Research On Optimization Algorithm For Depth Video Intra Coding In 3D-HEVC

In recent years,with the rapid development of multimedia technologies,the stereo video is gradually becoming a new research focus in the video field.Compared with 2D video,the sence of stereo video is more real and vivid.However,the huge data of stereo video also brings big difficulty to data storage and transmission.For this,the Joint Collaborative Team on 3D Video Coding Extension Development has established a new generation of stereo video coding standard,which is called 3D-HEVC.Although 3D-HEVC achieves higher compression rate,its application is limited due to the great increase in coding complexity.Therefore,the research on how to effectively reduce the coding complexity of 3D-HEVC is quite meaningful.In this thesis,the HEVC standard is briefly introduced at the first stage,followed by a series of key technologies of 3D-HEVC.After that,two algorithms are presented to reduce the computational complexity of depth map intra coding.To reduce the computational complexity of Coding Unit(CU)splitting process,a fast CU size decision algorithm for depth map intra coding is proposed.Firstly,the depth range of current Largest Coding Unit(LCU)is predicted by the smallest depths and maximal depths of spatially adjacent LCUs;Secondly,the gradient matrix of current CU is calculated by the texture analysis algorithm based on the concept of texture primitive,then the CU splitting process is early terminated when the sum of gradient values in gradient matrix is smaller than a given threshold.Experimental results show that,compared with HTM-13.0,the proposed algorithm achieves an average depth map coding time reduction of 45.55% with a small bitrate lost of 0.63% for synthesized views under all intra scenario.Moreover,a fast intra prediction mode selection algorithm for depth map intra coding is also proposed.Firstly,the texture features and the smallest low computational rate distortion cost of Planar and DC are used to skip the search of angular modes in RMD process for Prediction Units(PUs)of large size and PUs of small size,respectively;Secondly,the number of intra preduction modes in candidate list is reduced according to the correlation between the best mode and first mode in candidate list;Finally,the current PU is divided into two areas by a strategy,and then the mean of pixs in each area is calculated.If the absolute value of the difference between two means is smaller than a given threshold,DMMs is skipped.Experimental results have shown that,compared with HTM-13.0,the proposed algorithm can reduce 39.20% of depth map encoding time with negligible distortion for the synthesized views.In the end,the research content of this thesis is concluded,and the following research is also forecasted.