CN106937116B - Low-complexity video coding method based on random training set adaptive learning - Google Patents
Low-complexity video coding method based on random training set adaptive learning Download PDFInfo
- Publication number
- CN106937116B CN106937116B CN201710153936.1A CN201710153936A CN106937116B CN 106937116 B CN106937116 B CN 106937116B CN 201710153936 A CN201710153936 A CN 201710153936A CN 106937116 B CN106937116 B CN 106937116B
- Authority
- CN
- China
- Prior art keywords
- frame
- rate distortion
- size
- value
- coding
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
Classifications
-
- H—ELECTRICITY
- H04—ELECTRIC COMMUNICATION TECHNIQUE
- H04N—PICTORIAL COMMUNICATION, e.g. TELEVISION
- H04N19/00—Methods or arrangements for coding, decoding, compressing or decompressing digital video signals
- H04N19/10—Methods or arrangements for coding, decoding, compressing or decompressing digital video signals using adaptive coding
- H04N19/134—Methods or arrangements for coding, decoding, compressing or decompressing digital video signals using adaptive coding characterised by the element, parameter or criterion affecting or controlling the adaptive coding
- H04N19/146—Data rate or code amount at the encoder output
- H04N19/147—Data rate or code amount at the encoder output according to rate distortion criteria
-
- H—ELECTRICITY
- H04—ELECTRIC COMMUNICATION TECHNIQUE
- H04N—PICTORIAL COMMUNICATION, e.g. TELEVISION
- H04N19/00—Methods or arrangements for coding, decoding, compressing or decompressing digital video signals
- H04N19/10—Methods or arrangements for coding, decoding, compressing or decompressing digital video signals using adaptive coding
- H04N19/102—Methods or arrangements for coding, decoding, compressing or decompressing digital video signals using adaptive coding characterised by the element, parameter or selection affected or controlled by the adaptive coding
- H04N19/12—Selection from among a plurality of transforms or standards, e.g. selection between discrete cosine transform [DCT] and sub-band transform or selection between H.263 and H.264
- H04N19/122—Selection of transform size, e.g. 8x8 or 2x4x8 DCT; Selection of sub-band transforms of varying structure or type
-
- H—ELECTRICITY
- H04—ELECTRIC COMMUNICATION TECHNIQUE
- H04N—PICTORIAL COMMUNICATION, e.g. TELEVISION
- H04N19/00—Methods or arrangements for coding, decoding, compressing or decompressing digital video signals
- H04N19/10—Methods or arrangements for coding, decoding, compressing or decompressing digital video signals using adaptive coding
- H04N19/169—Methods or arrangements for coding, decoding, compressing or decompressing digital video signals using adaptive coding characterised by the coding unit, i.e. the structural portion or semantic portion of the video signal being the object or the subject of the adaptive coding
- H04N19/17—Methods or arrangements for coding, decoding, compressing or decompressing digital video signals using adaptive coding characterised by the coding unit, i.e. the structural portion or semantic portion of the video signal being the object or the subject of the adaptive coding the unit being an image region, e.g. an object
- H04N19/176—Methods or arrangements for coding, decoding, compressing or decompressing digital video signals using adaptive coding characterised by the coding unit, i.e. the structural portion or semantic portion of the video signal being the object or the subject of the adaptive coding the unit being an image region, e.g. an object the region being a block, e.g. a macroblock
-
- H—ELECTRICITY
- H04—ELECTRIC COMMUNICATION TECHNIQUE
- H04N—PICTORIAL COMMUNICATION, e.g. TELEVISION
- H04N19/00—Methods or arrangements for coding, decoding, compressing or decompressing digital video signals
- H04N19/50—Methods or arrangements for coding, decoding, compressing or decompressing digital video signals using predictive coding
- H04N19/503—Methods or arrangements for coding, decoding, compressing or decompressing digital video signals using predictive coding involving temporal prediction
- H04N19/51—Motion estimation or motion compensation
- H04N19/567—Motion estimation based on rate distortion criteria
-
- H—ELECTRICITY
- H04—ELECTRIC COMMUNICATION TECHNIQUE
- H04N—PICTORIAL COMMUNICATION, e.g. TELEVISION
- H04N19/00—Methods or arrangements for coding, decoding, compressing or decompressing digital video signals
- H04N19/60—Methods or arrangements for coding, decoding, compressing or decompressing digital video signals using transform coding
- H04N19/61—Methods or arrangements for coding, decoding, compressing or decompressing digital video signals using transform coding in combination with predictive coding
Landscapes
- Engineering & Computer Science (AREA)
- Multimedia (AREA)
- Signal Processing (AREA)
- Physics & Mathematics (AREA)
- Discrete Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Compression Or Coding Systems Of Tv Signals (AREA)
Abstract
The invention discloses a kind of low-complexity video coding methods based on random training set adaptive learning.The present invention includes study coding method and fast encoding method.Video sequence is grouped by frame per second first, the first two frame of every group of video frame is study coded frame, is used for parameter learning, and subsequent frame is fast coding frame.According to the video coding parameter that study coded frame obtains, optimize the CU division methods and PU division methods of Video coding.When making CU division, the rate distortion computation work directly division for dividing or skipping current CU is terminated in advance for the CU of non-minimum size.When making intra prediction, PU corresponding to the CU of minimum dimension 8 × 8 determines its partition mode or skips the rate distortion computation of current PU directly to determine its partition mode in advance.Therefore, it is divided present invention substantially reduces the CU in HEVC Video coding and determines to divide the complexity determined with PU.Under the premise of keeping coding quality, the code efficiency of HEVC is effectively improved.
