Effective multiview video coding using a scalable depth map

Abstract

In multiview video coding, the consideration to send a depth map or not is an important issue. If viewsynthesis is operated in the decoder part, a depth map of reasonable resolution is needed. Otherwise, the transmission of a depth map is overhead. Scalability of the depth map is a solution to meet this trade-off. In this paper, depth-scalability is adopted to multireference MVC. In multi-referencing, the warped view is added as a referencing picture in the current referencing scheme. The proposed scheme results in improved bitrate efficiency in lower layer depth coding in MVC. It is also shown that in the focus of inter-view synthesis in the decoder part, the bitrates are decreased by about 100-500 kbps in the 2nd and 1st layers, and the PSNR is decreased by about 1-3 dB. Therefore, application adaptive depth map scalable coding is an effective way to obtain coding gain.