CN109743575A - A kind of DVC-HEVC video transcoding method based on naive Bayesian - Google Patents
A kind of DVC-HEVC video transcoding method based on naive Bayesian Download PDFInfo
- Publication number
- CN109743575A CN109743575A CN201811479353.9A CN201811479353A CN109743575A CN 109743575 A CN109743575 A CN 109743575A CN 201811479353 A CN201811479353 A CN 201811479353A CN 109743575 A CN109743575 A CN 109743575A
- Authority
- CN
- China
- Prior art keywords
- dvc
- hevc
- naive bayesian
- video
- training
- 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.)
- Pending
Links
Landscapes
- Compression Or Coding Systems Of Tv Signals (AREA)
Abstract
The present invention provides a kind of DVC-HEVC code-transferring method and its transcoder based on naive Bayesian, relates generally to divide for the higher coding unit of HEVC cataloged procedure complexity in transcoder and carries out high-speed decision.Utilize prediction residual obtained by DVC decoding end, motion vector and texture-rich degree, a kind of sorter model based on naive Bayesian is established using the method for machine learning, avoid the high traversal iteration rate-distortion optimization process of complexity in HEVC, high-speed decision is carried out to CU partition mode, to reduce the computation complexity of its coding side.The experimental results showed that the present invention compared with traditional cascade code-transferring method, greatly reduces the scramble time in the case where code efficiency and Y-PSNR (PSNR) loss very littles.
Description
Technical field
The present invention relates to the Video Transcoding Technology problems in field of picture communication, more particularly, to one kind based on simple pattra leaves
This distributed video coding is to the Video Transcoding Technology between HEVC standard.
Background technique
With the rapid development of the technologies such as digital communication, integrated circuit, the video communication between mobile terminal device is obtained
It is widely applied, while having expedited the emergence of such as wireless monitor video, remote scene commander various new video communication services.However,
For specific environments such as mountain area disaster or Military Applications, generally there is equipment power consumption, memory capacity etc. to be limited
Feature.Conventional video coding standard (as H.26X serial) has coding side complicated using the hybrid encoding frame of prediction plus transformation
And the feature that decoding end is simple.Compared to H.264/AVC, video encoding standard (High Efficiency Video of new generation
Coding, HEVC) it is based on identical video coding framework [1], a variety of optimization processings are taken, such as based on the coding of quaternary tree
Block partition structure, the intra prediction mode of different angle, motion-vector prediction technology based on adaptable search etc., are guaranteeing
The compression ratio of left and right is doubled under the premise of same video quality.Distributed video coding (Distributed Video
Coding, DVC) it is different from traditional code mode, it is turned by the way of absolute coding combined decoding, by complexity from coding side
Move to decoding end.Therefore, DVC is communicated with low-power-consumption video of the transcoding of traditional code mode between mobile terminal device and is mentioned
A kind of extremely effective realization approach is supplied.
Transcoding process is in the nature first decoding re-encoding, assumes responsibility for the treatment process of two high complexities, calculation amount is extremely
It is huge, seriously affect the real work efficiency of Video transmission system.Therefore, to avoid excessively high communication delay, how to guarantee
Under the premise of video quality is basically unchanged, the computation complexity for reducing transcoding process is particularly important.Academic circles at present for DVC to
Conventional video standards transcoding has carried out extensive work, and majority has been achieved for better progress.Alberto
Corrales-Garc í a etc. divides according to the residual error of the block size size and DVC decoding end side information of macro block when H.264 encoding
Cloth have the characteristics that higher similitude this, realize the acceleration of mode selection algorithm.When J.L.Martinez etc. is decoded using DVC
The motion vector of generation reduces computation complexity when H.264 encoding in motion estimation process.Rong Song etc. is based on the excellent of cloud computing
Gesture proposes a kind of low complex degree based on cloud transcoding video communication system end to end.The above method can be to a certain degree
Upper acceleration DVC-H.264 transcoding process, but the research about DVC-HEVC transcoding is still in the elementary step.
