Generating Style-based Palm Vein Synthetic Images for the Creation of Large-Scale Datasets

Individuals recognition through their biometric traits is an essential component of modern society. The recent literature includes several works based on palm vein recognition for individual identification, being a very active research field in the last five years. However, the publicly available datasets are very limited and have a small number of subjects, which limits to conduct scalability tests on large-scale databases. In this work, we propose a novel specific domain application for stylebased GAN architecture (StyleGAN) for generating synthetic palm vein images. Moreover, we present the largest dataset of palm vein images of the state-of-the-art at this moment, comprising of 10,000 subjects with 6 samples per each. Experimental results show that generated images look very realistic based on different metrics for measuring them against prior real datasets. The proposed dataset, called Synthetic Style-based Palm Vein Database (Synthetic-sPVDB), is publicly available on the website of our laboratory.