Warping-driven mode selection for depth error concealment

Xue Zhang, Yao Zhao, Chunyu Lin, Huihui Bai, Chao Yao, Anhong Wang

Abstract: This paper deals with the recovery of corrupted depth maps in loss-prone networks. Different from color maps, depth maps are not directly used for display but served for view synthesis process. Therefore, the conventional concealment methods which focus on reducing the distortion of reconstructed color maps, are not suitable for the corrupted depth maps. In this paper, a novel mode selection method for depth error concealment is proposed to efficiently reconstruct the lost depth blocks. To this end, three modes, i.e., INTRA, INTER, and MV sharing, are applied as candidate modes to achieve the concealed depth blocks. In order to provide acceptable synthesized views, the optimal mode selection is based on a new distortion function, in which warping process is taken into account. Experimental results show that the proposed method outperforms the conventional schemes up to 0.65 dB on the synthesized views, at block loss rate up to 20%. Most important of all, the subjective quality of synthesized views are much better due to the consideration of warping process in the mode selection.