Remove scratches from old photos

Dust on a scan is easy to see and tedious to paint: hundreds of specks, a hair, a scratch from a sleeve. The proposal does the tedious part. A pseudo-median seven pixels across erases anything up to about three pixels wide; whatever it erased, beyond the threshold, is a candidate defect. Candidates are then kept only if they are small (a speck) or long and thin (a scratch or a hair).

That shape test is what keeps texture out — grain, gravel and foliage pass the difference test everywhere and fail the shape test — but it is a heuristic, and stars in a night sky, catchlights in eyes and sparkle on water look exactly like dust to it. So it proposes and you review: take back anything that belongs to the picture with the restore brush, then erase.

1. 1Open the scan. The proposal runs straight away. The live line says how many specks and scratches it found.
2. 2Look at what was proposed. Zoom to 1:1 and pan. Catchlights, stars and highlights are the usual false finds.
3. 3Take back what belongs to the picture. Restore brush, or Take out with the rectangle.
4. 4Adjust the threshold. Lower finds fainter dust; higher finds only the obvious. Press Dust & scratches again to re-propose.
5. 5Erase. The edge march closes each thin defect from its edges. A large tear is better re-rolled with the patch search.

• It cannot tell dust from a star, a catchlight or a sparkle on water. That is why it only proposes.
• Scratches wider than a few pixels and torn corners are removals, not dust — paint them and use the patch search.
• On a heavily textured scan most of the frame passes the test; the page says so and proposes nothing rather than marking the whole picture.
• Faded colour, mould and silvering are not defects it looks for.