2020-047 – Secure Architecture for Biometric Authentication

Thisinvention provides a significantly more secure biometric authenticationapproach than currently existing approaches. This is achieved by a constructionthat allows concatenating any pre-existing neural network-based biometricauthentication mechanism with a fuzzy extractor, without altering the expectedaccuracy. The resulting architecture thus achieves the combined performance ofthe former, and security of the latter. Currentstate-of-the-art approaches to biometric authentication that rely on either score-basedor neural-network-based methods are vulnerable to attacks in which a user withroot (or admin) privileges can successfully log in as any of the other users.This vulnerability is similar to that incurred when storing passwords alongsideusernames on the authenticator machine, which has long been considered anunacceptable security practice. This invention overcomes vulnerability to suchattacks and renders considerably more security to biometric authentication systems. Aarushi Gupta-Sheth aarushi@ksu.edu 785-532-3907