Politecnico di Milano tackles deepfakes

The Politecnico di Milano strengthens its commitment to combating digital disinformation with the completion of two European projects focused on deepfake detection: FF4ALL and FUN-Media.

The research was carried out by the Image and Sound Processing Lab (ISPL) of the Department of Electronics, Information and Bioengineering, coordinated by professors Stefano Tubaro and Paolo Bestagini, with contributions from researchers Sara Mandelli and Luca Comanducci.

The projects represent a significant step towards the development of reliable technologies for digital content verification and user protection.

FF4ALL focused on the analysis of techniques used to generate and distribute synthetic images and videos.

The lab developed new tools for detecting artificial faces, combining 3D information and structural features to improve model robustness, even in the presence of compression and post-processing. Among the outcomes is the creation of the WILD dataset, which collects images generated by state-of-the-art models.

FUN-Media addressed the detection of audio deepfakes, an emerging and still underexplored threat.

Researchers developed new architectures based on “Mixture of Experts” models, enabling improved detection performance in complex and evolving scenarios. Additional work focused on anomaly-based detection and attribution techniques to identify the generative technology behind synthetic audio.

In a rapidly evolving landscape where generative AI enables highly realistic manipulated content, these results contribute to the development of essential tools to counter disinformation, manipulation, and social engineering risks.

“Alongside detection, the project also addressed the issue of attribution, namely the identification of the generative technology responsible for creating an audio content” said Paolo Bestagini, professor at the Department of Electronics, Information and Bioengineering.