Summary of the invention
The purpose of the present invention is accelerate HEVC cataloged procedure in DVC-HEVC transcoder.The present invention is for complexity in HEVC
Higher coding unit (Coding Unit, CU) partition process analyzes the correlated characteristic information knot generated in DVC decoding process
The method for closing machine learning establishes the quick partitioning model of CU, proposes a kind of DVC-HEVC video code conversion based on naive Bayesian
Method.Compared to traditional cascade Transcoding Scheme, the feelings of method of the invention in code efficiency and Y-PSNR loss all very littles
Under condition, the computation complexity that HEVC is encoded in transcoder can be greatly reduced.
The basic idea of the invention is that using the division of self-characteristic and coding unit of video content, often there is phases
Association system, by analyzing its connection, the texture using prediction residual obtained by DVC decoding end, motion vector and the frame is rich
Fu Du, the method based on machine learning establish a kind of sorter model based on naive Bayesian, avoid complicated in HEVC
High traversal iteration rate-distortion optimization process is spent, high-speed decision is carried out to CU partition mode, to reduce its coding side
Computation complexity.
HEVC defines a set of completely new image segmentation mode, includes coding unit, predicting unit, converter unit, with this
It is more flexible that coding mode, thus itself correlation properties and HEVC of video sequence are efficiently selected according to video content characteristic
CU partition mode has very big correlation in coding.Typically for video content have compared with horn of plenty details characteristic region, one
As be divided into lesser coding unit, coding quality is promoted with this, reduces distortion, it is on the contrary then be divided into biggish unit, can
To reduce encoder complexity, code efficiency is improved.In addition, the region more violent for motion transform between adjacent video frames,
It is larger with residual values to show as motion vector, such region is typically divided into lesser unit, avoids cataloged procedure from causing excessively high
Distortion.Therefore, the correlated characteristic information of above-mentioned analysis all can be used as the foundation of CU partition mode, while these characteristic informations are all
It can be obtained in DVC decoding process.
Naive Bayes Classification is the classic algorithm in machine learning field, is a kind of classification learning method for having supervision,
With attribute conditions independence assumption.Equipped with the possible category label of N kind, i.e. y={ C1,C2,...,CN, there are n to each
Sample x={ the f of feature1,f2,...,fn, it selects that posterior probability P (Cx) maximum classification can be made as optimal classification classification.Belong to
Property conditional independence assumption be usually unable to satisfy in a practical situation, but Naive Bayes Classification still show high-precision and efficiently
Rate, compared with other algorithms, it has the characteristics that, and algorithm is simple, classification error rate is small.The present invention is by prediction residual information, texture
The input feature vector of richness and motion vector information as model carries out classification based training based on Naive Bayes Classification Algorithm and obtains
Optimal classification pattern function can carry out CU using disaggregated model quickly to divide decision.
In the design of DVC-HEVC transcoder, the committed step for improving transcoding real-time is how to efficiently use DVC decoding
The relevant information generated in the process accelerates HEVC cataloged procedure.The present invention utilizes obtainable prediction residual in DVC decoding process
Training frame number K is arranged, to WZ frame in the input parameter of information, texture-rich degree and motion vector information as CU partitioning model
The characteristic information of preceding K frame obtains optimal classification device model using Naive Bayes Classification Algorithm, and WZ frame thereafter then utilizes this point
Class model carries out quick CU piecemeal.By means of the invention it is also possible to which it is higher to skip computation complexity in HEVC coding module
Layer-by-layer rate-distortion optimization process, to achieve the purpose that reduce HEVC encoder complexity.
