After an evaluation, GNOME has moved from Bugzilla to GitLab. Learn more about GitLab.
No new issues can be reported in GNOME Bugzilla anymore.
To report an issue in a GNOME project, go to GNOME GitLab.
Do not go to GNOME Gitlab for: Bluefish, Doxygen, GnuCash, GStreamer, java-gnome, LDTP, NetworkManager, Tomboy.
Bug 779212 - Halos with the Gaussian blur plugin
Halos with the Gaussian blur plugin
Status: RESOLVED OBSOLETE
Product: GIMP
Classification: Other
Component: General
git master
Other All
: Normal normal
: ---
Assigned To: GIMP Bugs
GIMP Bugs
: 777612 (view as bug list)
Depends on:
Blocks:
 
 
Reported: 2017-02-25 10:21 UTC by Alberto Ferrante
Modified: 2018-05-24 17:28 UTC
See Also:
GNOME target: ---
GNOME version: ---


Attachments
Test file (1.43 MB, image/x-xcf)
2017-02-28 12:46 UTC, Alberto Ferrante
Details

Description Alberto Ferrante 2017-02-25 10:21:56 UTC
In some cases the Gaussian blur plugin produces some halos that should not be there. Halos are not shown in preview, but they are when the filter is applied (see: https://bugzilla.gnome.org/attachment.cgi?id=343983 ).
The problem appears only on masks (both in 8 and 16-bit images).

This problem has been there for awhile, I reported it also in a GEGL bug [https://bugzilla.gnome.org/show_bug.cgi?id=777612], but it is probably a GIMP bug as it occurs only on masks and not on layers.
Comment 1 Michael Natterer 2017-02-25 10:33:54 UTC
By "masks", do you mean channels (saved selections) or layer masks?
Comment 2 Alberto Ferrante 2017-02-25 10:42:40 UTC
Layer masks
Comment 3 Michael Natterer 2017-02-25 20:30:36 UTC
Thanks, can you attach a small image where this happens, with instructions
how to reproduce? I can't make that halo happen here.
Comment 4 Ell 2017-02-26 13:27:17 UTC
This issue is not directly related to Gaussian blur, or to layer masks per se, but to the conversion of low-precision linear light values to gamma corrected values.  Since the linear->gamma curve is very sparse at the low end, any loss of precision would create banding at the low-value range when converting to perceptual.

I'm guessing that Alberto is using an 8-bit perceptual image, which is why the issue doesn't happen with layer content, and is seeing the banding when showing the mask, rather than applying it to the layer, since masks are always linear.  Note that the linear->gamma conversion doesn't happen when using the mask for compositing, so this issue isn't *that* critical.

OT, but I'm much more concerned of what's going to happen when we switch normal mode to linear compositing by default, which will make low-precision alpha channels very problematic for similar reasons (to see this, in an 8-bit image, create an empty layer over a black background, and fill some of its area with white; apply a large blur, and switch the layer composite space to linear.)
Comment 5 Elle Stone 2017-02-26 14:27:22 UTC
Are you saying that alpha channels, including masks on layers, are only stored using 8-bit precision, even if the image itself is at 32f? Where is the code is this set?
Comment 6 Ell 2017-02-26 14:51:22 UTC
I never said that :)
Comment 7 Øyvind Kolås (pippin) 2017-02-27 19:03:25 UTC
*** Bug 777612 has been marked as a duplicate of this bug. ***
Comment 8 Alberto Ferrante 2017-02-28 12:46:03 UTC
Created attachment 346901 [details]
Test file

Here is an xcf file in which I have experienced the problem (which does not appear in all files, even though I could not find any specific reason as it appears with 8 and 16 bit files, both grayscale and color).

To reproduce the problem: add a new layer, add a black layer mask, paint in white, apply gaussian blur. The problem is not visible in the preview, but it is visible once the filter is applied.
Comment 9 GNOME Infrastructure Team 2018-05-24 17:28:34 UTC
-- GitLab Migration Automatic Message --

This bug has been migrated to GNOME's GitLab instance and has been closed from further activity.

You can subscribe and participate further through the new bug through this link to our GitLab instance: https://gitlab.gnome.org/GNOME/gimp/issues/1059.