In collaboration with Prof. V.S. Subrahmanian (University of Maryland at College Park), a formal framework for integrating heterogeneous similarity measures is being developed. A relational similarity algebra has been introduced, which provides an abstract view of the search engines, together with operators to enrich and combine the engines' output. It allows the specification of queries involving different search engines, based on the similarity between certain database object features and the query objects (which can be, among other things, text, images, video/audio signals). Such queries are optimized through formal techniques based on algebraic equivalences. For this purpose, cost models are being developed.