View Synthesis And Video Coding For 3D Video

Recently, Multi-View-plus-Depth(MVD)-based 3D video format has emerged to replace the Stereoscopic-based 3D video format, and it has been applied in the new generation of 3D video system. The MVD-based 3D video system uses the given texture and depth data to reconstruct many new viewpoints by Depth-Image-Based-Rendering(DIBR) technique. In this thesis, we focus on the research related on MVD video coding and view synthesis in the MVD-based 3D video system. It is worth noticing that, how to improve the quality of the synthesized video in the process of view synthesis and video coding is the key of this thesis. The main contributions and innovative research results include:· A depth enhancement method for minimizing the hole effect was proposed. First of all, this thesis analyzed the process of view synthesis, and concluded some reasons which caused in the holes of the reconstructed image. Then two holes of categories are classified. For the holes with discrete distribution, an adaptive filter was used to enhance the depth map for minimizing the number of holes in the reconstructed image. For the disocclusion regions, we proposed a directional hole-filling method. Experimental results proved that the proposed method could achieve some gains on subjective visual quality.· A view synthesis enhancement algorithm combined with temporal information was proposed. We proposed to fill the discocclusions in the synthesized view by using the temporal correlation information. To select the information, the motion mode of the foreground objects was taken into account. While for the possible complex movement, we proposed the Foreground-Depth-Correlation to recover the back-ground information which were occluded by the moving objects. The proposed approach further improved the quality of the reconstructed video in both of subjec-tive and objective evaluations.· A depth sampling and Rate-Distortion optimization scheme were proposed. In the texture-depth resolution asymmetric coding framework, our goal was to minimize the view-synthesis-distortion. Then we designed a depth sampling optimization scheme. For the Rate-Distortion optimization in the 3D video coding system, we not only considered the coding-caused distortion, also the depth-sampling-caused distortion was included. Finally, experimental results proved the validity of pro-posed scheme by comparing with 3D video coding standardization.