Face forgery detection with a fused attention mechanism

Abstract

In recent years, the technology of forged faces has become more and more sophisticated, and the human eye cannot even distinguish these forged products. The fake face images or videos generated by this series of technologies are widely disseminated on the Internet, causing a serious impact on society, thus drawing attention to DeepFake detection, and more research is also inclined to this, but The current research has the problem that the extracted artifact features are relatively single, which leads to the relatively low performance of the artifact detection algorithm. To solve the limitations of the existing methods, the DeepFake detection method fused with attention mechanism is proposed, which extracts the global and local features of the face respectively. Artifact features are found in multiple regions of the face. The method is trained on the FaceForensics++ dataset, and the detection accuracy is improved in different network structures.