New research explores AI manipulation attacks on face biometric systems

A pair of new research papers address sophisticated fraud attempts on biometric systems using AI, in one case to carry out morph attacks, and in the other using synthetic data to carry out template inversion attacks.

The Darmstadt University of Applied Sciences, known as Germany’s Silicon Valley, has released a paper on training algorithms for Morphing Attack Detection (MAD).

The paper proposes two different methods based on transfer-transfer for automatically creating digital print-scan face images and using such images in the training of a MAD algorithm. The method can reach an Equal Error Rate (EER) of 3.84 percent and 1.92 percent on the FRGC/FERET database when including our synthetic and texture-transfer print-scan with 600 dpi to handcrafted images, respectively.

The paper was authored by researchers at the Biometrics and Internet Security Research Group and published on ArXiv.

Anticipating synthetic data attacks

Another research paper, published by the IEEE Transactions on Biometrics, Behavior, and Identity Science, proposes a new method for template inversion attacks against facial recognition systems using synthetic data. Among the authors of the paper is Sebastien Marcel, head of the Biometrics Security and Privacy group at Idiap Research Institute in Switzerland.

“Our experiments show that the trained model with synthetic data can be used to reconstruct face images from templates extracted from real face images,” the paper notes.

The method outperforms previous methods of attacks against facial recognition models on four different face datasets, including the MOBIO, LFW, AgeDB, and IJB-C datasets. It also outperforms methods on high-resolution 2D face reconstruction from facial templates and achieves competitive results with SOTA face reconstruction methods, according to the study. The generated face images were also tested in practical presentation attacks against facial recognition systems.

The results of the paper and materials to reproduce it are available on Gitlab.

Idiap has worked extensively on template inversion attacks, while work on face morphing attacks continues in various venues, including the European Commission’s iMars project.

Article Topics

biometrics  |  biometrics research  |  face biometrics  |  Idiap  |  morphing attack  |  Morphing Attack Detection (MAD)  |  synthetic data  |  template inversion attack