Question about Panorama Tools interpolators...
Ken Warner
If I want to warp using Panorama Tools so I can use PT's
interpolators, which interpolator will give the sharpest
and at the same time smoothest result?
I used sinc256 that seemed pretty good -- but I don't want to
test each one -- lazy I guess.
Jim Watters
> If I want to warp using Panorama Tools so I can use PT's
> interpolators, which interpolator will give the sharpest
> and at the same time smoothest result?
>
> I used sinc256 that seemed pretty good -- but I don't want to
> test each one -- lazy I guess.
>
Anything over Spline 64 is likely overkill.
--
Jim Watters
http://photocreations.ca
Hans
bicubic.
You can easy triple stitching time by using many of the others.
Hans
Hans
Hans
Erik Krause
> If I want to warp using Panorama Tools so I can use PT's
> interpolators, which interpolator will give the sharpest
> and at the same time smoothest result?
If your result size is smaller than the recommended maximum size use one
of the anti-aliasing interpolators. With all others you get jaggies:
http://wiki.panotools.org/PanoTools_Anti_Aliasing_Filters
--
Erik Krause
http://www.erik-krause.de
Ken Warner
interpolator, I get good sharpening of the image but I also
get more noise artifacts in the sky.
From the discussion here, I assume that is relative to the
amount of negative values in the interpolator window. Some
negative values at the extrema is good for sharpening but
too much tends to produce noise artifacts in the sky.
So from the graph on the PanoTools_Anti_Aliasing_Filters page,
it looks like Blackman/Bessel Filter would be a good choice.
That's what I'm thinking now. But I most likely am wrong.
But the PT interpolators are really pretty slow.
Comments?
John Houghton
> The reason I asked this question is because when I use a sinc based
> interpolator, I get good sharpening of the image but I also
> get more noise artifacts in the sky.
stitching and then sharpen using edge sharpening techniques, which
avoid sharpening bland areas like the sky so that noise artifacts are
not accentuated.
John
Ken Warner
tried before. Good idea. I've read about that way of sharpening
but GIMP makes that sort of a multi step operation. I'll
look at that again.
Keith Martin
How does this sort of thing work with regard to the distorted areas
in equirects?
k
Ken Warner
edge (aka smart) sharpening in GIMP.
John Houghton
rather than the equirectangular image because of the resultant odd
effects in the stretched areas at the nadir and zenith. Try it both
ways to see for yourself. I confess I usually sharpen the
equirectangular image though.
John
John Houghton
> Idunno... I'm still trying to figure out how to do
> edge (aka smart) sharpening in GIMP.
http://www.gimp.org/tutorials/Smart_Sharpening/
John
Ken Warner
the stuff I do.
Erik Krause
> How does this sort of thing work with regard to the distorted areas
> in equirects?
Since a lack of sharpness usually distributes evenly across an image, it
is best to sharpen the unwarped images, then use an interpolator that
preserves sharpness but doesn't sharpen itself.
Joergen Geerds
> Since a lack of sharpness usually distributes evenly across an image, it
> is best to sharpen the unwarped images, then use an interpolator that
> preserves sharpness but doesn't sharpen itself.
I don't agree that "lack of sharpness" is equally distributed across
the source image, in most cases the center is sharper than the corners
of the source.
IMO it's best to leave off any sharpening of the source images, and
apply the sharpening process as the last step of the production, i.e.
the cube faces for VRs or the flat view for i.e. print. but I
completely agree with you to use an interpolator that doesn't mess
with the data too much.
joergen
Erik Krause
> I don't agree that "lack of sharpness" is equally distributed across
> the source image, in most cases the center is sharper than the corners
> of the source.
Well, you're absolutely right. I should write "distributed in a lens
specific pattern". A lens dependent sharpening algorithm would take this
into account. However, after warping there's no chance to remove this
kind of blur individually.
On the other hand: PTGui blender weights the center more than the
corners, which is generally good for the sharpness - and would justify a
general not too strong pre-warping sharpening.
Eric O'Brien
me. That is, most transformations (warping) in the stitching process
would be "cancelled out" when view the final image dynamically (the
final image is also an equirectangular). The distribution of the
sharpening effect would essentially be the same as it was in the
source images.
But if I sharpen a source image that is from a fisheye lens, it seems
that the distribution of the sharpening results will *not* be
cancelled out. I think a similar situation would be: I want to
*print* an image shot with a fisheye lens, but I'll remap a certain
portion to equirectangular. it seems that the majority of the
sharpening should be applied to the remapped (equirectangular)
version, not the source (fisheye) version.
eo
> --
Eric O'Brien
"neutral?" Or most neutral?
eo
Roger D. Williams
I use DXO when it has modules for both the cameras and lenses I use. DXO
just came out with a module for the Tamron 11-17 rectilinear zoom which
I use most often for non-panorama shots. I've been revisiting photos
that I previously developed from RAW using manual settings, and I must
say the auto DXO settings usually produce results as good as, and
sometimes better than, my own time-consuming work.
One of the features that the modules give is a "remove lens softness"
option, which applies lens-specific sharpening. This is quite subtle
but produces a noticeable improvement in the corners of my images taken
with this lens. Having read this exchange between Erik and Joergen, the
next time I use my Nikon 10.5mm full-frame fisheye I will see what
effect it has on panoramas stitched from the images.
Combined with the auto correction for colour aberration, which also
works very well, I imagine it could help precise control-point setting
in the peripheral areas of the individual images.
I recently got the Pentax 11-17 zoom fisheye (same optics as the Tokina
of that ilk) and I see that DXO is to support the Pentax K-x I am using.
I can hardly wait for when they introduce support for this lens/camera
combination!
Roger W.
--
Business: www.adex-japan.com
Pleasure: www.usefilm.com/member/roger
Erik Krause
> But if I sharpen a source image that is from a fisheye lens, it seems
> cancelled out. I think a similar situation would be: I want to
> *print* an image shot with a fisheye lens, but I'll remap a certain
> portion to equirectangular. it seems that the majority of the
> sharpening should be applied to the remapped (equirectangular)
> version, not the source (fisheye) version.
"Distribution" is wrong. What I meant is the following: Lens blur is
normally not directional, a point is usually blurred to a circle, not to
an oval or a line. Hence usual sharpening algorithms have a radius and
work in a circle around a single pixel.
In equirectangular projection single source pixels are distorted to a
shorter or longer line. Blur circles are distorted to more or less flat
ovals. A circular sharpening algorithm will work more or less correctly
in vertical direction but the nearer you get to the poles the worse will
be the effect in horizontal direction.
If you sharpen cubic output this applies too, but to a smaller degree.
The original round blur circles are distorted to ovals too, but
certainly less extreme as in equirectangular projection. In addition
there might be problems sharpening edge pixels, since the algorithm
can't know what's behind the edge...
Ken Warner
sharpening artifacts are near the poles. Especially in the sky
outdoors. It's a difficult problem to solve without specially
designed software for sharpening equirectangular images that
compensates for the lateral stretching of pixels at the poles.
I just sharpen the equirect a little and live with the artifacts.