Lightweight face mask detection and face recognition

Abstract

The Covid-19 pandemic has abruptly changed our daily life: face mask and wearing a mask have become popular, and are occasionally obligatory for some public activities. As a result, several methods were proposed to address the problem of face mask detection and/or face recognition. While these methods showed promising results, they often require high computational resource due to sophisticated deep learning models. In this paper, we will propose lightweight methods to detect face masks and recognize human faces simultaneously. Our proposed methods are based mainly on Random Forest and MobileNetV2, and are trained and tested with our own dataset collected from Vietnamese faces. The experiment shows that Random Forest can address face mask detection and face recognition with high accuracy. Moreover, a combination of Random Forest and MobileNetV2 can still improve the performance of detection and recognition while keeping the method at relatively low computational complexity.