Description
Technical field
The invention belongs to efficient video coding (HEVC) technical fields, more particularly to one kind is based on random training collection adaptive
The low-complexity video coding method of study.
Background technique
In recent years, as the visual field of people is come into high definition, ultra high-definition video (resolution ratio reaches 4K × 2K, 8K × 4K) application,
Video compression technology receives huge challenge.In addition, hair of the miscellaneous Video Applications also with network and memory technology
Exhibition continues to bring out.To video compression performance, more stringent requirements are proposed with high definition trend for the diversification of Video Applications.For this purpose, 2010
Year April ITU-T Video Coding Experts Group (VCEG) and the Motion Picture Experts Group (MPEG) of ISO/IEC set up Video coding
Joint group (JCT-VC) jointly formulates video encoding standard of new generation, efficient video coding HEVC (High was completed in 2013
Efficiency Video Coding) standard, also referred to as H.265.H.264/AVC compared to video encoding standard before,
Video encoding standard before the code efficiency of HEVC standard is compared is obviously improved, but the complexity encoded simultaneously is also substantially
Degree improves, and model selection complexity caused by especially flexible block divides is multiplied.HEVC has been image Definition of Division more
Flexible block partition mode, including coding unit (CU), predicting unit (PU), converter unit (TU).CU size in HEVC has
64 × 64,32 × 32,16 × 16,8 × 8, are denoted as 2N × 2N, wherein N=32, and 16,8,4.Wherein 64 × 64 CU is defined as
Maximum coding unit (LCU), using quad-tree partition structure, each LCU recursive can be divided into 4 equal-sized CU,
Until the smallest CU (8 × 8).In order to find the CU splitting scheme of optimization, encoder has to consider all dividing conditions.
Fig. 1 illustrates how a LCU is divided into various sizes of CU mode.Each related PU of CU, wherein PU is in frame
The basic unit of prediction and inter-prediction, all information relevant to prediction are all defined in PU.Fig. 2 gives intra prediction
PU partition mode.For the CU of a 2N × 2N, there are two types of corresponding intraprediction unit PU partition modes: 2N × 2N and N
× N, it is that the smallest 8 × 8 size just will use that N × N mode therein, which only works as CU, and the CU of other sizes only uses 2N × 2N's
PU mode.The HEVC Video coding of standard traverses various CU and PU partition modes, using rate-distortion optimization (RDO) technology from numerous
The smallest mode of rate distortion costs is chosen in mode as optimal mode.Because HEVC encoder needs to be traversed for all possible CU
And PU, the complexity of this ergodic process is higher, and calculation amount is too big, brings difficulty to Video Coding.
Summary of the invention
The purpose of the present invention is propose a kind of based on random training for the complicated high disadvantage of existing HEVC Video coding
The low-complexity video coding method of collection adaptive study is distorted by the determination method and its rate that simplify CU and PU partition mode
The calculation amount of cost reduces the complexity of coding while guaranteeing coding quality, is particularly suitable for quick Video Coding
Occasion.Existing fast encoding method more depends on the extraction and definition for making parameter to the feature of each video, no
Same video is generally required by repeatedly attempting and comparing the value to determine parameter.In contrast, the method for the present invention being capable of basis
Different videos adaptively learns video coding parameter, and these features are applied in subsequent fast coding frame, significantly
Improve the universality of method for video coding.
