The challenge: In the past decades, photographs and videos of events were regarded as reliable evidence for both news media and forensic investigations.
However, recent developments in AI technology now make it possible, with minimal technological knowledge or skills, to create extremely convincing manipulated or synthesised images, audio and video data, often called DeepFakes.
This fact is a serious threat to the trustworthiness of news in the media and forensic processes in criminal investigations.
Therefore, development of reliable methods to detect manipulations and distinguish bonafide from synthesised data is of utmost importance.
In addition, especially for forensic investigations, simple detection is not sufficient.
An explanation of the traces of manipulation and tools used for the synthesis or manipulation should be produced as well.
In the EU project DEFORM, 14 research institutes, universities and companies from 6 different countries aim to develop methods and tools for explainable detection of DeepFakes using Foundation models and other techniques.
At the University of Twente, we are looking for a PhD researcher who will work in close collaboration with the other partners on detection of fake videos using inconsistencies in video as well as audio.