Using a diffusion-like process for clustering
Orly Yadid-Pecht, Moshe Gur · 2002
A simple clustering method using a neural net, which implements a diffusion-like process, is suggested. The implementation requires basic elements, numbered as the number of pixels, that work in parallel. The units can be viewed as simple "neurons", requiring only a small number of local connections. In spite of its simplicity, this implementation has several advantages over commonly used fuzzy clustering methods. Specifically, it is not dependent on initial conditions and it provides the "typicality" notion that is lacking in the well known Fuzzy C means and its derivatives.>