Description du sujet de thèse

Documents play a critical role in communication, record-keeping, and decision-making across various domains, yet their integrity is increasingly threatened by sophisticated digital manipulation techniques. Traditional forensic approaches leverage device-specific artifacts and content-based analysis to verify the
authenticity of documents and detect forgeries. However, existing methods often lack
statistical rigor, leading to uncertainty in detection reliability. This project aims to advance forensic document analysis by developing \textit{a contrario} detection methods based on large deviation theory. By modeling the document image formation process and analyzing intrinsic structures such as text-line alignment and redundancy, we seek to establish a principled framework for anomaly detection. Our approach will enable the quantification of detection reliability, reducing false positives and improving forensic decision-making. Still, the challenge of the thesis is not only to develop the theory and the algorithms but also to publish them online in an open form, so that anyone can submit a suspicious document and get an automatic scientific report on detected anomalies

Contexte de travail

The project will be done at Centre Borelli ENS Paris-Saclay. There are planned
collaborations with the ''Laboratorio Del Plata'' in Uruguay to spend at least a month per year working there.

Contraintes et risques

WITHOUT