Abstract

In this project, we introduce and present a new search method for fast nearest-neighbor search in high-dimensional feature space, which is called Comb algorithm . Most similarity search techniques map the data objects into high-dimensional feature space. The similarity search corresponds to a nearest-neighbor search in the feature space. Fagin and Threshold algorithms are two known methods that perform for nearest-neighbor search with one query point. On the other hand, the method we present works on parallel systems that are identical. We provide an alternative solution with several query points searching in parallel identical systems in as many copies as query points are defined. The algorithm is a trade-off between space storage (multiple copies of the multidimensional system), computation resources, and query execution time.

Document Downloads