Scalable video coding using Wyner-Ziv codes

Abstract

This papers proposes a solution to the problem of scalable predictive coding of video by posing it as a variant of the Wyner-Ziv side information problem. We model the predictor for each frame of a video sequence as being generated by passing the video frame through a hypothetical communication channel. Through this interpretation, the problem of encoding a video frame is recast as the problem of correcting for the errors introduced by the hypothetical communication channel. Based on this, we propose a video compression algorithm that employs channel codes for source compression. We propose a bit-plane factorization approach that facilitates fast decoding of the video frames. In particular we use a bank of dependent LDPC decoders to decode each video frame. Though we only discuss the application of SNR scalable coding of video in this paper, the proposed approach is directly applicable to temporal scalability, spatial scalability and error resilient communication of video. Empirical results demonstrate that the proposed approach is approximately 4 dB superior to the conventional approach to scalable video coding.