[Insight-users] Geodesic dilate/erosion

Quan Chen quanchen at gmail.com
Tue Feb 1 13:26:07 EST 2005


Well, my purpose is to perform some sort of fast marching
segmentation.  I was not sure whether this geodesic dilation could
give me better result than fast marching.  Will these two method
generate similiar results on a binary mask image if the geodesic
algorithms can run user defined iteration?

Anyway, my application is doing some CT segmentation.  Now I use
fastmarching to get a initial guess and use deformable3DMeshFilter to
refine the estimate.  The results are pretty good.  BTW, I tried to
use GradientVectorFlow as input to deformable3dMeshFilter but I find
it doesn't perform very well.  I suspect it is because I didn't set
the correct parameter.  But I really don't quite understand what
SetNoiseLevel() and SetTimeStep() function in
GradientVectorFlowImageFilter do.
I didn't observe much change in the final segmentation results when I
modify those parameters.  And for SetIterationNum(), 1 iteration gives
me the very good result (similiar to no GVF) and more iteration just
make it worse.  So it appears the GVF do more harm than good in my
case.

BTW, in GenerateData() function in
itkGradientVectorFlowImageFilter.txx, there is a line "m_TimeStep =
0.2/m_NoiseLevel; ".  So it seems that the SetTimeStep() function
don't have real effect.  I am wondering if it was forget to be removed
by the original programmer after his debugging.  A somewhat similiar
issue occurs in the itkCannyEdgeDetectionImageFilter.txx where there
is only one SetThreshold() function to set m_Threshold but the
variable used in the algorithm is m_UpperThreshold and
m_LowerThreshold.  The function GenerateData() have the line " 
m_UpperThreshold = m_Threshold;
  m_LowerThreshold = m_Threshold/2.0; "
I believe there should be SetUpperThreshold() and SetLowerThreshold()
function to set it seperately, as
http://www.cee.hw.ac.uk/hipr/html/canny.html shows difference it can
make by using different Upper and Lower threshold combination.

On Mon, 31 Jan 2005 14:17:00 -0500, Miller, James V (Research)
<millerjv at crd.ge.com> wrote:
> It is true that there is only grayscale versions of the geodesic dilation
> and erosion.  However, they can operate on binary data without any issues.
>  
> The techniques do only provide a "run one iteration" or "run to convergence"
> capability. I have only seen geodesic morphology algorithms that do one or
> the other.  Are you aware of any geodesic morphology algorithms that run for
> a specified number of iterations?  I only set up to run once or run to
> convergence since all the published algorithms I saw only did those.
>  
> Jim
> -----Original Message-----
> From: Quan Chen [mailto:quanchen at gmail.com]
> Sent: Wednesday, January 26, 2005 4:18 PM
> To: insight-users at itk.org
> Subject: [Insight-users] Geodesic dilate/erosion
> 
> Is it ture that the ITK geodesic dilate/erosion only have grayscale version?
>  And the GrayscaleGeodesicDilateImageFilter and
> GrayscaleGeodesicErodeImageFilter only provide two options, run one
> iteration or run till convergence.  I am wondering if it is possible to
> provide a function that can dilate or erode the user defined number of
> iterations. 
>  
>  
>  
>


More information about the Insight-users mailing list