ResNet-50 for Classifying Indonesian Batik with Data Augmentation

Abstract

Batik is composed of various artistic images and patterns, which are called batik motifs. The diversity of batik motifs is influenced by the culture of a region which has a philosophical meaning. Indonesia as a country of cultural diversity has unique batik motifs in each region. Manual identification of batik motifs requires special knowledge and experiences from experts. Various methods are applied to classify images, among others is the Convolutional Neural Network (CNN) method. This study classifies batik images by applying deep learning using the Convolutional Neural Network (CNN) method with ResNet architecture. The number of original batik image dataset consists of 300 images with 50 classes. Augmentation process produce 1200 new image with the same number of classes. Testing scenario compare the accuracy between original data and augmented data with ratio 80:20 for data training and testing. The confusion matrices shows the model provides the highest accuracy performance at 96%.