High accuracy character recognition algorithm using fourier and topological descriptors

Abstract

This paper presents a new two-stage character recognition algorithm yielding accuracies in excess of 98%. In the first stage a primary classification using Fourier descriptors is derived. The Fourier descriptors are obtained from a Fourier analysis of the derived boundaries of the digitized image of the characters. Due to the rotational invariance of the Fourier descriptors, no distinction is made between the characters 2 and 5, 6 and 9 and occasionally between 4 and 7. In order to resolve this ambiguity, a secondary classification that utilized topological descriptors is initiated. This combined two-stage algorithm yields high recognition accuracies while retaining computational simplicity.