CN104333756A - HEVC (High Efficiency Video Coding) prediction mode fast selection method based on time domain correlation - Google Patents

Abstract

The invention discloses a method for quickly selecting an HEVC prediction mode based on time-domain correlation, which mainly solves the problem of high computational complexity of prediction mode selection in the high-efficiency video coding standard HEVC inter-frame coding process. The implementation steps are: (1) Divide the motion intensity of the video sequence into three states: slow motion, moderate motion, and fast motion; (2) Statistically calculate the best prediction of the coding unit under the two motion intensities of slow motion and moderate motion (3) According to the probability relationship, construct a candidate prediction mode table; (4) According to the candidate prediction mode table, encode the inter-frame coding unit to obtain the best predictive mode. On the premise that the video rate-distortion performance is approximately constant, the present invention effectively reduces the computational complexity of prediction mode selection, improves the speed of prediction mode selection, reduces encoding time, and can be used for real-time video compression.

Description

A Fast Selection Method of HEVC Prediction Mode Based on Temporal Correlation

technical field

The invention belongs to the field of video processing, in particular to the generation of candidate prediction modes of inter-frame coding units in high-performance video coding HEVC, which can be used in the HEVC inter-frame coding process.

Background technique

At present, basically all high-quality multimedia video applications, such as digital video, video on demand and DVD, are still using the advanced video coding standard H.264/AVC formulated in 2003. However, with the rapid development of video applications, various smart terminals, such as smartphones and tablet computers, have become tools for mass entertainment, and the demand for ultra-high-definition video is also growing. People's demand for high-quality, high-resolution video Higher and higher. In order to meet the ever-increasing demands of video applications, a new generation of video coding standards has emerged. In 2010, JCT-VC, a video coding joint group, began to develop a new generation of high-efficiency video coding standard HEVC, and officially completed it in February 2013. HEVC aims to double the compression efficiency while ensuring the subjective quality of the video remains unchanged, greatly reducing the bandwidth of the transmitted video signal.

The HEVC coding framework still follows the hybrid coding framework of the H.26x series standard, using the classic block-based hybrid coding model, joint motion compensation prediction, transform coding and high-efficiency entropy coding. Compared with the previous video coding standards, the HEVC coding unit adopts a flexible quadtree structure, ranging in size from 64×64 to 8×8, which can efficiently encode, predict and transform large-size image blocks. The experimental configuration conditions of the HEVC standard include three types: full I-frame configuration, low-latency configuration, and random access configuration. During random access configuration, hierarchical B frames are used, and there are 4 time domain layers in total.

Prediction mode selection is a key technology of HEVC, and its purpose is to select the best prediction mode from multiple candidate prediction modes for each inter coding unit to obtain the best prediction accuracy. HEVC uses a rate-distortion optimization model to select the best prediction mode.

The prediction mode selection method of the inter coding unit in the HEVC standard should first select SKIP, Inter 2N×2N, Inter 2N×N, Inter N×2N, Inter 2N×nU, Inter 2N×nD, Inter nL×2N, Inter nR× 2N, Intra 2N×2N, Intra N×N these 10 candidate prediction modes are traversed, and then the prediction mode with the smallest rate-distortion cost is taken as the best prediction mode of the current coding unit. The process of selecting the best prediction mode makes the complexity of HEVC coding rise sharply, which brings great difficulties to the real-time application of HEVC. Therefore, it is necessary to reduce HEVC coding complexity under the premise of ensuring that the coding performance is basically unchanged.

So far, the HEVC prediction mode fast selection methods that have been proposed are as follows:

The proposal JCTVC-F045 "Early termination of CU encoding to reduce HEVC complexity" simplifies the judgment of the prediction mode by the value of the coding block flag CBF. This method proposes that when the coded block flag CBF=0 of other inter-frame prediction modes except Inter N×N, the selection process of the remaining prediction modes will be terminated and no calculation will be performed, thereby reducing the coding complexity. This method is called coded block flag fast mode CFM method.

