The amount of multimedia content shared everyday online recently underwent a dramatic increase. This, combined with the stunning realism of fake images that can be generated with AI-based technologies, undermines the trustworthiness of online information sources. In this work, we tackle the problem of preserving media trustworthiness online from two different points of view. The first one consists in assessing the human ability to spot fake images, focusing in particular on synthetic faces, which are extremely realistic and can represent a severe threat if used to disseminate fake news. A perception study allowed us to prove for the first time how people are more prone to question the reality of authentic pictures rather than the one of last-generation AI-generated images. Secondly, we focused on social media forensics: our goal is to reconstruct the history of an image shared or re-shared online as typically happens nowadays. We propose a new framework that is able to trace the history of an image over multiple sharings. This framework improves the state of the art and has the advantage of being easily extensible with new methods and thus adapt to new datasets and scenarios. In fact, in this environment of fast-paced technological evolution, being able to adapt is fundamental to preserve our trust in what we see.
| Identifer | oai:union.ndltd.org:unitn.it/oai:iris.unitn.it:11572/334996 |
| Date | 29 March 2022 |
| Creators | Lago, Federica |
| Contributors | Lago, Federica, Boato, Giulia |
| Publisher | Università degli studi di Trento, place:TRENTO |
| Source Sets | Università di Trento |
| Language | English |
| Detected Language | English |
| Type | info:eu-repo/semantics/doctoralThesis |
| Rights | info:eu-repo/semantics/openAccess |
| Relation | firstpage:1, lastpage:146, numberofpages:146 |
Page generated in 0.0024 seconds