Effects of Face Image Degradation on Recognition with Vision Transformers: Review and Case Study

Vision Transformers have been recently introduced and achieved great success in the last two years beating convolutional neural networks (CNN) in many computer vision applications including Face Transformer for face recognition. However, one of the rising concerns is how sensitive these models are to various types of image degradations, especially when used for biometric authentication and person identification. This research aims at reviewing related work and presenting a case study to investigate the robustness of a fine-tuned version of Face Transformer under different pose, age, and gender variations for several degradations such as noise, blur, and resolution reduction. Several experiments have been conducted to train and test the model using two benchmark datasets (VGGFace2 and LFW) and benchmark its performance with FaceNet.