Boundary Recovery of Depth Map for Synthesis View Optimization in 3D Video

Depth map generally includes structural errors and blurs near boundaries due to the limitation of depth acquisition technique. These distortions severely degrade the quality of synthesis views in 3D video systems. In this paper, we propose a method to accurately recover the boundaries of the depth map with the guidance of its associated texture image. First, we detect the structural error regions of the depth map by considering its structural consistency with the texture image. Then, we incorporate the information of the detected structural error regions into the weight term of the general weighted median filter for depth map recovery. Experiments are conducted on a set of depth maps. The results show that the boundaries of the depth maps with the proposed method are more accurate and clear than that of the other methods. Consequently, it well improves the subjective quality of the synthesis views.