The proposal JCTVC-G543 "Early SKIP Detection for HEVC" proposes to search for the Inter 2N×2N mode before detecting the SKIP mode, and then detect the motion vector difference DMV and the coding block flag CBF of the Inter 2N×2N mode. If the DMV is equal to (0, 0), and the CBF is equal to 0, then the best prediction mode of the current coding unit is set to SKIP mode, and the remaining prediction modes are no longer traversed, thereby greatly reducing the coding complexity. This method is called the early SKIP detection ESD method.

In the paper "Novel Fast PU Decision Algorithm for the HEVC VideoStandard", Jong-Hyeok Lee et al. proposed a fast method based on temporal-spatial domain and encoding depth correlation. This method uses the motion complexity to divide the coding unit into different motion areas: if the motion complexity of the current coding unit is less than the threshold Th1, it is judged that the current coding unit is a slow motion area, and only the SKIP and Inter 2N×2N modes are traversed; if the current If the motion complexity of the coding unit is greater than the threshold Th1 and less than the threshold Th2, then it is judged that the current coding unit is a moderate motion area, and the SKIP, Inter 2N×2N, Inter 2N×N and Inter N×2N modes are traversed, otherwise the current coding unit is judged to be In the fast-moving area, all 10 prediction modes are traversed, thus skipping redundant prediction modes and effectively reducing the coding complexity.

The above methods all reduce the coding complexity of HEVC to a certain extent. However, none of these methods have conducted a detailed analysis of temporal correlation, so that the speed of HEVC prediction mode selection has not been further improved.

Contents of the invention

The purpose of the present invention is to address the deficiencies of the above-mentioned existing technologies, and propose a method for quickly selecting HEVC prediction modes based on time-domain correlation, so as to further reduce coding complexity and improve prediction while ensuring that the video compression performance is basically unchanged. Mode selection speed.

The technical idea of realizing the object of the present invention is: in the prediction mode selection process, according to the best prediction mode of the adjacent coding unit in the time domain, the candidate prediction mode of the current coding unit is given, and as many redundant prediction modes as possible are skipped, and then Select the best prediction mode from the candidate prediction modes to improve the encoding speed, and its implementation steps include the following:

(1) The motion intensity of the video sequence is divided into three states: slow motion, moderate motion, and fast motion;

(2) Determine the correlation between the best prediction mode of a CU and the best prediction modes of its neighboring CUs in the temporal domain:

2.1) Input the video sequence and the experimental configuration conditions, starting from the second non-I frame, for each inter-frame coding unit, use the number of this coding unit to search for the coding unit at the same position in the reference frame, and get the temporal adjacent coding unit;

2.2) Determine the category of experimental configuration conditions:

If the experimental configuration condition is low latency, for the four different quantization parameter ranges [20,26], [27,31], [32,36], [37,41], the statistics of slow motion and moderate motion The probability relationship between the best prediction mode of the coding unit of these two motion intensities and the best prediction mode of its neighboring coding units in the temporal domain;

If the experimental configuration condition is random access, in the four time domain layers of 1, 2, 3, and 4, [20,26], [27,31], [32,36], [37, 41] For these four different quantization parameter ranges, the probability relationship between the best prediction mode of the slow-moving coding unit and the moderate-moving coding unit and the best prediction mode of the adjacent coding unit in the time domain is counted;

(3) According to the probability relationship between the best prediction mode of the coding unit and the best prediction mode of the adjacent coding unit in the time domain under the conditions of experimental configuration conditions, quantization parameter range, motion intensity and temporal layering, construct low-latency configuration candidates Prediction mode table and random access configuration candidate prediction mode table:

3.1) The best prediction modes for selecting adjacent coding units in time domain are SKIP, Inter 2N×2N, Inter 2N×N, InterN×2N, Inter 2N×nU, Inter 2N×nD, Inter nL×2N, Inter nR× For each of the 10 prediction modes 2N, Intra 2N×2N, and Intra N×N, calculate the probability that the best prediction mode of the current coding unit is these 10 prediction modes, and sort them in descending order of probability;

3.2) Starting from the prediction mode with the highest probability, select the prediction mode with the sum of the probability not less than 90% and the least number as the candidate prediction mode, and set the total number of candidate prediction modes to no more than 5, so as to construct the candidate prediction mode table;

