Incorporating feature point-based motion hypotheses in distributed video coding

Emerging mobile video applications introduce demands (e.g. low encoding complexity) which do not fit well to conventional video coding (e.g. H.264/AVC, MPEG-4). Upcoming coding schemes, following the distributed video coding (DVC) paradigm, are well suited for mobile video encoding due to their low encoder complexity. The performance of a DVC scheme is mainly influenced by the Slepian-Wolf coding performance and the side information quality. The side information is commonly obtained by temporal interpolation, where fast inhomogeneous motion is still challenging. In this paper, fast motion is negotiated by applying a multiple motion hypotheses pixel-based temporal interpolation method. Global and local motion estimation are incorporated. Simulation results show side information quality improvements by up to 0.6 dB for the sequence Stefan. Furthermore, state-of-the-art side information generation approaches are outperformed by 1.2 dB in terms of overall RD performance.