Extension of Graph Distance Based SOM to Higher SOM and its Application to Gesture Recognition

Abstract

In the field of pattern recognition, it is useful that a manifold is estimated from data distribution. A graph distance based SOM (GSOM) is used for precise estimation of the manifold. Furthermore, GSOM is extended to higher order SOM to classify multiple manifolds. a gesture is defined as a set of patterns, and spans a manifold in a pattern space. In this paper, a new gesture recognition method which is based on GSOM based higher SOM is proposed.