Reduced resolution update for depth map compression on the basis of texture and predicted image

Shiori Sugimoto, Shinya Shimizu, Hideaki Kimata, Akira Kojima

Abstract: In 3D video compression, depth map is normally compressed by resolution reduction. To avoid artifacts in views synthesized with the reduced resolution depth map, up-sampling approaches based on special properties of depth map have been proposed. Most previous studies do not apply rate-distortion control and the quality of depth map and synthesized views is not always guaranteed because their re-sampling processes are out of the coding loop. This paper proposes a novel inloop reduced resolution depth coding based on the scheme of a reduced resolution update mode. This is achieved by encoding prediction residuals at reduced resolution, which was obtained by performing prediction using full resolution references. We also propose an effective method for reduction of resolution for residuals and reconstruction of depth map based on Texture and predicted depth map. Experiments showed that it is possible to achieve bitrate reductions of up to 2.09% and about 1.10% on average for seven sequences.