ImageShield: A System for Image Tampering Detection, Localization, and Recovery
Published in , 1900
This paper introduces ImageShield, an end-to-end system for image authentication, tampering localization, and semantic content recovery that is robust against advanced AI-enabled attacks. ImageShield employs a novel dual-embedding strategy that separately embeds cryptographically protected authentication data and high-capacity recovery information using specialized neural networks. Extensive experiments and a human user study demonstrate that ImageShield significantly outperforms prior state-of-the-art methods in robustness, localization accuracy, and restored image quality under realistic attacks and transformations.