[OSGeo-Discuss] OSGeo ISODATA implementation?

[Nikos Alexandris]

Hi list!

I am searching for an OSGeo implementation (at the very least simply FOSS will 
do) of the ISODATA clustering algorithm [*].  I want to avoid using a 
commercial tool.  Currently I have reviewed the following "options":


GRASS-GIS

GRASS has a modified k-means implementation in its  <i.cluster> module [1].  
However, it requires at least two raster maps to run.  The ISODATA algorithm 
is known to run even on a single raster map.


OpenEV

The "old" (but very good) OpenEV (is convenient since it reads directly grass 
raster maps and) has an integrated ISODATA-based classification tool [2].  
While testing with single Landsat bands ranging  [0,255] and and NDVI image 
[-0.1,1.0], it get stuck somehow after some iterations complaining bout 
"ZeroDivisionError: float division".  Don't know if there any restrictions 
concerning the input data format.


OrfeoToolBox

OrfeoToolBox has also a clustering algorithm [3] but I am unsure that it is an 
implementation of the ISODATA.  Can't trace a clean documentation about it.  
It seems as yet another k-means based algorithm.


Fast ISODATA implementation (by David Mount)

(Not tested yet!)


Any other candidates?
Warmest regards, Nikos

--
[*] ISODATA:

a) <http://www.dtic.mil/cgi-
bin/GetTRDoc?Location=U2&doc=GetTRDoc.pdf&AD=AD0699616>
b) 
<http://en.wikipedia.org/wiki/Multispectral_pattern_recognition#ISODATA_method>

[1] <http://grass.osgeo.org/grass64/manuals/i.cluster.html>

[2] <http://openev.sourceforge.net/>

[  Sometimes it fails giving following message:
--%<---
Traceback (most recent call last):
  File "/geo/osgeo/binaries/FWTools-2.0.6/lib/python2.2/site-packages/gtk.py", 
line 125, in __call__
    ret = apply(self.func, a)
  File "./../tools/isodata.py", line 529, in classify_cb
    d /= d_tresh
ZeroDivisionError: float division
--->%--  ]

[3] Mean-shift clustering: <http://orfeo-
toolbox.org/CookBook/CookBooksu17.html#x31-430002.4.4> and 
 <http://gracilis.carleton.ca/CUOSGwiki/index.php/Image_Classification_Tutorial_using_Orfeo_Toolbox#Clustering>

[4] <http://www.cs.umd.edu/~mount/Projects/ISODATA/>
-------------- next part --------------
A non-text attachment was scrubbed...
Name: signature.asc
Type: application/pgp-signature
Size: 230 bytes
Desc: This is a digitally signed message part.
URL: <http://lists.osgeo.org/pipermail/discuss/attachments/20130223/082adc2c/attachment.sig>