CN105120291B

CN105120291B - A kind of adaptive Fast video coding method based on variance

Info

Publication number: CN105120291B
Application number: CN201510483491.4A
Authority: CN
Inventors: 梁凡; 谢立威
Original assignee: Sun Yat Sen University
Current assignee: Sun Yat Sen University
Priority date: 2015-08-07
Filing date: 2015-08-07
Publication date: 2018-04-10
Anticipated expiration: 2035-08-07
Also published as: CN105120291A

Abstract

The invention discloses a kind of adaptive Fast video coding method based on variance, including：Calculate present frame CU variance；PU partition mode detections are carried out using the quick decision algorithm of PU partition modes based on variance according to present frame CU variance, obtain the optimal rate distortion costs that the minimum PU partition modes of rate distortion costs and CU are not divided；CU division judgements are carried out using the quick decision algorithm of CU partition modes based on variance according to present frame CU variance, determine the minimum CU partition modes of rate distortion costs；Encoded according to the minimum PU partition modes of rate distortion costs and CU partition modes.It is combined present invention employs the quick decision algorithm of PU partition modes based on variance and the quick decision algorithm of CU partition modes based on variance to carry out the selection of CU and PU partition mode, greatly reduce encoder complexity, coding rate is improved, can be widely applied to field of video encoding.

Description

A kind of adaptive Fast video coding method based on variance

Technical field

The present invention relates to field of video encoding, especially a kind of adaptive Fast video coding method based on variance.

Background technology

HEVC (High Efficiency Video Coding, high-performance video coding) is the VCEG by ITU-T (Video Coding Experts Group) and ISO/IEC MPEG (Moving Picture Experts Group) exist The Video coding joint group JCT-VC (Joint Collaborative Team on Video Coding) that in January, 2010 sets up The video encoding standard of new generation formulated.In January, 2013, HEVC formally turns into newest video coding international standard.

On the whole, there is more outstanding coding efficiency compared to H.264/AVC video encoding standard, HEVC standard, Can be in the case where keeping equal well as subjective video quality, the code check of average saving about 50%；But HEVC volume solution is complicated Degree is also higher by much than encoding and decoding complexity H.264/AVC.Therefore, HEVC standard is led applied to various multimedia application Domain, cost can be reduced, but there is also very big challenge, particularly in terms of real-time application and mobile video.

HEVC has still continued to use H.264/AVC block-based hybrid coding frame as generation standard H.264/AVC Frame, difference are that HEVC for predictive coding, transition coding, quantization encoding and entropy code careful effectively change Enter, while introduce new video coding technique, as shown in Figure 1.In an encoding process, input picture can be divided into one to HEVC The coding unit of series, for each coding unit, encoder can travel through all possible coding mode combination, during which can carry out Prediction, change quantization and entropy code, and corresponding process of reconstruction, until selecting the minimum coding mode group of rate distortion costs Close, this process is to be divided by carrying out complicated recurrence to coding unit to complete, that is, four points of trees (Quad-tree) Cataloged procedure.In addition, after the completion of the coding per two field picture, its reconstruction image can pass through block elimination filtering and adaptive sample bias SAO (Sample Adaptive Offset), so as to further remove the distortion of reconstruction image.

But point tree-encoding structure of new coding techniques, particularly four is used, the coding for substantially increasing HEVC is complicated Degree.Match with the coding structure of four points of trees, HEVC introduces many new concepts, including code tree unit CTU (Coding Tree Unit), coding unit CU (Coding Unit), predicting unit PU (Prediction Unit) and converter unit TU (Transform Unit).Wherein, CTU tetra- divides the main encoding process that tree-encoding process is HEVC encoders, essentially all of HEVC coding modes can all be established in this process.

When carrying out four points of tree-encodings, HEVC encoders need to travel through the node that is possible to that coding unit four divides tree, and For each node, it is necessary to travel through all possible PU partition modes, as shown in Figure 2.And divide mould for every kind of possible PU Formula, it is also necessary to travel through all nodes of RQT (Residue Quadtree, square residual error quad-tree structure), whole ergodic process Carried out in the way of depth-first, generally require to realize by the recursive call of function.According to common encoding setting, Coding unit four divides the node number up to 85 of tree, and each node needs up to 8 kinds of the PU partition modes detected, and RQT Node number can be with up to 21, therefore coding unit four divides tree, PU partition modes and RQT to form numerous coding mode groups Close.Therefore will be during four points of tree-encodings, according to RDO (Rate Distortion Optimization, rate-distortion optimization Block) optimal coding mode combination is selected, the plenty of time can be consumed so that meeting during by HEVC for application in real time and mobile terminal Run into great challenge.

In summary, a kind of reduce on the premise of video compression efficiency and coding quality is kept substantially is needed badly in the industry to encode Complexity and the method for video coding for improving coding rate.

The content of the invention

In order to solve the above-mentioned technical problem, the purpose of the present invention is：It is low fast with coding rate to provide a kind of encoder complexity , the adaptive Fast video coding method based on variance.

