[Insight-users] initial values to Expectation Maximization of Gaussian Mixture Model

[Audette, Michel]

Hi all, 

I would like to find out the following. Is it feasible to initialize an Expectation Maximization of Gaussian Mixture Model on the basis of statistics computed from a few training points of each class, perhaps assigning a certain weighting to the original parameters? 

Thank you for your kind consideration.

Best regards, 

Michel Audette, Ph.D.
Innovation Center Computer Assisted Surgery (ICCAS)
Philipp-Rosenthal-Strasse 55
Leipzig, Germany
Phone: ++49 (0) 341 / 97 - 1 20 13
Fax: ++49 (0) 341 / 97 - 1 20 09