https://bugs.kde.org/show_bug.cgi?id=487997
Noah Davis <noaha...@gmail.com> changed:
What |Removed |Added
----------------------------------------------------------------------------
Resolution|WAITINGFORINFO |---
Status|NEEDSINFO |CONFIRMED
Ever confirmed|0 |1
--- Comment #13 from Noah Davis <noaha...@gmail.com> ---
(In reply to Eamonn Rea from comment #8)
> Created attachment 170190 [details]
> Region Screenshot - YouTube [Poor Quality]
(In reply to Eamonn Rea from comment #9)
> Created attachment 170191 [details]
> Current Monitor (Cropped) - YouTube [Good Quality]
Besides the obvious image size difference, this example is hard to see
differences with even after downloading. I think I do see some subtle ringing
artifacts around the text.
(In reply to Eamonn Rea from comment #10)
> Created attachment 170192 [details]
> Region Screenshot - Panel [Poor Quality]
(In reply to Eamonn Rea from comment #11)
> Created attachment 170193 [details]
> Current Monitor (Cropped) - Panel [Good Quality]
I can see that the second image is larger and has some ringing artifacts.
The ringing artifacts seem to be from the Lanczos scaling algorithm used when
fractional scaling is used. When screen images are scaled back down to their
original size, that should be pretty much dealt with. I could try a bicubic
(often just called cubic) or bilinear (often just called linear) scaling
algorithm instead.
Lanczos is the fanciest algorithm available in OpenCV, but can cause ringing
artifacts where abrupt color changes occur. Bicubic and Bilinear might be
blurrier than Lanczos. Bilinear might have problems when scaling beyond 2x, but
might be better than bicubic at scales below that.
If you're interested in these kinds of things, check out this gallery:
https://en.wikipedia.org/wiki/Comparison_gallery_of_image_scaling_algorithms
Unfortunately, it doesn't have an example of the Lanczos algorithm.
--
You are receiving this mail because:
You are watching all bug changes.
