on Saturday, July 15, 2006, 20:36:00 you wrote:
BP> This is really a question for Thomas Rauscher, but the problem is
BP> very generic so I'll ask the panotools-devel list too.
BP> I want to apply a sharpening filter to panoramas, since my Peleng
BP> isn't so great in this department.
But still better than the Sigma, but this is a bit OT ;)
BP> I've looked at my workflow and it seems that the best place to
BP> apply an unsharp mask should be after creating the final
BP> equirectangular panorama.
The problem is how to define the area for the blur filter because its
not a ellipse/circle or a rectangle so the brute force method I used
for the antialiasing filters is to go back into the original,
calculate the distance there and apply those distance values to the
filter. This makes them independent from the actual remapping.
I guess the artefacts here appear because the filters are applied as a
rectangle not a circle. For the reconstruction this is good enough
(size~1 pixel, max error sqrt(2)) but for larger areas this is not
For those problems I guess its necessary to map the pixels onto a
sphere and calculate the distance there and remap it back to the
image. This can be speeded up a bit because you know the exact
(and simple) remapping formulas (no lens correct,...).
I think it would be nice to have a whole toolbox of operators on the
final equirectangular which take the distortion into account (for
example tonemapping!) because every image processing step which takes
the surrounding into account fails at the poles. You can discuss if it
makes sense to run a Canny filter on a pano but the problem is the
same for every of them.
I guess one question should be answered first: Should this toolbox
work only on equirectangulars (its only one easy transformation,...)
or should it be implemented for all possible transformations (slower,
but possible more useful)?
Just my thoughts.