Digitising film archives

Dust and scratch removal: infrared, software and restraint

Violet petals with pale centres and fine dark markings stand against green foliage.
Purple flower in close-up.Photograph by Robert Harper. Robert Harper Photograph Library, Archive.London. From the archive, shown for illustration.

A film scanner can see dust that the eye cannot, using light the eye cannot see. It works well on some films, badly on others, and archive standards say the result does not belong in the master file.

There are three ways to get dust and scratches out of a film scan. The first is not to let them in, by cleaning the film. The second is infrared detection, in which the scanner maps the defects with a lamp the picture is invisible to. The third is software or a retoucher working on the finished file. They are not interchangeable. Infrared is by far the most efficient where it works, and it does not work on every film. Software works on everything and cannot tell dust from detail. And for archive masters, the published standards advise using neither.

How does infrared dust removal work?

The method was developed by Applied Science Fiction, later Kodak's Austin Development Center, and sold as Digital ICE. A scanner equipped with it has two light sources: the normal one for the red, green and blue image, and an infrared lamp. The film is scanned once with each.

The trick lies in how film dyes behave. The cyan, magenta and yellow dyes of ordinary colour film are largely transparent to infrared, so in the infrared pass the photograph all but vanishes. Dust, hairs, fingerprints and scratches do not vanish: they block or scatter infrared as they do visible light. The infrared image is therefore a map of everything that is on the film but not part of the picture. Software fills each mapped area from its surroundings and leaves the rest of the image alone.

That last point is the method's great merit. Because the repair is confined to places where a physical defect was detected, clean areas of the picture are not softened. Software working on a finished file has no such map and has to guess.

Other makers built equivalents on the same principle: Canon's FARE, the infrared mode in LaserSoft's SilverFast software, called iSRD and introduced in 2008, and the "infrared clean" filter in VueScan. Flatbed scanners with transparency units often include an infrared channel as well.

Which films does infrared cleaning work on?

Film Infrared cleaning Why
E-6 colour transparencies Works well Dye image, transparent to infrared
C-41 colour negatives Works well Dye image
Chromogenic black-and-white (for example Ilford XP2 Super) Works Developed in C-41; the image is dye
Conventional black-and-white negatives Does not work The image is metallic silver, which blocks infrared
Kodachrome Unreliable Cyan dye absorbs near infrared; strong surface relief

Black-and-white film. In a conventional negative the picture is made of silver particles, and silver treats infrared much as dust does. The infrared pass shows the image itself, and the software sets about erasing it. Infrared cleaning must be switched off for silver film. The practical consequences are covered in how to scan black-and-white negatives.

Kodachrome. Kodachrome is a dye image, but an unusual one. The absorption of its cyan dye extends into the near infrared, making that layer partly opaque to the infrared lamp, and the film has a pronounced relief on its emulsion side. Both leak picture information into the defect map. The documented symptom is a slight loss of sharpness when infrared cleaning is used. A few scanners were engineered to handle it, notably Nikon's LS-9000 ED with its "Digital ICE Professional" mode, and some later software makes specific allowance for Kodachrome. Otherwise the choice is a light setting, checked carefully, or none.

Everything else. Even on well-behaved film, infrared repair can go wrong where fine, opaque picture detail sits against a dark ground. It is worth examining a few frames at full size, with and without cleaning, before running a batch. The compare two images tool shows the difference between two versions with a slider and a difference view.

What are the limits of infrared cleaning?

  • It must happen at capture. The Canadian Heritage Information Network's scanning guide stresses that infrared has to be applied during the scan. Once the film is back in its sleeve the defect map cannot be recreated.
  • It finds only what is on the film. Dust that was on the film in the camera, or on the original subject, is part of the photograph and stays.
  • Large debris defeats it. VueScan's own guidance is to blow film clean first, because particles can be too large for the infrared filter to repair convincingly. The larger the hole, the more the software has to invent.
  • Deep scratches may survive. A scratch that has removed emulsion has removed picture. Infrared can find it; nothing can know what was there.
  • Cameras do not have it. Camera-scanning rigs record no infrared channel, which is one of the real costs of that method, as camera scanning film: a practical set-up guide explains.
  • It slows the scan, since the film is read twice and the repair is computed.

What can software do without an infrared channel?

Two kinds of tool exist, and they should not be confused.

Automatic dust filters look for small, sharp, isolated specks and blur or replace them. They cannot distinguish a speck from a highlight on a diamond, a star, a catchlight in an eye or the grain of the film. Applied strongly, they smear texture across the whole frame. Applied lightly and locally, they have their uses on derivative files.

