CN105430391B - The intraframe coding unit fast selecting method of logic-based recurrence classifier - Google Patents
The intraframe coding unit fast selecting method of logic-based recurrence classifier Download PDFInfo
- Publication number
- CN105430391B CN105430391B CN201510885775.6A CN201510885775A CN105430391B CN 105430391 B CN105430391 B CN 105430391B CN 201510885775 A CN201510885775 A CN 201510885775A CN 105430391 B CN105430391 B CN 105430391B
- Authority
- CN
- China
- Prior art keywords
- downwards
- logic
- dividing
- value
- coding unit
- 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
Landscapes
- Compression Or Coding Systems Of Tv Signals (AREA)
Abstract
The present invention provides a kind of intraframe coding unit fast selecting method of logic-based time rule classifier, the method is modeled as the classification problem of " dividing downwards " and " not dividing downwards " two classes for whether a coding unit divides downwards, returns rule classifier using logic to solve this two classes classification problem.Three video sequences of off-line learning obtain logic and return rule classifier optimal logic time rule coefficient, select optimal decision feature by calculating the validity of multiple candidate features.The logic obtained using off-line learning is returned rule classifier and is quickly selected intraframe coding unit, to skip other unnecessary predictive coding mode computations, HEVC encoder frames interior coding complexity can be effectively reduced, the real-time application of real-time HEVC encoder is conducive to.
Description
Technical field
The invention belongs to technical field of video coding, are based on a new generation video encoding standard High Efficiency
Video Coding (HEVC) specifically quickly selects the coding unit of the intra prediction of HEVC.
Background technique
In order to solve the impact that the data transfer rate of high definition and ultra high-definition video sharp increase is brought to network transmission, international telecommunication
The video of Union Telecommunication Standards department (International Telecommunication Union-Telecom, ITU-T)
Coding Experts group (Video Coding Experts Group, VECG) and International Organization for standardization/International Electroteclinical committee member
Meeting (International Organization for Standardization/International Electro-
Technical Commission, ISO/IEC) dynamic image expert group (Moving Picture Experts Group,
MPEG Video coding joint working group (Joint Collaborative Team on Video) has been set up in January, 2010
Coding, JCT-VC), which has formulated the video compression standard HEVC (High of new generation with higher compression efficiency
Efficiency Video Coding)。
As video encoding standard of new generation, HEVC (H.265) still falls within the hybrid encoding frame of prediction plus transformation, and
Technological innovation is carried out to the modules of the frame, including flexible division mode, finer intra prediction, completely new
Merge mode, motion-vector prediction competition-based, the sub-pixel interpolation based on discrete cosine transform, sampling point adaptive equalization
Filter (Sample Adaptive Offset, SAO), Tile and pursuit coding parallel processing technique etc..These new technologies
So that H.264/AVC HEVC code efficiency ratio is doubled.However, the optimum choice of huge and flexible coding parameter collection,
But also the complexity of the encoder of HEVC sharply increases, which prevent application and the popularizations of HEVC standard.Therefore, guaranteeing to regard
While frequency image coding quality, the fast and efficiently code optimization algorithm for studying HEVC is most important.
Summary of the invention
Under conditions of guaranteeing coding efficiency, the choosing of HEVC intraframe coding dividing elements is quickly carried out
It selects to reduce encoding operation complexity.It is fast that a kind of intraframe coding unit that logic-based returns classifier is provided based on this purpose
Fast selection method can greatly reduce the encoder complexity of HEVC.
In order to achieve the above object, the intraframe coding unit that the present invention provides a kind of logic-based recurrence classifier quickly selects
The division of coding unit is modeled as classification problem by selection method, by directly analyzing the feature of inputting video data, with simple
And efficient logic returns the quick predict that rule sorting criterion is divided, so as to avoid the traversal search based on rate-distortion optimization
Algorithm reduces the computation complexity of encoder.And excavate and divide closely related input image data feature with coding unit,
Select optimal feature subset as decision feature.
The invention mainly comprises two parts, first part is that off-line learning obtains logistic regression classifier, and second part is
The logistic regression classifier that study obtains is used to CU to divide in selection course.
