Image manipulation

Detection and analysis of manipulated, forged, or falsified images used in misinformation and disinformation campaigns. Image manipulation encompasses a spectrum from subtle edits (cropping, splicing, color correction) to elaborate fabrications (deepfakes, AI-generated content), and from intentional forgeries to out-of-context usage of authentic images.

Manipulation types

• Authentic but out-of-context: An unmanipulated image is repurposed to support a false claim (e.g., archival photo presented as current event).
• AI-generated: Images created entirely by generative models without a real-world source.
• Deepfakes: Face or body swap using neural synthesis, typically applied to video or photos of people.
• Staged: Deliberately posed or crafted images designed to mislead (e.g., fake statistics visualizations, mock screenshots).
• Spliced/edited: Pixel-level manipulation combining elements from multiple images or adding/removing objects.
• Low-end forgery: Simple edits (caption overlays, crude splicing) without sophisticated tools.

Key papers

• MMMMM: A Unified Taxonomy for Investigating the Mechanisms of Multilingual MultiModal Misinformation — Taxonomy of multimodal misinformation mechanisms explicitly categorizes image manipulation types (AI-generated, deepfake, staged, low-end source) as one axis; analyzes how specific manipulation types co-occur with narrative topics and linguistic mechanisms across languages.

Related topics

• Deepfakes and synthetic media — deepfakes are a specific class of AI-generated manipulation targeting faces and bodies
• Multimodal fake news detection — multimodal detection systems exploit visual inconsistencies and text-image mismatches to identify manipulated images
• Image forensics — forensic techniques for detecting manipulation evidence in pixels or metadata