CN106713935B - A kind of HEVC block division fast method based on Bayesian decision - Google Patents
A kind of HEVC block division fast method based on Bayesian decision Download PDFInfo
- Publication number
- CN106713935B CN106713935B CN201710014097.5A CN201710014097A CN106713935B CN 106713935 B CN106713935 B CN 106713935B CN 201710014097 A CN201710014097 A CN 201710014097A CN 106713935 B CN106713935 B CN 106713935B
- Authority
- CN
- China
- Prior art keywords
- decision
- current
- value
- risk
- divided
- 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
- 238000000034 method Methods 0.000 title claims abstract description 31
- 230000008859 change Effects 0.000 claims abstract description 22
- 238000001514 detection method Methods 0.000 claims abstract description 9
- 239000000284 extract Substances 0.000 claims abstract description 4
- 239000013598 vector Substances 0.000 claims description 27
- 230000006870 function Effects 0.000 claims description 23
- 230000008569 process Effects 0.000 claims description 14
- 238000012549 training Methods 0.000 claims description 6
- 239000011159 matrix material Substances 0.000 claims description 4
- 238000005457 optimization Methods 0.000 claims description 4
- 238000000605 extraction Methods 0.000 claims description 3
- 230000005484 gravity Effects 0.000 claims description 3
- 239000000203 mixture Substances 0.000 claims description 3
- 230000007423 decrease Effects 0.000 claims description 2
- 241000208340 Araliaceae Species 0.000 claims 1
- 235000005035 Panax pseudoginseng ssp. pseudoginseng Nutrition 0.000 claims 1
- 235000003140 Panax quinquefolius Nutrition 0.000 claims 1
- 235000008434 ginseng Nutrition 0.000 claims 1
- 238000005192 partition Methods 0.000 description 19
- 230000009466 transformation Effects 0.000 description 7
- 230000006835 compression Effects 0.000 description 4
- 238000007906 compression Methods 0.000 description 4
- 238000013139 quantization Methods 0.000 description 4
- 230000002123 temporal effect Effects 0.000 description 4
- 230000003044 adaptive effect Effects 0.000 description 3
- 230000008901 benefit Effects 0.000 description 3
- 230000007246 mechanism Effects 0.000 description 3
- 238000010586 diagram Methods 0.000 description 2
- 239000004744 fabric Substances 0.000 description 2
- 238000001914 filtration Methods 0.000 description 2
- 238000012545 processing Methods 0.000 description 2
- 238000012360 testing method Methods 0.000 description 2
- 241001269238 Data Species 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 230000005540 biological transmission Effects 0.000 description 1
- 238000009795 derivation Methods 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 230000000694 effects Effects 0.000 description 1
- 238000011478 gradient descent method Methods 0.000 description 1
- 238000007689 inspection Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 238000005070 sampling Methods 0.000 description 1
- 235000015170 shellfish Nutrition 0.000 description 1
- 238000000638 solvent extraction Methods 0.000 description 1
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
- H04N19/60—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
-
- 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
- H04N19/119—Adaptive subdivision aspects, e.g. subdivision of a picture into rectangular or non-rectangular coding blocks
-
- 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
- H04N19/134—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
- H04N19/149—Data rate or code amount at the encoder output by estimating the code amount by means of a model, e.g. mathematical model or statistical model
-
- 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
Landscapes
- Engineering & Computer Science (AREA)
- Multimedia (AREA)
- Signal Processing (AREA)
- Physics & Mathematics (AREA)
- Algebra (AREA)
- General Physics & Mathematics (AREA)
- Mathematical Analysis (AREA)
- Mathematical Optimization (AREA)
- Pure & Applied Mathematics (AREA)
- Compression Or Coding Systems Of Tv Signals (AREA)
Abstract
The invention discloses a kind of, and the HEVC block based on Bayesian decision divides fast method.The invention firstly uses the scene change detections based on average gray difference value to be divided into on-line study stage and fast stage by stage for video sequence.Secondly for the video frame that on-line study stage and occurrence scene convert, depth is divided at each, extracts the J of CU respectivelyinterAnd JintraIt is characterized value, mixed Gauss model is established with this, the design parameter of model is determined according to the EM algorithm of K-Means algorithm initialization；For being in the CU to be divided of fast stage by stage, extract its characteristic value, the conditional probability whether divided is found out according to mixed Gauss model, finally finds out the lesser decision of risk, the judgment basis whether divided as current CU using the Bayesian formula of minimum risk.Present invention reduces algorithm complexities, and the scramble time can be greatly reduced.
Description
Technical field
The invention belongs to high-definition digital video compression coding and decoding technical fields more particularly to a kind of based on Bayesian decision
HEVC (High Efficiency Video Coding, efficient video coding) block divides fast method.
Background technique
In the fast development of digital video application industrial chain, in face of Video Applications constantly to fine definition, high frame per second, height
The trend that compression ratio direction is developed, the limitation of previous generation video compression standard agreement H.264/AVC constantly highlight.Therefore, by
The tissue JCT-VC that International Telecommunication Union (ITU-T) and Motion Picture Experts Group (MPEG) joint are set up proposes next-generation video
Encoding and decoding standard --- H.265/HEVC.Its target is under the premise of identical picture quality, and compression ratio ratio is H.264/AVC high-grade
It is secondary to double.
HEVC is not broken through in terms of coding principle and basic structure, and H.264/AVC almost the same, i.e. prediction plus transformation
Block encoding mode；On Ciphering details and H.264/AVC also very close to comprising intra prediction, inter-prediction, estimation
With the coding/decoding modules such as compensation, orthogonal transformation, quantization, loop filtering, entropy coding and reconstruction.But with H.264/AVC compare
Compared with HEVC almost takes important corrective measure in each coding link, as intra prediction supports 35 kinds of modes, frames
Between prediction introduce Merge mode, change quantization support be up to 32 × 32 converter units, entropy coding use CABAC (Context
Adaptive Binary Arithmetic Coder, the adaptive binary arithmetic coding based on context) algorithm, new introducing
Adaptive sampling point compensation loop filtering etc..
HEVC encoder uses coding tree unit (CTU, Coding Tree Unit), coding unit (CU, Coding
Unit), predicting unit (PU, Prediction Unit), converter unit (Transform Unit, TU) structure, enable HEVC
It is enough that the video of different resolution and application environment is encoded.When HEVC is encoded, piece image can be divided into multiple mutual
Nonoverlapping CTU, each CTU are divided into one or more CU；When carrying out in frame or inter prediction encoding, a CU can
One or more PU is divided into selection；In transformation, quantization operation, a CU can be divided into one or more TU.
CU is the basic unit of HEVC coding, using recursive division mode.HEVC support 8 × 8,16 × 16,32 × 32,
The CU of 64 × 64 this 4 kinds of sizes, corresponding division depth is respectively 3,2,1,0.