The technical solution adopted for the present invention to solve the technical problems is：

A kind of adaptive Fast video coding method based on variance, including：

A, present frame CU variance is calculated；

B, PU division moulds are carried out using the quick decision algorithm of PU partition modes based on variance according to present frame CU variance Formula detects, and obtains the optimal rate distortion costs that the minimum PU partition modes of rate distortion costs and CU are not divided；

C, CU divisions are carried out using the quick decision algorithm of CU partition modes based on variance according to present frame CU variance to sentence Certainly, the minimum CU partition modes of rate distortion costs are determined；

D, encoded according to the minimum PU partition modes of rate distortion costs and CU partition modes.

Further, the step A, it is specially：

Present frame CU variance is calculated, the calculation formula of the variance of the CU is：

Wherein, the size of CU corresponding regions is N × M, { p_ij| 0≤i≤N-1,0≤j≤M-1, i ∈ Z, j ∈ Z } represent CU All original pixel values in corresponding region,It is the average of all original pixels in CU corresponding regions, and v is CU corresponding regions The variance of interior original pixels.

Further, the step B, it includes：

B1, judge present frame coding configuration whether be AI or AI10 coding configuration, if so, then perform step B2, conversely, Then perform step B3；

B2, judge whether present frame meets the direct Rule of judgment that CU needs divide, if so, step B4 is then skipped to, conversely, Step B3 is then performed, the direct Rule of judgment that the CU needs to divide is:

Wherein, v_curFor present frame CU variance, β is the first adaptive Dynamic gene, v_maxIt is the most generous of present frame CU Difference, bitdepth are internal bit depths used in encoder；

B3, PU partition mode detections, elimination factor distortion carried out using the quick decision algorithm of PU partition modes based on variance The optimal rate distortion costs that the PU partition modes and CU of Least-cost do not divide；

B4, directly judge that present frame CU needs to be divided, and recursive traversal present frame CU CU.

Further, the step B2, it includes：

B21, update p value be ++ p, the p be former frame four divide detected in tree-encoding CU sum；

B22, judge whether present frame meets the direct Rule of judgment that CU needs divide, if so, being in renewal q value then ++ Step B4 is performed after q, conversely, step B3 is then performed, wherein, q is that former frame four is divided in tree-encoding and meets that CU needs divide straight Connect the CU sums of Rule of judgment.

Further, the described first adaptive Dynamic gene β self-adapting regulation method is：

If q is less than the 10% of p, the β value divided with present frame four before tree-encoding after once adaptively adjusting subtracts 0.1 gained β value of the value as current CU；

Before if 10% of q more than or equal to p and 15% less than or equal to p, current CU β value divide tree-encoding for present frame four The once β value after adaptive adjustment；

If q is more than the 15% of p, with the β value that present frame four divides before tree-encoding after once adaptively adjusting plus 0.1 gained β value of the value as current CU；

If β is less than 0.1, current CU β value is set to 0.1；

Wherein, the initial value of the first factor-beta adaptively adjusted is 1.

Further, the step B3, it includes：

B31, SKIP patterns are tested, obtain the rate distortion costs under SKIP patterns；

B32, the variance for calculating two PU corresponding to Part_Nx2N partition modes, and judge that the two PU variance is It is no to be not less than given PU variance threshold values, if so, step B34 is then performed, conversely, then performing step B33；

B33, Part_Nx2N partition modes are tested, obtain the rate distortion costs under Part_Nx2N partition modes, Then step B34 is performed；

B34, the variance for calculating two PU corresponding to Part_2NxN partition modes, and judge that the two PU variance is It is no to be not less than given PU variance threshold values, if so, step B36 is then performed, conversely, then performing step B35；

B35, Part_2NxN partition modes are tested, obtain the rate distortion costs under Part_2NxN partition modes, Then step B36 is performed；

B36, AMP patterns are tested, select the minimum PU partition modes of rate distortion costs and CU do not divide it is optimal Rate distortion costs, the AMP patterns include PART_nRx2N partition modes, PART_nLx2N partition modes, PART_2NxnU and drawn Merotype and PART_2NxnD partition modes.

Further, the given PU variance threshold values are the half of present frame CU variance.

Further, the step C, it includes：

C1, judge whether present frame CU variance less than present frame does not divide CU average variance, if so, then performing step C2, conversely, then performing step C4；

C2, update m value be ++ m, the division judgment condition whether present frame CU meets CU is then judged, if so, then existing Renewal n value is ++ step C3 is performed after n, conversely, the division judgment condition for then performing step C4, the CU is：

Wherein,It is the average variance that present frame does not divide CU, and α is the second adaptive Dynamic gene, m is previous CU variances are less than in the cataloged procedure of frameCU number, n be meet in the cataloged procedure of former frame CU division judgement CU number of condition；

C3, according to the rule of simplification of setting present frame CU CU ergodic processes are simplified；

C4, the optimal rate distortion costs summation for obtaining all sub- CU of present frame CU, and judge that present frame CU is not divided most Whether good rate distortion costs are less than or equal to all sub- CU of present frame CU optimal rate distortion costs summation, if so, then showing current Frame CU need not be divided, conversely, then showing that present frame CU needs to use partition structure corresponding to sub- CU.

Further, the described second adaptive Dynamic gene α self-adapting regulation method is：

If n is less than the 70% of m, the α values divided with present frame four before tree-encoding after once adaptively adjusting subtract 0.5 gained α value of the value as current CU；

Before if 70% of n more than or equal to m and 75% less than or equal to m, current CU α values divide tree-encoding for present frame four The once α values after adaptive adjustment；

If n is more than the 75% of m, with the α values that present frame four divides before tree-encoding after once adaptively adjusting plus 0.5 gained α value of the value as current CU；

If α is less than 1, current CU α values are set to 1；

Wherein, the initial value of the second factor-alpha adaptively adjusted is 6.

