Identifying Fake and Real Images by Using Masked Face Periocular Region

In this paper, we focus on the face spoofing of masked images to determine whether a masked person is real or fake. We developed a dataset of spoofed masked images generated using the DALL.E 2 tool, performed the ROI extraction using the CNN-DLib detector, and extracted the features using BoVw-sift. We applied deep learning and machine learning algorithms. XGBoost and Xception achieved the highest accuracy of 92% and 94% to determine whether the images were real or fake. The approach was tested on the real-world masked face recognition dataset (RMFRD). This shows that periocular information can predict whether the masked image is real or fake.