Efficient Algorithms for Similarity Search in Axis-Aligned Subspaces.

Many applications-such as content-based image retrieval, subspace clustering, and feature selection-may benefit from efficient subspace similarity search. Given a query object, the goal of subspace similarity search is to retrieve the most similar objects from the database, where the similarity distance is defined over an arbitrary subset of dimensions (or features) - that is, an arbitrary axis-aligned projective subspace. Though much effort has been spent on similarity search in fixed subspaces, relatively little attention has been given to the problem of similarity search when the dimensions are specified at query time. In this paper, we propose several new methods for the subspace similarity search problem. Extensive experiments are provided showing very competitive performance relative to state-of-the-art solutions.