Program options:
iap: k-means clustering
options:
  -a            accounting (generates .html output data file)
  -d            destroy empty classes 
  -i <mode>     initialization mode:
     random     pixels randomly distributed among classes
     distrib    initial means distributed across color space
     hier       hierarchical initialization method
  -o <tag>      output tag (output image filename is <image>.<tag>.ppm)
  -p <n>        floating-point precision set to n bits
                (must be between 1 and 16, defaults to 4)
  -r            empty classes randomly reinitialized
  -xy <percent> spatial data weighting factor (0-100, defaults to 0).
Initialization methods: Treatment of empty classes:
At the end of an iteration, new means are computed by dividing the channel and xy sums by the number of pixels in the class. In the case where a class has no pixels in it, this would result in division by 0. This program has three options for dealing with these empty classes:

Other information:

These pages created by Valerie Ohm (valerie@ece.neu.edu) at the Rapid Prototyping Laboratory at Northeastern University
Contact: Prof. Miriam Leeser (mel@ece.neu.edu)