An image stored in image database systems is assumed to be associated
with some content-based meta-data about that image, that is, information
about objects in the image and absolute/relative spatial relationships
among them. An image query for such an image database system can generally
be handled in two ways: exact picture matching and approximate picture
matching. In this paper we show the intractability of matching of spatial
relationships between a query image and an image stored in the database.
In particular, our results suggest that one would not expect to have
polynomial-time algorithms for finding the exact picture-matching and
computing the maximal similarity between a query picture and a database
picture, unless P = NP.