The present invention includes study coding method and fast encoding method, it is characterised in that presses video sequence when intraframe coding
Frame per second is grouped, the first two frame of every group of video is study coded frame, is used for parameter learning, and subsequent frame is fast coding frame, root
According to the parameter that study coded frame obtains, CU (coding unit) determination method and PU (predicting unit) optimized in intraframe coding determines
Method.The minimum dimension of CU is 8 × 8, and for the CU of non-minimum size, judgement terminates in advance CU and divides or skip current CU's
Rate distortion computation is made directly to divide.And when making intra prediction, for the CU of minimum dimension, corresponding PU partition mode has 2N
× 2N and N × N；And the CU of other sizes only has 2N × 2N one mode, so making PU to it to the CU of minimum dimension and dividing mould
The optimization of formula.
The coding method of study coded frame is:
Step (1), study coded frame first frame are for calculating threshold tauN, random to select for the CU of each 2N × 2N size
Take nNA pixel is gathered as training, and the gray scale difference mean value τ of each pixel pair of different size CU is calculatedN；
Step (2), study the second frame of coded frame are for calculating threshold value Tr1N、Tr2NAnd RD1_trN、RD2_trN, for every
The CU of a 2N × 2N size, randomly selects nNA pixel records do not divide and divide pixel in CU block respectively to composition training set
Threshold tau is greater than to gray scale differenceNNumber qk,N, take all qk,NThe threshold value Tr1 that is not divided as each size CU of average valueNWith draw
Divide threshold value Tr2N；Meanwhile recording each size CU respectively and not dividing mean value with the rate distortion costs under dividing condition, obtain each ruler
The threshold value RD1_tr of very little CU rate distortion costsNAnd RD2_trN。
The threshold value that PU partition mode corresponding for minimum dimension CU, all threshold value calculation methods and above-mentioned CU are divided
Calculation method is identical.
The coding method of fast coding frame:
Step (I) randomly selects n from CUNA pixel pair seeks the gray scale difference value of each pixel pair, by itself and step (1)
Obtained threshold tauNIt compares, calculates and be greater than τNPixel pair number qk,N；By qk,NThe threshold value Tr1 obtained with step (2)NWith
Tr2NIt is compared, CU is divided into three classes: I class, II class and Group III；
Step (II) optimizes the CU of non-minimum size to CU partition mode work according to CU classification results；To minimum ruler
Very little CU optimizes its corresponding PU partition mode work.Wherein:
If CU is I class, if rate distortion costs are less than threshold value RD1_trN, for the CU of non-minimum size, terminate in advance CU
It divides；Otherwise, current CU is divided into 4 sub- CU and calculates the rate distortion costs of every sub- CU, then determines whether current CU divides.
If rate distortion costs are less than threshold value RD1_trN, determine that the corresponding PU partition mode of CU of minimum dimension is 2N × 2N in advance, skip
The PU partition mode of subsequent N × N size calculates；Otherwise, it calculates PU mode and is the rate distortion costs of N × N, then determine PU mould
Formula.
If CU is II class, for non-minimum size CU, the rate for it being divided into every sub- CU after four sub- CU is first calculated
Distortion cost, when the sum of rate distortion costs of this four sub- CU are less than threshold value RD2_trNWhen, then skip the rate distortion generation of current CU
Valence calculating directly determines that current CU is divided into four sub- CU；Otherwise, the rate distortion costs of current CU are calculated, then determine that current CU is
No division.For minimum dimension CU, when its corresponding PU is divided into the sum of rate distortion costs of four sub- PU less than threshold value RD2_
trNWhen, then the PU rate distortion computation of 2N × 2N is skipped, directly judgement PU partition mode is N × N；Otherwise, PU partition mode is calculated
For the rate distortion costs of 2N × 2N, then determine PU mode.
If CU is Group III, the HEVC Video coding of standard is executed.
