Video Coding Optimization Method Based On Space-time Correlation

HEVC(High Efficiency Video Coding)is a new video coding standard established by JCT-VC in 2013.Compared with the former AVC standard,HEVC guarantees the video quality while reduces the code rate,which is suitable for high-resolution video coding.But at the same time,it also increases the coding complexity,which has a serious impact on its applications.SHVC(HEVC scalability extension)is developed based on HEVC.So it also has the problem of high coding complexity.Because SHVC is composed of basic layer and multiple enhancement layers,so there is another problem of inter layer coding redundancy.In view of the above problems,this paper exploits the space-time correlation characteristics between HEVC and SHVC.And the fast encoding methods of HEVC and SHVC are proposed,which effectively reduce the encoding complexity while ensures the efficiency of video coding.The main contributions in this article include:1.The probability distribution of space-time characteristics of coding unit division and prediction mode selection in HEVC and SHVC is studied,and the feasibility of the fast encoding method based on space-time correlation is analyzed statistically.2.To solve the problem of high complexity of HEVC inter coding,a fast HEVC inter coding method based on decision tree is proposed.The fast inter coding method includes fast CU division method,fast PU mode selection method and fast TU division method.Firstly,the feature sequence is selected and the relevant features of the decision tree are extracted.Then,the decision tree of CU depth division,the decision tree of PU inter mode selection and the decision tree of TU depth division are obtained by off-line training.The training decision tree is used to realize HEVC fast inter coding,including CU early termination division method based on time domain similarity,CU early skipping division method and early termination division method based on spatial domain similarity,PU fast mode selection method based on spatial domain similarity and TU early termination division method based on spatial domain similarity.The simulation results on the HM16.9 software show that under the low delay P mode and random access mode,the average time saving can reach 48.9% and 55.3% respectively.But the impact on the encoding quality is very small.3.In view of the high complexity of SHVC inter coding,a fast SHVC inter coding method is proposed,which combines coding depth prediction and Bayesian decision.It is divided into three parts: depth prediction method based on the correlation between layers and space,CU fast division method based on the minimum Bayesian risk and PU inter mode division method based on themaximum Bayesian posterior probability.Firstly,the coding depth range of enhancement layer is predicted,then CU fast division and PU fast mode selection method are used to improve the coding performance.The simulation results on SHM12.0 show that under low delay mode and random access mode,it can save 64% and 63% coding time for 1.5x space scalable coding respectively,63% and 63% coding time for 2.0x space scalable coding respectively as well as 64% and 64%coding time for quality scalable coding respectively.It is demonstrated that the proposed algorithm is effective.