PU is the basic unit of HEVC prediction, it is specified that the prediction mode of CU, all coded datas relevant to prediction are all
By PU transmission.For a CU having a size of 2N × 2N, the optional mode of PU has 2 kinds: 2N × 2N and N × N in frame；Interframe
The optional mode of PU has 8 kinds, including 4 kinds of symmetric patterns: 2N × 2N, 2N × N, N × 2N, N × N and 4 kinds of asymmetric moulds
Formula: 2N × nD, 2N × nU, nR × 2N, nL × 2N.When encoding use Merge when, residual information be not required to it is to be encoded,
For the PU mode used at this time for 2N × 2N skip mode, PU partition mode is as shown in Figure 1.
TU is the basic unit of HEVC transform and quantization, and size is also flexibly to change, in the form of quaternary tree into
Row recursive subdivision.When CTU is having a size of 64 × 64, in each CU TU quaternary tree recurrence divide as shown in Fig. 2, wherein CU draw
Point indicated by the solid line, the division of TU is represented by dashed line, and coded sequence is indicated with alpha code.
Similar with other international encoding standards, HM encoder is also based on rate-distortion optimization for the decision of block size
(Rate-Distortion Optimization's, RDO), in HEVC standard encryption algorithm, the division process description of CU is such as
Under:
(1) depth of current CU is arranged is zero；
(2) rate distortion costs (the Rate-Distortion Cost, RD of the Skip mode in the case where currently dividing depth are calculated
Cost)Jskip, and update current optimal partition mode (BestCuMode=Skip) and the smallest rate distortion costs
(MinRDcost=Jskip)；
(3) the rate distortion costs J of the Merge_2Nx2N mode in the case where currently dividing depth is calculatedmerge_2Nx2N, and with before
The smallest rate distortion costs out compare, if Jmerge_2Nx2N< MinRDcost updates current optimal partition mode
(BestCuMode=Merge_2Nx2N) and the smallest rate distortion costs (MinRDcost=Jmerge_2Nx2N)；
(4) the rate distortion costs J of the Inter_2Nx2N mode in the case where currently dividing depth is calculatedInter_2Nx2N, and with before
The smallest rate distortion costs out compare, if JInter_2Nx2N< MinRDcost updates current optimal partition mode
(BestCuMode=Inter_2Nx2N) and the smallest rate distortion costs (MinRDcost=JInter_2Nx2N)；
(5) if the depth of current CU is the maximum division depth that encoder allows, and the size of CU is bigger than 8 × 8, then asks
The rate distortion costs J of Inter_NxN modelnter_NxN, and compared with the smallest rate distortion costs obtained before, if
Jlnter_NxN< MinRDcost updates current optimal partition mode (BestCuMode=Inter_NxN) and the distortion of the smallest rate
Cost (MinRDcost=JInter_NxN), if being unsatisfactory for above-mentioned CU depth and size requirement, skip step 5；
(6) the rate distortion costs J of the Inter_2NxN mode in the case where currently dividing depth is calculatedInter_2NxN, and with before
The smallest rate distortion costs out compare, if Jinter_2NxN< MinRDcost updates current optimal partition mode
(BestCuMode=Inter_2NxN) and the smallest rate distortion costs (MinRDcost=JInter_2NxN)；
(7) the rate distortion costs J of the Inter_Nx2N mode in the case where currently dividing depth is calculatedInter_Nx2NIf JInter_Nx2N<
MinRDcost updates current optimal partition mode (BestCuMode=Inter_Nx2N) and the smallest rate distortion costs
(MinRDcost=JInter_Nx2N)；
(8) rate distortion costs whether current CU needs to calculate AMP mode are calculated, if BestCuMode=Inter_
2Nx2N or BestCuMode=Inter_2NxN, Test_Hor condition are that very, otherwise Test_Hor is false；If
BestCuMode=Inter_2Nx2N or BestCuMode=Inter_Nx2N, Test_Ver condition are true, otherwise Test_
Ver is false.
(9) if Test_Hor condition is true, continuation next step；Otherwise, step 12 is jumped to；
(10) the rate distortion costs J of the Inter_2NxnU mode in the case where currently dividing depth is calculatedInter_2NxnUIf
JInter_2NxnU< M inRDcost updates current optimal partition mode (BestCuMode=Inter_2NxnU) and the smallest rate
Distortion cost (MinRDcost=JInter_2NxnU)；
(11) the rate distortion costs J of the Inter_2NxnD mode in the case where currently dividing depth is calculatedInter_2NxnDIf
JInter_2NxnD< MinRDcost updates current optimal partition mode (BestCuMode=Inter_2NxnD) and the smallest rate
Distortion cost (MinRDcost=JInter_2NxnD)；
(12) if Test_Ver condition is true, continuation next step；Otherwise, step 15 is jumped to；
(13) rate for calculating the Inter_nLx2N mode in the case where currently dividing depth is distorted valence JInter_nLx2NIf JInter_nLx2N
< MinRDcost updates current optimal partition mode (BestCuMode=Inter_nLx2N) and the smallest rate distortion costs
(MinRDcost=JInter_nLx2N)；
(14) the rate distortion costs J of the Inter_nRx2N mode in the case where currently dividing depth is calculatedInter_nRx2NIf
JInter_nRx2N< MinRDcost updates current optimal partition mode (BestCuMode=Inter_nRx2N) and the smallest rate
Distortion cost (MinRDcost=JInter_nRx2N)；
(15) the rate distortion costs J of the Intra_2Nx2N mode in the case where currently dividing depth is calculatedIntra_2Nx2NIf
JIntra_2Nx2N< MinRDcost updates current optimal partition mode (BestCuMode=Intra_2Nx2N) and the smallest rate
Distortion cost (MinRDcost=JIntra_2Nx2N)；
(16) if the current division depth of CU is the permitted depth capacity of encoding setting, its rate of lntra_NxN is calculated
Distortion cost Jlntra_NxNIf Jlntra_NxN< MinRDcost updates current optimal partition mode (BestCuMode=lntra_
) and the smallest rate distortion costs (MinRDcost=J NxNlntra_NxN)；
(17) if the maximum that the current division depth ratio PCM mode of CU allows divides, depth is small, calculates the rate of PCM mode
Distortion cost JPCMIf JPCM< MinRDcost then updates current optimal partition mode (BestCuMode=PCM) and minimum
Rate distortion costs (MinRDcost=JPCM)；
(18) if BestCuMode is permitted no more than coding with what is centered by the current size of Skip mode or CU
Perhaps minimum dimension, jumps to step 19；Otherwise, current CU will be divided into four sub-blocks, and the depth of each sub-block is than current
Block is big by 1.Then to each Sub-CU iteration (1)-(18) algorithm steps, the MinRDcost of sub-block is found out respectively, after adding up
Enabling is Jsub-CU。
(19) compare the rate distortion costs before dividing and being divided into after sub-block, if MinRDcost < JSub-CU, current to divide deeply
CU under degree does not continue to divide, and updates the smallest rate distortion costs (MinCucost=MinRDcost)；Otherwise current CU is divided
For four Sub-CU, and update MinCuCost=JSub-CU。
The algorithm steps that are determined by above-mentioned CU size it is found that encoder will determine the quad-tree partition mode of a CTU,
It not only to determine the division depth of CU, also to traverse all possible division side PU, TU under a certain depth, finally therefrom select
The smallest coding mode of RD Cost and corresponding division depth, therefore the computation complexity of HEVC block partition process is high.
