Robust Techniques for Evaluating Biometric Cryptographic Key Generators
Lucas Ballard
Abstract
Humans are unable to generate and remember strong secrets, and thus
have difficulty managing cryptographic keys. To address this problem,
numerous proposals have been suggested to enable people to reliably
generate high-entropy cryptographic keys from measurements of their
physiology or behavior. Typically, evaluators argue that these
Biometric Cryptographic Key Generators (BKGs) achieve some notion of
security, for example, that the biometric input resists forgery, or
that the keys have high entropy. Unfortunately, despite these
arguments, many BKGs succumb to attacks in practice.
The goal of this work is to understand why typical security arguments
fail to identify practical attacks. We revisit the security
requirements of BKGs and show that common arguments overlook practical
subtleties. We provide examples of such oversights by examining three
general classes of adversaries. First, we study the impact of humans
who can replicate other users' biometrics with high accuracy, and
demonstrate why typical evaluation techniques fail to identify these
forgers. Second, we explore Generative techniques that combine
information about a target user with population statistics to create
forgeries. We show that these forgeries can subvert BKGs with high
likelihood. Third, we propose an algorithm that probabilistically
enumerates the key space of a BKG to find a target user's key. We
analyze two BKGs, and show that for each, our algorithm has at least a
15% chance of predicting ostensibly 40-bit keys on its first
guess. Our exposition brings to the forefront practical ways of
thinking about BKG security, and provides a framework for evaluators
to study BKGs with adversarial techniques.
Finally, we present Randomized Biometric Templates (RBTs), a BKG that
outputs keys with at least as much entropy as keys derived from
passwords. RBTs extract entropy not only from biometric inputs, but
also from a novel source: how these inputs are measured. Analysis
with our strengthened evaluation techniques show that for some users,
RBTs result in dramatically stronger keys. In our experiments, 40%
of the users were able to generate keys that were at least 2
30
times stronger than keys derived from passwords alone.
[
pdf]
Readers