Newsgroups: sci.image.processing
Path: cantaloupe.srv.cs.cmu.edu!das-news2.harvard.edu!news2.near.net!news.mathworks.com!news.alpha.net!uwm.edu!math.ohio-state.edu!jussieu.fr!lcr.thomson-csf.fr!news.thomson-lcr.fr!najman
From: najman@thomson-lcr.fr (Laurent Najman)
Subject: Re: Watershed algorithm shortcut?
Message-ID: <D0FtEx.LwM@news.thomson-lcr.fr>
Sender: news@news.thomson-lcr.fr
Nntp-Posting-Host: briseis
Organization: Thomson LCR
References: <3bvdpg$sp8@news.CCIT.Arizona.EDU> <D0F63v.6K0@news.otago.ac.nz>
Date: Wed, 7 Dec 1994 11:04:09 GMT
Lines: 52


In article <D0F63v.6K0@news.otago.ac.nz>, nmein@bifrost.otago.ac.nz (Nick Mein) writes:

   Te-Shen (Dickson) Liang (liang@mipl.ece.arizona.edu) wrote:

   >  You said "get good results". I wonder if the results you got consists only
   >  the catchment basins. That is, there is no watershed pixel after the final
   >  flooding ? 

   Thats right.

   >  If that is true, isn't it one of the reason for the displaying
   >  of the watershed pixels is to demonstrate the result of segmentation ?

   Well, I suppose that my question is - exactly what _is_ the significance
   of the WSHED pixels? Are they only "cosmetic", or are they necessary for
   the correct operation of the algorithm?

The WATERSHED value is needed to have a correct result, i.e. without
deviation, but in practice, these kinds of deviation do not happen
frequently. If you apply the watershed algorithm on the gradient, you
obtain a lot of small regions, so the deviation is not important. If
you apply first a filter on the gradient, like a geodesical
reconstuction, large plateaus appear, and so large deviations may
happen. Furthermore, if you REALLY need a correct result, you should
use an auxilliary distance function for the propagation on the
plateaus. See Vincent's thesis for more information.


@PhdThesis{Vincent90,
  author = 	 "Vincent, L.",
  title = 	 "{Algorithmes Morphologiques \`a Base de Files
		  d'Attente et de Lacets~: Extension aux Graphes}",
  school = 	 "\'Ecole des Mines",
  year = 	 1990,
  address =	 "Paris, France",
  month =	 "May",
  type =	 "Th\`ese"
}


Laurent Najman

-- 
================================================================================
	Laurent Najman
	Thomson-CSF - Laboratoire Central de Recherches
	Domaine de Corbeville 91404 - Orsay Cedex / France
	Tel : (33-1) 69 33 09 17 / Fax : (33-1) 69 33 08 65
	Telex : THOM 616 780 F
	E-mail :     najman@thomson-lcr.fr
================================================================================
