Unsupervised Symmetric Polygon Mesh Mapping The Dualism of Mesh Representation and Its Implementation for Many Layered Self-Organizing Map Architectures

Abstract

With this paper we present a fully automated semantic shape similarity detection based on N-rings with further potential for shape synthesis in a topological correct feature space. Therefore a way of symmetric encoding of geometry, optimized for the use as feature-vector in self-organizing maps, is introduced. Furthermore we present a modified kernel for the detection of the best matching unit in self-organizing maps especially designed for a data topology differing from the default predecessor/successor structure. Finally we provide the results of a conducted experiment clustering building blocks of an area in Zürich, Switzerland.