The present invention has the beneficial effect that:
Basic principle of the invention is to carry out normal encoding to the first two frame of every group of video frame, extracts HEVC Video coding ginseng
Number.Then since third frame Video coding, the threshold information divided with PU is divided using the CU that the first two frame obtains, CU is taken
It terminates in advance and divides or skip the strategy that the rate distortion costs calculating work of current CU directly divides, and to CU pairs of minimum dimension
The PU mode answered determines in advance or skips that the rate distortion costs of current PU calculate the strategy for making Direct Model judgement, so that view
The computational complexity of frequency coding reduces, and the time of Video coding is greatly saved.By measuring, Video coding matter is being kept
Under the premise of amount, it can be saved on video encoding time using low-complexity method of the invention than standard HEVC coding method
40% or more, substantially increase the efficiency of Video coding.Further it is proposed that method be a kind of adaptive coding staff
Method can be directed to different videos, extract its coding characteristic by study coded frame and be applied in subsequent fast coding frame,
This makes coding method proposed by the present invention have good flexibility and applicability.
Detailed description of the invention
Fig. 1 is the CU partition mode schematic diagram of HEVC；
Fig. 2 is the PU partition mode schematic diagram of HEVC intra prediction；
Fig. 3 is that the CU of 2N × 2N is divided into four schematic diagrames for waiting blocks；
Fig. 4 is present invention coding flow chart；
Fig. 5 is the coding flow chart for learning coded frame；
Fig. 6 is the coding flow chart of fast coding frame.
Specific embodiment
Present invention will be further explained below with reference to the attached drawings and examples.
As shown in figures 1 to 6, the low-complexity video coding method based on random training set adaptive learning, is regarded using HEVC
The HM model of frequency coding, the HEVC Video coding universal test condition (JCTVC-H1100) that test condition uses JCT-VC to formulate,
Use the full frame interior coding configuration file encoder_intra_main.cfg for the efficient video coding that HM model carries.
Fig. 4 show the total coding flow chart of the present invention, and before each video sequence coding, video sequence is first pressed frame per second
It is grouped, the first two frame of every group of video is study coded frame, and the HEVC coding method of operation standard traverses all 2N × 2N (N
=32,16,8) the corresponding PU partition mode of minimum dimension CU of CU partition mode and 2N × 2N (N=4) obtains Video coding
Parameter is simultaneously applied to subsequent fast coding frame to realize low encoding complexity.As shown in figure 5, the coding of study coded frame
Step is:
Step (1), study coded frame first frame are for training threshold tauN, random to select for the CU of each 2N × 2N size
Take nNA pixel pair calculates the pixel under different CU sizes to gray scale difference mean value τN, the wherein selection method of pixel pair are as follows:
N is randomly selected from CUNTo pixel pair, wherein each pair of pixel is to being located at four different divided block regions
It is interior, and two-dimentional Poisson distribution is obeyed in position of the pixel in sub-block region.To one having a size of 2N × 2N CU, such as Fig. 3
Shown, the region CU can be divided into four equal-sized sub-block region A, B, C, D in advance, and each pixel is different to being located at
Sub-block region, therefore one shares 6 kinds of combinations, (A, B), (A, C), (A, D), (B, C), (B, D), (C, D), each interior choosing of region combination
The pixel logarithm taken is identical.Each pixel obeys the two-dimentional Poisson spread about central point in corresponding sub-block region
Distribution, by taking a-quadrant as an example, a-quadrant centre coordinate isThe selection of the coordinate (X, Y) of pixel is obeyed:
nNValue can be adjusted according to CU size, for the CU of a size 2N × 2N, nNValue are as follows:
nN=3 × N × m (3)
Above formula m is positive integer, reasonably adjusts pixel to number by m value.It is determined by experiment, m value is 3；
Step (2), study the second frame of coded frame are for calculating threshold value Tr1N、Tr2NWith threshold value RD1_trN、RD2_trN.It is right
In the CU of each 2N × 2N size, n is randomly selectedNA pixel pair, is trained them, and study obtains threshold value Tr1N、Tr2NWith
Threshold value RD1_trN、RD2_trN.Wherein threshold value Tr1N、Tr2NTraining step are as follows:
(2a) is according to the n randomly selected out of CUNTo pixel to the threshold tau obtained with study coded frame first frameN, compare
The grey value difference of each pixel pair, the two gray scale difference are greater than τNWhen, δk,NValue is 1, and otherwise value is 0, calculating formula are as follows:
Above formula g (i, j) and g (u, v) respectively indicates a CU pixel to the pixel for being located at coordinate position (i, j), (u, v)
Gray value, τNFor adaptive gray difference threshold, learn to obtain by first frame.
