Multi-Model Face Liveness Detection Via Gaze Detection and Convolutional Neural Networks
Derek Apgar, Muhammad Rizwan Abid · 2022
Spoofing attacks pose a significant threat to facial recognition systems. There are many types of attacks that can be carried out, ranging from photo attacks to mask attacks. Facial recognition systems must defend themselves against such attacks with cutting-edge anti-spoofing countermeasures. The current Covid-19 outbreak has increased the need for facial recognition, exacerbating the issue of secure facial recognition systems. Many existing liveness detection methods fail to defend against all types of attacks because they exclude mask attacks. This is a flaw in many methods, and because this is the only method of defense, there is no redundancy. In this paper, we propose a multi-model liveness detection model that employs gaze detection and Convolutional Neural Networks (CNN) to provide increased defense against multiple types of spoofing attacks while also increasing robustness. We were able to achieve 100 percent accuracy against photo and video attacks using gaze detection, and 95.28 percent accuracy against photos, videos, and mask attacks using a CNN. The multi-model system achieved an aggregate accuracy of 99.85 percent.