Abstract
In a general multipoint video conferencing, the participants focus on the other participants rather than the background, and they will pay more attention on the active conferees. Therefore, we propose a bit-allocation method which may distribute the constrained bit-rates to each object based on its characteristics. From the rate and distortion model proposed in ITU-T TMN8, we derive an object-based R-D optimized bit allocation equation, which only considers spatial activity. Then we introduce the factors of temporal activity and object size into the bit allocation equation to derive the joint bit-allocation method. Simulation results show that the image quality of interest is improved while the image quality of less interest is degraded. In this thesis, two multipoint virtual conferencing environments are proposed to provide a more realistic conference, the conferencing with two-dimensional virtual scene and the conferencing with three-dimensional virtual environment. In the conferencing with two-dimensional virtual scene, the objects are resized and composed onto a pre-designed virtual scene. In the conferencing with three-dimensional virtual environment, the objects are manipulated onto a three-dimensional environment based on their specific locations. An object-segmentation procedure is also proposed in this thesis to discriminate the objects from the video sequence based on the background subtraction, morphological operations, and row-column scan method.