PCA-Tree NNS with two approximation methods and annulus bound method

Abstract

By the successive use of principal component analysis (PCA), database is partitioned into clusters in the preprocessing step of PCA-Tree nearest neighbor search algorithm [1]. In the search step, the algorithm first chooses a leaf node, which is likely to include the nearest neighbor point. Other leaf nodes which are also likely to include the nearest neighbor point are searched by the back tracking approach. The search performance is significantly improved by sorting the data on a leaf node to leaf node basis and updating the threshold value by the minimum distance found so far. The threshold is updated by the e-approximate nearest neighbor approach together with a fixed threshold approach. A further improved performance is achieved by the additional use of the annulus bound approach.