Style transfer of ink wash painting based on deep convolutional neural network and feature scaling
Jia Zeng · International Journal of Information and Communication Technology · 2025
This study proposes AdaFSNet, a neural style transfer model tailored for mural images with intricate textures and cultural motifs. Leveraging adaptive feature scaling, AdaFSNet enhances style-content fusion while maintaining chromatic and structural integrity. Trained on 200 diverse samples from WikiArt, InkWash, and MuralSet, the model is evaluated on both seen and unseen mural styles. It achieves a PSNR of 25.8, SSIM of 0.86, and LPIPS of 0.17, outperforming baseline models such as AdaIN and MSG-Net. AdaFSNet demonstrates strong generalisation in zero-shot settings, offering practical value for digital heritage conservation and stylisation of culturally significant artwork.