Getting the photograph off the paper

Resolution: scan for the output, with room

A flatbed scanner’s dpi setting is real, unlike a JPEG’s ppi tag: it determines how many samples are taken per inch of the original.

The arithmetic you need:

• A 6×4 inch print reprinted at the same size needs 300 dpi minimum. Scan at 600 so you have room to crop, straighten and correct.
• A 6×4 print to be enlarged to A4: that is roughly a 2.5× enlargement, so scan at 800 to 1200 dpi.
• A 35 mm negative or slide is about one inch across, so a usable file needs 2400 to 4000 dpi, and that requires a scanner with a transparency unit. A flatbed’s stated 9600 dpi is usually interpolated marketing rather than real optical resolution; look for the optical figure.
• Above the scanner’s optical resolution there is no gain, only file size. The sensor cannot sample more finely than it can sample.

Scan in 16-bit if the option exists. An old print is often faded and low in contrast, which means the correction will be a heavy stretch, which is exactly the case for extra bit depth.

Save the raw scan as a TIFF and never overwrite it. That file is the closest thing to the original that will exist once the print fades further.

Moiré, and the print that was already printed

Scanning a photographic print from a lab is straightforward. Scanning something from a magazine, newspaper or book is not, because that image is made of a regular grid of halftone dots.

When the scanner’s sampling grid meets the printing screen’s grid, the two interfere and you get moiré: coloured wave patterns across the image. This is the same aliasing mechanism as the demosaic moiré from the first module, and it is not a fault.

Three fixes, in order of quality:

1. Scan at a much higher resolution than you need, then downsample. Scanning at 1200 dpi and reducing to 300 averages many samples per printed dot, which removes the interference properly. This is the best result and the largest files.
2. Use the scanner software’s descreen setting, naming the original type (newspaper, magazine, book). It applies a targeted blur. Fast, and it costs sharpness.
3. Rotate the original a few degrees on the glass, scan, and rotate back in software. Changing the angle between the two grids breaks the interference pattern. Crude, effective, free.

In the editor, a small Gaussian blur followed by sharpening also suppresses residual moiré, and a Median filter at 1 to 2 pixels is often better than a blur because it removes the dot pattern while keeping edges.

Dust and scratches

Clean the glass and the print before anything else. A microfibre cloth and, for the print, a soft brush. Two minutes here saves an hour of healing.

Infrared dust removal (sold as Digital ICE, FARE, or similar) uses a fourth infrared channel to detect physical defects, because dust and scratches block infrared while the image dyes do not. It is excellent and it has two exceptions that catch everybody: it does not work on traditional silver black-and-white negatives, or on Kodachrome, because in both the image itself is opaque to infrared and gets detected as damage. Turning it on for those produces a smeared mess.

For prints, infrared removal generally does not apply: it is a film-scanning feature. Prints are cleaned by hand in the editor.

Flatness, and photographing instead

A curled print will not sit on the glass and scans with a soft band. Put it under a heavy book overnight, or scan with the lid pressed down and a weight on top.

No scanner? A phone works, and the technique matters:

• Even light, no direct source. Two lamps at 45 degrees from opposite sides, or an overcast window, or outdoors in open shade. A single overhead light produces a glare patch.
• Camera square to the print. Tilt introduces keystone which then has to be corrected, and correcting it resamples.
• Avoid the phone’s own shadow, which is why the light comes from the sides.
• Use the main camera, not the ultrawide, whose distortion bends the print’s edges.
• Shoot DNG if you can, for the bit depth on a faded original.
• Crop and correct perspective afterwards. Most phone galleries and Adobe Scan do this automatically, and doing it manually with a perspective transform gives a better result.

A phone photograph in good light beats a bad scan, and both beat a photograph taken at an angle under a ceiling light.

Free software for scanning

XSane and Simple Scan on Linux, NAPS2 on Windows, and VueScan across platforms (paid, and it supports very old scanners whose manufacturers abandoned drivers years ago, which is often the only thing that will).

GIMP can drive a scanner directly through XSane. For negatives, darktable has a negadoctor module that inverts and corrects colour negative film scans properly, accounting for the orange mask: this is a hard job done well, and free.

What the scan cannot recover

A scan captures what is on the paper now. If the print has faded, the scan captures the faded version; if it has yellowed, the scan captures the yellow.

Some of that is correctable, which is the next lesson. Some is not: where a dye has faded to nothing, there is no residual signal to amplify, and lifting that area produces noise and paper texture rather than a picture. A print left in a window for thirty years may have a corner with no image left.

And resolution cannot exceed what the print holds. A small print from a 1970s compact camera on cheap paper may contain very little detail; scanning it at 2400 dpi produces a large file of magnified paper grain. Scan it at 600, and spend the time on the tonal restoration instead.