Interactive Performance Optimization for Multiview Coding based 3D Video

An increasing demand is currently witnessed for Three-Dimensional (3D) video since it provides more real perception, interactivity and novel visual enjoyment for multimedia applications. The emerging interactive 3D video application not only brings 3D depth impression of a scene, but also allows users to have an interactive selection of the viewpoint and direction within certain ranges. However, multiview videos captured simultaneously from multiple cameras at slightly different views or angles are required for representation of genuine 3D world video content and thus a tremendously huge amount of storage space, transmission bandwidth and computational resources for compression are needed as compared with traditional mono-view videos. To compress the large volume of multiview video data efficiently, Multiview Video Coding (MVC) has been developed based on the state-of-the-art H.264/AVC (Advanced Video Coding) standard. However, due to the complex coding structure and the multi-reference prediction techniques, the improvement of coding efficiency is at the cost of degradation of interactive performances, including view scalability, temporal/view random access, computational complexity, delay and buffer, which are quite essential to support interactive view switching functionalities and for 2D/3D display device adaptation. In this project, several innovative techniques, including interactive MVC prediction structure, fast multi-reference prediction and MVC rate control schemes, are proposed to improve the interactive performances. This will enable MVC more applicable and adaptable to the interactive video markets such as interactive 3DTV, free viewpoint TV, live 3D broadcasting, immersive teleconference, etc.