Classification of Malaysian and Indonesian Batik Designs Using Deep Learning Models

Abstract

Batik is a traditional textile art form native to Southeast Asia, especially prominent in Malaysia and Indonesia, where unique patterns reflect significant cultural value. The intricate designs of batik, often embodying floral, geometric, and symbolic elements, make automated classification challenging and time intensive. This study presents a method for classifying Malaysian and Indonesian batik patterns using deep learning models. A curated dataset of 1,825 batik images was compiled, consisting of 949 Indonesian batik images and 876 Malaysian batik images. Three popular Convolutional Neural Network (CNN) architectures: MobileNet v2, YOLO-v8, and LeNet-5 were evaluated based on classification accuracy, loss, and training efficiency. Results show that YOLO-v8 achieved the highest accuracy at 98.80%, followed by MobileNet v2 with 97.79%, and LeNet-5 with 92.94%. These findings indicate that CNN models can effectively distinguish between Malaysian and Indonesian batik designs, offering valuable applications in cultural preservation and industry documentation. Future work could focus on refining these models for real-time use and expanding the dataset to capture additional regional variations in batik design.