[Insight-users] Watershed plateau problem

[Aymeric Perchant]

Hi Josh,

in fact, I am not working on a gradient image because my problem is 
quite specific...
One central problem is when your image does not have enough grey level, 
and is not "lower complete": each non minimal pixel
 should have a lower neighbor. I just found that there exists algorithms 
to transform an image with plateaus into a lower complete image,
perhaps it could be added to ITK ? I have just began to work on it :  I 
will be abble to share it soon if you beleive it could be usefull for 
ITK...

Aymeric

Joshua Cates wrote:

>Hi Aymeric,
>
>Yes the ITK watersheds implementation is not morphological watersheds as
>you describe so if you need that particular algorithm it will have to be
>implemented separately.  You may be able to reuse some of the components
>of the existing watershed filter.
>
>In our experience thus far with the ITK implementation and medical images,
>we have found that plateaus typically only result when the edge image used
>as input to the watershed filter is thresholded.  In other words, we only
>find plateaus that we artificially create in the data.  This makes them
>less of an issue because any such thresholding is usually done to remove
>low, background values.
>
>
>Josh.
>
>______________________________
> Josh Cates			
> School of Computer Science	
> University of Utah
> Email: cates@sci.utah.edu
> Phone: (801) 587-7697
> URL:   http://www.sci.utah.edu/~cates
>
>
>On Mon, 19 May 2003, Aymeric Perchant wrote:
>
>  
>
>>Having read the fabulous InsigthSoftwareGuide.pdf I discovered that the
>>watershed implementation was a "top down", "steepest gradient descent"
>>implementation that works well on float, or even color images.
>>
>>A morphological watershed on a binary image should be a skiz (skeleton
>>by influence zone). The ITK watershed does not have this feature. The
>>real problem is that it does not behave well on plateaus : a given
>>plateau in an image can only belong to one segment. Whereas it should be
>>split according to what nearest minima a pixel belong to.
>>
>>How can I adress this plateau problem with itk ? Do I have to write a
>>morphological watershed, like the one Vincent & Soille proposed (IEEE
>>PAMI, Vol 13 No 6, june 1991 : Watershed in digital spaces: An efficient
>>algoritm based on immersion simulations) ?
>>
>>Thanks,
>>
>>
>>Aymeric Perchant
>>
>>
>>_______________________________________________
>>Insight-users mailing list
>>Insight-users@public.kitware.com
>>http://public.kitware.com/mailman/listinfo/insight-users
>>
>>    
>>
>
>
>  
>