Multiple descriptions Wyner–Ziv video coding

Abstract

This paper proposes a multiple descriptions Wyner–Ziv video coding (MDWZVC) architecture, which combines the merits of multiple descriptions coding (MDC) and Wyner–Ziv video coding (WZVC). All of the Wyner–Ziv frames (WZFs) are sub-sampled into four low resolution parallel phases to generate four descriptions, which are encoded and transmitted to the decoder independently. In addition, a successive refined side information (SI) algorithm, which is suitable for the video coding architecture is proposed to exploit both temporal and spatial correlations of the video frames. The simulation results show the SI quality, the peak signal to noise ratio (PSNR) and the rate-distortion (R-D) performance improved with the increasing numbers of received descriptions. By updating the SI quality and reconstruction results of the already decoded descriptions, the R-D performance improved further more. When all of the four descriptions correctly received, the R-D performance improvement of the proposed MDWZVC architecture over DISCOVER codec can be up to 1.8dB, which indicates the effectiveness of the proposed architecture.