An enhanced approach to character recognition by Fourier descriptor

Abstract

A new algorithm of utilizing Fourier descriptors (FDs) as unique features in representation and classification of contours is proposed. It enhances the description of local information and distinguishes similar contours. The characteristic of this algorithm is to represent the object by several sets of FDs which represent different portions of the object in contrast to only one set of FDs which represents the whole object. The authors use a model-based approach in the recognition stage in which these sets, say k sets, of FDs of the input numeral will be matched with each predefined model of the numeral class. It is shown that a higher accuracy rate can be achieved by using a multicategory classifier incorporated with an artificial neural network classifier. Finally, an experiment on numeral recognition by the proposed algorithm is reported.<>