Hybrid Model for Detecting Deepfake Videos

Innovations that can produce Deepfake videos are developing quickly. These videos are simple to and leave no obvious signs of alteration. Even though forensic detection in high-quality video datasets has produced fantastic results, there is always a need for more research into the forensics of compressed videos. Compressed videos are common in unofficial forums, such as those on Facebook, WeChat, and Instagram. Therefore, a crucial challenge is how to identify compressed Deepfake videos. In this project, we propose a two-stream method to compute each frame of compressed Deepfake videos. The compressed data has lot of unwanted data which needs to be trimmed, the proposed model performs the filtering to remove the unwanted data on each frame. When combined with scores from the two streams, our proposed technique performs better than the existing techniques in compressed mode.