Cancelable random masking with deep learning for secure and interpretable finger vein authentication

In the area of identity verification and authentication, biometrics has emerged as a reliable means of recognizing individuals based on their unique behavioral or physical characteristics. Finger vein authentication, with its robustness, resistance to spoofing, and stable patterns, has gained significant attention as a biometric modality. This paper introduces a novel framework that integrates Cancelable Random Masking (CRM) with a lightweight deep learning model for secure and interpretable finger vein authentication. The CRM technique transforms biometric templates using cryptographic random masks, ensuring cancelability, revocability, and privacy. These transformed templates are then processed by a convolutional neural network (CNN) designed to learn discriminative features directly from masked inputs without relying on handcrafted feature extraction. Our method enhances transparency by making the transformation process interpretable and provides strong security against template inversion and adversarial attacks. Results conducted on three publicly available databases demonstrate the proposed framework's superior performance in terms of accuracy, robustness, and resistance to spoofing and replay attacks. This is the first framework to integrate CRM within a deep learning model, satisfying all cancelable biometric criteria while enabling real-time, interpretable, and secure finger vein authentication.