Occluded Face Recognition Based on Attention Mechanism and Damaged Feature Masking

Face recognition has made great achievements due to the continuous development of deep learning techniques and the arising of large-scale face datasets. However, these are only applicable to face recognition without occlusion. Face recognition encounters great obstacles under constrained conditions with occlusion. To handle this problem, this paper proposes a novel occluded face recognition method based on the bottleneck attention module (BAM) and the mask generator. BAM is embedded in backbone network to extract more distinguishable features, while the mask generator is designed to clean the corrupted feature, so that face recognition can be realized under the constraint conditions of occlusion. Experimental results on the public datasets indicate the proposed method is superior to the comparison methods.