Shape Preserving RGB-D Depth Map Restoration

The RGB-D cameras have enjoined a great popularity these years. However, the quality of the depth maps obtained by such cameras is far from perfect. In this paper, we propose a framework for shape preserving depth map restoration for RGB-D cameras. The quality of the depth map is improved from three aspects: 1) the proposed region adaptive bilateral filter (RA-BF) smooths the depth noise across the depth map adaptively, 2) by associating the color information with the depth information, incorrect depth values are adjusted properly, 3) a selective joint bilateral filter (SJBF) is proposed to successfully fill in the holes caused by low quality depth sensing. Encouraging performance is obtained through our experiments.