MSML: Enhancing Occlusion-Robustness by Multi-Scale Segmentation-Based Mask Learning for Face Recognition

In unconstrained scenarios, face recognition remains challenging, particularly when faces are occluded. Existing methods generalize poorly due to the distribution distortion induced by unpredictable occlusions. To tackle this problem, we propose a hierarchical segmentation-based mask learning strategy for recognition, enhancing occlusion-robustness integrating segmentation representations of occlusion into in latent space. We present novel multi-scale (MSML) network, which consists branch (FRB), an (OSB), and elaborate feature masking (FM) operators. With guidance learned OSB, FM operators can generate masks eliminate mistaken responses introduced occlusions purify contaminated facial features at multiple layers. way, proposed MSML network effectively identify remove from levels aggregate visible areas. Experiments on verification under synthetic or realistic demonstrate effectiveness our method compared state-of-the-art methods.