SynCoLFinGer: Synthetic Contactless Fingerprint Generator

1

SynCoLFinGer:Synthetic Contactless Fingerprint Generator

Jannis Priesnitz∗, Christian Rathgeb∗, Nicolas Buchmann†, Christoph Busch∗∗Hochschule Darmstadt, Germany†Freie Universitat Berlin, Germany

[email protected]

Abstract—We present the first method for synthetic generationof contactless fingerprint images, referred to as SynCoLFinGer.To this end, the constituent components of contactless finger-print images regarding capturing, subject characteristics, andenvironmental influences are modeled and applied to a synthet-ically generated ridge pattern using the SFinGe algorithm. Theproposed method is able to generate different synthetic samplescorresponding to a single finger and it can be parameterizedto generate contactless fingerprint images of various qualitylevels. The resemblance of the synthetically generated contactlessfingerprints to real fingerprints is confirmed by evaluating bio-metric sample quality using an adapted NFIQ 2.0 algorithm andbiometric utility using a state-of-the-art contactless fingerprintrecognition system.

Index Terms—Mobile Biometrics, Fingerprint Recognition,Contactless Fingerprint, Synthetic Generation

I. INTRODUCTION

Biometric system development and evaluation can benefitfrom the use of synthetic biometric data [1]. On the onehand, synthetic biometric data can be used for algorithmtraining where the acquisition of real samples is cost andtime intensive. On the other hand, biometric systems canbe evaluated with easily manageable synthetic data, which isgenerally not limited by data protection regulations.

In the scientific literature, two main approaches to syntheticbiometric data generation can be distinguished:

• Modeling: biometric signals or features may be mod-eled using hand-crafted methods specifically designedfor biometric characteristics. Such models usually requireknowledge about statistical properties of biometric dataand have been proposed for various biometric character-istics, e.g. contact-based fingerprints [2] or finger vein[3].

• Deep learning: Generative Adversarial Network (GAN)[4] architectures have been found to be suitable to gener-ate realistic image data including biometric samples, e.g.face [5] or iris images [6].

Focusing on fingerprint recognition [7], some approachesto synthetically generate contact-based fingerprints have beenproposed [2]. Pioneer work in this field has been doneby Cappelli et al. [8] who proposed Synthetic FingerprintGenerator (SFinGe). Starting from the positions of cores anddeltas, the SFinGe algorithm exploits a mathematical flowmodel to generate a consistent directional map. Subsequently,a density map is created on the basis of some heuristiccriteria and the ridge-line pattern. Furthermore, the minutiae

(a) real (b) SynCoLFinGer

Fig. 1. Examples of a real and synthetic contactless fingerprint sample:(a) contactless sample from the ISPFDv2 database (cropped), (b) contactlesssynthetic sample generated by the proposed method.

are created through a space-variant linear filtering; the outputis a near-binary clear fingerprint image. Finally, specific noiseis added to produce a realistic gray-scale representation ofthe fingerprint. The latter step allows for the generation ofmultiple mated samples of a single finger. Up until now,continuous improvements have been applied to the SFinGealgorithm1. Similarly, further approaches to model contact-based fingerprints have been proposed by different researchlaboratories, e.g. in [9]. More recently, GAN approacheshave been employed for synthetically generating contact-based fingerprints, e.g. in [10], [11]. In such deep learning-based approaches, the generation of several mated samplesrepresents a great challenge. Additionally, GANs may generateunrealistic ridge patterns in terms of blurred and ridge-valleytransitions as well as frictions in the ridges.

Besides contact-based fingerprint recognition, the develop-ment of contactless systems represents a growing research area[12], [13]. Contactless fingerprint capturing devices overcomesome problems of contact-based ones like latent fingerprintsof previous users (ghost fingerprints) or hygienic concerns.Since the contactless fingerprint capturing process is differentcompared to the contact-based process, the resulting biometricsamples are also different. Contactless fingerprints do notexhibit an elastic deformation caused by pressing the fingeron a sensor plate. Moreover, many contactless fingerprint

1Biometric System Laboratory – University of Bologna: http://biolab.csr.unibo.it/

arX

iv:2

110.

