Add known observations as deformation prior

15 views
Skip to first unread message

Maia R.

unread,
Jan 20, 2021, 5:53:50 PMJan 20
to scalismo
Hi,
I applied ICP-based model fitting to my project as explained in Tutorial 11 (Model fitting with Iterative Closest Points).
In this tutorial, uniformly distributed points on the surface are used to establish correspondence. I applied this method to my project, but for some cases, I don't get good fitting though increasing the number of iterations.
In this article, 
it explains that known observations can be used a deformation prior to improve the registration (as far as I understand).
How can I include my predefined anatomical landmarks as deformation prior in the ICP-approach ? As I'd like to get as much as registration accuracy as possible, does the combination of those dense points  with my very sparse known landmarks improve the fitting ?
Also, in Tutorial 12, the approach (gradient-based registration) do not use dense points anymore. Would it better to use both landmarks and dense point as in Tutorial 11 to improve the accuracy ?
Thank you very much.
Best regards,
Maia

Maia R.

unread,
Jan 21, 2021, 10:58:04 AMJan 21
to scalismo

Hi,
After reading more carefully the paper, I think I understand how to incorporate my landmarks in combination with the set of dense points in the registration.
Maia 

Marcel Luethi

unread,
Jan 21, 2021, 2:14:08 PMJan 21
to Maia R., scalismo
Dear Maia

Sorry for the late reply.
I am happy to hear that you found a solution. In case it still helps, the idea is to build a posterior model using the landmarks, and then to use this posterior model (which is just another statistical shape model) in the registration process. This works for both the ICP as well as the parametric registration approach. It would, however, break down if you optimized for pose and shape at the same time.

Best regards,

Marcel

--
You received this message because you are subscribed to the Google Groups "scalismo" group.
To unsubscribe from this group and stop receiving emails from it, send an email to scalismo+u...@googlegroups.com.
To view this discussion on the web visit https://groups.google.com/d/msgid/scalismo/7a634fc9-368d-4355-ba15-6ee5b3f22d1cn%40googlegroups.com.

V R

unread,
Jan 21, 2021, 3:11:56 PMJan 21
to Marcel Luethi, scalismo
Dear Marcel,
Thank you a lot for the additional explanation. 
Best regards
Maia

Reply all
Reply to author
Forward
0 new messages