Automated Methods for Image and Video Forgery Detection

Abstract

Image and video manipulation has been improving substantially in the latest years due to rapid technological improvements. Creating realistic fake video footage is now available to anyone regardless of the level of expertise. While it is mostly used for harmless entertainment or useful applications, this increases the chances of it being used with malicious intent. Therefore, automated methods of forgery detection are needed. This thesis reviews state-of-the-art image and video forgery detection models, as well as the backbones that are often used for the task. It is concluded that while the detection accuracy is decent, it is not enough yet to be used in serious cases. Moreover, image forgery detection models often suffer from poor, varying testing. Video forgery detection models, on the other hand, simply do not have a sufficient prediction accuracy.