Appearance
Documenting scientific imaging paradata means recording the process behind every image — the wavebands, settings, calibrations and processing decisions — not just describing the object. The best practice is to capture this at the moment of work into a structured, machine-readable record that travels with the masters, because paradata is what lets a future scholar reproduce your result, distinguish a real feature from an artefact, and decide whether to trust the reading. Metadata says what the file is; paradata says why you should believe it.
What is paradata and why does imaging need it?
Metadata answers "what, who, when." Paradata answers "how and why": which lamp, what exposure, which registration method, which PCA component, what contrast stretch. In multispectral and scientific imaging the final picture is a constructed visualisation — false colour and PCA deliberately reshape the data — so without the recipe, a convincing image is indistinguishable from an artefact. Paradata is the difference between evidence and decoration. The London Charter for computer-based visualisation makes exactly this point: process transparency is what gives a result authority.
What is the minimum you must record?
Treat the following as non-negotiable for every capture:
text
[ ] Object & folio identifier
[ ] Every waveband: wavelength + illumination source + mode (UVF/UVR/IRR...)
[ ] Exposure, aperture, ISO per band
[ ] Capture geometry: distance, angle, polarisation state
[ ] Calibration: white reference, dark frame, colour target used
[ ] Equipment: camera, lens, filters, lights (make/model)
[ ] Software + versions (capture and processing)
[ ] Processing steps in order, with parameters
[ ] Operator and dateIf any line is missing, the capture cannot be reproduced or defended — which in research practice means it cannot be cited.
Where should paradata live?
Spread it redundantly so it survives copying:
| Layer | What goes there | Format |
|---|---|---|
| Embedded | Camera/exposure basics | EXIF / XMP in the TIFF |
| Sidecar | Bands, calibration, pipeline | JSON or CSV next to masters |
| Project | Conventions, equipment, intent | README / data dictionary |
A note in a paper lab book is the first thing to vanish. Records co-located with the data, in formats both people and scripts can read, are the ones that endure.
How do you structure a machine-readable record?
A compact JSON sidecar per object scales to a whole collection and can be validated:
json
{
"object": "MS_0042", "folio": "017v",
"captured": "2026-02-11T10:24:00Z", "operator": "E. Reed",
"equipment": {"camera": "modified full-spectrum body",
"lens": "60mm apo", "lights": "narrowband LED panel"},
"bands": [
{"id": "b03", "wavelength_nm": 365, "mode": "UVF", "exposure_ms": 4000},
{"id": "b09", "wavelength_nm": 940, "mode": "IRR", "exposure_ms": 250}
],
"calibration": {"white": "white_20260211.tif", "dark": "dark_20260211.tif"},
"processing": [
{"step": "flat_field", "tool": "ImageJ 1.54"},
{"step": "register", "method": "SIFT"},
{"step": "pca", "component": 2, "tool": "scikit-learn 1.4"}
]
}Keep a JSON Schema for it so a quick validation run catches missing fields before deposit.
How do you document an enhancement honestly?
Every enhancement must be a recipe someone else can replay. Record input bands, alignment, statistical transform, parameters and final contrast, and always retain the unenhanced master alongside. The test of honest paradata: a colleague with your masters and your record should regenerate your published image pixel-for-pixel. If they cannot, the documentation is incomplete and the claim is unverifiable.
How do you keep it consistent across a collection?
Standardise the vocabulary up front. Borrow event semantics from PREMIS, capture parameters from FADGI or Metamorfoze, and process principles from the London Charter, then write a one-page data dictionary defining every field and allowed value. Consistency is what turns a hundred individual records into a queryable, comparable dataset rather than a hundred private notations.
Key Takeaways
- Paradata documents process (how and why); metadata documents the object (what and when).
- Constructed visualisations like false colour and PCA are untrustworthy without the recipe.
- Record bands, illumination, exposure, geometry, calibration, equipment, software and every processing step.
- Store paradata redundantly: embedded EXIF/XMP, a JSON/CSV sidecar, and a project README.
- Make it machine-readable and validate it against a schema before deposit.
- An enhancement is honest only if it is reproducible from the master plus the recorded pipeline.
- Standardise on PREMIS, FADGI/Metamorfoze and London Charter vocabulary, with a data dictionary.
Frequently Asked Questions
What is paradata, and how does it differ from metadata?
Metadata describes the resource (what it is, who made it, when). Paradata records the process that produced it — the decisions, settings, calibrations and reasoning behind a capture or enhancement. Paradata is what lets someone judge whether to trust a result.
What is the minimum paradata for a multispectral capture?
At minimum: every waveband and illumination source, exposure per band, capture geometry, calibration frames used, software and versions, and each processing step applied (registration, flat-field, PCA, contrast). Without these a result cannot be reproduced or challenged.
Where should paradata live so it does not get lost?
Embed what you can in the file (EXIF/XMP) and keep a human- and machine-readable sidecar (JSON or CSV) next to the masters, plus a project-level README. Redundant, co-located records survive copying far better than notes in a lab book.
Do I need paradata if the image looks self-explanatory?
Yes. An enhanced false-colour or PCA image looks convincing precisely because processing shaped it; without the recipe nobody can tell revelation from artefact. Paradata is what separates evidence from a pretty picture.
Is there a standard for imaging paradata?
There is no single mandated schema, but you can align with PREMIS for events, FADGI/Metamorfoze for capture parameters, and the London Charter principles for documenting process. Pick a vocabulary, apply it consistently, and define your terms.
How do I document an enhancement so it stays honest?
Record the exact pipeline — input bands, alignment method, statistical transform, parameters and contrast settings — and keep the unenhanced master beside it. Anyone should be able to regenerate your enhanced image from the master and the recipe.