CN103533354B - A kind of HD video fast encoding method - Google Patents

Abstract

The invention discloses a kind of HD video fast encoding method, whole process is divided into off-line training step and the online operational phase of threshold value by it, in off-line training step, obtain size in every class video sequence be the unified optimum segmentation threshold value of the coding unit of 64 × 64, size be unified optimum segmentation threshold value and the unified optimum segmentation threshold value of coding unit that size is 16 × 16 of the coding unit of 32 × 32；In the online operational phase of threshold value, it is respectively the rate distortion costs value of the coding unit of 64 × 64,32 × 32,16 × 16 with size in pending HEVC HD video according to optimum segmentation threshold value, determine that in pending HEVC HD video, size is respectively the coding unit of 64 × 64,32 × 32,16 × 16 the need of continuing segmentation, this mode can be in the case of distortion performance loss be less, save a large amount of scramble time, i.e. on the premise of ensureing decoded video objective quality, it is possible to efficiently reduce the scramble time.

Description

A kind of HD video fast encoding method

Technical field

The present invention relates to the processing method of a kind of video signal, especially relate to a kind of HD video fast encoding method.

Background technology

Along with increasing to video quality and high definition and ultra high-definition digital video demand of people, international electricity in 2010 Letter alliance's Video Coding Experts group (ITU-T Video Coding Experts Group, VCEG) and ISO and Motion Picture Experts Group of International Electrotechnical Commission (ISO/IEC Moving Picture Experts Group, MPEG) sets up Video coding joint specialist group (Joint Collaborative Team on Video Coding) the research next generation efficiently regard Frequency coding (High Efficiency Video Coding Standard, HEVC) standard.HEVC standard is to continue H.264/AVC After video encoding standard, the video encoding standard of a new generation, its main target is under conditions of same video content quality, Save the code check of about 50%, and solve H.264/MPEG-4AVC two basic problems present in standard: high-resolution and Parallel processing structure problem.

HEVC still uses and similar hybrid coding architecture, but has carried out substantial amounts of in HEVC Technological innovation, such as segmentation based on large scale quadtree coding unit and residual coding structure, multi-angle infra-prediction techniques, height Precision motion compensation technology etc..Owing to HEVC have employed the recursive encoding structure of based on quaternary tree 4 layers, maximum coding unit (Coding Unit, CU) to the size of minimum code unit be respectively 64 × 64,32 × 32,16 × 16 and 8 × 8, and HEVC in Coding unit still use rate-distortion optimization model, coding unit employs 35 kinds of intra prediction modes, therefore HEVC standard In the composition of each forced coding unit need the calculating through 85 rate-distortion optimization models, compared in H.264/AVC 16 × 16,8 × 8 two kinds of coded macroblocks and 9 kinds of intra prediction modes, the large scale quadtree coding unit in HEVC and multi-angle frame Interior Predicting Technique brings substantial amounts of computation complexity and time complexity.

Summary of the invention

The technical problem to be solved is to provide a kind of HD video fast encoding method, and it is after ensureing decoding On the premise of video objective quality, it is possible to efficiently reduce the scramble time required for HD video.

The present invention solves the technical scheme that above-mentioned technical problem used: a kind of HD video fast encoding method, its It is characterised by comprising the following steps:

1., from the standard test sequences of at least N' type of HEVC HD video, N number of video sequence is the most arbitrarily selected Selected all video sequences are constituted a training set by row, and this training set comprises N' × N number of video sequence, wherein, 3≤ N'≤6, N >=1；

First, determine that in each video sequence in training set, size is respectively 64 × 64,32 × 32,16 × 16 The cutting state of coding unit；Then, in each video sequence in statistics training set, size is respectively 64 × 64,32 The depth value of the coding unit of × 32,16 × 16, rate distortion costs value, four coding unit of next layer rate distortion costs value it With, cutting state；Then, it is respectively 64 × 64,32 by training set belongs to size in of a sort all video sequences The statistical data of the coding unit of × 32,16 × 16 is stored in a file, total N' file, will store the institute of the i-th class The file of statistical data having the coding unit that size in video sequence is respectively 64 × 64,32 × 32,16 × 16 is designated as F_i, wherein, 1≤i≤N'；

3. according to step 2. in be stored in the statistical data in each file, obtaining size in every class video sequence is In the unified optimum segmentation threshold value of the coding unit of 64 × 64, every class video sequence, size is the coding unit of 32 × 32 Unified optimum segmentation threshold value and every class video sequence in size be 16 × 16 the unified optimum of coding unit Segmentation threshold；

4. the coding unit that each size is 64 × 64 in pending HEVC HD video carries out traversal compile Code, in traversal cataloged procedure, it is first determined the type of pending HEVC HD video；Then high according to pending HEVC The type of clear video, determines that in pending HEVC HD video, size is respectively the volume of 64 × 64,32 × 32,16 × 16 The optimum segmentation threshold value of code unit；Then according to size in pending HEVC HD video be respectively 64 × 64,32 × 32, the rate distortion costs value of the coding unit of 16 × 16 and the optimum segmentation threshold value of correspondence, determine that pending HEVC high definition regards In Pin, size is respectively the coding unit of 64 × 64,32 × 32,16 × 16 the need of continuing segmentation.

Described step detailed process 2. is:

2.-1, video sequence currently pending in training set is defined as current video sequence；

2.-2, the coding unit that all sizes are 64 × 64 in current video sequence is handled as follows:

2.-2a, the coding unit that currently pending size is 64 × 64 is defined as current coded unit；②- 2b, according to the rate distortion costs value of current coded unit and the rate distortion generation of four coding unit of next layer of current coded unit It is worth sum, determines the cutting state of current coded unit, if the rate distortion costs value of current coded unit is more than or equal to The rate distortion costs value sum of four coding unit of next layer of current coded unit, it is determined that current coded unit continues to divide Cutting, being divided into four sizes is the coding unit of 32 × 32, and determines that the cutting state of current coded unit is for splitting, so Rear execution step 2.-2c；If the rate distortion costs value of current coded unit is less than four, next layer volume of current coded unit The rate distortion costs value sum of code unit, it is determined that current coded unit does not continues to segmentation, and determines current coded unit Cutting state, for not split, then performs step 2.-2c；2.-2c, by the pending volume that size is 64 × 64 of the next one Code unit, as current coded unit, is then back to step 2.-2b and continues executing with, until all sizes in current video sequence Size be 64 × 64 coding unit be disposed, then perform step 2.-3；

2.-3, the coding unit that all sizes are 32 × 32 in current video sequence is handled as follows:

2.-3a, the coding unit that currently pending size is 32 × 32 is defined as current coded unit；②- 3b, according to the rate distortion costs value of current coded unit and the rate distortion generation of four coding unit of next layer of current coded unit It is worth sum, determines the cutting state of current coded unit, if the rate distortion costs value of current coded unit is more than or equal to The rate distortion costs value sum of four coding unit of next layer of current coded unit, it is determined that current coded unit continues to divide Cutting, being divided into four sizes is the coding unit of 16 × 16, and determines that the cutting state of current coded unit is for splitting, so Rear execution step 2.-3c；If the rate distortion costs value of current coded unit is less than four, next layer volume of current coded unit The rate distortion costs value sum of code unit, it is determined that current coded unit does not continues to segmentation, and determines current coded unit Cutting state, for not split, then performs step 2.-3c；2.-3c, by the pending volume that size is 32 × 32 of the next one Code unit, as current coded unit, is then back to step 2.-3b and continues executing with, until all sizes in current video sequence Size be 32 × 32 coding unit be disposed, then perform step 2.-4；

2.-4, the coding unit that all sizes are 16 × 16 in current video sequence is handled as follows:

2.-4a, the coding unit that currently pending size is 16 × 16 is defined as current coded unit；②- 4b, according to the rate distortion costs value of current coded unit and the rate distortion generation of four coding unit of next layer of current coded unit It is worth sum, determines the cutting state of current coded unit, if the rate distortion costs value of current coded unit is more than or equal to The rate distortion costs value sum of four coding unit of next layer of current coded unit, it is determined that current coded unit continues to divide Cutting, being divided into four sizes is the coding unit of 8 × 8, and determines that the cutting state of current coded unit is for splitting, then Perform step 2.-4c；If the rate distortion costs value of current coded unit is less than four, next layer coding of current coded unit The rate distortion costs value sum of unit, it is determined that current coded unit does not continues to segmentation, and determines dividing of current coded unit State of cutting, for not split, then performs step 2.-4c；2.-4c, by the pending coding that size is 16 × 16 of the next one Unit, as current coded unit, is then back to step 2.-4b and continues executing with, until all sizes in current video sequence are big Little be 16 × 16 coding unit be disposed, then perform step 2.-5；

2.-5, in statistics current video sequence, size is respectively the coding unit of 64 × 64,32 × 32,16 × 16 Depth value, rate distortion costs value, the rate distortion costs value sum of four coding unit of next layer, cutting state；

2.-6, using pending video sequence next in training set as current video sequence, be then back to step 2.- 2 continue executing with, until all video sequences in training set are disposed, obtain size in each video sequence in training set Size is respectively the statistical data of the coding unit of 64 × 64,32 × 32,16 × 16；

2.-7, by training set belongs to size in of a sort all video sequences be respectively 64 × 64,32 × 32, The statistical data of the coding unit of 16 × 16 is stored in a file, total N' file, will store all videos of the i-th class In sequence, the file of the statistical data that size is respectively the coding unit of 64 × 64,32 × 32,16 × 16 is designated as F_i, its In, 1≤i≤N'.

