Feature Extraction and Classification for Online Handwritten Gurmukhi Character Recognition

Abstract

Online handwriting character recognition is gaining attraction from the researchers across the world because with the advent of touch based devices, a more natural way of communication is being explored. Online system is real time processing in which characters are recognized as they are written. There are various issues associated with online recognition process. Due to variations in handwriting, it is very difficult to achieve high degree of accuracy. Therefore, the research work presented in the thesis aims to develop an efficient system to recognize the input natural handwriting. Script for which recognition is done is Gurmukhi script. Stroke based approach is followed for online recognition of handwritten Gurmukhi characters because of the uniqueness of the strokes in comparison to characters. In the present work, 32 stroke classes have been considered and implemented for online character recognition of Gurmukhi script. Three types of features are extracted, namely Spatiotemporal features, Tangential features and Spectral features. In the thesis work, a hybrid method consisting of multiple features has been proposed to improve the performance of the recognizer. Two types of hybridization have been obtained. First, by combining Spatiotemporal and Tangential features and second, by combining Spatiotemporal and Spectral features. Three different types of classifiers which are K-Nearest Neighbor (KNN), Multilayer Perceptron (MLP), and Support Vector Machines (SVM) have been used for recognition. Recognition is implemented using two methods, namely cross validation technique and percentage split method. Highest accuracy is achieved using MLP and SVM using the hybrid features. KNN also observed good accuracy rate.