Research On Side Information Refinement Method For Distributed Video Coding

Distributed Video Coding (DVC) is a new paradigm for video compression. Comparedwith traditional video coding technology, DVC exploits the source statistics in the decoder,and hence, the encoder can be very simple, at the expense of a more complex decoder. Thequality of SI significantly influences the coding performance in DVC scheme. The moreaccurate the side information is, the higher the quality of reconstructed frames is and thelower the DVC rate is. But in most situations, only a crude estimation is generated to “guess”the original WZ frames, so there is still a large PSNR gap between the Wyner-Ziv codec andthe conventional video codec.This paper aims to research the side information refinement methods of Wyner-Zivcoding. In this paper, a simple denosing way to utilize the information of SI and the decodedKey frames to refine the current SI is proposed and also introduced into iterative sideinformation refinement scheme. The main work and achievements are as follows:1. This paper systematically expounds the basic principles and key technologies ofdistributed video coding (DVC), and then implements the Pixel-Domain and DCT-DomainDVC system based on turbo codec. The RD performance of DVC is discussed and comparedwith H.264intra coding.2. According to the correlation noise modeling between side information (SI) and sourcein DVC, a novel SI refinement method—SI denosing is proposed, which consider theoriginal SI as a noisy version of the source, and get preferable SI by motion compensationmethod. In this motion compensation method, OBME, WCI and OBMC are used to reducethe blocking artifact and get high quality motion compensated frames. Experimental resultsshow that SI quality and RD performance have better gains for various quality of SI.3. Iterative side information refinement methods which make use of additionalinformation from partial decoded Wyner-Ziv (WZ) frame can iteratively improve RDperformance by utilizing motion compensation method. The denoising method is proposed toenhance the accuracy of block-matching and improve the quality of motion compensatedinterpolation, which is able to improve the effective of iterative side information refinementmethod. In this paper, iterative side information refinement methods based on denosing for Pixel-Domain and DCT-Domain DVC are respectively proposed. Experimental results showthat the proposed strategy is better in both side information and RD performance, whilemaintaining the advantage of low algorithm complexity.