A Comparative Study of Anti-Spoofing Detection Based on Deep Learning
My Abdelouahed Sabri, El Khoukhi Hasnae, Ennouni Assia, Aarab Abdellah · 2023
In the developing field of biometrics, technology is improving our ability to automatically recognize persons and protect access. The face is an excellent option for an authentication application because most devices can be equipped with a camera. Fraudsters can falsify biometric data by using, for example, photos of faces instead of the real face. So, it is essential to confirm that the biometric is that of a real, live person. Face anti-spoofing is one of the methods to combat this type of fraud. The objective of this work is to propose a comparative study between 5 powerful deep learning architectures for anti-spoofing namely, DenseNet201, DenseNet169, VGG16, MiniVGG, InceptionV3, and ResNet50. We used in this study the ROSE-Youtu Face Liveness Detection dataset. Experimental results show the effectiveness of the use of deep learning and especially the MiniVGG architecture in terms of maximum accuracy and traditional performance evaluation measures.