Image Processing Reference
In-Depth Information
Fig. 8.2
Maximum likelihood classification results using equal prior probabilities
are attempted. For example, consider the case of Bristol in Fig.
8.1
where only the
northwestern part of the city is shown for clarity. A visual comparison of urban land
use coverage between equal priors (Fig.
8.2
) and unequal priors (Fig.
8.3
) reveals
interesting patterns. More pixels have been classified as detached (single dwelling)
and semi-detached (duplex dwelling) housing under conditions of equal prior prob-
abilities, particularly across the more affluent parts of the west and north east of the
sample area. At the same time, more pixels are classified as terraced (partitioned,
row housing) under unequal prior conditions. In both cases, class estimates from
unequal (stratified and modified) priors are closer to census estimates, and this is
indicative of the flexibility of modified prior probabilities and their ability to incor-
porate spatial information from beyond the spectral domain. Individual census
tracts can further highlight differences in classification between equal and unequal
priors. The census tracts 09DDGH21 and 09DDFA01 demonstrate how lack of
