Optimizations For HEVC Inter Codec

The latest video coding standard,High Efficiency Video Coding(HEVC),developed by ITU-T and ISO/IEC,achieves the significant compression improvement over H.264/Advanced Video Coding(AVC)with the competitive objective quality.It mainly stems from coding optimization technology,including CU/PU/TU hierarchical quad-tree coding scheme,sample adaptive offset(SAO),adaptive loop filter(ALF),and so on.However,the complex coding structure increases computational complexity intensively for the encoder.Researchers now optimize the coding complexity mainly aimed at software coding,while in consideration of the hardwire encoder,these algorithms have drawbacks in two aspects: Firstly,the CTU-grain maximum encoding complexity is not decreased;Secondly,the CU level data dependency incurs the cumbersome to encoding parallelism.This paper is devoted to relieving the coding complexity for HEVC inter mode.The hierarchical CU/PU/TU composite structure is the most computational module in inter coding.So our proposed algorithm execute the optimization in CU and PU level:first,fast fractional motion estimation(FME)algorithm using IME RD-Costs based error surface fitting scheme;second,CNN based CU partition mode decision algorithm.The contributions of our proposals include:· We analyse the error distribution feature in FME process by using statistics data,and fit the error surface with proper function.Then use this model to obtain the best 1/4-pixel motion vector(MV)directly.this method could take a trade-off between the computational complexity and the prediction accuracy.· The relationship between image texture and coding mode is discovered;CNN is introduced to classify CUs into different coding modes,which reduces the maximum number of CU candidate mode.Our algorithm speeds up the HEVC inter coding by 45.0%,whereas the incurred compression loss is BDBR=+2.91%,or equivalently BDPSNR=-0.1114 dB quality loss.· With the CTU pipeline architecture,the parallelism of the RDO processing will not be deteriorated by our fast algorithm.Our algrithm provide better performance and broader application than others.