Combining phase-diversity with spatial-diversity multihypothesis motion compensation

Abstract

Abstract-Discrete Wavelet Transform (DWT) has been proved to be a very efficient tool for data compression. Redundant Wavelet Transform (RDWT) conquers the "shift-variant" problem in DWT and enables the motion estimation and motion compensation (ME/MC) in wavelet domain. Also, RDWT produces growing number of phases at each level of the transform, each phase contributes to the ME/MC procedure by providing a unique hypothesis. Based on the previous successful usage of phase-diversity multihypothesis in redundant wavelet transform domain, this paper presents a technique to combine the spatial-diversity multihypothesis with the transform domain multihypothesis. We take the overlapped block motion compensation (OBMC) as one example from spatial domain multihypothesis. Experiment results show that the combination of the two classes of multihypothesis outperforms either method implemented alone. Further more, choose different subbands for OBMC in order to simplify the system is also discussed.