GRAPH BASED METHODS FOR LOCALISATION BY A SKETCH
Matthias Kopczynski, Monika Sester · 2005
This paper describes how a sketched location can be identified in a large collection of reference data. This is done by encoding the information of both datasets in a graph based structure, which preserves the topological properties of the sketched area. Finding the correspondence is a matching problem. In order to reduce complexity first a set of possible correspondences are identified using the following approach. From these graphs the frequency of relevant objects and relations is extracted and represented as a histogram. Distance measures for probability distributions can be used to judge if they are similar enough to represent the same location. Considerations on the information contents are leading to a way how to choose object types and relations used for the matching process. The distance measures are the fundamental tool for an efficient retrieval strategy. Since they declare a metric space, they can be used in a metric tree. Only promising locations are then considered in the matching process.