A Novel Facial Manipulation Detection Method Based on Contrastive Learning

Abstract

Nowadays, numerous synthesized face-swapping videos generated by face forgery algorithms have become an emerging problem, which promotes facial manipulation detection to be a significant topic. With the development of face forgery algorithms, some fake face images or videos generated by those strong forgery algorithms are very realistic, which have brought much difficulty to facial manipulation detection. In this paper, we present a novel facial manipulation detection method based on contrastive learning. We analyze the texture features of manipulated facial images and propose to compare and learn the features of the whole face and the center face in order to get more general features. We calculate the similarity and distribution distance between the whole face and the center face. The experiments implemented on FaceForensics++ dataset demonstrate that the proposed method achieves outstanding results and can learn the general features.