Masked Faces Recognition Using Deep Learning Models and the Structural Similarity Measure

Abstract

Wearing a mask is an important element to prevent infection with Corona disease. With the widespread adoption of face masks as a preventive measure, traditional face recognition systems encounter challenges in accurately identifying individuals. In this paper, we proposed a methodology that uses different deep learning models with pretrained weights on ImageNet to extract features and the structural similarity measure (SSIM) to recognize masked faces. Ten deep learning models were used, which are VGG16, VGG19, ReseNet50, Inception, InpectionV3, MobileNet, DenseNet201, NasNetMobile, EfficientNetB7, and InceptionResNet. The classification accuracy is used to evaluate the performance of each model. VGG-16, VGG-19, MobileNet and EfficientNetB7 gave the best results with an accuracy of 98\(\%\) which means that these methods are more appropriate for masked face recognition.