ENSEMBLING DEEP CONVOLUTIONAL NEURAL NEWORKS FOR BALINESE HANDWRITTEN CHARACTER RECOGNITION

Abstract

While deep learning has proven its performance in various problems and applications, it also opens opportunities in a new way to promote the heterogeneity of cultures and heritages. Balinese script is a cultural heritage in Bali, where it is used to write on palm-leaf manuscripts and contains essential information. Most manuscripts were damaged due to age and lack of maintenance, so a digitalization technique should be developed. In this study, we propose an ensemble of deep convolutional neural networks to recognize the handwritten characters in the Balinese script. We extensively compared various deep convolutional neural network architectures, and the results showed that our ensemble methods achieved the state of the art.