Paper 2024/100

FiveEyes: Cryptographic Biometric Authentication from the Iris

Abstract

Despite decades of effort, a stubborn chasm exists between the theory and practice of device-level biometric authentication. Deployed authentication algorithms rely on data that overtly leaks private information about the biometric; thus systems rely on externalized security measures such as trusted execution environments. The authentication algorithms have no cryptographic guarantees. This is particularly frustrating given the long line of research that has developed theoretical tools—known as fuzzy extractors—that enable secure, privacy preserving biometric authentication with public enrollment data (Dodis et al., SIAM Journal of Computing 2008). Unfortunately, the best known constructions either: 1. Assume that bits of biometrics are i.i.d. (or that all correlation is captured in pairs of features (Hine et al., TIFS 2023)), which is not true for the biometrics themselves or for features extracted using modern learning techniques, or 2. Only provide substantial true accept rates with an estimated security of $32$ bits for the iris (Simhadri et al., ISC 2019) and $45$ bits for the face (Zhang, Cui, and Yu, ePrint 2021/1559). This work introduces FiveEyes, an iris key derivation system powered by technical advances in both 1) feature extraction from the iris and 2) the fuzzy extractor used to secure authentication keys. FiveEyes’ feature extractor’s loss focuses on quality for key derivation. The fuzzy extractor builds on sample-then-lock (Canetti et al., Journal of Cryptology 2021). FiveEyes’ fuzzy extractor uses statistics of the produced features to sample non-uniformly, which significantly improves the security vs. true accept rate (TAR) tradeoff. Irises used to evaluate TAR and security are class disjoint from those used for training and collecting statistics. We state assumptions sufficient for security. We present various parameter regimes to highlight different TARs: 1. $65$ bits of security (equivalent to $87$ bits with a password) at $12$% TAR, and 2. $50$ bits of security (equivalent to $72$ bits with a password) at $45$% TAR. Applying known TAR (Davida et al., IEEE S&P 1998) amplification techniques additively boosts TAR by $30$% for the above settings.

Note: Substantial editorial work and more discussion of prior work.