Research On Low Complexity Coding For 3D-HEVC

As the main carrier of information presentation and dissemination,video has become an important engine to satisfy people's needs for better lives and promote the development of the information technology industry,economic,and social innovation.Among them,three-dimensional(3D)video can provide a highly immersive and realistic stereoscopic visual experience,playing an important role in the film industry,home entertainment industry,medicine,education,and other fields,as well as showing broad application prospects.In order to meet people's needs for visual comfort and freedom,the multi-view video plus depth(MVD)format came into being.It introduces the multi-view depth map based on the multi-view texture map,so that to synthesize new virtual viewpoints between original ones to provide multi-view video sequences with a wider range of viewpoints.In February 2015,the Joint Collaborative Team on 3D Video Coding Extension Development(JCT-3V)issued the 3D extension of High Efficiency Video Coding(3D-HEVC)standard,aimed at increasing the coding efficiency for MVD format.However,while 3D-HEVC guarantees the performance of the encoder and the quality of the synthesized virtual viewpoint,the computational complexity of the encoder increases sharply due to the introduction of many advanced encoding technologies,which seriously hinders the application and promotion of the 3D-HEVC standard.Therefore,how to adopt effective methods to perform high-quality and efficient compression encoding on video is a new challenge faced by the current 3D video processing technology.This paper takes the 3D-HEVC standard as the research object and HTM as the research platform,with the goal of accelerating intra and inter coding processes,the most time-consuming part,in 3D-HEVC,and strives for new breakthroughs on the theoretical research and application of 3D video lowcomplexity coding.The main research contents and results of this paper are as follows:1.A fast decision of intra prediction unit(PU)mode of depth map based on edge detection is proposed.To simplify candidate mode set reasonably,this paper designs a simplified strategy of rough pattern decision candidate pattern set based on edge detection based on the correlation between the optimal mode of the current intra prediction unit and its texture edge direction and texture complexity.The fast algorithm designed and implemented based on this strategy is compared with the 3D-HEVC standard test model HTM-16.0,which can accelerate the intra coding process of the depth map while ensuring the quality of the depth map and the synthesized view.2.A low-complexity intra coding algorithm for depth maps based on support vector machine(SVM)is proposed.In order to further reduce the complexity of depth map intra coding,this paper analyzes the correlation in CU depth and PU optimal mode decision between depth map texture complexity,texture feature attributes then design the CU depth fast decision and DMMs fast decision mechanism.Compared with the 3D-HEVC standard test model HTM-16.0,the fast intra coding algorithm implemented based on the two strategy designs can reduce the complexity of intra coding on the premise of ensuring the quality of depth maps and synthetic viewpoints reconstruction.In addition,the fusion of CU depth fast selection and PU mode fast decision for intra coding is realized,which achieves more coding complexity reduction and verifies the universality and accuracy of the proposed algorithm.3.A fast CU depth decision algorithm based on e Xtreme Gradient Boosting(XGBoost)for 3D-HEVC dependent view inter coding is proposed.To decrease the computational complexity in 3D-HEVC inter CU depth traversal,the correlation in CU depth decision between texture complexity,viewpoint and view CU depth selection is explored.Further,this paper builds a fast decision mechanism for the depth range of inter CU based on data mining and machine learning methods.Compared with HTM-16.0,this fast inter coding algorithm can f achieves more coding complexity reduction while maintaining the quality of depth map,texture and synthetic viewpoint reconstruction.