Specific mainly includes following procedure step:
(1) after DVC code stream decoding obtain rebuild frame sequence, by reconstruction sequence non-key frame divide training frames and
Test frame;
(2) HEVC coding is carried out to training frames, while extracts supervision message of its CU division result as model training when,
In conjunction with prediction residual, texture-rich degree and the motion vector obtained in DVC decoding process as input feature vector, 64 are established respectively
× 64,32 × 32,16 × 16 training sets obtain the CU partitioning model of each size;
(3) by test frame sequence characteristic information input (2) in each size disaggregated model, obtain CU whether continue to
The result of lower division;
CU traversal recurrence division in HEVC accounts for 90% of computation complexity or more, therefore it is entire that the present invention is improved
The highest place of computation complexity in HEVC Video coding in DVC-HEVC transcoder.The step of most critical of the present invention is basis
Obtainable prediction residual, texture-rich degree and motion vector characteristic information in DVC decoding process are based on Naive Bayes Classification
Algorithm establishes optimal classification device model, and test frame is allowed quickly to carry out the selection of CU partition mode by characteristic information.Cause
This, in terms of the computation complexity, the method for the present invention is conceived in DVC-HEVC transcoder computation complexity in HEVC Video coding
In place of improved most critical.
Detailed description of the invention
Fig. 1 is that the present invention is based on the DVC-HEVC video transcoding method system block diagrams of naive Bayesian;
Fig. 2 is that the present invention is based on the streams of the quick division methods of CU of the DVC-HEVC video code conversion test frame of naive Bayesian
Cheng Tu;
Fig. 3~6 are the rate distortion curve figure of the method for the present invention and traditional Cascade algorithms, wherein Fig. 3 is
The rate distortion curve of BasketballDrill;Fig. 4 is the rate distortion curve of BQMall;Fig. 5 is the rate distortion curve of Johnny;
Fig. 6 is the rate distortion curve of FourPeople.
Specific embodiment
The present invention is described in further detail with reference to the accompanying drawings and embodiments, it is necessary to, it is noted that below
Embodiment is served only for that the present invention is described further, should not be understood as limiting the scope of the invention, fields
Personnel be skillful at according to foregoing invention content, some nonessential modifications and adaptations are made to the present invention and are embodied,
Protection scope of the present invention should be still fallen within.
In conjunction with Fig. 1 and Fig. 2, the DVC-HEVC video transcoding method based on naive Bayesian, comprising the following steps:
(1) DVC code stream is decoded, obtains rebuilding frame sequence.The non-key frame in frame sequence will be rebuild and carry out frame point
Group is divided into training frames and test frame, and training frame number is arranged herein and is;
(2) HEVC coding is carried out to training frames, while extracts supervision message of its CU division result as model training when,
In conjunction with prediction residual, texture-rich degree and the motion vector obtained in DVC decoding process as input feature vector, 64 are established respectively
× 64,32 × 32,16 × 16 training sets obtain the CU partitioning model of each size;
(3) test frame image is divided into LCU (64 × 64), carries out feature extraction, test set is established, as disaggregated model
Input.First in the disaggregated model of input 64 × 64, if the posterior probability P (C divided downwardsS| x) it is greater than after dividing downwards
Test probability P (CN| x), then (4) are entered step, otherwise, enters step (7);
(4) block that LCU is continued to be divided into 32 × 32, extracts its feature, in the disaggregated model of input 32 × 32, if downwards
Posterior probability P (the C of divisionSX) it is greater than the posterior probability P (C divided downwardsN| x), then (5) are entered step, otherwise, entered step
(7);
(5) block for continuing to be divided into 16 × 16, extracts its feature, in the disaggregated model of input 16 × 16, if dividing downwards
Posterior probability P (CSX) it is greater than the posterior probability P (C divided downwardsN| x), then (6) are entered step, otherwise, enters step (7);
(6) continue to be divided into 8 × 8 CU block;
(7) current CU is sized to final CU macroblock mode, CU divides the selection completed and enter PU mode, continues
Next step cataloged procedure.
Specifically, in the step (1), in conjunction with transcoding process it is found that current HEVC coded frame is after DVC is decoded
The reconstructed frame of WZ frame.Due to being encoded when DVC is encoded to the HEVC that key frame frame uses, it is right not need in transcoding again
Key frame is encoded.
It in the step (2), has been selected herein based on Naive Bayes Classification Algorithm, there are two classes altogether in training pattern
Not, respectively CNAnd CS, judge whether current CU depth continues to divide downwards, wherein CSCurrent CU is represented to continue downwards " division ",
CNRepresent current CU " not dividing ".
