Deepfake: definitions, performance metrics and standards, datasets, and a meta-review

Area of Science:

• Artificial Intelligence
• Computer Vision
• Digital Media Forensics

Background:

• Deep learning advancements have fueled the creation and manipulation of realistic synthetic media, leading to the emergence of deepfake technology.
• The proliferation of deepfakes presents significant challenges across various domains, necessitating a thorough understanding of the field.

Purpose of the Study:

• To provide a comprehensive overview of deepfake technology.
• To analyze key aspects including definitions, performance metrics, and datasets.
• To conduct a meta-review of recent deepfake survey papers to identify challenges and recommendations.

Main Methods:

• Literature review of research papers and resources in English.
• Meta-review of 15 selected deepfake survey papers published since 2020.
• Analysis of definitions, performance metrics, datasets, challenges, and recommendations.

Main Results:

• Identified and categorized different definitions of deepfake.
• Cataloged commonly used performance metrics and standards for deepfake detection.
• Compiled a list of relevant deepfake-related datasets.
• Synthesized findings from 15 survey papers, highlighting key challenges and offering recommendations.

Conclusions:

• This paper offers the most comprehensive review of deepfake technology to date.
• Understanding deepfake definitions, metrics, datasets, and challenges is crucial for addressing its impact.
• The findings provide a valuable resource for researchers and practitioners in the field.