Figure 3

gSPC processing of simulated data using different K. The clusters detected for different K are shown as ellipsoids. On the first level K = 2, three most compact clusters are detected by the algorithm. The representative cases from these clusters will stay for further step of analysis, K = 6. For K = 6 two clusters are detected and, eventually for K = 20 one cluster will be detected. Thus the gSPC algorithm detects embedded clusters. The representative cases are used to restore hierarchical organization of clusters shown at the bottom of the figure. Notice, that for the upper clusters, the same sequence is the representative one for K = 2 and K = 6.