Described step detailed process 3. is:

3.-1, it is respectively 64 × 64,32 × 32,16 × 16 by storing size in all video sequences of the i-th class The file F of the statistical data of coding unit_iAs current file, wherein, 1≤i≤N'；

3.-2, find out from current file in the rate distortion costs value of the coding unit that all sizes are 64 × 64 Minima and maximum, correspondence is designated asWithBy big for all sizes identical for rate distortion costs value in current file Little be 64 × 64 coding unit composition one first set, by big for all sizes that rate distortion costs value in current file is k1 First set of the little coding unit composition being 64 × 64 is designated as Ω_k1, wherein,Then by each The rate distortion costs value of the coding unit that each size is 64 × 64 in one set and corresponding four, next layer coding Rate distortion costs value sum one first combination of composition of unit, by Ω_k1In 1 size of jth be the coding of 64 × 64 The first combination that the rate distortion costs value of unit forms with the rate distortion costs value sum of corresponding four coding unit of next layer It is designated asWherein, 1≤j1≤J1, J1 represent Ω_k1In the coding unit that size is 64 × 64 that comprises Total number,Represent Ω_k1In four, next layer coding of the coding unit that 1 size of jth is 64 × 64 The rate distortion costs value sum of unit；Again for each first set, from all first combinations, select the value of second element Different all first combinations, will be for Ω_k1The m1 selected first combination is designated asWherein, 1 ≤ m1≤M1≤J1, M1 represent for Ω_k1Total number of the first combination selected,Represent for Ω_k1Selected Four coding unit of next layer of the coding unit that size is 64 × 64 of second element representative in m1 first combination Rate distortion costs value sum；

The minimum in the rate distortion costs value of the coding unit that all sizes are 32 × 32 is found out from current file Value and maximum, correspondence is designated asWithBy all sizes identical for rate distortion costs value in current file it is Coding unit one second set of composition of 32 × 32, by all sizes that rate distortion costs value in current file is k2 be Second set of the coding unit composition of 32 × 32 is designated as Ω_k2, wherein,Then by each second collection The rate distortion costs value of the coding unit that each size is 32 × 32 in conjunction and corresponding four coding unit of next layer Rate distortion costs value sum composition one second combination, by Ω_k2In 2 sizes of jth be the coding unit of 32 × 32 The second combination that forms of the rate distortion costs value sum of rate distortion costs value and corresponding four coding unit of next layer be designated asWherein, 1≤j2≤J2, J2 represent Ω_k2In the coding unit that size is 32 × 32 total that comprise Number,Represent Ω_k2In four coding unit of next layer of the coding unit that 2 sizes of jth are 32 × 32 Rate distortion costs value sum；Again for each second set, the value selecting second element from all second combinations is each not Identical all second combinations, will be for Ω_k2The m2 selected second combination is designated asWherein, 1≤m2 ≤ M2≤J2, M2 represent for Ω_k2Total number of the second combination selected,Represent for Ω_k2The m2 selected The rate of four coding unit of next layer of the coding unit that size is 32 × 32 of second element representative in second combination Distortion cost value sum；

The minimum in the rate distortion costs value of the coding unit that all sizes are 16 × 16 is found out from current file Value and maximum, correspondence is designated asWithBy all sizes identical for rate distortion costs value in current file it is Coding unit one the 3rd set of composition of 16 × 16, by all sizes that rate distortion costs value in current file is k3 be 3rd set of the coding unit composition of 16 × 16 is designated as Ω_k3, wherein,Then by each 3rd collection The rate distortion costs value of the coding unit that each size is 16 × 16 in conjunction and corresponding four coding unit of next layer Rate distortion costs value sum composition one the 3rd combination, by Ω_k3In 3 sizes of jth be the coding unit of 16 × 16 The 3rd combination that forms of the rate distortion costs value sum of rate distortion costs value and corresponding four coding unit of next layer be designated asWherein, 1≤j3≤J3, J3 represent Ω_k3In the coding unit that size is 16 × 16 total that comprise Number,Represent Ω_k3In four coding unit of next layer of the coding unit that 3 sizes of jth are 16 × 16 Rate distortion costs value sum；Again for each 3rd set, the value selecting the 3rd element from all 3rd combinations is each not Identical all 3rd combinations, will be for Ω_k3The m3 selected the 3rd combination is designated asWherein, 1≤m3 ≤ M3≤J3, M3 represent for Ω_k3Total number of the 3rd combination selected,Represent for Ω_k3The m3 selected The rate of four coding unit of next layer of the coding unit that size is 16 × 16 of second element representative in 3rd combination Distortion cost value sum；

3.-3, basisWithCalculating for all sizes in the i-th class video sequence is 64 × 64 The unified optimum segmentation threshold value of coding unit, is designated asWherein, T1 represents current In file, all sizes are the unified segmentation threshold of the coding unit of 64 × 64, and R (T1) represents when segmentation threshold is T1 In current file, all sizes are the unified increment rate of the rate distortion costs value of the coding unit of 64 × 64, $R\left(T1\right)=\underset{k1={\mathrm{rd}}_{64\×64}^{\min }}{\overset{T1}{\mathrm{\Σ}}}\underset{m1=1}{\overset{M1}{\mathrm{\Σ}}}f(k1,m1)\×p(w_{\mathrm{Split}},k1,m1)+\underset{k1=T1+1}{\overset{{\mathrm{rd}}_{64\×64}^{\max }}{\mathrm{\Σ}}}\underset{m1=1}{\overset{M1}{\mathrm{\Σ}}}g(k1,m1)\×p(w_{\mathrm{Non}-\mathrm{Split},}k1,m1),$ $f(k1,m1)=(k1-{\left({\mathrm{rd}}_{64\×64}^s\right)}_{m1})/{\left({\mathrm{rd}}_{64\×64}^s\right)}_{m1},$ F (k1, m1) represents and works as ${\mathrm{rd}}_{64\×64}^{\min }\≤k1\≤T1$ TimeCorresponding chi Very little size is the increment rate of the rate distortion costs value of the coding unit of 64 × 64, p (w_Split, k1, m1) represent and work asTimeThe probability of the corresponding coding unit that size is 64 × 64 segmentation, $g(k1,m1)=({\left({\mathrm{rd}}_{64\×64}^s\right)}_{m1}-k1)/k1,$ g (k1, m1) represents and works as $T1+1\≤k1\≤{\mathrm{rd}}_{64\×64}^{\max }$ TimeThe rate distortion of the corresponding coding unit that size is 64 × 64 The increment rate of cost value, p (w_Non-Split, k1, m1) represent and work asTimeCorresponding size It is the ameristic probability of coding unit of 64 × 64,Represent and take the T1 value making R (T1) corresponding time minimum；

According toWithCalculating for all sizes in the i-th class video sequence is the coding list of 32 × 32 The unified optimum segmentation threshold value of unit, is designated asWherein, during T2 represents current file All sizes are the unified segmentation threshold of the coding unit of 32 × 32, and R (T2) represents that segmentation threshold ought be above when being T2 In part, all sizes are the unified increment rate of the rate distortion costs value of the coding unit of 32 × 32, $R\left(T2\right)=\underset{k2={\mathrm{rd}}_{32\×32}^{\min }}{\overset{T2}{\mathrm{\Σ}}}\underset{m2=1}{\overset{M2}{\mathrm{\Σ}}}f(k2,m2)\×p(w_{\mathrm{Split}},k2,m2)+\underset{k2=T2+1}{\overset{{\mathrm{rd}}_{32\×32}^{\max }}{\mathrm{\Σ}}}\underset{m2=1}{\overset{M2}{\mathrm{\Σ}}}g(k2,m2)\×p(w_{\mathrm{Non}-\mathrm{Split},}k2,m2),$ $f(k2,m2)=(k2-{\left({\mathrm{rd}}_{32\×32}^s\right)}_{m2})/{\left({\mathrm{rd}}_{32\×32}^s\right)}_{m2},$ F (k2, m2) represents and works as ${\mathrm{rd}}_{32\×32}^{\min }\≤k2\≤T2$ TimeCorresponding chi Very little size is the increment rate of the rate distortion costs value of the coding unit of 32 × 32, p (w_Split, k2, m2) represent and work asTimeThe probability of the corresponding coding unit that size is 32 × 32 segmentation, $g(k2,m2)=({\left({\mathrm{rd}}_{32\×32}^s\right)}_{m2}-k2)/k2,$ g (k2, m2) represents and works as $T2+1\≤k2\≤{\mathrm{rd}}_{32\×32}^{\max }$ TimeThe rate distortion of the corresponding coding unit that size is 32 × 32 The increment rate of cost value, p (w_Non-Split, k2, m2) represent and work asTimeCorresponding size It is the ameristic probability of coding unit of 32 × 32,Represent and take the T2 value making R (T2) corresponding time minimum；