Described method includes following steps：
Step 1：The multiple video sequences of offline statistical learning, using each coding unit of HEVC Plays process code
(CU), it and records optimal CU under different quantization parameters (QP) and different coding depth and is divided into " divide downwards " and " does not draw downwards
Point " when candidate feature value；
Step 2：The selection of logistic regression classifier input feature vector, step 1 is counted offline obtained candidate feature value into
Row compares, and selects best input feature vector；
Step 3:The best input feature vector obtained according to step 2, acquires logistic regression classifier with gradient ascent algorithm
Optimal logic return rule coefficient；
Step 4:The logistic regression classifier that step 3 is obtained is used CU and is divided in selection course；
Step 5：All prediction modes of this layer of CU are encoded, and acquire the value that logic returns rule input feature vector；
Step 6：The value for the input feature vector that step 5 acquires is substituted into logistic regression classifier determine whether current CU is downward
It divides, if it is decided that current CU is that " not dividing downwards " then goes to step 8, otherwise goes to step 7；
Step 7：CU continues to be divided into 4 identical sub- CU downwards；
Step 8：The coding for terminating present encoding tree unit (CTU), carries out the coding of next CTU.
Compared with prior art, the present invention has following beneficial effect：
The present invention is the CU cutting selection method that logic-based returns rule classification, and the division of CU is modeled as two class classification problems
That is " division downwards " and " not dividing downwards ", by directly analyzing the feature of inputting video data, with simple and efficient logic
It returns the quick predict that rule sorting criterion is divided and greatly reduces the meter of intraframe coding under the premise of guaranteeing coding quality
Calculate complexity.It is verified, after having used the invention, at the reference encoder device HM16.0 of HEVC, for multiple video sequences
Column encoder, coding rate averagely has 55.5% speed to be promoted, and BD-rate (code rate under phase homogenous quantities) only has 1.29%
Increase.
Detailed description of the invention
Upon reading the detailed description of non-limiting embodiments with reference to the following drawings, other feature of the invention,
Objects and advantages will become more apparent upon：
Fig. 1 is that the logic of one embodiment of the invention returns the result schematic diagram of rule classification；
Fig. 2 is that the logic-based of one embodiment of the invention returns rule classification CU division decision flowchart in advance；
Fig. 3 is figure compared with one embodiment of the invention is distorted (RD) performance curve with Raw encoder rate.
Specific embodiment
The present invention is described in detail combined with specific embodiments below.Following embodiment will be helpful to the technology of this field
Personnel further understand the present invention, but the invention is not limited in any way.It should be pointed out that the ordinary skill of this field
For personnel, without departing from the inventive concept of the premise, various modifications and improvements can be made.These belong to the present invention
Protection scope.
The present invention is based on video encoding standard High Efficiency Video Coding (HEVC) of new generation, provide one
The method that kind quickly selects the coding unit of the intra prediction of HEVC.Specific implementation step is as follows：
Step 1：3 video sequences of offline statistical learning, and various parameters are recorded, specifically：
Using HEVC Plays process code CU, and record optimal CU under different quantization parameters (QP) and different coding depth
The value of candidate feature when being divided into " dividing downwards " and " not dividing downwards ".Wherein candidate feature includes directly predicting (DC) mould
The variance of the prediction error of formula, the quantization parameter of the variance of the prediction error of current layer optimization model, current layer optimization model
Bit needed for variance, the rate distortion costs value of current layer optimization model, current layer optimization model encode.
In this step, 3 video sequences of offline statistical learning also can be set as needed it in other embodiments
His video sequence number, this essential influence no on realization of the invention.
Step 2：The selection of logistic regression classifier input feature vector, step 1 is counted offline obtained candidate feature value into
Row compares, and selects best input feature vector, specifically：
Because the feature selecting of input data is most important to the design of classifier, if correlation is not high or even uncorrelated
Input data, not but not help to classify, can make classifier inaccurate instead.On the contrary, if selecting good feature, it can
To more fully understand data, the storage overhead for sample can be reduced, it is possible to reduce the time of training and prediction, can be improved
Estimated performance.Therefore, feature selecting is shown for the intraframe coding unit fast selecting method that logic-based returns classifier
It obtains more important.F-score is to measure the two set i.e. evaluation of the data variance of " division downwards " and " not dividing downwards "
Standard, it calculates simply, is defined as follows：
Wherein,It is whole samples, " not downwards divide " class sample respectively, inputs in " dividing downwards " class sample
The mean value of the ith feature of vector, n+, n-Be respectively the sum of " not downwards divide " class sample, " dividing downwards " class sample it is total
Number,Be respectively " not downwards divide " class, in " dividing downwards " class k-th of sample ith feature.It can be obtained from above formula
Out, mean value of the ith feature in two classes and the difference of the mean value of all samples are bigger, and respectively ith feature in classification
Variance it is smaller, the value of F (i) is bigger, then the separation property of ith feature and classification is bigger, i.e., this feature has very well classification
Directive function.F (i) value of candidate feature is calculated separately, then input of maximum two features of selected value as classifier
Feature.
