Abstract
Detecting manipulated facial images and videos is an increasingly importanttopic in digital media forensics. As advanced face synthesis and manipulationmethods are made available, new types of fake face representations are beingcreated which have raised significant concerns for their use in social media.Hence, it is crucial to detect manipulated face images and localize manipulatedregions. Instead of simply using multi-task learning to simultaneously detectmanipulated images and predict the manipulated mask (regions), we propose toutilize an attention mechanism to process and improve the feature maps for theclassification task. The learned attention maps highlight the informativeregions to further improve the binary classification (genuine face v. fakeface), and also visualize the manipulated regions. To enable our study ofmanipulated face detection and localization, we collect a large-scale databasethat contains numerous types of facial forgeries. With this dataset, we performa thorough analysis of data-driven fake face detection. We show that the use ofan attention mechanism improves facial forgery detection and manipulated regionlocalization.