This paper details master's thesis work involving research and investigation into the approach of self-organizing maps for clustering of data, more specifically, clustering of image data, and how this can be used in understanding image composition. This work will build upon ideas which have previously been explored, such as using self organizing maps for identifying and grouping different regions of an image which may possess similar features. A large part of this research is based upon experimentation with a variety of topological models of the self-organizing map network and investigation into what advantages these different topologies afford the network in terms of its clustering capabilities. |