@Article{cmc.2026.077803,
AUTHOR = {Yuanjing Luo, Xichen Tan, Yinuo Jiang, Zhiping Cai},
TITLE = {ArtFlow: Flow-Based Watermarking for High-Quality Artwork Images Protection},
JOURNAL = {Computers, Materials \& Continua},
VOLUME = {},
YEAR = {},
NUMBER = {},
PAGES = {{pages}},
URL = {http://www.techscience.com/cmc/online/detail/26428},
ISSN = {1546-2226},
ABSTRACT = {With increasing artwork plagiarism incidents, the necessity of using digital watermarking technology for high-quality artwork copyright protection is evident. Current digital watermarking methods are limited in imperceptibility and robustness. To address this, based on comprehensive copyright protection research, we develop a novel watermark framework named ArtFlow, using Invertible Neural Networks (INN). Our framework treats watermark embedding and recovery as inverse image transformations, implemented through forward and reverse processes of INN. To ensure high-quality watermark embedding, we utilize frequency domain transformations and attention mechanisms to guide the watermark into high-frequency areas of the image that have greater protective weighting. These areas are attractive to plagiarizers yet have minimal impact on the artistic integrity of the artwork itself. For strong plagiarism-resistant, we design a noise layer that includes various infringement methods—transmission, plagiarism action, and camera-shooting—to train robust watermark recovery process. Additionally, an image quality enhancement module is introduced to minimize the distortions that may arise from infringement before the watermark recovery. Experimental results across four datasets confirm that our ArtFlow surpasses existing advanced watermarking methods.},
DOI = {10.32604/cmc.2026.077803}
}