Development of Passive Liveness Detection System Based on Deep Learning LivenessNet to Overcome Face Spoofing

This paper is focused on developing a real-time and robust passive liveness detection system to overcome a consequential challenge on existing face recognition systems, face spoofing. Our passive liveness detection system is a web-based system that is able to classify between real and 2D spoofing facial images. Our system consists of face detection and liveness detection process. For face detection, we conduct a comparative evaluation of pre-trained models to obtain the best-performing model. For liveness detection, we use LivenessNet, a lightweight Convolutional Neural Network (CNN) architecture. Using LivenessNet, we conduct an experimental evaluation of its hyperparameters value to obtain the most accurate model. The final result of our system can achieve 97.17% validation accuracy, 0% False Positive Rate, 5.68% False Negative Rate, and an inference time of 0.25 ms per frame.