Anything that would help to spread control points farther beyond
clumps in the center of the image for fisheye images... would greatly
reduce stitching errors, I think.
ei
On Nov 7, 2010, at 2:58 AM, kfj wrote:
On 31 Okt., 03:14, Tom Sharpless <tksharpl...@gmail.com> wrote:
It is a fact that SIFT does not find as many matches as we would like
in wide angle and especially fisheye images. But that problem will
not be solved by better matching of affine transformations, as it is
due to image deformations, characteristic of those lenses, that are
not affine transformations.
I've thought about the matter and come up with this train of thought:
If you look at two corresponding small sections of, e.g., two fisheye
images taken with different orientation, the transformation from one
to the other is not, but resembles an affine transformation. The
smaller the bits you compare, the closer the semblance. So looking at
affine transformations helps, since a particular affine transformation
locally approximates the actual nonaffine transformation.
If the lens geometry is known, though, the choice of which affine
transformation of which subsection of the image to feed into a feature
generating algorithm could be narrowed down to that affine
transformation that best locally approximates the non-affine
transformation, thus gaining some efficiency.
I admit that this is more of a gut feeling than something I could
easily demonstrate mathematically, but maybe someone with a firmer
mathematical background could give a comment?
with regards
Kay
--
You received this message because you are subscribed to the Google Groups "Hugin and
other free panoramic software" group.
A list of frequently asked questions is available at:
http://wiki.panotools.org/Hugin_FAQ
To post to this group, send email to hugin-ptx@googlegroups.com
To unsubscribe from this group, send email to
hugin-ptx+unsubscr...@googlegroups.com
For more options, visit this group at http://groups.google.com/group/hugin-ptx
