Typewritten Gurmukhi Character Recognition using Deep Learning Architecture

Abstract

In this era of digitalization, most businesses, work, and even daily life activities are influenced by technology. Nowadays, most textual data, including books, novels, old thesis/ dissertation books, and holy books, are available in digital form. Moreover, millions of pages containing typewritten text and images have been digitally preserved thanks to the ever-increasing availability of scanning and storage equipment, which has inspired archivists and historians to digitize many historical data. However, they rapidly understood that making historical records accessible to the academic community and the public requires more steps than just digitalization, even if that step is perhaps the simplest. Moreover, most historical documents are in the typical native languages, so systems need to recognize the characters. In this paper, the dataset of typewritten Gurmukhi characters is used to develop a Gurmukhi typewritten text recognition system. This dataset is collected from various typewritten thesis documents, books, etc. The recognition method is based on deep learning architecture called a deep convolutional neural network. The performance of the proposed system is tested in terms of accuracy, and it achieved an overall accuracy of 97.47%, which is suitable for the text recognition system.