Research Of Key Problems On Video Coding And Streaming

With the development of network technology, Internet has transformed our lives and the way we communicate, how we learn, how we work and spend free time, in essence– it has more or less changed every aspect of human society one can think of. Since video is one of the most compelling ways to present information, video steaming is expected to dominate the future network traffic, satisfying people's immediate requirements of study and entertainment. However, video streaming on Internet have some new challenges on video compression and network transmission. First, video compression for streaming should be adaptive on quality of service to satisfy people's different requirements on quality, delay and interaction. Second, video streaming over Internet should provide congestion control, bandwidth adaptation and video smoothing. Third, video compression should support different decoder capacity and different requirements of play back. All these factors require the cooperation of video compression and network scheduling to satisfy different needs of streaming applications. Scalable video coding (such as MPEG-4 FGS and JVT SVC) and streaming is one of the most promising technologies. In this thesis, we investigate the key problem of scalable video coding and streaming: rate-distortion analysis and approximation, with which we also design some simple and effective smooth quality reconstruction algorithms of video streaming. The main contributions of this thesis are as follows:Generalized Gaussian distribution (GGD) is a nice model for the DCT coefficients and wavelet coefficients in theory and practice. With information and quantization theory, we mathematically prove that the derivative of GGD distortion-rate function exists and is not related to the variance of the distribution. So if we gain the relation between the the derivative of GGD distortion-rate function and the GGD shape parameter, then we can reconstruct the rate-distortion function.The bitplane-based FGS of MPEG-4 has been accepted as a standard coding tool for video streaming applications. To support smooth quality reconstruction under multiple rate constraints of transmission, it's important to obtain the actual rate-distortion (R-D) function or distortion-rate function of each enhancement layer (EL) frame. In this thesis, firstly, we introduce generalized Gaussian distributions to model the distributions of DCT coefficients of an EL frame. Then, according to quantization theory, we analyze the distortion-rate function of the generalized Gaussian model, with which we conclude that for the actual FGS coding, the derivative of actual distortion-rate function usually decreases continuously as the rate increases, and then begins to increase slowly at a comparatively high bit rate. Finally, guided by the...