[Insight-users] active contour level set segmentation

[Amy C]

Hello,
Is this the right list to send questions about code in Review? I'm sorry if
it isn't. Question is aimed towards anyone familiar with this particular
submission.

I've been looking at the Active Contours Without Edges code (level set
segmentation) in publication 323 of the insight journal which is currently
in the 'Review' folder of the itk code. However I'm looking at the
(20-Dec-09) CVS version since the version linked to by
insight-journal.orgis old and has a couple of errors that have since
been fixed.

I wanted to know about AtanRegularizedHeavisideStepFunction ( H_eps(x) = 1/2
+ 1/pi * atan(x/eps) ), the header file evaluates the derivative as
   1/pi*1/(1+x^2/eps^2),
but I think it should be
  1/(pi*eps) * 1/(1+x^2/eps^2).
Is this correct?

Also, could anyone tell me if the VolumeRegularizationTerm comes from a
particular paper or something?
(RegionBasedLevelSetFunction->ComputeVolumeRegularizationTerm()). The
doxygen does say what it is (in the PDE), but I wondered what it
corresponded to in the energy functional. (It looks like it encourages the
volume of the level set to be a particular value. The original papers use
positive inside/negative outside for level sets which is opposite to the
common convention. I just want to verify the PDE for myself by deriving it
from the functional, so that I can stop being confused about the flipped
positive and negative signs in the code :p). And this term is multiplied by
a factor of 2 in the PDE (as well as the user-supplied weight) which seems
strange, since most squared terms in a functional would have a factor of .5
to counteract this.

Cheers,
Amy
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://www.itk.org/pipermail/insight-users/attachments/20091222/9af9dc38/attachment-0001.htm>