Efficient Optical Deep Learning Model Based on Cycle-GAN for Secure Face Recognition

Abstract

In the past few decades, biometrics authentication has been a popular practice. In place of the conventional passwords, biometric traits are used to secure access to computers. However, if these traits are used in their original form, they can only be used once. Therefore, it is perfect to use a biometric template that can be altered if it is attacked. This can be achieved through cancelable biometrics. Improving the security and privacy of biometric authentication is the goal of cancelable biometrics. It is effective to construct cancelable biometric templates using deep learning. To ensure safe biometric information in confirmation schemes, an efficient encryption algorithm is developed in this research. The face is the biometric taken into consideration in this paper. We utilize the Cycle-Generative Adversarial Networks (Cycle GANs), and a confusion Baker map. The Cycle GAN has a generator and a discriminator. The generator alters the original image to create an unreal one, and the discriminator tries to determine whether it came from the generator or not. A set of biometric faces were used for training of this model. The generated templates have been utilized to evaluate the cancelable biometric system. Moreover, the noise effect has been taken into consideration. Several metrics have been considered to test the system including Equal Error Rate (EER), False Accept Rate (FAR), False Reject Rate (FRR), and Area under the Receiver Operator Characteristics curve (AROC).