Improved multi-view depth estimation for view synthesis in 3D video coding

Abstract

Multi-view depth estimation is an important part in the virtual view synthesis and depth map coding within the 3DV/FTV applications. However, existing depth estimation methods generally have difficulty to generate good depth results around object edges or in areas with less texture. To solve this problem, we propose an improved algorithm to obtain a smooth and accurate depth map for view synthesis in 3D video coding. The proposed method aims to preserve details in the depth map and simultaneously smooth depths in object boundary occluded area, less texture and discontinuity regions, to avoid blurring in the synthesized views. Experimental results show a significant improvement of the initial input depth maps and coding efficiency, as well as a reduction of view synthesis artifacts.