Enhanced Augmented CNN Model Using Optimized Optical Character Recognition For Off-Line Telugu Characters

Abstract

The Indian Constitution acknowledges the significance of Telugu, Tamil, Malayalam, and Kannada as languages. Worldwide, over 90 million people speak Telugu, a language from South India. Digitalizing books and unstructured documents are one of the applications for Telugu optical character recognition (OCR), which improves human-to-human communication. The training of optical character recognition systems is more extensive for international languages such as English and German than for regional languages such as Telugu, Tamil, Malayalam, etc. The main difficulty in developing TOCR is the large number of unique characters used in the Telugu language. This paper makes two contributions to help with this: (i) a collection of Telugu characters written by hand, and (ii) a convolutional neural network (CNN) model with enhancements to identify the hand-written characters in the scanned text.