<title>Median regularization of iterative image restoration methods</title>

Abstract

This paper introduces an approach to the regularization of the iterative restoration methods based on the median filtering. The comparative analysis of low-pass and median regularization is performed. The median filtering is shown to be more efficient as the regularization in the case of noise with mixed distribution. The nonlinearity of the iterative method is provided by the constrain on non-negativity that makes possible to solve the problem of band-limited extrapolation. The use of median regularization does not require to chose the regularization parameter in contrast to Tikhonov regularization. However, the window size is to be chosen according to the noise level and could be considered as a parameter for the adaptive regularization to preserve edges according to the masking effect of human vision system.