Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

TetraLoss: Improving the Robustness of Face Recognition against Morphing Attacks

Mathias Ibsen, Lázaro J. González-Soler, Christian Rathgeb, Christoph Busch

TL;DR

Morphing-based attacks threaten border-control face recognition, prompting the need for morph-aware defenses. The authors propose TetraLoss, a loss function that trains an embedding adapter on anchor, positive, negative, and morphed samples to separate morphed embeddings from contributor identities while preserving verification accuracy. Across ArcFace, MagFace, and AdaFace on FERET/FRGCv2, the method yields substantial RIAPAR improvements (e.g., >45 percentage points at $FMR=0.1\%$) and remains effective with or without a differential MAD detector, aligning with ISO/IEC 30107-3 evaluation standards. This framework can be added on top of existing FR systems to boost morphing robustness in high-security deployments, enabling safer real-world operation with minimal disruption to current architectures.

Abstract

Face recognition systems are widely deployed in high-security applications such as for biometric verification at border controls. Despite their high accuracy on pristine data, it is well-known that digital manipulations, such as face morphing, pose a security threat to face recognition systems. Malicious actors can exploit the facilities offered by the identity document issuance process to obtain identity documents containing morphed images. Thus, subjects who contributed to the creation of the morphed image can with high probability use the identity document to bypass automated face recognition systems. In recent years, no-reference (i.e., single image) and differential morphing attack detectors have been proposed to tackle this risk. These systems are typically evaluated in isolation from the face recognition system that they have to operate jointly with and do not consider the face recognition process. Contrary to most existing works, we present a novel method for adapting deep learning-based face recognition systems to be more robust against face morphing attacks. To this end, we introduce TetraLoss, a novel loss function that learns to separate morphed face images from its contributing subjects in the embedding space while still achieving high biometric verification performance. In a comprehensive evaluation, we show that the proposed method can significantly enhance the original system while also significantly outperforming other tested baseline methods.

TetraLoss: Improving the Robustness of Face Recognition against Morphing Attacks

TL;DR

Morphing-based attacks threaten border-control face recognition, prompting the need for morph-aware defenses. The authors propose TetraLoss, a loss function that trains an embedding adapter on anchor, positive, negative, and morphed samples to separate morphed embeddings from contributor identities while preserving verification accuracy. Across ArcFace, MagFace, and AdaFace on FERET/FRGCv2, the method yields substantial RIAPAR improvements (e.g., >45 percentage points at ) and remains effective with or without a differential MAD detector, aligning with ISO/IEC 30107-3 evaluation standards. This framework can be added on top of existing FR systems to boost morphing robustness in high-security deployments, enabling safer real-world operation with minimal disruption to current architectures.

Abstract

Face recognition systems are widely deployed in high-security applications such as for biometric verification at border controls. Despite their high accuracy on pristine data, it is well-known that digital manipulations, such as face morphing, pose a security threat to face recognition systems. Malicious actors can exploit the facilities offered by the identity document issuance process to obtain identity documents containing morphed images. Thus, subjects who contributed to the creation of the morphed image can with high probability use the identity document to bypass automated face recognition systems. In recent years, no-reference (i.e., single image) and differential morphing attack detectors have been proposed to tackle this risk. These systems are typically evaluated in isolation from the face recognition system that they have to operate jointly with and do not consider the face recognition process. Contrary to most existing works, we present a novel method for adapting deep learning-based face recognition systems to be more robust against face morphing attacks. To this end, we introduce TetraLoss, a novel loss function that learns to separate morphed face images from its contributing subjects in the embedding space while still achieving high biometric verification performance. In a comprehensive evaluation, we show that the proposed method can significantly enhance the original system while also significantly outperforming other tested baseline methods.
Paper Structure (17 sections, 3 equations, 4 figures, 4 tables)

This paper contains 17 sections, 3 equations, 4 figures, 4 tables.

Table of Contents

  1. INTRODUCTION
  2. RELATED WORK
  3. Face Recognition Systems under Morphing Attacks
  4. Detection of Face Morphing Attacks
  5. PROPOSED FRAMEWORK
  6. Model Architecture
  7. TetraLoss
  8. EXPERIMENTAL SETUP
  9. Face Recognition Systems
  10. Databases
  11. Training Details
  12. Experimental Protocols
  13. Metrics
  14. Generalized Full-System Evaluation
  15. Feasibility Study
  16. ...and 2 more sections

Figures (4)

  • Figure 1: Examples of a morphed face image (b) created from subject 1 (a) and subject 2 (c). Taken from Scherhag-FaceMorphingAttacks-TIFS-2020.
  • Figure 2: Example of reference and probe images from FERET (a, b) and FRGCv2 (c, d).
  • Figure 3: Examples of the different morphing tools and their corresponding bona fide subjects in the used dataset.
  • Figure 4: t-SNE visualisation of the original AdaFace embeddings (a, c) and the modified TetraLoss embeddings (b, d) using ResNet100 (a, b) and ResNet50 (c, d).