Penelitian Pendahuluan Transliterasi Citra Aksara Bali Menggunakan Ciri Momen Invarian dan Algoritma Klasifikasi SVM atau CNN

Abstract

The lontar manuscript is one of the cultural heritages that must be preserved. The lontar manuscript contains many valuable things but is considered no longer exciting and challenging to learn. The study aims to develop a handwritten Balinese script transliteration system from digitizing lontar manuscripts. The peculiarity of this research is the use of research objects and the combination of algorithms used in transliteration. The method used is machine learning with SVM and CNN classification algorithms. 1001 Balinese script images in lontar manuscripts were used as training data. Using the CNN algorithm, an accuracy of 86.42% is obtained, and an accuracy of 82.32% obtains in the SVM algorithm. The model testing was carried out with 18 digitized script images from printed books and obtained an accuracy of 23.53% using the SVM algorithm. The low accuracy value of the testing data is thought to be due to the different shape of the handwritten script imagery with the training data used. This research opens opportunities to be developed by adding training data from various forms of images from different sources. This study also shows that machine learning approaches with SVM and CNN algorithms can potentially be used in developing Balinese script image transliteration systems.