Efficient image retrieval based on minimal spatial relationships

Abstract

Visual interfaces are known to be effective for retrieving images from databases based on spatial relationships between objects in the image. For efficient image indexing and visual querying, the images are represented using 2D strings, which are derived from symbolic projections of image objects. However, with this approach, it is sometimes difficult to describe the spatial relationships between objects in an image exactly. That is, ambiguities may arise in the representation of the image that inherently captures the 2D projection of the 3D real world, which leads to uncertainty during the retrieval of images. In order to remove these ambiguities, images can be referred to using the spatial location algebra reflecting their spatial relationships in the 3D space. In this paper, we present a unified representation of spatial objects for both topological and directional relationships based on Allen's temporal interval algebra. We also describe a set of reduction rules, which minimizes those relationships. Overall, this scheme can easily be integrated into any multimedia database system using a simple inference engine to provide better precision and flexibility in image retrieval.