hello jim, thank you for your reply.
you are right, and my scanner does have a descreening feature.
however, you may happen to work with images that were scanned by others, who didn't know about that feature, or didn't care to activate it, or whatever.
i experimented a bit with the tools that AP already offers, and achieved quite satisfactory results applying the following filters:
1) fft filter to the original image
2) frequency separation filter to the fft-filtered image
3) gaussian blur to the high frequency resulting layer
4) unsharp mask to the low frequency resulting layer
i'm no expert in this field, so there may probably be a better approach, and maybe more than one.
in addition, all of the above steps require some experimenting to find the parameters that best configure the filters.
for example, i found that the optimal configuration of the separation filter causes all (or at least most) of the noise that the fft filter could not remove to go to the high frequency layer. this is also the maximum detail desired on it, so you can start with a 0 radius and look at the preview to find the value where the noise stops to become more and more visible and stays constant. usually the optimal radius value appears to be between 0.8 and 1.5.
i guess that a new descreening filter could be implemented as a sequence of filters, with 4 controls:
- precision of the fft filter
- radius of the frequency separation
- radius of the gaussian blur
- radius of the unsharp mask
i feel (but i could be wrong) that there may be some correlation among the parameter values, especially among the three radius values: this could simplify the user interface of the filter, or at least allow to set some meaningful default values.
optionally, the filter could offer to display the fft window and manually paint over the frequency nodes.
i know that this is far more simple to put in words than into lines of code... but i'd really be happy to see something like this in one of the next releases.