New Depth Coding Approaches For 3D Video Compression

The 3D video provides us a novel way to show a natural scene of the real world from the recorded multimedia signals. Therefore, it has been becoming so popular in the multimedia signal processing. The most used compression method for the 3D video is the multi-view plus depth(MVD) based coding approach in which both the texture information and the depth information are processed and it is so necessary to compress the texture information and depth information independently. Since the depth image is composed of almost smooth region and sharp boundary, which contains more redundancy and important information concentrated in the boundary region. Based on these features of the depth image, a new image coding algorithm is proposed in the article, aiming at efficiently compressing depth information and reducing the information redundancy of depth map. Meanwhile, the depth image which is processed by our algorithm will be used for the virtual image rendering, further testing the effect of new depth coding algorithm in virtual image rendering. Depth image based rendering(DIBR) algorithm and our new hole filling algorithm will be applied to the virtual image rendering in 3D video. To achieve this goal, we try to improve the coding efficiency for the MVD-based 3D video coding by using our proposed several different methods. Firstly, we propose two novel transform domain down-ward conversion(TDDC) methods, i.e., the smart padding based method and the smart quantization based method, to help compress the depth map for the 3D video, respectively. Both methods may work competitively with the traditional compression method via a rate-distortion optimization(RDO) principle to get a better coding performance. Secondly, we optimize our proposed TDDC methods via an interpolation-directed optimization way and some better reconstruction results have been obtained for the smart padding based method after such an optimization. Finally, we design a novel hole-filling method based on the multi-direction interpolation to improve the quality of the reconstructed view image. Simulation results show that some significant quality gains have been achieved by using our proposed methods in the 3D video coding.