Pahlsson Descreen artifact and possible patch

[flugi]

When using fx_pahlsson_descreen, I noticed horizontal/vertical artifacts on the output. I investigated the issue, created a custom copy of the script and edited the pahlsson_descreen_channel function on the mask using this modification:

 # Make sure the center is not masked.
# display[mask]
 #ellipse[mask] 50%,50%,7%,7%,255,255,255,255    <-- original mask
 ellipse[mask] 50%,50%,40%,1%,0,255,255,255
 ellipse[mask] 50%,50%,1%,40%,0,255,255,255
# display[mask]

so instead of circular 7% mask, I used a cross-like mask. The results are much better for me. The scan is 400 ppi.

Without extensive testing I cannot tell if this modification creates different kinds of artifacts, should both be used, or should it be an option passed to the command, just wanted you to know that I found some cases with a solution you may want to use in the future.

[dtschump]

In the referenced commit, I've tried to add an option to select the Masking type.
Is that OK for you ?

(I didn't see much difference in this example, but it's probably the wrong example).

[flugi]

I am glad you think my modification is integration worthy, and thank you for your work.

I have no demand and I am not in a rush. I expected more research before acting, my "40%" is literally the first tested value without any optimization. I have not yet seen artifact with the patch, but maybe I will in a few days in usage.

The "Image" link in your message refers this page in mouseover tooltip, but sometimes while I type this message the browser dynamically resizes the page as an image will be loaded here and then jumps back. I don't know if my browser have an issue or the image link is wrong..

I learned the gmic script syntax just today, so I can not give you code review on your commit. As user, I prefer numeric parameters, so maybe both mask could be used with three parameters: the circle radius, the cross length and cross width, so the original would be 7,0,0 and mine would be 0,40,1 and future experience on possible new artifacts will not need new commits but fit into these parameters, what do you think?

[dtschump]

OK, so following your suggestions, here's my proposal :

GreycLab/gmic-community@40e5565

And here is how the modified filter looks like when run fron the G'MIC-Qt plug-in :

You can already test it by clicking on the "Refresh Filters" button from the G'MIC-Qt interface.
Could you please do that and tell me is that looks good for you ?

[flugi]

works in the sense that produces output which are not containing the old artifacts, however, when I run 0,40,0 and 1,40,0, none of those matches exactly my previous modification but both should, the 1% radius circle should be covered by the 40%,1% ellipse, so should not differ from the no-circle case, so the output should be exactly the same, but even the two new outputs are different. Maybe I don't understand how ellipses are drawn and overlaps are additive or something. Today I will have no time to test but eventually I will look into this.

Also, the preview looks much worse that the output, especially zoomed in 300%. The filter is very slow, so I understand that the preview is not exact, and I have no idea how to solve this.

For users, I think it is nice as it is. As a developer, I tend to stick to the case until I understand what happens :)

On another note, I smuggled around the blur parts of the filter and found some yet non conclusive clues how the fact that the mask and blur patches contain strong edges around the DOG output blobs within the blur blobs and some effort to get rid of those produces outputs which feels better but I have not enough data to show if it is indeed better or just my few samples worked better. Anyways, it is possible that I bother you with further ideas in the future regarding the details.

[dtschump]

Thanks for your comments.
I'm not the original developer of this filter so my knowledge about it is rather limited.
I've tried to modify it a little bit so that your suggestion can be added, but I'm not sure if I did it the best way possible.
At least, if this is already a slight improvement, I'm fine with that.
If you have more time to investigate/improve the script, just let me know.

Thanks again !