(4) According to the candidate prediction mode table, select the candidate prediction mode, and encode the inter coding unit to obtain the best prediction mode:

4.1) Input the video sequence and experimental configuration conditions, and use the HEVC standard prediction mode selection method to encode the I frame and the first non-I frame coding unit to obtain the best prediction mode for each coding unit;

4.2) Starting from the second non-I frame, for each inter-frame coding unit, determine the category of its motion intensity:

If the motion intensity of the current coding unit is slow motion, perform step 4.3); if the motion intensity of the current coding unit is moderate motion, perform step 4.4); otherwise, use the HEVC standard prediction mode selection method to encode the coding unit to obtain the most best predictive model;

4.3) The category of experimental configuration conditions for judging slow movement:

If the experimental configuration condition is low delay, according to the quantization parameter QP and the best prediction mode of the adjacent coding unit in the time domain, select the corresponding candidate prediction mode from the candidate prediction mode table of the slow moving coding unit under the low delay configuration. traverse to get the best prediction mode;

If the experimental configuration condition is random access, according to the time-domain layering, quantization parameter QP and the best prediction mode of the neighboring coding units in the time domain, select the corresponding The candidate prediction modes are traversed to get the best prediction mode;

4.4) The category of experimental configuration conditions for judging moderate movement:

If the experimental configuration condition is low delay, according to the quantization parameter QP and the best prediction mode of the adjacent coding unit in the time domain, select the corresponding candidate prediction mode from the candidate prediction mode table of the moderately moving coding unit under the low delay configuration. traverse to get the best prediction mode;

If the experimental configuration condition is random access, according to the time-domain layering, quantization parameter QP and the best prediction mode of the adjacent coding units in the time domain, select the corresponding prediction mode from the candidate prediction mode table of the moderately moving coding unit under the random access configuration. The candidate prediction modes are traversed to get the best prediction mode.

Compared with existing methods, the present invention has the following advantages:

(a) The present invention skips the redundant prediction mode according to the time domain correlation, so that the calculation complexity of the prediction mode selection is small, and a large amount of coding time is reduced;

(b) In the present invention, when selecting the candidate prediction mode, the selection is made according to different motion intensities, so that the mode selection result is more accurate.

Description of drawings

Fig. 1 is the realization flowchart of the present invention;

Fig. 2 is a schematic diagram of the positions of surrounding coding unit sets of the current coding unit.

detailed description

The present invention will be further described in detail with reference to the accompanying drawings and embodiments. This embodiment is carried out on the premise of the technical solution of the present invention, and detailed implementation and specific operation process are given, but the protection scope of the present invention is not limited to the following embodiments.

With reference to Fig. 1, the realization steps of the present invention are as follows:

Step 1: Classify the motion intensity of the video sequence.

1a) Establish a motion vector set {mv ₁ , mv ₂ , mv ₃ , mv ₄ , mv ₅ } for the current coding unit CU ₀ ,

Where mv _i is the motion vector of the coding unit CU _i around the current coding unit CU ₀ , mv _i = ( _xi , y _i ), i=1,...,5, _xi and y _i are respectively the transverse direction of the motion vector coordinates and ordinates; the set of coding units around CU ₀ includes its left neighboring coding unit CU ₁ , upper neighboring coding unit CU ₂ , upper right neighboring coding unit CU ₃ , upper left neighboring coding unit CU ₄ and time domain phase Adjacent coding unit CU ₅ , as shown in FIG. 2 ;

1b) Calculate the length l of each motion vector in the motion vector set:

l(mv _i )＝|x _i |+|y _i |;

1c) select the maximum value L of length in the motion vector length collection that step 1b) obtains:

L=max{l(mv ₁ ), l(mv ₂ ), l(mv ₃ ), l(mv ₄ ), l(mv ₅ )},

1d) Set slow motion threshold L0 and moderate motion threshold L1:

1d1) Input video sequences with a resolution of 416×240 and experimental configuration conditions;

