Multiple description Wyner-Ziv video coding based on decoder multi-resolution motion refinement

Abstract

In this paper, we consider multiple description video coding (MDVC) for robust transmission over packet loss networks. With the motion compensated prediction, it is known that MDVC suffers from predictive mismatch in presence of transmission losses, where the reference frame received at the decoder may be different from that used at the encoder. Some works have been presented to solve the mismatch problem by applying distributed source coding (DSC) techniques to the MDVC. However, due to limitations of motion estimation techniques used (at either encoder or decoder side) in these DSC based MDVC schemes, coding efficiency is still relatively low. Recently, a decoder multi-resolution motion refinement (MRMR) scheme in the wavelet domain was proposed for more efficient distributed video coding. We thereby apply the decoder MRMR scheme in the MDVC to construct MRMR based MDVC for higher coding efficiency. The MRMR based scheme can flexibly use any available frame or their combinations at the decoder to generate a better reference, thus decreasing coding rate required for the distributed source decoding. Experimental results demonstrate better coding efficiency of the MRMR based scheme in the case of transmission loss.