Step 3：The best input feature vector obtained according to step 2, acquires logistic regression classifier with gradient ascent algorithm
Optimal logic return rule coefficient.Gradient rise method is based on the idea that be：The maximum value of certain function is found, the best way is edge
The gradient direction of function f (w) seek, calculation formula is as follows：
Wherein vector w is optimal parameter to be found, and α is the size moved along gradient direction,Indicate gradient,Table
Show the gradient of function f (w), f (w) is defined as follows：
F (w)=w0x0+w1x1+…+wnxn (3)
Wherein wi, i=0,1 ... n are the one-component of vector w, xi, i=0,1 ... n are input feature vector values.Formula (2) will
It is iterated execution always, the error range of permission is reached until reaching specified the number of iterations or gradient rise method.
Step 4：The logistic regression classifier that step 3 is obtained is used CU and is divided in selection course.
Step 5：All prediction modes of this layer of CU are encoded, and acquire the value that logic returns rule input feature vector, specifically
It is：
Start all prediction modes of coding present encoding depth layer CU, and obtains the optimal coding mould of present encoding depth
Formula calculates the value for the input feature vector that step 2 selects, and obtains the characteristic value of input classifier.
Step 6：The value for the input feature vector that step 5 acquires is substituted into logistic regression classifier determine whether current CU is downward
It divides, if it is decided that current CU is that " not dividing downwards " then goes to step 8, otherwise goes to step 7, specifically：
The input feature vector value that step 5 acquires is substituted into logistic regression classifier and determines whether current CU divides downwards, is sentenced
It fixes then as follows：
Wherein CunsplitAnd Csplit" not dividing downwards " and " dividing downwards " are respectively indicated, x is that the input that step 5 obtains is special
Value indicative.p(Csplit| x) indicate that the posterior probability of " dividing downwards ", calculation formula are as follows：
Wherein z is the linear combination of characteristic value x, and calculation formula is as follows：
Z=w0x0+w1x1+…+wnxn (5)
Wherein wi, i=0,1 ... n are that the optimal logic that step (3) acquires returns rule coefficient, xi, i=0,1 ... n are input feature vectors
Value.And the posterior probability p (C of " not dividing downwards "unsplit| it x) calculates as follows：
p(Cunsplit| x)=1-p (Csplit|x) (6)
If it is determined that current CU is that " not dividing downwards " then goes to step (8), 7 are otherwise gone to step.
Step 7：CU continuation is divided into 4 identical sub- CU downwards, specifically：
Current layer CU continuation is divided into 4 identical sub- CU downwards, into the intra prediction mode of next coding depth layer
Coding.
Step 8：The coding for terminating present encoding tree unit (CTU), carries out the coding of next CTU.
Fig. 1 is the result schematic diagram that logic proposed by the invention returns rule classification, and Fig. 2 is proposed by the invention based on patrolling
It collecting back rule classification CU and divides and shift to an earlier date decision flowchart, Fig. 3 is then present invention figure compared with the RD performance of original reference encoder, from
Figure shows that RD performance is not decreased obviously it can be seen that two RD curve co-insides.It is verified, after having used the invention,
Under the reference encoder device HM16.0 of HEVC, averagely there is 55.5% speed to be promoted for multiple video sequence coding devices, and BD-
Rate (code rate under phase homogenous quantities) only has 1.29% increase.
Specific embodiments of the present invention are described above.It is to be appreciated that the invention is not limited to above-mentioned
Particular implementation, those skilled in the art can make various deformations or amendments within the scope of the claims, this not shadow
Ring substantive content of the invention.
