A Novel Virtual View Distortion Estimation for Depth Maps Coding

Abstract

an efficient virtual view distortion estimation algorithm for depth maps coding is proposed in this paper, in which the virtual view distortion is calculated based on the analysis of occlusion relationship change. It is judged whether the depth coding errors alter the occlusion relationship. If true, wrong occluded pixels or wrong un-occluded pixels are further distinguished. The wrong occluded pixels should be unoccluded in the virtual view, but are occluded for the depth map coding error. While the wrong un-occluded pixels are reverse. All the pixels in depth maps are classified into four types: the base pixels which neither are occluded nor alter the occlusion relation, the occluded pixels, the wrong occluded pixels and the wrong unoccluded pixels. The virtual distortion is calculated according to the characteristics of different depth map pixel types. Experimental results show that the proposed virtual distortion estimation algorithm is effective. When the traditional Rate-Distortion Optimization (RDO) in depth map coding is replaced by the estimating algorithm proposed in this paper, the average coding bitrates reduce about 20%. It demonstrates the better subject quality also. Keywords-Three dimensional video (3DV); depth coding; virtual view distortion; Rate-Distortion Optimization (RDO); Multiview plus depth (MVD)