Fuzzy selection filters for image restoration with neural learning

In this paper, a novel class of nonlinear filters, called rank conditioned fuzzy selection (RCFS) filters, is proposed to improve the filtering performance and adaptive capability of rank conditioned rank selection (RCRS) filters. The output of this new class of nonlinear filters is determined by the center of gravity of a selected fuzzy rank which is a fuzzy set of the ranks of the observation samples. Since the membership functions in each fuzzy rank of the RCFS filters are trained by neural learning, they are more adaptive for noise filtering. The computer simulation shows that the RCFS filters have better filtering capability than the RCRS filters and other conventional filters.