DeepFake detection

Deepfakes refer to all multimedia content that is synthetically altered or created using generative models in the field of Deep Learning. The main techniques used for this purpose are Generative Adversarial Networks (GANs) or Autoencoders. These techniques are continuously improving and allow for the generation of incredibly realistic images, making it possible to swap faces, add accessories like glasses, hats, or beards, change facial expressions, age or look younger individuals. Some networks are trained to generate completely synthetic faces. In addition to individual images, it is also possible to generate videos and allow the creation of fake speeches that have never happened before. All of this has made the situation worrying because it is possible to create deepfakes easily using applications or web apps and for this reason our research focuses on the study of technologies capable of recognizing these manipulations in the most complex scenarios.

PRA Lab studies and develops solutions for detecting deepfakes and morphing attacks, exploiting the quality inconsistency between frames and information in the frequency domain.

Latest Releted Publications

Active projects

Biometrics research topics