A simplified knn search is proposed for the R-tree family which reduces the number of nodes accessed when a query point is close to or within the clustered data points in a database. We specified that with only MINDIST values, the knn search can be achieved effectively, and giving a faster response than with the MINMAXDIST computation. Roussopoulos et al. knn search and our simplied search are compared with Hilbert R-trees indexing an image feature database. Our method shows a consistent improvement with fewer nodes accessed in different numbers of dimensions.
However, our method shows that there is hardly any improvement for synthetic random databases and knn query points far from clustered data points. The reason for this is that it is not possible to forsee the real minimum distance between a query point and a data object within an MBR without exploring down to the leaf nodes.
In future work, our algorithm will be presented in more detail with the psuedocode,
further reduction on distance computation, and a more generalised knn implementation that can offer the
next k nearest matches effectively on the basis of the kth distance value.