Deep Convolutional Neural Network for Handwritten Bangla and English Digit Recognition

As handwritten digit recognition is one of the most complex tasks in pattern detection for a machine, many works were done and many methods were applied to solve this problem. But, traditional methods have reached the final destination and deep learning techniques are now showing the new effective path in this field. This paper describes the application of artificial intelligence to work out the problem of handwritten English & Bangla digit recognition. The most effective model called Convolutional Neural Network (CNN) is used in this work to detect English and Bangla handwritten digits. After building up this model which contains 19 layers, MNIST and NumtaDB data set of the handwritten digit is used for training. At the stage of testing, constructed CNN model shows the fruitful result for both English and Bangla handwritten digit recognition segments. The trained network is highly capable to give the expected result. The constructed CNN model gets 98.48% accuracy for English handwritten digit recognition and 86.76% accuracy for handwritten Bangla digit recognition.