Sparse representation for image prediction

This paper addresses the problem of closed-loop spatial image prediction based on sparse signal representation techniques. The basis functions which best approximate a causal neighborhood are used to extrapolate the signal in the region to predict. Two iterative algorithms for sparse signal representation are considered: the Matching Pursuit algorithm and the Global Matched Filter. The predicted signal PSNR achieved with these two methods are compared against those obtained with the directional predictive modes of H.264/AVC.