A motion compensation method using least squares motion estimation filter in wavelet domain

In this paper, a new motion compensation (MC) method in discrete wavelet transform (DWT) domain is proposed to realize hierarchical MC. In order to dissolve shift-invariant or phase shift problem caused by down sampling in DWT, a locally adaptive interpolation filter, called MC filter is applied. This filter predicts DWT coefficients in current frame from the coarsely motion compensated DWT coefficients in reference frame, using motion vector estimated from lower resolution image. Simulation results show that the proposed MC filter is effective to compensate the phase shift caused by down sampling and the proposed hierarchical MC has higher coding gain compared to conventional non-hierarchical MC methods.