next up previous
Next: Connected-Component Labelling Up: Characteristic Neighbourhood Previous: Characteristic Neighbourhood

Unimodal Thresholding

[87] yields an optimal threshold for an adequate separation of a unimodal distribution. No heuristic parameter is involved in the process. In the structural identification, this algorithm is applied to separate a small number of decisive, high energy clique families from the rest ones.

The thresholding algorithm draws a straight line from the peak of the histogram to the last non-empty bin. The threshold is at the point on the histogram curve having the maximal distance to the straight line. The algorithm is illustrated in Fig 6.7.



dzho002 2006-02-22