According toWithCalculating for all sizes in the i-th class video sequence is the coding list of 16 × 16 Unit unified optimum segmentation threshold value, be designated asWherein, during T3 represents current file All sizes are the unified segmentation threshold of the coding unit of 16 × 16, and R (T3) represents that segmentation threshold ought be above when being T3 In part, all sizes are the unified increment rate of the rate distortion costs value of the coding unit of 16 × 16, $R\left(T3\right)=\underset{k3={\mathrm{rd}}_{16\×16}^{\min }}{\overset{T3}{\mathrm{\Σ}}}\underset{m3=1}{\overset{M3}{\mathrm{\Σ}}}f(k3,m3)\×p(w_{\mathrm{Split}},k3,m3)+\underset{k3=T3+1}{\overset{{\mathrm{rd}}_{16\×16}^{\max }}{\mathrm{\Σ}}}\underset{m3=1}{\overset{M3}{\mathrm{\Σ}}}g(k3,m3)\×p(w_{\mathrm{Non}-\mathrm{Split},}k3,m3),$ $f(k3,m3)=(k3-{\left({\mathrm{rd}}_{16\×16}^s\right)}_{m3})/{\left({\mathrm{rd}}_{16\×16}^s\right)}_{m3},$ F (k3, m3) represents and works as ${\mathrm{rd}}_{16\×16}^{\min }\≤k3\≤T3$ TimeCorresponding chi Very little size is the increment rate of the rate distortion costs value of the coding unit of 16 × 16, p (w_Split, k3, m3) represent and work asTimeThe probability of the corresponding coding unit that size is 16 × 16 segmentation, $g(k3,m3)=({\left({\mathrm{rd}}_{16\×16}^s\right)}_{m3}-k3)/k3,$ g (k3, m3) represents and works as $T3+1\≤k3\≤{\mathrm{rd}}_{16\×16}^{\max }$ TimeThe rate distortion of the corresponding coding unit that size is 16 × 16 The increment rate of cost value, p (w_Non-Split, k3, m3) represent and work asTimeCorresponding size It is the ameristic probability of coding unit of 16 × 16,Represent and take the T3 value making R (T3) corresponding time minimum；

-4 3., make i=i+1, by store size in all video sequences of next class be respectively 64 × 64,32 × 32, The file of the statistical data of the coding unit of 16 × 16, as current file, is then back to step and 3.-2 continues executing with, until N' Individual file is all disposed, obtain optimum segmentation threshold value that size in every class video sequence is the coding unit of 64 × 64, Size during size is the optimum segmentation threshold value of the coding unit of 32 × 32 and every class video sequence in every class video sequence Size is the optimum segmentation threshold value of the coding unit of 16 × 16, wherein, in i=i+1 "=" it is assignment.

Described step 4. in determine in pending HEVC HD video size be respectively 64 × 64,32 × 32, The coding unit of 16 × 16 the need of the process continuing segmentation is:

4.-1, the coding unit that all sizes are 64 × 64 in pending HEVC HD video is carried out as follows Process:

4.-1a, the coding unit that currently pending size is 64 × 64 is defined as current coded unit；④- 1b, optimum segmentation threshold value according to the rate distortion costs value of current coded unit and coding unit that size is 64 × 64, Determine that current coded unit, the need of continuing segmentation, if the rate distortion costs value of current coded unit is more than size is The optimum segmentation threshold value of the coding unit of 64 × 64, it is determined that current coded unit continues segmentation, is divided into four sizes It is the coding unit of 32 × 32, then performs step 4.-1c；If the rate distortion costs value of current coded unit is less than or equal to Size is the optimum segmentation threshold value of the coding unit of 64 × 64, it is determined that current coded unit does not continues to segmentation, then Perform step 4.-1c；4.-1c, using the pending coding unit that size is 64 × 64 of the next one as present encoding list Unit, is then back to step 4.-1b and continues executing with, until all sizes in pending HEVC HD video are 64 × 64 Coding unit be disposed, then perform step 4.-2；

4.-2, the coding unit that all sizes are 32 × 32 in pending HEVC HD video is carried out as follows Process:

4.-2a, the coding unit that currently pending size is 32 × 32 is defined as current coded unit；④- 2b, optimum segmentation threshold value according to the rate distortion costs value of current coded unit and coding unit that size is 32 × 32, Determine that current coded unit, the need of continuing segmentation, if the rate distortion costs value of current coded unit is more than size is The optimum segmentation threshold value of the coding unit of 32 × 32, it is determined that current coded unit continues segmentation, is divided into four sizes It is the coding unit of 16 × 16, then performs step 4.-2c；If the rate distortion costs value of current coded unit is less than or equal to Size is the optimum segmentation threshold value of the coding unit of 32 × 32, it is determined that current coded unit does not continues to segmentation, then Perform step 4.-2c；4.-2c, using the pending coding unit that size is 32 × 32 of the next one as present encoding list Unit, is then back to step 4.-2b and continues executing with, until all sizes in pending HEVC HD video are 32 × 32 Coding unit be disposed, then perform step 4.-3；

4.-3, the coding unit that all sizes are 16 × 16 in pending HEVC HD video is carried out as follows Process:

4.-3a, the coding unit that currently pending size is 16 × 16 is defined as current coded unit；④- 3b, optimum segmentation threshold value according to the rate distortion costs value of current coded unit and coding unit that size is 16 × 16, Determine that current coded unit, the need of continuing segmentation, if the rate distortion costs value of current coded unit is more than size is The optimum segmentation threshold value of the coding unit of 16 × 16, it is determined that current coded unit continues segmentation, is divided into four sizes It is the coding unit of 8 × 8, then performs step 4.-3c；If the rate distortion costs value of current coded unit is less than or equal to chi Very little size is the optimum segmentation threshold value of the coding unit of 16 × 16, it is determined that current coded unit does not continues to segmentation, then holds Row step 4.-3c；4.-3c, using the pending coding unit that size is 16 × 16 of the next one as current coded unit, It is then back to step 4.-3b continue executing with, until all sizes in pending HEVC HD video are 16 × 16 Coding unit is disposed.

Compared with prior art, it is an advantage of the current invention that: whole process is divided into off-line training step by the inventive method Operational phase online with threshold value, in each video sequence in off-line training step, first statistics training set, size is divided Be not 64 × 64,32 × 32,16 × 16 the depth value of coding unit, rate distortion costs value, four coding unit of next layer Rate distortion costs value sum, cutting state, then to obtain size in every class video sequence be the system of the coding unit of 64 × 64 The optimum segmentation threshold value of one, size are the unified optimum segmentation threshold value of the coding unit of 32 × 32 and size is The unified optimum segmentation threshold value of the coding unit of 16 × 16, and in the online operational phase of threshold value, according to optimum segmentation threshold value with In pending HEVC HD video, size is respectively the rate distortion generation of the coding unit of 64 × 64,32 × 32,16 × 16 Being worth, determining in pending HEVC HD video that size is respectively the coding unit of 64 × 64,32 × 32,16 × 16 is No needs continues segmentation, and this mode can be saved a large amount of scramble time, i.e. exist in the case of distortion performance loss is less On the premise of ensureing decoded video objective quality, it is possible to efficiently reduce the scramble time required for HD video.

Accompanying drawing explanation

Fig. 1 is the example schematic diagram of HEVC intraframe coding quadtree coding structure；

When Fig. 2 is QP=32 the rate distortion costs value of the coding unit in one video sequence of B class increment rate R (T) and The graph of a relation of the segmentation threshold T of coding unit；

Fig. 3 a be A class a video sequence in size be respectively 64 × 64,32 × 32,16 × 16 coding unit Optimum segmentation threshold value with quantization parameter QP and the graph of a relation of the size variation of coding unit；

Fig. 3 b be B class a video sequence in size be respectively 64 × 64,32 × 32,16 × 16 coding unit Optimum segmentation threshold value with quantization parameter QP and the graph of a relation of the size variation of coding unit；

Fig. 3 c be C class a video sequence in size be respectively 64 × 64,32 × 32,16 × 16 coding unit Optimum segmentation threshold value with quantization parameter QP and the graph of a relation of the size variation of coding unit；

BQTerrace test video sequence is encoded respectively by Fig. 4 for utilizing HEVC standard algorithm and the inventive method Distortion performance comparison diagram；

Fig. 5 is the FB(flow block) of the inventive method.

Detailed description of the invention

Below in conjunction with accompanying drawing embodiment, the present invention is described in further detail.

In HEVC video encoding standard, intraframe coding uses quadtree coding structure, and Fig. 1 gives HEVC intraframe coding The example schematic diagram of quadtree coding structure.HEVC intraframe coding quadtree coding structure uses the coding unit of 4 layers of size, greatly Little by respectively 64 × 64,32 × 32,16 × 16,8 × 8, it is determined that the optimum combination of a coding unit needs the rate distortion of 85 times Cost calculates, and to ask for the rate distortion costs value of minimum, thus brings huge time complexity.When table 1 gives QP=32 The segmentation situation statistical table of the coding unit in one video sequence of B class, from table 1 it follows that work as the big of coding unit The little coding unit having 70.49%, 41.51%, 12.22% when being 64 × 64,32 × 32,16 × 16 respectively is to need not continue to point Cut, if these coding units can be judged in advance, then the substantial amounts of scramble time can be saved.

