Comparison of HMM- and SVM-based stroke classifiers for Gurmukhi script

Abstract

With the evolution of touch-based devices, development of handwriting recognition systems has received attention from many researchers. An online handwriting recognition system for Gurmukhi script is proposed in this paper. In this work, 74 stroke classes have been identified and implemented for character recognition of Gurmukhi script. Seventy-two different combinations of SVM- and HMM-based stroke classifiers with five different features have been experimented. The results of recognition of 35 basic characters of Gurmukhi script on a data set of 1750 Gurmukhi characters written by 10 writers have been reported using three best classifiers and a voting-based classifier built with the help of these classifiers. A character recognition rate of 96.7 % has been achieved using the voting-based classifier, whereas a recognition rate of 96.4 % has been achieved with an HMM-based classifier.