| Recent advances in 3D content acquisition and display technologies have brought significant market potential to multiview video systems. Many of the first 3D video systems are likely to use the Multiview Video Coding (MVC) extension of the H.264/AVC standard as the compression format. While MVC exploits the correlation between views for improved compression efficiency, it provides limited flexibility for free viewpoint generation in the decoding and rendering process. The multiview video plus depth (MVD) representation which is formed by a multiview texture video and its corresponding per-pixel depth map is one of the most prominent techniques to enable flexible view synthesis in the rendering end, which would bring many 3D applications into practice such as free viewpoint video, autostereoscopic video and advanced stereo display. However, per-pixel depth map sequences require extra bandwith or storage space that MVD data must be efficiently compressed before it is transmitted or stored.A joint coding framework for the multiview texuture video and depth map in MVD data representation is proposed in this dissertation to replace the conventional compression scheme of coding texture video and depth map seperately with MVC codec. The muliview texture video and its corresponding depth map share a very similar pattern of edges and silhouettes which lead to the very similar motion fields and disparity fields. In this dissertation, such redundancy is eliminated by utilizing the inter-layer motion prediction mechanism in the Scalable Video Coding (SVC) extension of H.264/AVC under a multiview video coding scenario.Besides the proposed joint coding framework, a real codec is also implemented for the MVD data representation. Base on JSVM, the reference software of SVC standard, we not only add the support for multiview video coding, but also remove some useless modules that hardly affect the coding efficiency of multiview depth map. Compared with the conventional MVC based coding scheme, our codec has many advances such as high coding efficiency, low computational complexity, excellent backward compatibility and strong expandability. It can be utilized all by itself to compress MVD data, or integrated with other 3D compression techniques, which would improve the coding performance even more.Great coding gain can be achieved by our joint coding scheme for the MVD representation which is also referred to as Joint Multiview Video plus Depth Coding (JMVDC). The performance tests of our JMVDC codec show that, under the same bitrate, our best performance exceed the conventional coding scheme by 1.34dB. |
PDF Full Text Request