(2b) calculates random training set nNMiddle δk,NThe pixel that value is 1 is to number:
(2c) completes the compressed encoding of video frame according to the HEVC method for video coding of standard, is divided according to each size CU
Situation calculates q when not dividing and divide under study each CU size of the second frame of coded framek,NMean value, respectively as threshold value Tr1N、
Tr2N。
Threshold value RD1_trN、RD2_trNCalculation method are as follows:
When coding to study the second frame of coded frame, the case where not dividing and divide according to each size CU, record
The rate distortion costs with size CU each under dividing condition are not divided and seek its mean value, are denoted as mean_RD_nonsplit respectivelyNWith
mean_RD_splitN.In addition, counting the rate distortion costs that each size does not divide with divides CU respectively, maximum value is remembered respectively
For max_RD_nonsplitNAnd max_RD_splitN, minimum value is denoted as min_RD_nonsplit respectivelyNAnd min_RD_
splitN.The then CU rate distortion costs threshold value RD1_tr having a size of 2N × 2NN、RD2_trNAre as follows:
Wherein, α and β is adjusting parameter, and value range is α ∈ [0,1], and β ∈ [0,1] is determined by experiment here, and α takes
It is 0.05 that value, which is 0.2, β value,.
The threshold value that PU partition mode corresponding for minimum dimension CU, all threshold value calculation methods and above-mentioned CU are divided
Calculation method is identical.
The subsequent frame for learning coded frame is fast coding frame.As shown in fig. 6, the third frame from every group of video is initially quick
Coded frame, coding step are:
Step (I) randomly selects n from CUNA pixel pair obtains the characteristic parameter of current CU to study by pixel
qk,N, calculation method with study coded frame coding step (2) in (2a) and (2b) it is identical.By qk,NWith threshold value Tr1NAnd Tr2N
Be compared, CU is divided into three classes: I class, II class and Group III, specific classifying rules is:
Step (II) optimizes the CU of non-minimum size to CU partition mode work according to CU classification results；To minimum ruler
Very little CU optimizes its corresponding PU partition mode work；Wherein:
If CU is I class, if rate distortion costs are less than threshold value RD1_trN, for the CU of non-minimum size, terminate in advance CU
It divides；Otherwise, current CU is divided into 4 sub- CU and calculates the rate distortion costs of every sub- CU, then determines whether current CU divides.
If rate distortion costs are less than threshold value RD1_trN, determine that the corresponding PU partition mode of CU of minimum dimension is 2N × 2N in advance, skip
The PU partition mode of subsequent N × N size calculates；Otherwise, it calculates PU mode and is the rate distortion costs of N × N, then determine PU mould
Formula.
If CU is II class, for non-minimum size CU, the rate for it being divided into every sub- CU after four sub- CU is first calculated
Distortion cost, when the sum of rate distortion costs of this four sub- CU are less than threshold value RD2_trNWhen, then skip the rate distortion generation of current CU
Valence calculating directly determines that current CU is divided into four sub- CU；Otherwise, the rate distortion costs of current CU are calculated, then determine that current CU is
No division.For minimum dimension CU, when its corresponding PU is divided into the sum of rate distortion costs of four sub- PU less than threshold value RD2_
trNWhen, then the PU rate distortion computation of 2N × 2N is skipped, directly judgement PU partition mode is N × N；Otherwise, PU partition mode is calculated
For the rate distortion costs of 2N × 2N, then determine PU mode.
If CU is Group III, the HEVC Video coding of standard is executed.
