The goal of the REVEAL project is to address the problem of digital image forensics and to provide reference algorithms to the academic community and to the forensics experts. We will tackle the problems of image tempering detection, image tempering localization and large scale image analysis, which are the main requirements of the Defals challenge. In order to do so, our methodology relies on several core components:
1. the acquisition of a large image database that will be used in order to train models designed for forensics purposes; this database will also be automatically processed in order to generate examples of manipulated images,
2. the development of forensics algorithms that will be able to scale with respect to the large diversity of image sources and image numbers. This will be possible by specific coding principles and the access to HPCs. We plan also to release our algorithms to the academic community by the development of a dedicated platform, but also to release a forensic private server especially dedicated to forensics experts,
3. the diversification of the forensics strategies that will be developed, some relying on feature extractions, other relying on the training of statistical models, some possibly tutored by forensics experts, other fully automatic using deep learning strategies, some supervised, other unsupervised, . . .
4. the strong commitment of the academic researchers at participating at the two milestones of the Defals challenge, particularly the project leader who has already been confronted to the organization and the participation in two challenges in data-hiding, and the two other partners who have participated to previous challenges in data-hiding,
5. the guaranty to have a continuous transfer between the two academic partners (CRIStAL and GIPSA) and the technological institute B-COM in order to develop products dedicated to forensics experts,
6. the participation members of the DxO company as scientific advisors in the scientific discussions and the different developments of the project, DxO is one leading company in image development and photographic device characterization.
The REVEAL project is naturally decomposed into tasks that will lead to the development of state of the art algorithms in forensics, and to the participation in the contest. Task T0 (Coordination and communication) is the general task of the project. The aim of this task is to ensure the smooth running of the project, but also to deal with internal and external communication subtasks. Then, three tasks are dedicated to research. The first one (T1) aims at building very large datasets of original and manipulated images. The two next ones (T2 and T3) respectively address the two core parts of the research project which are the development of forensics tools that are either based on feature extraction algorithms, or on statistical models. The goal of these two tasks will be to provide state-of-the-art methods for the Defals contest. This is achieved by the conduction of a dedicated task on participating in the two Defals submissions (T4). Another output of Defals, linked with T4, is the generation of two forensics platforms, one open for academics and for DGA, and an another dedicated to forensics experts.
Monsieur Patrick Bas (Centre de Recherche en Informatique, Signal et Automatique de Lille)
The author of this summary is the project coordinator, who is responsible for the content of this summary. The ANR declines any responsibility as for its contents.
CRIStAL Centre de Recherche en Informatique, Signal et Automatique de Lille
GIPSA-Lab Grenoble Images Sgnal Parole Automatique
Institut de Recherche Technologique B-Com
Help of the ANR 326,580 euros
Beginning and duration of the scientific project: February 2017 - 42 Months