3D video coding with depth modeling modes and view synthesis optimization

This paper presents efficient coding tools for depth data in depth-enhanced video formats. The method is based on the high-efficiency video codec (HEVC). The developed tools include new depth modeling modes (DMMs), in particular using non-rectangular wedgelet and contour block partitions. As the depth data is used for synthesis of new video views, a specific 3D video encoder optimization is used. This view synthesis optimization (VSO) considers the exact local distortion in a synthesized intermediate video portion or image block for the depth map coding. In a fully optimized 3D-HEVC coder, VSO achieves average bit rate savings of 17%, while DMMs gain 6%in BD rate, even though the depth rate only contributes 10% to the overall MVD bit rate.