Digital media forensics

Digital media forensics encompasses techniques for detecting and authenticating manipulated images, videos, and audio. This includes analyzing compression artifacts, camera sensor patterns, frequency-domain signatures, and behavioral inconsistencies to identify synthetic or edited media.

Core approaches

• Frequency-domain analysis: Detecting artifacts in Fourier/frequency space introduced by GAN compression or video encoding
• Camera fingerprinting: Identifying camera sensor noise patterns and lens-specific artifacts
• Behavioral analysis: Detecting unnatural eye movements, facial expressions, head poses, or audio-visual synchronization
• Biological signal analysis: Monitoring heartbeat, blood flow, and eye-blink patterns for temporal consistency
• Copy-move forgery detection: Identifying regions copied within images
• Sensor pattern noise: Leveraging photo response non-uniformity (PRNU) to detect source camera or digital manipulation

Challenges

Digital media forensics faces an arms race with generation techniques: as detection methods improve, synthesis methods advance to evade them. Compression, transcoding, and platform-specific processing degrade detection signals. Evaluating methods across diverse datasets with inconsistent metrics limits reproducibility.

Key papers

• Rana et al. (2022) — systematic review of 112 deepfake detection studies across multiple forensic approaches
• FaceForensics++ — benchmark dataset for evaluating facial manipulation detection
• DFDC — large-scale deepfake detection benchmark