Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Greedy-DiM: Greedy Algorithms for Unreasonably Effective Face Morphs

Zander W. Blasingame, Chen Liu

TL;DR

This work proposes a greedy strategy on the iterative sampling process of DiM models which searches for an optimal step guided by an identity-based heuristic function, and finds that the proposed algorithm is unreasonably effective, outperforming all other morphing algorithms compared.

Abstract

Morphing attacks are an emerging threat to state-of-the-art Face Recognition (FR) systems, which aim to create a single image that contains the biometric information of multiple identities. Diffusion Morphs (DiM) are a recently proposed morphing attack that has achieved state-of-the-art performance for representation-based morphing attacks. However, none of the existing research on DiMs have leveraged the iterative nature of DiMs and left the DiM model as a black box, treating it no differently than one would a Generative Adversarial Network (GAN) or Varational AutoEncoder (VAE). We propose a greedy strategy on the iterative sampling process of DiM models which searches for an optimal step guided by an identity-based heuristic function. We compare our proposed algorithm against ten other state-of-the-art morphing algorithms using the open-source SYN-MAD 2022 competition dataset. We find that our proposed algorithm is unreasonably effective, fooling all of the tested FR systems with an MMPMR of 100%, outperforming all other morphing algorithms compared.

Greedy-DiM: Greedy Algorithms for Unreasonably Effective Face Morphs

TL;DR

This work proposes a greedy strategy on the iterative sampling process of DiM models which searches for an optimal step guided by an identity-based heuristic function, and finds that the proposed algorithm is unreasonably effective, outperforming all other morphing algorithms compared.

Abstract

Morphing attacks are an emerging threat to state-of-the-art Face Recognition (FR) systems, which aim to create a single image that contains the biometric information of multiple identities. Diffusion Morphs (DiM) are a recently proposed morphing attack that has achieved state-of-the-art performance for representation-based morphing attacks. However, none of the existing research on DiMs have leveraged the iterative nature of DiMs and left the DiM model as a black box, treating it no differently than one would a Generative Adversarial Network (GAN) or Varational AutoEncoder (VAE). We propose a greedy strategy on the iterative sampling process of DiM models which searches for an optimal step guided by an identity-based heuristic function. We compare our proposed algorithm against ten other state-of-the-art morphing algorithms using the open-source SYN-MAD 2022 competition dataset. We find that our proposed algorithm is unreasonably effective, fooling all of the tested FR systems with an MMPMR of 100%, outperforming all other morphing algorithms compared.
Paper Structure (34 sections, 4 theorems, 21 equations, 7 figures, 14 tables, 3 algorithms)

This paper contains 34 sections, 4 theorems, 21 equations, 7 figures, 14 tables, 3 algorithms.

Table of Contents

  1. Introduction
  2. Prior Work
  3. DiM
  4. Greedy-DiM
  5. Greedy Search
  6. Greedy Optimization
  7. Theoretical Analysis
  8. Experimental Setup
  9. Datasets
  10. FR Systems
  11. Metrics
  12. Results
  13. Vulnerability Study
  14. Detectability Study
  15. Conclusions
  16. ...and 19 more sections

Key Result

Theorem 3.1

Given a sequence of monotonically descending timesteps, $\{t_n\}_{n=1}^N$, from $T$ to $0$, the DDIM solver to the Probability Flow ODE, and a heuristic function $\mathcal{H}$, the locally optimal solution admitted by Greedy-DiM* at time $t_n$ is globally optimal.

Figures (7)

  • Figure 1: Example of a morph created using Greedy-DiM. Samples are from the FRLL dataset frll.
  • Figure 2: Overview of a single step of the Greedy-DiM* algorithm. Proposed changes highlighted in green.
  • Figure 3: Illustration of the search space in $\mathbb{R}^2$ of different DiM algorithms at a single step. Purple denotes Morph-PIPE/Greedy-DiM-S, red denotes Greedy-DiM-S continuous, and green denotes Greedy-DiM*. Note the search spaces of the algorithms other than Greedy-DiM* lie in a low dimensional manifold.
  • Figure 4: Comparison of DiM morphs on the FRLL dataset.
  • Figure 5: The RSM and Transferability metrics.
  • ...and 2 more figures

Theorems & Definitions (8)

  • Theorem 3.1
  • proof
  • Theorem 3.2
  • proof
  • Theorem A.1
  • proof
  • Theorem A.2
  • proof