Shape descriptors based handwritten character recognition engine with application to Kannada characters

Abstract

In this paper, we discuss the implementation of shape based features, namely, Fourier descriptors and chain codes, for performing optical character recognition of binary images with application to Kannada handwritten characters. Invariant Fourier descriptors and normalized chain codes are obtained as features from preprocessed Kannada character binary images. Well known SVM classifier is used for recognition purpose. As an initial step towards recognition of handwritten characters, we have performed experiments on handwritten Kannada character numerals and vowels. The result computation is done using five-fold cross validation. The mean performance of the recognition system with the two shape based features together is 98.45% and 93.92%, for numeral characters and vowels, respectively. Further, the mean recognition rate of 95% is obtained for both vowels and characters taken together. © 2011 IEEE.