Iterative methods for restoring noisy images

Abstract

An iterative image restoration algorithm is presented that is based on the best linear mean-square estimate of the Wiener filter technique and the iterative approach of the first-order, stationary, linear Jacobi method. The iterative algorithm incorporates a priori knowledge concerning the image and noise statistics directly into the iterative procedure. It also assumes the image covariance to be described by a separable first-order Markov field and exploits the resulting Toeplitz structure. The usefulness and validity of this algorithm are demonstrated by implementation and testing on real images. The advantages offered by the algorithm are that it is computationally efficient, since only a small number of computations per pixel per iteration is required, it requires a small amount of memory, it has a fast rate of convergence, and it exhibits neither high noise sensitivity nor significant loss of resolution. >