Abstract

Batik has a variety of varied motifs, each region in Indonesia has certain characteristics on batik motifs. Based on literature studies theuse of backpropagation neural network methods to recognize complex patterns has a satisfactory rate of success. The purpose of this research is to develop and apply neural networks that are fast, precise and accurate to classify batik designs and patterns. Types of batik motifs typical of Central Java that are used include; Truntum from Solo, Warak Ngendhog from Semarang, Sekar Jagad from Lasem, Burnt from Pati, and Jlamprang from Pekalongan. The image first undergoes RGB color feature extraction based on mean values of R, G, and B, and Gabor filter texture characteristics. The tests were carried out using 90 batik images, 60 batik images for training data and 30 batik images for testing data. The results of the study concluded that the best parameter settings were, the number of hidden layer 30 neurons in the first layer and 15 in the second layer, with 6 input layers and 5 output layers. Gabor filter with 90º orientation angle and wavelength 4 become the best combination in this study. From the results of training and testing results obtained an average accuracy of 93.3% in all batik classes in Central Java.