The following geometrical proximity concepts are discussed: relative
closeness and geographic closeness. Consider a set V =
{v1,v2,...,vn} of distinct points in a two-dimensional
space. The point vj is said to be a relative neighbour of
vi if
dp(vi,vj) ≤ max{dp(vj,vk), dp(vj,vk)} for all vk ∈ V,
where dp denotes the distance in the Lp metric, 1
≤ p ≤ ∞. After dividing the space around the point vi
into eight sectors (regions) of equal size, a closest point to vi
in some region is called an octant (region, or
geographic) neighbour of vi. For any Lp metric, a
relative neighbour of vi is always an octant neighbour in some
region at vi. This gives a direct method for computing all relative
neighbours, i.e. for establishing the relative neighbourhood
graph of V. For every point vi of V, first search for the
octant neighbours of vi in each region, and then for each octant
neighbour vj found check whether the point vj is also a relative
neighbour of vi. In the Lp metric, 1 < p < ∞, the total
number of octant neighbours is shown to be Θ(n) for any set of
n points; hence, even a straightforward implementation of the above
method runs in Θ(n2) time. In the L1 and L∞
metrics the method can be refined to a Θ(n log n + m)
algorithm, where m is the number of relative neighbours in the
output, n − 1 ≤ m ≤ n (n − 1). The L1 (L∞)
algorithm is optimal within a constant factor.
@article{Kat1988J,
author = {Jyrki Katajainen},
title = {The region approach for computing relative neighbourhood graphs in
the $L_p$ metric},
journaltitle = {Computing},
volume = {40},
number = {2},
year = {1988},
pages = {147--161},
}