Research On Fast Mode Decision For Depth Video Coding

Three-dimensional(3D) stereoscopic video can give the audience more sense of the real stereo environment, and received widespread attention among film and TV industry. Multi-view video plus depth(MVD) is one of the main 3D formats. It usually consists of a series of texture videos as well as the corresponding depth videos, so that virtual views, located between coded views, can be generated by using depth image based rendering(DIBR). MVD videos should not only compress the texture videos, but also the depth videos. Traditional H.264/AVC and the newest high efficiency video coding(HEVC) have proposed effective compress methods, but also bring huge coding complexity. In order to solve this problem, a number of researchers have proposed fast mode decision methods. Considering that the depth video and texture video have different characteristics, this thesis makes a research on mode decision process in depth video coding, and emphatically explores the influence of inter-view correlation and inter-component correlation on fast mode decision process.Each frame is divided into a lot of 16 × 16 macro blocks in traditional H.264/AVC. The motion estimation and other coding processes are done in macro blocks one by one. For motion estimation, H.264/AVC gives a variety of inter frame and intra frame prediction modes, and the best mode is selected by comparing the rate distortion cost after the traversal of all modes. The redundancy in the mode decision is the concerning part in this thesis. At the same time, considering the inter-view correlation and inter-component correlation in MVD, an early terminated mode decision method is proposed on the JMVC platform, taking both of the correlations into consideration. Experimental results show that, the proposed method can reduce the coding complexity with almost the same PSNR and bitrate.3D-HEVC is an emerging standard which supports MVD, and is also an extension of the HEVC. It adopts the same coding scheme as HEVC. Every encoding block(LCU) is divided into coding tree units(CTU), each CU conducts the prediction process. The mode decision is also very complicated in the prediction process. This thesis makes a research on the newly added MERGE mode in HEVC on the HTM platform, and explores the influence of inter-view and inter-component correlation on early terminated MERGE mode decision. Experimental results show that the proposed method can reduce the coding time without significant degradation in PSNR.