1d2) Starting from the first non-I frame, judge the category of the experimental configuration condition:

If the experimental configuration condition is low delay, then output the motion vector of the video sequence, according to the size of the motion vector and the video resolution, and refer to the threshold T0 of the slow motion state and the threshold T1 of the moderate motion state in the Zeng method, set the threshold L0 and L1;

If the experimental configuration condition is random access, output the motion vectors of different time-domain layers of the video sequence, according to the size of the motion vector and video resolution, and refer to the threshold T0 of the slow motion state and the threshold T1 of the moderate motion state in the Zeng method , setting the thresholds L0 and L1 of different time domain layers;

For the Zeng method, see the article "A Novel Fast ModeDecision for the H.264/AVC Based on Local MacroblockMotion Activity" by Huanqiang Zeng, Canhui Cai et al.;

1d3) Repeat step 1d1) and step 1d2) to obtain the slow motion threshold L0 and moderate motion threshold L1 of video sequences with resolutions of 832×480, 1280×720, 1920×1080, and 2560×1600, respectively, to obtain the final videos with different resolutions Sequential slow movement threshold L0 and moderate movement threshold L1, as shown in Table 1 and Table 2:

Table 1 Low latency configuration slow motion state threshold L0 and moderate motion state threshold L1

video sequence resolution Threshold L0 Threshold L1 416×240 2 8 832×480, 1280×720 3 12 1920×1080 8 32

Table 2 Random access configuration slow state threshold L0 and moderate motion state threshold L1

1e) According to the experimental configuration conditions, video resolution and time domain layering, select the slow motion threshold L0 and the moderate motion threshold L1 from Table 1 or Table 2, and combine the maximum length L obtained in step 1c) with the threshold L0 and Threshold L1 for comparison:

If L<L0, define the exercise intensity as slow exercise;

If L0≤L<L1, the exercise intensity is defined as moderate exercise;

If L≥L1, the exercise intensity is defined as fast exercise.

Step 2: Determine the correlation between the best prediction mode of the CU and the best prediction modes of its neighboring CUs in time domain.

2a) Input the video sequence BQSquare and the experimental configuration conditions, starting from the second non-I frame, for each inter-frame coding unit, use the number of this coding unit to search for the coding unit at the same position in the reference frame, and get the temporal phase Adjacent coding unit;

2b) Determine the category of experimental configuration conditions:

If the experimental configuration condition is low latency, test the first 100 frames of the video sequence BQSquare, for the four different quantization parameters [20,26], [27,31], [32,36], [37,41] Range, respectively statistics the probability relationship between the best prediction mode of the coding unit with slow motion and moderate motion intensity and the best prediction mode of the adjacent coding unit in time domain;

If the experimental configuration condition is random access, test the first 65 frames of the video sequence BQSquare, in the four time domain layers of 1, 2, 3, and 4, respectively [20,26], [27,31], [32,36], [37,41] these four different quantization parameter ranges, statistically the best prediction mode of the coding unit with slow motion and moderate motion intensity and the best prediction of the temporal adjacent coding unit the probability relationship of the pattern;

2c) Repeat step 2a) and step 2b) respectively to obtain the probability relationship between the best prediction mode of the coding unit of the video sequence BasketballPass and Johnny and the best prediction mode of the adjacent coding unit in the time domain, and the probability of the above three video sequences The relationship is averaged to obtain the final low-latency configuration and random access configuration. The probability relationship between coding units with different motion intensities and the best prediction modes of adjacent coding units in the time domain is given below. Only the quantization parameters for low-latency configurations are given below. The probability relationship when the range is [20,26], and the probability relationship when the time domain layer is 1 and the quantization parameter range is [20,26] under the random access configuration, as shown in Table 3 to Table 6:

Table 3 Probability relationship between low-latency configuration slow-moving coding units and the best prediction modes of neighboring coding units in time domain

Table 4 Probability relationship between low-latency configuration moderate motion coding unit and the best prediction mode of adjacent coding units in time domain

Table 5 Probability relationship between random access configuration slow motion coding unit and the best prediction mode of adjacent coding units in time domain

Table 6 Probability relationship between the optimal prediction mode of the moderately moving coding unit and the adjacent coding unit in the time domain with random access configuration

In the above Table 3-Table 6, mode 0 represents SKIP, mode 1 represents Inter 2N×2N, mode 2 represents Inter 2N×N, mode 3 represents Inter N×2N, mode 4 represents Inter 2N×nU, and mode 5 represents Inter 2N ×nD, mode 6 represents InternL×2N, mode 7 represents Intra nR×2N, mode 8 represents Intra 2N×2N, and mode 9 represents Intra N×N.