Claims (3)
1. the intraframe coding unit fast selecting method that a kind of logic-based returns classifier, it is characterised in that including walking as follows
Suddenly：
Step 1：The multiple video sequences of offline statistical learning, using each coding unit CU of HEVC Plays process code, and remember
It records optimum code unit CU under different quantization parameter QP and different coding depth and is divided into " dividing downwards " and " not dividing downwards "
When candidate feature value；
Step 2：Step 1 is counted obtained candidate feature value offline and compared by the selection of logistic regression classifier input feature vector
Compared with selecting best input feature vector；
Step 3：The best input feature vector obtained according to step 2 acquires logistic regression classifier most with gradient ascent algorithm
Excellent logic returns rule coefficient；
Step 4：The logistic regression classifier that step 3 is obtained is used coding unit CU and is divided in selection course；
Step 5：All prediction modes of this layer of coding unit CU are encoded, and acquire the value that logic returns rule input feature vector；
Step 6：The value for the input feature vector that step 5 is acquired substitutes into logistic regression classifier whether determine current coded unit CU
It divides downwards, if it is decided that current coded unit CU is that " not dividing downwards " then goes to step 8, otherwise goes to step 7；
Step 7：Coding unit CU continues to be divided into 4 identical sub- coding unit CU downwards；
Step 8：The coding for terminating present encoding tree unit CTU, carries out the coding of next coding tree unit CTU；
In step 2：The selection of logistic regression classifier input feature vector, F-score be measure two set i.e. " downwards division " and
The evaluation criterion of the data variance of " not dividing downwards ", it is defined as follows：
Wherein,It is whole samples, " not downwards divide " class sample, input vector in " dividing downwards " class sample respectively
Ith feature mean value, n+, n-It is the sum of the sum of " not downwards divide " class sample, " dividing downwards " class sample respectively,Be respectively " not downwards divide " class, in " dividing downwards " class k-th of sample ith feature；It can be obtained from above formula,
Mean value of the ith feature in two classes and the difference of the mean value of all samples are bigger, and respectively in classification ith feature side
Difference is smaller, and the value of F (i) is bigger, then the separation property of ith feature and classification is bigger, i.e., this feature has good finger to classification
Lead effect；F (i) value of candidate feature is calculated separately, then maximum two features of selected value are as logistic regression classifier
Input feature vector.
2. logic-based according to claim 1 returns the intraframe coding unit fast selecting method of classifier, feature
It is：In step 3：The optimal logic for acquiring logistic regression classifier returns rule coefficient, and using Optimal gradient rise method, gradient rises
Method is：The maximum value for finding certain function is sought along the gradient direction of function f (w), calculation formula is as follows：
Wherein vector w is the optimal parameter to be found, and α is the size moved along gradient direction,Indicate gradient,It indicates
The gradient of function f (w), f (w) are defined as follows：
F (w)=w0x0+w1x1+…+wnxn (3)
Wherein wi, i=0,1 ... n are the one-component of vector w, xi, i=0,1 ... n are input feature vector values, formula (2) will always by
Iteration executes, and the error range of permission is reached until reaching specified the number of iterations or gradient rise method.
3. -2 described in any item logic-baseds return the quick selecting party of intraframe coding unit of classifier according to claim 1
Method, it is characterised in that：In step 6：The value for the input feature vector that step 5 is acquired, which substitutes into logistic regression classifier, determines current compile
Whether code unit CU divides downwards, if it is decided that current coded unit CU is that " not dividing downwards " then goes to step 8, is otherwise gone to step
7, decision criteria is as follows：
Wherein CunsplitAnd Csplit" not dividing downwards " and " dividing downwards " are respectively indicated, x is the input feature vector that step 5 obtains
Value；If the ratio that the likelihood ratio for being selected as " not dividing downwards " is selected as the probability of " dividing downwards " is big, it will be selected to be and " do not draw downwards
Point ", otherwise it is selected as " dividing downwards ".
