GUC100 Multi-scanner Fingerprint Database for In-House (Semi-Public) Performance and Interoperability Evaluation Davrondzhon Gafurov, Patrick Bours, Bian Yang and Christoph Busch Norwegian Information Security Lab Gjøvik University College P.O. Box 191, 2802 Gjøvik, Norway {Firstname.Lastname}@hig.no Abstract—In this paper, we describe the GUC100 multi- scanner fingerprint database that has been created for inde- pendent and in-house (semi-public) performance and interop- erability testing of third party algorithms. The GUC100 was collected by using six different fingerprint scanners (TST, L-1, Cross Match, PreciseBiometrics, Lumidigm and Sagem). Over several months fingerprint images of all 10 fingers from 100 subjects on all 6 scanners were acquired. In total, GUC100 con- tains almost 72.000 fingerprint images. The GUC100 database enables to evaluate various performances and interoperability settings by taking into account different influencing factors such as fingerprint scanner, image quality and so on. The GUC100 data set is freely available to other researchers and practitioners provided that they conduct their testing in the premises of the Gjøvik University College in Norway, or alternatively submit their algorithms (in compiled form) to run on GUC100 by researchers in Gjøvik. We applied a commercial fingerprint verification algorithm on GUC100 and the reported results indicate that GUC100 is a challenging database. I. I NTRODUCTION The two important aspects in performance evaluation of fingerprint recognition algorithms (and other biometrics in general) are the availability of independent databases and desirably testing bodies too. The advantages of such databases and third party testing bodies are that firstly it allows more direct and unbiased benchmarking of different algorithms and secondly it increases a trustworthiness of the performance report, since developers do not have a direct access to the database for tuning algorithm’s parameters to adapt to the database. However, creating and distributing large scale databases publicly is not an easy task because of the involved costs and time as well as jurisdictional limits. Due to the nature of the collected data (i.e. human physiology), creation and distribution of the large scale biometric databases raise privacy concerns and may not be permitted by data protection authorities in some countries (especially in Europe). Even if data collection is permitted, usually it is requested to destroy collected data after the completion of the project, e.g. as in [1]. Nevertheless, in the biometric community several finger- print databases were established for research purposes [2], [3], [4], [5], [6]. Previously public databases were provided by NIST which consists of thousands of fingerprint images [2]. However, these images are rolled ones i.e. scanned from inked tenprint paper card. In addition in the context of the MINEX project NIST composed a large scale fingerprint data set for in-house testing of algorithms [7]. The database series FVC200x [3], [4], [5], [6] were designed for the Fingerprint Verification Competition (FVC) where several competing algorithms were tested on them. There are also some multi-modal databases, where the fingerprint is col- lected as one of the modalities [8]. This paper describes a multi-scanner fingerprint database, which has been created for independent and in-house perfor- mance and interoperability testing. In the rest of the paper, we will refer to this database as GUC100 1 . In exploitation of this database we follow - due to privacy regulations in Norway - the principal of ”If the data cannot travel to the algorithm then the algorithm shall travel to the data”. This means that copies of GUC100 database cannot be distributed to other parties outside of GUC campus. However, algorithm developers are free to visit GUC and perform training and testing of their algorithm in its premises, or submit their (binary code) fingerprint recognition algorithm to GUC team for testing. The GUC100 database is intended for technology testing which is an offline evaluation of biometric components using a pre-existing corpus [9]. Although evaluation of products from a single biometric supplier is essential from the supplier’s perspective, testing of scenarios, where products (e.g. sensor, minutia extractor, minutia comparator) are provided by different suppliers is very important for both integrators and operators to proof the interoperability prior to component integration and/or system roll-out. This refers to the settings where e.g. the enrolment and verification fingerprint images are acquired by different capture devices. For instance, in a biometric passport case, the document issued by a country where the enrolment image is captured by one scanner shall be able to be verified by another country where the probe image is very likely to be acquired by a different scanner. The GUC100 database provides greater potential to evaluate performance 1 GUC stands for Gjøvik University College 2010 International Conference of Computational Science and Its Applications 978-0-7695-3999-7/10 $26.00 © 2010 IEEE DOI 10.1109/ICCSA.2010.71 303 2010 International Conference on Computational Science and Its Applications 978-0-7695-3999-7/10 $26.00 © 2010 IEEE DOI 10.1109/ICCSA.2010.71 303