Implementing a neural network for a progressive fuzzy clustering algorithm

A neural network based scheme for the detection of fuzzy cluster prototypes from normalised histogram of a real world image is described. These prototypes can be subsequently processed outside of the neural network to enable viable real time fuzzy clustering application. The direct mapping of the histogram data whilst simple to implement, is demonstrated to suffer from errors related to a bias condition associated with weight distribution of the network. The proposed method mitigates this problem by using a conventional backpropagation neural network with output responses trained to five points of a fuzzy membership function. Test responses from this network yielded less than 5 percent error for the prototype centres