Step 3: According to Table 3-Table 6, construct a low-latency configuration candidate prediction mode table and a random access configuration candidate prediction mode table.

3a) The best prediction modes for a coding unit and its neighboring coding units in time domain include: SKIP, Inter 2N×2N, Inter2N×N, Inter N×2N, Inter 2N×nU, Inter 2N×nD, Inter nL×2N, Inter There are 10 prediction modes: nR×2N, Intra 2N×2N, and IntraN×N; select the best prediction mode of adjacent coding units in the time domain as SKIP mode, and calculate the best prediction modes of the current coding unit as the above 10 prediction modes Probability of , and sorted by probability from large to small;

3b) Repeat step 3a) to select the other 9 prediction modes for the best prediction mode of adjacent coding units in the time domain, and calculate the probability that the best prediction mode of the current coding unit is the above 10 prediction modes, and rank the probability from large to small sort;

3c) Starting from the prediction mode with the highest probability, select the prediction mode with the sum of the probability not less than 90% and the least number as the candidate prediction mode, and set the total number of candidate prediction modes to no more than 5, so as to construct the candidate prediction mode table, as shown in the table 7-Table 10;

Table 7 Candidate prediction mode table of low-latency configuration and slow-moving coding unit

Table 8 Low-latency configuration and moderate motion coding unit candidate prediction mode table

Table 9 Table of candidate prediction modes for random access configuration slow motion coding unit

Table 10 Random access configuration motion moderate coding unit candidate prediction mode table

Step 4: According to Table 7-Table 10, select candidate prediction modes, encode the inter-coding units of each video sequence in Table 11, and obtain the best prediction mode.

4a) Input a video sequence, use the prediction mode selection method of the HEVC standard to encode the coding unit of the I frame and the first non-I frame, and obtain the best prediction mode of each coding unit;

4b) Starting from the second non-I frame, for each inter-frame coding unit, judge its motion strength:

If the motion intensity of the current coding unit is slow motion, perform step 4c); if the motion intensity of the current coding unit is moderate motion, perform step 4d); otherwise, use the HEVC standard prediction mode selection method to encode the coding unit to obtain the most best predictive model;

4c) The category of experimental configuration conditions for judging slow movement:

If the experimental configuration condition is low delay, then according to the quantization parameter QP and the best prediction mode of the adjacent coding unit in the time domain, select the corresponding candidate prediction mode in Table 7 for traversal, and obtain the best prediction mode;

If the experimental configuration condition is random access, then according to the time domain layering, quantization parameter QP and the best prediction mode of the adjacent coding unit in the time domain, select the corresponding candidate prediction mode in Table 9 and traverse to obtain the best prediction mode ;

4d) The category of experimental configuration conditions for judging moderate movement:

If the experimental configuration condition is low delay, then according to the quantization parameter QP and the best prediction mode of the adjacent coding unit in the time domain, select the corresponding candidate prediction mode in Table 8 for traversal, and obtain the best prediction mode;

If the experimental configuration condition is random access, according to the time-domain layering, quantization parameter QP and the best prediction mode of the adjacent coding unit in the time domain, select the corresponding candidate prediction mode in Table 10 and traverse to obtain the best prediction mode .

Effect of the present invention can be further illustrated by following simulation:

1. Experimental environment

Use the VS2010 coding environment and test with the reference software HM16.0. The experimental configuration conditions are low-latency configuration and random access configuration.

