<![CDATA[This study aims to innovate automatic coloring of batik patterns using deep learning models. Specifically, it compares the performance of Generative Adversarial Network (GAN) with pretrained Caffe-based Convolutional Neural Networks (CNN) in coloring images of Madura and Bali batik. The dataset consists of 388 Madura batik images for training, 97 for validation, and 20 distinct images of both Bali and Madura batik for testing. This dataset was obtained through web scraping from batik posts on social media platforms like Instagram, Bing Image Search using specific keywords, and Kaggle, followed by a manual combination and cleaning process. The GAN model was trained with varying epochs (40, 80, 150), while the CNN utilized pretrained Caffe weights. Evaluation was conducted using Peak Signal-to-Noise Ratio (PSNR), Fréchet Inception Distance (FID), Mean Squared Error (MSE), and Structural Similarity Index (SSIM). The results indicate that the GAN model with 150 epochs outperformed the CNN, achieving a PSNR of 29.702, an FID of 84.016, an MSE of 511.8812, and an SSIM of 0.9925, demonstrating superior color creation and artistic detail in batik. Conversely, the CNN model exhibited lower performance, with a PSNR of 28.218, an FID of 200.271, and an SSIM of 0.7925, indicating its limitations in preserving the intricate patterns and colors of batik. This research demonstrates the applicability of GAN in automatic batik coloring, potentially providing innovative solutions for the batik industry while maintaining the cultural and artistic integrity of traditional designs.]]>
Copyrights © 2025
