Generalized Deepfake Detection Using Identity, Behavioral, and Geometric Signatures

Trust in online media is increasingly compromised by deepfake multimedia, which undermines the authenticity of shared content. Existing detection techniques often perform well only on specific types of deepfakes, limiting their generalization ability and making them vulnerable in real-world applications. To address this, we propose a novel deepfake detection framework featuring an effective feature descriptor that integrates deep identity, behavioral, and geometric (DBaG) signatures, along with a classifier named DBaGNet. The DBaGNet classifier utilizes the extracted DBaG signatures and applies a triplet loss objective to enhance generalized representation learning for improved classification. These comprehensive DBaG signatures capture both facial geometry inconsistencies and behavioral cues, improving the detection of diverse deepfake types and enhancing generalization. We evaluate our approach using six benchmark deepfake datasets: WLDR, CelebDF, DFDC, FaceForensics++, DFD, and NVFAIR. Cross-dataset evaluations demonstrate significant performance gains over several state-of-the-art methods.