Análise da efetividade de um sistema anti-spoofing para reconhecimento facial com a classificação de imagens estéreo através de uma rede neural convolucional

Abstract

The rise in the use of facial recognition technologies for authentication and security has hastened the need to develop effective methods to combat facial spoofing, known as face Presentation Attacks (PAs). This work addresses the challenge of Face Anti-Spoofing (FAS) and proposes a stereo vision-based approach to enhance the detection of 2D facial spoofing attacks. The research details the implementation of a Convolutional Neural Network (CNN) built from the Transfer Learning technique, which uses disparity images to add a depth dimension to the analysis. Throughout the work, the performance of the proposed model is also compared with an existing open-source model, the Silent-Face-Anti-Spoofing, providing insights into potential improvements. The results show that the exploration of disparity images can increase robustness and generalization compared to traditional approaches based solely on RGB images, achieving an accuracy of 100% in a controlled environment.