Handwritten Bengali Digit Recognition Using Machine Learning Approach

Abstract

Hand written digit recognition in Bengali language is one of the important starting point for creating an OCR (Optical Character Recognition) in the Bengali language. Nevertheless, sometimes absence of massive and fair data collection, recognition in Bangla digit was not build already. In our work, an honest and fair data source known as NumtaDB is make use for recognition Bengali digits. Moreover, troublesome endeavor is added to get the solid presentation and good precision for fair, common, natural, gigantic and particularly more extended NumtaDB dataset. In that case, various sorts of preprocessing frameworks are make use for planning images and significant CNN (Convolutional Neural Network) is make use for the request of representation in this paper. Furthermore, LeNet-5 architecture based CNN (Convolutional Neural Network) model has specified superb execution. Our accomplished testing exactness 97.5% which is an adequate outcome for massive and honest NumtaDB dataset contrasting with other one-sided datasets. A numerous of preprocessing of images is as well as significant before preparing. We make use of some preprocessing strategies for hazy and loud images. Yet these are not sufficient for the first class. Examination of the system gets out the EMNIST and MNIST datasets were performed so as to comfort the evaluation.