Claims (4)
1. the low-complexity video coding method based on random training set adaptive learning, it is characterised in that including learning coding staff
Video sequence is grouped by method and fast encoding method, when intraframe coding by frame per second, the first two frame of every group of video is that study is compiled
Code frame is used for parameter learning, and subsequent frame is fast coding frame, according to the parameter that study coded frame obtains, optimizes in intraframe coding
CU partition mode determination method and PU partition mode determination method；The minimum dimension of CU is 8 × 8, for non-minimum size
CU judges whether to terminate in advance the rate distortion computation work directly division that CU divided or skipped current CU；It is right when making intra prediction
In the CU of minimum dimension, corresponding PU partition mode has 2N × 2N and N × N；And the CU of other sizes only has 2N × 2N a kind of
PU partition mode makees the optimization of PU partition mode judgement to it for the CU of minimum dimension；
The coding method of study coded frame is:
Step (1), study coded frame first frame are for calculating threshold tauN, for the CU of each 2N × 2N size, randomly select nNIt is a
Pixel is gathered as training, and the gray scale difference mean value τ of each pixel pair of different size CU is calculatedN；
Step (2), study the second frame of coded frame are for calculating threshold value Tr1N、Tr2NAnd RD1_trN、RD2_trN；For each 2N ×
The CU of 2N size, randomly selects nNA pixel records do not divide and divide in CU block pixel to gray scale respectively to composition training set
Difference is greater than threshold tauNNumber qk,N, take qk,NThe threshold value Tr1 that is not divided as each size CU of average valueNWith division threshold value Tr2N；
Meanwhile recording each size CU respectively and not dividing mean value with the rate distortion costs under dividing condition, obtain each size CU rate distortion
The threshold value RD1_tr of costNAnd RD2_trN；
The threshold calculations that PU partition mode corresponding for minimum dimension CU, all threshold value calculation methods and above-mentioned CU are divided
Method is identical；
The coding method of fast coding frame is:
Step (I) randomly selects n from CUNA pixel pair seeks the gray scale difference value of each pixel pair, it is obtained with step (1)
Threshold tauNIt compares, calculates and be greater than τNPixel pair number qN；By qNThe threshold value Tr1 obtained with step (2)NAnd Tr2NIt carries out
Compare, CU is divided into three classes: I class, II class and Group III；
Step (II) optimizes the CU of non-minimum size to CU partition mode work according to CU classification results；To minimum dimension
CU optimizes its corresponding PU partition mode work；Wherein:
If CU is I class, if rate distortion costs are less than threshold value RD1_trN, for the CU of non-minimum size, terminate in advance CU division；
Otherwise, current CU is divided into 4 sub- CU and calculates the rate distortion costs of every sub- CU, then determines whether current CU divides；If rate
Distortion cost is less than threshold value RD1_trN, determine that the corresponding PU partition mode of CU of minimum dimension is 2N × 2N in advance, skip subsequent
N × N size PU partition mode calculate；Otherwise, it calculates PU mode and is the rate distortion costs of N × N, then determine PU mode；
If CU is II class, for non-minimum size CU, the rate distortion for it being divided into every sub- CU after four sub- CU is first calculated
Cost, when the sum of rate distortion costs of this four sub- CU are less than threshold value RD2_trNWhen, then skip the rate distortion costs meter of current CU
Calculation directly determines that current CU is divided into four sub- CU；Otherwise, the rate distortion costs of current CU are calculated, then determine whether current CU draws
Point；For minimum dimension CU, when its corresponding PU is divided into the sum of rate distortion costs of four sub- PU less than threshold value RD2_trN
When, then the PU rate distortion computation of 2N × 2N is skipped, directly judgement PU partition mode is N × N；Otherwise, calculating PU partition mode is
The rate distortion costs of 2N × 2N, then determine PU mode；
If CU is Group III, the HEVC Video coding of standard is executed；
The CU classifying rules of the coding method step (I) of fast coding frame is:
2. the low-complexity video coding method as described in claim 1 based on random training set adaptive learning, feature
It is, the pixel in step (1) and step (I) is to randomly selecting method are as follows:
N is randomly selected from CUNA pixel is to forming random training set, wherein each pixel is to being located at the four of identical size
In a sub- CU, and two-dimentional Poisson distribution is obeyed in position of the pixel in sub- CU；For a CU having a size of 2N × 2N, nN's
Value are as follows:
nN=3 × N × m (1)
Above formula m is positive integer, adjusts the pixel in random training set to number by setting m value.
3. the low-complexity video coding method as described in claim 1 based on random training set adaptive learning, feature
It is to learn coding method step (2) threshold value Tr1 of coded frameNAnd Tr2NCalculation method be:
A. according to the n randomly selected out of CUNA pixel pair, the gray value differences of more each pixel pair, when the two gray scale difference is greater than τN
When, δk,NValue is 1, and otherwise value is 0:
Above formula g (i, j) and g (u, v) respectively indicates the pixel gray value of coordinate position (i, j), (u, v) in CU, τNIt is learned to pass through
Practise the threshold value that coded frame first frame learns；
B. random training set n is calculatedNMiddle δk,NThe pixel that value is 1 is to number:
C. each size CU is counted respectively not divide and the q under dividing conditionk,NAverage value, obtain threshold value Tr1NAnd Tr2N。
4. the low-complexity video coding method as described in claim 1 based on random training set adaptive learning, feature
It is, learns the rate distortion costs threshold value RD1_tr of the coding method step (2) of coded frameNAnd RD2_trNCalculation method be:
It according to each size CU dividing condition, records the rate distortion costs that each size CU is not divided and divided and seeks its mean value, respectively
It is denoted as mean_RD_nonsplitNAnd mean_RD_splitN, in addition, counting the rate that each size CU is not divided with divided respectively
Distortion cost, maximum value are denoted as max_RD_nonsplit respectivelyNAnd max_RD_splitN, minimum value is denoted as min_ respectively
RD_nonsplitNAnd min_RD_splitN；The then CU rate distortion costs threshold value RD1_tr having a size of 2N × 2NNAnd RD2_trNMeter
It calculates are as follows:
Wherein, α and β is adjusting parameter, and value range is α ∈ [0,1], β ∈ [0,1].
