watershed

[Emmanuelle Gouillart]

Hi Daniel,

you might want to ask the question on the cell profiler mailing-list as
well since it's their implementation that was included in scikit-image
(cellprofiler-dev <cellprofiler-dev at broadinstitute.org>).

Cheers
Emmanuelle

On Wed, May 20, 2015 at 04:06:52PM +1000, Juan Nunez-Iglesias wrote:
> Hi Daniel,

> Actually there's a known bug [1]_ in the scikit-image implementation (which is
> originally from CellProfiler). I haven't figured it out yet, nor do I know what
> the exact differences are to the paper. (And most images are unaffected.) Are
> you looking to use the implementation or are you trying to implement your own
> as a learning strategy? If the latter, you might want to have a look at my own
> implementation [2]_ in gala, for which I tried to follow the paper exactly. But
> it's way slower.

> Juan.

> .. [1]Â https://github.com/scikit-image/scikit-image/issues/803
> .. [2]Â https://github.com/janelia-flyem/gala/blob/master/gala/morpho.py#L212

> On Wed, May 20, 2015 at 7:00 AM, Daniel Bliss <daniel.p.bliss at gmail.com> wrote:

>     Hi all,

>     I'm trying to implement the watershed algorithm described in this paper in
>     Python.  Looking at the watershed function in skimage.morphology, it's not
>     clear to me whether this function is an exact implementation of Vincent &
>     Soille's method.  The docstring for morphology says, "Some ideas taken from
>     . . ." and lists an older (by one year) paper by Soille. Â "Some ideas"
>     seems to suggest that the implementation is not exact, but where the
>     differences are is not clear.  I've read through the morphology source
>     code, as well as the C code it draws upon, but I'm getting tangled up
>     trying to translate between Vincent and Soille's description and the C/
>     Python code in skimage.  Can anyone provide a clear and complete
>     explanation of what exactly skimage.morphology.watershed does relative to
>     the Vincent and Soille 1991 method?

>     Many thanks,
>     Dan