Machine Learning based Recognition of Bangla Handwritten Characters

In this paper, we propose a method to recognize Bangla handwritten characters using CNN based machine learning techniques. Our proposed model takes input of single Bangla character images and recognizes it. We worked with 50 basic Bangla characters, 10 numeral digits and 13 special characters. A total of 24231 images from CMATERDB dataset have been used here. Our proposed model achieved 99.06% accuracy on alphabets, 99.75
^%
accuracy on digits and 99.15% accuracy on special characters. We look forward to upgrade this model to a complete Bangla handwritten speech recognition system in future.