Binary biometric feedback can still enable template reconstruction
Original title: How Much Feedback Is Needed to Break Template Protection? The Geometric Vulnerability of Biometric Authentication
A geometric attack using only accept or reject, plus a facial diffusion prior, cuts queries by about five times.
The paper studies reconstruction of protected biometric templates when systems return only an accept or reject decision, rather than similarity scores. It presents a geometric attack that uses that binary feedback, plus distribution-aware methods for score-releasing systems that the authors say need about five times fewer queries than recent score-based attacks. Adding a pretrained facial diffusion prior to the binary attack also reduces queries by about five times, the authors report. They argue the authentication decision itself leaks geometric information regardless of the template-protection mechanism.
Unconfirmed: Query reductions and perfect reconstruction are the authors' reported results, not independently confirmed here.
- Prior attacks used released similarity scores; this targets binary decisions.
- Score-based methods reportedly need about five times fewer queries than recent attacks.
- A facial diffusion prior also cuts binary-attack queries by about five times.
- Authors claim reconstruction up to numerical precision, independent of template protection.
Impact: reconstruct protected biometric templates from accept or reject feedback
From the original article
1 min read at huggingface.co
Biometric authentication systems increasingly use privacy-preserving mechanisms to protect sensitive templates. However, the authentication interface itself may still leak information that enables reconstruction attacks. Prior attacks exploit released similarity scores, whereas more advanced systems reveal only the binary accept/reject decision.
By Eliron Rahimi, Margarita Osadchy, Orr Dunkelman · huggingface.co
A short excerpt; the article and its images belong to the publisher. Summary, key facts and assessment above are AI-written from the article and can be wrong.