Further, the step C3, it is specially：

The two CU for the being located at present frame CU surfaces and the left side respectively dividing condition setting present frame according to A and L CU's simplifies traversal flag bit bFast, and then present frame CU CU ergodic processes are simplified：If A and L are without CU Division, then it is true to set bFast, then directly judges that present frame CU need not be divided, and now, no longer travels through present frame CU's Sub- CU；If for A with having that a CU is not divided in L and another CU is divided, setting bFast is false, then judges to work as Whether the optimal rate distortion costs that previous frame CU is not divided are less than or equal to the optimal ratio distortion generation that all sub- CU of present frame CU are not divided Valency summation, if it is then judging that present frame CU need not be divided, now, previous frame CU CU is no longer traveled through, if it is not, Stop the sub- CU of traversal present frame so after present frame CU to be only divided into 4 sub- CU；If A and L are divided, set Put bFast and be false, then recursive traversal present frame CU, and judge all sub- CU of present frame CU optimal ratio distortion generation Whether valency summation is less than or equal to the optimal rate distortion costs summation that all sub- CU of present frame CU are not divided, if it is then judging Present frame CU need not be divided, and now, no longer travel through previous frame CU CU, if not, then present frame CU uses sub- CU pairs The partition structure answered.

The beneficial effects of the invention are as follows：It is related to present in the local feature of image based on the coding efficiency of predictive coding Property, employ the quick decision algorithm of PU partition modes based on variance and the quick decision algorithm of CU partition modes based on variance It is combined to carry out the selection of CU and PU partition mode, first according to CU variance and PU variance come look-ahead CU and PU Division court verdict, greatly reduce coding unit four divide tree node, PU partition modes and RQT nodes travel time and time All previous number, on the premise of HEVC video compression efficiencies and coding quality is kept substantially, encoder complexity is greatly reduced, improved Coding rate, the high request of application and mobile terminal in real time can be met.Further, specially carried for AI and AI 10 coding configurations Supply in four points of tree-encodings to divide CU look-ahead method, further reduced and compiled by the CU direct Rule of judgment for needing to divide The complexity of code.Further, it is additionally arranged present frame CU CU ergodic processes are simplified according to the rule of simplification of setting Step, the time decision of CU partition modes is further reduced, coding rate is faster.

Brief description of the drawings

The invention will be further described with reference to the accompanying drawings and examples.

Fig. 1 is the coding framework of HEVC video encoding standards；

Fig. 2 is the PU partition mode schematic diagrames of HEVC video encoding standards；

Fig. 3 is a kind of overall flow figure of the adaptive Fast video coding method based on variance of the present invention；

Fig. 4 is step B of the present invention flow chart；

Fig. 5 is step B2 of the present invention flow chart；

Fig. 6 is step B3 of the present invention flow chart；

Fig. 7 is step C of the present invention flow chart；

Fig. 8 is the CU structural representations of one or two of object of the embodiment of the present invention；

Fig. 9 is the flow chart of PU partition mode quick decision algorithm of the embodiment of the present invention one based on variance；

Figure 10 is the flow chart of adaptive quick HEVC coding method of the embodiment of the present invention one based on variance.

Embodiment

Reference picture 3, a kind of adaptive Fast video coding method based on variance, including：

A, present frame CU variance is calculated；

Preferred embodiment is further used as, the step A, it is specially：

Reference picture 4, preferred embodiment is further used as, the step B, it includes：

Wherein, select the process for the optimal rate distortion costs that CU is not divided and select the minimum PU division moulds of rate distortion costs The process of formula is similar, and the process for selecting the optimal ratio distortion generation that CU is not divided is：Encoder assumes that CU without division, obtains it Rate distortion costs under different predictive modes, the rate distortion costs of minimum are therefrom selected to be lost as the optimal ratio that CU is not divided True cost；The process for selecting the minimum PU partition modes of rate distortion costs is：Encoder assumes that PU needs to be divided, and obtains it Rate distortion costs under different predictive modes, therefrom select the PU partition mode conducts corresponding to the rate distortion costs of minimum The minimum PU partition modes of rate distortion costs.

Reference picture 5, preferred embodiment is further used as, the step B2, it includes：

Wherein, ++ p refers to that p adds 1 certainly, ++ q refers to that q adds 1 certainly.

Preferred embodiment is further used as, the self-adapting regulation method of the first adaptive Dynamic gene β is：

If β is less than 0.1, current CU β value is set to 0.1；

Wherein, the initial value of the first factor-beta adaptively adjusted is 1.

Because four points of tree-encoding units comprise at least a CU, therefore present frame four divides the process of tree-encoding to carry out at least β value once adaptively adjusts.

Reference picture 6, preferred embodiment is further used as, the step B3, it includes：

Preferred embodiment is further used as, the given PU variance threshold values are the half of present frame CU variance.

Reference picture 7, preferred embodiment is further used as, the step C, it includes：

Wherein, ++ m refers to that m adds 1 certainly, ++ n refers to that n adds 1 certainly.

