Restoring Old Photographs: Where Recovery Ends and Invention Begins

Restoring an old photograph looks like a technical task: remove the scratches, even out the contrast, bring back sharpness. In practice it hides two fundamentally different operations, and telling them apart matters.

Recovery versus invention

Recovery is working with information that is present in the image but hard to see. Faded contrast can be pulled back: the data never left, it is merely compressed into a narrow range. Dust and scratches can be removed: the image survives around them.

Invention is filling in what is not there. A torn corner, a face faded to white, a lost fragment. No algorithm "recovers" these: it invents plausible content.

The distinction is fundamental for a family archive: in the first case you get a more legible original, in the second a new image partly based on the original.

What current tools do

• Damage removal: scratches, creases, stains, glue marks. Works well, especially on backgrounds and clothing.
• Contrast and tone recovery: almost always safe — the information is there.
• Sharpening and detail enhancement: partly recovery, partly invention — the line depends on how far it goes.
• Colourisation: entirely invention. The colour of clothing and eyes is not stored in a black-and-white photograph.
• Reconstructing lost fragments: entirely invention.

The order of work

1. Digitisation. A scanner, not a phone: reflections and keystone add artefacts that later cannot be separated from the image.
2. Preserve the original. The scan file is never edited — everything happens on copies.
3. Geometry: correct skew, crop damaged borders.
4. Damage: remove scratches and stains, starting with background and finishing with the face.
5. Tone and contrast: expand the range without clipping.
6. Sharpness: carefully, stopping before halos appear.

The face is the last and most careful zone

Everything done to a face affects recognisability. Aggressive sharpening changes features, heavy noise reduction erases micro-texture, automatic "enhancement" pulls proportions towards statistically expected ones.

A practical rule: do the minimum on the face and always compare with the original at the same scale. If the person looks different after processing, the processing went too far.

Colourisation

Colourising a black-and-white photograph is the most visible and most contested operation. The algorithm assigns colours statistically: sky blue, grass green, skin flesh-toned. The colour of a dress, of eyes, of the wall behind is a guess.

That does not make colourisation pointless: it restores emotional accessibility, and many families value exactly that. But noting "colours are approximate" is honest and costs nothing.

Working with several copies

If one photograph survives in several prints of differing condition, that is a gift: damaged areas of one can be taken from another. Such restoration is genuine, with no invention.

The same applies to frames from one series: a neighbouring shot shows what the fabric texture or background looked like and tells you what to rebuild.

Storing the result

A practical scheme for a family archive: keep three files — the untouched scan, the restored version, and the colourised one if made. Label which is which.

In twenty years nobody will remember which details were on the original and which were painted in. A label solves that permanently.

Realistic expectations

A small group photograph will not become a portrait: enlarged faces will be invented. A badly faded frame will not regain its original contrast without loss. A lost fragment will be filled plausibly but not correctly.

Restoration makes an archive legible and printable. It does not turn it into modern photography — and should not.