0914

4v1

[cs

.CV

] 1

8 O

ct 2

021

2

Fig. 2. Overview of SynCoLFinGer. First step: simulation of a contactless capturing, second step: generation of subject-related characteristics, third step:simulation of environmental influences.

capturing devices are based on commodity equipment likesmartphones, which produce RGB images.

Despite the growing interest in contactless fingerprint recog-nition, so far no approaches to synthetic contactless fingerprintgeneration have been published. In this work, we propose thefirst (to the best of the authors’ knowledge) method for gener-ating synthetic contactless fingerprints named SynCoLFinGer.Starting from a synthetically generated ridge pattern (using theSFinGe software [2]), the constituent components of contact-less fingerprint images with respect to capturing conditions,subject characteristics, as well as environmental influencesare modeled to generate synthetic contactless fingerprint sam-ples. By varying certain changeable parameters biometricvariance and variation in sample quality are simulated whichenables the generation of various mated synthetic samplescorresponding to a single finger. The generated contactlessfingerprint samples exhibit good visual resemblance to realimages, see Fig. 1 for an illustration. Furthermore, the real-ism of the synthetically generated contactless fingerprints isverified by estimating sample quality employing an adaptedNFIQ 2.0 algorithm and biometric utility using a state-of-the-art contactless fingerprint recognition system. To facilitatereproducible research and future experiments, the source codeof SynCoLFinGer will be made publicly available.

This work is organized as follows: the proposed syntheticcontactless fingerprint generation method is described in detailin Sect. II. Experimental evaluations are presented in Sect. III.Finally, conclusions are drawn in Sect. IV.

II. PROPOSED METHOD

The proposed SynCoLFinGer method aims at generatinga segmented fingertip. For this reason, the detection andsegmentation of a fingertip and the rotation into an uprightposition is considered as solved processing steps. This is areasonable assumption considering the detection and segmen-tation performance of state-of-the-art algorithms, e.g. [14].Our method utilizes the well-known contact-based fingerprintgeneration algorithm SFinGe [2]. For our method we use anintermediate result of the SFinGe algorithm which generatesonly the ridge line characteristic of a 500 dpi live scannedfingerprint but applies no further processing to the sample.

We apply three main steps to the SFinGe ridge pattern (seeFig. 2):

(a) SFinGe ridge pattern (b) Contactless ridgepattern

Fig. 3. Visualization of the fingerprint deformation with deformation vectorsshown in red.

1) Deformation of the ridge line pattern to simulate acontactless capturing scenario (see Sect. II-A).

2) Generation of subject-related characteristics (seeSect. II-B).

3) Simulation of environmental influences (see Sect. II-C).This workflow ensures that SynCoLFinGer is able to robustlycreate a set of mated samples. Further, the generation of afingerprint image is highly controllable and easy to extend.

A. Contactless Capturing

The simulation of a contactless captured fingerprint includesthree processing steps (see Figs. 3, 4):

• Fingerprint deformation• Rotation• Ridge line thinning and blurringThe fingerprint deformation transforms the SFinGe ridge

pattern into a contactless fingerprint ridge pattern. Our de-formation method implements a warp transformation whichdistorts the pixel along a deformation vector within a certainrange around the vector, see Fig. 3 for an example. Thedistortion vectors are equally arranged around the contact-based ridge pattern and are facing towards the fingerprint

3

(a) Contactless ridgepattern

(b) Rotation(rolling)

(c) Ridge line thin-ning and blur

Fig. 4. Processing of the deformed ridge pattern to simulate effects relatedto the capturing process.

template. The size of the contact-based ridge pattern definesthe amount of deformation vectors, the length of the vectorsand the distance between the ridge pattern and the vectors.The length of the distortion vectors in the upper area of thefingerprint is higher compared to the left and right border ofthe finger to represent the natural shape of a human finger.

