Publication

Abstract : The growth of deepfakes, or hyper-realistic synthetic media, demands effective detection algorithms to ensure the validity of visual information. This study presents a unique deepfake detection model based on a hybrid architecture that takes advantage of the characteristics of both Meso-5 and XceptionNet. Meso-5, which is well-known for its capacity to efficiently capture mesoscopic picture characteristics, is carefully paired with Xception-Net’s outstanding feature extraction capabilities. This novel strategy tries to achieve higher deepfake detection accuracy than previous approaches. We assess the proposed model’s performance using the publicly available Deepfake Detection Challenge (DFDC) dataset, which was published by Facebook in 2018. Our results show that this hybrid technique works well, with an impressive accuracy of 81.24% in detecting deepfakes. These findings demonstrate our model’s ability to mitigate the deepfake danger and build a more dependable future for visual information.