Large-scale palm vein recognition on synthetic datasets

During the last decade, palm vein recognition has gained the attention of the research community on biometric systems since it presents high security. However, its applications on massive individuals identification are limited mainly because publicly available datasets have very small numbers of subjects. In this context, synthetic datasets are helpful to evaluate the performance and scalability of biometric systems on large-scale databases. Thus, the present work evaluates CNN-based models on two self-created synthetic datasets. For this purpose, we implemented two end-to-end CNN architectures based on AlexNet and Resnet32. Besides, we created two large-scale synthetic datasets by using a Sty1eGAN-based model and a specific method based on biological transport networks, which are comprised of 10,000 and 2,000 individuals, respectively. The generated datasets are the largest of the state-of-the-art and were validated by using different quantitative metrics in order to measure their visual quality and realism comparing to real images. The experimental results show the applicability and quality of the proposed synthetic databases in order to evaluate the efficiency and scalability of palm vein recognition methods.