Pengenalan Pola Karakter Tulisan Tangan Aksara Bali Menggunakan Fitur Zoning, Direction, dan Backpropagation

Abstract

Balinese script has a character with high similarity. Identifying classes between characters requires an optimal pattern recognition model by maximizing the use of feature extraction methods. The use of feature extraction methods in the dataset aims to obtain the characteristic value of each character, in the case of Balinese script data, a method with detailed feature retrieval is used using the zoning method. This study also added the additional factor of the direction feature. The learning method uses a neural network with a backpropagation algorithm. Tests on character data get the highest accuracy in the combination of 16x16 zone ICZ + ZCZ zoning features with the addition of direction features that are 91.18%, ZCZ zone 16x16 zoning and direction with 86.82% accuracy, and ICZ zoning 16x16 and direction zones with 82.43% accuracy. The highest increase in accuracy is found in the ZCZ feature with a difference of addition of 4.36%. The implementation of the model in word testing has an accuracy of 66.2 % and the results of segmentation testing are 97.33 %.