A MODIFIED SELF-ORGANIZING NEURAL-NET FOR SHAPE EXTRACTION

Some modifications on Kohonen's self-organizing feature map are discus sed to make it suitable for finding skeletons of binary images. In Koh onen's feature map, the set of processors and their neighbourhoods are fixed and do not change in the learning process. This may pose proble ms when the set of input vectors represents a prominent shape. The ref erence vectors or weight vectors lying in zero-density areas are affec ted by input vectors from all the surrounding parts of the non-zero di stribution [5]. Hence a shape extraction problem requires a dynamic ch ange in the network topology, In the present paper, to overcome the li mitations of Kohonen's feature maps, we propose a mechanism in which t he set of processors and their neighbourhoods change adaptively during learning, to extract the shape of a binary object in the form of a sk eleton.