The Method Of Side Information Estimation Based On Universal Prediction

With the development of multimedia communication technology, new video applications, such as wireless video surveillance, wireless PC cameras, mobile video phones and multimedia sensor networks are emerging continuously. These new applications require low computational complexity coding techniques. Under this circumstance, the researches on video coding with low computational complexity are becoming increasingly urgent, and more and more efforts are put on distributed video coding. For distributed coding, Slepian-Wolf theory and Wyner-Ziv theory are the main theoretic basis. Distributed video coding can reduce encoder computational complexity, and it makes the design of low complexity video codec become possible.For a distributed video coding system, the coding and decoding efficiency depends in a great extent on the quality of side information estimated at the decoder side, so the study of effective side information estimation methods is of much significance for distributed video coding. The aim of this paper is to try to find an effective way to extract side information to reduce the decoder complexity while maintaining quality of side information on the decoder side.First this paper reviews the theoretic background and research development in distributed video coding. Then, on the basis of analyzing the limitations of side information estimation based on motion compensated interpolation, this paper puts emphasis on the research on universal prediction based side information estimation. This kind of method does not depend on any motion model and could greatly reduces the decoder complexity. However, the performance of universal prediction based side information estimation method is not good enough when there are obvious inter-frame motion. To overcome this, this paper proposes a new improved universal prediction based side information estimation method. In the new method, frames are divided into blocks first. Then, the blocks are classified into two classes according to motion activity. For a block which does not move apparently, universal prediction is used to estimate the side information in it, while, for a block which has apparent motion, the motion compensated interpolation is used to estimate the side information in it. By combining the above two side information estimation methods, not only is the quality of estimated side information greatly improved but also the computational complexity is reduced simultaneously. The experimental results on testing video sequences verify the effectiveness of the new side information estimation method proposed in this paper.