Page 1 of 2

sigmoidized EWA quadratic B-spline windowed Jinc 3-lobe

Posted: 2012-09-24T07:45:00-07:00
by NicolasRobidoux
Using my usual coterie of enlargements, between 7x and 32x, used in viewtopic.php?f=22&t=21804, here are the results with sigmoidized (contrast = 7.5) EWA (Elliptical Weighted Averaging) quadratic B-spline-windowed Jinc 3-lobe:

Code: Select all

magick input_small.png -colorspace RGB +sigmoidal-contrast 7.5 -define filter:window=Quadratic -define filter:lobes=3 -distort resize 800% -sigmoidal-contrast 7.5 -colorspace sRGB EWAQuadraticJincSig7p5.png
(Of course you can replace "magick" by "convert".)
Image
Image
Image
Image
Image
Image
Image
Image
As expected, it has no noticeable second halo (given that it is a 3-lobe method), and it is more jaggy and more blurry than, say, EWA LanczosSharp and EWA LanczosRadius3.
That's the price for an insignificant second halo.

Re: sigmoidized EWA quadratic B-spline windowed Jinc 3-lobe

Posted: 2012-09-24T08:03:33-07:00
by NicolasRobidoux
In viewtopic.php?f=1&t=20992&start=225#p90394, I show the results of EWA quadratic B-spline windowed Jinc 3-lobe, with my two favorite downsampling tests, the fly image, and the backpack image. Of course, sigmoidization is omitted.

Re: sigmoidized EWA quadratic B-spline windowed Jinc 3-lobe

Posted: 2012-09-24T11:30:48-07:00
by BryantMoore
@NicolasRobidoux
Could you perhaps test your filter on the original image I have uploaded in this post?

viewtopic.php?p=90497#p90497

:)

Re: sigmoidized EWA quadratic B-spline windowed Jinc 3-lobe

Posted: 2012-09-26T12:42:34-07:00
by NicolasRobidoux
I found an amusing name for this filter: Ka-ching ("QUAdratic-windowed JINC" -> "Ka-ching").
Good name, because this 3-love filter is "right in the money" as far as 2-lobe filters go.
(Amusing to me at least.)

Re: sigmoidized EWA quadratic B-spline windowed Jinc 3-lobe

Posted: 2012-09-26T13:28:13-07:00
by BryantMoore
I like that name! :D

Re: sigmoidized EWA quadratic B-spline windowed Jinc 3-lobe

Posted: 2012-10-01T09:22:25-07:00
by NicolasRobidoux
I bet you anything that Quadratic windowing would work well with tensor Sinc 3-lobe filtering. Will have to try.
(Spline too?)
P.S. It works well, but the only thing it has on EWA Quadratic-windowed Jinc 3 is that it is interpolatory and a tiny bit sharper (which may be good for very minor resizing operations).

Code: Select all

convert {input} -define filter:filter=Sinc -define filter:lobes=3 -define filter:window=Quadratic -resize {size} {output}
The spline version is almost identical. Unremarkable except for that it has basically no halo.

Re: sigmoidized EWA quadratic B-spline windowed Jinc 3-lobe

Posted: 2012-10-01T11:35:02-07:00
by NicolasRobidoux
Maybe I was a bit harsh on tensor Quadratic-windowed Sinc 3.
Compared to my favorite tensor windowed-Sinc, namely Ginseng, it has much less halo, but it does not suppress moire as well.
Basically, it is a trade-off between less halo (tensor Quadratic-windowed Sinc 3 wins) and moire suppression and a bit of sharpness (tensor Ginseng wins).
In the balance of things, probably not worth pushing into the "pantheon".

Re: sigmoidized EWA quadratic B-spline windowed Jinc 3-lobe

Posted: 2012-10-12T17:07:58-07:00
by NicolasRobidoux
Hann windowing also works well with EWA Jinc with the deblur that gives Radius 3.
P.S. And Spline windowing works well with EWA Jinc with the deblur that gives Radius 4.

