A new representation and similarity retrieval algorithm for spatial databases

Content-based retrieval of images is an important task in applications of image databases. A major class of users' requests requires retrieving the images in the database that are spatially similar to the query image. Most systems of previous approaches provided search capability making use of a data structure to represent the features of relationships among the objects in a picture, such as 2D-String and 2D-PIR. The 2D-String described the directional relationships by a two-dimensional string. The 2D-PIR used a three-dimensional notation to represent the topological and directional relationships. In this paper, we propose a new representation that integrates not only the direction relationships but also the distance relationships. The measuring function and the similarity retrieval algorithm based on the new representation is designed. We also show that the new similarity retrieval algorithm will have a better precision than the method of 2D-PIR by examples.