Preferred embodiment is further used as, the self-adapting regulation method of the second adaptive Dynamic gene α is：

If α is less than 1, current CU α values are set to 1；

Wherein, the initial value of the second factor-alpha adaptively adjusted is 6.

Because four points of tree-encoding units comprise at least a CU, therefore present frame four divides the process of tree-encoding to carry out at least α values once adaptively adjust.

Preferred embodiment is further used as, the step C3, it is specially：

The present invention is described in further detail with reference to Figure of description and specific embodiment.

Embodiment one

Reference picture 8,9 and 10, the first embodiment of the present invention：

During CTU four points of tree-encodings, HEVC encoders can select the minimum volume of rate distortion costs according to RDO criterions Pattern, so as to preferably remove the redundant data in video sequence, this means that CU partition mode and drawing for PU Merotype depends on its corresponding encoding efficiency.Redundant data in video sequence be based on time redundancy and spatial redundancy, And infra-frame prediction or inter prediction are applied on PU by HEVC encoders by using CU specific PU partition modes, so as to Remove time redundancy and spatial redundancy, that is to say, that CU and PU encoding efficiency depends greatly on the property of predictive coding Energy.

According to the principle of predictive coding, it is known that the local feature of its coding efficiency and image exist it is very big related Property, therefore for the CU of some depth, using the encoding efficiency of specific PU partition modes with the CU's and each PU Characteristics of image is closely related.And the variance of regional area can reflect CU and PU characteristics of image to a certain extent, also It is to say variance and CU and PU partition mode selection there is correlation.Correlation analysis based on more than, the present invention propose Adaptive Fast video coding algorithm based on variance, the algorithm include two parts, are the CU division moulds based on variance respectively The quick decision algorithm of formula and the quick decision algorithm of PU partition modes based on variance.Below one by one to the two algorithms and by this two The binary encoding algorithm that individual algorithm is formed is introduced.

(1) the quick decision algorithm of CU partition modes based on variance

As previously noted, the main purpose of CTU four points of tree-encodings is to ensure that it is given that image fault is no more than In the case of value, the redundant data of image-region corresponding to CTU is removed as much as possible, and especially time redundancy and space is superfluous It is remaining.

Infra-frame prediction is the main tool for removing spatial redundancy, and inter prediction is then the main work for removing time redundancy Tool, therefore infra-frame prediction and inter prediction be also into main code instrument used in four points of tree-encodings, its coding efficiency can be with Largely influence judgement of four points of tree-encodings for CU partition modes.

The general principle of infra-frame prediction is the encoded reconstruction pixel prediction pixel to be encoded using present frame, transmission or Store original pixels and the difference of the predicted value.And the general principle of inter prediction is to utilize encoded reconstruction frames prediction present frame Pixel to be encoded, the difference of transmission or storage original pixels and the predicted value.Therefore it is pre- by application infra-frame prediction and interframe Survey, it is possible to remove the spatial redundancy and time redundancy in video sequence.

However, in block-based hybrid encoding frame, respectively there is the hypothesis of an acquiescence with inter prediction in infra-frame prediction. Infra-frame prediction assumes that all pixels in estimation range have identical grain direction；Inter prediction is then assumed in estimation range All pixels belong to same object, share same motion vector.In four points of tree-encodings of reality, all possible knot The included region of point differ establish a capital meet these it is assumed that but for those meet these nodes assumed, predictive coding Good encoding efficiency can be obtained, it is believed that need not continue to detect these nodes child node (namely need not pair CU carries out subdivided)；And for those are unsatisfactory for the node of these hypothesis, the encoding efficiency of predictive coding is excessively poor, can Directly to judge that these nodes must divide, it is not required to detect these nodes again.Further, it is also possible to know the knot for meeting that these assume Point often has the variance of very little, and the nodes for being unsatisfactory for these hypothesis often have a very big variance, thus CU variance and CU division judgement has correlation, and CU division court verdict can be predicted according to CU variance.

Variance calculates the original pixels that can use CU corresponding regions or reconstruction pixel, in order to avoid rebuilding pixel institute The distortion of introducing, the present embodiment calculates its variance using the original pixels of CU corresponding regions, shown in equation below (1) and (2)：

Based on more than analyze, this algorithm by the variance according to CU come look-ahead CU division adjudicate, even formula (3) into Vertical, encoder would consider that the CU need not be divided, and no longer detect its sub- CU.

In formula (3), α adjustment can be carried out before the coding of every frame, it is assumed that m is the variance in the cataloged procedure of former frame It is less thanCU number, and n be in the cataloged procedure of former frame cause formula (3) set up CU number, α initial value It is set to 6.α Adaptive adjusting algorithm is as follows：

1) if 70%, αs of the n less than m will subtract 0.5 on the basis of preceding once adaptive adjustment；

2) if 75%, αs of the n more than m will add 0.5 on the basis of preceding once adaptive adjustment；

3) if α is less than 1, α is set to 1.

