Invisible Image Watermarks Are Provably Removable Using Generative AI
Invisible watermarks safeguard images' copyright by embedding hidden messages only detectable by owners. They also prevent people from misusing images, especially those generated by AI models. We propose a family of regeneration attacks to remove these invisible watermarks. The proposed attack method first adds random noise to an image to destroy the watermark and then reconstructs the image. This approach is flexible and can be instantiated with many existing image-denoising algorithms and pre-trained generative models such as diffusion models. Through formal proofs and empirical results, we show that all invisible watermarks are vulnerable to the proposed attack. For a particularly resilient watermark, RivaGAN, regeneration attacks remove 93-99% of the invisible watermarks while the baseline attacks remove no more than 3%. However, if we do not require the watermarked image to look the same as the original one, watermarks that keep the image semantically similar can be an alternative defense against our attack. Our finding underscores the need for a shift in research/industry emphasis from invisible watermarks to semantically similar ones. Code is available at https://github.com/XuandongZhao/WatermarkAttacker.
PDF Abstract
MS COCO
DiffusionDB
