Character recognition using fuzzy rules extracted from data

This paper describes a character recognition method based on fuzzy rules extracted directly from training data. The fuzzy rules with variable fuzzy regions are defined by activation hyperboxes which show the existence regions of data for a given class and by inhibition hyperboxes which inhibit the existence of data for that class. These rules are extracted from numerical data by recursively resolving overlaps between two classes pair by pair. Using this methodology, a character recognition system with a high recognition performance can be developed and the development effort is relatively low. In this paper, the authors apply the method to develop a fuzzy classifier for recognition of Japanese hiragana (phonetic system) characters, which are mainly composed of curved strokes, of vehicle licence plates. Since feature extraction of these characters is difficult, the authors investigate two different recognition approaches which use different input vectors. One approach directly uses pixel values of a gray scale image as input to the classifier, while another approach uses the central moments calculated from a gray scale image as input. The results and comparison of these two approaches are presented in this paper