Synthetic Occluded Masked Face Recognition using Convolutional Neural Networks
Ian James H. Recto, Madhavi Devaraj · 2022
Wearing a face mask is the norm during the COVID–19 pandemic and is advised for enclosed spaces such as workplaces. In face recognition, a face mask is considered a partial occlusion which degrades recognition accuracy. This study focuses on the occlusion factor by a variety of face mask designs. This study aims to mitigate the impact of face masks as an occlusion on a face recognition system. We superimposed a synthetic face mask and black occlusions on top of the face images (FI). FaceNet, a deep convolutional neural network, was used to extract facial embeddings. The faces were classified using a support vector machine. We experimented with different scenarios by using different training sets and testing sets, contains differing mask designs. It achieved a performance of recognizing occluded lower FI with an average accuracy rate of 98.93% in a controlled environment.