PREMIER: PREserving Media trustworthiness in the artificial Intelligence Era

The credibility of digital media is put at risk by the availability of easy-to-use media editing tools, and, recently, by the development of AI-techniques that permit to create visually plausible fake videos with no or minimum intervention of the users.

To goal of PREMIER is to devise a new class of techniques capable of distinguishing fake from original videos. The new techniques will be developed by following a hybrid approach whereby forgery detectors based on deep learning are coupled with model-based methods, so to cope with the problems raising when deep learning is used within a multimedia forensics scenario. Looked-for solutions will aim at decreasing the amount of data needed for training, at providing ways to interpret the results of the forensic analysis, and at improving the security of the developed tools in the presence of an informed forger aiming at evading fake media detection.

The developed tools will be integrated into a web-based demonstrator whereby users can assess the authenticity of their own videos. A dataset of original and fake videos will be built to train and test the techniques developed during the project. The dataset will be made available to the international research community.