Detecting Forged Sentinel-2 Images Through Parallax-Based Cloud Analysis

Matthieu Serfaty, Quentin Bammey, Tina Nikoukhah, Rafael Grompone von Gioi, Carlo de Franchis

Abstract: The availability and significance of satellite imagery in our world is continuously growing. Satellite images now play a crucial role in various applications such as weather forecasting, greenhouse gas monitoring, agricultural crop health assessment, and external security. However, this also exposes them to malicious attacks aimed at hiding or manipulating information. Forensic analysis is, therefore, a necessary shield against disinformation and disruption attempts in these areas. While forensic analysis of photographs has received considerable academic attention in recent years, the same cannot be said for satellite imagery. In this study, we present two methodologies to create realistic datasets of images forged with added clouds, that may inconspicuously hide information. We show results produced by state-of-the-art forensic methods are unable to detect the forged satellite images. To overcome this problem, we propose a parallax-based method to detect inconsistent satellite images.