Computational complexity of Similarity Retrieval in Iconic Image Database

Abstract

Iconic indexing is an important tool for image database. For similarity retrieval of iconic image database, there are 3types of similarity. Only one of them is isometric, hence transitive. We present polynomial time algorithms to find maximum similar subpicture of two given pictures when the similarity is isometric. For nonisometric types, we show that the problems are NP-complete, even when all symbols are distinct, by reducing the 3CNF satisfiability problem to them.