On Fast Non-metric Similarity Search by Metric Access Methods

The retrieval of objects from a multimedia database employs a measure which defines a similarity score for every pair of objects. The measure should effectively follow the nature of similarity, hence, it should not be limited by the triangular inequality, regarded as a restriction in similarity modeling. On the other hand, the retrieval should be as efficient (or fast) as possible. The measure is thus often restricted to a metric, because then the search can be handled by metric access methods (MAMs). In this paper we propose a general method of non-metric search by MAMs. We show the triangular inequality can be enforced for any semimetric (reflexive, non-negative and symmetric measure), resulting in a metric that preserves the original similarity orderings (retrieval effectiveness). We propose the TriGen algorithm for turning any black-box semimetric into (approximated) metric, just by use of distance distribution in a fraction of the database. The algorithm finds such a metric for which the retrieval efficiency is maximized, considering any MAM.