Studies On New Video Coding Techniques

Publication of CCITT Recommendation H.261 in the end of 1990 was the milestone that the video coding techniques were put into real use. Recently further improvement of the video coding techniques in coding efficiency, complexity (especially low power decoding), error resilience and scalability is required for mobile multimedia applications. One of the trends is to develop in depth. That is to develop the waveform coding technique by applying Shannon information theory and signal processing principles so as to improve the performance of prediction coding, transform coding and entropy coding. The other trend is to develop in extension. That is to develop new techniques including structure coding technique and analysis-synthetic coding technique by applying complex information theory and also combining the video coding techniques with the techniques of artificial intelligence and computer graphics. The said complex information theory consists of Shannon information theory, comprehensive information theory and human vision system theory. MPEG-4 and JVT/H.26L, which are the new video coding techniques in these two trends respectively, are studied in this thesis. Achievements are listed below.1. Based on the user-assisted video object segmentation tool of the Electronics and Telecommunications Research Institute (ETRI) Korea, a video object tracking approach is proposed to segment video objects by contour localization of multi-objective optimization. Experimental results on several head-and-shoulders video test sequences show this approach improves the accuracy of motion estimation and compensation for object contour pixels and also enhances the adaptability to complicated motions including translation, scaling, rotation and distortion. The segmentation results of this approach are better to a little extent than that of the ETRI's tool.2. To further enhance the adaptability to video scenes containing complicated contents and motions, especially to deal with the occlusion problem and the aperture problem, a more effective video object segmentation approach is proposed. In this approach the above process is combined with template matching so as to refine...