[Insight-users] Canny-Edge Level Set Segmentation

Luis Ibanez luis.ibanez at kitware.com
Mon Jan 15 16:40:15 EST 2007


Hi Anna,

The choice of the threshold value is entirely image-dependent.

This parameter defines how strong an edge should be in the image
in order to be considered a Canny edge. The value of the threshold
parameter is applied to the Gradient Magnitudes of the input image.

Therefore a rule of thumb for selecting this number is to take your
input image, and apply to it a Gaussian-Smoothed Gradient Magnitude
(with the same variance value that you are using here), and then
probe the image in regions that you consider have edges that should
be considered (maybe because they are anatomically significant).

By selecting a value 7.0 we are telling the Canny filter that all
pixels with gradient magnitude greater or equal to 7.0 should be
considered to be significant edges.

If you use a threshold value that is too high, then you get less
Canny edges, and they will be the strongest edges in the image.

If you use a threshold value that is too low, then you get a lot
of Canny edges, many of which may not be anatomically significant
and may even be just related to the image noise.

The user-friendly way of setting this value could be to have an
interactive application where you move a slider for selecting
the threshold and it shows you the effect on the image.


You can use the application

    InsightApplications/GaussianFilter

in order to get a feeling on the typical Gradient Magnitude
values of your image.

You could also devise a semi-automatic mechanism based on
evaluating the histogram of the gradient magnitude image.

Yet another way could be to have an interactive "region-of-
interest" selectino where a use selects a region with edges
that are considered important, and a region with edges that
are to be neglected. Then your program could use the values
of Gradient magnitude from both regions in order to compute
a reasonable threshold.


Note that the typical difficulty is that the thresholds that
are suitable for some regions of the image are not so for
other regions.


    Regards,


       Luis


-------------------
Anna Vignati wrote:
> Hi all,
> 
> I've got a question about Canny-Edge Level Set Segmentation. In the
> example:
> 
> Examples/Segmentation/CannySegmentationLevelSetImageFilter.cxx
> 
> The value of the parameter "threshold" is 7.0. Do you know what is the
> rational behind this choice? I mean, How can I choose the value for this
> parameter for a given input image?
> 
> I was searching through the mailing-list archive but I couldn't find 
> anything about it.
> 
> Thank you in advance.
> Anna.
> 
> _______________________________________________
> Insight-users mailing list
> Insight-users at itk.org
> http://www.itk.org/mailman/listinfo/insight-users
> 


More information about the Insight-users mailing list