Robust masked face recognition using CBAM and FFT for global and local feature enhancement

Face recognition (FR) methods have made remarkable progress with the advent of deep learning technologies. Many stateof- the-art FR models are now based on Convolutional Neural Networks (CNNs). However, significant challenges persist, including variations in pose, illumination, and age. Another critical issue is occluded face recognition, which has gained attention, particularly during the COVID-19 pandemic, as mask-wearing has become widespread to reduce infection risks. Additionally, various accessories worn in daily life further complicate recognition tasks by introducing diverse occlusions. To address these challenges, we propose a model that integrates CBAM (Convolutional Block Attention Module) and FFT (Fast Fourier Transform). This approach leverages CBAM to refine spatial and channel attention while using FFT to extract frequency-domain features from the visible parts of the face. Experimental results demonstrate the efficiency of our model, achieving accuracies of 99.86% on the LFW dataset, 99.57% on the M-LFW dataset, and 96.70% on the RMFD dataset.