MaskOut: A Data Augmentation Method for Masked Face Recognition

Weiqiu Wang, Zhicheng Zhao, Hongyuan Zhang, Zhaohui Wang, Fei Su

Abstract: Deep learning methods have achieved great performances in face recognition. However, the performances of deep learning methods deteriorate in case of wearing a mask. Recently, due to the world-wide COVID-19 pandemic, masked face recognition attracts more attention. It is non-trivial and urgent to improve the performances in masked face recognition. In this work, a simple and effective data augmentation method, named MaskOut, is proposed. MaskOut replaces a random region below the nose of a face with a random mask template to mask out original face features. Our method is computing and memory efficient and convenient to combine with other methods. The experimental results show that the performances in masked face recognition are improved by a large margin with MaskOut. Besides, we construct a real-life masked face dataset, named MCPRL-Mask, to evaluate the performance of masked face recognition models.