The video sequence details of the experimental test are shown in Table 11:

Table 11 Video sequence details

2. Experimental content

All the video sequences in Table 11 are encoded using the method of the present invention, the fast method of CFM, the fast method of ESD and the fast method of Lee, and record the coding time and rate-distortion performance estimator BD-PSNR. The coding performance comparison of the present invention and these three fast methods is as table 12-table 14, and wherein table 12 is the comparison of the coding performance of the inventive method and the CFM fast method, and table 13 is the comparison of the coding performance of the inventive method and the ESD fast method, Table 14 is the comparison of the encoding performance of the method of the present invention and the Lee fast method.

Among them, in Table 12-Table 14 Indicates the amount of time change between the method of the present invention and the existing rapid selection method, and "-" indicates that the method of the present invention is faster than the existing rapid method in terms of time. BD-PSNR represents the difference between the luminance peak signal-to-noise ratio PSNR-Y of the two methods at the same given code rate, and its unit is dB. "-" indicates that the method of the present invention is lower than the existing fast method PSNR-Y .

Table 12 The method of the present invention compares with the fast method of CFM

It can be seen from Table 12 that, compared with the CFM fast method, the method of the present invention has an average reduction of 0.07dB in BD-PSNR in low-latency configuration, and the average encoding time is increased by 4.69%. In the case of a 0.07dB reduction, the encoding time is increased by an average of 8.43%.

Table 13 The inventive method compares with ESD fast method

It can be seen from Table 13 that, compared with the ESD fast method, the method of the present invention has an average reduction of 0.095dB in BD-PSNR in low-delay configuration, and the average encoding time is increased by 13.81%. In the case of a reduction of 0.10dB, the encoding time is increased by an average of 11.68%.

Table 14 The inventive method compares with Lee's fast method

It can be seen from Table 14 that compared with the Lee fast method, the method of the present invention speeds up the encoding time by 4.22% on average when the BD-PSNR is reduced by 0.04dB on average in random access configuration.

To sum up, the present invention utilizes time-domain correlation to skip redundant prediction modes, and further improves the speed of prediction mode selection under the condition that the BD-PSNR is basically the same.

The foregoing description is a preferred example of the present invention, obviously researchers in the field can make various modifications and replacements to the present invention with reference to the preferred examples of the present invention and accompanying drawings, and these modifications and replacements all should fall within the protection scope of the present invention .

Claims (2)

1., based on a HEVC predictive mode fast selecting method for relativity of time domain, it is characterized in that, comprise the steps:

(1) exercise intensity of video sequence is divided into: motion is slow, moderate, motion these three kinds of states fast of moving;

(2) correlation of the optimum prediction mode of coding unit and the optimum prediction mode of its time domain adjacent encoder unit is determined:

2.1) input video sequence and experimental configuration condition, from the 2nd non-I frame, to each interframe encode unit, uses the numbering of this coding unit in reference frame, search for the coding unit with its same position, obtains time domain adjacent encoder unit;

2.2) classification of judgment experiment configuration condition:

If experimental configuration condition is low time delay, then to [20,26], [27,31], [32,36], [37,41] these four kinds different quantization parameter scopes, respectively statistics motion slowly, the optimum prediction mode of coding unit of these the two kinds of exercise intensities moderate that move and the probabilistic relation of the optimum prediction mode of its time domain adjacent encoder unit;

If experimental configuration condition is Stochastic accessing, then in 1,2,3,4 these four time domain layerings, respectively to [20,26], [27,31], [32,36], the different quantization parameter scope in [37,41] these four kinds, statistics motion slowly, the optimum prediction mode of coding unit of these the two kinds of exercise intensities moderate that move and the probabilistic relation of the optimum prediction mode of its time domain adjacent encoder unit;

(3) the experimentally optimum prediction mode of coding unit and the probabilistic relation of its time domain adjacent encoder unit optimum prediction mode in configuration condition, quantization parameter scope, exercise intensity and these situations of time domain layering, builds low time delay configuration candidate modes table and Stochastic accessing configuration candidate modes table:

