Depth boundary filtering for view synthesis in 3D video

This paper presents a boundary sharpening method for depth maps to improve synthesis view quality. In general, coded depth maps exhibit noise and artifacts around object boundaries, leading to ineffective view synthesis. In our approach, gradient information is used to extract depth boundary regions. Afterward, filtering based on distance, similarity, and direction is performed on such regions to replace depth values. The proposed algorithm was implemented on 3DV-ATM v0.3 as post-processing to coded depth maps. Experimental results showed 5.23% compared to the anchor results of 3DV-ATM v0.3. Subjective quality was improved as well.