WearMask in COVID-19: Identification of Wearing Facemask Based on Using CNN Model and Pre-trained CNN Models

Abstract

The rapid worldwide spread of Coronavirus Disease 2019 (COVID-19) has led to global outbreak. To prevent the further spreading of this virus thus ensuring people’s health safety, wearing face mask is mandatory in all public places strictly imposed by the respective governments of the countries all over the world. To assess the situation whether people are wearing mask, automatic identification of face mask in public places is a timely solution and that’s where our ‘WearMask in COVID-19’ project primarily stands for. As part of our project, to identify whether people wear mask or not, this article analyzes and compares the effectiveness of using CNN architecture built from scratch with pre-trained ResNet50 and MobileNet-v2 models. These models are trained on a combined face mask image dataset having mostly low resolution simulated masked face images and some high resolution real masked face images. The models work through three steps: image pre-processing, image feature extraction and classification using the models above, and model evaluation. Experimental results show that an in-house CNN structure that we have developed from scratch performs better than the ResNet50 model and is comparable to the performance of the MobileNet-v2 model.