HANDWRITTEN CHARACTER RECOGNITION USING CONVOLUTIONAL NEURAL NETWORK IN THE CONTEXT OF TAMIL LANGUAGE

Abstract

English language is having a high impact on handwritten character recognition. It is a difficult task to generate a model for character recognition especially for south Asian languages while they are having curves and dots in their shapes and the collection of characters as compound characters. Among other South Asian languages (e.g.: - Hindi, Malayalam, Telugu, etc.) Tamil characters are unique, because of the curves and dots they are having on each characters. These unique attributes make it hard to build a model to recognize Tamil characters. Another challenging task is to recognize handwritten characters than the printed characters, since the handwriting of each others are differing from each. Because of that little attention gained to build a model for Tamil handwritten character recognition. Convolutional Neural Network (CNN) and Deep Learning concepts are playing a major part in character recognition with more efficient image classification. This study mainly targeting the Tamil handwritten character recognition using CNN. The model was implemented in python programming language using Google colaboratory platform. The performance of the model was evaluated in each training and testing of the dataset. The maximum accuracy level reached when dataset reaches 50 character classes. Finally the overall accuracy was shown as 94.26% for 247 Tamil characters. Around 125 thousand data used for the model. Considering other similar systems this model shows the maximum performance.