Photo-montage and counterfeit images are not a recent phenomenon. As soon as images have been used within a political or economical context, the reality and the authenticity of recorded scenes become a legitimate question. The emergence of digital data did not really modify that basic context.
The modification of digital images is now a fact, in particular in the domain of cyber-criminality. Modifications can be naive (slight modifications done in order to remove spots present in a facial pictures), controversial (removal of visual object's defaults on a online commercial website) or may have strong societal impacts (improbable meeting between two important politician persons).
This project takes place in the domain of image forensics. The main goal is to certify if an image is a clean or a doctored image. The associated decision process must be as reliable as possible because the digital proof of falsification is really credible only if the method of detection returns very limited number of errors.
In a first step, it is proposed to develop some methods to detect malicious modifications of digital images from two complementary approaches, a first approach based on the modelisation of the digital image acquisition process and a second approach based on machine learning. Considering that the two approaches are scientifically complementary, it will be then proposed to fuse them in order to form a unique detector of digital image integrity.
The DEFACTO consortium (UTT, EURECOM and SURYS) covers all required expertise to complete the DEFALS challenge. The domains of expertise are:
- Development of effective detection tools specific to image forensics;
- Good understanding of existing methods to generate malicious modifications as well as associated counter-forensic approaches associated with digital multimedia data;
- Significant knowledge in physical mechanisms attached to image acquisition process;
- Image processing and modeling;
- Knowledge of economical and societal issues related to digital image forgeries;
- Important expertise in several related domains that are steganography and digital watermarking, and sensor identification;
- Skills in machine learning.
Monsieur Florent Retraint (Université de Technologie de Troyes)
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.
UTT/ICD Université de Technologie de Troyes
Help of the ANR 399,963 euros
Beginning and duration of the scientific project: February 2017 - 42 Months