It is very useful to be able to search a large database for objects based on their similarity to a given query object. There is a wide array of indexing structures available which have been studied well in an attempt at increasing query efficiency, but most techniques run into significant problems when the objects being indexed are very complex or high in dimensionality. One of the most promising ideas is to map the objects as points in an arbitrary k-d vector space, and then build the indexing structure using these points. These query results can be used as a "filter" to attain the true similarity query set in a space of much lower dimensionality, significantly lowering the computational cost. This thesis provides analysis and experimental data to explore the usefulness of FastMap, one such dimensionality reduction technique, as it is applied to point and spatial access methods as a filter. |