MaBaybay-OCR: A Matlab-based Baybayin optical character recognition package

Abstract

Optical character recognition (OCR) is a state-of-the-art technology that allows automated detection and recognition of text from scanned documents and other images. While OCR has highly developed in popular writing systems like Roman, Brahmic, and Han scripts, there is presently a lack of technological integration for Baybayin scripts $-$ a precolonial Filipino writing system. This gap in recent advancements may be due from the script’s antiquity. However, ongoing efforts led by Philippine government and cultural institutions, are actively working towards promoting Baybayin as a means of cultural revival and heritage preservation. In this work, a Matlab-based Baybayin OCR (MaBaybay-OCR) package is introduced to automate the process of transliterating the Baybayin to its modern Filipino Latin form. MaBaybay-OCR takes raw Baybayin text image as an input and performs image analytic strategies to generate correct transliteration results. The software system implements essential pre-processing techniques such as binarization, segmentation, and computation of the character’s features of interest to precisely isolate each Baybayin character. With the use of support vector machine (SVM) classifiers, the platform can accurately identify every Baybayin character’s corresponding Latin equivalent and can realize the Baybayin word transliteration by concatenating each character recognition result. The discrimination of Baybayin from Latin or Roman texts is a distinctive feature of this recognition package. All Matlab source codes are available in a public repository for reproducibility, and a user-friendly graphical user interface (GUI) for convenient usage is provided. To the best of our knowledge, this is the first software program that offers direct transliteration of Baybayin texts up to block-level. It is expected that this work will promote the Baybayin script and contribute towards its positive cultural exposure in the Philippines.