After α adaptive adjustment is completed, m and n need to be reset to 0, and this can be updated in the cataloged procedure of present frame Two values.In addition, in order to obtainPreceding 10 CTU tetra- that this algorithm can collect present frame divide CU are not divided in tree-encoding Variance, using the average value of these variances as

In addition, this algorithm also encodes for AI and AI 10 divides shifting to an earlier date for CU in the corresponding four points of tree-encodings of configuration provides Forecasting Methodology, if formula (4) is set up, it will directly judge that current CU needs to divide, no longer it detected.

v_cur＞ v_max·β (4)

Wherein, v_maxValue such as formula (5) shown in：

Before every frame coding, this algorithm can adaptively adjust β.It is assumed that p, which is former frame four, divides the CU detected in tree-encoding Sum, q are that former frame four divides in tree-encoding the CU sums for meeting formula (4), and β initial value is 1, then β adaptive adjustment is calculated Method is as follows：

1) if 10%, βs of the q less than p will subtract 0.1 on the basis of preceding once adaptive adjustment；

2) if 15%, βs of the q more than p will add 0.1 on the basis of preceding once adaptive adjustment；

3) if β is less than 0.1, β is set to 0.1.

After completing β adaptive adjustment before every frame coding, p and q are reset to 0, and compile in four points of trees of present frame The two values are updated in code.The CU division judgement fast algorithms of variance are namely based on above.

(2) the PU division judgement fast algorithms based on variance

According to analysis above, when PU partition modes can cause all PU in CU all to meet that the acquiescence of inter prediction is false Regularly, inter prediction can obtain extraordinary encoding efficiency under the PU partition modes, you can to think the PU partition modes It is exactly optimal partition mode, otherwise encoding efficiency of the inter prediction under the PU partition modes is excessively poor, you can to think this PU partition modes are not optimal partition modes.

And the variance of PU corresponding to PU partition modes can reflect whether the PU partition modes meet inter prediction very well It is assumed that.Fig. 8 shows the CU for including two objects A and B, and dotted line represents both borders, it is assumed that the pixel value in A is a, B In pixel value be b, it is known that Part_Nx2N partition mode can cause A and B all meet inter prediction it is assumed that Now both variances are all 0, and Part_2NxN partition mode then causes corresponding PU to be all unsatisfactory for the vacation of inter prediction Fixed, now two PU variance is all 0.25 (a-b)².So if PU variance is more than some threshold value and (in the present embodiment will The threshold value is set to the half of CU variances), it is believed that corresponding PU partition modes are not optimal partition modes.

This algorithm will predict the court verdict of PU partition modes by PU variances corresponding to PU partition modes, such as Fig. 9 institutes Show.During PU partition modes are detected, encoder calculates current CU variance first, then tests SKIP patterns (wherein, SKIP patterns use the PU of Part_2N × 2N partition modes), obtain rate distortion costs corresponding to SKIP patterns.Subsequent encoder The variance of two PU corresponding to Part_Nx2N partition modes is first calculated, if both of which is not less than given PU variance threshold values, The detection of Part_Nx2N partition modes is just directly skipped, otherwise will obtain the rate distortion costs under the partition mode.Next, Encoder equally first calculates the variance of two PU corresponding to Part_2NxN partition modes, if being neither less than given PU Variance threshold values, just directly skip the detection of Part_2NxN partition modes, otherwise obtain the rate distortion costs under the partition mode. Last test AMP patterns, and selected from SKIP patterns, Part_Nx2N partition modes, Part_2NxN partition modes and AMP patterns Go out the minimum pattern of rate distortion costs.

(3) the adaptive Fast video coding algorithm based on variance

As shown in Figure 10, above-mentioned two algorithm is combined, is exactly proposed by the invention based on the adaptive of variance Fast video coding algorithm, specific implementation process are：

Preceding 10 CTU of present frame first are used for Data Collection, and CTU below is applied into this algorithm.The four of CTU Divide during tree-encoding, encoder first calculates CU variance.If using AI or AI 10 to encode configuration, renewal p's Value, and detect whether it meets formula (4), if it is satisfied, then renewal q value, directly judges that current CU needs to divide, recurrence time Go through its sub- CU；Otherwise the flow of application drawing 9, the optimal rate distortion costs that the CU is not divided are obtained.

Next, if encoder uses interframe encode to configure, or on occurring in AI or AI 10 coding configurations Formula (4) the ungratified situation stated, then the quick decision algorithm of PU partition modes is being applied to obtain the optimal ratio that the CU is not divided After distortion cost Rd, whether the variance for testing CU is less than the average variance for not dividing CU, if it is not, its sub- CU of recursive traversal, Obtain optimal rate distortion costs summation J corresponding to sub- CU_subIf R_dLess than or equal to J_sub, judge that the CU need not be divided, otherwise The CU uses partition structure corresponding to sub- CU.

If CU variance is less than the average variance for not dividing CU, m value is updated, whether judgment formula (3) is set up, such as Fruit is set up, and updates n value, it is assumed that A and L is directly over the CU and the CU of the equal size on the left side, if A and L are not drawn Point, setting bFast is true, and it is false otherwise to set bFast.If bFast is true, directly judge that CU is not divided, no Sub- CU is traveled through again.If bFast is false and A is with there is a CU not divide in L, then judges that the optimal ratio that CU is not divided loses Whether true cost is less than or equal to the optimal rate distortion costs summation that sub- CU is not divided, if it is not, CU only then is divided into 4 sons CU, no longer travels through sub- CU, otherwise judges that CU need not be divided and no longer be traveled through sub- CU.If bFast is false and A and L enter Division is gone, then the sub- CU of recursive traversal, according to sub- CU optimal rate distortion costs summation, judge whether the CU needs to divide.

