modes, thereby reducing the computational complexity of video coding. The above two methods have been added to the latest HEVC reference software HM7.0. In addition, in the current HEVC standard, for the current CU, the rate-distortion cost modes first detected are PART_2N×2N, PART_2N×N and PART_N×2N, if the rate-distortion cost value of the mode PART_2N×N is less than PART_N× For the rate-distortion cost value of the 2N mode, continue to detect the rate-distortion cost of the mode PART_2N×nU and PART_2N×nD, and skip the calculation of the rate-distortion cost of the mode PART_nL×2N and PART_nR×2N; on the contrary, when the mode PART_2N ×N rate-distortion cost value is greater than the rate-distortion cost of PART_N×2N mode, then continue to detect the rate-distortion cost of mode PART_nL×2N and PART_nR×2N, and skip the rate-distortion cost of mode PART_2N×nU and PART_2N×nD The calculation of distortion cost, this method also greatly reduces the computational complexity of HEVC video coding. All of the above methods reduce the computational complexity of HEVC video coding to a certain extent, but the current inter-frame prediction mode of HEVC video coding still has a large redundancy in the time domain.

Compared with H.264/AVC, HEVC has greatly improved the video compression ratio, but the computational complexity of video coding has increased a lot. Although many scholars have proposed some fast video coding methods for H.264/AVC , and a fast video coding method for HEVC, but the fast video coding method of H.264/AVC is not suitable for HEVC, and the current fast video coding method of HEVC still has some distance to the goal of applying it to real-time communication.

Contents of the invention

Aiming at the current situation and shortcomings of the high-performance video coding inter-frame mode judgment method in the prior art, the purpose of the present invention is to provide a new high-performance video coding inter-frame mode judgment method based on temporal correlation to reduce HEVC video coding Computational complexity, to achieve the goal of applying it to real-time communication.

The basic idea of the present invention is to use the similarity of PU mode selection between adjacent frames to effectively select the PU mode of the current CU according to the PU mode adopted by the large-sized CU in the previous frame, and skip some unlikely ones. The CU block size and PU prediction mode reduce the PU modes that need to be traversed, thereby reducing the number of rate-distortion cost calculations, and finally achieving the purpose of reducing the computational complexity of HEVC video coding.

The high-performance video coding inter-frame mode decision method based on time-domain correlation provided by the present invention includes prediction mode configuration and prediction mode selection. In the prediction mode configuration, the CU segmentation depth is not greater than 4, and the PU adopts symmetric and asymmetric comprehensive prediction. Mode or only symmetric prediction mode is used. In the prediction mode selection, the sum of the total rate-distortion cost of the current depth CU is compared with the sum of the total rate-distortion cost of the previous layer CU. If it is smaller than the upper layer, further Take the quadtree divided into 4 CUs with a lower level of depth, otherwise terminate the quadtree division, and the prediction mode selection includes the following steps:

(1) Detect the size of the CU block corresponding to the frame before the current CU block. If the size of the current CU block is smaller than the size of the corresponding CU block, enter the following step (2), otherwise traverse all PUs in the current CU block mode, and divide the quadtree into 4 deeper CU blocks, and repeat the above process for each CU block of the deeper layer;

(2) Determine whether the PU mode corresponding to the CU block in the previous frame is PART_2N×2N, if so, the current CU block only detects the rate-distortion cost of the PU mode of PART_2N×2N, and enters the following step (6), otherwise Enter the following step (3);

(3) Determine whether the PU mode corresponding to the CU block in the previous frame is PART_nL×2N or PART_nR×2N, and if so, the current CU block only detects the rate-distortion cost of the two PU modes of PART_N×2N and PART_2N×2N, And enter the following step (6), otherwise enter the following step (4);

(4) Determine whether the PU mode corresponding to the CU block in the previous frame is PART_2N×nU or PART_2N×nD, if so, the current CU block only detects the rate-distortion cost of the two PU modes of PART_2N×N and PART_2N×2N, And enter the following step (6), otherwise enter the following step (5);

(5) Detect the current CU block to detect the rate-distortion cost of all PU modes, and enter the following step (6);

(6) Determine whether the size of the current CU block is 1/4 of the size of the corresponding CU block in the previous frame. If so, the current CU block will no longer be divided into quadtrees; otherwise, the current CU block will be further adopted as a quadtree Divide into 4 deeper CU blocks, and repeat step (1) for each CU block of the deeper layer.

