Depth-based direct mode for multiview video coding

摘要

Multiview video plus depth sequence is considered as an efficient 3D video format for supporting advanced stereoscopic and auto-stereoscopic multiview displays. In order to encode this video format, several modes are commonly employed with rate distortion optimization technique. Specifically, direct mode is an efficient mode to encode homogeneous or stationary regions without encoding any additional motion information. However, in the format, the conventional direct mode causes inaccuracy of motion parameters, i.e., a motion vector and a reference index. Thus, we propose a new method for deriving motion parameters of direct mode, which settles the inaccuracy problem, based on depth information and priority of neighboring blocks determined by their relative position. The proposed direct mode achieves an improvement in coding performance when compared with two state-of-the-art direct modes for multiview video coding. The experimental results show that the proposed method yields average bitrate savings of 4.6%, up to 13.7% in terms of the Bjontegaard Delta metric.