[Insight-users] RGBGibbsPriorFilter
Barry Branham
bhb at branham.ws
Mon Oct 5 23:31:59 EDT 2009
I have a project to segment the left ventricle endo and epi walls from
MRI images. After reading the ItkSoftwareGuide.pdf, the book "Insight
into Images" and various papers on segmentation, several by Dimitris
Metaxas and Ting Chen. I decided to try the hybrid method using Gibbs
Prior and Deformable models.
I've found several issues with RGBGibbsPriorFilter:
1. Although advertised as for 2 or 3D, It didn't work on 2D images
(which I need because my images are 8mm apart). I had to add an
ImageDimension check in Allocate() and set m_ImageDepth = 1 if Dim==2.
2. Has ''#define RGBGibbsPriorFilterNeedsDebugging 1' and only does
energy minimization if not #defined. So are left with labeling based on
membership functions from the classifier.
3. Has 'SetTrainingImage()', but doesn't use it.
4. Allocates a 'm_MediumImage', but doesn't use it.
5. Only uses the membership value of the first membership (seed)
function, so can only use one seed.
6. Calls the classifier Update() function, gets the resulting classified
image, then relabels it
using membership function 1 (1st seed) so classifier results are ignored.
I conclude that RGBGibbsFilter is unfinished. I don't understand how
it's not in the Review category. Can anyone comment on this or know if
anyone has done further work on this filter?
Thanks,
Barry Branham
More information about the Insight-users
mailing list