Answers are grounded in this paper and cite their sources.
Protocol and data of a PRISMA 2020 systematic literature review of image forgery detection and localization across manipulated, face-forged, AI-generated, and document images (2015-2026). Timing statement: the protocol (1_protocol.zip) was written before any screening decision, but was deposited here on 15 November 2026, after screening and selection; the review was therefore not registered before screening. The protocol files are released as written, with only placeholders and administrative details completed. All deviations from the protocol, including those made after screening began, are listed in AMENDMENTS.md. Contents: protocol v2.0, search strings, screening and extraction guides, codebook and seed corpus (1_protocol.zip); deduplicated records (20,523), stage-1 and stage-2 screening decisions, the selection exported in Rayyan format, and the final core set of 100 studies (2_search_and_screening.zip); extraction sheet, quality scores and independent re-extraction (3_extraction_and_quality.zip); Python scripts (4_scripts.zip). Abstracts and keywords from bibliographic databases are not redistributed. See README.md. Version 2.1 (15 November 2026): cue family recoded for 18 core studies so that C5 covers only features of pre-trained models (extraction files regenerated); open-sources-only full-text access recorded as a deviation from EC6; the 20 top-ranked 2025-2026 studies outside the core set added; stage-1 recall on the 98 seed studies reported; scripts and figure data updated. The protocol is unchanged. Details in README.md and AMENDMENTS.md. Version 2.2 (15 November 2026): records of each source at every selection step added (figure data, figure, and script); no data, decision, or protocol file changed.
No takes yet. Share an insight, caveat, or question.
Ali et al. (2026) studied this question.
Synapse has enriched 8 closely related papers on similar clinical questions. Consider them for comparative context: