Provenance Detection for AI-Generated Images: Combining Perceptual Hashing, Homomorphic Encryption, and AI Detection Models

2025-03-14

AI Summary

Traditional watermarking methods for provenance tracing of AI-generated sensitive images suffer from vulnerability to social-media-induced distortions (e.g., filters, compression, screenshots) and are increasingly susceptible to removal or forgery due to the open availability of generative models. Method: This paper proposes a triple-security framework for AI-image provenance tracing, comprising: (1) DinoHash—a robust perceptual hashing model leveraging DINOV2 features; (2) a privacy-preserving matching mechanism integrating multi-party fully homomorphic encryption (MP-FHE); and (3) unified support for both registered-database lookup and unknown-content detection. Contribution/Results: Experiments demonstrate that DinoHash achieves a 12% improvement in average bit accuracy and significantly enhanced TPR/FPR trade-offs; AI-generated image classification accuracy rises by 25%. The framework jointly strengthens robustness against real-world distortions, cryptographic privacy guarantees, and generalization across diverse content and deployment scenarios.

Technology Category

• Data Mining & Knowledge Management: Representing, Reasoning, and Using Provenance, Trust
• Computer Vision: Adversarial Attacks & Robustness
• Machine Learning: Privacy

Application Category

• Security and Privacy: Data transparency and provenance
• Web Mining and Content Analysis: Web data provenance, reliability, and authenticity
• Systems and Infrastructure for Web, Mobile and WoT: Applied ML and AI for Web-based mobile applications

Abstract

As AI-generated sensitive images become more prevalent, identifying their source is crucial for distinguishing them from real images. Conventional image watermarking methods are vulnerable to common transformations like filters, lossy compression, and screenshots, often applied during social media sharing. Watermarks can also be faked or removed if models are open-sourced or leaked since images can be rewatermarked. We have developed a three-part framework for secure, transformation-resilient AI content provenance detection, to address these limitations. We develop an adversarially robust state-of-the-art perceptual hashing model, DinoHash, derived from DINOV2, which is robust to common transformations like filters, compression, and crops. Additionally, we integrate a Multi-Party Fully Homomorphic Encryption~(MP-FHE) scheme into our proposed framework to ensure the protection of both user queries and registry privacy. Furthermore, we improve previous work on AI-generated media detection. This approach is useful in cases where the content is absent from our registry. DinoHash significantly improves average bit accuracy by 12% over state-of-the-art watermarking and perceptual hashing methods while maintaining superior true positive rate (TPR) and false positive rate (FPR) tradeoffs across various transformations. Our AI-generated media detection results show a 25% improvement in classification accuracy on commonly used real-world AI image generators over existing algorithms. By combining perceptual hashing, MP-FHE, and an AI content detection model, our proposed framework provides better robustness and privacy compared to previous work.

Problem

• Detect source of AI-generated images securely.
• Improve robustness against image transformations.
• Enhance privacy with homomorphic encryption.

Innovation

• DinoHash: robust perceptual hashing for transformations
• Multi-Party Fully Homomorphic Encryption for privacy
• Improved AI-generated media detection accuracy