Slight rotations along the rolling axis occur because of thelevel of freedom during the finger presentation. SynCoLFinGersimulates these rotations by re-using the deformation algo-rithm. Here, the deformation vectors are facing horizontally inthe opposite direction of the rolling. This leads to a stretchingof one side and a compression on the other side of thefingerprint template which simulates a rolling of the finger (seeFig. 4 b). Our method simulates a rotation of up to approx.seven degrees of rolling.

Subsequently, the ridge lines are thinned by an erosionalgorithm to achieve a realistic ridge line appearance in thecontactless sample. Further, the image is blurred. A Gaussianblur is applied to the whole sample to smooth the transitionsbetween ridges and valleys. The depth-of-field of a contactlesscapturing device is usually only a few millimeters whichcauses an out-of-focus blur at the border of the finger [15].This property is simulated by a blurring of the finger borderregions (see Fig. 4 c).

B. Subject-related Characteristics

The generation of realistic contactless fingerprint images re-quires the simulation of certain subject-specific characteristics.These characteristics refer to (see Fig. 5):

• Regions of low contrast• Skin color• Skin tone variationMost contactless fingerprint images show regions of low

contrast. These are mainly caused by worn out fingers or der-matological issues. To simulate such properties, local regionsof the fingerprint image are blurred (Fig. 5 a).

To acquire a set of fingertip skin colors we manually anal-ysed a subset of the ISPFDv2 database [16]. More precisely,we randomly selected 25 subjects. For each subject, ten datapoints within the fingerprint region are averaged to a skin colorrepresentation (c.f. Fig. 5 b).

(a) Regions of lowcontrast

(b) Finger shape andskin tone

(c) Skin tone variation

Fig. 5. Simulation of subject-related characteristics.

The color of a fingertip is not perfectly homogeneousthroughout the whole surface. Skin impurities or damagescause variations of the skin tone. This is simulated by a localbrightness and color variations. Here, a simple filter slightlyvaries the brightness and color of the finger area (Fig. 5 c).

C. Environmental Influences

Numerous environmental influences need to be considered,such as variations in the capturing scenario, technical pre-conditions of the capturing device, or particles on the fingersurface. SynCoLFinGer implements some of the most relevantenvironmental challenges (c.f. Fig. 6):

• Brightness and color variation• Shadow• Dirt• Illumination• Camera noise

The skin color impression in contactless capturing scenariosalso depends on the environment. Here, the illumination ofthe surrounding area and the background of the image havean influence on the skin tone appearance. This is simulatedthrough a general variation of the skin tone.

Another aspect which is closely related to illumination isthe shadow on the finger area. SynCoLFinGer simulates theshadow as darker regions at the border of a finger image. Here,a shadow mask implements the illumination of the fingertipwith a single light source (Fig. 6 (a)).

Ridge line inversion refers to the switching of parts of theridge line structure from bright to dark because of illuminationand reflection properties of the skin. In brighter regions theridge lines appear brighter than the fingertip whereas indark regions the ridge lines appear darker (Fig. 6 (b)). Thefingerprint sample is adjusted accordingly to simulate thisproperty. For this purpose, the shadow mask is used to havean indication of bright and dark regions in the finger area.

Camera noise is directly related to illumination and thecapturing device. The adding of camera noise simulates acapturing device which operates under challenging light con-ditions (Fig. 6 (c)). During the simulation, the illuminationinfluences the amount of noise generated, whereas darkerpreconditions lead to more noise.

4

(a) Shadow (b) Inversion ofridge lines

(c) Camera noise (d) Dirt (e) Finger image

Fig. 6. Simulation of environmental influences like shadow and camera noise.

(a) High quality (b) Medium quality (c) Low quality

Fig. 7. Three synthetically generated samples of a single finger with differentquality presets: (a) high, (b) medium (b) and (c) low quality.

Dirt refers to temporary particles on the finger surface whichmay influence the recognition performance. SynCoLFinGerimplements the presence of dirt by particles which are ran-domly distributed over the finger region (Fig. 6 (d)).

