Bangla Handwritten Character Recognition Using Convolutional Neural Network

Abstract

This report presents “Bangla Handwritten Character Recognition using Convolutional Neural Network”. Sample training data, scanned using a modest scanner. Pre-processing steps that follows are skew angle detection and correction, noise removal, line, word and character separation. The separated characters are then fed into a 10 layer Convolutional Neural Network for training. Finally, this network is used to recognize handwritten Bangla scripts. For collection of data we have developed a form that helped us getting the most out of data. The form was designed in such manner so that it helps us during the preprocessing. In preprocessing we segmented each character separately and fed that data to our own developed 10 layered neural network.