Retrieving similar pictures from iconic databases using G-tree

G-tree, combining the features of both grids and B-trees, is a disk-based spatial data structure for organizing multidimensional data. This paper presents an efficient access method for retrieving similar pictures from iconic databases using G-tree. The proposed method preserves the merits of G-tree, and hence indexing the pictures in the database, processing range queries for similarity retrieval, inserting/deleting a picture into/from the database, and effective memory utilization can be achieved in practical applications. In order to improve the acceptability of similarity retrieval as compared to previous works, we define a similarity function by the usage of query ranges, and each range corresponds to one object in the query. This similarity function reflects both the direction and the distance knowledge among objects in the pictures. Besides, the result of similarity retrieval can be further refined by simply adjusting the query ranges or increasing the threshold value of the predefined similarity function. (~) 1997 Elsevier Science B.V.