PaperlayerView exact revision on arXiv

Source-linked AI summary

Direct attacks using fake images in iris verification

Virginia Ruiz-Albacete, Pedro Tome-Gonzalez, Fernando Alonso-Fernandez, Javier Galbally, Julian Fierrez, Javier Ortega-Garcia

arXiv:2111.00178v1cs.CV

TL;DR

Biometric traits cannot be replaced, and biometric systems are vulnerable to external attacks, motivating this study of direct attacks on iris verification. The authors created fake iris images from BioSec irises and evaluated two attack scenarios; both had success rates of 50% or higher when images were correctly segmented.

  • Problem

    Biometric traits lack secrecy and cannot be replaced if stolen, while biometric systems remain vulnerable to external attacks.

    Source: §1 · pp. 1–2

  • Method

    The authors created fake images from BioSec iris images and compared two direct-attack scenarios with normal operation using a publicly available recognition system.

    Source: §1 · p. 2, §4 · p. 9

  • Results

    About 40% of fake images were correctly segmented, and both evaluated attacks had success rates of 50% or higher.

    Source: §4 · p. 10

  • Takeaways & Limitations

    The results support considering countermeasures such as liveness detection for direct attacks on iris systems.

    Source: §4 · p. 10

    The authors identify evaluation with another type of iris sensor as future work, bounding the reported findings to the sensor used in this study.

    Source: §4 · pp. 9–10

Abstract

from arXiv · show

In this contribution, the vulnerabilities of iris-based recognition systems to direct attacks are studied. A database of fake iris images has been created from real iris of the BioSec baseline database. Iris images are printed using a commercial printer and then, presented at the iris sensor. We use for our experiments a publicly available iris recognition system, which some modifications to improve the iris segmentation step. Based on results achieved on different operational scenarios, we show that the system is vulnerable to direct attacks, pointing out the importance of having countermeasures against this type of fraudulent actions.

1 Introduction

Biometric recognition offers identification without passwords or tokens, but biometric traits cannot be replaced and systems remain vulnerable to external attacks. This paper studies direct attacks on iris-based verification.

1 Introduction

  • Biometric recognition identifies people from anatomical or behavioral characteristics; iris recognition is described as reliable and the iris as stable over a lifetime.Unlike passwords or cards, biometric traits lack secrecy and cannot be replaced if stolen.Source: §1 · p. 1
  • The eight identified attack points divide into direct and indirect groups: direct attacks such as gummy fingers need no system-specific knowledge, while indirect attacks require additional information about the system’s internal workings.Direct attacks occur in the analog domain, outside the system’s digital limits, so digital signatures and watermarking cannot be used.Source: §1 · p. 2
  • The study focuses on direct attacks on iris-based verification systems and builds a synthetic-iris database using images generated from 27 users of the BioSec multi-modal baseline corpus.Source: §1 · p. 2

2 Fake Iris Database

The paper creates fake iris images by printing real iris images and recapturing them at an iris sensor. It tests image-preparation options and builds a 432-image database from 27 users.

2 Fake Iris Database

  • Real iris images were preprocessed, printed on paper with a commercial printer, and presented to an iris sensor to create fake images.The three-step process is illustrated as preparation, printing, and sensor capture.Source: §2 · p. 3

2.1 Fake iris generation method

  • Inkjet printing on high-resolution paper with Open-TopHat preprocessing gave the best segmentation results among the tested options.Image quality depended on paper type, and preprocessing was needed because the camera did not correctly capture unmodified printed images.Source: §2.1 · pp. 3–4

2.2 Database

  • The database contains 432 fake images from 27 users, two eyes, four images, and two sessions, alongside corresponding real images.Fake-image acquisition used the same LG IrisAccess EOU3000 camera as BioSec.Source: §2.2 · p. 4

3 Experiments

The experiments use a publicly available iris-recognition system and compare normal operation with two attacks using printed fake irises. The system is vulnerable to both attacks at every tested operating point, although fake-image segmentation limits the images entering the comparison.

3.1 Recognition system

  • The recognition system segments iris and pupil boundaries, normalizes around the pupil center, encodes iris patterns with Log-Gabor wavelets, and matches templates using masked Hamming distance.Rotational shifts are tested and the lowest Hamming distance is retained.Source: §3.1 · pp. 5–7

3.2 Experimental Protocol

  • The protocol treats 54 eyes as users across two sessions, comparing real-iris operation with fake-enrollment/fake-test and real-enrollment/fake-test attacks.The attack comparisons use matchings defined for each scenario.Source: §3.2 · pp. 7–8

3.3 Results

  • For the original database, 348 of 432 images were correctly segmented (80.56%); for the fake database, 166 of 432 were correctly segmented (38.43%).Source: §3.3 · p. 8
  • Both attacks achieved success rates of 50% or higher at all four normal-operation FAR operating points, considering only correctly segmented images.Attack 1 had higher success rates, especially as the normal-operation FAR increased.Source: §3.3 · p. 9, Table 2 · p. 9

4 Conclusion

The study evaluates direct attacks using printed fake irises created from BioSec images and presented to an iris sensor. Both tested attacks were successful at rates of 50% or higher when fake images were correctly segmented; liveness detection and other sensors are identified as future directions.

  • The fake-iris database uses images from 54 eyes, printed with a commercial printer and presented to the same iris camera used in BioSec.Preprocessing, printer type, and paper type were studied when selecting the image-creation combination.Source: §4 · p. 9
  • When fake images were correctly segmented, both evaluated attacks had success rates of 50% or higher.The attacks compared fake enrollment and access with real enrollment followed by fake-iris access.Source: §4 · pp. 9–10
  • Liveness detection and evaluation with another iris sensor, including OKI’s handheld sensor, are proposed as future work.Suggested liveness cues include light reflections and eye or pupil responses.Source: §4 · p. 10
Loading 2111.00178v1…