Reproduce “Protego: User-Centric Pose-Invariant Privacy Protection Against Face Recognition-Induced Digital Footprint Exposure” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Ziling Wang, Shuya Yang, Jialin Lu, Ka-Ho Chow
Protego introduces a user-centric, pose-invariant privacy protection method that safeguards individuals against face recognition systems. It generates a unique 3D facial signature to perturb images, achieving robust protection even when query images are also protected, while maintaining visual naturalness across varying poses in photos and videos.
Key insights
- Maximizing the determinant of the Gram matrix for protected image features effectively makes FR models 'hypersensitive' to subtle variations, preventing protected-to-protected matches.
- Learning perturbations in a canonical 3D UV map space enables inherent pose-invariance, ensuring consistent visual quality and protection across diverse head orientations.
- A small, diverse ensemble of FR models, selected via focal diversity, is sufficient to train a user-specific PPT with strong black-box transferability to unseen FR systems.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.