A neural net model for unsupervised pattern classification and its application to image segmentation

This paper describes a new neural net model for unsupervised pattern classification, which is known as generalized entropy mapping (GEM) net. The frame-work of generalized information entropic theory is described to represent the characteristics and performance of a GEM-net. The organization of a GEM-net is hierarchical. The principal contributions of the paper are mainly the following two aspects: (1) establishing the global optimization net based on generalized entropy measurement; and (2) a scheme of self-organizing cluster validation by means of unsupervised parallel recursive algorithm is proposed. The preliminary experimental results show that the performance of a GEM net is efficient