Summary of the invention
The present invention is for above-mentioned the technical problems existing in the prior art, the correlation according to video sequence in time-space domain,
Propose a kind of HEVC block division fast method based on Bayesian decision.In the present invention, regard CU division operation as two points
Class problem, i.e. W={ ωN, ωP, wherein ωNIndicate that current CU does not continue to divide, ωpIt indicates that current CU continues to divide, and uses
Least risk Bayes decision solves two classification problems.
The technical solution adopted by the present invention to solve the technical problems includes the following steps:
Step 1, off-line learning obtain loss function CNPAnd CPNAnd prior probability P (ωN) and P (ωP)。
When doing decision using the Bayesian decision criterion of minimal error rate, if the division result decision error of CU, i.e. shellfish
The judgement whether this decision of leaf divides current depth CU is different from the result of canonical algorithm, will cause under distortion performance
Drop.Meanwhile the CU different with characteristic value for size, the degree that distortion performance caused by the result of decision of mistake declines is not yet
Together.Therefore, the present invention proposes that foundation is based on using the decline of erroneous decision bring distortion performance as bayes risk function
The Bayes decision-making model of minimum risk, to reduce the loss of erroneous decision bring distortion performance, complete rate distortion costs letter
Number is shown below:
JFRD=(SSEluma+ωchroma×SSEchroma)+λmode×Bmode
Wherein, SSElumaIndicate the pixel difference error sum of squares between the original picture block and reconstruction image block of brightness；
SSEchromaIndicate the pixel difference error sum of squares between the original picture block and reconstruction image block of carrier chrominance signal, ωchromaFor color
Spend the weight coefficient of component, λmodeIt is Lagrange's multiplier, BmodeRequired bit number when being to current coding mode.
For the CU of 64 × 64,32 × 32,16 × 16 sizes, loss matrix C is obtained by off-line learning training respectively, is damaged
Matrix is lost to be defined as follows shown in formula:
C hereinPNAnd CNPIt is defined as follows:
JFRD, NAnd JFRD, PBe respectively current CU to be divided do not continue to divide with continue to be divided into four sub-blocks it is corresponding completely
Rate distortion costs.CNPIt indicates to assume ωNIt is true, i.e., judgement hypothesis ω when rate distorted result is " not dividing "PIt sets up, i.e., CU is adopted
Brought value of risk when taking " division " decision；CPNIt indicates to assume ωPIt is very, to be brought when taking " not dividing " decision to CU
Value of risk.CNPAnd CPNMore than or equal to 0, (C when decision is correctNNOr CPP), the penalty values of decision are 0.
Prior probability P (ωR) it is by making ω under statistics current depthRThe specific gravity that the CU of decision accounts for all CU obtains
, i.e.,
P(ωR)=current depth belongs to ωRCU number/current depth CU sum；
Wherein R is N or P；
Step 2, the sequence of pictures for being passed to, judge whether it is in the on-line study stage or whether scene has occurred
Variation carries out depth division to it according to HM canonical algorithm if present frame is in on-line study stage either scene change frame,
The characteristic value for extracting and storing each depth, jumps to step 7；It is no to then follow the steps 3.
Scene change is the severe degree in order to judge video image motion, when there is scene change in video sequence, one
As for i.e. show that violent variation has occurred in the content of video, will be lost between adjacent video frame temporal correlation (depending on
Temporal redundancy of the frequency at switching disappears).In this case, the block based on Bayesian decision divides fast algorithm and discomfort
With once the threshold value for judging whether to divide will change because scene converts, based on offline statistics and on-line study
Bayesian decision will cannot make correct decisions according to the characteristic value extracted.Therefore the invention proposes examined based on scene change
The on-line study stage of survey and fast alternate mechanism stage by stage.For preceding N frame after N frame before sequence and scene change, Dou Yaochong
New characteristic value of extracting establishes new mixed Gauss model.
Since gray value of image can be obtained before encoding present frame by simple computation, if adjacent image is in same
One scene, adjacent frame correlation is stronger, and gray value is not much different；Conversely, adjacent picture frame is in identical bits when occurrence scene transformation
The gray value for setting pixel usually differs larger.
2-1. measures the severe degree of video motion with average gray difference value, i.e., whether occurrence scene converts, video sequence
The average gray difference value calculating of pixel of adjacent two frame in same position is shown below in column:
In formula, P, Q are picture traverse and height, Gn(i, j) is the gray value at n-th frame pixel (i, j), Gn-1(i, j)
For the gray value at the (n-1)th frame pixel (i, j).If G is usually smaller without scene switching.If occurrence scene converts, G is logical
Often it increased dramatically.
In conclusion can be set one for judging whether the detection threshold value of occurrence scene transformation, if G is greater than detection threshold
Value λsc, then it is assumed that scene converts, otherwise it is assumed that scene does not change；λscFor detection threshold value, value is often empirical value.Threshold
The size of value determines the accuracy of scene change, excessive to will lead to missing inspection and too small, will lead to erroneous detection, therefore λscValue should close
Reason.The present invention sets 6 for the frame number in on-line study stage, and threshold value is set as 9, i.e., thinks that occurrence scene is converted as G > 9, otherwise
There is no scene changes.Once scene converts, the statistical parameter of feature vector need be just reevaluated.
Step 3 judges whether picture is the first frame image of fast stage by stage, if it is not, executing step 4；Otherwise it mentions
The characteristic value for taking on-line study stage or scene change frame to store, and mixed Gauss model, the system of model are established based on characteristic value
Meter parameter is estimated to expectation maximization (EM) algorithm of K-Means algorithm initialization.
If the observation information of each CU n dimensional feature vector X=[x1, x2, x3…xn] describe, p (X | ωR), R ∈ N,
P }, it is ωRThe conditional probability density of feature vector, X is observed under state.State ωRPosterior probability p (ωR| X) it can be by pattra leaves
This formula acquires
In order to measure the decision bring risk of mistake, present invention introduces loss function CRQ, indicate rate-distortion optimization result
For decision R, the risk being mistakenly classified as in the case of decision Q.Consider that the Bayes rule in the case where risk becomes:
RN(ωN| X)=CNN p(X|ωN)P(ωN)+CPNp(X|ωP)P(ωP)
RP(ωP| X)=CPPp(X|ωP)P(ωP)+CNP p(X|ωN)P(ωN)
RR(X) it is conditional risk, illustrates that characteristic value X is judged to ωRThe mean value lost when class, CNPIt indicates to assume ωNFor
Very, i.e., ω is assumed in judgement when rate distorted result is " not dividing "PIt sets up, i.e., brought risk when taking " division " decision to CU
Cost.
