SECURE BIOMETRIC AUTHENTICATION

Abstract

In this thesis, we present a method for secure biometric authentication based on facial fea- tures. Our secure biometric authentication is designed against insider and direct attacks. It prevents a malicious (insider) server from cheating the system to learn the user’s plaintext biometric data. It also reduces the chances of unauthorized users from accessing the system through conducting (direct) attacks on the input sensors. Our method leverages Π-nets to obtain an embedding of the facial image in a high-dimensional vector-space. The similarity between this biohash (the probe) and pre-registered biometrics (the template) is computed using a two-level homomorphic encryption scheme. The verification phase of the proposed scheme is processed in 0.52s with an overall communication cost of 99 KB. Our proto- col is largely orthogonal to the biohash used for authentication and can be extended using modalities other than RGB images, e.g., to improve robustness against spoofing attacks. We therefore also present a survey into face data sets with a taxonomy focusing on aspects such as multi-modality, ethnicity, size, and applications of the data sets. This will allow readers of the thesis to gauge publications related to their work quickly and based on multiple indicators.

Date of Award	2024
Original language	American English
Awarding Institution	HBKU College of Science and Engineering