Effective Attention-Based Mechanism for Masked Face Recognition

Research on facial recognition has recently been flourishing, which led to the introduction of many robust methods. However, since worldwide outbreak COVID-19, people have had regularly wear masks, thus making existing face methods less reliable. Although normal are nearly complete, masked (MFR)—which refers recognizing identity an individual when a mask—remains most challenging topic in this area. To overcome difficulties involved MFR, novel deep learning method based convolutional block attention module (CBAM) and angular margin ArcFace loss is proposed. In method, CBAM integrated with neural networks (CNNs) extract input image feature maps, particularly region around eyes. Meanwhile, used as training function optimize embedding enhance discriminative for MFR. Because insufficient availability images model training, study data augmentation generate from common dataset. The proposed was evaluated using well-known version LFW, AgeDB-30, CFP-FP, real mask MFR2 verification datasets. A variety experiments confirmed that offers improvements MFR compared current state-of-the-art