A neuro-fuzzy graphic object classifier with modified distance measure estimator

Abstract

The paper analyses issues leading to errors in graphic object classifiers. The distance measures suggested in literature and used as a basis in traditional, fuzzy, and Neuro-Fuzzy classifiers are found to be not suitable for classification of non-stylized or fuzzy objects in which the features of classes are much more difficult to recognize because of significant uncertainties in their location and gray-levels. The authors suggest a neuro-fuzzy graphic object classifier with modified distance measure that gives better performance indices than systems based on traditional ordinary and cumulative distance measures. Simulation has shown that the quality of recognition significantly improves when using the suggested method.