A Sundanese Characters Recognition Based on Backpropagation Neural Network Approach

摘要

The Sundanese script is the original script of the archipelago inheritance which should be recognized and preserved. This study is to recognize the Sundanese script pattern with the Backpropagation Neural Network (BPNN) algorithm. This research starts from image acquisition, image preprocessing process which consists of normalization of size, thresholding, image sharpening, binarization and thinning, feature extraction, analysis by the BPNN method. This test obtained an accuracy rate in recognizing Sundanese script patterns of 95.23%. The most optimal variant of network architecture is the variation of learning rate of 0.02 and the number of hidden layers of 90 layers so that an MSE of 4.99×10-10 is obtained with a learning time of 2 minutes 45 seconds. So, the BPNN method can be recommended to recognize Sundanese script patterns.