Master’s thesis “Tribraide”: Kai Altwicker develops a fusion-based detector for AI-generated images
Master’s thesis “Tribraide”: Kai Altwicker develops a fusion-based detector for AI-generated images
2026-04-29
With “Tribraide”, Kai Altwicker has developed a detector for AI-generated images in his master’s thesis at the Institute of Media and Imaging Technology at TH Köln that goes beyond conventional single-model approaches. The work combines spatial-domain, frequency, and reconstruction analyses and exploits the complementarity of these feature spaces to detect synthetic content. The thesis was supervised by Prof. Dr. Jan Salmen and Prof. Dr. Gregor Fischer.
Approach
Instead of relying on a single detector, Tribraide fuses three complementary views of the image: a spatial-domain analysis, a frequency-spectrum analysis, and a reconstruction-error analysis. Each view responds to different traces of generative methods — structural artefacts, characteristic frequency patterns, and inconsistencies under back-projection. Merging the feature spaces yields a classifier that outperforms individual models in both robustness and detection rate.
Robustness under realistic image alterations
A central finding concerns the stability of the method under typical image-processing steps such as compression and noise. Whereas many detectors lose discriminative power even under moderate JPEG compression, Tribraide remains reliable thanks to the fusion of the three feature spaces — a property profile that is essential for practical use in media forensics and image authentication.
Continuation at TH Köln
Kai Altwicker remains at TH Köln and is now a doctoral researcher in the group of Prof. Dr. Pascal Cerfontaine (AI4Science). A detailed article on the master’s thesis is available via this link.