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201710153936.1A CN106937116B (en) | 2017-03-15 | 2017-03-15 | Low-complexity video coding method based on random training set adaptive learning |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201710153936.1A CN106937116B (en) | 2017-03-15 | 2017-03-15 | Low-complexity video coding method based on random training set adaptive learning |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN106937116A CN106937116A (en) | 2017-07-07 |
| CN106937116B true CN106937116B (en) | 2019-08-27 |
Family
ID=59433739
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201710153936.1A Active CN106937116B (en) | 2017-03-15 | 2017-03-15 | Low-complexity video coding method based on random training set adaptive learning |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN106937116B (en) |
Families Citing this family (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109510987B (en) * | 2017-09-15 | 2022-12-06 | 华为技术有限公司 | Method and device for determining coding tree node division mode and coding equipment |
| CN109788287B (en) * | 2017-11-10 | 2022-10-25 | 腾讯科技（深圳）有限公司 | Video encoding method, video encoding device, computer equipment and storage medium |
| CN108174208B (en) * | 2018-02-12 | 2020-05-12 | 杭州电子科技大学 | Efficient video coding method based on feature classification |
| CN108777794B (en) * | 2018-06-25 | 2022-02-08 | 腾讯科技（深圳）有限公司 | Image encoding method and apparatus, storage medium, and electronic apparatus |
| CN110139097B (en) * | 2019-04-19 | 2023-01-06 | 中南大学 | Method for self-adaptive mode sequence regulation in video coding |
| CN111355956B (en) * | 2020-03-09 | 2023-05-09 | 蔡晓刚 | Deep learning-based rate distortion optimization rapid decision system and method in HEVC intra-frame coding |
| CN112291562B (en) * | 2020-10-29 | 2022-06-14 | 郑州轻工业大学 | Fast CU partition and intra mode decision method for H.266/VVC |
| CN112437310B (en) * | 2020-12-18 | 2022-07-08 | 重庆邮电大学 | VVC intra-frame coding rapid CU partition decision method based on random forest |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN104125469A (en) * | 2014-07-10 | 2014-10-29 | 中山大学 | Fast coding method for high efficiency video coding (HEVC) |
| CN104320658A (en) * | 2014-10-20 | 2015-01-28 | 南京邮电大学 | HEVC (High Efficiency Video Coding) fast encoding method |
| CN105721865A (en) * | 2016-02-01 | 2016-06-29 | 同济大学 | Fast decision algorithm for dividing HEVC inter-frame coding unit |
| CN105791826A (en) * | 2016-05-11 | 2016-07-20 | 南京大学 | Data mining-based HEVC inter-frame fast mode selection method |
-
2017
- 2017-03-15 CN CN201710153936.1A patent/CN106937116B/en active Active
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN104125469A (en) * | 2014-07-10 | 2014-10-29 | 中山大学 | Fast coding method for high efficiency video coding (HEVC) |
| CN104320658A (en) * | 2014-10-20 | 2015-01-28 | 南京邮电大学 | HEVC (High Efficiency Video Coding) fast encoding method |
| CN105721865A (en) * | 2016-02-01 | 2016-06-29 | 同济大学 | Fast decision algorithm for dividing HEVC inter-frame coding unit |
| CN105791826A (en) * | 2016-05-11 | 2016-07-20 | 南京大学 | Data mining-based HEVC inter-frame fast mode selection method |
Non-Patent Citations (3)
| Title |
|---|
| 《Fast CU Size Decision and Mode Decision Algorithm for HEVC Intra Coding》;Linquan Shen et.al.;《IEEE Transactions on Consumer Electronics》;20130228;第207-213页 * |
| 《Fast intra mode decision algorithm for HEVC based on dominant edge assent distribution》;Yao yingbiao, Li Xiaojuan, Lu Yu;《MULTIMEDIA TOOLS AND APPLICATIONS》;20160228;第1963-1981页 * |
| 3D_HEVC帧间编码的低复杂度算法;马然等;《电视技术》;20151231;第39卷(第13期);第7-13页 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN106937116A (en) | 2017-07-07 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN106937116B (en) | Low-complexity video coding method based on random training set adaptive learning | |
