Segmentation Based Approach For Off-line Handwritten Sinhala Word Recognition From Touch Screen Gestures

The traditional way of using pen and paper to take notes is getting over by the touch screen devices. These devices provide more options to the users to enhance their productivity while taking notes. The ability to recognize and validate the words written on the touch screens facilitates further capabilities to the users. Hence, in this paper, we describe a segmentation-based approach combined with an n-gram model for the recognition and validation of the Sinhala words written on touch screens. We compare the results of 6 commonly used machine learning models to find the best performing classifier for recognizing individual characters of words. The classifiers are trained to identify 19 different Sinhala characters. Based on the results, Convolution Neural Network (CNN) based word classifier stands ahead of other classifiers.