In order to guarantee expected risk minimum, when must guarantee to do each decision, all make conditional risk minimum, it is such to determine
Plan is referred to as Least risk Bayes decision, and the rule of decision is as follows: if RN(ωN|X)<RP(ωP| X), then ω=ωN, decision
For ωN, indicate the decision that " not dividing " is taken current CU；Otherwise the decision of " division " is taken current CU.
3-1. feature extraction and selection
Come as described above, needing to extract feature vector from observation so that decision is reasonable, for the space-time of reflecting video
Feature, our selected characteristic vectors are JinterAnd Jintra, it is obtained by following formula:
JLRD=SAD+ λpred×Bpred
Jinter=argminP ∈ Inter { JLRD(P)}
Jintra=argminP ∈ Intra { JLRD(P)}
λ hereinpredIt is Lagrange's multiplier, BpredTo information all under current prediction mode, (such as division mode, mode are believed
Breath, transformation coefficient etc.) required bit number, P indicate that the candidate of PU divides when being encoded.The calculating of characteristics extraction is multiple
Miscellaneous degree is almost negligible, because the information of rate distortion costs can be obtained directly from cataloged procedure.
Feature vector, X is obtained from the picture in on-line study stage, if training picture is I frame, the dimension of feature vector
Just drop to 1, X=Jintra, because I frame can only use intra prediction mode.
The probability density function of 3-2. on-line study
In order to Bayesian decision is applied to quick CU partitioning algorithm, need each on-line study stage estimate feature to
The statistical parameter of amount.In the method for proposition, we are according to RN(ωN| X) and RP(ωP| X) value it is current to determine whether to divide
CU to be divided；By Least risk Bayes formula it is found that compare RR(ωR| X) size, need first to acquire p (X | ωN)P
(ωN) and p (X | ωP)P(ωP) value.
In the case of each " division " or " not dividing " for dividing depth, likelihood function is all obeyed dimensional Gaussian and is divided at random
Cloth, i.e. p (X | ωR) can by the Gaussian probability-density function weighted sum approximate representation of two independent probabilities so that
Wherein i is i-th of single Gauss model in mixed Gauss model.The parameter of each Gauss model is θi=(μi, σi),
μiAnd σiIndicate class ωiFeature vector mean vector and variance, the mixing probability function of feature vector, X is
Wherein Θ=(θ1,θ2) be mixed Gauss model distribution parameter collection, P=(p (ω1),p(ω2)) it is corresponding height
The prior probability collection of this model.The parameter of mixed Gauss model with the EM algorithm of K-Means algorithm initialization by being determined.
Step 4, the characteristic value for extracting current CU, acquire p (X | ωR), R ∈ { N, P }.
The characteristic value X for extracting CU to be divided acquires the letter on the Gaussian mixtures of same depth according to locating depth
Numerical value, i.e. p (X | ωN) and p (X | ωP)。
Step 5 decides whether to divide current CU, if dividing, executes mistake described in step 1- step 5 to each Sub-CU
Journey；Otherwise current CU is not divided, jumps to step 6.
Calculating is comparedWithSize, ifIt is larger, current CU is divided into four Sub-CU,
And step 1-5 is executed to each Sub-CU；Otherwise, current CU is not divided.
Step 6 encodes the CU, if current frame image CU divided completion, executes step 7；Otherwise, to next
A CU executes process described in step 1-6.
Fast algorithm is divided according to the block based on Bayes and encodes current CU, then to next CU traversal step 1-6.If working as
All CU have completed to divide in preceding picture frame, then follow the steps 7.Step 7 executes 2-7 step institute to next frame image
Process is stated, until having encoded all video frames.
For a frame image of new incoming, fast algorithm processing equally is divided with the block based on Bayes, until having encoded
All video frames.
So far, entirely the HEVC block division fast method based on Bayesian decision terminates, for the division of a certain depth CU.
Beneficial effect
The method have the advantages that
(1) present invention does high-speed decision to the division mode of current depth CU by Bayesian decision, reduces CU and draws
Divide algorithm complexity, the HEVC scramble time can be greatly reduced.
(2) present invention employs the scene change detection mechanism based on average gray difference value, it is ensured that the stability of method
And accuracy.
(3) using of the invention the experimental results showed that, compared with the original algorithm of HM13.0, the present invention, which announces method, to be reduced
44.5% scramble time, and coding bit rate BD-Rate only increases by 0.94%.
Detailed description of the invention
Fig. 1: the H.265/HEVC partition mode of PU.
Fig. 2: CU, the quad-tree partition of TU.
Fig. 3: on-line study stage and fast replace schematic diagram stage by stage.
Fig. 4: the HEVC block based on Bayesian decision divides fast algorithm main-process stream.
Specific embodiment
By taking the block partition process of 16 sequences such as People on street as an example, the inventive method of proposition is carried out detailed
Explanation.
It is as follows that HEVC block based on Bayesian decision divides the step of fast algorithm:
Step 1, off-line learning obtain loss function CNPAnd CPNAnd prior probability P (ωN) and P (ωP)。
The CU of 64 × 64,32 × 32,16 × 16 sizes is damaged by off-line learning training respectively according to the following formula
Lose Matrix C and prior probability:
JFRD=(SSEluma+ωchroma×SSEchroma)+λmode×Bmode
Wherein, SSEluma, SSEchromaIndicate the picture between the original picture block and reconstruction image block of luminance and chrominance information
Plain difference error sum of squares, ωchromaFor the weight coefficient of chromatic component, λmodeIt is Lagrange's multiplier, BmodeIt is to present encoding
Required bit number when mode.JFRD, NAnd JFRD, PIt is that current CU to be divided does not continue to divide and continues to be divided into four respectively
The corresponding complete rate distortion costs of sub-block.
Prior probability P (ωR) it is by making ω under statistics current depthRThe specific gravity that the CU of decision accounts for all CU obtains
, i.e.,
P(ωR)=current depth belongs to ωRCU number/current depth CU sum
Step 2, the sequence of pictures for being passed to, judge whether it is in the on-line study stage or whether scene has occurred
Variation carries out depth division to it according to HM canonical algorithm if present frame is in on-line study stage either scene change frame,
The characteristic value for extracting and storing each depth, jumps to step 7；It is no to then follow the steps 3.
