Fast Algorithm Research Of HEVC Based On Decision Trees And System Implementation

With the popularization of high definition and ultra definition video recently,the video coding and transmission of these type of video content is more and more important.However,the coding efficiency of traditional video coding standard H.264/AVC for high resolution video cannot meet the basic requirements.Therefore,in order to significantly improve the coding efficiency,the group of JCT-VC is established by Video Coding Experts Group of ITU-T and Moving Picture Experts Group of ISO/IEC,who jointly proposed the newest generation of video coding standard,which called High Efficiency Video Coding(HEVC).Compared with H.264,the improved coding technology of HEVC is applied to the various modules of coding framework,more importantly,the coding efficiency of HEVC is increased greatly by using some new coding tools.With the adoption of these tools,meanwhile,the encoding time complexity is also increased,which seriously hinders the promotion and application of HEVC coding standard.Therefore,the optimization method for intra coding and inter coding of HEVC is studied in my article,and also realize the build of demo platform for real-time coding and transmission of the multi-view videos via the x265 library,which meets video coding standard of HEVC,then the main works of this article are concluded as following:(1)Because of the great time complexity of quad-tree partitioning algorithm for coding unit based on rate distortion optimization in HEVC,a novel decision tree prediction model is proposed,which is trained by the specific feature data and offline.Then the trained model is applied in the video coding standard HEVC,which is used for the optimization of the partition of intra-frame coding unit(CU).As a result,the time complexity of intra-frame coding unit is decreased efficiently by introducing the decision tree prediction model;(2)In the process of inter-frame coding,the current coding units of HEVC have to traverse various prediction modes of prediction units.While for inter-frame coding,ten different partition types are defined by HEVC,including Skip/Merge,2N×2N,N×N,N×2N,2N×N et al.The feature selection method,namely F-Score ranking,is used to optimize the feature set,based on the statistical results,then an optimal decision tree model is trained via these feature subsets,on which based,a fast decision algorithm is proposed for prediction mode;(3)The build of demo platform for real-time coding and transmission of the multi-view videos is realized based on MFC frame which consists of server and receiver.The server is used for real-time video coding via x265 library,where FFmpeg is used for code framework,multithreading technology is applied to multi-view coding,and the stream is transferred via local area network(LAN)by adopting Real-time Transport Protocol(RTP);In the receiver,the received network packets are decoded into video data that is transferred to displaying thread to realize the real-time display of multi-channel video stream.