DEEPFAKE DETECTION VIA DIGITAL FORENSIC ARTIFACT ANALYSIS: A MULTI-ARTIFACT FRAMEWORK

Abstract

Deepfake detection is a new field with rapid development in synthetic media technology manipulation in audio, video, and photos. Traditional single-artifact detection approaches depend on face inconsistency and audio distortions' uniqueness. Yet, deepfake detection is a sophisticated problem worth a multi-faceted analysis in effectively identifying manipulated multimedia. In this work, a new multi-artifact deepfake detection technique is proposed, leveraging synergy between heterogeneous digital forensics artifacts, such as pixel analysis, metadata analysis, and temporal discrepancies. By synergistically combining such heterogeneous artifacts, the proposed technique fortifies deepfake model robustness and accuracy, providing a multi-faceted deepfake detection technique for synthetic media. The proposed technique is evaluated with real deepfake datasets, such as FaceForensics++ and DeepfakeTIMIT datasets, providing rich diversity in manipulations in a variety of modalities. Experimental performance validates that the proposed technique outperforms single-artifact detection frameworks, providing high performance in deepfake detection in uncontrolled and controlled environments. In this work, the necessity for multi-faceted analysis in countering deepfake multimedia and its impact on security, trust, and privacy is emphasized.