Faster Region-based Convolutional Neural Network for Mask Face Detection

We present two-stage detection approach, Faster Region-based Convolutional Network, Faster R-CNN for masked face detection. In this face detection, we localize the face object in the image and classify the face object based on mask occlusion on the face. There are three classes: a face without any mask, the second is an incorrectly masked face, and the third is a face with correctly masking. The first stage of the detection is finding the candidate regions of the target object. This stage uses the Region Proposal Network (RPN). Then, the candidate regions are fed into the last pooling layer of the Faster R-CNN identified as the ROI Pooling layer. The model is trained using MAFA and AFLW datasets. The mean Average Precision of trained model for all classes is 0.73, with the highest accuracy is obtained by the face without mask class, and the lowest accuracy is the incorrectly masked face class.