Impact of the key frames quality on the overall Wyner-Ziv video coding performance

Distributed video coding (DVC) is a new paradigm based on the slepian-wolf and Wyner-Ziv theorems where the temporal redundancy is mainly exploited at the decoder and not anymore at the encoder as in conventional predictive video coding. One of the most popular Wyner-Ziv video coding architectures, the stanford architecture, splits the video frames into two types of frames: key frames and Wyner-Ziv (WZ) frames. Key frames are typically coded using existing intra coding techniques; these key frames are used at the decoder to estimate the WZ frames generating the so-called side information. Since the side information quality strongly depends on the key frames, this paper investigates the impact of the key frames quality on the WZ frames coding and on the overall video codec RD performance. Results show that there is an upper bound for the key frames quality after which the overall RD performance does not improve. The results will be very useful to develop an efficient encoder rate control solution for an unidirectional DVC codec.