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201510885775.6A CN105430391B (en) | 2015-12-04 | 2015-12-04 | The intraframe coding unit fast selecting method of logic-based recurrence classifier |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201510885775.6A CN105430391B (en) | 2015-12-04 | 2015-12-04 | The intraframe coding unit fast selecting method of logic-based recurrence classifier |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN105430391A CN105430391A (en) | 2016-03-23 |
| CN105430391B true CN105430391B (en) | 2018-11-16 |
Family
ID=55508286
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201510885775.6A Active CN105430391B (en) | 2015-12-04 | 2015-12-04 | The intraframe coding unit fast selecting method of logic-based recurrence classifier |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN105430391B (en) |
Families Citing this family (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US10382770B2 (en) * | 2017-02-06 | 2019-08-13 | Google Llc | Multi-level machine learning-based early termination in partition search for video encoding |
| CN107071496B (en) * | 2017-05-14 | 2020-01-17 | 北京工业大学 | Fast depth selection method for H.265/HEVC inter-frame coding unit |
| SG11202005658VA (en) * | 2017-12-14 | 2020-07-29 | Beijing Kingsoft Cloud Network Technology Co Ltd | Coding unit division decision method and device, encoder, and storage medium |
| CN108629181A (en) * | 2018-05-11 | 2018-10-09 | 湖南大学 | The Cache attack detection methods of Behavior-based control |
| CN108768986B (en) * | 2018-05-17 | 2020-09-08 | 中国科学院信息工程研究所 | Encrypted traffic classification method, server and computer readable storage medium |
| 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 |
| 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 |
| CN111654698B (en) * | 2020-06-12 | 2022-03-22 | 郑州轻工业大学 | Fast CU partition decision method for H.266/VVC |
Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN104796693A (en) * | 2015-04-01 | 2015-07-22 | 南京邮电大学 | Rapid HEVC CU deep partition coding method |
| CN104837019A (en) * | 2015-04-30 | 2015-08-12 | 上海交通大学 | AVS-to-HEVC optimal video transcoding method based on support vector machine |
Family Cites Families (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US11178407B2 (en) * | 2012-11-19 | 2021-11-16 | Texas Instruments Incorporated | Adaptive coding unit (CU) partitioning based on image statistics |
-
2015
- 2015-12-04 CN CN201510885775.6A patent/CN105430391B/en active Active
Patent Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN104796693A (en) * | 2015-04-01 | 2015-07-22 | 南京邮电大学 | Rapid HEVC CU deep partition coding method |
| CN104837019A (en) * | 2015-04-30 | 2015-08-12 | 上海交通大学 | AVS-to-HEVC optimal video transcoding method based on support vector machine |
Non-Patent Citations (3)
| Title |
|---|
| An Complexity Reduction Scheme With Fast Prediction Unit Decision For HEVC Intra Encoding;M.U.K.Khan,M.Shafique,J.Henkel;《International Conference on Image Processing(ICIP)》;20130930;全文 * |
| An Fast CU Size Decision and Mode Decision Algorithm for HEVC Intra Coding;L.Shen,Z.Zhang;《IEEE Transactions on Consumer Electronics》;20130228;全文 * |
| Content based Hierarchical Fast Coding Unit Decision Algorithm For HEVC;J.Leng,L.Sun,T.IKENAGA,et.al.;《International Conference on Multimedia and Signal Processing (CMSP)》;20110531;全文 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN105430391A (en) | 2016-03-23 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN105430391B (en) | The intraframe coding unit fast selecting method of logic-based recurrence classifier | |
| Zhang et al. | Fast coding unit depth decision algorithm for interframe coding in HEVC | |
| Hu et al. | Fast HEVC intra mode decision based on logistic regression classification | |
| KR101344200B1 (en) | Dynamic selection of motion estimation search ranges and extended motion vector ranges | |
| US9031130B2 (en) | Image prediction/encoding device, image prediction/encoding method, image prediction/encoding program, image prediction/decoding device, image prediction/decoding method, and image prediction decoding program | |
| CN104796694B (en) | Optimization intraframe video coding method based on video texture information | |
| CN111462261B (en) | Fast CU partitioning and intra-frame decision method for H.266/VVC | |
| Wu et al. | HG-FCN: Hierarchical grid fully convolutional network for fast VVC intra coding | |
| CN105981382A (en) | Hash-Based Encoder Decisions For Video Coding | |
| CN105306947B (en) | video transcoding method based on machine learning | |
| CN105706450A (en) | Encoder decisions based on results of hash-based block matching | |
| CN103797803A (en) | Method and apparatus for entropy encoding/decoding | |
| CN1283291A (en) | Apparatus and method for performing scalable hierarchical motion estimation | |
| KR101772252B1 (en) | Motion prediction method | |
| CN111479110B (en) | Fast affine motion estimation method for H.266/VVC | |
| Westland et al. | Decision trees for complexity reduction in video compression | |
| Chen et al. | Intra frame rate control for versatile video coding with quadratic rate-distortion modelling | |
| CN106534849A (en) | Fast HEVC interframe coding method | |
| Saldanha et al. | Learning-based complexity reduction scheme for VVC intra-frame prediction | |
| Chiang et al. | A multi-pass coding mode search framework for AV1 encoder optimization | |
| Ruiz et al. | A unified architecture for fast HEVC intra-prediction coding | |
| Yang et al. | Fast intra encoding decisions for high efficiency video coding standard | |
| Hemanth et al. | A pattern-based artificial bee colony algorithm for motion estimation in video compression techniques | |
| Liu et al. | Video coding and processing: a survey | |
| CN106878754B (en) | A kind of 3D video depth image method for choosing frame inner forecast mode |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| C06 | Publication | ||
| PB01 | Publication | ||
| C10 | Entry into substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |