Research On HEVC Fast Inter Prediction Algorithm Based On Analyses Of Coding Unit Features

With the rapid development of information technology,digital video is moving towards a larger and clearer goal,which leads to people's more urgent demand for spiritual entertainment.However,limited capacity and bandwidth put forward higher requirements for video storage and transmission.Facing the severe test,the Joint Collaboration Team on Video Coding?JCT-VC?launched a new generation of High Efficiency Video Coding?HEVC?in 2013.Compared with the previous generation of H.264 /AVC standard,HEVC has improved the compression efficiency by 50%,but the application of various cuttingedge technologies has doubled the complexity of coding,restricting the promotion and development of HEVC standard in mobile terminal devices.Therefore,how to improve the time efficiency of coding and ensure a small quality loss has been the focus of attention.As the most time-consuming part of the coding process,inter prediction takes up about 70% of the time,so improving the inter prediction process has become a hot research field.On this basis,this paper proposes a fast scheme for CU depth partition and PU mode decision with the highest redundancy in the inter prediction process.In view of the fast partition scheme of CU depth,this paper firstly introduces the HEVC depth partition process based on quadtree in detail,and carries out statistics and analysis according to the part with high redundancy,and proposes a fast inter-frame CU depth selection algorithm based on multi-feature of time domain.This algorithm divides the depth into two categories: large size depth?depth 0 and depth 1?and small size depth?depth 2 and depth 3?.For large-size depth,the mean depth information of time-domain CU is used to predict the optimal depth of current CU,and the adaptive thresholds to terminate and skip the current depth are set respectively.For the small size depth,the complexity characteristics of the current CU texture,edge and structure were extracted,and the decision was made by the decision tree model which was off-line trained in advance.Experiments show that the fast CU depth partitioning algorithm can save 38% of time and 1.16% of BDBR under LowDelay_P configuration.For the quick decision scheme of PU mode,this paper proposes Skip mode and Merge/Inter₂N×2N modes respectively based on the decision process of PU mode.For Skip mode,we combine the texture and motion characteristics of PU block itself,and decide in advance according to the relationship between this mode and adjacent depth.For Merge mode and Inter₂N×2N mode,we use the consistency of the four sub-blocks to distinguish these two modes from the rest of the modes,and then choose the one with the lowest rate distortion cost.Experiments show that the algorithm can reduce the coding time by 39.8% and the BDBR loss by 1.01% under LowDelay_P configuration.Finally,this paper integrated the algorithms of CU and PU,and obtained 52.33% acceleration ratio and 1.84% BDBR loss under LowDelay_P configuration.The algorithm proposed in this paper can give a layered solution based on the information of CU depth and the structural characteristics of PU mode.Combining with the characteristics of the image and the classification rules,the classification model is used to make decisions,and good results are achieved.Therefore,it provides an effective reference for the future study of HEVC interframe fast algorithm.