Masked Face Recognition Using Transfer Learning Approaches

Face recognition is a subfield of artificial intelligence science that uses different biometric features of human faces to recognize people. Face recognition systems are widely used due to highly achieved recognition accuracy reaching almost 99.73%. However, there are several challenges and limitations existing especially for real-time face recognitions. Some of these challenges occurred during the global outbreak of the COVID-19 pandemic in December 2019. In this paper, a transfer learning approach proposes to overcome the challenges of masked face recognition by using a pre-trained deep convolution neural network (DCNN) model. The modified version of the CASIA dataset with generated synthetic masks dataset is used for evaluating the proposed model. The results of evaluation metrics were promises where the overall model accuracy rate reaches 93%. Thus, the proposed model showed the ability of DCNN model to recognize masked face images efficiently.