Performance analysis of flexible zone based features to classify Hindi numerals

Abstract

Various parameters available for human identification but hand written documents always proven its importance. Although many pieces of work had been done for recognition of roman script but only few attempts have been tried for Indian scripts. Hindi is the most popular language of India which is written and encoded using Devnagri script. Zoning is basically the superimposition of nxn grid on the image of character. In this paper we have compared the various zoning alternatives. The performance of fixed boundary and flexible boundary is evaluated. The average distance and average angle of black pixel of the skeleton of the image pattern is used as feature. For recognition of the digits MLP based classifier SVM based classifier is used. Performance for SVM is better than SVM. © 2011 IEEE.