Online Kannada Character Recognition Using SVM Classifier

The objective of the paper is to recognize online handwritten Kannada Characters using SVM Classifier. The online character acquisition process involves the capturing of data using iball 5540U Pen Tablet. The sensor picks up the pen-tip movements and also the pen-up/down switching. Kannada script consists of 49 phonemic letters, divided into three groups: 13 vowels, 34 consonants, and other two special characters. 2940 samples of Kannada base characters consisting of single stroke, double stroke and triple stroke characters are used to build the database. Support Vector Machine (SVM) classifier is used to recognize the characters. Seven features are extracted during preprocessing of the data samples. Of these three best features viz. normalized coordinates, normalized trajectory, and normalized deviation are selected based on k-fold cross validation for different train to test data ratio. These features give an average recognition rate of 97.14%, 97.55%, and 92.65% respectively for SVM classifier with RBF kernel.