Fuzzy ordering theory and its use in filter generalization (Q5957448)

The main purpose of this paper is to develop the concepts of fuzzy spatial and rank (SR) orderings, fuzzy observation samples and order statistics, and to utilize these new concepts in the generalization of median, rank condition rank selection (RCRS), and weighted median filters. In detail, the paper surveys firstly the fundamental concepts of the strict and fuzzy SR ordering relations. Several key properties of fuzzy SR ordering and fuzzy order statistics are then developed to show the advantages of this more general approach. The unbiased behaviour of the fuzzy median filtering provides an improved edge preservation and noise smoothing when compared to the crisp, or standard median filter. Fuzzy relations are used to extend the class of RCRS filters to the class of fuzzy RCRS filters. The fuzzy RCRS filters incorporate rank order, spatial order, and sample spread information into the filtering process. Finally, fuzzy weighted median filters are defined as a straightforward extension of weighted median filters. Optimization procedures of the parameters for the newly defined fuzzy filters are developed and discussed. Simulation numerical results illustrate the relative performance of crisp and fuzzy based filtering methods, indicating that fuzzy filters converge to crisp filters in highly impulsive environments, have advantages over the crisp filters in additive noise cases, and provide robustness to changing image statistics.