Fraunhofer-Gesellschaft

Publica

Hier finden Sie wissenschaftliche Publikationen aus den Fraunhofer-Instituten.

On the Detection of GAN-Based Face Morphs Using Established Morph Detectors

 
: Debiasi, Luca; Damer, Naser; Moseguí Saladié, Alexandra; Rathgeb, Christian; Scherhag, Ulrich; Busch, Christoph; Kirchbuchner, Florian; Uhl, Andreas

:

Ricci, Elisa (Ed.):
Image Analysis and Processing - ICIAP 2019 : 20th International Conference, Trento, Italy, September 9-13, 2019, Proceedings, Part II
Cham: Springer Nature, 2019 (Lecture Notes in Computer Science 11752)
ISBN: 978-3-030-30644-1 (Print)
ISBN: 978-3-030-30645-8 (Online)
pp.345-356
International Conference on Image Analysis and Processing (ICIAP) <20, 2019, Trento>
European Commission EC
H2020-Research and Innovation Program; 690907; IDENTITY
English
Conference Paper
Fraunhofer IGD ()
Lead Topic- Smart City; Research Area: Computer vision (CV); biometric; face recognition; spoofing attacks; CRISP

Abstract
Face recognition systems (FRS) have been found to be highly vulnerable to face morphing attacks. Due to this severe security risk, morph detection systems do not only need to be robust against classical landmark-based face morphing approach (LMA), but also future attacks such as neural network based morph generation techniques. The focus of this paper lies on an experimental evaluation of the morph detection capabilities of various state-of-the-art morph detectors with respect to a recently presented novel face morphing approach, MorGAN, which is based on Generative Adversarial Networks (GANs). In this work, existing detection algorithms are confronted with different attack scenarios: known and unknown attacks comprising different morph types (LMA and MorGAN). The detectors’ performance results are highly dependent on the features used by the detection algorithms. In addition, the image quality of the morphed face images produced with the MorGAN approach is assessed using well-established no-reference image quality metrics and compared to LMA morphs. The results indicate that the image quality of MorGAN morphs is more similar to bona fide images compared to classical LMA morphs.

: http://publica.fraunhofer.de/documents/N-559030.html