Multi-instance Finger Vein-based Authentication with Secured Templates

Abstract

The illegitimate access to biometric templates is one of the major issues to be handled for authentication systems. In this work, we propose to use two instances of finger vein images which inherits the advantages of a robust multi-modal biometric authentication system without needing different sensors. Two local texture feature extraction methods are experimented on standard finger-vein datasets. Fused discriminating features with reduced dimension lowers down the system computational cost. A cancelable template protection scheme as Gaussian Random Projection based Index-of-Max is then applied for embedding privacy and security to the templates. Foremost template protection properties like revocability, non-invertibility and unlinkability are observed to be significantly obeyed by the proposed system with considerable authentication performance. Recognition performance of the proposed methods are compared with some previously executed finger vein systems and observed to be less complex and overperforming on the combined basis of authentication and template protection. Thus, the proposed system utilizes multiple evidence and provides a balanced performance with respect to authentication, template protection and computational cost.