A Geometric Framework for Modelling Similarity Search.pdfVIP

A Geometric Framework for Modelling Similarity Search∗ Vladimir Pestov School of Mathematical and Computing Sciences, Computer Sciences Laboratory, RSISE, Victoria University of Wellington, Australian National University, P.O. Box 600, Wellington, New Zealand † Canberra, ACT 0200, Australia ‡ 9 vova@mcs.vuw.ac.nz 9 http://www.vuw.ac.nz/∼vova 9 1 n u Abstract ten highly ingenious, mathematical techniques. This could J We suggest a geometric framework for modelling similarity well be the case with the technique of metric transform as 1 search in large and multidimensional data spaces of gen- applied to histogram indexing for image search by colour 2 eral nature, formed by the concept of the similarity work- content. Even here the gap separating theory and practice ] load, which is a probability metric space Ω (query domain) of database research (discussed in [14] in a highly colour- R with a distinguished finite subspace X (dataset), together ful manner) has to be bridged yet. Nevertheless, as the size I with an assembly of concepts, techniques, and results from of datasets grows exponentially with time, attempts to un- . s metric geometry. As some of the latter are being currently derstand the underlying, very complex, geometry of sim- c reinvented by the database community, it seems desirable to ilarity search through joint efforts of mathematicians and [ try and bridge the gap between database research and the comput