Empirical Performance Analysis of Deep Convolutional Neural Networks Architectures for Face Liveness Detection

Abstract

In the realm of biometrics, face recognition (F.R.) is one of the most exciting new developments. In the past decade, computer vision and artificial intelligence advancements have improved face recognition systems by several orders of magnitude. Many attacks can be launched against these systems, such as the low-cost and low-effort Presentation attacks. Face liveness detection is gaining momentum in research. With the advent of deep learning, observing the performance of pre-trained DCNN Architectures for Face Liveness Detection will be interesting. The paper proposes an Empirical Performance analysis of eight pre-trained DCNN Architectures allies VGG16, VGG19, ResNet50, InceptionResNetV2, MobileNetV2, DenseNet201, InceptionV3, Xception. The experimentation carried out on the NUAA dataset (120000 images) & and the Replay attack dataset has shown that the better performance is given by DenseNet201, closely followed by MobileNetV2. This study offers significant contributions to the understanding of the dynamic nature of face recognition technologies.