Incremental Self-Organizing Map (iSOM) in categorization of visual objects

Abstract

We present a modification of the well-known Self-Organizing Map (SOM) in which we incrementally allocate the neuronal nodes to progressively added new stimuli. Our incremental SOM (iSOM) aims at the situation when a stimulus, or percept, is represented by a number of neuronal nodes a typical case in biological situation when the redundancy of representation of data is important. The iSOM is applied to categorization of visual objects using the recently introduced feature vector based on the angular integral of the Radon transform [10]. © 2012 Springer-Verlag.