The step (3) is based on attribute conditions independence assumption into (5), and posterior probability can be described as follows:
Wherein n is characterized number, xiThe value for being x in ith feature.P (C) is prior probability, by training set
The frequency that Different categories of samples occurs is estimated.P (x | C) it is conditional probability, also referred to as " likelihood ".P (x) is to all categories value phases
Together.Therefore Naive Bayes Classifier can be expressed as follows.
For the validity for proving inventive algorithm, we have carried out experimental verification to it, and result is as illustrated in figures 3-6.Figure
3~6 be the DVC-HEVC video transcoding method of the invention based on Naive Bayes Classification and traditional cascade transcoding algorithm
Rate distortion curve comparing result, compares that detailed process is as follows:
(1) to video sequence carry out DVC encoding and decoding, the HEVC test video of video sequence selection criteria, their title,
Resolution ratio is respectively as follows: BasketballDrill (832 × 480), FourPeople (1280 × 720), and frame per second is 30 frames/second.
Wherein, quantization step (QP) value takes 22,26,30,34 respectively.
(2) it opens simultaneously the program of two methods and sets identical configuration file, the selection of HEVC coding parameter version
HM16.5, quantization step (QP) value take 22,26,30,34 respectively.The present invention will be compared with traditional cascade transcoding algorithm.
Select three kinds of video coding performances herein: Y-PSNR (PSNR), bit rate and scramble time, (wherein PSNR embodied view
The objective video quality of frequency, video encoding time embody the computation complexity of coding) it is compared analysis, for more intuitive table
Now mentioned transcoding accelerates situation herein, is evaluated using following three indexs performance gap:
Δ PSNR=PSNRHerein-PSNRCascade
Wherein, Δ PSNR indicates the difference of method and tradition cascade transcoding algorithm Y-PSNR of the invention, Δ BR table
Show that the percentage of method and tradition cascade transcoding algorithm bitrate difference of the invention, Δ T indicate method and tradition of the invention
Cascade the percentage of transcoding algorithm time difference.
(3) it inputs DVC obtained in 2 identical steps 1 and rebuilds video sequence;
(4) Video coding is carried out to it respectively;
(5) by tradition cascade transcoding algorithm to video sequence in HEVC in the way of under carry out Video coding;
(6) by the method for the present invention to video sequence in HEVC in the way of under carry out Video coding;
(7) two programs export the video sequence after Video coding and respective bit rate, PSNR value and total respectively
Video encoding time, the results are shown in Table 1 for above three index, statistics display the method for the present invention and tradition cascade transcoding side
Method mean change 1.413% in terms of bit rate, averagely reduces 0.046dB, in encoding computational complexity in terms of PSNR
Aspect reduces 44.142%.Experiment results show the method for the present invention compared with traditional cascade transcoding algorithm, in video
Compression ratio (declining degree by bit rate to embody) and video quality (being worth decline degree to embody by PSNR) lose very little
Under the premise of, the computation complexity for largely reducing Video coding (declines degree by the scramble time to embody, such as 1 institute of table
Show).
1 inventive algorithm of table is compared with tradition cascades transcoding algorithm
Claims (3)
1. a kind of DVC-HEVC video transcoding method based on naive Bayesian, it is characterised in that mainly include that following procedure walks
It is rapid:
(1) reconstruction frame sequence is obtained after DVC code stream decoding, the non-key frame in reconstruction sequence is divided into training frames and test
Frame;
(2) HEVC coding is carried out to training frames, while extracts supervision message of its CU division result as model training when, in conjunction with
Prediction residual, texture-rich degree and the motion vector obtained in DVC decoding process as input feature vector, establish 64 × 64 respectively,
32 × 32,16 × 16 training sets obtain the CU partitioning model of each size;
(3) by the disaggregated model of each size in the characteristic information input (2) for testing frame sequence, obtain whether CU continues to draw downwards
The result divided.
2. the DVC-HEVC video transcoding method based on naive Bayesian as described in claim 1, it is characterised in that in step
(1) it does not need to carry out transcoding to key frame again when transit code.
