Description
In the DECOuVErT project, we are interested in the technique of deepfakes, which synthesizes images of humans based on artificial intelligence. The focus will be on the problem of generalization. We plan to approach the subject of deepfake detection as a problem of anomaly detection between consecutive frames of a manipulated video by studying two perspectives:
- colorimetry - we will explore the possibility of using different color spaces to improve the deepfake detection methods proposed in recent articles or in the DeepfakeBench benchmark.
- video encoding and compression - In addition to I-frames, we will use predicted frames, particularly P-frames, to improve the detection of recent deepfakes.
Involved reserchers
- Iuliia Tkachenko - Principal investigator in LIRIS, Lyon
- Alain Trémeau - Collaborator in Laboratoire Hubert Curien, Saint-Etienne
- William Puech - Collaborator in LIRMM, Montpellier
- Master internship on the topic "Détection de Deepfakes en analysant la palette des couleurs d’image " in Laboratoire Hubert Curien (01/02/2026 - 19/07/2026)
Link via FIL projects
Publications related to the project
- K. Ali Javaid, A. Trémeau, I. Tkachenko, “The effect of color-space representations on deepfake detection: a benchmark study”, Computational Color Imaging Workshop, Satellite workshop within ICIP2026, September 2026, Tampere, Finland.[pdf]