Roop Goyal and
Max Egenhofer Seventh International Symposium on Spatial and Temporal Databases, Los Angeles, CA,
C. Jensen, M. Schneider, B. Seeger, V. Tsotras(eds.), Lecture Notes in Computer Science, Vol. 2121, Springer-Verlag, pp. 36-55, July 2001.
Abstract
Like people who casually assess similarity between spatial scenes
in their routine activities, users of pictorial databases are often
interested in retrieving scenes that are similar to a given scene,
and ranking them according to degrees of their match. For example,
a town architect would like to query a database for the towns that
have a landscape similar to the landscape of the site of a planned
town. In this paper, we develop a computational model to determine
the directional similarity between extended spatial objects, which
forms a foundation for semantically meaningful spatial similarity
operators. The model is based on the direction-relation matrix. We
derive how the similarity assessment of two direction-relation
matrices corresponds to determining the least cost for transforming
one direction-relation matrix into another. Using the
transportation algorithm, the cost can be determined efficiently
for pairs of arbitrary direction-relation matrices. The similarity
values are evaluated empirically with several types of movements
that create increasingly less similar direction relations. The
tests confirm the cognitive plausibility of the similarity model.
PDF
Roop Goyal and
Max Egenhofer