In the above technical solution, the CU segmentation depth is preferably 2-4, more preferably 4. Among the symmetric and asymmetric integrated prediction modes and the symmetric integrated prediction mode that can be adopted by the PU, the symmetric and asymmetric integrated prediction modes are preferentially used.

In the above technical solution, the rate-distortion cost can be determined by the following formula:

J _mode =(SAD _luma +w _chroma ×SAD _chroma )+λ _mode ×B _mode

In the formula, J _mode is the rate-distortion cost, SAD _luma is the mean square difference between the original image brightness and the predicted image brightness, SAD _chroma is the mean square difference between the original image chromaticity and the predicted image chromaticity, w _chroma is the weight of chromaticity distortion, λ _mode represents the Lagrangian multiplier, and B _mode represents the number of encoded bits in this mode.

According to the above method of the present invention, a video encoder that executes the above high-performance video coding inter-frame mode decision method based on temporal correlation can be programmed.

The present invention is based on following train of thought analysis and finishes:

When the PU mode of the corresponding block CU is PART_2N×2N, it means that the texture of the corresponding block should be relatively smooth or all moving objects in the area where it is located have the same motion vector, according to the height between two adjacent frames Similarity, the current block area should also have similar attributes, therefore, for the current CU block, it is only necessary to detect the PU mode PART_2N×2N.

When the PU mode of the corresponding block CU is PART_2N×N or PART_N×2N, in this case, according to the similarity between two adjacent frames, there is no PU in the corresponding block corresponding to the CU in the current block Therefore, at this time, it cannot be inferred that the corresponding block area is smooth or that the moving objects in it all have the same motion vector. Therefore, for the current CU block, there is no PU mode that can be referred to, and all PU modes must be traversed.

When the PU mode of the corresponding block CU is PART_nL×2N or PART_nR×2N, according to the time domain correlation between two adjacent frames and the motion direction in the time domain, on the one hand, it means that the moving object in the current block has a large The probability is divided into left and right parts. On the other hand, at this time, the dividing line of the PU mode PART_N×2N of some CUs in the current block will correspond to the dividing line of the PU mode PART_nL×2N or PART_nR×2N of the corresponding block CU, and then Or, according to the correlation between the PU modes of each layer in the current block (the current block in the current frame with the same position and size as the corresponding CU block) and the correlation of the PU modes of adjacent CUs under the same CU layer, in the current block In the PU mode of each layer CU, the probability of PART_N×2N is relatively high. Therefore, the current block should detect the PU mode PART_N×2N. In addition, in order to ensure the continuity between the CUs of each layer and ensure the video coding performance, the PU mode PART_2N×2N should also be taken into consideration. Therefore, in this case, only two PU modes are traversed: PART_N×2N and PART_2N×2N. Similarly, when the PU mode of the corresponding block CU is PART_2N×nU or PART_2N×nD, it means that the moving object in the current block has a high probability of being divided into left and right parts. Therefore, the current block should detect that the PU mode is PART_2N× N and PART_2N×2N.

Finally, when the block size of the corresponding CU block is 2N×2N, and the smallest CU size in the current block is N/4×N/4, according to the temporal correlation between two adjacent frames, regardless of the size of the corresponding CU block What kind of PU mode is it? It is impossible for the dividing line of the PU mode of the CU whose size is N/4×N/4 in the current block to correspond to the dividing line of the PU mode of the corresponding CU. That is, the current CU block size is not It is very likely that N/4×N/4 will appear. Therefore, for a CU of this size, its rate-distortion cost may not be detected to save the computational complexity of video coding.

Compared with the HEVC video coding standard, the method of the present invention can greatly reduce the computational complexity of video coding, and has little loss in video coding compression ratio and video quality. The fundamental basis of the video compression method is to achieve the information volume of the original whole video with less data by reducing various correlations in the video. The method of the present invention analyzes the time domain between two adjacent frames in the video. The correlation between the above PU modes, by judging the PU mode of the CU corresponding to the previous frame, and the correlation between adjacent CU blocks in the current block, the best PU mode of the current block is judged, thereby skipping other PU mode, from the perspective of the correlation of PU modes between adjacent frames, the method of the present invention removes the redundancy between PU modes, and reasonable redundancy removal can not only remove the required The video encoding calculation tasks performed will not lose the amount of video information at the same time, so the video compression ratio and video quality will basically not be reduced.