The different masks have different properties in termsof persistence. Some properties like the skin color or thefingerprint characteristic itself are highly persistent, whereasothers like illumination may change more drastically. Someproperties like regions of low contrast caused by worn-outfingers are semi-persistent and change after days or weeks.For this reason semi-persistent masks are also equipped withan alteration factor which may represent the time between twocapturing sessions.

III. EVALUATION

In this section we evaluate the suitability of our proposedmethod in a biometric recognition scenario. First, we visually

inspect the generated database and discuss its properties.Second, we evaluate the sample quality and the biometricrecognition performance. Here, we used the workflow pro-posed in one of our previous works [17] to process thefingerprint images and to estimate sample quality in termsof NFIQ 2.0 scores and biometric performance in terms ofthe standardised metrics False Match Rate (FMR), False Non-Match Rate (FNMR) [18], and Equal Error Rate (EER).

A. Synthetic Database Generation

To evaluate our algorithm, we generate a database of syn-thetic contactless fingerprint samples. To show the achievablevariation three subsets are created:

• High: sample which are captured under good conditions,i.e. samples of high quality.

• Medium: samples which are captured under good con-ditions but where the fingertip is slightly rotated or thefingerprint has a low contrast.

• Low: challenging samples with huge regions of lowcontrast, strong rotation exhibiting huge variation in termsof illumination, i.e. samples of low quality.

In addition, changeable properties described in the previoussection are varied randomly to obtain several mated contact-less fingerprint samples from a single ridge pattern. Fig. 7illustrates an example of the above described quality variationsfor mated samples. The generated database consists of 1, 000different fingerprints (SFinGe samples). For each category andfingerprint six samples are generated which results in a totalamount of 18, 000 samples.

B. Experimental Results

Requirements for the assessment of synthetic fingerprintgeneration have recently been defined by Makrushin et al.[19]. Stipulated properties include data anonymity, sufficientlyhigh image resolution, diversity and uniqueness. These re-quirements are fulfilled through the use of the SFinGe algo-rithm. The requirement of controllable generation is met by theconceptual design of SynCoLFinGer described in the previoussection.

Realistic appearance represents another key requirement forsynthetic fingerprint images [19]. A visual inspection of thegenerated database reveals that the SynCoLFinGer samplesshow a good level of realism. Examples of syntheticallygenerated images are depicted in Fig. 8. Important aspects likea realistic impression of the ridge pattern, regions of low con-trast, skin tone, and environmental influences are representedin the generated samples. Moreover, the biometric variationbetween mated samples realistically simulates the capturingsessions under different conditions, c.f. Fig. 7. Because ofthe controllable design of the algorithm, generated samplesshow realistic ridge patterns, in contrast to some recentlyproposed GAN-based approaches for generating contact-basedfingerprints. Nevertheless, it should be noted that the generatedimages can still be distinguished from real ones (c.f. Fig. 8).Note that this is also caused by the missing background of thegenerated fingerprint images.

5

(a) Real

(b) Synthetic – high quality

(c) Synthetic – medium quality

(d) Synthetic – low quality

Fig. 8. Examples of real and synthetic contactless fingerprint images generated by SynCoLFinGer: (a) real images, synthetic images of (b) high, (b) mediumand (c) low quality.

In the next step, we investigate the most important require-ment on whether the generated fingerprint images reflect thebasic characteristics of real contactless fingerprint data. To thisend, we evaluate biometric sample quality using an adaptedNFIQ 2.0 algorithm and biometric utility using a state-of-the-art contactless fingerprint recognition system.

