Comparative Analysis of AlexNet, Resnet-50, and Inception-V3 Models on Masked Face Recognition

Abstract

Since the outbreak of the coronavirus pandemic in December 2019, there has been increased interest in developing better facial recognition systems. This stems from the need to protect everyone from the spread of the virus. However, the measures taken to prevent the spread of the virus pose a challenge to security and surveillance systems as existing systems are unable to match faces with masks more efficiently. For this study, a custom dataset was generated due to the unavailability of a large face dataset for masked face recognition, and the existing datasets focused on Caucasians (white race faces) while Aethiopians (black race faces) were neglected. In this study, a comparative analysis was conducted between the AlexNet, ResNet-50, and Inception-V3 models to recognize faces with surgical masks, fabric masks, and N95 masks. The results of the study showed that the CNN models achieve excellent recognition accuracy for masked and unmasked faces. Analysis of the models' performance showed that the AlexNet model achieved 95.7%, ResNet-50 achieved 97.5%, and Inception-V3 also achieved 95.5%. From the study, ResNet-50 performed better than Inception-V3 and AlexNet models in recognizing masked faces.