Feature Fusion and Ranking for Face Forgery Detection

In recent years, face forgery detection has attracted more and more attention due to the negative influence of social opinion brought by face forgery technology. Although a lot of recent work has achieved good results, the faces generated by new forging methods are often not well detected. In this paper, we present a Feature Fusion and Ranking (FFR) method to further improve the performance of face forgery detection. Specifically, considering almost identical appearance between fake face and real face, our FFR method designs a multi-feature fusion structure to well exploit features from different backbones for enhancing detection ability. At the same time, in order to improve the discriminative ability of the proposed FFR, we design a feature ranking block and the corresponding ranking loss from the idea of metric learning. Experiments on intra-dataset and cross-dataset evaluation demonstrate the effectiveness of our proposed FFR.