<title>Fast association-rule-based similarity search in 3D models</title>

Abstract

Advances in automated data collection tools in design and manufacturing have far exceeded our capacity to analyze this data for novel information. Techniques of data mining and knowledge discovery in large databases promise computationally efficient and accurate means to analyze such data for patterns and similar structures. In this paper, we present a unique data mining approach for finding similarities in classes of 3D models, using discovery of association rules. PCA is first performed on the 3D model to transform it along first principal axis. Transformed 3D model is then sliced and segmented along multiple principal axes, such that each slice can be interpreted as a transaction in a transaction database. Association-rule discovery is performed on this transaction space for multiple models and common association rules among those transactions are stored as a representative of a class of models. We have evaluated the performance of association rules for efficient representation of classes of shape models. The method is time and space efficient, besides presenting a novel paradigm for searching content dependencies in a database of 3D models.