Utilizing Generative Adversarial Networks (GANs) for Finger Vein Biometrics’ Spoofing Attack Detection

Abstract

Finger vein modality is one of the emerging modalities since 2004 they are, unlike other traits such as face, hidden under the skin making the forgery of such modality a difficult task and making privacy more reachable. Finger vein holds a very high distinctive value between different individuals as well as between different fingers of a single individual. The usage of the finger vein is a very good step toward more security and less threats to a biometric system. Hence, this work utilized Generative Adversarial Networks (GANs) for finger vein spoofing attack mitigation with the aim of increasing the security of biometric authentication systems. GANs are deep learning networks proposed by Ian Goodfellow in 2014 and are one of many deep learning areas which acquired increasing attention in the recent years. One of the usages of GANs is the generation of synthetic biometric images to increase the classification accuracy percentage of a model. Different biometric factors, either physical or behavioral factors, could be used with GANs such as fingerprint However, finger vein biometrics are not as popular in this realm, and this could be seen from the literature published in this area. This work created a GAN model for finger vein biometric dataset to help in increasing the accuracy of detecting synthetic samples aiding in mitigation of finger vein biometric spoofing attack. This work also provided a deployable model with an accuracy of 99.83% for correctly classifying samples of real and generated synthetic data.