Synthesizing Spaces for Restoring Membership Functions in Fuzzy Image Recognition Problems
В. И. Васильев, V. I. Sushko · Journal of Automation and Information Sciences · 1996
A method for specifying and restoring membership functions in fuzzily specified image (concept) recognition problems is proposed. The problem of restoring a membership function reduces a procedure of search for metric attributes subspace in which the order of the training sample objects, when ordered by their distance from the standard object of the given concept, coincides with the order of objects that is defined on the set of membership in this concept. The operational quality and reliability of the restored membership function on new data are guaranteed to be no worse than pre-specified values.