View Issue Details [ Jump to Notes ] [ Print ]
IDProjectCategoryView StatusDate SubmittedLast Update
0011098ITKpublic2010-08-05 04:452010-10-21 12:31
Reporterroodzeek 
Assigned ToBradley Lowekamp 
PrioritynormalSeveritymajorReproducibilityalways
StatusclosedResolutionfixed 
PlatformOSOS Version
Product VersionITK-3-20 
Target VersionFixed in VersionITK-3-20 
Summary0011098: itkHessianToObjectnessMeasureImageFilter.h incorrect sorting of eigenvalues
DescriptionTested for 3D case.
Obtained completely different final filter results compared with MATLAB implementation of the same algorithm.

According to the Frangi's article where the methodology is described, the eigenvalues are to be sorted |l1|<=|l2|<=|l3|, then parameters rA = |l2|/|l3| and rB = |l1|/(|l2|*|l3|)^(1/2) can be calculated.

Line 148 of itkHessianToObjectnessMeasureImageFilter.h defines sorting function
[...]
return vnl_math_abs(a) > vnl_math_abs(b);
[...]
for sorting the eignevalues in itkHessianToObjectnessMeasureImageFilter.txx line 95 using std::sort().

This causes the eigenvalues to be sorted |l1|>=|l2|>=|l3| thus giving improper filter results.
Additional InformationAttached files in dicom format:

vedbase.vtk.dcm - source image.
Processed using: sigma = 1.25, alpha = 0.5, beta = 0.5, gamma = 30.

INCORRECT_vedbase.vtk-vess,s1.25,a0.5,b0.5,g30.dcm - results obtained using original implementation of the filter from ITK 3.20.

CORRECT_vedbase.vtk-vess,s1.25,a0.5,b0.5,g30.dcm - results obtained after correcing line 148 of itkHessianToObjectnessMeasureImageFilter.h file.

TagsNo tags attached.
Resolution Date
Sprint
Sprint Status
Attached Fileszip file icon datafiles.zip [^] (2,001,340 bytes) 2010-08-05 04:45

 Relationships

  Notes
(0021660)
roodzeek (reporter)
2010-08-05 04:50

 Forgot to write the suggested solution that worked:
Line 148 of itkHessianToObjectnessMeasureImageFilter.h should be:

return vnl_math_abs(a) < vnl_math_abs(b);
(0021661)
Bradley Lowekamp (developer)
2010-08-05 08:27

 Presumably these results were correct with ITK 3.18?
(0021663)
Bradley Lowekamp (developer)
2010-08-05 10:26

 On Aug 5, 2010, at 9:01 AM, Luca Antiga wrote:

Hi Brad,
 I confirm. In fact, with the comparison function currently in the
code the vector is sorted in descending order, not ascending.
If you fix it, it might be worth fixing any affected release (probably
only the latest).
Thanks for the feedback

Luca
(0021706)
Bradley Lowekamp (developer)
2010-08-11 10:02

 The patch was committed and test was added:
http://itk.org/gitweb?p=ITK.git;a=commitdiff;h=73f66bdba383a14f32868bf5695c60710f3b48bd;hp=eb1ee9cd4251a55406444387328b7d7fe1c4fc50 [^]

This work has been integrated into the 3.20.0 "release" branch along with the development "master" branch.
(0021707)
Bradley Lowekamp (developer)
2010-08-11 10:03

 This fix has been committed to the 3.20.0 release, along with the master.

 Issue History
Date Modified Username Field Change
2010-08-05 04:45 roodzeek New Issue
2010-08-05 04:45 roodzeek File Added: datafiles.zip
2010-08-05 04:50 roodzeek Note Added: 0021660
2010-08-05 08:26 Bradley Lowekamp Status new => assigned
2010-08-05 08:26 Bradley Lowekamp Assigned To => Bradley Lowekamp
2010-08-05 08:27 Bradley Lowekamp Note Added: 0021661
2010-08-05 10:26 Bradley Lowekamp Note Added: 0021663
2010-08-11 10:02 Bradley Lowekamp Note Added: 0021706
2010-08-11 10:03 Bradley Lowekamp Note Added: 0021707
2010-08-11 10:03 Bradley Lowekamp Status assigned => resolved
2010-08-11 10:03 Bradley Lowekamp Fixed in Version => ITK-3-20
2010-08-11 10:03 Bradley Lowekamp Resolution open => fixed
2010-10-21 12:31 Gabe Hart Status resolved => closed

Copyright © 2000 - 2018 MantisBT Team