[Insight-users] possible bug in itkScalarSinglePhaseSparse2DTest.cxx

[Juan Cardelino]

In addition, the dense version of this filter shows the same
behaviour. In addition, when changing the area regularization term, I
only see effects when setting  a value of 1000, and still the results
are only slightly different.
Finally, using the same parameters, the resulting contour of the
sparse version is very rugged compared with the dense version.
I'd expect some differences between them due to different numerical
discretization, but maybe it is too much.
Any comments?
Best regards,
                   Juan

On Wed, Apr 20, 2011 at 4:21 PM, Juan Cardelino
<juan.cardelino at gmail.com> wrote:
> Dear all,
>           I'm using the Chan and Vese segmentation method and I am
> currently testing ITK's implementation. The code in the review folder
> works in the provided example and also with my images, but I'm
> experiencing a strange behaviour: it seems that it gives always the
> same results regardless of some of the input parameters given. Namely,
> changing the following parameters (in the sparse version) doesn't
> affect the result at all:
>  double curvature_weight = 0.;
>  double reinitialization_weight = 0.;
>  double l1 = 1.;
>  double l2 = 1.;
> In my tests, I've changed the curvature weight to higher values, to
> get a smoother contour, but the result does not change (diff'ing the
> images gives 0 differences). In addition, I've tried with the
> reinitialization weight and without any luck.
> Thinking it twice it seems reasonable, because in a narrow band
> approach you won't use a reinitialization term. In addition, the
> curvature term seems redundant with the area and volumen
> regularization terms. I tried to find out the explanation of these
> parameters, but the Insight Journal Submission has different names for
> them.
> So, can someone clarify on these points and tell me the difference
> between the length regularization and the curvature term? I thought
> the should be the same, but they have different parameters.
> Thanks in advance.
> Best regards,
>                   Juan
>