iCaps-Dfake: An Integrated Capsule-Based Model for Deepfake Image and Video Detection

摘要: Fake media is spreading like wildfire all over the internet as a result of great advancement in deepfake creation tools and huge interest researchers corporations are showing to explore its limits. Now anyone can create manipulated unethical forensics, defame, humiliate others or even scam them out their money with click button. In this research new detection approach, iCaps-Dfake, proposed that competes state-of-the-art techniques video addresses low generalization problem. Two feature extraction methods combined, texture-based Local Binary Patterns (LBP) Convolutional Neural Networks (CNN) based modified High-Resolution Network (HRNet), along an application capsule neural networks (CapsNets) implementing concurrent routing technique. Experiments have been conducted on large benchmark datasets evaluate performance model. Several metrics applied experimental results analyzed. The model was primarily trained tested DeepFakeDetectionChallenge-Preview (DFDC-P) dataset then Celeb-DF examine capability. achieved Area-Under Curve (AUC) score improvement 20.25% models.