An Automated System for Recognizing Isolated Handwritten Bangla Characters using Deep Convolutional Neural Network
Md. Nahid Hasan, Rafi Ibn Sultan, Mohammad Kasedullah · 2021
Among various handwritten character recognition of different languages, Bangla stands as one of the most challenging tasks. Because of its unique texture and morphologically complex structure often classification models do not provide the expected result as one hopes to have. In this research, a deep novel Convolutional Neural Network (CNN) of 11 layers is proposed. Among the layers, 6 are convolutional layers. This deep CNN model was trained to classify Bangla basic isolated characters of 50 classes (each representing a character). The research utilized a publicly available dataset, CMATERdb 3.1.2, for its classification purpose. The model performed relatively better than the current research on this field and wielded a 98.03% accuracy on the test dataset. This improvement leads us to believe that more accurate results can be achieved in the future by working with other such kinds of datasets or by tweaking the existing model.