Average obtained NFIQ 2.0 scores for synthetically gener-ated fingerprint images at different quality level are depictedin Fig. 9 and listed in Table I along with a comparison toNFIQ 2.0 scores achieved on popular real datasets. It can beobserved that the sample quality of the SynCoLFinGer imagesis within the range of real contactless fingerprint databases orslightly better. As expected, the three categories are showingdifferent biometric accuracy. The high-quality samples havean average NFIQ 2.0 above 40. For medium-quality sam-ples, the average NFIQ 2.0 score is dropping by approx. 5points. The lowest NFIQ 2.0 score of approximately 28 isobtained on the low-quality samples. The standard deviationof NFIQ 2.0 scores is in a similar range compared to the real

databases. This shows that SynCoLFinGer is able to producea similar amount of variance in terms of sample quality. Wesee that the high-quality database has the lowest standarddeviation, whereas the low-quality category has the highest.This represents the parameter settings in our configurationfiles and can be adapted easily. Real contactless databasescontain samples with different ridge line frequency, especiallyif they are captured in unconstrained setups e.g. ”ISPFDv1natural” and ”Own unconstrained”). Databases captured inmore constrained scenarios show much higher NFIQ 2.0 scores(e.g. ”PolyU” or ”Own constrained”). Especially, the distancebetween the capturing device and the finger needs to be keptconstant to approximate a fingerprint size which is equivalentto 500 dpi contact-based fingerprints (which is expected bythe NFIQ 2.0 algorithm). The SFinGe templates on which ourmethod is based simulate a 500 dpi capturing device whichis also favored by NFIQ 2.0. Even if the samples are highlydeformed, the size of a fingerprint and the ridge line frequencyin the central area of the fingerprint contribute to a high

6

high

medium low

white

natural

session1

session2

constrained

unconstrained

0

10

20

30

40

50

60avg.NFIQ

2.0Score

synthetic ISPFDv1 PolyU CB2CL own Database

Fig. 9. Overview of NFIQ 2.0 scores obtained from different contactlessdatabases: synthetic fingerprint image generated by SynCoLFinGer (blue),the contactless sub-databases of the ISPFDv1 database (green), the contactlesssub-databases of the PolyU database (yellow) and a database acquired withour own capturing device (red).

TABLE IAVERAGE NFIQ 2.0 SCORES AND BIOMETRIC PERFORMANCE OBTAINEDFROM THE SYNTHETICALLY GENERATED DATABASE COMPARED TO REAL

CONTACTLESS FINGERPRINT DATABASES.

DB Subset NFIQ 2.0 score EER (%)

SynCoLFinGehigh 40.68 (±7.93) 1.13

medium 34.92 (±10.04) 3.07low 27.80 (±12.66) 6.44

PolyU [23] session 1 47.71 (±10.86) 3.91session 2 47.08 (±13.21) 3.17

ISPFDv1 [16] white 17.73 (±8.40) 30.48natural 12.16 (±8.52) 31.38

Own database [17] constrained 44.80 (±12.36) 11.16unconstrained 36.13 (±14.05) 29.75

NFIQ 2.0 score. It is noteworthy that NFIQ 2.0 was foundto be suboptimal for biometric sample quality assessment oncontactless fingerprint data and, hence, its application requiresspecific pre-processing [20]. Alternatively, quality assessmentalgorithms specifically designed for contactless fingerprintdata could be employed, e.g. [21], [22].

The evaluation in terms of biometric recognition perfor-mance shows that SynCoLFinGer samples are working wellin an open-source contactless fingerprint recognition workflow(see Table I). As expected, the three categories are showingdifferent biometric accuracy. The high, medium, and low-quality samples reveal EERs of 1.13%, 3.07%, and 6.44%,respectively. This wanted behaviour is directly caused bythe parameter settings in the configuration presets and isobservable across different decision threshold configurationsas shown in the DET plots of Fig. 10. It is observable thatespecially databases acquired in unconstrained condition (e.g.”ISPFDv1 natural” and ”Own unconstrained”) have muchhigher EERs compared to the SynCoLFinGer samples. A pos-sible explanation for this is a slight variance of the segmentedfingertip, which results in a shifted minutiae map. Because theSynCoLFinGer samples are not suffering from this variancethe recognition accuracy is expected to be much higher.

