FROM: Resilient Face Recognition under Masked and Occluded Environment

Recent advancements in deep convolutional neural networks have significantly impacted face recognition. However, existing models designed for face recognition and verification struggle with occluded face images, common in real-world scenarios. These models face challenges in scaling their architecture to generalize and identify individuals, especially with masked or partially obscured faces. The difficulties in face recognition, particularly with occluded faces, arise from limited datasets with mask-occluded faces and a lack of designs addressing feature corruption due to occlusions. The variety of masking patterns in facial photos, from fully masked faces to partially masked and unmasked ones, adds complexity to face recognition algorithms. To address these challenges, this research work proposes FROM (Face Recognition with Occlusion Masks), This innovative framework leverages dynamically learned masks and incorporates Deep Convolutional Neural Networks explicitly tailored for facial recognition applications. CNN is chosen on testing basis of different size on the basis of overfitting and underfitting. The objective is to effectively detect and handle occluded or distorted facial features caused by masks. The proposed methodology is rigorously evaluated using a substantial dataset comprising occluded and masked face images. Rigorous evaluation on datasets, including LFW, Megaface Challenge 1, RMF2, AR dataset, Indian Masked Faces in the Wild (IMFW), and simulated occluded/masked datasets, underscores its generalization to face recognition tasks.