Digitising film archives

Restoring faded colour slides digitally: honest or invented?

Hedgerows divide green and golden fields into irregular shapes.
Fields from the air.Photograph by Robert Harper. Robert Harper Photograph Library, Archive.London. From the archive, shown for illustration.

A faded slide still contains most of its picture, out of balance. Putting the balance back is arithmetic. Putting back what the dye no longer holds is guesswork, and the two should never be saved under the same name.

A faded colour slide can be made to look right again in a digital copy, and in most cases the method is plain arithmetic: the three dyes have weakened by different amounts, so the three colour channels of the scan are rescaled by different amounts. That is an honest correction, because it works only with what the film still holds. Beyond a certain point there is nothing left to rescale, and any colour or detail that appears has been supplied by the software or the operator. The skill in restoration is partly technical and partly a matter of knowing which side of that line a given picture is on, and saying so.

What has actually happened to a faded slide?

A colour transparency is three layers of dye: cyan, magenta and yellow. Each dye breaks down at its own rate, mainly under the influence of heat, and the Image Permanence Institute is categorical that the process is irreversible: when a dye fades, the image information it carried is gone from the film. The causes are set out in why colour slides fade.

Two things follow. Density falls, so the slide looks thin and flat. And because the dyes do not fall together, the balance shifts. If cyan and yellow have gone faster than magenta, the picture turns magenta or purple. The damage is uneven between layers but, within a layer, broadly proportional: Rudolf Gschwind and Franziska Frey, in a 1997 paper from the University of Basel and the Image Permanence Institute, found that dyes normally fade in proportion to their original density, and that fading can be described as a linear transformation of the three dye densities.

That proportionality is the reason correction is possible at all. The picture is still there, written more faintly in some layers than in others.

What can be recovered honestly?

Whatever dye remains can be rebalanced. In a scan, cyan dye is seen mainly by the red channel, magenta by the green and yellow by the blue. Stretching each channel so that its darkest and lightest values return to full range undoes most of a uniform fade. The Image Permanence Institute describes the result accurately: digital colour restoration can approximate the original colour balance.

The Basel work went further than simple stretching. It fitted a matrix of twelve coefficients to the behaviour of real dyes, measured in accelerated ageing tests, so that the side absorptions of each dye and the staining of the film base were corrected as well. It also called for at least ten bits per channel in the scan. The reason is easy to see: stretching a weak signal widens the gaps between its tonal steps.

What this class of correction does is put existing information back into proportion. Nothing is added. Two operators using the same method on the same scan will arrive at nearly the same picture, and that repeatability is a fair test of honesty.

Where does correction turn into invention?

At four points, in rising order of seriousness.

When a dye has gone. Gschwind and Frey state the limit plainly: if the dyes of a photograph are completely destroyed, reconstruction is no longer possible. A layer with no density left has no gradation to stretch. Software can still paint the area a believable colour, but that colour comes from somewhere other than the slide.

When there is no reference. The same paper notes that most photographs contain no known reference colours, and that with a single slide and no reference there are too many degrees of freedom. Was the wall cream or white? The corrected file gives an answer. It may not be the right one.

When damage is uneven. Stains, spots, mould and patchy fading do not follow the proportional model, and the paper notes they need retouching. Retouching is local invention by definition.

When a generative tool is used. Machine-learning restoration does not rescale the film's own data. It replaces weak or missing areas with content predicted from other images. It can turn a ruined slide into a handsome picture. It can also give a product the wrong label, a face the wrong eye colour or a fabric a pattern it never had. For a family album that may be acceptable. For an archive whose value lies in being a record, the output is an illustration and has to be labelled as one.

How do you restore a faded slide step by step?

  1. Scan faithfully. Sixteen bits per channel, no automatic colour, no clipping. The master should look as faded as the slide. If the scanner is profiled, so much the better: see colour management for film scanning.
  2. Work on a copy. The master is never overwritten.
  3. Look at the histogram. In a faded slide the three channels occupy different ranges. The image histogram tool shows them side by side.
  4. Set each channel's end points. Move the black and white points of red, green and blue separately to the ends of their own data. Use the unexposed film edge as black if it is in the scan.
  5. Neutralise. Find something that should be grey or white and remove the remaining cast, as the white balance correction tool does with a click.
  6. Restore contrast and saturation modestly. Fading lowers both; over-correction is the commonest fault.
  7. Compare. Put master and correction side by side, for instance with the compare two images tool, and ask whether every change can be justified by the film.
  8. Save separately and describe it. A different filename, and a note of what was done.

