Contactless fingerprinting is an emerging approach to biometric authentication that allows users to scan their fingerprints without touching a scanner. Due to the limited amount of contactless fingerprint data available and the security risks associated with sharing real individuals' fingerprints, it is valuable to explore methods of generating synthetic data that can be used in place of - or in conjunction with - real data to develop and evaluate contactless fingerprinting systems. In this paper, we present and evaluate synthetic fingerphotos generated using StyleGAN2-ADA and StyleGAN3, existing image generation architectures. We evaluate the realism, privacy preservation, and variety of the synthetic fingerphotos by comparing their biometric feature statistics to those of real fingerphotos, computing match scores between real and synthetic fingerphotos, and computing match scores between different synthetic fingerphotos. This paper provides a quantitative comparison point for future evaluations of synthetic fingerphotos. The evaluation code is made available at https://github.com/cmillerlynch/fingerphoto-gan.
Figures & tables
Figure 1: Synthetic fingerphotos generated using StyleGAN2-ADA (top) and StyleGAN3 (bottom). From left to right, the generated images represent index, middle, ring, and little fingers from left hands.
Figure 2: The image processing pipelines that were compared. enhance-CLAHE was selected to process fingerphotos for compatibility with contact-based evaluation and matching tools.
Figure 3: DET curves achieved by VeriFinger on fingerphotos processed using each image processing pipeline.
Real
StyleGAN2-ADA
StyleGAN3
Mean
SD
Mean
SD
Mean
SD
Mean Minutiae Quality
63.70
4.18
62.27
3.90
61.97
3.96
Ridge Ending Minutiae Count
40.38
10.62
40.94
10.25
38.67
9.96
Bifurcation Minutiae Count
45.71
11.26
49.63
11.75
45.06
10.75
Percentage of Bifurcation Minutiae
53.13%
6.52%
54.82%
5.79%
53.87%
6.27%
NFIQ2 Score [ 26 ]
29.80
21.85
30.28
21.82
27.32
22.21
Table 1: Comparison of biometric features of the real fingerphotos in the training dataset and the synthetic fingerphotos generated by the StyleGAN2-ADA and StyleGAN3 models, as measured by VeriFinger and NFIQ2. The enhance-CLAHE image processing pipeline was applied to all fingerphotos prior to feature extraction.
Figure 4: VeriFinger match score distributions for privacy evaluation. The more similar the real and synthetic distributions, the more distinct the synthetic fingerprint patterns are from the real ones in the training dataset.
Figure 5: VeriFinger match score distributions for variety evaluation. The more similar the real and synthetic distributions, the better the synthetic fingerphotos model the variety of fingerprint patterns in the training dataset.
Figure 6: A visualiztion that compares the false accept rates for synthetic fingerphotos at a given threshold to the expected false accept rate for real fingerphotos. The closer each curve is to the diagonal line, the better the corresponding model achieves the corresponding goal. This visualization method provides more information about matching results and the tails of the distributions than histograms.