Masked Face Detection using Multi-Graph Convolutional Networks

In epidemic situations, as in the current COVID-19 pandemic, wearing face-masks is one of the most effective practices imposed to protect people lives. This will be one of the daily-life routines for a prolonged period, especially in public areas. Therefore, there is a demand to provide an efficient face detection method to help in dealing with such abnormal situations where people wearing masks are under monitoring. In this paper, we propose a deep learning model based on multi-graph convolutional networks (MGCN) to accurately detect people wearing masks. Unlike conventional GCNs, the proposed model includes many convolutional filters to produce multi-graph structure in which we use a 4D facet tensor as an input function and a convergence layer to capture multiple face expressions. This multi-graph version of spectral convolution transforms the extracted facial relief and generalizes image frequencies using graph rows and columns eigenvalues. The proposed architecture is simple yet efficient with several layers, including multi-graph convolutional, max pooling, dropout and softmax. We evaluate the performance of masked-faces detection on the publicly available real-world masked face dataset (RWMFD). The experimental results show an accuracy of 97.9%, which proves the efficiency of our proposed model in detecting people wearing facemasks.