In this paper knowledge-based (re)clustering and some of it applications in two domains was presented. A clustering algorithm when guided by knowledge can be iteratively applied to focus-of-attention patterns to enhance unsupervised classification and achieve pattern labeling. Over-clustering provides useful information, as well as limiting the degree of under-segmentation. The idea has been demonstrated with a fuzzy c-means clustering algorithm (FCM), but can easily be implemented using other clustering methods.
ieeetr