Double Branch Face Forgery Detection Method for Low Quality Video
Existing detection methods have certain limitations when dealing with low-quality fake facial videos,such as poor detection performance in compressed low-quality videos,poor generalization performance,and a decrease in detection accuracy.In order to improve the accuracy and generalization of the detection network,this paper proposes to combine semantic information and noise information,and proposes a dual tributary network.While focusing on the Semantic informa-tion of the image,this paper can show the inconsistency between the forged area and the real area through high-frequen-cy noise information.The inconsistency exposed by high-frequency noise information focuses on the forgery traces in the image semantic information.It enhance the interactivity and integration between semantic information and high-frequency in-formation through interaction modules.To evaluate,this paper traines and testes the model on the FaceForensics++dataset,and evaluated its cross dataset generalization performance on the Celeb DF dataset.
万方数据
维普
