Hi,
Am 04.04.19 um 15:05 schrieb P. Altenburg:
As it is, I've noticed the radius of the blur on (parametric) masks is
200 at max.
I would love it if this radius could be really big, like a quarter of
the image width or something.
I agree that a maximal blur radius of 200 can be too small in some
situations. In particular, when one considers the ever growing pixel
counts of modern cameras. However, border effects may become an issue
when the kernel size is of the order of the image width/height. One has
to chose boundary conditions carefully. Furthermore, due to boundary
effects the current pixel pipe does not allow (afaik) to calculate a
realistic preview when zoomed in, i.e. only a section of the whole image
is shown. This would require to process to whole image at the chosen
zoom level and then crop the view to the size of the preview window.
If you think: "But with the blur radius at 1000, you'd need to sample
almost a million pixels for every pixel that you want to blur!", that's
not necessary; the program could easily subsample like 10% or less of
those 1000x1000 pixels and be accurate enough.
Actually, this is not an issue. darktable utilizes an implementation of
the Gaussian filter that features a computational complexity that is
independent of the kernel size. I am unsure how accurate these
so-called recursive filters (which are approximate procedures) are when
the kernel size is of the order of the image width/height. I expect
that these methods are not applicable in this case.
Heiko
--
-- Number Crunch Blog @ https://www.numbercrunch.de
-- Cluster Computing @ https://www.clustercomputing.de
-- Social Networking @ https://www.researchgate.net/profile/Heiko_Bauke
___________________________________________________________________________
darktable developer mailing list
to unsubscribe send a mail to darktable-dev+unsubscr...@lists.darktable.org