Re: sigmoidized EWA quadratic B-spline windowed Jinc 3-lobe

Posted: 2012-10-16T19:32:57-07:00
by anthony
NicolasRobidoux wrote:Hann windowing also works well with EWA Jinc with the deblur that gives Radius 3.
P.S. And Spline windowing works well with EWA Jinc with the deblur that gives Radius 4.
I thought mixing the new LanczosRadius with a windowing function would perhaps let you do this.
But the order in which 'special filters' overrides defaults, prevents it from working. Sorry, this will be a DIY.

Re: sigmoidized EWA quadratic B-spline windowed Jinc 3-lobe

Posted: 2012-10-17T04:46:52-07:00
by NicolasRobidoux
OK with DYI, esp. if it's pain in the neck for you to "name".
----
And there was a mistake in the above:
Spline windowing with EWA Jinc Radius 3 works well (not Radius 4).

Re: sigmoidized EWA quadratic B-spline windowed Jinc 3-lobe

Posted: 2012-10-17T05:25:24-07:00
by NicolasRobidoux
This being said, EWA Spline-windowed Jinc Radius 3 is an expensive way of getting a scheme which is very close to EWA Robidoux. (Noticeably better when downsampling? It's a little sharper and hopefully has "moire fighting" whiskers.)

Re: sigmoidized EWA quadratic B-spline windowed Jinc 3-lobe

Posted: 2012-10-18T16:39:54-07:00
by henrywho
NicolasRobidoux wrote:This being said, EWA Spline-windowed Jinc Radius 3 is an expensive way of getting a scheme which is very close to EWA Robidoux. (Noticeably better when downsampling? It's a little sharper and hopefully has "moire fighting" whiskers.)
Yes, definitely better "moire fighting".... if you are talking about "-define filter:window=Spline -define filter:lobes=3 -distort resize"

Re: sigmoidized EWA quadratic B-spline windowed Jinc 3-lobe

Posted: 2012-10-18T17:23:47-07:00
by anthony
henrywho wrote:
NicolasRobidoux wrote:This being said, EWA Spline-windowed Jinc Radius 3 is an expensive way of getting a scheme which is very close to EWA Robidoux. (Noticeably better when downsampling? It's a little sharper and hopefully has "moire fighting" whiskers.)
Yes, definitely better "moire fighting".... if you are talking about "-define filter:window=Spline -define filter:lobes=3 -distort resize"

You do know that that is the exact definition of a 3 lobe Parzen Windowed Filter (though its normal default is 4 lobes)

Code: Select all

    -filter Parzen -define filter:lobes=3 -distort resize ...
Verification...

Code: Select all

convert xc: -define filter:verbose=1 -define filter:window=Spline -define filter:lobes=3 -distort resize 200% null: | grep '^#'
# Resampling Filter (for graphing)
#
# filter = Jinc
# window = Cubic
# support = 3.23832
# window-support = 3.23832
# scale-blur = 1
# practical-support = 3.23832
# B,C = 1,0

Code: Select all

convert xc: -define filter:verbose=1 -filter Parzen -define filter:lobes=3 -distort resize 200% null: | grep '^#'
# Resampling Filter (for graphing)
#
# filter = Jinc
# window = Cubic
# support = 3.23832
# window-support = 3.23832
# scale-blur = 1
# practical-support = 3.23832
# B,C = 1,0
Note the windowing function is Cubic and B,C values are the same.

And While I have not graphed it, it is likely to be similar to other sharpened filters, but without the blur shift of the zero crossing.

Nicholas, this may be a candidate for what I talked about in Qualitive classification of Cylindrical Resampling Filters, as it has different zero crossing point than the other sharpened filters. It may mean that the first zero crossing is not a good candidate for classifying filters.

Re: sigmoidized EWA quadratic B-spline windowed Jinc 3-lobe

Posted: 2012-10-18T18:39:12-07:00
by NicolasRobidoux
Yes, I knew it was Parzen-windowed Jinc 3-lobe with no deblur. (Probably should have called it that, too.)
Going back and forth between communities, and my terminology "slips" as a result.

Re: sigmoidized EWA quadratic B-spline windowed Jinc 3-lobe

Posted: 2012-10-18T18:44:18-07:00
by NicolasRobidoux
henrywho wrote: Yes, definitely better "moire fighting".... if you are talking about "-define filter:window=Spline -define filter:lobes=3 -distort resize"
Yes, I was.
It looks like "thin whiskers" do give something, at least in EWA.