| CN110062234B (en) | Perceptual video coding method based on just noticeable distortion of region | |
| CN107623850B (en) | A kind of quick screen contents encoding method based on temporal correlation | |
| CN103546749B (en) | Method for optimizing HEVC (high efficiency video coding) residual coding by using residual coefficient distribution features and bayes theorem | |
| CN103888762B (en) | Video coding framework based on HEVC standard | |
| CN104243997B (en) | Method for quality scalable HEVC (high efficiency video coding) | |
| CN105141954A (en) | HEVC interframe coding quick mode selection method | |
| CN101710993A (en) | Block-based self-adaptive super-resolution video processing method and system | |
| CN105578181B (en) | The fast intra mode decision and block matching method of screen content compression in a kind of HEVC | |
| Kong et al. | Object-detection-based video compression for wireless surveillance systems | |
| CN108924554A (en) | A kind of panorama video code Rate-distortion optimization method of spherical shape weighting structures similarity | |
| CN104853191B (en) | A kind of HEVC fast encoding method | |
| CN104601992B (en) | SKIP mode quick selecting methods based on Bayesian Smallest Risk decision | |
| CN104333754B (en) | Based on the SHVC enhancement-layer video coding methods that predictive mode is quickly selected | |
| CN103596004A (en) | Intra-frame prediction method and device based on mathematical statistics and classification training in HEVC | |
| CN105430415A (en) | Fast intraframe coding method of 3D-HEVC depth videos | |
| CN111314698A (en) | Image coding processing method and device | |
| CN104378636A (en) | Video image coding method and device | |
| CN110365982A (en) | The different transform of intraframe coding selects accelerated method in a kind of multipurpose coding | |
| CN104038769B (en) | Rate control method for intra-frame coding | |
| CN108769696A (en) | A kind of DVC-HEVC video transcoding methods based on Fisher discriminates | |
| CN104333755B (en) | The CU based on SKIP/Merge RD Cost of B frames shifts to an earlier date terminating method in HEVC | |
| CN107690069B (en) | Data-driven cascade video coding method | |
| CN101202915A (en) | Method and apparatus for selecting frame inner forecasting mode | |
| CN108881905B (en) | Probability-based intra-frame encoder optimization method |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant | ||
| TR01 | Transfer of patent right | ||
| TR01 | Transfer of patent right |
Effective date of registration: 20201223 Address after: 314500 Tongxiang, Jiaxing, Zhejiang, Wutong Street East Road (East) 55, Tongxiang chamber of Commerce Building 1 unit 1702, 1703 room -A-218 Patentee after: Jiaxing Baoqiao Machinery Technology Co.,Ltd. Address before: Room 3003-1, building 1, Gaode land center, Jianggan District, Hangzhou City, Zhejiang Province Patentee before: Zhejiang Zhiduo Network Technology Co.,Ltd. Effective date of registration: 20201223 Address after: Room 3003-1, building 1, Gaode land center, Jianggan District, Hangzhou City, Zhejiang Province Patentee after: Zhejiang Zhiduo Network Technology Co.,Ltd. Address before: 310027 No.2 street, Xiasha Higher Education Park, Hangzhou, Zhejiang Province Patentee before: HANGZHOU DIANZI University |
|
| TR01 | Transfer of patent right | ||
| TR01 | Transfer of patent right |
Effective date of registration: 20210119 Address after: 053100 room 113, development and Reform Bureau, 125 Jianshe North Road, Zaoqiang County, Hengshui City, Hebei Province Patentee after: Hebei Xionghua Technology Co.,Ltd. Address before: 314500 Tongxiang, Jiaxing, Zhejiang, Wutong Street East Road (East) 55, Tongxiang chamber of Commerce Building 1 unit 1702, 1703 room -A-218 Patentee before: Jiaxing Baoqiao Machinery Technology Co.,Ltd. |