OCR for Printed Bangla Characters Using Neural Network

Optical Character recognition is a buzzword in the field of computing. Artificial neural networks are being used to recognize characters for a long time. ANN has the ability to learn and model non-linear and complex relationships, which is really important because in real life, many of the relationships between inputs and outputs are non-linear as well as complex. Research in the field of OCR with Bangla language is not as vast as the English language. So, there is a scope of research in this area. It can be used to search and scan hundreds of Bangla documents within seconds and can easily manipulate the data. It is developed for various purpose like for vision impaired person where OCR software can help turn books, magazines and other printed documents into accessible files that they can listen. The limitation of traditional OCR are sufficient dataset is not available, all different font of characters are not available and there are lots of complex and similar shape characters for which accuracy not good. In our research, we first tried to make a dataset large enough so that we can train our neural network as they require big data to train. We built our own dataset of 2,97,898 Bangla single character images of different fonts . Then for implementing neural network we used Scikit-learn’s multi-layer perceptron classifier and we also implemented our own multi-layer feed forward back propagation neural network using a machine learning framework named Tensorflow. We have also built a GUI application to demonstrate the recognition of Bangla single character images.