Research integrity

Reading figure integrity signals without over-claiming

Error level analysis, copy-move detection and noise profiling tell you where to look. They do not tell you that an image is fraudulent — and the difference matters.

Last updated 8 May 2026 · 6 min read

The Figure Integrity service runs several independent forensic passes over an uploaded figure: error level analysis to surface regions recompressed differently from their surroundings, copy-move detection to find duplicated blocks within one image, noise profiling to spot inconsistent sensor noise, and perceptual hashing to compare a figure against others you have scanned.

Each pass produces a signal, not a verdict. A western blot band that lights up under copy-move detection may be a splice, or may be a genuinely repeated loading control. A high ELA response often just means the figure was saved twice.

The right way to use the output is as triage: it tells an editor or reviewer which panel to look at first and what to ask the authors for. The raw data — original acquisition files, uncropped gels — settles the question, not the scan.

We deliberately word the editorial assessment in terms of what to request next rather than what happened, because an integrity accusation based on a compression artefact damages a career for nothing.

← All articles