Multichannel regularized recovery of compressed video sequences

In this paper, we propose a multichannel regularized recovery approach to ameliorate coding artifacts in compressed video. The major advantage of the proposed approach is that both temporal and spatial correlations in a video sequence can be exploited to complement the compressed video data. In particular a temporal regularization term is introduced to enforce smoothness along the motion trajectories defined by the transmitted motion vectors for motion compensation. Several forms of temporal regularization with different computational complexity are considered. Based on the proposed approach, recovered images are obtained from the compressed data using the well-known gradient-projection algorithm. Moreover, an iterative algorithm is proposed for the determination of regularization parameters at the coder side. A number of numerical experiments using several H.261 and H.263 compressed streams are presented to evaluate the performance of the proposed recovery algorithms. Results from these experiments demonstrate that the use of temporal regularization ran yield significant improvement in the quality of the recovered images-in terms of both visual evaluation and objective peak-signal-to-noise (PSNR) measure