The segmentation statistical table of the coding unit in one video sequence of B class during table 1QP=32

The size of coding unit	The probability (%) of segmentation	Ameristic probability (%)
			64×64	87.78	12.22
32×32	58.49	41.51

16×16

29.51

70.49

In order on the premise of ensureing decoded video objective quality, it is possible to efficiently reduce the volume required for HD video The code time, the present invention proposes a kind of HD video fast encoding method, and it is divided into off-line training step and threshold value to use online Two stages, wherein step 1., step be 3. 2. off-line training step with step, 4. step uses online for threshold value, by step Rapid the most 1., step 2. 3. can obtain the optimum segmentation threshold value of the size for coding unit with step, by step 4. by optimum During segmentation threshold is applied to actual coding.The FB(flow block) of the inventive method is as it is shown in figure 5, it specifically includes following step Rapid:

1., from the standard test sequences of at least N' type of HEVC HD video, N number of video sequence is the most arbitrarily selected Selected all video sequences are constituted a training set by row, and this training set comprises N' × N number of video sequence, wherein, 3≤ N'≤6, N >=1.

At present, HEVC HD video standard test sequences has A, B, C, D, E and F six standard test sequences of type, 2 video sequences can be respectively selected from the standard test sequences of this six type, the most altogether by 12 video sequences during practical operation Row composing training collection.

First, determine that in each video sequence in training set, size is respectively 64 × 64,32 × 32,16 × 16 The cutting state of coding unit；Then, in each video sequence in statistics training set, size is respectively 64 × 64,32 The depth value of the coding unit of × 32,16 × 16, rate distortion costs value, four coding unit of next layer rate distortion costs value it With, cutting state；Then, it is respectively 64 × 64,32 by training set belongs to size in of a sort all video sequences The statistical data of the coding unit of × 32,16 × 16 is stored in a file, total N' file, will store the institute of the i-th class The file of statistical data having the coding unit that size in video sequence is respectively 64 × 64,32 × 32,16 × 16 is designated as F_i, wherein, 1≤i≤N'.Here, HEVC HD video has four layers of coding unit, size is 64 × 64 respectively, 32 × 32, 16 × 16,8 × 8, the size of minimum coding unit is 8 × 8, has only to statistics size in the methods of the invention and is The depth value of 64 × 64,32 × 32,16 × 16 these three layers of coding units, rate distortion costs value, next layer four coding unit Rate distortion costs value sum, cutting state.

In this particular embodiment, step detailed process 2. is:

2.-1, video sequence currently pending in training set is defined as current video sequence.

2.-2, the coding unit that all sizes are 64 × 64 in current video sequence is handled as follows:

2.-2a, the coding unit that currently pending size is 64 × 64 is defined as current coded unit；②- 2b, according to the rate distortion costs value of current coded unit and the rate distortion generation of four coding unit of next layer of current coded unit It is worth sum, determines the cutting state of current coded unit, if the rate distortion costs value of current coded unit is more than or equal to The rate distortion costs value sum of four coding unit of next layer of current coded unit, it is determined that current coded unit continues to divide Cutting, being divided into four sizes is the coding unit of 32 × 32, and determines that the cutting state of current coded unit is for splitting, so Rear execution step 2.-2c；If the rate distortion costs value of current coded unit is less than four, next layer volume of current coded unit The rate distortion costs value sum of code unit, it is determined that current coded unit does not continues to segmentation, and determines current coded unit Cutting state, for not split, then performs step 2.-2c；2.-2c, by the pending volume that size is 64 × 64 of the next one Code unit, as current coded unit, is then back to step 2.-2b and continues executing with, until all sizes in current video sequence Size be 64 × 64 coding unit be disposed, then perform step 2.-3.

The coding unit that a size is 64 × 64 in one video sequence of B class when table 2 gives QP=32 Depth value, rate distortion costs value, the rate distortion costs value sum of four coding unit of next layer and cutting state.

The degree of depth of the coding unit that a size is 64 × 64 in one video sequence of B class during table 2QP=32 Value, rate distortion costs value, the rate distortion costs value sum of four coding unit of next layer and cutting state list

Table 2 illustrates when the size of coding unit is 64 × 64, if the rate distortion costs value of this coding unit is 91024, when the rate distortion costs value sum of its four coding units of next layer is 90429, due to 91024 > 90429, the most really This coding unit fixed needs to continue segmentation；If the rate distortion costs value of this coding unit is 90898, its four, next layer coding When the rate distortion costs value sum of unit is 90961, due to 90898 < 90961, therefore this coding unit needs not continue to segmentation.

2.-3, the coding unit that all sizes are 32 × 32 in current video sequence is handled as follows:

2.-3a, the coding unit that currently pending size is 32 × 32 is defined as current coded unit；②- 3b, according to the rate distortion costs value of current coded unit and the rate distortion generation of four coding unit of next layer of current coded unit It is worth sum, determines the cutting state of current coded unit, if the rate distortion costs value of current coded unit is more than or equal to The rate distortion costs value sum of four coding unit of next layer of current coded unit, it is determined that current coded unit continues to divide Cutting, being divided into four sizes is the coding unit of 16 × 16, and determines that the cutting state of current coded unit is for splitting, so Rear execution step 2.-3c；If the rate distortion costs value of current coded unit is less than four, next layer volume of current coded unit The rate distortion costs value sum of code unit, it is determined that current coded unit does not continues to segmentation, and determines current coded unit Cutting state, for not split, then performs step 2.-3c；2.-3c, by the pending volume that size is 32 × 32 of the next one Code unit, as current coded unit, is then back to step 2.-3b and continues executing with, until all sizes in current video sequence Size be 32 × 32 coding unit be disposed, then perform step 2.-4.

2.-4, the coding unit that all sizes are 16 × 16 in current video sequence is handled as follows:

2.-4a, the coding unit that currently pending size is 16 × 16 is defined as current coded unit；②- 4b, according to the rate distortion costs value of current coded unit and the rate distortion generation of four coding unit of next layer of current coded unit It is worth sum, determines the cutting state of current coded unit, if the rate distortion costs value of current coded unit is more than or equal to The rate distortion costs value sum of four coding unit of next layer of current coded unit, it is determined that current coded unit continues to divide Cutting, being divided into four sizes is the coding unit of 8 × 8, and determines that the cutting state of current coded unit is for splitting, then Perform step 2.-4c；If the rate distortion costs value of current coded unit is less than four, next layer coding of current coded unit The rate distortion costs value sum of unit, it is determined that current coded unit does not continues to segmentation, and determines dividing of current coded unit State of cutting, for not split, then performs step 2.-4c；2.-4c, by the pending coding that size is 16 × 16 of the next one Unit, as current coded unit, is then back to step 2.-4b and continues executing with, until all sizes in current video sequence are big Little be 16 × 16 coding unit be disposed, then perform step 2.-5.

2.-5, in statistics current video sequence, size is respectively the coding unit of 64 × 64,32 × 32,16 × 16 Depth value, rate distortion costs value, the rate distortion costs value sum of four coding unit of next layer, cutting state.In HEVC high definition In video, size be the depth value of the coding unit of 64 × 64 be 0, size is the degree of depth of the coding unit of 32 × 32 Value is 1, size be the depth value of the coding unit of 16 × 16 be 2, size is the depth value of the coding unit of 8 × 8 It is 4.

2.-6, using pending video sequence next in training set as current video sequence, be then back to step 2.- 2 continue executing with, until all video sequences in training set are disposed, obtain size in each video sequence in training set Size is respectively the statistical data of the coding unit of 64 × 64,32 × 32,16 × 16.

2.-7, by training set belongs to size in of a sort all video sequences be respectively 64 × 64,32 × 32, The statistical data of the coding unit of 16 × 16 is stored in a file, total N' file, will store all videos of the i-th class In sequence, the file of the statistical data that size is respectively the coding unit of 64 × 64,32 × 32,16 × 16 is designated as F_i, its In, 1≤i≤N'.

3. according to step 2. in be stored in the statistical data in each file, obtaining size in every class video sequence is In the unified optimum segmentation threshold value of the coding unit of 64 × 64, every class video sequence, size is the coding unit of 32 × 32 Unified optimum segmentation threshold value and every class video sequence in size be 16 × 16 the unified optimum of coding unit Segmentation threshold.

In this particular embodiment, step detailed process 3. is:

3.-1, it is respectively 64 × 64,32 × 32,16 × 16 by storing size in all video sequences of the i-th class The file F of the statistical data of coding unit_iAs current file, wherein, 1≤i≤N'.

3.-2, find out from current file in the rate distortion costs value of the coding unit that all sizes are 64 × 64 Minima and maximum, correspondence is designated asWithBy big for all sizes identical for rate distortion costs value in current file Little be 64 × 64 coding unit composition one first set, by big for all sizes that rate distortion costs value in current file is k1 First set of the little coding unit composition being 64 × 64 is designated as Ω_k1, wherein,Then by each The rate distortion costs value of the coding unit that each size is 64 × 64 in one set and corresponding four, next layer coding Rate distortion costs value sum one first combination of composition of unit, by Ω_k1In 1 size of jth be the coding of 64 × 64 The first combination that the rate distortion costs value of unit forms with the rate distortion costs value sum of corresponding four coding unit of next layer It is designated asWherein, 1≤j1≤J1, J1 represent Ω_k1In the coding unit that size is 64 × 64 that comprises Total number,Represent Ω_k1In four, next layer coding of the coding unit that 1 size of jth is 64 × 64 The rate distortion costs value sum of unit；Again for each first set, from all first combinations, select the value of second element Different all first combinations, will be for Ω_k1The m1 selected first combination is designated asWherein, 1 ≤ m1≤M1≤J1, M1 represent for Ω_k1Total number of the first combination selected,Represent for Ω_k1Selected Four coding unit of next layer of the coding unit that size is 64 × 64 of second element representative in m1 first combination Rate distortion costs value sum.