Video motion is measured using the average gray difference value of pixel of two frame adjacent in video sequence in same position
Severe degree, calculating are shown below:
In formula, P, Q are picture traverse and height, Gn(i, j) is the gray value at n-th frame pixel (i, j), Gn-1(i, j)
For the gray value at the (n-1)th frame pixel (i, j).If G is usually smaller without scene switching.If occurrence scene converts, G is logical
Often it increased dramatically.Once scene converts, the statistical parameter of feature vector need be just reevaluated.Video is judged according to following formula
Whether occurrence scene changes frame:
Scene change is the severe degree in order to judge video image motion, when there is scene change in video sequence, one
As for i.e. show that violent variation has occurred in the content of video, will be lost between adjacent video frame temporal correlation (depending on
Temporal redundancy of the frequency at switching disappears).In this case, the block based on Bayesian decision divides fast algorithm and discomfort
With once the threshold value for judging whether to divide will change because scene converts, based on offline statistics and on-line study
Bayesian decision will cannot make correct decisions according to the characteristic value extracted.Therefore the invention proposes examined based on scene change
The on-line study stage of survey and fast alternate mechanism stage by stage.Schematic diagram is as shown in Figure 3.N frame before sequence and scene are become
N frame, will extract characteristic value again and establish new mixed Gauss model before after changing.
Step 3 judges whether picture is the first frame image of fast stage by stage, if it is not, executing step 4；Otherwise it mentions
The characteristic value for taking on-line study stage or scene codes frame to store, establishes mixed Gauss model, the system of model based on these data
Meter parameter is estimated to the EM algorithm of K-Means algorithm initialization.
Assuming that the observation information of each CU n dimensional feature vector X=[x1, x2, x3…xn] describe, our selected characteristics
Vector is JinterAnd Jintra, it is obtained by following formula:
JLRD=SAD+ λpred×Bpred
Jinter=argminP ∈ Inter { JLRD(P)}
Jintra=argminP ∈ Intra { JLRD(P)}
λ hereinpredIt is Lagrange's multiplier, BpredTo information all under current prediction mode, (such as division mode, mode are believed
Breath, transformation coefficient etc.) required bit number, P indicate that the candidate of PU divides when being encoded.
Feature vector, X is obtained from the picture in on-line study stage, if training picture is I frame, the dimension of feature vector
Just drop to 1, X=Jintra, because I frame can only use intra prediction mode.
In the case of each " division " or " not dividing " for dividing depth, likelihood function is all obeyed dimensional Gaussian and is divided at random
Cloth, i.e. p (X | ωR) can by the Gaussian probability-density function weighted sum approximate representation of two independent probabilities so that
Wherein i is i-th of single Gauss model in mixed Gauss model.The parameter of each Gauss model is θi=(μi, σi),
μiAnd σiIndicate class ωiFeature vector mean vector and variance, the mixing probability function of feature vector, X is
Wherein Θ=(θ1,θ2) be mixed Gauss model distribution parameter collection, P=(p (ω1),p(ω2)) it is corresponding height
The prior probability collection of this model.
The parameter of mixed Gauss model in order to obtain, defining log-likelihood function is
Wherein m indicates the dimension of feature vector.Parametric solution needs to find out model parameter (Θ0, P0) make formula logarithm
Likelihood function value is maximum, then the statistical parameter estimation of mixed model should meet following formula:
The present invention is estimated with statistical parameter of the EM algorithm to mixed Gauss model.Solution is divided into two using EM algorithm
A step: assume initially that Gauss model parameter be it is known, the weight for estimating each Gauss model is removed using parameter；Second
Step, obtains weight and then goes to estimate its parameter using the weight of Gauss model.
The initial value for assuming initially that the mixed Gauss model parameter of the characteristic value of CU block is Θ(0), prior probability initial value is
P(0), set the mixed Gauss model estimates of parameters of t step then as Θ(t)And P(t), then the calculating process of t+1 step is as follows:
(1) E step (calculating expectation): according to the estimated value Θ of current mixed Gauss model parameter t step iteration(t)And P(t), calculate in known current characteristic value, the decision of CU belongs to ωiPosterior probability estimated value:
Obviously, if characteristic value xm(m=1,2) ωiP (the ω of classi|xm,θi)≥P(ωi|xm,θj), j ≠ i, then xm
Belong to ωiA possibility that class, is just very big.
(2) M step (maximizing expectation): obtaining the posterior probability that each data sample belongs to each subclass, this step benefit
The statistical parameter estimation formulas of formula mixed model is solved with gradient descent method.To obtain Θ, estimation of the P in+1 step of kth.It will be right
Number likelihood function brings above formula posterior probability calculated into respectively to Θ and P derivation, can obtain after arrangement:
It is thereinIndicate the probability number for belonging to the characteristic value of the i-th class when+1 step iteration of kth.EM algorithm passes through continuous
The parameter of iteration E step and M step improved model, until the parameter of mixed Gauss model can be obtained in convergence.The present invention
The initial value of middle EM algorithm has K-Means algorithm to determine, and uses Euclidean distance for the metric algorithm of distinctiveness ratio,
Its formula is as follows:
Wherein A={ a1,a2,…,an, B={ a1,a2,…,anIt is two data element entries, respectively have n a mensurable
Characteristic attribute.
K-Means algorithm arbitrarily chooses center of the k object as initial clustering first from sample.Then according to sample
Distinctiveness ratio between cluster centre judges which cluster each sample belongs to.Then the cluster centre of each cluster is updated, is repeated
The iteration above process, until criterion function is restrained, it is shown that specific step is as follows:
The initial cluster center point for randomly selecting k cluster first isThen it is straight to repeat following procedure
To convergence:
(1) calculate each conditional probability p (X | ωi) belonging to class:
(2) in each class j, such mass center is recalculated:
K is the number of cluster, needs to specify before cluster.c(i)Represent conditional probability p (X | ωi) with distance in k class most
That close class, c(i)∈{1,2,3,…,k}.Mass center μjIndicate the cluster centre position of j-th of class, iteration step (1)
With step (2) until cluster centre no longer changes or varies less, in the present invention, k=2 is taken.
Step 4, the characteristic value for extracting current CU, acquire p (X | ωR), R ∈ { N, P }.
The characteristic value X for extracting CU to be divided acquires the letter on the Gaussian mixtures of same depth according to locating depth
Numerical value, i.e. p (X | ωN) and p (X | ωP)。
Step 5 decides whether to divide current CU, if dividing, executes process described in step 1-5 each Sub-CU；It is no
It does not divide current CU then, jumps to step 6.
Calculating is comparedWithSize, ifIt is larger, current CU is divided into four Sub-CU,
And step 1-5 is executed to each Sub-CU；Otherwise, current CU is not divided.
Step 6 encodes the CU, if current frame image CU divided completion, executes step 7；Otherwise, to next
A CU executes process described in step 1-6.
Fast algorithm is divided according to the block based on Bayes and encodes current CU, then to next CU traversal step 1-6.If working as
All CU have completed to divide in preceding picture frame, then follow the steps 7.
Step 7 executes process described in 2-7 step next frame image, until having encoded all video frames.