Overall, the equal error rates and NFIQ 2.0 scores showthat SynCoLFinGer is able to generate synthetic contactless

0.001 0.01 0.1 1 5 20 400.001

0.01

0.1

1

5

20

40

False Match Rate (in %)

False

Non

-Match

Rate(in%) high medium low

Fig. 10. Detection Error Trade-off (DET) curves for synthetically generatedcontactless fingerprint images at different sample quality levels.

fingerprint images which closely reflect the characteristics ofreal contactless fingerprint data in terms of sample quality andbiometric utility.

IV. CONCLUSION

This work introduced SynCoLFinGer, a synthetic contact-less fingerprint generator. To the best of our knowledge,SynCoLFinGer represents the first approach towards the gener-ation of synthetic fingerprint images. In the proposed method,properties of contactless fingerprints with respect to the captur-ing process, subject-related characteristics, and environmentalinfluences were simulated. It was shown that the proposedmethod is able to generate synthetic mated samples at variousquality levels. In experiments, it was demonstrated that thesynthetically generated contactless fingerprints reflect the basiccharacteristics of real contactless fingerprint data in terms ofbiometric sample quality and biometric utility.

Our proposed approach is expected to open new possibilitiesfor further contributions where different avenues for futureresearch could be considered, e.g.:

• Improvement of appearance: further developments areneeded to improve the amount of visual appearance ofthe generated images. Although we achieved a certainbiometric variance further variations may be simulated,especially the low-quality samples could be designed tobe even more challenging. Moreover, the set of appliedoperations could be extended to additionally simulatefurther properties like scratches or dermatological issues.Additional developments will be facilitated by providingthe implementation of SynCoLFinGer as open-sourceimplementation.

• Synthetic samples for interoperability studies: since Syn-CoLFinGer utilizes SFinGe ridge patterns, it is possibleto generate synthetic contactless and contact-based fin-gerprint samples of single fingers. This could be of highinterest for researchers working on the interoperabilitybetween contact-based and contactless capturing devices.

7

ACKNOWLEDGEMENTS

This research work has been funded by the German FederalMinistry of Education and Research and the Hessian Ministryof Higher Education, Research, Science and the Arts withintheir joint support of the National Research Center for AppliedCybersecurity ATHENE.

REFERENCES

[1] D. J. Buettner, “Biometric sample synthesis,” in Encyclopedia of Bio-metrics, S. Z. Li and A. Jain, Eds. Springer, 2009, pp. 116–122.

[2] R. Cappelli, “Sfinge,” in Encyclopedia of Biometrics, S. Z. Li andA. Jain, Eds. Springer, 2009, pp. 1169–1176.

[3] F. Hillerstrom, A. Kumar, and R. Veldhuis, “Generating and analyzingsynthetic finger vein images,” in Int’l Conf. of the Biometrics SpecialInterest Group (BIOSIG’14), 2014, pp. 1–9.

[4] I. Goodfellow, J. P.-A. M. Mirza, B. Xu, D. Warde-Farley, S. Ozair,A. Courville, and Y. Bengio, “Generative adversarial nets,” in Advancesin Neural Information Processing Systems (NIPS), 2014.

[5] T. Karras, S. Laine, and T. Aila, “A style-based generator architecturefor generative adversarial networks,” in Conf. on Computer Vision andPattern Recognition (CVPR’19), 2019, pp. 4396–4405.

[6] S. Yadav, C. Chen, and A. Ross, “Synthesizing iris images using rasganwith application in presentation attack detection,” in Conf. on ComputerVision and Pattern Recognition Workshops (CVPRW’19), 2019, pp.2422–2430.

[7] D. Maltoni, D. Maio, A. Jain, and S. Prabhakar, Handbook of FingerprintRecognition, 1st ed. Springer-Verlag, 2009.

[8] R. Cappelli, A. Erol, D. Maio, and D. Maltoni, “Synthetic fingerprint-image generation,” in Int’l Conf. on Pattern Recognition. (ICPR’00),2000, pp. 471–474 vol.3.