The minimum in the rate distortion costs value of the coding unit that all sizes are 32 × 32 is found out from current file Value and maximum, correspondence is designated asWithBy all sizes identical for rate distortion costs value in current file it is Coding unit one second set of composition of 32 × 32, by all sizes that rate distortion costs value in current file is k2 be Second set of the coding unit composition of 32 × 32 is designated as Ω_k2, wherein,Then by each second collection The rate distortion costs value of the coding unit that each size is 32 × 32 in conjunction and corresponding four coding unit of next layer Rate distortion costs value sum composition one second combination, by Ω_k2In 2 sizes of jth be the coding unit of 32 × 32 The second combination that forms of the rate distortion costs value sum of rate distortion costs value and corresponding four coding unit of next layer be designated asWherein, 1≤j2≤J2, J2 represent Ω_k2In the coding unit that size is 32 × 32 total that comprise Number,Represent Ω_k2In four coding unit of next layer of the coding unit that 2 sizes of jth are 32 × 32 Rate distortion costs value sum；Again for each second set, the value selecting second element from all second combinations is each not Identical all second combinations, will be for Ω_k2The m2 selected second combination is designated asWherein, 1≤m2 ≤ M2≤J2, M2 represent for Ω_k2Total number of the second combination selected,Represent for Ω_k2The m2 selected The rate of four coding unit of next layer of the coding unit that size is 32 × 32 of second element representative in second combination Distortion cost value sum.

The minimum in the rate distortion costs value of the coding unit that all sizes are 16 × 16 is found out from current file Value and maximum, correspondence is designated asWithBy all sizes identical for rate distortion costs value in current file it is Coding unit one the 3rd set of composition of 16 × 16, by all sizes that rate distortion costs value in current file is k3 be 3rd set of the coding unit composition of 16 × 16 is designated as Ω_k3, wherein,Then by each 3rd collection The rate distortion costs value of the coding unit that each size is 16 × 16 in conjunction and corresponding four coding unit of next layer Rate distortion costs value sum composition one the 3rd combination, by Ω_k3In 3 sizes of jth be the coding unit of 16 × 16 The 3rd combination that forms of the rate distortion costs value sum of rate distortion costs value and corresponding four coding unit of next layer be designated asWherein, 1≤j3≤J3, J3 represent Ω_k3In the coding unit that size is 16 × 16 total that comprise Number,Represent Ω_k3In four coding unit of next layer of the coding unit that 3 sizes of jth are 16 × 16 Rate distortion costs value sum；Again for each 3rd set, the value selecting the 3rd element from all 3rd combinations is each not Identical all 3rd combinations, will be for Ω_k3The m3 selected the 3rd combination is designated asWherein, 1≤m3 ≤ M3≤J3, M3 represent for Ω_k3Total number of the 3rd combination selected,Represent for Ω_k3The m3 selected The rate of four coding unit of next layer of the coding unit that size is 16 × 16 of second element representative in 3rd combination Distortion cost value sum.

3.-3, basisWithCalculating for all sizes in the i-th class video sequence is the volume of 64 × 64 The unified optimum segmentation threshold value of code unit, is designated asWherein, represent ought be above for T1 In part, all sizes are the unified segmentation threshold of the coding unit of 64 × 64, and R (T1) represents and works as when segmentation threshold is T1 In preceding document, all sizes are the unified increment rate of the rate distortion costs value of the coding unit of 64 × 64, $R\left(T1\right)=\underset{k1={\mathrm{rd}}_{64\&times;64}^{\min }}{\overset{T1}{\mathrm{\&Sigma;}}}\underset{m1=1}{\overset{M1}{\mathrm{\&Sigma;}}}f(k1,m1)\&times;p(w_{\mathrm{Split}},k1,m1)+\underset{k1=T1+1}{\overset{{\mathrm{rd}}_{64\&times;64}^{\max }}{\mathrm{\&Sigma;}}}\underset{m1=1}{\overset{M1}{\mathrm{\&Sigma;}}}g(k1,m1)\&times;p(w_{\mathrm{Non}-\mathrm{Split},}k1,m1),$ $f(k1,m1)=(k1-{\left({\mathrm{rd}}_{64\&times;64}^s\right)}_{m1})/{\left({\mathrm{rd}}_{64\&times;64}^s\right)}_{m1},$ F (k1, m1) represents and works as ${\mathrm{rd}}_{64\&times;64}^{\min }\&le;k1\&le;T1$ TimeCorresponding chi Very little size is the increment rate of the rate distortion costs value of the coding unit of 64 × 64, p (w_Split, k1, m1) represent and work asTimeThe probability of the corresponding coding unit that size is 64 × 64 segmentation, $g(k1,m1)=({\left({\mathrm{rd}}_{64\&times;64}^s\right)}_{m1}-k1)/k1,$ g (k1, m1) represents and works as $T1+1\&le;k1\&le;{\mathrm{rd}}_{64\&times;64}^{\max }$ TimeThe rate distortion of the corresponding coding unit that size is 64 × 64 The increment rate of cost value, p (w_Non-Split, k1, m1) represent and work asTimeCorresponding size is 64 The ameristic probability of coding unit of × 64,Represent and take the T1 value making R (T1) corresponding time minimum.

According toWithCalculating for all sizes in the i-th class video sequence is the coding list of 32 × 32 The unified optimum segmentation threshold value of unit, is designated asWherein, during T2 represents current file All sizes are the unified segmentation threshold of the coding unit of 32 × 32, and R (T2) represents that segmentation threshold ought be above when being T2 In part, all sizes are the unified increment rate of the rate distortion costs value of the coding unit of 32 × 32, $R\left(T2\right)=\underset{k2={\mathrm{rd}}_{32\&times;32}^{\min }}{\overset{T2}{\mathrm{\&Sigma;}}}\underset{m2=1}{\overset{M2}{\mathrm{\&Sigma;}}}f(k2,m2)\&times;p(w_{\mathrm{Split}},k2,m2)+\underset{k2=T2+1}{\overset{{\mathrm{rd}}_{32\&times;32}^{\max }}{\mathrm{\&Sigma;}}}\underset{m2=1}{\overset{M2}{\mathrm{\&Sigma;}}}g(k2,m2)\&times;p(w_{\mathrm{Non}-\mathrm{Split},}k2,m2),$ $f(k2,m2)=(k2-{\left({\mathrm{rd}}_{32\&times;32}^s\right)}_{m2})/{\left({\mathrm{rd}}_{32\&times;32}^s\right)}_{m2},$ F (k2, m2) represents and works as ${\mathrm{rd}}_{32\&times;32}^{\min }\&le;k2\&le;T2$ TimeCorresponding chi Very little size is the increment rate of the rate distortion costs value of the coding unit of 32 × 32, p (w_Split, k2, m2) represent and work asTimeThe probability of the corresponding coding unit that size is 32 × 32 segmentation, $g(k2,m2)=({\left({\mathrm{rd}}_{32\&times;32}^s\right)}_{m2}-k2)/k2,$ g (k2, m2) represents and works as $T2+1\&le;k2\&le;{\mathrm{rd}}_{32\&times;32}^{\max }$ TimeThe rate distortion of the corresponding coding unit that size is 32 × 32 The increment rate of cost value, p (w_Non-Split, k2, m2) represent and work asTimeCorresponding size It is the ameristic probability of coding unit of 32 × 32,Represent and take the T2 value making R (T2) corresponding time minimum.

According toWithCalculating for all sizes in the i-th class video sequence is the coding list of 16 × 16 Unit unified optimum segmentation threshold value, be designated asWherein, during T3 represents current file All sizes are the unified segmentation threshold of the coding unit of 16 × 16, and R (T3) represents that segmentation threshold ought be above when being T3 In part, all sizes are the unified increment rate of the rate distortion costs value of the coding unit of 16 × 16, $R\left(T3\right)=\underset{k3={\mathrm{rd}}_{16\&times;16}^{\min }}{\overset{T3}{\mathrm{\&Sigma;}}}\underset{m3=1}{\overset{M3}{\mathrm{\&Sigma;}}}f(k3,m3)\&times;p(w_{\mathrm{Split}},k3,m3)+\underset{k3=T3+1}{\overset{{\mathrm{rd}}_{16\&times;16}^{\max }}{\mathrm{\&Sigma;}}}\underset{m3=1}{\overset{M3}{\mathrm{\&Sigma;}}}g(k3,m3)\&times;p(w_{\mathrm{Non}-\mathrm{Split},}k3,m3),$ $f(k3,m3)=(k3-{\left({\mathrm{rd}}_{16\&times;16}^s\right)}_{m3})/{\left({\mathrm{rd}}_{16\&times;16}^s\right)}_{m3},$ F (k3, m3) represents and works as ${\mathrm{rd}}_{16\&times;16}^{\min }\&le;k3\&le;T3$ TimeCorresponding chi Very little size is the increment rate of the rate distortion costs value of the coding unit of 16 × 16, p (w_Split, k3, m3) represent and work asTimeThe probability of the corresponding coding unit that size is 16 × 16 segmentation, $g(k3,m3)=({\left({\mathrm{rd}}_{16\&times;16}^s\right)}_{m3}-k3)/k3,$ G (k3, m3) represents and works as $T3+1\&le;k3\&le;{\mathrm{rd}}_{16\&times;16}^{\max }$ TimeThe rate of the corresponding coding unit that size is 16 × 16 The increment rate of distortion cost value, p (w_Non-Split, k3, m3) represent and work asTimeCorresponding size Size is the ameristic probability of coding unit of 16 × 16,Represent and take the T3 value making R (T3) corresponding time minimum.

