Parametric modeling of blurred images for image restoration

Abstract

Almost all of parameter estimation schemes for image restoration to date, attempt to model the true image as a autoregressive model and the point spread function as a moving average model and assume the symmetry of the point spread function in order to reduce the computational complexity. Autoregressive process builds the true image by passing a Gaussian white noise process through a filter and may result in unstable systems and optimization of parameters could be trapped in local minima. In this article a different approach is presented with simulation results where initial white Gaussian process is replaced by scaled degraded image avoiding optimization problems.