Off-line Balinese Handwritten Character Recognition based on Backpropagation Neural Network

Abstract

In this paper an off-line character recognition system for Balinese alphabet based on backpropagation neural networks is developed. The process of the system can be grouped into four stages: image acquisition, pre-processing and segmentation, feature extraction using PCA algorithm and classification based on Artificial Neural Networks using Resilient Backpropagation algorithm. A total of 1,980 samples of images are considered for experiments, and the overall accuracy is found to be 96.708% when the network is set to have 100 nodes on its hidden layer. The total samples trained and divided into three groups: 60% as trained set, 20% as validation set and the rest 20% are used as test set. The novelty of the proposed method is that, it is thinning free, fast and reliable with the degree of accurate above 95 percent.