Research And Implementation On HEVC Frame Processing Based On Spatio-temporal Correlation

With the development of high definition videos and ultra high definition videos,the video coding standard also is evolving.HEVC standard was announced in 2013.Compared with H.264/AVC,HEVC standard can save about half bitrate,and the quality of HEVC is as same as H.264/AVC.However,it is an urgent problem that how to speed up the encoder.The research on HEVC standard includes two main aspects: intra prediction and inter prediction.In this paper,the partition of intra code blocks and the fast selection of multiple reference frames are studied.The main contents and innovation points are as follows:?1?In this paper,HEVC coding framework,new coding technologies and coding block partitioning process are introduced.A fast algorithm of coding block partition has been studied.A new fast algorithm based on spatio-temporal correlation is proposed.Computational complexity of HEVC is improved because of the ratio distortion optimization.The depth of spatial CTU and temporal collocated CTUs determine whether current coding blocks skip the calculation of CU with zero depth.Then this paper uses the depth of collocated CTU to predict the depth range of current coding block.At last,the ratio of rate distortion value between collocated CUs,the ratio of standard deviation between collocated CUs determine whether the current CU splits or not.Experiment result shows that,compared with HM16.7,video bitrates increase 1.4% on average,encoding time saves about 40% with negligible video quality loss.The proposed algorithm achieves good results.?2?This paper introduces the muluti reference frame management technique of HEVC.The multi reference frame selection based on spatio-temporal correlation also has been studied.Firstly,The order of inter prediction modes was changed.PART₂N×2N is executed immediately after merge mode.Secondly,The best reference frames of PART_N×2N is used to build up the candidate reference frame list of PART₂N×2N.Thirdly,The reference frames from frame zero to the maximum of best reference frames of PART_N×2N are added to the candidates of PART₂N×N.Finally,frame zero,the best reference frame of temporal collocated PU,spatial neighbouring PU and corresponding PU between different prediction modes are added to the reference frame list.This list is sorted by probability in desending order.AMP mode just use first two frames as the reference frame list.Experiment results shows that this proposed algorithm can save 18% encoding time in lowdelay configuration and video quality is not changed obviously.This algorithm improves the real-time video compression.