PhD defense of Matthieu SERFATY

Matthieu SERFATY

Fighting Misinformation in Satellite Imagery

Abstract

Satellite imagery plays a central role in environmental monitoring, urban planning, agriculture, and security. With its growing strategic value comes the risk of falsification, which could mislead decision makers and undermine trust. While image forensics has achieved strong results for consumer photography, the specificities of satellite acquisition—pushbroom sensors, orthorectification, radiometric calibration—render these methods largely ineffective.

This thesis investigates the adaptation of forensic techniques to remote sensing. Controlled experiments on forged Sentinel-2 datasets show that existing tools fail to detect even simple manipulations, confirming that photographic cues such as JPEG artefacts or demosaicing traces are absent in satellites. To address more realistic threats, the study focuses on cloud insertion, a natural and highly deceptive form of forgery. Existing tools prove ineffective, leading to the development of a parallax-based method that exploits the sequential acquisition of spectral bands to reveal inconsistencies.

Beyond scene-level manipulations, the thesis explores sensor-specific fingerprints. Residual analysis uncovers recurring dark line artefacts aligned with detector geometry, while the well-known Photo-Response Non-Uniformity (PRNU) is adapted to Sentinel-2 data. Though weaker than in consumer cameras, PRNU remains sufficiently stable to serve as a basis for source attribution and tampering detection.

Overall, the thesis provides new forged datasets, a benchmark evaluation, a parallax-based detection method, and exploratory studies of sensor artefacts and PRNU. These results demonstrate that satellite image authentication requires approaches grounded in sensor physics rather than direct reuse of photographic tools.

Key Words

Forgery detection, satellite imagery, image forensics

Supervision

• Rafael GROMPONE VON GIOI

Jury

• Patrick BAS, Professeur, Université de Lille, Rapporteur
• Béatrice PINEL-PUYSSéGUR, Chargée de recherche, CEA-DAM, Rapporteur
• Pascal MONASSE, Professeur, Imagine/LIGM, Ecole des Ponts, Examinateur
• Teddy FURON, Professeur, INRIA Rennes, Examinateur

Publication