Sinhala Inscription Character Recognition Model Using Deep Learning Technologies

Abstract

Manual inscription character reading is a time-consuming task. The identification of a character takes nearly one month in the manual process. Characters have evolved into different shapes over the centuries. Archeology experts analyze all these shapes one by one to recognize a character. Reading inscriptions directly using manual procedures would be time-consuming and inefficient due to a lack of consistency. Automating the process of character recognition will be a huge advantage for both archeology experts and the general public. This is the main objective of this research, which focuses on developing a solution using an optical character recognition module to recognize ancient Sinhala inscription characters. The period from 10 A.D. to 12 A.D. was selected to limit the scope of the study. The final output of the research study has two components. The OCR module facilitates proving the recognized characters when the user inputs a scanned image of the inscription. The GIS module is used to present a map for inscription site tracking features that facilitate users' visits to the locations of inscriptions. Mainly, three OCR solutions were developed based on template matching, artificial neural networks (ANN), and convolutional neural networks (CNN). After evaluating each OCR solution, the best-result OCR solution was further implemented.