Realistic Batik Motif Synthesis Using GAN with Convolutional Attention and Super-Resolution Discriminators

Abstrak

Batik is a local Indonesian cultural tradition composed of intricate patterns that represent distinct aspects of cultural identity, philosophy, and indigenous wisdom, and is a prominent textile art form. Despite its cultural wealth, it is challenging to conserve and digitally reinterpret Batik patterns, as generating visually convincing, stylistically faithful motifs remains an ongoing project. In order to overcome these shortcomings, this study presents an enhanced generative model named BatikGAN, wherein a Convolutional Block Attention Module (CBAM) is added to the generator, and an Enhanced Super-Resolution GAN (ESRGAN) is employed in the discriminator. CBAM improves the generator's ability to focus on salient spatial and channel features, whereas ESRGAN enables more robust evaluation of image realism at high resolutions. The model was initially trained and subsequently validated on the Batik Nitik dataset, which comprises 120 curated images depicting traditional motifs. Multiple training cases were analyzed by switching the number of residual blocks to explore the trade-off between computational cost and visual quality. Out of the models evaluated, the BatikGAN CE6 had the best results in terms of the lowest Fréchet Inception Distance (FID) of 13.691, maximum Peak Signal-to-Noise Ratio (PSNR) of 24.637, and best Structural Similarity Index (SSIM) of 0.897. These findings show that BatikGAN CE6 can provide high-quality Batik patterns that closely match those of manually created forms. This study contributes to a new direction in processing for generating digital Batik and to the evolution of Batik, and is seen as a promising intersection of cultural heritage and state-of-the-art deep learning techniques.

Identifier

• OpenAlex: https://openalex.org/W7128902042
• DOI: https://doi.org/10.62527/joiv.10.1.4040