KDANet: optical recognition for Bangla language using deep neural networks

Abstract

When images of printed or handwritten are converted; be it mechanically or electronically to an editable text format, this is called optical character recognition. Bangla is one of the most complex languages as it has so many characters and digits. Moreover the Bangla language has about 300 composite characters. That is why the extraction of characters from images is more difficult for Bangla compared to other languages. Deep learning has recently developed good capabilities for extracting high-level features from an image kernel. This paper will propose a custom model KDANet and compare with some popular deep learning models that can recognize handwritten Bangla characters written in various and distinct handwriting styles. These systems learn more accurate and inclusive features from large-scale training datasets than earlier feature extraction techniques.