[9] Q. Zhao, A. K. Jain, N. G. Paulter, and M. Taylor, “Fingerprintimage synthesis based on statistical feature models,” in Int’l Conf. onBiometrics: Theory, Applications and Systems (BTAS’12), 2012, pp. 23–30.

[10] M. S. Riazi, S. M. Chavoshian, and F. Koushanfar, “Synfi: Automaticsynthetic fingerprint generation,” arXiv preprint arXiv:2002.08900,2020.

[11] M. A.-N. I. Fahim and H. Y. Jung, “A lightweight gan network for largescale fingerprint generation,” IEEE Access, vol. 8, pp. 92 918–92 928,2020.

[12] R. D. Labati, A. Genovese, V. Piuri, and F. Scotti, “Touchless fingerprintbiometrics: A survey on 2d and 3d technologies,” Journal of InternetTechnology, vol. 15, no. 3, p. 328, 2014.

[13] J. Priesnitz, C. Rathgeb, N. Buchmann, C. Busch, and M. Margraf, “Anoverview of touchless 2d fingerprint recognition,” EURASIP Journal onImage and Video Processing, vol. 2021, no. 1, pp. 1–28, 2021.

[14] J. Priesnitz, C. Rathgeb, N. Buchmann, and C. Busch, “Deep learning-based semantic segmentation for touchless fingerprint recognition,” inInternational Conference on Pattern Recognition. Springer, 2021, pp.154–168.

[15] R. Raghavendra, C. Busch, and B. Yang, “Scaling-robust fingerprintverification with smartphone camera in real-life scenarios,” in 2013 IEEESixth International Conference on Biometrics: Theory, Applications andSystems (BTAS). IEEE, 2013, pp. 1–8.

[16] A. Malhotra, A. Sankaran, A. Mittal, M. Vatsa, and R. Singh, “Finger-photo authentication using smartphone camera captured under varyingenvironmental conditions,” in Human Recognition in Unconstrained En-vironments, M. D. Marsico, M. Nappi, and H. Proenca, Eds. AcademicPress, 2017, pp. 119 – 144.

[17] J. Priesnitz, R. Huesmann, C. Rathgeb, N. Buchmann, and C. Busch,“Mobile touchless fingerprint recognition: Implementation, performanceand usability aspects,” arXiv preprint arXiv:2103.03038, 2021.

[18] ISO/IEC JTC1 SC37 Biometrics, ISO/IEC 19795-1:2021. InformationTechnology – Biometric Performance Testing and Reporting – Part 1:Principles and Framework, International Organization for Standardiza-tion, June 2021.

[19] A. Makrushin, C. Kauba, S. Kirchgasser, S. Seidlitz, C. Kraetzer, A. Uhl,and J. Dittmann, “General requirements on synthetic fingerprint imagesfor biometric authentication and forensic investigations,” in Proceedingsof the 2021 ACM Workshop on Information Hiding and MultimediaSecurity, ser. IH&MMSec ’21. ACM, 2021, p. 93–104.

[20] J. Priesnitz, C. Rathgeb, N. Buchmann, and C. Busch, “Touchless Fin-gerprint Sample Quality: Prerequisites for the Applicability of NFIQ2.0,”in Proc. Intl. Conf. of the Biometrics Special Interest Group (BIOSIG),2020.

[21] R. D. Labati, V. Piuri, and F. Scotti, “Neural-based quality measurementof fingerprint images in contactless biometric systems,” in The 2010International Joint Conference on Neural Networks (IJCNN), 2010, pp.1–8.

[22] G. Li, B. Yang, M. A. Olsen, and C. Busch, “Quality assessment forfingerprints collected by smartphone cameras,” in 2013 IEEE Conferenceon Computer Vision and Pattern Recognition Workshops, 2013, pp. 146–153.

[23] C. Lin and A. Kumar, “Matching contactless and contact-based conven-tional fingerprint images for biometrics identification,” IEEE Transac-tions on Image Processing, vol. 27, no. 4, pp. 2008–2021, 2018.