3.1) optimum prediction mode selecting time domain adjacent encoder unit is respectively SKIP, Inter 2N × 2N, Inter2N × N, Inter N × 2N, Inter 2N × nU, Inter 2N × nD, Inter nL × 2N, Inter nR × 2N, Intra2N × 2N, each in these 10 kinds of predictive modes of Intra N × N, the optimum prediction mode calculating current coded unit is the probability of these 10 kinds of predictive modes, and sorts from big to small by probability;

3.2) from the predictive mode of maximum probability, select probability sum is not less than 90% and the predictive mode of minimum number alternatively predictive mode, and sets candidate modes sum and be no more than 5 kinds, thus builds candidate modes table;

(4) according to candidate modes table, choose candidate modes, and interframe encode unit encoded, obtain optimum prediction mode:

4.1) input video sequence and experimental configuration condition, adopts the coding unit of predicting mode selecting method to I frame and the 1st non-I frame of HEVC standard to encode, obtains the optimum prediction mode of each coding unit;

4.2) from the 2nd non-I frame, to each interframe encode unit, the classification of its exercise intensity is judged:

If the exercise intensity of current coded unit, for moving slowly, performs step 4.3); If the exercise intensity of current coded unit is moderate for moving, perform step 4.4); Otherwise, adopt the predicting mode selecting method of HEVC standard to encode to coding unit, obtain optimum prediction mode;

4.3) judge to move experimental configuration condition classification slowly:

If experimental configuration condition is low time delay, then according to the optimum prediction mode of quantization parameter QP and time domain adjacent encoder unit, low time delay configuration under move slow coding unit candidate modes table in select corresponding candidate modes to travel through, obtain optimum prediction mode;

If experimental configuration condition is Stochastic accessing, then according to the optimum prediction mode of time domain layering, quantization parameter QP and time domain adjacent encoder unit, Stochastic accessing configuration under move slow coding unit candidate modes table in select corresponding candidate modes to travel through, obtain optimum prediction mode;

4.4) the experimental configuration condition classification that motion is moderate is judged:

If experimental configuration condition is low time delay, then according to the optimum prediction mode of quantization parameter QP and time domain adjacent encoder unit, low time delay configuration under move moderate coding unit candidate modes table in select corresponding candidate modes to travel through, obtain optimum prediction mode;

If experimental configuration condition is Stochastic accessing, then according to the optimum prediction mode of time domain layering, quantization parameter QP and time domain adjacent encoder unit, Stochastic accessing configuration under move moderate coding unit candidate modes table in select corresponding candidate modes to travel through, obtain optimum prediction mode.

2. as claimed in claim 1 based on the HEVC predictive mode fast selecting method of relativity of time domain, it is characterized in that: the exercise intensity of video sequence is divided in (1) by described step: motion is slow, moderate, motion these three kinds of states fast of moving, and carries out as follows:

1a) to current coded unit CU ₀set up motion vector set { mv ₁, mv ₂, mv ₃, mv ₄, mv ₅,

Wherein mv _icurrent coded unit CU ₀coding unit CU around _imotion vector, mv _i=(x _i, y _i), i=1 ..., 5, x _iand y _ibe respectively abscissa and the ordinate of motion vector; CU ₀coding unit around comprises its left side adjacent encoder unit CU ₁, top adjacent encoder unit CU ₂, top right-hand side adjacent encoder unit CU ₃, limit, upper left adjacent encoder unit CU ₄with time domain adjacent encoder unit CU ₅;

1b) the length l of each motion vector in calculating kinematical vector set:

l(mv _i)＝|x _i|+|y _i|，

1c) in step 1b) select the maximum L of length in the motion vector length set that obtains:

L＝max{l(mv ₁),l(mv ₂),l(mv ₃),l(mv ₄),l(mv ₅)}，

1d) by step 1c) the length maximum L that obtains compares with the slow threshold value L0 of motion and the moderate threshold value L1 that moves:

If L < is L0, then define exercise intensity for moving slowly;

If L0≤L < is L1, then define exercise intensity moderate for moving;

If L >=L1, then define exercise intensity motion fast.

Wherein, move slow threshold value L0 and the moderate threshold value L1 of motion, according to size and the video resolution of motion vector, and determines with reference to Zeng method.