Multiresolution Video Coding Based on Kalman Filtering Motion Estimation

Abstract

Motion estimation plays an important role for the coding of video signals in the framework of multiresolution. The conventional block-matching algorithms achieve only integer pixel accuracy, if interpolation is not performed. In this paper, a multiresolution video coding system based on Kalman filtering motion estimation is presented. The Kalman filter-based motion estimation employs the predicted motion and measured motion to obtain the best estimate of motion vector at the lowest resolution subimage. The motion vector is used as the initial estimate of other subimages and then refined with the conventional block-matching algorithm. The multiresolution motion-compensated difference of each subband is encoded by lattice vector quantization according to their statistical properties, respectively. Simulation results show that the proposed coding system not only improves the performance but also reduces the error propagation significantly when compared to the multiresolution video coding system based on conventional block-matching algorithms.