Image Processing Reference
In-Depth Information
instruments are not fully optimized in their spectral, radiometric, and spatial
resolutions to handle urban phenomena. High separability is achieved through
proper selection of resolutions according to the level of detail targeted in the
information extraction process in a particular study. Spatial properties of fea-
tures may lead to the use of techniques that also incorporate their spectral
properties. Features of interest may be represented as part of a single pixel in
a coarse spatial resolution image, thus resulting in spectral mixing of multiple
land cover materials in a single image pixel and the use of spectral unmixing
models. Similarly, features or objects of interest may be represented as mul-
tiple pixels in high spatial resolution imagery. Thus, the improper selection of
spectral and spatial resolutions can hinder the ability to extract information
from imagery at the necessary level of detail.
Although there is an apparent need for careful selection of resolution char-
acteristics according to the nature and goals of urban studies, there is no unan-
imous agreement on the sole importance of a specific resolution characteristic
for selection of suitable imagery. Some researchers have observed that spatial
resolution has generally been the single most important technical characteristic
for urban remote sensing applications (Welch
1982
). However, Jensen and
Cowen (
1999
) discussed a rating system for determining the usefulness of
imagery that emphasized the importance of other technical characteristics and
argued that spatial resolution alone cannot determine the usefulness of imagery.
While higher resolution generally means greater resolving ability in the spa-
tial, radiometric, spectral, or temporal domains, requirements such as storage,
processing time, and data exchange should be taken into account, especially
if these are the limiting factors.
Selection of the most appropriate or optimal image resolution should also
be based on the image processing and interpretation approach utilized. If
visual interpretation and manual mapping techniques are employed, then the
highest spatial resolution imagery available should generally be chosen.
Extraneous detail is normally filtered out or ignored through the human cog-
nitive element of visual interpretation. Of course, areal coverage and cost
considerations may need to be traded-off with high spatial resolution, when
making such a choice. If semi-automated, per-pixel image classification
approaches are to be implemented, then moderate resolution imagery (e.g.,
SPOT HRV or Landsat TM) may provide superior land use and land cover
mapping results for urban and urbanizing areas, relative to high spatial reso-
lution image data (Chen et al
2004
). Conversely, land use and land cover
classification or linear feature (e.g., road) extraction based primarily on tex-
ture measures will tend to yield superior results when derived from high
spatial resolution image data (Chen and Stow
2002
).
