[ptx] autopano: first attempt at semi-automatic panorama stitching

[Sebastian Nowozin]

Hi Pablo and others :)


On Sun, Dec 28, 2003 at 03:25:55PM +0100, Pablo d'Angelo wrote:

> [...]
> If I remember correctly, this is also one of the interesting parts of the
> SIFT features.
 
  I admit of not having read up / understood the SIFT paper yet. I use an old
fashioned approach where the problem of creating fingerprint signatures is
divided clearly into two seperate parts:

  1. Feature indexing
  2. Feature fingerprint generation

I think the most important part is 1 and even my simple circular-map will do
fine for most panorama images at step 2.


> Your first correlation step you could use your rotation invariant, radial
> sum descriptor, but additionally you have additional information that can be
> exploited, like it is done with the RANSAC in Browns approach.
 
Have to read up on that ;) (but programming and trying out things is more fun
;)


Just for the part (1) I now completely dropped my idea of feature pixel
identification. I read up a bit on standard ways to do image feature indexing
and found an interesting PhD thesis:
http://www.cat.csiro.au/cmst/staff/jmr/thesis/

Whats interesting about it is that is has an easy to understand description of
a feature indexing algorithm called "Smith"'s edge and corner detection
algorithm. I got it implemented today and it works vastly better than my old
one. In the PhD thesis its compared to Harris's algorithm (which is what most
more modern algorithms base their idea on) and while it is quite good in
general, Harris's is more stable against slight image movements (which is what
I need for the panorama thing :)

I don't think one would need scale invariant and shear/stretch resistant
detection algorithms that much for panorama photography. If at all, scale
invariant, because of images "below your feet" which appear larger than images
containing the same features but at another angle.

If I understand Harris's algorithm I will implement it, but I will try to put
on a faster and more accurate version of my small program tomorrow using the
Smith algorithm.

I also found an interesting paper today that might be an alternative to the
[iccv2003] panorama paper you posted me:
http://citeseer.nj.nec.com/mikolajczyk02affine.html


> Alexandre hasn't posted it yet. we have to think if we want to introduce
> patented algorithms into hugin. hmm maybe its ok for us germans, at least for
> now ;)
 
If there is a good description of whats patented exactly one can surely work
around that.


ciao,
Sebastian Nowozin

-- 
nowozin at cs.tu-berlin.de --- http://user.cs.tu-berlin.de/~nowozin/