A Comprehensive Review of Deep-Learning-Based Methods for Image Forensics.

Area of Science:

• Computer Science
• Digital Forensics
• Artificial Intelligence

Background:

• The proliferation of accessible image editing tools has led to realistic image forgeries.
• Distinguishing between authentic and manipulated media is challenging for the human eye.
• The spread of doctored images poses risks, including public deception and legal implications.

Purpose of the Study:

• To provide a comprehensive literature review of image forensics techniques.
• To focus on deep-learning-based methods for image manipulation detection.
• To address the growing challenge of identifying sophisticated image forgeries, including Deepfakes.

Main Methods:

• Systematic review of existing image forensics literature.
• Categorization of techniques based on the type of image manipulation.
• In-depth analysis of deep-learning approaches for forgery detection.
• Review of relevant image databases and anti-forensic methods.

Main Results:

• Despite the ease of creating image forgeries, numerous detection techniques exist.
• Deep-learning methods show significant promise in identifying various image manipulations.
• The review covers detection of routine edits, intentional falsifications, camera identification, CG image classification, and Deepfake detection.

Conclusions:

• Image forensics is a critical field for verifying media authenticity.
• Deep-learning-based image forensics offers effective solutions against sophisticated forgeries.
• Further research is needed to combat the evolving landscape of doctored images and anti-forensic strategies.