Parent Topic: <A href="http://www.pcigeomatics.com/cgi-bin/pcihlp/CLWORKS|Classify">Classify</a><hr>
<H4>Unsupervised Algorithms</H4>
There are three different unsupervised classifiers available. They are:
<P>
<UL>
<LI> K-Means
<LI> IsoData
<LI> Fuzzy K-Means
</UL>
The Fuzzy K-Means unsupervised classifier is available only if the Neural 
Network package has been purchased. The licensing code for Neural 
Networks is ``NN''.
<P>
<H5>IsoData</H5>
This classifier performs unsupervised clustering using the ISODATA.
Up to 16 image channels can be analyzed and 255 clusters (classes) found 
using this program. The algorithm is controlled by the following parameters:
<P>
<br><b>Min Clusters</b><br>
Specifies the minimum number of clusters allowed. This parameter limits 
the number of clusters allowed during Lumping.  This parameter must be 
in the range from 1 to 255.  The default value is 16.
<P>
<br><b>Max Clusters</b><br>
Specifies the maximum number of clusters allowed. This parameter limits 
the total number of clusters allowed during splitting. This parameter must 
be in the range from 1 to 255. The default value is 16.
<P>
<br><b>Desired Clusters</b><br>
Specifies the number of clusters (classes) desired.  It should be noted 
that this is only an estimate. The final number of clusters may vary.
However, the user can limit the variation by setting the parameter's
Min Clusters and Max Clusters.  
This parameter must be in the range from 1 to 255. 
The default value is 16.
<P>
<br><b>Max Iterations</b><br>
Specifies the total number of iterations in calculating the 
cluster mean positions.
<P>
<br><b>Movement Threshold </b><br>
Specifies the movement threshold as a fraction of cluster means.
If the movement of all cluster means is less than Movement Threshold, the 
program has converged. The default is 0.01 or 1%.
<P>
<br><b>Min Threshold</b><br>
Specifies the minimum sample threshold.  The number of samples 
in a cluster domain is compared to Min Threshold in the following way:
 
When the number of samples in a cluster is less than the Min Threshold,
the cluster will be discarded and the total number of clusters
will be reduced by 1.  
 
For a cluster,
<P>
<UL>
<LI> if the standard deviation is greater than the Standard Deviation, AND
<LI> if the average distance of samples from the cluster center 
is greater than the overall average distance of the samples
from their respective cluster centers, AND 
<LI> if the total number of clusters is less than Max Clusters, AND
<LI> if the number of samples in the cluster is greater than 
2(Min Threshold+1), OR
<LI> if the number of clusters is less than or equal to Desired Clusters/2,
</UL>
then splitting of the cluster will occur.
<P>
<br><b>Standard Deviation</b><br>
If a cluster has a standard 
deviation greater than this parameter, splitting may occur.  
A value of 10.0 is reasonable.
<P>
<br><b>Lumping Threshold</b><br>
Specifies the lumping parameter which is used in the following way:<p>
If the distance between two cluster centers is less than this 
parameter, the total number of clusters is greater than Min 
Clusters, and the number of lumped pairs is less than Max Lump Pairs, 
then clusters will occur.
<P>
<br><b>Max Lump Pairs</b><br>
Specifies the maximum number of pairs of cluster centres which 
can be lumped during each iteration.
<P>
<br><b>Background</b><br>
Any pixel with this value will be ignored during the
classification and will be assigned to class 0 ( NULL class).
<P>
<H5>K-Means &amp; Fuzzy K-Means</H5>
This program uses the K-Means or Fuzzy K-means method to 
classify image data into
different clusters. Up to 16 image channels can be analyzed
and 255 clusters (classes) found using this program. The algorithm is
controlled by the following parameters:
<P>
<br><b>Max Class </b><br>
Specifies the number of clusters (classes) desired.
<P>
<br><b>Max Iteration</b><br>
Specifies the maximum number of iterations in calculating the
cluster mean positions.
<P>
<br><b>Min Threshold </b><br>
Specifies the movement threshold as a fraction of the cluster
means. If the movement of all cluster means is less than the minimum 
threshold, then the algorithm has converged. The default is 1%.
<P>
<br><b>Max Sample Size</b><br>
Specifies the number of
samples to collect on which to perform the
the iterative clustering. This defaults to 262144 if not provided
by the user. If the indicated number is larger than the total
number of pixels in the image, then all the pixels in the image
will be used.<p>
The time to compute each iteration is proportional to the number
of samples used. This means that using a lot more than the default 
number of samples
can make the clustering process much slower. Also, all the samples
are stored in memory, meaning that a large NSAM value can lead to 
higher memory requirements. With 262144 samples and five bands 
of eight bit input data, the program would require approximately 1.3MB of
memory while with NSAM set to 2000000, it would take 10MB of memory.
<P>
<br><b>Background </b><br>
Any pixel with this value will be ignored during the
classification and will be assigned to class 0 ( NULL class).
<P>
<hr>
Parent Topic: <A href="http://www.pcigeomatics.com/cgi-bin/pcihlp/CLWORKS|Classify">Classify</a><br>
<a href="http://www.pcigeomatics.com/dev_corner/pcihlp_gateway.html">About PCI Help Gateway</a>