GAN-based Privacy-Conscious Data Augmentation with Finger-Vein Images

Abstract

The lack of sufficient data for evaluation and development is a major problem in biometrics. A novel GAN-based data-augmentation method for finger-vein authentication is proposed and evaluated in this study. Based on the GAN model structure, a subnetwork is added that lowers the similarity between the real data used for training and the fake data from the generator; the fake data looks remarkably similar to the real data, and the correlation between the real and fake data is lowered. Because the real data and fake data are different individuals, the privacy of a particular person is not considered when examining authentication technologies using only generated fake data. Moreover, the possibility of improving the authentication accuracy is confirmed by using both real data and generated fake data for training. The effectiveness of the proposed method is proved experimentally.