If after renewal m value, formula (3) is invalid, then the sub- CU of same recursive traversal, is lost according to sub- CU optimal ratio True cost summation, judges whether the CU needs to divide.

Embodiment two

In order to verify algorithm performance proposed by the invention, the present invention realizes proposed algorithm based on HM16.0, and And substantial amounts of experiment is carried out.Follow-up all experiments of the invention are all carried out under identical experimental situation, of the invention Experimental situation is as follows：

Processor：Intel(R)Core(TM)i7-2600CPU@3.40GHz 3.41GHz；

Operating system：64 Windows 7；

Coding configuration：Test condition specified by CTC (Common Test Condition)

The present embodiment is by the way that the test result of algorithm proposed by the invention test result corresponding with HM16.0 is carried out Compare, so that the synthetically performance of evaluation algorithms, experimental data (wherein, table 1, table 2 and table 3 as shown in table 1 below, table 2 and table 3 In video type and video sequence be all from the test condition of HEVC standard defined).

Three evaluation indexes being specified using CTC of the present invention carry out the performance of measure algorithm, equation below (6), formula (7) and Shown in formula (8),

Wherein, Time_opt、Bitrate_optAnd PSNR_optIt is the coding speed corresponding to algorithm proposed by the invention respectively Degree, code check and signal to noise ratio PSNR, and Time_org、Bitrate_orgAnd PSNR_orgIt is coding rate, the code corresponding to HM16.0 respectively Rate and signal to noise ratio PSNR.

Table 1 lists experimental result of the algorithm of the present invention under AI and AI10 coding configurations.From the table, the present invention Algorithm AI coding configuration under can averagely reduce by 25.02% encoder complexity, while substantially with original encoder Maintain same coding efficiency.Under AI coding configurations, the coding that algorithm highest of the invention can reduce by 55.96% is complicated Degree, effectively optimizes coding rate.And the algorithm of the present invention can averagely reduce by 25.14% volume under AI10 coding configurations Code complexity, and coding efficiency is substantially such as original encoder, wherein, the coding that highest can reduce by 56.13% is complicated Degree.Specifically, algorithm of the invention can effectively optimize high definition and ultra high-definition video under AI and AI10 coding configurations The coding rate of sequence.

Experimental result of the algorithm of table 1 under AI and AI10 coding configurations

And table 2 then lists experimental result of the algorithm of the present invention under RA and RA10 coding configurations.Encoded and configured by RA Experimental data understand, algorithm of the invention can averagely reduce by 34.37% encoder complexity, and the increase of code check is far from To 1%, and PSNR change can be ignored substantially, and algorithm highest of the invention can reduce by 53.93% encoder complexity. The algorithm of the present invention is can be seen that from the data of RA coding configurations can reduce for ultra high-definition video sequence connect under in such a configuration Nearly 50% encoder complexity.And the coding that the algorithm of the present invention can averagely reduce by 34.42% under RA10 coding configurations is answered Miscellaneous degree, 54.07% encoder complexity can be reduced with original suitable coding efficiency, highest by maintaining.Therefore, with RA Coding configuration it is similar, RA10 coding configuration under, algorithm of the invention can effectively optimize ultra high-definition video sequence Coding rate.

Experimental result of the algorithm of table 2 under RA and RA10 coding configurations

Table 3 lists experimental result of the algorithm of the present invention under LB and LB10 coding configurations.From the data of the table, The algorithm of the present invention can averagely reduce by 37.10% encoder complexity under LB coding configurations, maintain and original phase When coding efficiency, highest can reduce by 57.11% encoder complexity.Similarly, algorithm of the invention can be with highly effective Ground is reduced for high definition and the encoder complexity of ultra high-definition video sequence.And encode and configure for LB10, algorithm of the invention With similar experimental result.Therefore on the whole, under LB and LB10 coding configurations, algorithm of the invention is in holding and originally While suitable coding efficiency, encoder complexity can be effectively reduced.

Experimental result of the algorithm of table 3 under LB and LB10 coding configurations

In summary, algorithm proposed by the present invention can keep encoding basically identical coding efficiency with original HEVC Meanwhile encoder complexity is significantly reduced, especially suitable for video sequence more than high definition resolution ratio.

Above is the preferable implementation to the present invention is illustrated, but the invention is not limited to the implementation Example, those skilled in the art can also make a variety of equivalent variations on the premise of without prejudice to spirit of the invention or replace Change, these equivalent deformations or replacement are all contained in the application claim limited range.

