Identification of Image Spoofing Using Deep Convolution Neural Network

Abstract

Image plays a prominent role in forensic and biometric system authentication. Biometrics has emerged as a reliable security system solution. However, biometric applications become more prevalent, criminals are developing techniques to circumvent them by simulating physical or behavioural characteristics of legal users (image spoofing attacks). Face recognition systems are extremely vulnerable to such frauds because they can easily be fooled with common printed facial photographs, despite being a promising characteristic due to its universality, acceptability, and presence of cameras almost everywhere. Modern approaches based on convolutional neural networks (CNNs) produce good results in detecting face spoofing. However, these methods ignore the importance of learning deep local features from each facial region, despite the fact that face recognition has shown that each facial region has distinct visual characteristics that can be used to detect face spoofing. In this paper, we propose a novel deep CNN architecture for such tasks that is trained in two steps. At first, each component of the neural network learns features from a specific facial region. Following that, the entire model is fine-tuned on the entire set of images. The results show that such pre-training allows the CNN to learn different local spoofing cues, improving the final model’s performance and convergence speed and outperforming state-of-the-art approaches.