Detection of Fake and Recontextualized Media against Disinformation

The recent development of Artificial Intelligence tools, which enables non-expert users to generate deepfake videos of extraordinary quality is raising increasing concerns about the credibility and the authenticity of digital media. The ability to create such videos, coupled with the widespread use of social media raises enormous concerns about the authenticity and credibility of what we see. The consequences of these fake media could be devastating to our society. In addition to deepfake manipulation, a significant challenge persists within the realm of fake media diffusion, the deceptive geographical recontextualization of photos. This deceptive practice involves the misrepresentation of images in news posts or articles, where an image depicting a war, flood, protest, or earthquake, is claimed to be captured in one specific location in the world, but it has been taken from an entirely different country or city. This manipulation fuels the dissemination of misinformation campaigns by exploiting the powerful impact of images in shaping public perception and belief. In response to these challenges, the thesis focuses on the development of AI-based algorithms. These algorithms aim to detect deepfake videos, a critical step in safeguarding the credibility of digital media. Additionally, the thesis addresses the deceptive recontextualization of photos by unveiling the true geographical locations where images were captured in the world. By tackling these issues, the thesis aims to contribute to the protection of our digital world from the adverse effects of misinformation campaigns and fake media diffusion.