[Insight-users] image registration metrics for cloud of points

[Karthik Krishnan]

On 1/24/07, José Santamaría López <jsantam at ugr.es> wrote:

  Could someone tell me how many metrics there exist
> in ITK working with cloud of points images or these
> ones with additional information such as normals or
> curvature information ?


"could of points image with normals / curvature information" is a bit vague.

Please clarify if you have a
  (1) point set or
  (2) mesh (point set with connectivity info)
  (3) a bunch of pixels in an image, that you're referring to as a point
set.

----

If what you have is a point set (1), then ITK offers you the metrics :
 - ICP
 - Hausdorff distance [The filter in ITK takes images as input, but if you
have a point set, you should find it really simple to do that with a forloop
].

---

If you have a mesh (2), nothing that I know of that takes advantage of
connectivity as well.

---

If you have a bunch of pixels in an image (3) ..

Where did those pixels come from ?
(a) If they are the points that represent the contour of a segmented object,
the hausdorff distance filter will work for you.
(b) You can compute the signed distance image of one object from the other
(the hausdorff distance is actually a max function of this distance) and
compute the mean signed distance.
(c) Narrow band to narrow band registration - Please Read
http://www.itk.org/cgi-bin/viewcvs.cgi/*checkout*/Papers/MedIASpecialIssue/NarrowBandToImageRegistration/NarrowBandToImageRegistration.pdf?rev=1.1&root=InsightDocuments

HTH
-karthik
-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://public.kitware.com/pipermail/insight-users/attachments/20070124/6ef7e9f0/attachment.htm