A Comparison of Deep Learning Models for Periocular Region-Based Authentication

Facial recognition has been widely used within our society, whether it is through identification, classification, or authentication. However, due to the COVID-19 pandemic, facial recognition must adapt to individuals wearing masks to determine if the person is authenticated or not. Facial recognition uses main biometric features, such as eyes, iris, nose, skin tone, and mouth, to accomplish its tasks given. It is possible to identify and authenticate a person using just the periocular region of their face, especially if they are masked. In this paper, we apply deep learning models for authentication based on periocular biometrics of face images with masks. We compared the performance of a basic Convolutional Neural Network (CNN), VGG16, GoogleNet, LeNet, and AlexNet on full face and periocular biometrics-based authentication on masked face images. The individuals will be classified as either authentic or not for a system. We found that all the models performed better on the periocular region of masked face images than on the full-face images. We also found that GoogleNet had the best performance among the models.