Research On Extrapolating Side Information Generation Algorithm In Distributed Video Coding

Distributed Video Coding (DVC) is a new video compression architecture. In DVCschemes, the frames are coded independently at encoder but jointly decoded at decoder. DVChas simple encoder and complex decoder, and is applicative to the video terminals whosecomputing speed and power consumption are limited. Side information generated at decoderis the prediction of WZ frame to be decoded. The accurate side information can reduce therequired stream bits from encoder and improve the Rate-Distortion (RD) performance of thesystem. Thus, the technology of the side information generation at the decoder plays a criticalrole in DVC systems. Compared with the interpolation method, extrapolation method forgenerating side information has a low decoding delay, and still maintains a good compressionefficiency with the increase of GOP. Because of these advantages, extrapolation method ismore practical. The main work and research results in this paper are focused on extrapolatingside information generation, which are as follows:1. On the basis of in-depth study of the various interpolating and extrapolating sideinformation generation methods, a new extrapolating side information generation algorithm isproposed. A set of techniques is used in algorithm, such as weighted median vector filter,Adaptive Search Range, Motion Projection, and hierarchical motion estimation. Experimentalresults show that, for video sequences with different types of motion, the extrapolating sideinformation algorithm proposed can generate good side information, and for video sequenceswith fast motion, the algorithm works especially well.2. Based on the proposed extrapolation algorithm, a novel iterative side informationgeneration algorithm has been proposed, and more information contained in the parity bits canbe utilized. For each WZ frame, first, the algorithm generates side information byextrapolation method, and parity bits are requested from encoder to decode the WZ frame.Second, using the decoded WZ frame, the next WZ frame is predicted and decoded. Finally,the adjacent forward and backward decoded frames are used to interpolate the sideinformation of the WZ frame, and the decoder requests more parity bits from the encoder to decode the WZ frame. The algorithm makes use of more information contained in the paritybits. Experimental results show that, the iterative side information generation algorithmproposed has improved the system RD performance.