next up previous
Next: Introduction Up: Fast k nearest neighbour Previous: Fast k nearest neighbour


A simplified k nearest neighbour (knn) search for the R-tree [6] family is proposed in this paper. This method is modified from the technique developed by Roussopoulos et al. [10]. The main approach aims to eliminate redundant searches when the data is highly correlated. We also describe how MINMAXDIST calculations can be avoided using MINDIST as the only distance metric which gives a significant speed up. Our method is compared with Roussopoulos et al.'s knn search on Hilbert R-trees [7] in different dimensions, and shows that an improvement can be achieved on clustered image databases which have large numbers of data objects very close to each other. However, our method only achieved a marginally better performance of pages accessed on randomly distributed databases and random queries far from clustered objects, but has less computation intensity.

Joseph Kuan
Wed Jun 3 13:57:27 BST 1998