3. the DVC-HEVC video transcoding method based on naive Bayesian as described in claim 1, it is characterised in that in step
(3) described according to prediction residual, texture-rich degree and motion vector as three kinds of characteristic values of input feature vector, establish based on Piao
The optimal classification device model of plain Bayes's classification so that test frame quickly carries out the selection of CU partition mode, and has skipped former calculation
Complicated layer-by-layer rate-distortion optimization process in method is based on attribute conditions independence assumption, and posterior probability can be described as follows:
Wherein n is characterized number, xiThe value for being x in ith feature.P (C) is prior probability, passes through samples all kinds of in training set
The frequency of this appearance is estimated.P (x | C) it is conditional probability, also referred to as " likelihood ".P (x) is identical to all categories value.Cause
This Naive Bayes Classifier can be expressed as follows.
Based on simple bass disaggregated model, the CU partitioning model of each size is obtained, characteristics or every unit information is extracted, utilizes formula
(2) decision is carried out, if the posterior probability P (C divided downwardsS| x) it is greater than the posterior probability P (C divided downwardsN| x), represents and continue
It " divides " downwards, it is on the contrary then represent " not dividing ".
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201811479353.9A CN109743575A (en) | 2018-12-05 | 2018-12-05 | A kind of DVC-HEVC video transcoding method based on naive Bayesian |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201811479353.9A CN109743575A (en) | 2018-12-05 | 2018-12-05 | A kind of DVC-HEVC video transcoding method based on naive Bayesian |
Publications (1)
Publication Number | Publication Date |
---|---|
CN109743575A true CN109743575A (en) | 2019-05-10 |
Family
ID=66358422
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201811479353.9A Pending CN109743575A (en) | 2018-12-05 | 2018-12-05 | A kind of DVC-HEVC video transcoding method based on naive Bayesian |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN109743575A (en) |
Cited By (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN111046766A (en) * | 2019-12-02 | 2020-04-21 | 武汉烽火众智数字技术有限责任公司 | Behavior recognition method and device and computer storage medium |
CN113259664A (en) * | 2021-07-15 | 2021-08-13 | 康达洲际医疗器械有限公司 | Video compression method based on image binary identification |
CN113347415A (en) * | 2020-03-02 | 2021-09-03 | 阿里巴巴集团控股有限公司 | Coding mode determining method and device |
CN115633171A (en) * | 2022-10-08 | 2023-01-20 | 重庆邮电大学 | SHVC-based fast CU decision algorithm |
Citations (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN103546749A (en) * | 2013-10-14 | 2014-01-29 | 上海大学 | Method for optimizing HEVC (high efficiency video coding) residual coding by using residual coefficient distribution features and bayes theorem |
CN106713935A (en) * | 2017-01-09 | 2017-05-24 | 杭州电子科技大学 | Fast method for HEVC (High Efficiency Video Coding) block size partition based on Bayes decision |
US20180295375A1 (en) * | 2017-04-05 | 2018-10-11 | Lyrical Labs Video Compression Technology, LLC | Video processing and encoding |
CN108769696A (en) * | 2018-06-06 | 2018-11-06 | 四川大学 | A kind of DVC-HEVC video transcoding methods based on Fisher discriminates |
-
2018
- 2018-12-05 CN CN201811479353.9A patent/CN109743575A/en active Pending
Patent Citations (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN103546749A (en) * | 2013-10-14 | 2014-01-29 | 上海大学 | Method for optimizing HEVC (high efficiency video coding) residual coding by using residual coefficient distribution features and bayes theorem |
CN106713935A (en) * | 2017-01-09 | 2017-05-24 | 杭州电子科技大学 | Fast method for HEVC (High Efficiency Video Coding) block size partition based on Bayes decision |
US20180295375A1 (en) * | 2017-04-05 | 2018-10-11 | Lyrical Labs Video Compression Technology, LLC | Video processing and encoding |
