Cancelable Biometric Template Generation using Eigenfeature Regularization

Cancelable biometrics addresses biometric data’s privacy and security concerns. We present two new cancelable biometrics template generation methods: RP-RegSt and RP-RegSb. The suggested approaches use random permutations and regularized eigenfeature extraction to generate cancelable biometrics templates, which can be reissued if compromised. We also show that applying random permutation to generate cancelable biometric templates enhances recognition accuracy. The suggested approaches are tested on six publicly accessible databases: three iris databases (UBIRIS.v1, CASIA-V1, and IITD Iris), two face databases (Georgia Tech and AT&T), and one ear database (IITD Ear). The superiority of the proposed methods is demonstrated by comparing them to three state-of-the-art random permutation-based cancelable biometric template generation techniques. The suggested approaches’ performance on challenging databases with substantial biometric image variation, such as Georgia Tech and UBIRIS, shows their robustness and efficacy. The privacy concern is addressed as the templates are irreversible (non-invertible) and immune to imposter attacks, while brute force analysis shows the templates are secure. The templates satisfy the diversity (unlinkability) and revocability properties.