Countering Malicious DeepFakes: Survey, Battleground, and Horizon

Area of Science:

• Computer Vision
• Artificial Intelligence
• Digital Forensics

Background:

• Deep generative approaches, termed DeepFake, enable sophisticated manipulation of facial appearance.
• Applications range from visual effects in film to the malicious generation of misinformation.
• The dual nature of DeepFake technology necessitates robust detection methods to identify synthetic media.

Purpose of the Study:

• To provide a comprehensive overview and analysis of DeepFake generation, detection, and evasion research.
• To clarify the 'battleground' dynamics between DeepFake creation and detection methodologies.
• To identify research challenges, opportunities, trends, and future directions in the field.

Main Methods:

• Systematic survey and analysis of over 318 research papers on DeepFake technology.
• Development of a taxonomy for DeepFake generation techniques.
• Categorization of DeepFake detection methods.
• Showcasing the adversarial interactions between generation and detection.

Main Results:

• Detailed taxonomy of DeepFake generation methods and categorization of detection approaches.
• Illumination of the adversarial relationship, or 'battleground', between DeepFake creation and detection.
• Identification of emerging trends, including DeepFake detection evasion strategies.

Conclusions:

• The rapid advancement of DeepFake technology necessitates continuous research in detection and evasion.
• Understanding the 'battleground' provides critical insights into the current landscape and future trajectory of DeepFake research.
• Interactive diagrams are provided to facilitate exploration of popular DeepFake generators and detectors.