Personalized zero‐watermark algorithm based on non‐uniform weighted reconstruction and WGAN

Abstract The image protection of non‐fungible tokens based on zero‐watermark methods has received widespread attention. However, on the one hand, existing zero‐watermark methods are often limited to complex texture changes in the host image, and the features for constructing the zero‐watermark are vulnerable to geometric attacks. On the other hand, a single watermark image cannot adapt to the diverse usage scenarios of non‐fungible tokens. This paper proposes a robust personalized zero‐watermark scheme to address the challenges above. Firstly, the image regions suitable for constructing zero‐watermarks are highlighted by the non‐uniform weighted reconstruction, and the U‐Net is introduced for feature extraction against the geometric attacks. At the same time, the watermark images generated by generative models that are suitable for the usage scenario have achieved zero‐watermark addition and protection for non‐fungible token image content. The proposed method has been experimentally verified to have a certain degree of robustness in geometric and non‐geometric attacks.