Analysis of scanned documents for integrity and authenticity checking

There is no universal method for detecting counterfeit images. Several techniques have been proposed but each has its limits. Among these methods, the so-called "forensic digital image" techniques offer an interesting solution. They aim either to verify the integrity of the image, or to provide proof of its authenticity by identifying the system which acquired it. To do this, they take advantage of how the acquisition systems generate their output. In this thesis, we are particularly interested in flatbed scanners as an acquisition system and we propose to study and develop "digital image forensics" techniques for scanned multi-type documents. We first proposed techniques to identify the scanner behind a scanned document based on a set of signatures "manually" extracted from images. Then, to face the limitations of these approaches, we focused on the automatic extraction of signatures from scanners through 1D and 2D neural networks. Subsequently, we developed a new approach, called "Device Linking",which determines whether two images were acquired by the same scanner or not. Finally, we provide two security mechanisms capable of detecting content manipulation of scanned data based on certain approaches of source scanner identification proposed previously. In order to validate the solutions proposed in real situations and make comparisons between them, we have built a database of scanned documents that we have made public.