For a frame image of new incoming, fast algorithm processing equally is divided with the block based on Bayes, until having encoded
All video frames.
So far, entirely HEVC block based on Bayesian decision divides fast method and terminates, for the division of a certain depth CU,
Algorithm flow chart is as shown in Figure 4.
The present invention is tested test environment and is compiled using HEVC standard identifying code HM13.0 in 7 system VS2010 of windows
16 cycle tests are counted and are tested by operation.With full I frame (All Intra, AI), arbitrary access (Random
Access, RA), low latency-P frame (LP), low latency-B frame (LB) be coding environment, setting 22,27,32,37 be used as QP values, benefit
The block of proposition is divided into fast algorithm with the Bjontegaard-Delta bit rate (YBD-rate) of brightness and HEVC standard block is drawn
Divide algorithm to compare, and count the encoding and decoding time, the results are shown in Table 1.
Algorithm proposed by the present invention averagely saves 46% under the configuration of AI compared to HM13.0 canonical algorithm as shown in Table 1
Above scramble time, and BD-Rate only has lost 1.08%；45.9% scramble time, BD- are averagely saved under RA configuration
Rate loss 1.12%；Scramble time under LB configuration reduces 42.8%, and BD-Rate only has lost 0.83%；LP configuration
Averagely save for 42.8% scramble time down, BD-Rate has lost 0.73%, and coding efficiency declines very little.
Table 1 divides fast algorithm and HEVC standard algorithm effect comparing result based on the block of Bayesian decision
Certainly, those of ordinary skill in the art is it should be appreciated that above embodiments are intended merely to illustrate this hair
It is bright, and be not intended as limitation of the invention, as long as within the scope of the invention, all to the variations of above embodiments, modification
Protection scope of the present invention will be fallen into.
Claims (1)
1. a kind of HEVC block based on Bayesian decision divides fast method, it is characterised in that regard CU division operation as two classification
Problem, i.e. W={ ωN, ωP, wherein ωNIndicate that current CU does not continue to divide, ωpIndicate that current CU continues to divide, and using most
Small risk Bayesian decision solves two classification problems, and the specific implementation steps are as follows:
Step 1, off-line learning obtain loss function CNPAnd CPNAnd prior probability P (ωN) and P (ωP)；
Step 2, the sequence of pictures for being passed to, judge whether it is in the on-line study stage or whether scene is become
Change, if present frame is in on-line study stage either scene change frame, depth division is carried out to it according to HM canonical algorithm, is mentioned
The characteristic value for taking and storing each depth, jumps to step 7；It is no to then follow the steps 3；
Step 3 judges whether picture is the first frame image of fast stage by stage, if it is not, executing step 4；Otherwise it extracts
Line learns the characteristic value in stage or the storage of scene change frame, and establishes mixed Gauss model, the statistics ginseng of model based on characteristic value
The EM algorithm to K-Means algorithm initialization is counted to estimate；
Step 4, the characteristic value for extracting current CU, acquire p (X | ωR)R∈{N,P}；
The characteristic value X for extracting CU to be divided acquires the functional value on the Gaussian mixtures of same depth according to locating depth,
I.e. p (X | ωN) and p (X | ωP)；
Step 5 decides whether to divide current CU, if dividing, executes process described in step 1- step 4 to each Sub-CU；It is no
It does not divide current CU then, jumps to step 6；
The judgement divided for current CU is as follows: calculating is comparedWithSize, ifIt is larger, it will work as
Preceding CU is divided into four Sub-CU, and executes step 1-5 to each Sub-CU；
Step 6 encodes the CU, if current frame image CU divided completion, executes step 7；Otherwise, to next CU
Execute process described in step 1-6；
Step 7 jumps to step 2 and handles next video frame, until having encoded all video frames；
Detailed process is as follows for step 1:
It proposes to establish using the decline of erroneous decision bring distortion performance as bayes risk function based on minimum risk
Bayes decision-making model, to reduce the loss of erroneous decision bring distortion performance, complete rate distortion costs function such as following formula institute
Show:
JFRD=(SSEluma+ωchroma×SSEchroma)+λmode×Bmode
Wherein, SSEluma, indicate the pixel difference error sum of squares between the original picture block of brightness and reconstruction image block；SSEchroma
Indicate the pixel difference error sum of squares between the original picture block and reconstruction image block of carrier chrominance signal, ωchromaFor chromatic component
Weight coefficient, λmodeIt is Lagrange's multiplier, BmodeRequired bit number when being to current coding mode；
For the CU of 64 × 64,32 × 32,16 × 16 sizes, loss matrix C is obtained by off-line learning training respectively, loses square
Battle array is defined as follows shown in formula:
C hereinPNAnd CNPIt is defined as follows:
JFRD, NAnd JFRD, PIt is that current CU to be divided does not continue to divide complete rate corresponding with continuing to be divided into four sub-blocks and loses respectively
True cost；CNPIt indicates to assume ωNIt is true, i.e., judgement hypothesis ω when rate distorted result is " not dividing "PIt sets up, i.e., CU is taken
Brought value of risk when " division " decision；CPNIt indicates to assume ωPBe it is true, it is brought when taking " not dividing " decision to CU
Value of risk；CNPAnd CPNMore than or equal to 0, (C when decision is correctNNOr CPP), the penalty values of decision are 0；
Prior probability P (ωR) it is by making ω under statistics current depthRWhat the specific gravity that the CU of decision accounts for all CU obtained, i.e.,
P(ωR)=current depth belongs to ωRCU number/current depth CU sum；
Wherein R is N or P；
Step 2 is implemented as follows:
The severe degree of video motion is measured with average gray difference value, i.e., whether occurrence scene converts, adjacent in video sequence
The average gray difference value calculating of pixel of two frames in same position is shown below:
In formula, P, Q are picture traverse and height, Gn(i, j) is the gray value at n-th frame pixel (i, j), Gn-1(i, j) is the
Gray value at n-1 frame pixel (i, j)；If G is usually smaller without scene switching；If occurrence scene converts, G is usually anxious
Increase severely big；
If G is greater than detection threshold value, then it is assumed that scene converts, otherwise it is assumed that scene does not change, i.e.,
λscFor detection threshold value, value is often empirical value, for judging whether that occurrence scene converts；Detection threshold value is set as 9, that is, work as G >
Think that occurrence scene converts when 9, otherwise there is no scene changes；
