[Insight-users] ActiveShapeModelCalculator flawed?

[Matthias Kirschner]

Hi,

I am currently implementing a class library for the active shape model 
in n dimensions. The current itk implementation works only in 2D. After 
examining the itk code in detail, it seems to me that there is a flaw in 
the current implementation of the PCA
(itkActiveShapeModelCalculator.txx, Revision 1.5):

After solving the eigenproblem, the eigenvalue vector (line 623) is 
flipped to sort the eigenvalues in decreasing order. The goal is to 
extract the eigenvectors that describe 98 % of  the variance. But since 
the order of the columns of the matrix is not reversed, the function
invoked in line 633 extracts the wrong vectors.

I'm not sure if I'm right, maybe I just didn't get the meaning of the 
code. Maybe somebody else has time to check it?


Thanks,
Matthias