I have a question about how initial cluster seeding is performed by the ISODATA algorithm as implemented in ENVI. The only description I've been able to find is that it "calculates class means evenly distributed in the data space." Does anyone know how many cluster centres are calculated during the initial seeding? As in, if 5 is the minimum number of classes specified by the user, does the algorithm calculate five initial cluster centroids that are evenly distributed throughout the data?
|