摘要：Wearing face masks in public spaces has becomean essential step to prevent the spread of COVID-19. Thisstep poses some challenges to conventional face recognitiondue to several reasons: 1) the absence of large real-worldmasked face recognition dataset, and 2) the loss of some visualcues due to the occlusion by the face masks. To addressthese challenges, this paper presents a real-world masked facerecognition dataset that consists of 80500 masked face images of161 subjects, referred to as MFRD-80K dataset. Every subjectcontributes 500 masked face images, which are then partitionedinto 60:20:20 for train, validation and test. Subsequently, weconduct some benchmark studies to evaluate the performanceof the existing face recognition and classification methodson the MFRD-80K dataset. The methods include k-NearestNeighbour, Multinomial Logistic Regression, Support VectorMachines, Random Forest, Multilayer Perceptron and Convolutional Neural Networks. Since the parameter settings affectthe performance of each method, a grid search is performed todetermine the optimal parameter settings. The empirical resultsdemonstrate that Convolutional Neural Network achieves thehighest test accuracy of 97.16% on MFRD-80K dataset.
关键词：masked face; masked face recognition; masked face recognition dataset; machine learning; classification; CNN
