Handwritten Kannada vowel character recognition using crack codes and Fourier descriptors

Abstract

In this paper, we present an efficient method for recognition of basic characters (vowels) of handwritten Kannada text, which is thinning free and independent of size of handwritten characters. Crack codes and Fourier descriptors are used for computing features. The recognition accuracy has been studied by comparing the performances of well known K-NN and SVM classifiers. Five-fold cross validation technique is used for result computation. Experiments are performed on handwritten Kannada vowels consisting of 6500 images with 500 samples for each class. The mean performance of the system with these two shape based features together is 91.24% and 93.73% for K-NN and SVM classifiers, respectively, demonstrating the fact that SVM performs better over K-NN classifier. The system methodology can be extended for the recognition of remaining set of Kannada characters. © 2011 Springer-Verlag.