Claims

A kind of 1. adaptive Fast video coding method based on variance, it is characterised in that：Including：

A, present frame CU variance is calculated；

B, PU partition mode inspections are carried out using the quick decision algorithm of PU partition modes based on variance according to present frame CU variance Survey, obtain the optimal rate distortion costs that the minimum PU partition modes of rate distortion costs and CU are not divided；

C, CU division judgements are carried out using the quick decision algorithm of CU partition modes based on variance according to present frame CU variance, really Determine the minimum CU partition modes of rate distortion costs；

D, encoded according to the minimum PU partition modes of rate distortion costs and CU partition modes；

The step B, it includes：

B1, judge present frame coding configuration whether be AI or AI10 coding configuration, if so, then perform step B2, conversely, then holding Row step B3；

B2, judge whether present frame meets the direct Rule of judgment that CU needs divide, if so, step B4 is then skipped to, conversely, then holding Row step B3, the direct Rule of judgment that the CU needs to divide are:

<mfenced open = "{" close = ""> <mtable> <mtr> <mtd> <msub> <mi>v</mi> <mrow> <mi>c</mi> <mi>u</mi> <mi>r</mi> </mrow> </msub> <mo>></mo> <msub> <mi>v</mi> <mrow> <mi>m</mi> <mi>a</mi> <mi>x</mi> </mrow> </msub> <mo>&CenterDot;</mo> <mi>&beta;</mi> </mtd> </mtr> <mtr> <mtd> <msub> <mi>v</mi> <mrow> <mi>m</mi> <mi>a</mi> <mi>x</mi> </mrow> </msub> <mo>=</mo> <mfrac> <msup> <mrow> <mo>(</mo> <msup> <mn>2</mn> <mrow> <mi>b</mi> <mi>i</mi> <mi>t</mi> <mi>d</mi> <mi>e</mi> <mi>p</mi> <mi>t</mi> <mi>h</mi> </mrow> </msup> <mo>-</mo> <mn>1</mn> <mo>)</mo> </mrow> <mn>2</mn> </msup> <mn>4</mn> </mfrac> <mo>,</mo> </mtd> </mtr> </mtable> </mfenced>

Wherein, v_curFor present frame CU variance, β is the first adaptive Dynamic gene, v_maxIt is present frame CU maximum variance, Bitdepth is internal bit depth used in encoder；

B3, PU partition mode detections are carried out using the quick decision algorithm of PU partition modes based on variance, select rate distortion costs The optimal rate distortion costs that minimum PU partition modes and CU does not divide；

B4, directly judge that present frame CU needs to be divided, and recursive traversal present frame CU CU；

The step B2, it includes：

B21, update p value be ++ p, the p be former frame four divide detected in tree-encoding CU sum；

B22, judge whether present frame meets the direct Rule of judgment that CU needs divide, if so, being in renewal q value then ++ after q Step B4 is performed, conversely, then perform step B3, wherein, q is that former frame four divides in tree-encoding meet that CU needs divide direct to sentence The CU sums of broken strip part；

The self-adapting regulation method of the first adaptive Dynamic gene β is：

If q is less than the 10% of p, the β value divided with present frame four before tree-encoding after once adaptively adjusting subtracts the value of 0.1 gained β value as current CU；

If 10% of q more than or equal to p and 15% less than or equal to p, current CU β value are divided before tree-encoding once for present frame four β value after adaptive adjustment；

If q is more than the 15% of p, the value with the β value that present frame four divides before tree-encoding after once adaptively adjusting plus 0.1 gained β value as current CU；

If β is less than 0.1, current CU β value is set to 0.1；

Wherein, the initial value of the first factor-beta adaptively adjusted is 1；

The step B3, it includes：

B31, SKIP patterns are tested, obtain the rate distortion costs under SKIP patterns；

The variance of two PU corresponding to B32, calculating Part_Nx2N partition modes, and judge whether the two PU variance is equal Not less than given PU variance threshold values, if so, step B34 is then performed, conversely, then performing step B33；

B33, Part_Nx2N partition modes are tested, obtain the rate distortion costs under Part_Nx2N partition modes, then Perform step B34；

The variance of two PU corresponding to B34, calculating Part_2NxN partition modes, and judge whether the two PU variance is equal Not less than given PU variance threshold values, if so, step B36 is then performed, conversely, then performing step B35；

B35, Part_2NxN partition modes are tested, obtain the rate distortion costs under Part_2NxN partition modes, then Perform step B36；

B36, AMP patterns are tested, select the optimal ratio that the minimum PU partition modes of rate distortion costs and CU are not divided and lose True cost, the AMP patterns include PART_nRx2N partition modes, PART_nLx2N partition modes, PART_2NxnU division moulds Formula and PART_2NxnD partition modes.
A kind of 2. adaptive Fast video coding method based on variance according to claim 1, it is characterised in that：It is described Step A, it is specially：

Present frame CU variance is calculated, the calculation formula of the variance of the CU is：

