Depth No-Synthesis-Error Time-Consistent Preprocessing for Depth Coding in FVV System

Abstract Multi-view view plus depth (MVD) is an effective format for 3D scene representation, where view synthesis based on depth-image-based rendering is employed to generate virtual views in free viewpoint video (FVV) systems. MVD signals must be effectively compressed before transmission. However, weak temporal correlation of depth map sequence, caused by inaccuracy depth estimation and capturing, seriously decreases its compression efficiency. Therefore, depth map sequence should be processed before encoding. At the same time, distortions in depth information may introduce geometry changes or occlusion variations in the synthesized view. Based on the above analyses, this paper proposes a depth no-synthesis-error time-consistent preprocessing algorithm to improve depth map sequences compression ratio.. Experimental results show that proposed algorithm can save the bit rate ranging from 14.17% to 74.55%, while the synthesized virtual view by the proposed algorithm has the same quality with the synthesized virtual view with the original depth map.