A parallel index supprorting concurrent queries for finding relevant remote sensing images

Abstract

Nearest neighbor (NN) query in multi-dimensional space is one of the key problems for searching relevant remote sensing images from a large gallery. Facing the concurrent queries, we propose a Parallel Compressed Vector Approximation Hashing (PCVAH) index in this paper. The PCVAH keeps the pointers to approximated vectors in a hashing style structure, uses neighboring masks for filtering. The neighboring masks are sets of mask vectors indicating grids close to the query point. And then access the accurate vectors to calculate the final NN results. It handles several concurrent queries in parallel when filtering and access the accurate vectors together. Theoretical analysis and experiments confirm that the PCVAH parallel query method is of high parallel efficiency and time efficiency. And more important it is simple for practical implement in real applications.