Dust map#

Sensor dust appears in every frame at the same place, as a soft dark spot. A focus stacker does not merely keep it — it makes it worse than in any single frame, because the region-energy rule reads the spot's edge as detail and picks it. The spot comes out sharper than it ever was.

There is a second, quieter problem: a fixed pattern present in every frame is exactly what pulls an intensity-based registration toward zero shift. Dust degrades your alignment as well as your image.

lapstack's answer is Helicon Focus's: shoot a dust map, and take the spots out of every frame as it is decoded, before alignment.

Shooting the map#

One frame of an evenly lit, featureless surface, thoroughly out of focus, at the stack's aperture:

• A white wall, the sky, or a sheet of paper
• Defocused enough that the surface itself contributes nothing
• The same aperture as the stack, because dust shadows sharpen and darken as you stop down, and move slightly

That frame shows nothing but the dust.

Using it#

In the browser#

The Dust map section of the parameter panel. Load the frame; the spots are found in the worker and outlined on a preview (click the preview for a larger view in a new tab). The threshold, margin and mode take effect immediately, so you can tune them against the preview.

The map then applies to every frame the engine decodes: the run, the DFR and WAV renders, the slabs, the Source view and the retouch brush. The filmstrip's thumbnails show the frames as shot, unretouched, so you can compare.

A project file records the map's name, and the file is matched from the frames added when the project is opened again.

On the command line#

lapstack --dust-map dust.tif --save-dust-map spots.png -o stacked.tif frames/*.tif

The controls#

How spots are found#

The map's luminance is taken at half resolution — one REDUCE, which also tames the noise — and divided by its own large-scale background. That background is a plane fitted under each cell four pyramid levels up, by a first-order normalized convolution, so the illumination's falloff is followed right out to the frame's edges while a small spot is not. It is estimated twice, the second time with the first pass's spots weighted out.

A pixel darker than that background by more than the threshold is dust. The connected components, grown by the margin, are the spots. A component under 16 pixels is treated as noise, and a blob wider than a quarter of the frame is not dust and is reported as such.

The run logs the spots found — count, size, and share of the frame.

fill or flat#

fill (the default, and Helicon's way) interpolates each spot from its surroundings by pull–push: the window's pyramid is built with the dust weighted out, and on the way back down every hole takes the coarser level's value. The result is a smooth patch that meets its edges cleanly. Whatever detail was under the spot is gone, replaced by a plausible continuation.

flat divides the spot by the attenuation the map measured — a flat-field correction. It keeps whatever detail lies under the spot, which is strictly better when it works. It works when the map was shot at the stack's aperture and lighting; when it was not, you get a ring.

Use fill unless you shot a careful map and the spots fall on detail you care about.