Research On Content-based Inter Prediction Techniques

With the rapid development of information technology, the demand for multimedia communication is growing day by day. Especially the multimedia applied in wireless mobile condition is becoming the key componet of the third and fourth generation mobile communication system. As the inevitable part of the multimedia technology, video coding plays a vital role in the development of the multimedia communication. Co-sponsored by the National Science Fund of China——"Research on General Rate-Distortion Model on Wireless Channel"(Grant No.: 60372067), cooperated project with HuaWei company——"Research of Proposals for AVS Video Coding Standard"and Youth Chenguang Program of Wuhan——"Research on Joint Power-Rate-Distortion Model for Wireless Mutimedia Communication"(Grant No.: 20045006071-18), this thesis coducts an in-depth research on the inter prediction technology based on content. Meanwhile, we have participated in the AVS (Audio Video Coding Standard Work group of China) workgroup including proposing the technical proposals and the standard establishment.In recent years, toward the contradiction between the calculation capability requirement and limited energy supply, how to balance the coding efficiency and complexity becomes a hot issue in video coding field. As the key technology in video coding, inter prediction acts a decisive role for coding efficiency and complexity. Inter prediction is deduced to be an ideal cut-in point to this solution. Based on the experiment and theory analysis on typical video sequences, the useful correlation and content characteristics in inter prediction is analyzed to improve the motion vector prediction part and search scheme for sub-pixel search. In the motion vector prediction part, different from the traditional spatial prediction, an adaptive spatio-temporal prediction scheme is proposed. In the proposed scheme, the predicted motion vector is choosed from spatial neighbouring or temporal correlated block by the correlation degree of the neighbouring motion vectors. Experiment results show that with little complexity increasement, the predicted motion vectors are more accurate compared to the traditional method. In the sub-pixel search part, by analysis of the probability distributing of best search position, the sub-pixel points are selected utilizing the sub-pixel information in the previous and current frames. Experiment results show that the search points reduced more than 70% compared to the full search while keep the coding efficiency.In addition, macroblock based AVS-P2 Golden C decoder is introduced for the bright foreground of IPTV and video surveillance application. Compared to the frame based decoder, the Golden C decoder can reduce both temporal and spatial complexity effectively, also in favor of the parallel operation of macroblocks. The Golden C decoder lays foundation for the embedded codec on special flat (e.g. DSP).Taking advantage of the correlation and content characteristic in video sequences, the improved inter prediction algorithm achieved acceptable tradeoff between coding efficiency and complexity. It provides an effective solution to the bottleneck of mobile video application in the next generation and holds extensive application value in the AVS industrialization process.