Bias and Explainability in Image Forensics
Forensic detectors inherit biases from their training data and can rely on cues that have little to do with the manipulation itself. We examine how these systems perform across demographic groups, what they actually attend to, and how data diversity shapes their ability to generalise, with the aim of detectors whose decisions are fair and interpretable.
Publications
Shades of Generalization: Diversity Aware Open Set Deepfake Detection
2026 IEEE 20th International Conference on Automatic Face and Gesture Recognition (FG)
·
25 May 2026
·
doi:10.1109/FG67764.2026.11557037
Exploring Saliency Bias in Manipulation Detection
2024 IEEE International Conference on Image Processing (ICIP)
·
27 Oct 2024
·
doi:10.1109/ICIP51287.2024.10648063
A Relative Data Diversity Measure for Synthetic Face Images
2024 IEEE International Joint Conference on Biometrics (IJCB)
·
15 Sep 2024
·
doi:10.1109/IJCB62174.2024.10744512
In-group and Out-group Performance Bias in Facial Retouching Detection
2022 IEEE International Joint Conference on Biometrics (IJCB)
·
10 Oct 2022
·
doi:10.1109/IJCB54206.2022.10007942