[Insight-users] Problem with ImageGaussianModelEstimator

Jimmy Wong good_piggy at msn.com
Thu Dec 9 07:02:54 EST 2004


Dear all,

I have a problem when I am trying to use the ImageGaussianModelEstimator to 
estimate the segmented data models/parameters.

Let me put it in this way: first, I use k-means classifier to cluster the 
image (4 classes), then I have the label image, with component in {0, 1, 2, 
3}. After that, I want to estimate the parameters for these four gaussian 
distributions. I print the results out and find that, for class 0, there is 
no parameters, zero mean and zero variance.

And I read the doxygen documentation, found this "Any non zero, non negative 
value is considered a valid label". I just wonder, class 0 is not a valid 
label for this estimator. In this case, there are only 3 clusters in this 
model.

OK, now I use the shift filter to add -1 to the label image, and {0, 1, 2, 
3} becomes {-1, 0, 1, 2}. Run the estimator, I got the result as expected. I 
can only get two sets of parameters from that two 'valid' cluster {1, 2}.

But, when I change -1 to +1, I encounter the problem. No exception prompted, 
only the windows style dialog, with the words something like -- we have 
encounter the problem with the xxx.exe application, bla bla bla.

Anyone has this problem? How to solve it?

Thank you

Zhimin




More information about the Insight-users mailing list