Improved depth maps coding efficiency of 3D videos

Abstract

Immersive 3D video services demand the transmission of the viewpoints' depth map together with the texture multiview video to allow arbitrary reconstruction of intermediate viewpoints required for free-view navigation and 3D depth perception. The Multi-view Video Coding (MVC) standard is generally used to encode these auxiliary depth maps and since their estimation process is highly computational intensive, the coding time increases. This paper proposes a technique that exploits the multi-view geometry together with the depth map itself to calculate more accurate initial compensation vectors for the Macro-blocks' estimation. Starting from a more accurate position allows for a smaller search area, reducing the computations required during depth map MVC. Furthermore, the SKIP mode is extended to predict also the disparity vectors from the neighborhood encoded vectors, to omit some of them from transmission. Results demonstrate that these modifications provide an average computational reduction of up-to 87% with a bitrate saving of about 8.3% while encoding an inter-view predicted viewpoint from a depth map multi-view video.