[Insight-users] Re: Watershed success

[Luis Ibanez]

Hi Robert,

There are two Watershed plugins in VolView,
both of them under the group:

       "Segmentation Level Set"

The first plugin to try is the

     Watershed RGB Module (ITK)

This module only has the Watershed filter
followed by the color encoding filter.
It is up to you to prepare the input image,
typically by running a gradient magnitude
filter on it. With this filter you may want
to start exploring the image at water levels
around 0.2, and then increase the water level
if you get too many small regions, or decrease
the water level if you get regions that are
too large.

Once you feel comfortable with this plugin,
you can use the

     Watershed Module (ITK)

which has a pre-processing and post-processing
steps.  In preprocessing it computes the gradient
magnitude of the input image for you. Then runs
the Watershed filter on it and finally use the
seed point that you provided in order to extract
the specific watershed where the seed point is
located.  The output of this module is a binary
mask representing only the basin associated to
the seed point. (Note that the seed point is not
directly used by the watershed filter, but by
the post-processing of its output.

----

Please let us know if you still find any
difficulties running these plugins.


   Thanks



      Luis



------------------------
Atwood, Robert C wrote:

> I could not use the watershed module in VolView , it did not do what I
> wanted. I don't quite understand how the module uses the watershed
> filter, it does not seem to return an image with the watersheds coloured
> as in Fig 9.9 of the software guide? Anyways, I took another stab at
> programming using the ITK filters directly, using the Watershed1 example
> and 9.2 of the guide but substituting my own algorithm instead of all
> that diffusion-gradient-filter stuff.
> 
> 
> I went back and coded a 'pipeline' to use the WatershedFilterType on my
> own generated "height-map" generated on my algorithm, and ... looks good
> so far, even without prefiltering and with just the first guess at
> parameters to use. (see attatched screen shot) Still some 'dumb-bell'
> shapes are not separated but , a bit of parameter adjustment (and manual
> line-drawing) should do the trick. (It starts as one connected region)
> 
> The result seems to come out flipped on some axes, though, I haven't
> figured out which ones yet.
> 
> 
> 
> --Robert
> 
> 
> 
> 
>