Enhancing Face Recognition in Masked Environments: A Deep Neural Network Investigation with Attention Modules

This research comprehensively investigates improving face recognition performance in masked environments. This study investigates the deep neural networks (DNN) models with Channel Attention Module (CBAM) on masked face recognition tasks using the VGGFace2 dataset. The study leverages the unique feature of residual connections in ResN et-50 to overcome vanishing gradients, thereby enhancing representation learning. The DNN with Attention aiming to improve face recognition performance in masked environments reveals significant accuracy improvements for Unmasked-Masked (UM) and All-Pair comparisons compared to less deep networks. Surprisingly, the results demonstrate that CBAM performs poorly, behaving similarly to a random classifier. The research demonstrates the efficacy of DNN s like ResN et-50 in challenging tasks. It underlines its potential to facilitate enhanced face recognition in masked scenarios. The research sheds light on the limitations of CBAM in handling masked face recognition scenarios. It highlights the need for guided attention explicitly targeting the region above masks and around the eyes and the nose bridge. Addressing these shortcomings paves the way for more effective and accurate masked face recognition models.