[Digikam-devel] [digikam] [Bug 342513] Add support for fractional radius in blur and unsharp mask filters [patch]

[pochini at shiny.it]

https://bugs.kde.org/show_bug.cgi?id=342513

--- Comment #3 from pochini at shiny.it ---
(In reply to Maik Qualmann from comment #2)
> Created attachment 90332 [details]
> blurtest.png
> 
> Hi pochini,
> 
> I also think this should be a double value of the radius. You seem to
> understand a lot of the blur kernel

Not really...


> maybe you could enhance it further? I
> once created a test image. Left image is from digiKam and right image is
> from Gimp or Krita. Look on the mouth.

Yes, most likely it is due the square kernel. Gimp and other programs feature a
gaussian blur filter instead that produces a much more pleasing "bokeh" (and
less artifacts that are not visible in your example). I can write a generic
convolution filter that can approximate the gaussian filter. It will be a lot
slower, though.
I think the blur filter in Digikam is useless. I never had the need to blur a
photo... quite the opposite. I enhanced it because it is used internally by the
unsharp mask filter and the integer-only radius does not allow the user (me :))
) to set the radius small enough. IMO radius>=1 is too large in most cases
because it creates halo artifacts around sharp edges.
If you think a good blur filter is useful, then I'll try to write something
better. Otherwise I think it is already good enough for the unsharp mask
filter, which is my primary goal.

-- 
You are receiving this mail because:
You are the assignee for the bug.