Algorithm Data for the Darmstadt Face Manipulation Detection Tests

Algorithm data from the Darmstadt Face Manipulation Detection Tests used in the paper Conditional Face Image Manipulation Detection: Combining Algorithm and Human Examiner Decisions, published in ACM IH&MMSec 2024.

Introduction

This repository provides algorithm scores for each trial from the Darmstadt Face Manipulation Detection Tests (DFMD 1 and DFMD 2) and coresponds to the algorithm scores used in [1].

The available files contain only algorithm scores. The human examiner scores and data used in [2] will be made available once this paper is published. Trial IDs are consistent across datasets, allowing cross-referencing of human and algorithm scores.

---

Algorithm

The algorithm implemented is based on the Differential Anomaly Detection Algorithm proposed in [3]. It leverages a Variational Autoencoder (VAE) with a subtraction fusion scheme. Hence it is only trained on bona fide (i.e., non-manipulated) face images. A high-level overview of the method is included below.

---

Data Overview

The repository includes the following files:

• dfmd1_algorithm_scores.csv: Contains algorithm scores for DFMD 1.
• dfmd2_algorithm_scores.csv: Contains algorithm scores for DFMD 2.

Data Columns

Column Name	Description
trial_id	Unique ID of the trial, which includes both a suspected and a trusted face image.
suspected_type	Type of suspected image in the trial (e.g., bonafide, morphing, face swap, or retouching).
score	Normalized algorithm score obtained on the trial.

---

Citation

If you use this data or find our work usefull, consider citing the following papers:

[1] Conditional Face Image Manipulation Detection: Combining Algorithm and Human Examiner Decisions

@inproceedings{Ibsen-DFMDManipulationDetectionHumanAlgFusion-2024,
  Author       = {M. Ibsen and R. Nichols and C. Rathgeb and D. J. Robertson and J. P. Davis and F. L{\o}v{\aa}sdal and K. Raja and R. E. Jenkins and C. Busch},
  Booktitle = {{ACM} Workshop on Information Hiding and Multimedia Security},
  Title        = {Conditional Face Image Manipulation Detection: Combining Algorithm and Human Examiner Decisions},
  Year         = {2024}
}

[2] The super-recogniser advantage extends to the detection of digitally manipulated faces

@misc{David-DFMDSuperRecogniserAdvantage-osf-2024,
 title={The super-recogniser advantage extends to the detection of digitally manipulated faces},
 url={osf.io/preprints/psyarxiv/ye7ph},
 publisher={PsyArXiv},
 author={J. P. Davis and R. Nichols and D. J. Robertson and M. Ibsen and others},
 year={2024},
 month={Nov}
}

[3] Differential Anomaly Detection for Facial Images

@inproceedings{Ibsen-PAD-DiffAnomalyDetection-WIFS-2021_1,
 Author = {M. Ibsen and L. J. Gonzalez-Soler and C. Rathgeb and P. Drozdowski and M. Gomez-Barrero and C. Busch},
 Booktitle = {{IEEE} Intl. Workshop on Information Forensics and Security ({WIFS})},
 Title = {Differential Anomaly Detection for Facial Images},
 Year = {2021},
}