SPATIAL RELATION REPRESENTATION AND PROCESSING

Abstract

Spatial relation representations have been the central topic for a variety of research fields such as Spatial Databases, Image Databases, and Geographic Information Systems etc. during the past few years. Typically, most existing results have considered direction and topology as two independent components of spatial relationships. Only loosely coupled pairwise representations, which, as will be discussed in this thesis, fail to accurately capture the genuine spatial relationships, have been proposed to combine these two kinds of relationships. This thesis studies the representation and processing of spatial knowledge. In particular, the thesis concentrates on qualitative spatial representation, similarity retrieval and reasoning. The work here extends and augments existing results. Looking al available results on qualitative spatial representations such as 2D string [21], 2D-C string [46], Symbolic Array [53], 4-inlersection [22], and others [34][51], the thesis presents an efficient scheme in which both directional and topological spatial relationships are naturally combined together. The proposed representation scheme, called 2D-2T, captures more accurate and complete spatial relationships that otherwise could not be captured or are distorted using available methods. Similarity retrieval based on 2D-2T representation has been addressed based on both qualitative and quantitative similarities, which take into account both topological and directional relations. A set of reasoning rules based on 2D-2T representation has been proposed and these rules are proven to be sound and more powerful than current work. Also, storage overhead of 2D-2T has been analyzed with a rich set of experiment data. ln addition, extensions on 2D-2T have been made to further handle complicated spatial scenes involving distance relationships and spatial regions with holes. Such an integrated framework can serve as a useful basis to the development of solutions for a variety of applications.