For quick results on viewing copies, the restore faded photo colours tool carries out steps 4 to 6 in the browser: it stretches each colour channel separately, with sliders for strength, contrast and saturation, and works in batches. Its limit is the limit of the method: it works on the scan and cannot recover detail that is gone. It outputs 8-bit files, so it is for access copies, not masters.

Why should the master file be left faded?

Because the master is the record, and corrections are opinions about the record. The US federal digitisation guidelines (FADGI, third edition, May 2023) say that digitised material should document the appearance of the original at the time of capture. Their list of practices to avoid in masters names colour restoration and digital infill directly, and they add that profiling a scanner cannot restore colour that is no longer present.

There is a practical argument too. Correction methods improve. A file that preserves the faded state at full bit depth can be reinterpreted in ten years with better tools and better knowledge of the film. A file that was "fixed" in the scanner preserves only today's guess.

How should restored versions be labelled?

An archive needs at least three tiers, and a caption or metadata field that says which one a viewer is looking at.

Tier What it is What may be said of it
Master The film as scanned, unadjusted "Scan of the original transparency in its present condition"
Corrected derivative Channels rebalanced using only the film's own data "Colour corrected for dye fading"
Reconstruction Retouched, infilled or generatively restored "Digitally reconstructed; colour and detail partly interpreted"

The middle tier is what most publication calls for. The third has its uses, in exhibition graphics or a book cover, provided nobody mistakes it for the first. Recording these distinctions in each file's metadata is part of ordinary cataloguing, described in IPTC photo metadata fields for archives.

Restoration and the Robert Harper archive

The Robert Harper Photograph Library's principal analogue format is 6 x 7 colour transparency film, within 52,684 unique analogue photographs by the inventory's count, which has not been checked frame by frame against the physical material. More than half of the archive is still on film and has yet to be scanned. The colour condition of the film has not been surveyed item by item, so no claim is made here that any of it has faded or that it has not. The principle will apply either way: masters record the film as found, and any correction is made on a copy and described.

The holdings are described in the analogue collections and listed in the archive inventory. Scanning laboratories, colourists and conservators with experience of dye-fade correction are invited to make contact through the digitisation partners page.

Frequently asked questions

Can faded slides be restored?

The film cannot be, but a scan of it can be corrected. Fading leaves the three colour dyes out of balance, and adjusting the red, green and blue channels of the scan separately brings the balance back towards the original. The correction works with whatever dye remains. Where a dye has gone completely, there is nothing to rebalance and the colour cannot be recovered.

How do I fix the red or magenta cast on old slides?

Scan at 16 bits per channel without automatic correction, then, on a copy, set the black and white points of each colour channel separately so that all three span the same range. That removes most of the cast. Fine-tune with a white balance adjustment on something that should be neutral, then restore contrast and a little saturation. Keep the uncorrected scan.

Is AI photo restoration accurate?

It is convincing, which is a different thing. Generative tools fill in colour, texture and detail from patterns learned on other photographs, not from the film in front of them. The output may look right and be wrong about the colour of a dress or the lettering on a label. Use it for illustration if you wish, label it clearly, and never treat it as evidence.

Should a faded slide be corrected in the scanner software?

No. Scanner corrections are applied before the file is saved and cannot be undone or inspected later. Standards for archival work ask for a master that records the film as it is, faded colour included. Make the correction afterwards on a separately saved copy, where it can be adjusted, compared with the master, and redone as tools improve.

How can I tell what colour a faded slide originally was?

Look for references inside and outside the picture. Inside: things that should be neutral, such as white paper, grey card, chrome or concrete, and the black of the film edge. Outside: a print or publication made from the slide when it was new, an unfaded duplicate, or other frames from the same roll. Without any reference, the correction is an informed estimate and should be described as one.

Sources

Published by Archive.London. Corrections are welcome under our editorial policy.

Help digitise the archive

52,684 analogue originals, mostly medium-format transparencies, are waiting to be scanned and catalogued. We are speaking to scanning, cataloguing and tagging partners.

See the digitisation briefWrite to the archive