Noise Based Deepfake Detection via Multi-Head Relative-Interaction

Deepfake brings huge and potential negative impacts to our daily lives. As the real-life videos circulated on Internet become more authentic, most existing detection algorithms have failed since few visual differences can be observed between an authentic video a one. However, forensic traces are always retained within synthesized videos. In this study, we present noise-based model, NoiseDF for short, which focuses underlying noise left behind particular, enhance RIDNet denoiser extract features from cropped face background squares of image frames. Meanwhile, devise novel Multi-Head Relative-Interaction method evaluate degree interaction faces backgrounds that plays pivotal role in task. Besides outperforming state-of-the-art models, visualization extracted has further displayed evidence proved robustness approach.