Securing Iris Recognition with Fully Homomorphic Encryption: Dimensionality Reduction in Iris Codes.

Biometric recognition systems are fundamental to modern identity management and access control. However, the security and privacy of these systems are severely compromised when adversaries gain access to stored biometric templates, given the immutable link between individuals and their biometric traits. This study presents a benchmark analysis of the trade-offs between dimensionality and system performance in the context of securing iris biometric templates using Fully Homomorphic Encryption (FHE). We examine the impact of Principal Component Analysis (PCA) as a dimensionality reduction technique to enable encrypted-domain computation while maintaining recognition accuracy and reducing computational overhead. In lieu of introducing a new dimensionality reduction technique, this study rigorously benchmarks the applicability of PCA in balancing computational efficiency and recognition accuracy for biometric systems operating under Fully Homomorphic Encryption (FHE) constraints. We evaluate our method on various iris databases, including CASIA-V1, CASIA-V3, UBATH, and IITD. Our approach achieves a 100% True Acceptance Rate (TAR) on CASIA-V1, CASIA-V3, and UBATH and a 99.38% TAR on the IITD database at a 0.1% False Accept Rate (FAR), with a feature dimensionality of 250. This work advances iris recognition security by combining privacy measures with dimensionality reduction for improved authentication accuracy.