i.cluster

[Sincak]

On Mon, 17 Aug 92 12:50:40 PDT Greg Gaston said:
>I am working with i.cluster and i.maxlik on multi-temporal GVI data for the
> former Soviet Union and I have some questions:
>
>- Has anybody worked out a method of establishing the "natural" number of
> clusters that occur in multi-image data?  Or a description of how changing the
> input parameters can change the results?
>I have experimented with the input parameters that i.cluster asks for and I
>have
> some interesting results, however I would like to be able to justify my
>results
> with more than "it just looks better".
>
Well it is rather methodical question than clearly connected to
GRAss . Isn't it? I think so that number of cluster depends on the
image and also very inportant question is what is cluster ?? Also
question is what kind of methods are You using ? I dont know what
kind of algorythm is build in GRASS ?? According to my small experience
very logical seems to be AMNO algorythm which is seaching for modes
- AMNO (Analysis of Modes) when we assume multimodal normal data distri-
bution. Well maybe You can try it. If You need more fell free to
touch me directly.
> indicate classes are closer than large numbers... but what are the units? When
> I specify a class separation of .7, what units am I using and can I use this
> matrix to compare the within class variation to the between class variation?
>
>Thanks, Greg G.
well the cluster evaluation is also not so easy. You have mentioned
intercluster evaluation - there is also intracluster evaluation using
compactness coef. If You would like to use it - I can send You a referenc
of it.
Comming back to AMNO - the input parameter is not a number of clusters
but coef. ALFA.
    Anyway if You are interested in methodical problems of image
processing - there was created list IMAGRS-L at CSEARN.BITNET which is devo-
ted to image processing of remotelly sensed data. It is truth that it
not very active but maybe You can find it useful(there is aprx.200people)
  Sincerely
******************************************************************
  I    I     I
  I----I     I    ----- it is    A H O J     in Slovak language
  I    I     I               &  Slovakia is a part of
                         Czecho-Slovakia in Central Europe
        Peter Sincak, Ph.D.
                 Kosice, Technical University of Kosice, E-Fac.
  Department of Cybernetic & Artificial Inteligence
  Czecho-Slovakia,     E-mail  - KUKO at CSPUNI12.BITNET
                                 KUKO at ACI.CVUT.CS
                       FAX     - 0042 95 343 92
                       Voice   - 0042 95 578 64
                  CITY AIR-plane connection = YES
*****************************************************************

----------------------------------------------------------------