This research paper addresses the challenge of recognizing faces masked due to the post-pandemic scenario. The researchers created a new dataset by selecting images from the Labelled Faces in the Wild (LFW) Dataset and simulating face masks on them. They compared four different models including Eigenface, Inception ResNet(v1), ResNet50, and VGG16 for masked face recognition. They achieved the best test accuracy of 95% on 50 identity MFR using data augmentation strategy and fine-tuning the models on their new dataset. The study also proposed the concept of Exponential Margin to implement real-world applications.
Publication date: 14 Nov 2023
Project Page: https://arxiv.org/abs/2311.07475
Paper: https://arxiv.org/pdf/2311.07475
