Clustering is a grouping of data with similar characteristics. Clustering is divided into hierarchical clustering and non-hierarchical clustering as mentioned as follows.
a. Hierarchical Clustering
The similarity of a cluster is evaluated using a "distance" measure. The minimum distance between clusters will give a merged cluster after repeated procedures from a starting point of pixel-wise clusters to a final limited number of clusters.
Figure 11.3.1 shows the general procedure of hierarchical clustering.
The distances to evaluate the similarity are selected from the following methods.
(1) Nearest neighbor method
Nearest neighbor with minimum distance will form a new merged cluster.
(2) Furthest neighbor method
Furthest neighbor with maximum distance will form a new merged cluster.
(3) Centroid method
Distance between the gravity centers of two clusters is evaluated for merging a new merged cluster.
(4) Group average method
Root mean square distance between all pairs of data within two different clusters, is used for clustering.
(5) Ward method
Root mean square distance between the gravity center and each member is minimized.
b. Non-hierarchical Clustering
At the initial stage, an arbitrary number of clusters should be temporally chosen. The members belonging to each cluster will be checked by selected parameters or distance and relocated into the more appropriate clusters with higher separability. The ISODATA method and K-mean method are examples of non-hierarchical clustering.
The ISODATA method is composed of the following procedures (see Figure 11.3.2).
(1) All members are relocated into the closest clusters by computing the distance between the member and the clusters.
(2) The center of gravity of all clusters is recalculated and the above procedure is repeated until convergence.
(3) If the number of clusters is within a certain specified number, and the distances between the clusters meet a prescribed threshold, the clustering is considered complete.
Copyright © 1996 Japan Association of Remote Sensing All rights reserved