Step 3 concrete implementation process is as follows:
If the observation information of each CU n dimensional feature vector X=[x1, x2, x3…xn] describe, p (X | ωR), R ∈ { N, P } is
ωRThe conditional probability density of feature vector, X is observed under state；State ωRPosterior probability p (ωR| X) it can be by Bayesian formula
It acquires
Introduce loss function CRQ, expression rate-distortion optimization result is decision R, the risk being mistakenly classified as in the case of decision Q；It examines
Bayes rule in the case where worry risk becomes:
RN(ωN| X)=CNNp(X|ωN)P(ωN)+CPNp(X|ωP)P(ωP)
RP(ωP| X)=CPPp(X|ωP)P(ωP)+CNPp(X|ωN)P(ωN)
RR(X) it is conditional risk, illustrates that characteristic value X is judged to ωRThe mean value lost when class, CNPIt indicates to assume ωNIt is very, i.e.,
ω is assumed in judgement when rate distorted result is " not dividing "PIt sets up, i.e., brought value of risk when taking " division " decision to CU；
In order to guarantee expected risk minimum, when must guarantee to do each decision, all make conditional risk minimum, such decision quilt
The rule of referred to as Least risk Bayes decision, decision is as follows: if RN(ωN|X)<RP(ωP| X), then ω=ωN, decision is
ωN, indicate the decision that " not dividing " is taken current CU；Otherwise the decision of " division " is taken current CU；
3-1. feature extraction and selection
Feature vector J is extracted from observationinterAnd Jintra, it is obtained by following formula:
JLRD=SAD+ λpred×Bpred
Jinter=argminP ∈ Inter { JLRD(P)}
Jintra=argminP ∈ Intra { JLRD(P)}
λ hereinpredIt is Lagrange's multiplier, BpredTo information all under current prediction mode；Required bit when being encoded
Number, P indicate that the candidate of PU divides；
Feature vector, X is obtained from the picture in on-line study stage, if training picture is I frame, the dimension of feature vector just drops
To 1, X=Jintra, because I frame can only use intra prediction mode；
The probability density function of 3-2. on-line study
According to RN(ωN| X) and RP(ωP| X) value determine whether to divide current CU to be divided；By Least risk Bayes formula
It is found that compare RR(ωR| X) size, need first to acquire p (X | ωN)P(ωN) and p (X | ωP)P(ωP) value；
In the case of each " division " or " not dividing " for dividing depth, likelihood function all obeys dimensional Gaussian random distribution, i.e. p
(X|ωR) by the Gaussian probability-density function weighted sum approximate representation of two independent probabilities, so that
Wherein i is i-th of single Gauss model in mixed Gauss model；The parameter of each Gauss model is θi=(μi, σi), μiAnd σi
Indicate class ωiFeature vector mean vector and variance, the mixing probability function of feature vector, X is
Wherein Θ=(θ1,θ2) be mixed Gauss model distribution parameter collection, P=(p (ω1),p(ω2)) it is corresponding Gaussian mode
The prior probability collection of type；The parameter of mixed Gauss model with the EM algorithm of K-Means algorithm initialization by being determined.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201710014097.5A CN106713935B (en) | 2017-01-09 | 2017-01-09 | A kind of HEVC block division fast method based on Bayesian decision |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201710014097.5A CN106713935B (en) | 2017-01-09 | 2017-01-09 | A kind of HEVC block division fast method based on Bayesian decision |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN106713935A CN106713935A (en) | 2017-05-24 |
| CN106713935B true CN106713935B (en) | 2019-06-11 |
Family
ID=58908184
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201710014097.5A Active CN106713935B (en) | 2017-01-09 | 2017-01-09 | A kind of HEVC block division fast method based on Bayesian decision |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN106713935B (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110730350A (en) * | 2019-09-25 | 2020-01-24 | 杭州电子科技大学 | SHVC (scalable high-speed coding) quick coding method combining coding depth estimation and Bayesian judgment |
Families Citing this family (21)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107547902B (en) * | 2017-07-19 | 2020-03-31 | 南京邮电大学 | Adaptive rate distortion optimization method for surveillance video coding |
| US10735736B2 (en) * | 2017-08-29 | 2020-08-04 | Google Llc | Selective mixing for entropy coding in video compression |
| CN108235018B (en) * | 2017-12-13 | 2019-12-27 | 北京大学 | Point cloud intra-frame coding optimization method and device based on Lagrange multiplier model |
| WO2019113903A1 (en) * | 2017-12-14 | 2019-06-20 | 北京金山云网络技术有限公司 | Coding unit division decision method and device, encoder, and storage medium |
| WO2020061008A1 (en) * | 2018-09-18 | 2020-03-26 | Google Llc | Receptive-field-conforming convolution models for video coding |
| CN109743575A (en) * | 2018-12-05 | 2019-05-10 | 四川大学 | A kind of DVC-HEVC video transcoding method based on naive Bayesian |
| CN109769119B (en) * | 2018-12-18 | 2021-01-19 | 中国科学院深圳先进技术研究院 | Low-complexity video signal coding processing method |
| CN109859248B (en) * | 2018-12-24 | 2024-03-19 | 上海大学 | Time domain difference-based secondary background modeling method |
| CN109714584A (en) * | 2019-01-11 | 2019-05-03 | 杭州电子科技大学 | 3D-HEVC depth map encoding unit high-speed decision method based on deep learning |
| CN109756719B (en) * | 2019-01-28 | 2020-06-26 | 华侨大学 | 3D-HEVC inter-frame rapid method based on CU partition Bayesian decision |
| WO2020190297A1 (en) * | 2019-03-21 | 2020-09-24 | Google Llc | Using rate distortion cost as a loss function for deep learning |
| WO2020242260A1 (en) * | 2019-05-31 | 2020-12-03 | 한국전자통신연구원 | Method and device for machine learning-based image compression using global context |
| CN110581993A (en) * | 2019-08-16 | 2019-12-17 | 中南大学 | Coding unit rapid partitioning method based on intra-frame coding in multipurpose coding |
| CN112822490B (en) * | 2019-11-15 | 2022-09-23 | 北京博雅慧视智能技术研究院有限公司 | Coding method for fast decision of intra-frame coding unit size based on perception |
| CN111107359A (en) * | 2019-12-16 | 2020-05-05 | 暨南大学 | Intra-frame prediction coding unit dividing method suitable for HEVC standard |
| CN111757129B (en) * | 2020-07-07 | 2023-07-07 | 福州大学 | VVC-oriented rapid intra-frame prediction method |
| CN112465664B (en) * | 2020-11-12 | 2022-05-03 | 贵州电网有限责任公司 | AVC intelligent control method based on artificial neural network and deep reinforcement learning |
| CN113129292B (en) * | 2021-04-27 | 2023-04-07 | 陕西师范大学 | Iterative Markov-based synthetic aperture radar image change detection method |
| WO2023024115A1 (en) * | 2021-08-27 | 2023-03-02 | Oppo广东移动通信有限公司 | Encoding method, decoding method, encoder, decoder and decoding system |
| CN114119789B (en) * | 2022-01-27 | 2022-05-03 | 电子科技大学 | Lightweight HEVC chrominance image quality enhancement method based on online learning |
| CN115334308B (en) * | 2022-10-14 | 2022-12-27 | 北京大学深圳研究生院 | Learning model-oriented coding decision processing method, device and equipment |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101833768A (en) * | 2009-03-12 | 2010-09-15 | 索尼株式会社 | Method and system for carrying out reliability classification on motion vector in video |
