Using Deep Learning and Adaptive Thresholding Approach for Image-based Baybayin to Tagalog Word Transliteration

The "Baybayin" writing system is a National Cultural Treasure in the Philippines of which a House bill 1022 was formulated in order to revive its use. However, as the new generations of people come, the languages have also evolved and developed making it more challenging to recognize and understand the Baybayin scripts. Thus, this study proposed an alternative approach for word-level Baybayin transliteration on text images. The method entails the following: (1) application of adaptive thresholding technique for Baybayin character extraction, (2) classification of every extracted character to its syllabic equivalent through the generated 97.62% accurate CNN-based Baybayin Optical Character Recognition model, (3) concatenation mechanism for constructing the transliterated word, and (4) word validation using Filipino word dictionary. Using 200 Baybayin word testing images, the transliteration achieved a satisfactory accuracy rate of 92.00%.