What the method of the invention improves is the most critical part of the calculation complexity in the whole video coding calculation process. In the entire video coding process, the calculation complexity of motion estimation (including integer pixel motion estimation and fractional pixel motion estimation) accounts for more than 50% (there are certain differences according to different configuration links), the most critical of the method of the present invention is According to the time-domain correlation between two adjacent frames, the detection process of the unlikely PU mode in each layer of CU blocks in the current block is skipped, that is, the rate-distortion cost is performed on each PU block in the PU mode The calculation and detection of the rate-distortion cost is selected to select a PU mode with the smallest rate-distortion cost. In the calculation of the rate-distortion cost, motion estimation is the most time-consuming encoding calculation process. Skipping several PU modes means skipping Therefore, in terms of computational complexity, the point where the method of the present invention starts is the most critical improvement in the video coding calculation process.

The method of the invention can reduce the computational complexity without additional hardware implementation cost. In many cases, video coding technology will eventually be embedded in hardware devices, including FPGA and DSP, etc. Therefore, the requirements for code operation cost and data storage hardware cost of the improved method are relatively high. The code that the method of the present invention needs to increase is seldom, mainly comprises several judgment sentences, aspect the memory required by hardware, because the object judged in the method of the present invention is the PU mode of the corresponding position CU block in the previous frame, and these modes The information of the present invention is originally stored in the data stream, and the method of the present invention does not bring additional data storage requirements. Therefore, if the method of the present invention is applied to hardware devices, no additional cost will be added to the manufacture of hardware devices, and at the same time it can Save power consumption.

Description of drawings

Figure 1 is a schematic diagram of the comparison between the HEVC fast inter-frame mode decision method based on time domain characteristics and the CU block method of the HM7.0 video coding standard, where (a) is the CU block method of the HM7.0 video coding standard, (b) ) is the CU block method in the HEVC fast inter-frame mode decision method based on temporal correlation;

Figure 2 is a schematic diagram of the comparison between the HEVC fast inter-frame mode decision method based on time domain characteristics and the PU prediction method of the HM7.0 video coding standard, where (a) is the PU prediction method of the HM7.0 video coding standard, and (b) is the PU prediction method of the HM7.0 video coding standard. PU prediction method in HEVC fast inter mode decision method based on temporal correlation;

Fig. 3 is a flow chart of HEVC fast inter-frame mode decision method based on temporal correlation.

Detailed ways

The present invention will be further described in detail below in conjunction with the examples. It must be pointed out that the following examples are only used to further illustrate the present invention and cannot be interpreted as limiting the protection scope of the present invention. Content, making some non-essential improvements and adjustments to the present invention for specific implementation shall still belong to the protection scope of the present invention.

1. Open the programs of the two algorithms at the same time and set the same configuration file. The reference software selects HM7.0, and the quantization step (QP) value is 27 and 32 respectively. The present invention will be compared with the method of the reference software algorithm HM7.0 of the HEVC video coding standard. And its three kinds of video coding performance: bit rate, peak signal-to-noise ratio (PSNR) and video coding time (PSNR reflects the objective video quality of the video, video coding time reflects the computational complexity of coding), comparative analysis, comparison The performance gap is evaluated with the following three indicators:

$\mathrm{<span\; class="notranslate">\; \&Delta;\; Bitrate</span>}<span\; class="notranslate">\; =</span>\frac{{\mathrm{<span\; class="notranslate">\; bit\; rate</span>}}_{\mathrm{<span\; class="notranslate">\; pro</span>}}<span\; class="notranslate">\; -</span>{\mathrm{<span\; class="notranslate">\; bit\; rate</span>}}_{\mathrm{<span\; class="notranslate">\; ref</span>}}}{{\mathrm{<span\; class="notranslate">\; bit\; rate</span>}}_{\mathrm{<span\; class="notranslate">\; ref</span>}}}<span\; class="notranslate">\; \&times;</span>\mathrm{<span\; class="notranslate">\; 100</span>}<span\; class="notranslate">\; \%</span>$

