Video Coding System Based On Texture Analysis And Synthesis

Video coding is the basis and key technology in multimedia communications, which will become an essential part in the communication systems for the next generation. Sponsored by Huawei Science and Technology Foundation"AVS Video Technology Proposal and Research on The Next Generation Video Coding", this thesis conducts an in-depth research on vision-based low bit rate video coding which aims to improve the compression performance while maintaining subject video quality. Meanwhile, our research group takes great effort in China's Digital Audio and Video Coding Standard (AVS) Workgroup to push the drafting of the AVS Standard of China.Over the last one decade, in order to achieve better transmission quality and compression efficiency, hybrid video coding has been successfully improving the video quality while decreasing the bit rate. For most of the state-of-the-art video compression standards, they all employ hybrid video coding structure to exploit spatial and temporal correlation to achieve compression. Currently, minor gain is achieved under the great cost of complexity increase. Nowadays, the visual information surrounding us expands extraordinarily; the cost to enjoy it has to increase because of the limited bandwidth and transmission ability, as well as effect by unstable environment. As a result, based on statistical analysis and human perception characteristic, video coding is still a promising and challenging research area.Our research focus on the relationship between image content and human being's perception, analysis the structure and property of texture, and propose vision-based low bit rate video coding algorithm. Firstly, this thesis propose a texture index algorithm based on the color space, through the conversion between different color space and the statistic information of macroblock based on the characteristic of direction property and color property, it can fix out texture region of video sequences through similarity determination.Secondly, according to the shortage of the above algorithm, we propose a new texture index algorithm which changes the minimal determinate unit size, it not only corresponds to current AVS standard, but also increases the correctness. Then it imports efficient EMD-L1 algorithm to further improve the accuracy of texture region.Thirdly, based on the excellent performance of Mean Shift on clustering and image segmentation, we use it to segment the video frame before coding, through the calculation of pixel gradient, it can find maximum convergent mode pixel and result in pixel clustering. It utilizes the result to fix out texture region.Finally, this thesis analyzes the integrated system, including system implementment, system architecture, texture analysis in encoder and texture synthesis in decoder. Then we propose a syntax modification method which changes little compared to AVS standard. The experiment result shows that the previous algorithm performs well in simple and smoothness texture region, while the later algorithm does well in complex texture region.