Coloring Pekalongan Batik Using a Madura Dataset: A Comparative Study of GAN and Caffe-Based CNN Models

Madura Batik, as one of Indonesia's valuable cultural heritages, is known for its unique characteristics involving the use of bright colors such as red, yellow, and green, as well as traditional motifs that often feature elements of nature like flowers, leaves, and animals. Each motif in Madura Batik reflects the rich philosophy, values, and stories of Madura culture. This batik is also famous for its production process, which is largely carried out manually using traditional dyeing techniques. However, with the advancement of technology, there is a growing need to integrate technological innovations into the batik dyeing process without losing its traditional essence. This research combines Generative Adversarial Networks (GAN) models and compares them with Caffe-based pretrained Convolutional Neural Networks (CNN) to create new color variations in Pekalongan batik images. The input for the models is grayscale batik images, which are then processed to generate colorful outputs. The dataset used consists of 519 Madura batik images, with a distribution of 80% for training, 20% for validation, and 10 images for testing. The preprocessing process includes resizing, normalization, and batching to accelerate model convergence. Performance evaluation is conducted using FID, MSE, PSNR, and SSIM metrics. The results show that the GAN model with 100 epochs produces better image quality compared to the Caffe-based pretrained CNN model, particularly in terms of visual and structural similarity. In conclusion, the GAN method offers great potential for innovation in batik coloring without compromising its traditional motifs.