[Insight-users] Chan-Vese Level Sets in ITK: where to tweak speed term for "deflation"?

[Audette, Michel A.]

Dear ITK users,

I'm playing with the Chan-Vese level sets example SingleLevelSetWhitaker; when I use the cells.png image, it seems to work fine: the contour is initialized completely outside of the cell shapes, and the front is inward moving, culminating in a fairly successful segmentation. When I try to substitute some other image, and initializing the model completely outside the structure of interest again, for whatever reason, the contour seems push _outwards_ . The only thing that I've changed is the input image, and the size given to the region class.

What would account for this type of behavior? To be frank, I don't understand where the speed term comes from (the so-called advection term in Malladi and Sethian...), so I don't know how to tweak this term in Chan-Vese .

Thanks for your kind support.

Cheers,

Michel

Michel Audette, Ph.D.
Assistant Professor,
Department of Modeling, Simulation and Visualization Engineering,
Old Dominion University,
Norfolk, VA.

________________________________
From: Audette, Michel A.
Sent: Friday, June 07, 2013 4:13 PM
To: insight-users at itk.org
Subject: 2 questions about Level Sets in ITK: triangulating isosurface and how to tune ShapeDetectionLevelSetSegmentation

Dear ITK users,

I have two questions on Level Sets in ITK.

Is there a Level Sets model that is triangulated, i.e.: produces a watertight polygonalization of the zero-level isosurface? Is there a way to produce this watertight triangulation? I know that VTK has Marching Cubes, but I have misgivings about this method producing a watertight result.

Second, I would like to use an example with phantom images that are narrow structures, similar to the Circle of Willis, but in 2D, and I would like to have a recommendation for working example code that I can use with this. I have tried ShapeDetectionLevelSetSegmentation, but I find this example produces a constant speed term, as opposed to a speed term that halts the model on high-gradient areas of the image.

Please advise.

Thanks for your kind consideration.

Michel

Michel Audette, Ph.D.
Assistant Professor,
Department of Modeling, Simulation and Visualization Engineering,
Old Dominion University,
Norfolk, VA.

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://www.itk.org/pipermail/insight-users/attachments/20130610/16052eea/attachment.htm>