[Insight-users] MutualInformationEuler2DRegistration

[Ying Du]

Hi,

I am trying to register some 2D images using the InsightApplication "MutualInformationEuler2DRegistration". It seems like that the result is very very sensitive to the parameters (the learning rate, rotation scale, translation scale). 

I tried the application on a very simple case, but could not find a set of parameters to get a very good aligned result (there are always a small part of the boundaries that are not aligned). The input fixed image and moving image are shown at the following link:
http://www.sci.utah.edu/~ydu/img_registration.htm

The moving image is rotated for 5 degree and translated for a few pixels. This is a pretty simple case, and Mutual information should be able to handle it well. I am wondering why the MutualInformationEuler2DRegistration application is so difficult to get a good registration result for such a simple case, and why it is so sensitive to the parameters? Is there any way to solve this "super sensitive to parameters" problem?

Many thanks.

Ying
-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://public.kitware.com/pipermail/insight-users/attachments/20040630/6b56130b/attachment.htm