Parallelized Contour Based Depth Map Coding in DIBR

Depth map is a critical factor in Depth-Image-Based-Rendering system. Conventionally the depth map is encoded with the block-based method, such as H.264/AVC or some MVC methods. The traditional coding strategy cannot guarantee the quality of the boundary of the objects in the depth map. The distortion on boundary will cause unrecoverable distortions in the synthesized view. To guarantee the precision of the object boundary, we proposed a Contour Based Depth map Coding (CBDC) method. In the proposal, depth maps are divided into several layers in the depth dimension and further segmented into regions. Each region is regarded to be made up of contours and interiors. We adopted a lossless vectorized method to represent the contour and a lossy dynamical modeling method to represent the interior. This paper is based on our previous work, which is a preliminary framework of the CBDC. In this paper, two further contributions are illustrated. One is a data structure improvement of contour bypassing. Redundant contour segments are bypassed under our proposed rules. Experimental result shows that the bypassing strategy reduces 40% to 60% bit cost for the CBDC. Compared with other contour based coding strategy, the improved CBDC achieves higher quality of the synthesized view and better coding efficiency. Compared with block-based method, CBDC achieves much higher image quality (3–15 dB) in high bitrate scenarios. The other contribution is about the parallel process design of CBDC. The layer-region structure is with relatively low data dependency. Taking that advantage, we implemented and evaluated the parallel structure of CBDC. Experimental results show that the parallelized process speeds up by as much as 1.8 to 2.5 times when 2 to 4 process units are employed. Up to 90% parallel efficiency is achieved when 2 process units is employed.