Abstract

This paper derives and compares two strategies for minimizing topological constraints in a query expressed by a visual example: (1) elimination of topological relations that are implied uniquely by composition and (2) restriction to topological relations that relate near-neighbor objects, as determined by a Delaunay triangulation. In both cases, the query processing approach is to solve a constraint satisfaction problem over a graph of binary topological relations. Individuals and the combination of the composition- and neighborhood-based strategies were implemented and compared with respect to their ability to reduce topological constraints, and with respect to the quality of the results obtained by a similarity-based searching that uses these pre-processing strategies. The main conclusion of this work is that similarity queries that are formulated in a visual language should exploit the metric characteristics of the configuration, even if only topological constraints are considered for making matches. © Springer-Verlag Berlin Heidelberg 2003.