-4 3., make i=i+1, by store size in all video sequences of next class be respectively 64 × 64,32 × 32, The file of the statistical data of the coding unit of 16 × 16, as current file, is then back to step and 3.-2 continues executing with, until N' Individual file is all disposed, obtain optimum segmentation threshold value that size in every class video sequence is the coding unit of 64 × 64, Size during size is the optimum segmentation threshold value of the coding unit of 32 × 32 and every class video sequence in every class video sequence Size is the optimum segmentation threshold value of the coding unit of 16 × 16, wherein, in i=i+1 "=" it is assignment.

Increment rate R (T) of the rate distortion costs value of the coding unit in one video sequence of B class when Fig. 2 gives QP=32 With the graph of a relation of the segmentation threshold T of coding unit, by the segmentation threshold T of coding unit from minimum rate distortion costs value to maximum Rate distortion costs value travels through, and obtains rate distortion costs value when making R (T) minimum, is optimum segmentation threshold value, if a volume The rate distortion costs value of code unit is less than corresponding optimum segmentation threshold value, then this coding unit is not split, and is otherwise divided into 4 Smaller coding unit.During in order to quantization parameter QP is respectively 22,27,32,37, the coding unit of each size is Excellent segmentation threshold fits to a smooth curve, needs to use QP cubic polynomial to be fitted, size is respectively 64 × 64, the matrix that the optimum segmentation threshold value of the coding unit of 32 × 32,16 × 16 is constituted is designated as $\left(\begin{array}{ccc}{T}_{64\&times;64}^{\mathrm{opt}}& T_{32\&times;32}^{\mathrm{opt}}& T_{16\&times;16}^{\mathrm{opt}}\end{array}\right),$ ${\left(\begin{array}{ccc}{T}_{64\&times;64}^{\mathrm{opt}}& T_{32\&times;32}^{\mathrm{opt}}& T_{16\&times;16}^{\mathrm{opt}}\end{array}\right)}^T=P\&CenterDot;{\left(\begin{array}{cccc}{\mathrm{QP}}^3& {\mathrm{QP}}^2& \mathrm{QP}& 1\end{array}\right)}^T,$ P represents correlation matrix, QP³Represent quantization parameter The cube of QP, other by that analogy, $P=\left(\begin{array}{cccc}{a}_{64}& b_{64}& c_{64}& d_{64}\\ a_{32}& b_{32}& c_{32}& d_{32}\\ a_{16}& b_{16}& c_{16}& d_{16}\end{array}\right),$ Wherein, a₆₄,b₆₄,c₆₄,d₆₄Represent that size is The optimum segmentation threshold value of the coding unit of 64 × 64Coefficient, a₃₂,b₃₂,c₃₂,d₃₂Represent that size is the volume of 32 × 32 The optimum segmentation threshold value of code unitCoefficient, a₁₆,b₁₆,c₁₆,d₁₆Represent that coding unit that size is 16 × 16 is Excellent segmentation thresholdCoefficient.Fig. 3 a give size in a video sequence of A class be respectively 64 × 64,32 × 32, the optimum segmentation threshold value of the coding unit of 16 × 16 is with quantization parameter QP and the graph of a relation of the size variation of coding unit, figure 3b gives size in a video sequence of B class and is respectively the optimum of the coding unit of 64 × 64,32 × 32,16 × 16 Segmentation threshold is with quantization parameter QP and the graph of a relation of the size variation of coding unit, and Fig. 3 c gives a video sequence of C class Middle size is respectively the optimum segmentation threshold value of the coding unit of 64 × 64,32 × 32,16 × 16 with quantization parameter QP and volume The graph of a relation of the size variation of code unit, obtains correlation matrix P according to curve-fitting results, specific as follows:

4. the coding unit that each size is 64 × 64 in pending HEVC HD video carries out traversal compile Code, in traversal cataloged procedure, it is first determined the type of pending HEVC HD video, is basis in actual process The resolution of pending HEVC HD video judges that pending HEVC HD video belongs to A, B, C, D, E and F six kind Any type in type；Then according to the type of pending HEVC HD video, pending HEVC HD video is determined Middle size is respectively the optimum segmentation threshold value of the coding unit of 64 × 64,32 × 32,16 × 16；Then according to pending In HEVC HD video, size is respectively the rate distortion costs value of the coding unit of 64 × 64,32 × 32,16 × 16 and right The optimum segmentation threshold value answered, determines that in pending HEVC HD video, size is respectively 64 × 64,32 × 32,16 × 16 Coding unit the need of continue segmentation.

In this particular embodiment, step 4. in determine in pending HEVC HD video size be respectively 64 × 64, the coding unit of 32 × 32,16 × 16 the need of the process continuing segmentation is:

4.-1, the coding unit that all sizes are 64 × 64 in pending HEVC HD video is carried out as follows Process:

4.-1a, the coding unit that currently pending size is 64 × 64 is defined as current coded unit；④- 1b, optimum segmentation threshold value according to the rate distortion costs value of current coded unit and coding unit that size is 64 × 64, Determine that current coded unit, the need of continuing segmentation, if the rate distortion costs value of current coded unit is more than size is The optimum segmentation threshold value of the coding unit of 64 × 64, it is determined that current coded unit continues segmentation, is divided into four sizes It is the coding unit of 32 × 32, then performs step 4.-1c；If the rate distortion costs value of current coded unit is less than or equal to Size is the optimum segmentation threshold value of the coding unit of 64 × 64, it is determined that current coded unit does not continues to segmentation, then Perform step 4.-1c；4.-1c, using the pending coding unit that size is 64 × 64 of the next one as present encoding list Unit, is then back to step 4.-1b and continues executing with, until all sizes in pending HEVC HD video are 64 × 64 Coding unit be disposed, then perform step 4.-2.

4.-2, the coding unit that all sizes are 32 × 32 in pending HEVC HD video is carried out as follows Process:

4.-2a, the coding unit that currently pending size is 32 × 32 is defined as current coded unit；④- 2b, optimum segmentation threshold value according to the rate distortion costs value of current coded unit and coding unit that size is 32 × 32, Determine that current coded unit, the need of continuing segmentation, if the rate distortion costs value of current coded unit is more than size is The optimum segmentation threshold value of the coding unit of 32 × 32, it is determined that current coded unit continues segmentation, is divided into four sizes It is the coding unit of 16 × 16, then performs step 4.-2c；If the rate distortion costs value of current coded unit is less than or equal to Size is the optimum segmentation threshold value of the coding unit of 32 × 32, it is determined that current coded unit does not continues to segmentation, then Perform step 4.-2c；4.-2c, using the pending coding unit that size is 32 × 32 of the next one as present encoding list Unit, is then back to step 4.-2b and continues executing with, until all sizes in pending HEVC HD video are 32 × 32 Coding unit be disposed, then perform step 4.-3.

4.-3, the coding unit that all sizes are 16 × 16 in pending HEVC HD video is carried out as follows Process:

4.-3a, the coding unit that currently pending size is 16 × 16 is defined as current coded unit；④- 3b, optimum segmentation threshold value according to the rate distortion costs value of current coded unit and coding unit that size is 16 × 16, Determine that current coded unit, the need of continuing segmentation, if the rate distortion costs value of current coded unit is more than size is The optimum segmentation threshold value of the coding unit of 16 × 16, it is determined that current coded unit continues segmentation, is divided into four sizes It is the coding unit of 8 × 8, then performs step 4.-3c；If the rate distortion costs value of current coded unit is less than or equal to chi Very little size is the optimum segmentation threshold value of the coding unit of 16 × 16, it is determined that current coded unit does not continues to segmentation, then holds Row step 4.-3c；4.-3c, using the pending coding unit that size is 16 × 16 of the next one as current coded unit, It is then back to step 4.-3b continue executing with, until all sizes in pending HEVC HD video are 16 × 16 Coding unit is disposed.

In order to verify the performance of the inventive method, use HEVC common test environment " I frame high efficient coding in frame ", On the server of InterCore i5-2500 second filial generation processor 3.30GHZ, 8GB internal memory, profit on HM9.2rc1 software platform The statistical result under each video sequence coding 100 frame condition is obtained by the inventive method.List A class respectively two regard table 3 Four video sequences of frequency sequence, five video sequences of B class and C class are 22,27,32,37 times at QP, are respectively adopted the present invention Method and HM canonical algorithm carry out encoding at time and BDBR(Delta bit rate) and BDPSNR(Delta peak signal-to-noise rate) on comparative result.

Table 3 is 22,27,32,37 times at QP, is respectively adopted the inventive method and each video sequence is carried out by HM canonical algorithm Coding comparative result on time and BDBR and BDPSNR

From table 3 it is observed that the inventive method can averagely save the scramble time of 31.71%, the simultaneously increase of BDBR The least on the impact of the objective quality of video with reducing of BDPSNR, this is because the resolution of C class video sequence is low, great majority Coding unit be required for being divided into four smaller coding units of next layer, so for the volume of C class video sequence The saving of code time is not as A class and B class video sequence.

