Image Processing Reference

In-Depth Information

clustering patterns (Residential 1, 2, 4 and 5). As expected, given the compact

architecture of terraced (row) housing in the UK, the most clustered neighborhoods

are the linear patterns of Residential-1 and Residential-4. However, although their

nearest-neighbor values may be very similar, it is plainly apparent that Residential-1

exhibits a far denser concentration of postal points. The same situation applies between

Residential-2 (an inner-city local government-owned estate) and Residential-5 (a

more affluent 1980s peripheral estate). Conversely, Residential-3 and Residential-4

have very similar densities yet somewhat dissimilar nearest-neighbor values. The

remaining pattern, Residential-6, is a highly affluent low-density area of Bristol with

large dwellings, and is the only one demonstrating a uniform tendency. Overall, what

is clear is that if nearest-neighbor and postal point densities are taken together they

are valuable measures for identifying and characterizing different residential types.

The same can also be applied to commercial postal points. Again, a strong clustering

pattern is indicative of linear developments in Commercial-1, but the city center

(Commercial-2) and peripheral estates (Commercial-4) exhibit explicit signs of

uniformity. As with the residential values if both nearest-neighbor and density

values are taken in combination then unequivocal differences can be revealed. Both

Commercial-1 and Commercial-3 (commercial development within residential

areas), and Commercial-2 and Commercial-4 have similar nearest-neighbor values

but contrasting densities.

The conventional nearest-neighbor statistic is effective for measuring clustering

patterns but it lacks the capacity to identify spatial arrangement. A variant of two-

dimensional nearest-neighbor analysis is the linear readjustment (
LR
) devised by

Pinder and Witherick (
1975
). Instead of measuring all observed distances between

neighboring points,
D
OBS
is determined from a linear sequence (
L
) of consecutive

points (
LN
) in all directions (essentially linear subsets), while,

L

=

(8.4)

LD

0.5

RAN

N

−

1

What constitutes a linear arrangement is of course dependent on scale. Any group-

ing of buildings regardless of configuration can be considered linear if the scale is

fine enough. As a consequence, a minimum of 20 buildings were identified manu-

ally but considered representative of a linear formation of both residential and com-

mercial postal points. Although at present this is a subjective process, the intention

is to automate future selections and calculate indices that are more representative

of each sample type.

Values for both
R
and
LR
are documented in Table
8.2
. On the whole, they are

very similar for both residential and commercial postal point distributions.

However,
LR
values are usually lower for linear patterns of points (Residential-1

through 5, and Commercial-1, 2 and 4) and higher for inherently random or uni-

form distributions (Residential-6 and Commercial-3). Another difficulty with the

linear nearest neighbor is the low numbers of points that are used for calculations.

This tends to reduce the significance of
LR
to levels of randomness demonstrated

by Residential-3 and Commercial-2 through 4.