Research On Inter Prediction Algorithm Of High Efficient Video Coding

High Efficient Video Coding(HEVC) has significant improvements compared with the last generation H.264/AVC in compression performance and parallel processing. However, the high performance of HEVC is at the cost of the higher computation complexity which restricts its promotion in practical. Therefore, how to ensure the video quality and reduce coding complexity simultaneously in practical application is a significant research content of HEVC. In this thesis, based on the research of HEVC standard and its key technologies, the improved algorithms is proposed to reduce complexity from block partitioning and motion estimation in inter prediction.Aiming at the problem of the high computation complexity of block patitioning in inter prediction, the improved algorithm is proposed from both macro and micro level. On the macro level, the depth range of the current coding unit(CU) is predicted through the depth information of spatial and temporal neighbors. Meanwhile, the depth difference between temporal and current CUs is adopted to further optimize CU depth. As a result, the computing on redundant recursize depth and computation complexity of CU are reduced. On the micro level, according to the structure similarity of temporal adjacent prediction units(PUs) in the same depth and the mode complexity of the adjacent PUs, candidate modes are predicted and filtered to reduce the traversal complexity of the PUs. Compared to HM14.0, experimental result shows that the proposed algorithm can save 50.61% computational complexity on average with the quality of vedio unchanged.Considering the high computation complexity in searching process of motion estimation, the algorithm is improved from two aspects, uni-prediction and bi-prediction. On the uni-prediction, the algorithm optimizes searching strategies by adopting different step searching for motion classification, the direction of decreasing distortion and the unbalance of object in horizontal and vertical direction for diamond searching. On the biprediction, the proposed algorithm reduces the search range by utilizing the correlation between motion vector prediction and motion vector of uni-prediction. Compared to HM10.0, experimental result shows that the proposed algorithm can save 10.13% computational complexity on average with the quality of vedio remain basically unchanged.In conclusion, the improvements of block partitioning and motion estimation in inter prediction show competitive performance under the guarantee of video quality. With neglectable loss in bit rate, the computation complexity is effectively reduced.