Several attempts have been made recently to generate synthetic fingerprint data. This has become necessary after legal changes in Europe and some US states in order to allow and continue long-term developments in the field of fingerprint biometrics. Apart from utilizing traditional methods (often based on Gabor filters), deep convolutional neural networks are widely used to generate synthetic fingerprint samples. The current study aims at comparing several publicly available synthetic fingerprint datasets with several datasets that consist of imprints taken from real people. To enable a comparison, first a detailed description of these datasets is carried out. Secondly, an available 4-level protocol is used, which is supposed to show similarities and/or differences between real and synthetic fingerprint samples in terms of quality assessment and non-mated as well as mated comparison scores’ behavior. Furthermore, a new synthetic FP dataset composed of 50k samples is created and made publicly available in the course of this study.

2D contactless to contact-based fingerprint (FP) comparison is a challenging task due to different types of distortion introduced during the capturing process. While contact-based FPs typically exhibit a wide range of elastic distortions, perspective distortions pose a problem in contactless FP imagery. In this work, we investigate three simple parametric warping models for contactless fingerprints — circular, elliptical and bidirectional warping — and show that these models can be used to improve the interoperability between the two modalities by simulating unfolding of a generic 3D model. Additionally, we employ score fusion as a technique to enhance the comparison performance in scenarios where multiple contactless FPs of the same finger are available. Using the simple circular warping, we have been able to decrease the Equal Error Rate (EER) from 1.79% to 0.78% and 1.82% to 1.31% on our dataset, respectively.

We study the impact of highly correlated image content on the estimated photo response non-uniformity (PRNU) of a sensor unit and its impact on the sensor identification performance. Based on eight publicly available finger vein datasets, we show formally and experimentally that the nature of finger vein imagery can cause the estimated PRNU to be biased by image content and lead to a fairly bad PRNU estimate. Such bias can cause a false increase in sensor identification performance depending on the dataset composition. Our results indicate that independent of the biometric modality, examining the quality of the estimated PRNU is essential before the sensor identification performance can be claimed to be good.