Appearance
Faded iron gall ink can be recovered with multispectral imaging because the visible brown colour disappears long before the chemistry does — iron residues and substrate damage remain in the page and respond differently to ultraviolet and infrared light than blank parchment. The reliable route is: capture a calibrated multi-band stack, register the bands, flat-field them, then visualise with ultraviolet fluorescence plus PCA. Below is the full step-by-step with sensible defaults and the mistakes that waste a session.
Why does faded ink still leave a trace?
Iron gall ink is acidic and oxidises over centuries; the chromophore that makes it black breaks down, so the eye loses the text. But the iron itself, and the localised degradation of the parchment under each stroke, persist. Under 365 nm ultraviolet excitation, healthy parchment fluoresces brightly while ink-affected areas quench that fluorescence — the writing returns as dark strokes on a luminous page. This is why a folio that looks blank can yield a clean reading in the right band.
What do I need before capturing?
- A camera with response across UV, visible and near-IR (often an astro-modified or filter-swappable body).
- Narrowband or filtered illumination at several wavelengths, including a 365 nm UV source.
- White and dark reference frames for flat-fielding.
- A rigid copy stand so geometry does not shift between bands.
If you lack a dedicated rig, a modified DSLR with UV-pass and IR-pass filters and steady lighting recovers a surprising amount; the constraints that matter are stable geometry and calibration, not brand.
How do you capture and calibrate the stack?
Shoot each band without moving the page, plus a flat-field set:
text
1. Mount folio flat; lock focus and aperture (do not refocus between bands).
2. Capture: UVF 365nm, UVR, blue, green, red, IR ~850nm, IR ~940nm.
3. Capture white reference (Spectralon/white card) and dark frame per band.
4. Flat-field each band: corrected = (raw - dark) / (white - dark)Flat-fielding removes the lamp's uneven illumination, which otherwise reads as a false brightness gradient that hides edge text.
How do you register the bands?
Because each exposure is separate, sub-pixel shifts creep in. Align before any analysis. In Fiji, Plugins > Registration > Linear Stack Alignment with SIFT, or in Python:
python
from skimage.registration import phase_cross_correlation
from scipy.ndimage import shift
offset, _, _ = phase_cross_correlation(reference_band, target_band, upsample_factor=20)
aligned = shift(target_band, offset)Unregistered bands turn into coloured ghosting that masquerades as ink — fix alignment first, always.
Which visualisation actually reveals the text?
Run two approaches and compare:
| Method | Strength | When to prefer |
|---|---|---|
| UV fluorescence band alone | Often instantly legible | First thing to check |
| False colour (UV/IR to RGB) | Interpretable, fast | Sharing a clear reading |
| PCA on the full stack | Pulls faintest strokes from noise | Severely faded passages |
A common finishing move is to compute PCA, pick the component where the script stands out, and blend it over a false-colour base:
python
from sklearn.decomposition import PCA
flat = stack.reshape(stack.shape[0], -1).T # pixels x bands
comps = PCA(n_components=4).fit_transform(flat)
pc2 = comps[:, 1].reshape(stack.shape[1:]) # often the text componentHow do you stay honest about the result?
Recovery reveals; it does not invent. Never paint in strokes, and never present an enhanced image without the unenhanced master alongside. Record every processing step — bands used, registration method, PCA component, contrast settings — as paradata so the reading is reproducible and a palaeographer can challenge it. An enhancement nobody can verify is not evidence.
Key Takeaways
- Faded iron gall ink survives chemically even when invisible; UV and IR reveal it.
- Start with 365 nm UV-induced fluorescence — degraded ink quenches it and reads as dark text.
- Lock geometry and flat-field every band before doing anything else.
- Register bands with SIFT or phase cross-correlation to avoid colour ghosting.
- Compare UV fluorescence, false colour and PCA; blend the best PCA component for legibility.
- Imaging is non-destructive revelation, not restoration — never invent strokes.
- Publish the unenhanced master and full paradata so the reading can be verified.
Frequently Asked Questions
Why does iron gall ink fade in a way that imaging can recover?
The visible brown-black colour fades as the ink degrades, but iron compounds and the substrate change remain in the page. Those residues respond differently to ultraviolet and infrared light than the surrounding parchment, so the lost text reappears in non-visible bands even when the eye sees nothing.
Which wavebands work best for faded iron gall ink?
Ultraviolet-induced fluorescence (around 365 nm excitation) and the blue-to-ultraviolet reflectance range are usually most effective, because degraded ink quenches fluorescence and shows up as dark text on a glowing parchment. Test the full band set, but start there.
Do I need a full multispectral rig or can I improvise?
A full multispectral system gives the cleanest result, but a modified camera with UV and IR filters plus controlled lighting recovers a great deal. The non-negotiables are stable geometry between bands and a flat-field calibration.
Should I use false colour or PCA for the recovery?
Try both. False colour is quick and interpretable; PCA often pulls the faintest strokes out of noise. Many recoveries end with a PCA component blended over a false-colour base for legibility.
Is image processing the same as restoration?
No, and the distinction is ethical. Imaging reveals what survives without touching the object; it must never invent strokes. Always keep and publish the unenhanced master so others can verify the reading.
What ruins a faded-ink recovery most often?
Misregistration between bands and uneven illumination. Both create artefacts that mimic or obscure faint text, so flat-fielding and band alignment matter more than any clever filter.