da/sec scientific talk on Biometrics

Topic: Improved Feature Type Transformation for Deep Fuzzy Vaults

by Hans Geißner
D19/2.03a, January 15, 2025 (Wednesday), 12.00 noon

Keywords — Biometric Template Protection, Biometric Cryptosystem, Fuzzy Vault, Face Recognition, Fingerprint Recognition, Iris Recognition

Abstract

„Biometric technologies, such as face, fingerprint, and iris recognition, play a crucial role in secure identity management systems. Ensuring privacy-preserving storage and comparison of biometric data is essential, as biometric information, unlike passwords, cannot be replaced if compromised. Biometric cryptosystems (BCS) address this challenge by protecting biometric data, though they often result in reduced recognition performance. This performance gap arises primarily from two factors: feature type transformation, which is required to meet the specific input needs of BCS, and the comparison between the probe and the protected template, which typically involves error-correction.
In this work, we analyse the performance gap in the fuzzy vault scheme, one of the most prominent examples of biometric cryptosystems (BCS). Apart from feature type transformation, we identify the variability of feature sets during comparison, along with the resulting variability in correctable errors, as the main contributing factor to this gap. To address this issue, we propose a novel feature type transformation that converts deep-learning-based, fixed-length real-valued feature vectors into fixed-length integer-valued feature sets. A case study across various biometric traits demonstrates that this approach mitigates the performance differential during comparison.
The findings of this study provide a foundation to reduce the performance gap in BCS, enhancing their efficiency and practical applicability.“