A Novel Defensive Strategy for Facial Manipulation Detection Combining Bilateral Filtering and Joint Adversarial Training

Facial manipulation enables facial expressions to be tampered with or identities replaced in videos. The fake videos are so realistic that they even difficult for human eyes distinguish. This poses a great threat social and public information security. A number of detectors have been proposed address this threat. However, previous studies shown the accuracy these is sensitive adversarial examples. existing defense methods very limited terms applicable scenes effects. paper proposes new strategy detectors, which combines passive method, bilateral filtering, proactive joint training, mitigate vulnerability against filtering method applied preprocessing stage model without any modification denoise input training starts from model, mixes various examples original train model. introduction can defends multiple attacks. experimental results show positively helps counter