Research On Fast Prediction Mode Selection Algorithm For HEVC

High-efficiency video coding technology H.265/HEVC(High Efficiency Video Coding)is the state-of-art video coding standard.Compared with the previous generation video coding standard,H.264/AVC(Advanced Video Coding),H.265/HEVC has better compression efficiency.Under the condition of the same distortion,the compression ratio is increased by 50% than that of H.264/AVC.The H.265/HEVC standard uses a series of new advanced technologies,which include CU(Coding Unit)recursive quadtree partitioning,supports smaller size macroblocks,and adds more intra prediction angle modes and advanced inter-frame prediction modes.With these new coding tools,they ensure high compression ratio,but the computational complexity of the encoder increases,resulting in a significant increase in coding time.In order to reduce the coding time,this thesis studies in-depth the prediction mode selection process of the new generation of efficient video coding standards,and proposes a fast prediction mode selection algorithm based on content and an inter-frame prediction unit partition mode fast selection algorithm based on motion characteristics.Inter prediction relies on the content similarity of current frame with reference frame to reduce the prediction residual.The degree of content similarity mostly determines whether the best prediction mode is intra prediction or inter prediction.In this paper,the pixel value statistics are performed on the current prediction block and the reference frame corresponding block.For a CU of size 16 × 16,8 × 8,the size is too small,and the statistical results may not accurately describe the content of the current block,so the CU is linearly interpolated and scaled,and then the statistical results after being scaled are presented as the probability models.The similarity analysis is performed on the two probabilistic models by JS divergence.Finally,the CU block is predicted by JS divergence before entering the intra-frame inter-frame prediction.The modes are roughly selected,and the selection results are used as candidate modes for the intra-frame and inter-frame prediction modes,thereby achieving the effect of reducing the cyclic redundancy of the prediction modes.The experimental results show that the algorithm reduces the coding time by 11.81% on average when the changes of BDPNSR and BDBR are small.In this paper,we start with looping over the inter prediction mode.The motion characterization of the current CU block is described by the motion vector MV(Motion Vector)of the four sub-PUs of the N × N,which are classified into unified motion characteristics and up-down motion characteristics,left-right motion characteristics,dispersion motion characteristics,and undefined motion characteristics,and then the inter prediction modes are classified into unified motion mode,up-down motion mode,left-right motion mode,and dispersion motion mode according to the characteristics of the inter prediction mode division.By relating the motion characteristics with the inter-frame prediction partitioning mode,a fast algorithm based on motion characteristics for inter-frame prediction mode partitioning is proposed.The experimental results show that the algorithm reduces the coding time by 31.2% in the low delay configuration,and reduces the coding time by 27.6% in the random access configuration with the minimal impact on the encoding performance of the encoder,which showed by RD(Rate-Distortion)curve.