In the example illustrated in Fig. 12 and 13, the entire original matrix undergoes the transformation and produces seven final clusters with a good selectivity and performance. Performance in this case serves as an indicator of a high "harmony level" in the given system of objects and variables. In systems of objects where at least one object appears to display high proximity to each of the primary clusters, the informographic analysis produces broadened peaks with drastically decreased resolution. Thus, the above example represents a particular case of infospectral analysis.

Clustering resolution can be improved by applying the following procedure. After a similarity matrix is ultimately divided into two primary sections, and an object with a proximity to both sections appears in one of the sections, each of the sections separately undergoes a new transformation. Upon further transformation of the two sections, the "foreign" object either goes to a separate cluster or - if appropriately - stays within a respective section and participates in the process of its hierarchical division.

The efficiency of this technique is demonstrated on the following example. Fig. 14 represents a similarity matrix for a complex graph of size 50. Fig. 15 shows both the informograms of the original matrix with the largely broadened peaks indicating a very low resolution, and the informograms of the groups obtained from the original matrix by cascade-dichotomous resolution. The latter informograms display good resolution resulting from high selectivity and performance. Detailed informograms of two selected groups, 1.1 (pink background) and 2.2.2.1 (blue background) are shown in Fig. 15. In this example of graph partitioning, the NBI-algorithm decides by itself which of the edges in the original graph have to be broken for achieving the most harmonious partitioning.