Handwritten Devanagari Character Recognition using Convolutional Neural Network

Abstract

In any pattern recognition task, feature extraction and classification stages handle the responsibility of recognizing the patterns accurately. Deep learning relives the task of feature extraction and extracts them automatically, thus reducing the programmer's burden. Deep learning is replacing other pattern recognition techniques recently. In applications like character recognition, which involves large amount of database and variability in the data, deep learning is the right choice to handle the challenges involved. In this paper, a system for handwritten Devanagari character recognition using Convolutional Neural Network is discussed. The recognition accuracy obtained is 91.23% for Devanagari characters and 100% for Devanagari numerals.