Table 3.

Indexing performance for nearest neighbor queries using weighted Euclidean distances on kd-tree. (k is the number of retrieved nearest neighbors, NT is the average number of node tests, $\mathit{\text{FNT}}=\frac{\mathit{\text{NT}}}{\mathit{\text{DBSize}}}$ is NT expressed as a fraction of database size, DA is the average number of disk accesses, $\mathit{\text{FDA}}=\frac{\mathit{\text{DA}}}{\mathit{\text{DBSize}}}$ is DA expressed as a fraction of database size. )

DB	k = 10				k = 20
Size	NT	FNT	DA	FDA	NT	FNT	DA	FDA
434	257.7	0.59	62.3	0.14	318.4	0.73	89.8	0.21
902	408.5	0.45	78.7	0.09	498.6	0.55	114.1	0.13
1654	600.1	0.36	113.6	0.07	719.2	0.44	160.2	0.10
2812	833.2	0.30	146.0	0.05	996.8	0.36	204.5	0.07