[Insight-users] ImageGaussianModelEstimator, possible bug

[Daniel Mace]

All,

I've been working with the ImageGaussianModelEstimator in conjunction 
with the MRF label model for segementation.  However, I kept getting 
runtime errors while running the ImageGaussianModelEstimator with my 
training image.  In my particular case, I was trying to segment 2 
objects (basically the background and the object of interest).  I set my 
Number of classes to two and had my label image with these two classes 
demarked (1 for the background, 2 for the object).  This is where I ran 
into the problem.  According to the EstimateGaussianModelParameters() it 
comments that it expects the label image to be labeled 1,2, N (where n 
is the number of classes), while the memory references be labeled from 
0, 1,..., n - 1.  However, there doesn't seem to be a correction for the 
label index to the correct memory reference.  As such, it tries to 
reference the n+1 unit of the array and throws an error.  At the same 
time, however, it doesn't allow you to just rewrite your label image 
from 0,1, ... n -1 units as line 219 does not allow labels of 0:

----------------------
if(classIndex > 0)
----------------------

The quick workaround I have done is to decrement the classIndex variable 
by one and rewrite the above line as

---------------------
if(classIndex >= 0)
---------------------

as well as line 214:

------------------------
if( classIndex > numberOfModels )
-------------------------

to

------------------------
if( classIndex >= numberOfModels )
-------------------------


However, I'm not sure if this was the intended purpose of the labeling.

Cheers,
Dan