Image Processing Reference
In-Depth Information
Fig. 10.1
Multi-scale object representation of an image scene. Note that the reporting level does
not necessarily have to be the highest aggregated level
of urban remote sensing there are several applications
demonstrating that object-based approaches are superior
to per-pixel analysis especially when urban land cover
classes, as airport, roads, etc., are to be separated (e.g.
Darwish et al.
2003
; Hofmann
2001
). In comparison to
most natural environments the built environment is
characterized by sharp, discrete boundaries and a high-
frequency change of different surfaces with similar reflec-
tance properties. Anthropogenic features can be reproduced
rather unambiguously in an object-based environment
through an iterative segmentation process. Shadowed areas
for example can be characterized by specific relationships
to neighboring objects and the shape of an object may
help in discerning between a roof and a road of similar
reflectance. In summary, object-based image analysis
combines GIS functionality and remote sensing techniques
by working with polygonal, homogeneous clusters instead of single pixels.
image segmen-
tation suits
the urban
environments due
to the nature of its
anthropogenic
features that are
characterized by
sharp, discrete
boundaries and a
high-frequency
change of different
surfaces with
similar reflectance
properties
10.2.3
Principles and Realization
A crucial step for object-based image analysis is image segmentation. Image segmen-
tation aims at delineating spectrally homogenous regions within an image to generate