$\mathrm{<span\; class="notranslate">\; \&Delta;\; PSNR</span>}<span\; class="notranslate">\; =</span>{\mathrm{<span\; class="notranslate">\; PSNR</span>}}_{\mathrm{<span\; class="notranslate">\; pro</span>}}<span\; class="notranslate">\; -</span>{\mathrm{<span\; class="notranslate">\; PSNR</span>}}_{\mathrm{<span\; class="notranslate">\; ref</span>}}$

$\mathrm{<span\; class="notranslate">\; \&Delta;Time</span>}<span\; class="notranslate">\; =</span>\frac{{\mathrm{<span\; class="notranslate">\; Time</span>}}_{\mathrm{<span\; class="notranslate">\; pro</span>}}<span\; class="notranslate">\; -</span>{\mathrm{<span\; class="notranslate">\; Time</span>}}_{\mathrm{<span\; class="notranslate">\; ref</span>}}}{{\mathrm{<span\; class="notranslate">\; Time</span>}}_{\mathrm{<span\; class="notranslate">\; ref</span>}}}<span\; class="notranslate">\; \&times;</span>\mathrm{<span\; class="notranslate">\; 100</span>}<span\; class="notranslate">\; \%</span>$

Wherein Bitrate _pro , PSNR _pro and Time _pro are the bit rate, PSNR and video encoding time of the algorithm of the present invention respectively, Bitrate _ref , PSNR _ref and Time _ref are the bit rate, PSNR and video encoding time of the HM7.0 standard algorithm respectively, ? Bitrate, ?PSNR, and ?Time are the differences in bit rate, PSNR, and video encoding time between the algorithm of the present invention and the HM7.0 standard algorithm, respectively.

2. In the HEVC video coding technology, the PU prediction mode can adopt a symmetric and asymmetric comprehensive prediction mode, or only a symmetric prediction mode. The invention is effective in the above two cases, but the symmetric and asymmetric comprehensive prediction is adopted The mode can reduce more computational complexity of video coding, that is, better algorithm effect can be obtained, so the present invention adopts the comprehensive prediction mode.

3. The encoded objects are standard HEVC test videos. Their names, resolutions and frame rates are: Fourpeople (1280×720, 60 frames per second), Johnny (1280×720, 60 frames per second), KristenandSara ( 1280×720, 60 frames per second), Cactus (1920×1080, 50 frames per second), Kimono1 (1920×1080, 24 frames per second) and ParkScene (1920×1080, 24 frames per second).

4. Input 2 identical video sequences;

5. Perform video encoding on two identical video sequences;

6. Use HEVC video encoder HM7.0 to encode video sequences in HEVC mode;

7. The algorithm of the present invention selects the PU mode of the current CU block according to the PU mode of the corresponding block in the previous frame;

8. In the prediction mode selection, compare the sum of the total rate-distortion cost of the current depth CU with the sum of the total rate-distortion cost of the previous layer CU. If it is smaller than the upper layer, further divide the quadtree into 4 CUs with a lower level of depth, otherwise the quadtree division is terminated. The specific prediction mode selection is as follows:

(1) Detect the size of the CU block corresponding to the frame before the current CU block. If the size of the current CU block is smaller than the size of the corresponding CU block, enter the following step (2), otherwise traverse all PUs in the current CU block mode, and divide the quadtree into 4 deeper CU blocks, and repeat the above process for each CU block of the deeper layer;

(2) Determine whether the PU mode corresponding to the CU block in the previous frame is PART_2N×2N, if so, the current CU block only detects the rate-distortion cost of the PU mode of PART_2N×2N, and enters the following step (6), otherwise Enter the following step (3);

(3) Determine whether the PU mode corresponding to the CU block in the previous frame is PART_nL×2N or PART_nR×2N, and if so, the current CU block only detects the rate-distortion cost of the two PU modes of PART_N×2N and PART_2N×2N , and go to the following step (6), otherwise go to the following step (4);

(4) Determine whether the PU mode corresponding to the CU block in the previous frame is PART_2N×nU or PART_2N×nD, if so, the current CU block only detects the rate-distortion cost of the two PU modes of PART_2N×N and PART_2N×2N, And enter the following step (6), otherwise enter the following step (5);

(5) Detect the current CU block to detect the rate-distortion cost of all PU modes, and enter the following step (6);

(6) Determine whether the size of the current CU block is 1/4 of the size of the corresponding CU block in the previous frame. If so, the current CU block will not be divided into quadtrees; otherwise, the current CU block will be divided into quadtrees Divide into 4 CU blocks, and repeat step (1) for each CU block.

