Research On HEVC Inter Prediction Algorithm Based On Fast Motion Search And Spatiotemporal Correlation

The principle of video coding is to remove the redundant information from the video,to represent a large number of original video data through a small number of bits.HEVC is the latest generation of video coding standard,which adopts a variety of new technologies to improve the coding efficiency,but it also causes the high computational complexity inevitably and brings a great challenge for real-time compression of video streams.This thesis focuses on the realization of motion estimation,the flexible CU quad-tree division and the selection of PU prediction modes.There is one common point in the three processes,which is to adopt the traversal search method to find an optimal result with the lowest RDcost.In a complete coding process,these three parts bring a very high computational complexity,taking up most of the coding time.In this thesis,we propose a fast motion search algorithm and a fast inter prediction method based on temporal and spatial correlation.The main work includes the following four aspects:Firstly,it studies the HEVC video coding standard and the contents of each coding module earnestly,especially researches the key technologies involved in the inter prediction process carefully.Secondly,this thesis studies the latest TZSearch search algorithm used in HEVC,analyzes the probability information of its final search point,and proposes a fast TZSearch search algorithm.The proposed algorithm only needs to search for 13 candidate points to get the best matching block,which can greatly reduce the motion search time.Thirdly,a fast decision algorithm for SKIP model is proposed.In this thesis,the characteristics of SKIP coding mode are analyzed in depth,and it is known that the distribution of SKIP model has strong temporal and spatial correlation.On this basis,this thesis use the distribution probability of spatiotemporal neighborhood SKIP model and the mean value of the RDcost to determine the current CU using SKIP mode quickly,saving the time to traverse the other PU mode.Finally,This thesis presents an early termination CU division strategy.SKIP mode is always used in the homogeneous regions,while flat areas are often encoded using large-size CUs.Based on this point,this article sets a threshold by the distribution probability and the mean RDcost value of the neighborhood SKIP mode,and proposes an early termination CU division algorithm,avoiding a series of unnecessary computational processes in the next layer of CU depth.In summary,this thesis has improved the three aspects of motion search,PU model rapid decision and CU depth division.Compared with HM16.0 official coding model,the three algorithms proposed in this thesis have superior performance,which can greatly reduce the coding time under the premise of ensuring coding quality.