Head Pose Estimation Patterns as Deepfake Detectors

The capacity to create ”fake” videos has recently raised concerns about the reliability of multimedia content. Identifying between true and false information is a critical step toward resolving this problem. On issue, several algorithms utilizing deep learning facial landmarks have yielded intriguing results. Facial are traits that solely tied subject’s head posture. Based on observation, we study how Head Pose Estimation (HPE) patterns may be utilized detect deepfakes in work. HPE studied based FSA-Net, SynergyNet, WSM, which among most performant approaches state art. Finally, using machine technique K-Nearest Neighbor Dynamic Time Warping, their temporal categorized as authentic or false. We also offer set experiments for examining feasibility techniques such patterns. findings reveal ability recognize deepfake video an pattern dependent methodology. contrary, performance less technique. Experiments carried out FaceForensics++ dataset, presents both identity swap expression examples. show FSA-Net effective feature extraction method determining whether belongs not. approach robust comparison created various methods different goals. In mean obtain 86% accuracy task 86.5% swap. These up possibilities future directions solving detection problem specialized approaches, known fast reliable.