70
2: Raghuveer M. Rao, Manoj K. Arora
sense. However, it is susceptible to impulsive disturbance if the distribution is
non-Gaussian with significant outlier probability.
Once the clustering is done, it may be observed that the clustering is not
entirely satisfactory in the sense that there may be too fine or too coarse
a clustering. Under such circumstances close clusters may be merged if the
clustering is too fine or the clustering procedure may be repeated requesting
a higher number of clusters if it is too coarse. These decisions are driven by
300
2SO
200
150
100
50
0
300
300
Fig. 2.10. a Input color image. b 3D scatter plot of R, G and B components. c Pseudocolor
map of unsupervised classification. For a colored version of this figure, see the end of the
book