Blind Image Restoration

Study of Identification Metod of the Image Model

Abstract

In this paper, we propose a 2-D image restoration and identification algorithm only using the blurred noisy image without any a priori knowledge of original image. The parametric canonical-formed state space model is identified by maximum likelihood method. For obtaining maximum likelihood estimates of model parameters, we apply the Expectation-Maximization (EM) algorithm with the initial estimates of obtained by Subspace identification methods. In addition, the stability of models and the convergence of parameters are guaranteed using the stabilization algorithm in the proposed algorithm.