Fig. 4 gives and utilizes HEVC standard algorithm and the inventive method to carry out BQTerrace test video sequence respectively The distortion performance of coding compares.Figure 4, it is seen that utilize the inventive method that BQTerrace test video sequence is entered The distortion performance curve of row coding loses with the rate utilizing HEVC standard algorithm to encode BQTerrace test video sequence True performance curve essentially coincides, and this fully indicates the inventive method and has almost identical rate distortion with HEVC standard algorithm Energy.

Binding analysis table 3 and Fig. 4, it is known that the inventive method can on the premise of conservation rate distortion performance is almost unchanged, The scramble time of average saving 31.71%, greatly reduce the complexity of coding.

Claims (3)

1. a HD video fast encoding method, it is characterised in that comprise the following steps:

1., from the standard test sequences of at least N' type of HEVC HD video, N number of video sequence is the most arbitrarily selected, will Selected all video sequences constitute a training set, and this training set comprises N' × N number of video sequence, wherein, 3≤N'≤6, N≥1；

First, determine that in each video sequence in training set, size is respectively the volume of 64 × 64,32 × 32,16 × 16 The cutting state of code unit；Then, in each video sequence in statistics training set size be respectively 64 × 64,32 × 32, the depth value of the coding unit of 16 × 16, rate distortion costs value, four coding unit of next layer rate distortion costs value it With, cutting state；Then, it is respectively 64 × 64,32 by training set belongs to size in of a sort all video sequences The statistical data of the coding unit of × 32,16 × 16 is stored in a file, total N' file, will store the institute of the i-th class The file of statistical data having the coding unit that size in video sequence is respectively 64 × 64,32 × 32,16 × 16 is designated as F_i, wherein, 1≤i≤N'；

3. according to step 2. in be stored in the statistical data in each file, obtain size in every class video sequence be 64 × In the unified optimum segmentation threshold value of the coding unit of 64, every class video sequence, size is the system of the coding unit of 32 × 32 In the optimum segmentation threshold value of one and every class video sequence, size is the unified optimum segmentation of the coding unit of 16 × 16 Threshold value；

Described step detailed process 3. is:

3.-1, it is respectively the coding of 64 × 64,32 × 32,16 × 16 by storing size in all video sequences of the i-th class The file F of the statistical data of unit_iAs current file, wherein, 1≤i≤N'；

3. from current file ,-2, find out the minima in the rate distortion costs value of the coding unit that all sizes are 64 × 64 And maximum, correspondence is designated asWithBy all sizes identical for rate distortion costs value in current file be 64 × All sizes that rate distortion costs value in current file is k1 are 64 × 64 by coding unit one first set of composition of 64 Coding unit composition first set be designated as Ω_k1, wherein,Then by each first set in every Individual size is the rate distortion costs value rate distortion generation with corresponding four coding unit of next layer of the coding unit of 64 × 64 It is worth sum one first combination of composition, by Ω_k1In the rate distortion costs of the coding unit that 1 size of jth is 64 × 64 The first combination that value forms with the rate distortion costs value sum of corresponding four coding unit of next layer is designated as Wherein, 1≤j1≤J1, J1 represent Ω_k1In total number of the coding unit that size is 64 × 64 of comprising,Table Show Ω_k1In the coding unit that 1 size of jth is 64 × 64 four coding unit of next layer rate distortion costs value it With；Again for each first set, select from all first combinations that the value of second element is different all first group Close, will be for Ω_k1The m1 selected first combination is designated asWherein, 1≤m1≤M1≤J1, M1 represent pin To Ω_k1Total number of the first combination selected,Represent for Ω_k1In the m1 selected first combination second The rate distortion costs value sum of four coding unit of next layer of the coding unit that size is 64 × 64 that element represents；

Find out from current file minima in the rate distortion costs value of the coding unit that all sizes are 32 × 32 and Maximum, correspondence is designated asWithBy all sizes identical for rate distortion costs value in current file be 32 × Coding unit one second set of composition of 32, by all sizes that rate distortion costs value in current file is k2 be 32 × Second set of the coding unit composition of 32 is designated as Ω_k2, wherein,Then by each second set The rate of rate distortion costs value and corresponding four coding unit of next layer of the coding unit that each size is 32 × 32 One second combination of composition of distortion cost value sum, by Ω_k2In the rate of the coding unit that 2 sizes of jth are 32 × 32 The second combination that distortion cost value forms with the rate distortion costs value sum of corresponding four coding unit of next layer is designated asWherein, 1≤j2≤J2, J2 represent Ω_k2In the coding unit that size is 32 × 32 total that comprise Number,Represent Ω_k2In four coding unit of next layer of the coding unit that 2 sizes of jth are 32 × 32 Rate distortion costs value sum；Again for each second set, the value selecting second element from all second combinations is each not Identical all second combinations, will be for Ω_k2The m2 selected second combination is designated asWherein, 1≤m2 ≤ M2≤J2, M2 represent for Ω_k2Total number of the second combination selected,Represent for Ω_k2The m2 selected The rate of four coding unit of next layer of the coding unit that size is 32 × 32 of second element representative in second combination Distortion cost value sum；

The minima in the rate distortion costs value of the coding unit that all sizes are 16 × 16 and is found out from current file Big value, correspondence is designated asWithIt is 16 × 16 by all sizes identical for rate distortion costs value in current file Coding unit one the 3rd set of composition, by the volume that all sizes are 16 × 16 that rate distortion costs value in current file is k3 3rd set of code unit composition is designated as Ω_k3, wherein,Then by each chi in each 3rd set Very little size is the rate distortion costs value rate distortion costs value with corresponding four coding unit of next layer of the coding unit of 16 × 16 Sum one the 3rd combination of composition, by Ω_k3In the coding unit that 3 sizes of jth are 16 × 16 rate distortion costs value with 3rd combination of the rate distortion costs value sum composition of corresponding four coding unit of next layer is designated asWherein, 1≤j3≤J3, J3 represent Ω_k3In total number of the coding unit that size is 16 × 16 of comprising,Represent Ω_k3In the coding unit that 3 sizes of jth are 16 × 16 four coding unit of next layer rate distortion costs value it With；Again for each 3rd set, select from all 3rd combinations that the value of the 3rd element is different all 3rd group Close, will be for Ω_k3The m3 selected the 3rd combination is designated asWherein, 1≤m3≤M3≤J3, M3 represent pin To Ω_k3Total number of the 3rd combination selected,Represent for Ω_k3In a m3 selected the 3rd combination second The rate distortion costs value sum of four coding unit of next layer of the coding unit that size is 16 × 16 that element represents；

3.-3, basisWithCalculating for all sizes in the i-th class video sequence is the coding list of 64 × 64 The unified optimum segmentation threshold value of unit, is designated as Wherein, institute during T1 represents current file The unified segmentation threshold of the coding unit having size to be 64 × 64, R (T1) represents current file when segmentation threshold is T1 In all sizes be 64 × 64 the unified increment rate of rate distortion costs value of coding unit, F (k1, m1) represents and works asTimeCorresponding size Size is the increment rate of the rate distortion costs value of the coding unit of 64 × 64, p (w_Split, k1, m1) represent and work asTimeThe probability of the corresponding coding unit that size is 64 × 64 segmentation,g (k1, m1) represents and works asTimeThe rate distortion of the corresponding coding unit that size is 64 × 64 The increment rate of cost value, p (w_Non-Split, k1, m1) represent and work asTimeCorresponding size It is the ameristic probability of coding unit of 64 × 64,Represent and take the T1 value making R (T1) corresponding time minimum；

According toWithCalculating for all sizes in the i-th class video sequence is the coding unit of 32 × 32 Unified optimum segmentation threshold value, is designated as Wherein, all chis during T2 represents current file Very little size is the unified segmentation threshold of the coding unit of 32 × 32, institute in current file when R (T2) expression segmentation threshold is T2 The unified increment rate of the rate distortion costs value of the coding unit having size to be 32 × 32, F (k2, m2) represents and works asTimeCorresponding chi Very little size is the increment rate of the rate distortion costs value of the coding unit of 32 × 32, p (w_Split, k2, m2) represent and work asTimeThe probability of the corresponding coding unit that size is 32 × 32 segmentation,g (k2, m2) represents and works asTimeThe rate distortion of the corresponding coding unit that size is 32 × 32 The increment rate of cost value, p (w_Non-Split, k2, m2) represent and work asCorresponding size It is the ameristic probability of coding unit of 32 × 32,Represent and take the T2 value making R (T2) corresponding time minimum；

According toWithCalculate for the coding unit that all sizes in the i-th class video sequence are 16 × 16 unified Optimum segmentation threshold value, be designated as Wherein, all sizes during T3 represents current file Size is the unified segmentation threshold of the coding unit of 16 × 16, and R (T3) represents all in current file when segmentation threshold is T3 Size is the unified increment rate of the rate distortion costs value of the coding unit of 16 × 16, F (k3, m3) represents and works asTimeCorresponding Size is the increment rate of the rate distortion costs value of the coding unit of 16 × 16, p (w_Split, k3, m3) represent and work as TimeThe probability of the corresponding coding unit that size is 16 × 16 segmentation, G (k3, m3) represents and works asTimeThe rate of the corresponding coding unit that size is 16 × 16 is lost The increment rate of true cost value, p (w_Non-Split, k3, m3) represent and work asTimeCorresponding size It is the ameristic probability of coding unit of 16 × 16,Represent and take the T3 value making R (T3) corresponding time minimum；

