Recent advances in the generation of convincing deep fakes have highlighted a problem that had been brewing for more than a decade as multimedia editing programs became more and more popular: today it is extremely easy to generate digital forgeries, to the point that it has affected the credibility of the media. Likewise, these advances have a huge economic impact, because they affect the identification and verification in digital onboarding or Know-Your-Customer systems.

 

Fortunately, in parallel with these advances in manipulations, there has been amazing progress in the development of tools for automatic detection whose usefulness goes beyond the classic

forensic uses. These tools exploit imperceptible properties of the multimedia objects, so they go much deeper than what a person can do with editing software. This talk aims to give a perspective on the state of the art in multimedia forensic analysis, from conventional statistical methods to recent advances produced by machine learning, both for integrity verification and for tampering detection. In addition, we will cover other less known aspects such as the identification of the image / video capture device or the determination of the moment of recording of a video from the fluctuations in the lighting of the room. We will also discuss the current open problems and related

research directions.

 

For many of these tools we will present practical examples derived from our Aidfork project (https://aidfork.uvigo.es/) whose products we have licensed to various companies around the

world.