Optical Character Recognition for Printed Javanese Script Using Projection Profile Segmentation and Nearest Centroid Classifier

Preservation of cultural heritage manuscripts has been a constant challenge in countries where the societies lacked interest in doing a serious preservation. Therefore, digitisation become the most feasible option to help in preserving many forms of cultural legacy. In terms of manuscript preservation, optical character recognition (OCR) is the best option aside from keeping only the photograph form of a manuscript. OCR opens further possibilities to preserve not only the text but also the knowledge or wisdom contained within, as it is possible to provide a transliteration or translation using the stored digital text. This study proposes an OCR for Javanese script, one of the digitally under-resourced language which faced a usage decline in Indonesia. The recognition is performed using projection profile analysis for the segmentation and binary image features to be classified using nearest centroid classifier (NCC). Printed books of various Javanese font size and type are scanned and used for training and testing. Experiment results shown that projection profile (PP) provides an acceptable segmentation result, namely 93.88% for line segmentation and 73.59% for character segmentation. As for the classification, the combination of binary image features and NCC yields the result of 60.6% accuracy in recognition. This is mainly caused by wide array of character compound classes (over 600) produced by the usage of PP segmentation, with which the selected features are not strong enough to distinguish the classes.