-4 3., make i=i+1, be respectively 64 × 64,32 × 32,16 by storing size in all video sequences of next class The file of the statistical data of the coding unit of × 16, as current file, is then back to step and 3.-2 continues executing with, until N' File is all disposed, and obtains optimum segmentation threshold value that size in every class video sequence is the coding unit of 64 × 64, every During in class video sequence, size is the optimum segmentation threshold value of the coding unit of 32 × 32 and every class video sequence, size is big The optimum segmentation threshold value of the little coding unit being 16 × 16, wherein, in i=i+1 "=" it is assignment；

4. the coding unit that each size is 64 × 64 in pending HEVC HD video is carried out traversal coding, In traversal cataloged procedure, it is first determined the type of pending HEVC HD video；Then regard according to pending HEVC high definition The type of frequency, determines that in pending HEVC HD video, size is respectively the coding list of 64 × 64,32 × 32,16 × 16 The optimum segmentation threshold value of unit；It is respectively 64 × 64,32 × 32,16 then according to size in pending HEVC HD video The rate distortion costs value of the coding unit of × 16 and the optimum segmentation threshold value of correspondence, determine chi in pending HEVC HD video Very little size is respectively the coding unit of 64 × 64,32 × 32,16 × 16 the need of continuing segmentation.

A kind of HD video fast encoding method the most according to claim 1, it is characterised in that described step tool 2. Body process is:

2.-1, video sequence currently pending in training set is defined as current video sequence；

2.-2, the coding unit that all sizes are 64 × 64 in current video sequence is handled as follows:

2.-2a, the coding unit that currently pending size is 64 × 64 is defined as current coded unit；2.-2b, root According to the rate distortion costs value of current coded unit and the rate distortion costs value of four coding unit of next layer of current coded unit Sum, determines the cutting state of current coded unit, if the rate distortion costs value of current coded unit is more than or equal to current The rate distortion costs value sum of four coding unit of next layer of coding unit, it is determined that current coded unit continues segmentation, point Being slit into four sizes is the coding unit of 32 × 32, and determines that the cutting state of current coded unit, for splitting, is then held Row step 2.-2c；If the rate distortion costs value of current coded unit is single less than four, next layer coding of current coded unit The rate distortion costs value sum of unit, it is determined that current coded unit does not continues to segmentation, and determines the segmentation of current coded unit State, for not split, then performs step 2.-2c；2.-2c, by the pending coding list that size is 64 × 64 of the next one Unit, as current coded unit, is then back to step 2.-2b and continues executing with, until all sizes in current video sequence Be 64 × 64 coding unit be disposed, then perform step 2.-3；

2.-3, the coding unit that all sizes are 32 × 32 in current video sequence is handled as follows:

2.-3a, the coding unit that currently pending size is 32 × 32 is defined as current coded unit；2.-3b, root According to the rate distortion costs value of current coded unit and the rate distortion costs value of four coding unit of next layer of current coded unit Sum, determines the cutting state of current coded unit, if the rate distortion costs value of current coded unit is more than or equal to current The rate distortion costs value sum of four coding unit of next layer of coding unit, it is determined that current coded unit continues segmentation, point Being slit into four sizes is the coding unit of 16 × 16, and determines that the cutting state of current coded unit, for splitting, is then held Row step 2.-3c；If the rate distortion costs value of current coded unit is single less than four, next layer coding of current coded unit The rate distortion costs value sum of unit, it is determined that current coded unit does not continues to segmentation, and determines the segmentation of current coded unit State, for not split, then performs step 2.-3c；2.-3c, by the pending coding list that size is 32 × 32 of the next one Unit, as current coded unit, is then back to step 2.-3b and continues executing with, until all sizes in current video sequence Be 32 × 32 coding unit be disposed, then perform step 2.-4；

2.-4, the coding unit that all sizes are 16 × 16 in current video sequence is handled as follows:

2.-4a, the coding unit that currently pending size is 16 × 16 is defined as current coded unit；2.-4b, root According to the rate distortion costs value of current coded unit and the rate distortion costs value of four coding unit of next layer of current coded unit Sum, determines the cutting state of current coded unit, if the rate distortion costs value of current coded unit is more than or equal to current The rate distortion costs value sum of four coding unit of next layer of coding unit, it is determined that current coded unit continues segmentation, point Being slit into four sizes is the coding unit of 8 × 8, and determines that the cutting state of current coded unit, for splitting, then performs Step 2.-4c；If the rate distortion costs value of current coded unit is less than four coding unit of next layer of current coded unit Rate distortion costs value sum, it is determined that current coded unit does not continues to segmentation, and determines the segmentation shape of current coded unit State, for not split, then performs step 2.-4c；2.-4c, by the pending coding unit that size is 16 × 16 of the next one As current coded unit, it is then back to step 2.-4b and continues executing with, until all sizes in current video sequence are The coding unit of 16 × 16 is disposed, then performs step 2.-5；

2.-5, in statistics current video sequence, size is respectively the degree of depth of the coding unit of 64 × 64,32 × 32,16 × 16 Value, rate distortion costs value, the rate distortion costs value sum of four coding unit of next layer, cutting state；

2.-6, using pending video sequence next in training set as current video sequence, it is then back to step and 2.-2 continues Continuous execution, until all video sequences in training set are disposed, obtains size in each video sequence in training set big The statistical data of the little coding unit being respectively 64 × 64,32 × 32,16 × 16；

2.-7, by training set belongs to size in of a sort all video sequences be respectively 64 × 64,32 × 32,16 × The statistical data of the coding unit of 16 is stored in a file, total N' file, will store all video sequences of the i-th class The file of the statistical data that middle size is respectively the coding unit of 64 × 64,32 × 32,16 × 16 is designated as F_i, wherein, 1≤ i≤N'。

A kind of HD video fast encoding method the most according to claim 1, it is characterised in that described step 4. in really In the most pending fixed HEVC HD video size be respectively the coding unit of 64 × 64,32 × 32,16 × 16 the need of The process continuing segmentation is:

4.-1, the coding unit that all sizes are 64 × 64 in pending HEVC HD video is located as follows Reason:

4.-1a, the coding unit that currently pending size is 64 × 64 is defined as current coded unit；4.-1b, root According to rate distortion costs value and the optimum segmentation threshold value of coding unit that size is 64 × 64 of current coded unit, determine and work as Front coding unit is the need of continuing segmentation, if the rate distortion costs value of current coded unit is 64 × 64 more than size The optimum segmentation threshold value of coding unit, it is determined that current coded unit continues segmentation, be divided into four sizes be 32 × The coding unit of 32, then performs step 4.-1c；If the rate distortion costs value of current coded unit is big less than or equal to size The optimum segmentation threshold value of the little coding unit being 64 × 64, it is determined that current coded unit does not continues to segmentation, then performs step Rapid 4.-1c；4.-1c, using the pending coding unit that size is 64 × 64 of the next one as current coded unit, then Return step 4.-1b to continue executing with, until the coding that all sizes are 64 × 64 in pending HEVC HD video Cell processing is complete, then performs step 4.-2；

4.-2, the coding unit that all sizes are 32 × 32 in pending HEVC HD video is located as follows Reason:

4.-2a, the coding unit that currently pending size is 32 × 32 is defined as current coded unit；4.-2b, root According to rate distortion costs value and the optimum segmentation threshold value of coding unit that size is 32 × 32 of current coded unit, determine and work as Front coding unit is the need of continuing segmentation, if the rate distortion costs value of current coded unit is 32 × 32 more than size The optimum segmentation threshold value of coding unit, it is determined that current coded unit continues segmentation, be divided into four sizes be 16 × The coding unit of 16, then performs step 4.-2c；If the rate distortion costs value of current coded unit is big less than or equal to size The optimum segmentation threshold value of the little coding unit being 32 × 32, it is determined that current coded unit does not continues to segmentation, then performs step Rapid 4.-2c；4.-2c, using the pending coding unit that size is 32 × 32 of the next one as current coded unit, then Return step 4.-2b to continue executing with, until the coding that all sizes are 32 × 32 in pending HEVC HD video Cell processing is complete, then performs step 4.-3；

4.-3, the coding unit that all sizes are 16 × 16 in pending HEVC HD video is located as follows Reason:

4.-3a, the coding unit that currently pending size is 16 × 16 is defined as current coded unit；4.-3b, root According to rate distortion costs value and the optimum segmentation threshold value of coding unit that size is 16 × 16 of current coded unit, determine and work as Front coding unit is the need of continuing segmentation, if the rate distortion costs value of current coded unit is 16 × 16 more than size The optimum segmentation threshold value of coding unit, it is determined that current coded unit continues segmentation, and being divided into four sizes is 8 × 8 Coding unit, then perform step 4.-3c；If the rate distortion costs value of current coded unit is less than or equal to size It is the optimum segmentation threshold value of the coding unit of 16 × 16, it is determined that current coded unit does not continues to segmentation, then performs step ④-3c；4.-3c, using the pending coding unit that size is 16 × 16 of the next one as current coded unit, then return Return step 4.-3b to continue executing with, until the coding list that all sizes are 16 × 16 in pending HEVC HD video Unit is disposed.