Privacy preservation of industrial access system users with cancellable face recognition based on inverse techniques

To strengthen security, we now use biometric technologies instead of passwords or tokens in a variety of authentication applications. Biometrics depend on parts of the human body. They have an advantage of not being lost. This paper seeks to safeguard biometrics within recognition systems by inducing intended distortions in biometric traits prior to saving them in a database for preventing hackers from accessing the actual biometrics. We can still use the biometrics, even if they are stolen or hacked in this situation by changing the distorted versions. Two approaches for face biometric distortion are presented in this paper, namely a regularization approach and a Linear Minimum Mean Square Error (LMMSE) approach. The regularization strategy begins by adding noise to the original faces, and then applying a regularized inverse on the noisy face images to create face images with amplified fixed noise patterns. The process of noise amplification is really a masking process that helps in the generation of cancellable templates. These versions of the face images can be used as templates for verification that can be cancelled and replaced with new templates if required. The other approach is cancellable face recognition based on the LMMSE algorithm, which begins with noise addition to the original faces and then application of LMMSE algorithm to the face images to obtain new templates with magnified fixed noise patterns. Finally, we compare between the two approaches in the cancellable face recognition application.