Manual spotting with a healing or clone brush is slow and accurate. A retoucher looks at each mark, decides whether it is a defect, and repairs it from nearby tone. For a selected picture going to print this remains the standard of quality. For fifty thousand frames it is unaffordable, which is why spotting is normally reserved for images that are actually being used.

Newer tools use machine learning to detect and fill defects. They can be effective, and they raise the same question as any generative repair: the filled area is a plausible invention, not a recovery. That distinction is explored in restoring faded colour slides digitally.

Should dust be removed from archival masters at all?

The cultural heritage standards say it should not. The US federal guidelines (FADGI, third edition, May 2023) hold that digitised material should document the appearance of the original at the time of capture. Their list of things to avoid in masters includes digital infill and the removal of blemishes in the original such as stains, dust or scratches, and they state that editing out artefacts should only be done to derivative files.

The Canadian guide takes a slightly different line, accepting infrared cleaning for colour transparencies because it is applied at capture and targets only the artefacts, while insisting that filters which alter the data should not be applied to an original scan afterwards. Its advice when unsure is to scan both with and without.

A workable policy for an archive reconciles the two:

  1. Clean the film physically before capture. See how to clean slides and negatives before scanning.
  2. Save an uncleaned master. If the scanner allows, save the infrared channel with it, so that cleaning can be done or redone later.
  3. Apply infrared or software cleaning to production copies only, and record that it was done.
  4. Spot by hand only the images selected for publication or print.
  5. State the policy in the scanning specification, so that every vendor quotes for the same thing. How to write a digitisation brief vendors can quote against covers the rest of that document.

When should a mark be left alone?

When it is evidence. A photographer's crop marks in wax pencil, a retoucher's dye on a transparency, a scratch that dates from the day of the shoot, a printer's registration notes on the rebate: all of these are part of the history of the object. An archive scan that removes them has removed information. Marks that carry no meaning, such as fresh dust from the scanning room, are a housekeeping failure and should be prevented, not painted out.

Dust, scratches and the Robert Harper archive

The film in the Robert Harper Photograph Library is mainly 6 x 7 colour transparency, with 35mm slides, black-and-white negatives and contact sheets, in an archive of 52,684 unique analogue photographs by its inventory's count, which has not been checked frame by frame against the physical material. Infrared cleaning is technically suited to the colour material and not to the silver negatives, so a scanning plan will need at least two dust policies. The film stocks are not recorded in the inventory, and the surface condition of the film has not been surveyed item by item. More than half of the archive is still on film and has yet to be scanned.

The holdings are described in the analogue collections and itemised in the archive inventory. Scanning vendors who can supply both untouched masters and cleaned derivatives are invited to make contact through the digitisation partners page.

Frequently asked questions

How does Digital ICE work?

The scanner makes two passes, one with ordinary light and one with an infrared lamp. Colour film dyes are largely transparent to infrared, while dust, hairs and scratches block or scatter it. The infrared pass therefore shows only the defects. Software uses it as a map and fills each marked area from the surrounding picture, leaving clean areas untouched.

Why does infrared dust removal not work on black-and-white negatives?

Conventional black-and-white film holds its image in metallic silver, and silver blocks infrared in much the same way as dust. The infrared pass shows the whole picture as a defect, and the software damages the image trying to repair it. Chromogenic black-and-white films, processed in C-41 colour chemistry, contain dye instead of silver and work normally.

Does infrared cleaning work on Kodachrome?

Imperfectly. Kodachrome's cyan dye absorbs into the near infrared and the film has a pronounced surface relief, so the infrared channel picks up parts of the image as well as dust. The result can be a slight loss of sharpness. Some scanners and software were designed to cope with it, and many operators use a lower setting or none.

Should dust removal be applied to archival master scans?

Standards say no. FADGI's guidelines ask that the master document the original as it was at capture and list removal of dust and scratches among edits to avoid there, reserving them for derivatives. If infrared cleaning is wanted, save an uncleaned master as well, or keep the infrared channel, so the repair can be reviewed or redone.

What is the best way to remove dust from film scans without infrared?

Prevention first: clean film and a clean scanner. After that, manual spotting with a healing or clone tool on a copy of the file is the most reliable method, because a person can tell a speck from a star or a catchlight. Automatic dust filters work by blurring small details and should be used lightly, on copies, and checked at full size.

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