Farsi handwritten character recognition with moment invariants

Abstract

This paper introduces an experimental evaluation of the effectiveness of utilizing various moments as pattern features in recognition of the handwritten Farsi characters. The moments that have been used are Zernike moments, pseudo Zernike moments, and Legendre moments. We have used an unsupervised neural network (ART2) for this application, so that the clusters are formed only based on inherent properties of pattern features. The performance of classification is dependent on the moment order as well as the type of the moment invariant, but the classification error rate was below 10% in all cases. The pseudo Zernike moments of order 5 had the best performance among all the moment invariants. Its error rate and discrimination factor were 3.06% and 96.92% respectively.