<mrow> <mfenced open = "{" close = ""> <mtable> <mtr> <mtd> <mrow> <mover> <mi>p</mi> <mo>&OverBar;</mo> </mover> <mo>=</mo> <mfrac> <mn>1</mn> <mrow> <mi>N</mi> <mi>M</mi> </mrow> </mfrac> <munderover> <mo>&Sigma;</mo> <mrow> <mi>i</mi> <mo>=</mo> <mn>0</mn> </mrow> <mrow> <mi>N</mi> <mo>-</mo> <mn>1</mn> </mrow> </munderover> <munderover> <mo>&Sigma;</mo> <mrow> <mi>j</mi> <mo>=</mo> <mn>0</mn> </mrow> <mrow> <mi>M</mi> <mo>-</mo> <mn>1</mn> </mrow> </munderover> <msub> <mi>p</mi> <mrow> <mi>i</mi> <mi>j</mi> </mrow> </msub> </mrow> </mtd> </mtr> <mtr> <mtd> <mrow> <mi>v</mi> <mo>=</mo> <mfrac> <mn>1</mn> <mrow> <mi>N</mi> <mi>M</mi> </mrow> </mfrac> <munderover> <mo>&Sigma;</mo> <mrow> <mi>i</mi> <mo>=</mo> <mn>0</mn> </mrow> <mrow> <mi>N</mi> <mo>-</mo> <mn>1</mn> </mrow> </munderover> <munderover> <mo>&Sigma;</mo> <mrow> <mi>j</mi> <mo>=</mo> <mn>0</mn> </mrow> <mrow> <mi>M</mi> <mo>-</mo> <mn>1</mn> </mrow> </munderover> <msup> <mrow> <mo>(</mo> <msub> <mi>p</mi> <mrow> <mi>i</mi> <mi>j</mi> </mrow> </msub> <mo>-</mo> <mover> <mi>p</mi> <mo>&OverBar;</mo> </mover> <mo>)</mo> </mrow> <mn>2</mn> </msup> </mrow> </mtd> </mtr> </mtable> </mfenced> <mo>,</mo> </mrow>

Wherein, the size of CU corresponding regions is N × M, { p_ij| 0≤i≤N-1,0≤j≤M-1, i ∈ Z, j ∈ Z } represent that CU is corresponding All original pixel values in region,It is the average of all original pixels in CU corresponding regions, and v is former in CU corresponding regions The variance of beginning pixel.
A kind of 3. adaptive Fast video coding method based on variance according to claim 1, it is characterised in that：It is described Given PU variance threshold values are the half of present frame CU variance.
A kind of 4. adaptive Fast video coding method based on variance according to claim 1, it is characterised in that：It is described Step C, it includes：

C1, judge whether present frame CU variance is less than the average variance that present frame does not divide CU, if so, step C2 is then performed, Conversely, then perform step C4；

C2, update m value be ++ m, then judge the division judgment condition whether present frame CU meets CU, if so, then updating n Value be ++ step C3 is performed after n, conversely, the division judgment condition for then performing step C4, the CU is：

<mrow> <msub> <mi>v</mi> <mrow> <mi>c</mi> <mi>u</mi> <mi>r</mi> </mrow> </msub> <mo>&CenterDot;</mo> <mi>&alpha;</mi> <mo><</mo> <msub> <mover> <mi>v</mi> <mo>&OverBar;</mo> </mover> <mrow> <mi>l</mi> <mi>e</mi> <mi>a</mi> <mi>f</mi> </mrow> </msub> <mo>,</mo> </mrow>

Wherein,It is the average variance that present frame does not divide CU, and α is the second adaptive Dynamic gene, m is in former frame CU variances are less than in cataloged procedureCU number, n is to meet CU division judgment condition in the cataloged procedure of former frame CU number；

C3, according to the rule of simplification of setting present frame CU CU ergodic processes are simplified；

C4, the optimal rate distortion costs summation for obtaining all sub- CU of present frame CU, and judge the optimal ratio that present frame CU is not divided Whether distortion cost is less than or equal to all sub- CU of present frame CU optimal rate distortion costs summation, if so, then showing present frame CU It need not divide, conversely, then showing that present frame CU needs to use partition structure corresponding to sub- CU.
A kind of 5. adaptive Fast video coding method based on variance according to claim 4, it is characterised in that：Described Two adaptive Dynamic gene α self-adapting regulation method is：

If n is less than the 70% of m, the α values divided with present frame four before tree-encoding after once adaptively adjusting subtract the value of 0.5 gained α values as current CU；

If 70% of n more than or equal to m and 75% less than or equal to m, current CU α values are divided before tree-encoding once for present frame four α values after adaptive adjustment；

If n is more than the 75% of m, the value with the α values that present frame four divides before tree-encoding after once adaptively adjusting plus 0.5 gained α values as current CU；

If α is less than 1, current CU α values are set to 1；

Wherein, the initial value of the second factor-alpha adaptively adjusted is 6.
A kind of 6. adaptive Fast video coding method based on variance according to claim 4, it is characterised in that：The step Rapid C3, it is specially：

The two CU for the being located at present frame CU surfaces and the left side respectively dividing condition setting present frame CU according to A and L Simplify traversal flag bit bFast, then present frame CU CU ergodic processes are simplified：If A and L are drawn without CU Point, then it is true to set bFast, then directly judges that present frame CU need not be divided, now, no longer travels through present frame CU CU；If for A with having that a CU is not divided in L and another CU is divided, setting bFast is false, is then judged current Whether the optimal rate distortion costs that frame CU is not divided are less than or equal to the optimal rate distortion costs that all sub- CU of present frame CU are not divided Summation, if it is then judging that present frame CU need not be divided, now, previous frame CU CU is no longer traveled through, if it is not, that Stop the sub- CU of traversal present frame after present frame CU to be only divided into 4 sub- CU；If A and L are divided, set BFast is false, then the sub- CU of recursive traversal present frame, and judges all sub- CU of present frame CU optimal rate distortion costs Whether summation is less than or equal to the optimal rate distortion costs summation that all sub- CU of present frame CU are not divided, if it is then judging to work as Previous frame CU need not be divided, and now, no longer travel through previous frame CU CU, if not, then present frame CU is corresponding using sub- CU Partition structure.