Attentive Semantic Exploring for Manipulated Face Detection

Zehao Chen,Hua Yang

DOI: https://doi.org/10.1109/icassp39728.2021.9414225

2020-01-01

Abstract:Face manipulation methods develop rapidly in recent years, whose potentialrisk to society accounts for the emerging of researches on detection methods.However, due to the diversity of manipulation methods and the high quality offake images, detection methods suffer from a lack of generalization ability. Tosolve the problem, we find that segmenting images into semantic fragments couldbe effective, as discriminative defects and distortions are closely related tosuch fragments. Besides, to highlight discriminative regions in fragments andto measure contribution to the final prediction of each fragment is efficientfor the improvement of generalization ability. Therefore, we propose a novelmanipulated face detection method based on Multilevel Facial SemanticSegmentation and Cascade Attention Mechanism. To evaluate our method, wereconstruct two datasets: GGFI and FFMI, and also collect two open-sourcedatasets. Experiments on four datasets verify the advantages of our approachagainst other state-of-the-arts, especially its generalization ability.