Random permutation‐based linear regression for cancelable biometrics

Abstract

AbstractThe security of biometric data in biometric‐based authentication systems is a significant concern. Cancellable biometrics aim to generate templates that can be replaced by new templates if compromised. We propose a new approach for generating cancellable biometric templates based on linear regression with random permutation. Our approach generates a virtual image for every biometric image by applying linear regression. In the next step, the cancellable biometric template is produced by randomly permuting each virtual image depending on a key assigned to each individual. If the template is compromised, it can be cancelled, and a new template can be generated by altering the key. Our method has shown superior performance compared to existing random permutation‐based methods in terms of authentication accuracy across six databases, encompassing face, iris, and ear, even when dealing with low‐resolution images. It performed well on challenging databases like UBIRIS and Georgia Tech, demonstrating the robustness of the proposed approach.