Efficient depth map compression for view rendering in 3D video

Abstract

A depth map represents three-dimensional (3D) scene information, which is used to synthesise virtual views in 3D video coding. Since the quality of synthesised virtual view highly depends on the quality of depth map, efficient depth coding is crucial for the 3D video system. However, depth map is not directly used for display but to generate virtual views. Compressing depth map using existing video coding techniques yields unacceptable distortions in rendered virtual view. Thus, the depth map needs to be compressed in a way that can minimise distortions in the rendered views. To solve this problem, we propose an efficient depth map compression method for view rendering based on view rendering distortion other than depth map distortion itself. Firstly, we derive relationship between distortion in coded depth map and the rendered view. Then, a region-based video characteristic distortion model is proposed to precisely estimate distortion in view synthesis. Finally, a multilateral filtering is applied as an in-loop filter to reduce rendering artefacts. Experimental results show that the proposed method achieves 2 dB coding gain in terms of PSNR and subjective quality improvement of synthesised views.