A Multimodal Hand Vein Database and Recognition System

Biometric studies over the years have made the usage of physiological traits in human authentication technologies popular around the world. More recent studies in this field have given rise to more reliable, faster and user friendly security systems. One of the relatively new area of this field is hand vein biometrics where vascular patterns of hands are used for human recognition. This has some advantages over other physiological traits such as inherent spoof-proof attribute, lack of occlusion and noninvasiveness. Additionally, vein patterns can be captured from different parts of the hand, which could in turn be used in a multimodal system. Multimodal systems are generally preferred because they ensure a more robust and secure system compared to unimodal frameworks. In general, this study introduced a hand vein database named FYO with multiple hand vein datasets for palm, dorsal and wrist vein for the purpose of implementing hand vein multimodal biometric systems. Subsequently, feature descriptors such as Histogram of Oriented Gradients, Gabor filter and Binarized Statistical Image Features, and Convolutional Neural Network models such as AlexNet, VGG-16, VGG-19 and ResNet-50 are applied to show the efficiency of the proposed methodologies. Varieties of architectures for improving the robustness of hand vein recognition systems in both unimodal and multimodal forms are proposed in this study. Additionally, all experiments performed with the datasets acquired are similarly carried out on datasets from publicly available databases such as Badawi, Bosphorus, PUT, Tongji Contactless Palm Vein database and VERA, while the performances of the proposed systems are effectively compared to similar studies in the field