Competition results show iris biometric liveness detection a work in progress

Competition results show iris biometric liveness detection a work in progress

The 2020 edition of the Iris Liveness Detection competition (LivDet-Iris) has been held, with significant increases in difficulty resulting in weaker spoof attack detection performance over the previous competitions.

The event is held to assess and highlight advances in presentation attack detection (PAD) for iris biometric systems. Organizers included Stephanie Schuckers of Clarkson University, Adam Czajka of the University of Notre Dame and Arun Ross of Michigan State University, as well as academics from universities around the world and the Idiap Research Institute in Switzerland.

Iris recognition sensors from LG, IrisGuard, Iris ID, and IriTech were used to collect biometric images. No training dataset was offered, and the competition analysts attribute the lower attack detection rate compared to previous years to the introduction of novel attack types, the increased complexity of the test datasets, and possible variability between training and test datasets.

The 2020 version of the competition introduced new types of attacks, specifically with the eyes of deceased people, prosthetic eyes and samples displayed on a screen. It also initiated the competition as an on-going project, with a testing protocol available via the Biometrics Evaluation and Testing (BEAT) platform hosted by Idiap. The performance of submissions was also compared with three baseline methods, provided by Notre Dame and MSU, and three open-source PAD methods. The two PAD baseline algorithms provided by MSU were found to be much more accurate for attack detection than any others.

Other attack methods considered include printed irises and patterned contact lenses.

The winning entry, from USACH/TOC (Universidad de Santiago – Chile and Chile’s TOC Biometrics) detected spoofs with an APCER (Attack Presentation Classification Error Rate) of 59.1 percent and a BPCER (Bona-fide Presentation Classification Error Rate) of 0.46 percent. The best performing algorithm in the first LivDet-Iris, which was held in 2013, had an APCER of 5.7 percent, and a BPCER of 28.6 percent. By the third and most recent competition in 2017, the best result was an APCER of 14.71 percent and a BPCER of 3.36 percent. Other entries in the 2020 competition, from “Competitor-3” and Fraunhofer IGD in Germany, also had much higher APCER scores than entries in previous years, and the Fraunhofer team’s entry was just slightly behind the USACH/TOC team’s in overall detection accuracy.

The paper will be presented at the International Joint Conference on Biometrics (IJCB 2020), which begins at the end of September.

This post was updated at 8:59 AM on Wednesday, September 9 2020 to include Adam Czajka as an organizer of the event.

Related Posts

Article Topics

 |   |   |   |   |   |   |   |   |   |   | 

Comments

Leave a Reply

This site uses Akismet to reduce spam. Learn how your comment data is processed.

Most Read This Week

Featured Company

Biometrics Research

Biometrics White Papers

Biometrics Events

Explaining Biometrics