[Insight-users] ExpectationMaximizationMixtureModelEstimator with ImageToListSampleAdaptor

[David Doria]

Has anyone used ExpectationMaximizationMixtureModelEstimator to
classify image pixels?

I tried to do so here:
http://www.vtk.org/Wiki/ITK/Examples/Broken/Statistics/ExpectationMaximizationMixtureModelEstimator_Image

But the cluster proportions end up as NaN:

Cluster[0]
    Parameters:
         [0, 0, 0, 5, 5, 5, 5, 5, 5, 5, 5, 5]
    Proportion:          -nan
....

I create a demo image consisting of only black, red, and green pixels.
Therefore, I set the number of classes to 3, expecting each one of
these colors to get its own class. I set the estimator to 12 variables
(3 for means and 9 for 3x3 covariance).

Would anyone mind skimming this example to make sure I haven't done
something silly? Any other comments?

Thanks,

David