SHAPE APPROXIMATION OF ARC PATTERNS USING DYNAMIC NEURAL NETWORKS

A shape representation technique for two-dimensional patterns using a dynamic variation of Kohonen's self-organizing feature maps is discuss ed. In Kohonen's map, the number of processors is fixed and is to be s pecified a priori. This number, if not properly chosen, can cause eith er wastage or shortage of processors. The problem is overcome, in this paper, by incorporating dynamic growing and shrinking capabilities in the network. Shape of only are patterns is considered here.