September 14, 2021

Combating Deep Fakes with Artificial Intelligence: Fraunhofer HHI launches BMBF-funded project FakeID

September 14, 2021

The Fraunhofer Heinrich Hertz Institute (HHI) and its "Vision and Imaging Technologies" department have joined the new consortial project FakeID. FakeID aims to use artificial intelligence (AI) video analytics to detect false and manipulated identities, so-called deep fakes. The project started in May 2021 and will run until April 2024. It is funded by the German Federal Ministry of Education and Research (BMBF) with an amount of 2.6 million euros.

The constant development of new image and video processing technologies means that the risk of manipulation and falsification of (audio-) visual material is also increasing. Apart from the devastating consequences that the manipulated material can have, prosecution of the creation and distribution of these technologically sophisticated deep fakes also poses a challenge.

The FakeID consortium project aims to approach these problems on a technical level by identifying and classifying potential attacks and forgeries of images and video data streams that have been created using different types of deep fakes. For this purpose, the researchers will apply image- and video-based artificial intelligence authentication methods, which will also enable the development of algorithms for the detection of fake identities. 

In order to recognize and assess attack scenarios and forgeries, the project is generating new insights into the diverse motivations and backgrounds for falsifying images and video data streams. These range from fraud to the manipulation of political decision-making processes. In addition, the project will develop approaches for detection and explanation as well as for the handling of these cases by police authorities and the judiciary.

The main contribution of Fraunhofer HHI's Vision and Imaging Technologies department will be the technical implementation and development of new methods for the generation and detection of deep fakes. The core objective is to create new detectors that reliably detect deep fakes using AI-based methods and exploit a wide range of information such as inconsistencies in lighting, facial expressions, movement or vital parameters such as pulse.

To increase the usability of the results for end users, the developed methods should remain interpretable and explainable regarding the decisions of the neural network. In order to train and evaluate the new detectors, Fraunhofer HHI will build up a reference database of deep fakes. For this purpose, the researchers will analyze and use existing databases and methods, as well as develop their own improved methods for synthesis and manipulation. This way they’ll be able to boost the robustness of the detectors against future attacks.

Alongside Fraunhofer HHI, Bundesdruckerei GmbH, Otto von Guericke University Magdeburg, Berlin School of Economics and Law, and BioID GmbH are also involved in the research project.