Multi-Layer Fusion Neural Network for Deepfake Detection

Recently, the spread of videos forged by deepfake tools has been widely concerning, and effective ways for detecting them are urgently needed. It is known that such artificial intelligence-aided forgery makes at least three levels artifacts, which can be named as microcosmic or statistical features, mesoscopic macroscopic semantic features. However, existing detection methods have not designed to exploited all. This work proposes a new approach more videos. A multi-layer fusion neural network (MFNN) capture artifacts in different levels. Features maps output from specially shallow, middle, deep layers, used statistical, mesoscopic, respectively, fused together before classification. FaceForensic++ dataset was train test method. The experimental results show MFNN outperforms other relevant methods. Particularly, it demonstrates advantage low-quality