Isolated Handwritten Balinese Character Recognition from Palm Leaf Manuscripts with Residual Convolutional Neural Networks

The versatility of machine learning tools creates new opportunities to preserve cultural heritage and promote cultural diversity. One important task for such preservation and promotion is the processing of local languages, of which the digitisation of traditional document written in the local scripts is a fundamental building block. We are hereby concerned with the recognition of isolated handwritten Balinese characters from palm leaf manuscripts. We propose a method based on a residual convolutional neural network to recognise handwritten characters written on palm leaf manuscripts in the Balinese script. The proposed method essentially consists of the combination of identity and convolution blocks. A comparative empirical performance evaluation, using a publicly available data set, shows that the proposed method improves on existing alternatives.