SimuFreeMark: a robust watermarking method without noise simulation that resists image editing attacks

SimuFreeMark is a robust image watermarking framework without noise simulation that resists conventional signal processing and emerging semantic editing attacks. The method exploits the inherent stability of the low-frequency components of images, using a pretrained VAE to embed watermarks into the deep feature space of the low-frequency components, with no noise simulation at all during training. Experiments show that it outperforms existing state-of-the-art methods under various conventional and semantic attacks while maintaining higher visual quality.