Convolutional Neural Network based Handwritten Devanagari Character Recognition

Abstract

Industry 4.0 might be thumping at our door, but there are millions of individuals who are still disconnected from even the most fundamental technologies. One of the major reasons for this disparity is the language gap. The majority of them don't know English, which is the de-facto language for most of the technologies. Only 12.6% of all Indians speak English according to the 2011 census. Language is the medium through which an individual expresses himself or herself. Education of the rural section of society is one of the most important things to accomplish if we aim to develop the country. Because of this, there is a need for a reliable software solution for recognizing the traditional scripts. This would also be helpful for the purpose of remote schooling in times of lockdown and quarantine.The paper proposes a methodology for recognition of handwritten ‘Devanagari’ characters. Devanagari being one of the most common scripts, is used in many Indian dialects. The Hindi language is also written in the Devanagari script as shown in Fig 1. This paper explores and analyzes the use of Deep learning techniques such as the Convolutional Neural Networks for the recognition of Devanagari characters. Inspired by the structure of the brain, CNNs classify characters by making use of neurons linked in various layers so as to achieve maximum efficiency. In this paper, 6 layers of neurons were used for the purpose of classifying Devanagari characters. An accuracy of 95.6% was achieved in this approach. Once recognized, the handwritten Devanagari characters can easily be translated into English or any other languages.