CN108769696A (en) * | 2018-06-06 | 2018-11-06 | 四川大学 | A kind of DVC-HEVC video transcoding methods based on Fisher discriminates |
Non-Patent Citations (2)
Title |
---|
ANTONIO JESÚS DÍAZ-HONRUBIA等: "Adaptive Fast Quadtree Level Decision Algorithm for H.264 to HEVC Video Transcoding", 《 IEEE TRANSACTIONS ON CIRCUITS AND SYSTEMS FOR VIDEO TECHNOLOGY》 * |
李亚泳: "基于编码单元特征分析的H.264到HEVC快速视频转码算法研究", 《中国优秀硕士学位论文全文数据库》 * |
Cited By (6)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN111046766A (en) * | 2019-12-02 | 2020-04-21 | 武汉烽火众智数字技术有限责任公司 | Behavior recognition method and device and computer storage medium |
CN113347415A (en) * | 2020-03-02 | 2021-09-03 | 阿里巴巴集团控股有限公司 | Coding mode determining method and device |
CN113259664A (en) * | 2021-07-15 | 2021-08-13 | 康达洲际医疗器械有限公司 | Video compression method based on image binary identification |
CN113259664B (en) * | 2021-07-15 | 2021-11-16 | 康达洲际医疗器械有限公司 | Video compression method based on image binary identification |
CN115633171A (en) * | 2022-10-08 | 2023-01-20 | 重庆邮电大学 | SHVC-based fast CU decision algorithm |
CN115633171B (en) * | 2022-10-08 | 2024-01-02 | 重庆邮电大学 | SHVC-based quick CU decision algorithm |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN110087087B (en) | VVC inter-frame coding unit prediction mode early decision and block division early termination method | |
CN109743575A (en) | A kind of DVC-HEVC video transcoding method based on naive Bayesian | |
CN103248893B (en) | From H.264/AVC standard to code-transferring method and transcoder thereof the fast frame of HEVC standard | |
CN111355956B (en) | Deep learning-based rate distortion optimization rapid decision system and method in HEVC intra-frame coding | |
Shen et al. | Ultra fast H. 264/AVC to HEVC transcoder | |
CN108347612B (en) | Monitoring video compression and reconstruction method based on visual attention mechanism | |
CN107018412B (en) | A kind of DVC-HEVC video transcoding method based on key frame coding unit partition mode | |
CN104079937B (en) | It is a kind of based on motion vector analysis by H.264 to the quick interframe code-transferring method and transcoding device of HEVC | |
EP2084911A2 (en) | Apparatus and method of reduced reference frame search in video encoding | |
CN108769696A (en) | A kind of DVC-HEVC video transcoding methods based on Fisher discriminates | |
CN106791828B (en) | Video transcoding method and its transcoder based on machine learning | |
CN105681797A (en) | Prediction residual based DVC-HEVC (Distributed Video Coding-High Efficiency Video Coding) video transcoding method | |
KR100211916B1 (en) | Method for determination of coding type and mode in the object coding based on shape information | |
CN104581170A (en) | Rapid inter-frame transcoding method for reducing video resolution based on HEVC | |
CN110677624B (en) | Monitoring video-oriented foreground and background parallel compression method based on deep learning | |
CN106412611A (en) | Complexity control method of efficient video encoding | |
CN110351552B (en) | Fast coding method in video coding | |
CN106878754B (en) | A kind of 3D video depth image method for choosing frame inner forecast mode | |
Li et al. | Fast CU decision algorithm based on texture complexity and CNN for VVC | |
Shindo et al. | Accuracy improvement of object detection in VVC coded video using YOLO-v7 features | |
CN102595137B (en) | Fast mode judging device and method based on image pixel block row/column pipelining | |
CN100586185C (en) | Mode selection method for transcoding 264 video to reduce resolving capability | |
Xing et al. | Surveillance video coding with quadtree partition based ROI extraction | |
Huang et al. | Beyond VVC: Towards perceptual quality optimized video compression using multi-scale hybrid approaches | |
CN111726636A (en) | HEVC (high efficiency video coding) coding optimization method based on time domain downsampling and frame rate upconversion |
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 | ||
RJ01 | Rejection of invention patent application after publication |
Application publication date: 20190510 |
|
RJ01 | Rejection of invention patent application after publication |