9. During the mode selection process, the rate-distortion cost formula is as follows:

J _mode =(SAD _luma +w _chroma ×SAD _chroma )+λ _mode ×B _mode

In the formula, J _mode is the rate-distortion cost, SAD _luma is the mean square difference between the original image brightness and the predicted image brightness, SAD _chroma is the mean square difference between the original image chromaticity and the predicted image chromaticity, w _chroma is the weight of chromaticity distortion, λ _mode represents the Lagrangian multiplier, and B _mode represents the number of encoded bits in this mode.

The distortion SAD _luma and SAD _chroma of brightness and chrominance can be obtained by the following two formulas respectively:

${\mathrm{<span\; class="notranslate">\; SAD</span>}}_{\mathrm{<span\; class="notranslate">\; luma</span>}}<span\; class="notranslate">\; =</span>\underset{\mathrm{<span\; class="notranslate">\; i</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; j</span>}}{\mathrm{<span\; class="notranslate">\; \&Sigma;</span>}}<span\; class="notranslate">\; |</span>{\mathrm{<span\; class="notranslate">\; Diff</span>}}_{\mathrm{<span\; class="notranslate">\; luma</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; i</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; j</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; |</span>$

${\mathrm{<span\; class="notranslate">\; SAD</span>}}_{\mathrm{<span\; class="notranslate">\; chroma</span>}}<span\; class="notranslate">\; =</span>\underset{\mathrm{<span\; class="notranslate">\; i</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; j</span>}}{\mathrm{<span\; class="notranslate">\; \&Sigma;</span>}}<span\; class="notranslate">\; |</span>{\mathrm{<span\; class="notranslate">\; Diff</span>}}_{\mathrm{<span\; class="notranslate">\; chroma</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; i</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; j</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; |</span>$

Among them, Diff _luma and Diff _chroma are:

Diff _luma (i,j)=BlockA _luma (i,j)-BlockAB _luma (i,j)

Diff _chroma (i,j)=BlockA _chroma (i,j)-BlockAB _chroma (i,j)

Among them, BlockA _luma and BlockB _luma are the pixel luminance values at the coordinate position (i, j) in the coding block and the prediction block respectively, and BlockA _chroma and BlockB _chroma are the coordinate positions in the coding block and the prediction block at (i, j) respectively. The pixel chromaticity value.

The chroma distortion weight w _chroma can be obtained by the following formula:

${\mathrm{<span\; class="notranslate">\; w</span>}}_{\mathrm{<span\; class="notranslate">\; chroma</span>}}<span\; class="notranslate">\; =</span>{\mathrm{<span\; class="notranslate">\; 2</span>}}^{<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; QP</span>}<span\; class="notranslate">\; -</span>{\mathrm{<span\; class="notranslate">\; QP</span>}}_{\mathrm{<span\; class="notranslate">\; chroma</span>}}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; /</span>\mathrm{<span\; class="notranslate">\; 3</span>}}$

Among them, QP and QP _chroma are the QP values of brightness and chroma, respectively.

The Lagrange multiplier λ _mode can be obtained by the following formula:

λ _mode =2 ^(QP-12)/3

10. In the algorithm of the present invention, the CU segmentation depth can be 2 to 4, but the larger the segmentation depth is, the more PU prediction modes will be skipped in the method, which will eventually reduce more video coding calculation complexity . Therefore, the CU partition depth of the present invention is 4.

11. The two programs respectively output video sequences after video encoding and their respective bit rates, PSNR values and total video encoding time. The results of the above three indicators are shown in Table 1-3. Statistics show that the algorithm of the present invention is better than the HEVC standard The bit rate has increased by 0.13-1.05%, and generally speaking, the large QP value is slightly higher than the small QP value, and the video quality PSNR value is reduced by 0.00-0.06dB. In the video encoding calculation The complexity is reduced by 22.38-58.36%. Generally speaking, compared with the HEVC video coding standard, the algorithm of the present invention greatly reduces the calculation of video coding under the premise that the video compression rate (reflected by the degree of bit rate reduction) and video quality loss is small. Complexity (see Table 1~3).

