Abstract: In order to well utilize the information in the observed images and the complementarities between the information to improve the results of the image restoration, this paper proposed a blind restoration algorithm for multichannel images. Firstly, the prior models of the original image, point spread functions and the observed images were reconstructed, from which the prior distributions of them were obtained; secondly, the Uamma distribution was used to describe the unknown model parameters; finally, based on the inference of the max posterior probability, the optimal original image and the point spread functions were estimated using the evidence analysis method. Compared with the single channel algorithms, the experiments show that the multichannel blind image restoration can obtain better results. Meanwhile,the proposed algorithm shows competitive performance on restored results compared with some state-of-the-art algorithms.

Key words: Image restoration,Bayesian framework, Prior models, Harmonic model, Channel co-primeness