duster

‘duster’ is a shiny webapp focused on removing large scale dust from old photographs. If you can’t face clicking on each of scores of dust spots in each of many images with some ‘heal’ tool, or your favourite photo tool leaves ‘dusted’ images looking widely blurred, then ‘duster’ might help.

Duster shows you exactly where it has detected dust. It only changes the pixels there.

Quick ‘how-to’

The quick steps are as follows. There is more explanation of each step in the following section.

1. Upload a jpeg. It will be shown, together with the dust detected using the default settings.
2. If there is a ‘black’ border, it will be mostly removed. If too much is being removed, reduce the edge crop strength.
3. Check the dust. If not enough is being found, increase the detection radius or reduce the detection threshold. It’s likely that some larger pieces will escape.
4. Check the dust again. If it is showing real structure from the image (eg mouth, fabric texture), increase the detection threshold.
5. If dust is found, but the result still looks dark or grey where the dust spot is, try increasing the replacement radius.
6. If you’re finding fine ‘rings’ like water droplet stains around dust, try fattening the dust.
7. When you’re happy, download the image to your browser’s default download folder.

Reset: returns to default values.

More details

The final image is shown on the left, the dust is shown on the right. If you choose that option, the original is shown below.

Border removal

This step removes any ‘plain’ border from the image, such as a black border from a scanned slide. A setting of 0 means that any edge row or column of pixels that are identical in intensity will be removed. (These could be all identical grey, or all black, say.)

In practice, particularly if the image is jpeg, even if it looks like the border is just black, it won’t be: there will be pixels very near black, but not black. Increasing the tolerance will allow for that ‘insignificant’ variation. What’s insignificant will vary between photos.

Dust in the final image

The dust that has been found is shown as bright dots on a black background. They will be white for black and white images, or colour for colour images. duster only changes the image for those bright pixels, the black areas will be unchanged.

If there is ‘obvious’ dust in the final, for example in patches of sky or skin, check the dust pattern. If there is no bright spot in the corresponding place, increase the detection radius or reduce the detection threshold.

Non-dust in the dust image

Increasing the detection radius is likely to bring real image structure into the dust image. Real, fine structure like fabric patterns, lines that are small gaps (between fingers, between lips), telegraph poles will appear in the dust image. If this non-dust is not very bright in the dust image, then increasing the threshold will remove it. Otherwise you need to find a detection radius that is a balance between finding dust and not finding real image features.

Dust remains grey

The pixels of ‘dust’ are replaced by the median pixel nearby. Normally, this will give a reasonable, non-dust value. Occasionally you might need to increase the radius for this median averaging to get a better result. (This is the slowest step of the process.)

‘Water stains’ around dust

Particularly if the source is a jpeg, the dust will have been smudged, and this can also produce a slightly-brighter ‘ring’ around each piece of dust. A clunky solution is to ‘fatten’ the dust: add extra pixels around each dust spot for replacement.

License & Acknowledgements

‘duster’ is subject to the license. The code is available on github.

It makes heavy use of the shiny and magick packages. Consult their documentation for the licenses which apply.