[08:48 Wed,2.May 2018 by Thomas Richter] |
https://www.slashcam.de/news/single/DeepFake--Gesichtstausch--nicht-nur--in-Pornos-per-14122.html (DeepFakes), i.e. faces exchanged in videos using deep learning algorithms, are becoming easier and easier to produce. Until now, one focus has been to let celebrities seemingly participate in erotic clips, but many other possibilities to use DeepFakes are conceivable, which have an even greater potential to harm people by apparently real videos and to re-terminate the confidence in the conclusiveness of moving pictures in this way. ;)
Researchers at the TU Munich have therefore developed a Deep Learning System to automatically recognize exchanged faces using Deep Learning. To this end, a database with a large number of fake videos was first created by the researchers FaceForensics, with the help of which a Deep Learning System was then trained in order to identify the fakes based on the differences between the DeepFake images and the originals. The new algorithm detects more forgeries than other methods - even if the videos are compressed, which makes the task much more difficult due to the poorer image quality. However, the new method is double-edged - it can not only be used to detect deep fakes, but also (quasi as feedback for a deep fake system) to improve such counterfeits. VY=Tle7YaPkO_k  more infos at bei arxiv.orgdeutsche Version dieser Seite: Neuer Algorithmus kann per DeepFake gefälschte Gesichter erkennen |
 |
|
|