@article{SU2025, 
author = {Xueping SU and Dandan SUN and Yunhong LI and Lina YAO and Yingxuan REN and Can WANG},
title = {Mask-occluded face recognition algorithm with integrated multi-view features},
year = {2025},
journal = {Journal of Northwest University (Natural Science Edition)},
volume = {55},
number = {2},
pages = {286-296},
keywords = {masked face recognition, BoTNet, attention augmentation},
url = {https://www.sciopen.com/article/10.16152/j.cnki.xdxbzr.2025-02-006},
doi = {10.16152/j.cnki.xdxbzr.2025-02-006},
abstract = {Wearing a mask is the most economical, effective and practical protective measure to prevent respiratory infectious diseases, but it also reduces the accuracy of face recognition. Therefore, a face recognition algorithm with mask occlusion combined with multi-view features is proposed. Firstly, BoTNet is used as the backbone feature extraction network to improve the recognition accuracy. Secondly, face attention augmentation model (FAAM) is introduced to generate a non-masked map of face areas (eyes, eyebrows and forehead) that are not covered by masks, and then the features of the masked map areas are accurately extracted to improve the performance of face recognition. In addition, the joint loss function Lface is designed to improve the convergence speed and performance of the model. Experiments are carried out on a public occlusion face data set with 500 000 face images of about 10 000 people. Compared with other algorithms, the recognition accuracy of this method is significantly improved, and it is 13.9% higher than that of the classic algorithm FaceNet.}
}