| CN102333213A (en) * | 2011-06-15 | 2012-01-25 | 夏东 | H.264 compressed domain moving object detection algorithm under complex background |
| CN103237222A (en) * | 2013-05-07 | 2013-08-07 | 河海大学常州校区 | Motion estimation method adopting multi-mode searching manner |
| CN105430396A (en) * | 2015-12-15 | 2016-03-23 | 浙江大学 | Video coding method capable of deciding sizes of coding blocks by means of classification |
Family Cites Families (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US7143353B2 (en) * | 2001-03-30 | 2006-11-28 | Koninklijke Philips Electronics, N.V. | Streaming video bookmarks |
| BRPI0918478A2 (en) * | 2008-09-04 | 2015-12-01 | Thomson Licensing | methods and apparatus for prediction refinement using implicit motion prediction |
-
2017
- 2017-01-09 CN CN201710014097.5A patent/CN106713935B/en active Active
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101833768A (en) * | 2009-03-12 | 2010-09-15 | 索尼株式会社 | Method and system for carrying out reliability classification on motion vector in video |
| CN102333213A (en) * | 2011-06-15 | 2012-01-25 | 夏东 | H.264 compressed domain moving object detection algorithm under complex background |
| CN103237222A (en) * | 2013-05-07 | 2013-08-07 | 河海大学常州校区 | Motion estimation method adopting multi-mode searching manner |
| CN105430396A (en) * | 2015-12-15 | 2016-03-23 | 浙江大学 | Video coding method capable of deciding sizes of coding blocks by means of classification |
Non-Patent Citations (2)
| Title |
|---|
| Fast Coding Unit Size Selection for HEVC based on Bayesian Decisio Rule;Xiaolin Shen et al.;《2012 Picture Coding Symposium》;20120509;第1-4页 * |
| Fast HEVC Intra Coding Unit Size Decision based on an Improved Bayesian Classification Framework;Xuefei Fang et al.;《PCS2013》;20131231;第273-276页 * |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110730350A (en) * | 2019-09-25 | 2020-01-24 | 杭州电子科技大学 | SHVC (scalable high-speed coding) quick coding method combining coding depth estimation and Bayesian judgment |
| CN110730350B (en) * | 2019-09-25 | 2021-08-24 | 杭州电子科技大学 | SHVC (scalable high-speed coding) quick coding method combining coding depth estimation and Bayesian judgment |
Also Published As
| Publication number | Publication date |
|---|---|
| CN106713935A (en) | 2017-05-24 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN106713935B (en) | A kind of HEVC block division fast method based on Bayesian decision | |
| CN111164978B (en) | Method and apparatus for encoding/decoding image and recording medium for storing bit stream | |
| Yan et al. | Group-based fast mode decision algorithm for intra prediction in HEVC | |
| EP4351139A2 (en) | Video encoding/decoding method and device, and recording medium storing bit stream | |
| CN103997646B (en) | Fast intra-mode prediction mode selecting method in a kind of HD video coding | |
| CN109996082A (en) | Method and apparatus for sharing candidate list | |
| Ruiz et al. | Fast intra mode decision algorithm based on texture orientation detection in HEVC | |
| WO2020123053A1 (en) | Image and video coding using machine learning prediction coding models | |
| US10554965B2 (en) | Motion-compensated partitioning | |
| MX2014000163A (en) | Prediction method and apparatus for chroma component of image using luma component of image. | |
| CN107360419B (en) | A kind of movement forward sight video interprediction encoding method based on perspective model | |
| MX2014000160A (en) | Method and apparatus for encoding motion information, and method and apparatus for decoding same. | |
| MX2014000046A (en) | Video encoding method using offset adjustments according to pixel classification and apparatus therefor, video decoding method and apparatus therefor. | |
| Ma et al. | Low complexity rate distortion optimization for HEVC | |
| CN109361920B (en) | Multi-scene-oriented inter-frame rapid prediction algorithm for adaptive decision tree selection | |
| WO2017027113A1 (en) | Super-transform video coding | |
| CN110519591B (en) | Method for quickly selecting prediction mode based on intra-frame coding in multipurpose coding | |
| CN103583043A (en) | Method and device for encoding video image, method and device for decoding video image, and program therefor | |
| 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 | |
| KR20200098433A (en) | Method and apparatus for encoding/decoding image and recording medium for storing bitstream | |
| Duanmu et al. | A novel screen content fast transcoding framework based on statistical study and machine learning | |
| CN109547798B (en) | Rapid HEVC inter-frame mode selection method | |
| WO2018023352A1 (en) | Fast motion estimation method based on online learning | |
| Zheng et al. | Fast H. 264/AVC to HEVC transcoding based on residual homogeneity |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant | ||
| TR01 | Transfer of patent right |
Effective date of registration: 20201215 Address after: Room 806, building 5, Wuhu navigation Innovation Park, Wanbi Town, Wanbi District, Wuhu City, Anhui Province Patentee after: Wuhu Qibo Intellectual Property Operation 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: 20201215 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: 310018 No. 2 street, Xiasha Higher Education Zone, Hangzhou, Zhejiang Patentee before: HANGZHOU DIANZI University |
|
| TR01 | Transfer of patent right | ||
| EE01 | Entry into force of recordation of patent licensing contract |
Application publication date: 20170524 Assignee: Hangzhou Qihu Information Technology Co.,Ltd. Assignor: Wuhu Qibo Intellectual Property Operation Co.,Ltd. Contract record no.: X2021330000547 Denomination of invention: A fast hevc block partition method based on Bayesian decision Granted publication date: 20190611 License type: Common License Record date: 20211028 |
|
| EE01 | Entry into force of recordation of patent licensing contract | ||
| EE01 | Entry into force of recordation of patent licensing contract |
Application publication date: 20170524 Assignee: Hangzhou Julu enterprise management consulting partnership (L.P.) Assignor: Wuhu Qibo Intellectual Property Operation Co.,Ltd. Contract record no.: X2021330000726 Denomination of invention: A fast hevc block partition method based on Bayesian decision Granted publication date: 20190611 License type: Common License Record date: 20211109 |
|
| EE01 | Entry into force of recordation of patent licensing contract | ||
| EC01 | Cancellation of recordation of patent licensing contract |
Assignee: Hangzhou Julu enterprise management consulting partnership (L.P.) Assignor: Wuhu Qibo Intellectual Property Operation Co.,Ltd. Contract record no.: X2021330000726 Date of cancellation: 20221103 Assignee: Hangzhou Qihu Information Technology Co.,Ltd. Assignor: Wuhu Qibo Intellectual Property Operation Co.,Ltd. Contract record no.: X2021330000547 Date of cancellation: 20221103 |
|
| EC01 | Cancellation of recordation of patent licensing contract |