Dear Johannes,
Am 15.10.18 um 10:08 schrieb johannes hanika:
this is an IIR filter (infinite impulse response, as opposed to the
straight forward FIR, finite impulse response, which would be for
instance a 5-tap b-spline filter) which approximates a gaussian. it's
especially well suited for large filter supports. why do you need to
know? for sports or you need to change / adapt it for something?
I plan to implement a guided filter
(https://arxiv.org/pdf/1505.00996.pdf
https://www.researchgate.net/publication/236228168_Guided_Image_Filtering
) as a reusable routine. This algorithm is already employed in the haze
removal module but it might find applications in other image processing
steps too. Currently I have a prototype for blend mask matting /
feathering, which I have mentioned on this list quite some time ago.
I had a look at darktable's Gaussian filter because I thought it might
be a good idea if the API of the guided filter is similar to the API of
the Gaussian filter. When I realized that it does work quite
differently from what I had expected I became curious.
After having a look at some papers (but lacking the time to read them
carefully) I have a rough idea how it works. I found several algorithms
that look similar to the one in gaussian.c but could not find a
reference that which refers to the particular method of gaussian.c.
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
