A general sparse image prior combination in super-resolution

Abstract

In this paper the Super-Resolution (SR) image registration and reconstruction problem is studied within the Bayesian framework using a general sparse image prior combination. The representation of the proposed priors as Scale Mixtures of Gaussians (SMG), leads to the introduction of variational parameters, for which degenerate distributions are assumed. In the proposed method all the problem unknowns are automatically estimated using variational techniques. An experimental comparison between the proposed and state of the art methods has been performed, on both synthetic and real images.