[Insight-developers] bug in itkGradientVectorFlowImageFilter

[Kishore Mosaliganti]

Hi,

itkGradientVectorFlowImageFilter does not make use of image spacing.
It assumes isotropic spacing of 1 in all dimensions. This is a
problem. In particular, the ITK implementation uses the definition of
'r' in Equation 17 which will change for anisotropic cases.

Also, the CFL condition for estimating the time-step was incorrectly
being computed. The original paper, "Snakes, Shapes, and Gradient
Vector Flow" shows the equation incorrectly in Equation 18. It is well
known that for the diffusion equation, it is (delta_x)^2 and not
delta_x in the denominator.

I corrected the code locally in my machine. Basically, I had to set
the spacing in new images that were allocated and for the
LaplacianImageFilter.

I am attaching the modified code.


Kishore
-------------- next part --------------
A non-text attachment was scrubbed...
Name: itkGradientVectorFlowImageFilter.h
Type: text/x-c++src
Size: 5715 bytes
Desc: not available
URL: <http://www.itk.org/mailman/private/insight-developers/attachments/20091005/fbbe15ec/attachment.cc>
-------------- next part --------------
A non-text attachment was scrubbed...
Name: itkGradientVectorFlowImageFilter.txx
Type: application/octet-stream
Size: 8469 bytes
Desc: not available
URL: <http://www.itk.org/mailman/private/insight-developers/attachments/20091005/fbbe15ec/attachment.obj>