Table 1 Comparison of bit rates between the algorithm of the present invention and the HM7.0 standard algorithm

Table 2 Comparison of PSNR values between the algorithm of the present invention and the HM7.0 standard algorithm

Table 3 Comparison of video encoding time between the algorithm of the present invention and the HM7.0 standard algorithm

Claims (4)

1. the high-performance video coding inter-frame mode decision method based on relativity of time domain, comprise prediction mode configuration and predictive mode selection, in prediction mode configuration, coding unit CU is split the degree of depth and is not more than 4, predicting unit PU adopts symmetry and asymmetric integrated forecasting pattern or only adopts symmetrical predictive mode, in predictive mode is selected, rate total for current depth coding unit CU-distortion cost sum and the total rate-distortion cost sum of last layer coding unit CU are compared, if less than upper strata, quad-tree partition is then taked to become the coding unit CU of 4 more next layer depth further, otherwise termination quad-tree partition, it is characterized in that described predictive mode is selected to comprise the following steps:

(1) size of current coded unit CU piecemeal former frame correspondence position coding unit CU piecemeal is detected, if current coded unit CU block size is less than the size of corresponding coding unit CU piecemeal, then enter step (2) below, otherwise all predicting unit PU patterns of traversal current coded unit CU piecemeal, and take quad-tree partition to become 4 darker one deck coding unit CU piecemeals, said process is repeated to more further each coding unit CU piecemeal;

(2) whether the predicting unit PU pattern judging former frame corresponding coding unit CU piecemeal is PART_2N × 2N, if, then current coded unit CU piecemeal only detects the rate-distortion cost of the predicting unit PU pattern of PART_2N × 2N, and enter step (6) below, otherwise enter step (3) below;

(3) whether the predicting unit PU pattern judging former frame corresponding coding unit CU piecemeal is PART_nL × 2N or PART_nR × 2N, if, then current coded unit CU piecemeal only detects the rate-distortion cost of PART_N × 2N and PART_2N × 2N two kinds of predicting unit PU patterns, and enter step (6) below, otherwise enter step (4) below;

(4) whether the predicting unit PU pattern judging former frame corresponding coding unit CU piecemeal is PART_2N × nU or PART_2N × nD, if, then current coded unit CU piecemeal only detects the rate-distortion cost of PART_2N × N and PART_2N × 2N two kinds of predicting unit PU patterns, and enter step (6) below, otherwise enter step (5) below;

(5) detect rate-distortion cost that current coded unit CU piecemeal detects all predicting unit PU patterns, and enter step (6) below;

(6) judge that whether the size of current coded unit CU piecemeal is 1/4 of former frame corresponding coding unit CU block size, if so, then current coded unit CU piecemeal no longer carries out quad-tree partition; Otherwise taked by current coded unit CU piecemeal quad-tree partition to become 4 darker one deck coding unit CU piecemeals further, step (1) process is repeated to more further each coding unit CU piecemeal.

2., as claimed in claim 1 based on the high-performance video coding inter-frame mode decision method of relativity of time domain, it is characterized in that coding unit CU splits the degree of depth is 2 ~ 4.

3., as claimed in claim 2 based on the high-performance video coding inter-frame mode decision method of relativity of time domain, it is characterized in that coding unit CU splits the degree of depth is 4.

4. the high-performance video coding inter-frame mode decision method based on relativity of time domain as described in one of claims 1 to 3, is characterized in that described rate-distortion cost is determined by following formula:

J _mode＝(SAD _luma+w _chroma×SAD _chroma)+λ _mode×B _mode

J in formula _modefor rate-distortion cost, SAD _lumafor the mean square deviation of original image brightness and predicted picture brightness, SAD _chromafor the mean square deviation of original image colourity and predicted picture colourity, w _chromafor the weights of chromatic distortion, λ _moderepresent Lagrange multiplier, B _moderepresent number of coded bits in this mode.