Handwritten Isolated Bangla Compound Character Recognition

Abstract

Bangla compound characters are a part of Bangla alphabet. Without developing a better recognizer for BangIa isolated compound characters Bangla OCR is incomplete. Whereas most of the researches have been conducted for Bangla basic characters and numerals, little has been done for handwritten Bangla compound characters. From this motivation, a convolutional neural network (CNN) based model has been developed to recognize handwritten isolated Bangla compound characters. The performance of the proposed model has been analyzed by training it over CMATERdb 3.1.3.3 and comparing the result over test dataset with other existing methods for handwritten Bangla compound character recognition. The result of the network showed 95.5% accuracy on the test dataset which is better than some current approaches.