Approximate retrieval approaches for incremental similarity searches

Similarity selections of objects in a very large database can be executed by an incremental search on the basis of their distance from a given point. To cope with this problem, indexing support and retrieval strategies, that are able to ensure good performance for different kinds of queries, need to be developed. In this work we propose incremental and approximate retrieval approaches for searching points in a d-dimensional metric space. Four new retrieval algorithms coupled with dynamical disk-based spatial structures are discussed and some experimental results are presented. In particular, two strategies named Chessboard and City Block respectively, implement approximate incremental searches on a grid file data structure and the others, heap queue and virtual tree, apply to hierarchical data structures such us the R-tree.