Secure Biometric Template Protection Using Qr-Embedded Visual Secret Sharing For Face Authentication

Abstract

Face authentication systems offer user-friendly and accurate user verification; nevertheless, the storage and management of biometric templates pose considerable security and privacy risks since the compromised biometric data cannot be restored or replaced with another password. The proposed framework is a QR-Embedded Visual Secret Sharing (QR-VSS) approach, which is aimed at providing biometric template protection through the integration of key-dependent template transformation, Visual Secret Sharing (VSS), and QR-based representation. According to the proposed concept, the facial features are acquired using ArcFace, protected with key-dependent transformation, separated into different shares, and embedded into QR codes. When it comes to authentication, the shares stored in QRs are recovered and reconstructed to receive the protected representation for verification. The performance of the proposed solution is estimated in terms of simulated biometric performance, security, QR robustness, computational complexity, overhead storage, and ablation tests. Simulated experiments prove that the proposed QR-VSS framework ensures the EER of 2.71%, TAR of 96.72%, and ROC-AUC of 0.987 with clear separation of similarity scores between impostors and genuine cases. Regarding security, the single-share reconstruction proves to be 0.0%, the original-template reconstruction equals 0.8%, 99.8% of reconstruction attempts were carried out with proper authorization, and 0.0% attacks were successful. Moreover, the relatively high QR decoding and authentication performance can be observed in case of mild-to-moderate degradation, although, increased computational and storage resources are considered the major drawbacks of the approach. The obtained results demonstrate the promising nature of the QR-VSS integration in order to combine face authentication with template protection. However, more experiments are needed.