Classification of Urban Areas: Inferring Land Use from the Interpretation of Land Cover - Remote Sensing of Urban and Suburban Areas - page 151

Fig. 8.6 Classified IKONOS image of Belfast subset: white areas represent pixels classified as
residential, dark gray areas classified as non-built, and light gray as roads. Residential ( open
squares ) and commercial ( shaded squares ) buildings from the COMPAS™ dataset are represented
in proportion with area attributes. Note the very high density, linear arrangement of residential
buildings at A; lower density, curvi-linear distribution at B; medium density, linear arrangement
at C; and very low density, more uniform configuration at D. Only one commercial pattern is
observable at E; and a school to the west of X
be seldom represented by convenient square point patterns, regardless of the area
attribute. Finally, a quick comparison with Fig. 8.5 reveals many misclassified pix-
els throughout the subset. Most of these are a result of shadows, a problem rarely
confronted in conventional spectral or even spatial classifications.
All building types have been identified visually and are therefore highly subjec-
tive. What is now needed is an automated approach that can more precisely and
objectively implement the comparison of sample nearest-neighbor indices of both
residential and commercial land use with classified imagery. But at the same time
be able to also accommodate these same nearest-neighbor indices within an
image classification methodology. Research is currently examining the feasibility
of an automated image pattern recognition system to facilitate both of these
objectives.