Image-Specific Protection Against Manipulation

The rapid development of Generative Models (GMs) in image synthesis poses challenges for accurately identifying manipulated images. Previous methods exploited a finite set of templates to proactively counter image manipulation, but this raises some concerns about potential vulnerabilities. A finite number of templates can provide a predictable exploit for malicious attackers, allow- ing them to reverse-engineer a template and evade detection by reapplying it to the modified image. This work presents a template-based detection system that integrates transformer- based models to generate personalised templates for each image. Our ap- proach enhances template protection while also improving the accuracy of manipulation detection. Furthermore, the generated template is employed to localize manipulated areas within an image, enabling the identification of specific regions that have been altered. Our solution achieves high detection accuracy and robust localization per- formance on both trained attributes and previously unseen attributes, while also outperforming existing solutions.