Are there techniques to register spherical images? Panning and tilting
of the original images would actually correspond to simple shifts on
the sphere, but what about roll?

On Apr 30, 5:21 am, damo suzuki <liquidt...@gmail.com> wrote:

Are there techniques to register spherical images? Panning and tilting
of the original images would actually correspond to simple shifts on
the sphere, but what about roll?

--------------------------------------------------------------------------- -----------------
damo:
Is it really a sphere, like you're trying to align two maps on a
globe?
I'm not sure how to do it but I remember seeing these and perhaps a
Mobius transformation or the Reimann sphere concept might be a lead
that might help you.

http://en.wikipedia.org/wiki/M%C3%B6bius_transformation

Here's a really cool video that shows how you can take a spherical
image and map it to a plane:

http://www.youtube.com/watch?v=JX3VmDgiFnY
The full version and more explanation is available at the professor's
web site:http://www.ima.umn.edu/~arnold/moebius/

Then, once your spherical images have been mapped to the plane, maybe
you can align the images in the plane using more traditional
registration techniques, and then map back onto your sphere.
Regards,
ImageAnalyst

On Apr 30, 3:31 pm, ImageAnalyst <imageanal...@mailinator.com> wrote:





On Apr 30, 5:21 am, damo suzuki <liquidt...@gmail.com> wrote:

Are there techniques to register spherical images? Panning and tilting
of the original images would actually correspond to simple shifts on
the sphere, but what about roll?

--------------------------------------------------------------------------- -----------------
damo:
Is it really a sphere, like you're trying to align two maps on a
globe?
I'm not sure how to do it but I remember seeing these and perhaps a
Mobius transformation or the Reimann sphere concept might be a lead
that might help you.

http://en.wikipedia.org/wiki/M%C3%B6bius_transformation

Here's a really cool video that shows how you can take a spherical
image and map it to a plane:

http://www.youtube.com/watch?v=JX3VmDgiFnY
The full version and more explanation is available at the professor's
web site:http://www.ima.umn.edu/~arnold/moebius/

Then, once your spherical images have been mapped to the plane, maybe
you can align the images in the plane using more traditional
registration techniques, and then map back onto your sphere.
Regards,
ImageAnalyst

Thanks, I'll take a look. Basically what I'D LIKE TO DO, but maybe is
not possible, is capture images from different views, remap them using
spherical projection and then mosaic them. Since a computationally
simple technique is what I'm looking for I was thinking about
spherical projection because so I could easily take care of pan and
tilt but don't know about roll. I don't think that on spheres a roll
of the input image is still a simple rotation about the image center.
Actually my panoramic technique is based on cylindrical warping but
then I can "easily" compensate only for panning (shift on a cylinder).
But I need something to compensate for roll.- Hide quoted text -

- Show quoted text -
-------------------------------------------------------------------------------------------------------

Are there techniques to register spherical images? Panning and tilting
of the original images would actually correspond to simple shifts on
the sphere, but what about roll?

You should look up Szeliski at Microsoft research.

He has an excellent tech report on how to do exactly what you are talk
about.  Its a 3-parameter spherical registration technique for doing
rotational panoramic mosaics.  He also has an excellent tutorial on
image registration which includes the spherical registration method.

http://research.microsoft.com/~szeliski/publications.htm

damo suzuki wrote:
Are there techniques to register spherical images? Panning and tilting
of the original images would actually correspond to simple shifts on
the sphere, but what about roll?