Page 33 - Digital Analysis of Remotely Sensed Imagery

6 Cha pte r O n e

still expensive to buy the most recent, very high spatial resolution
satellite data. High costs are also related to maintenance personnel. A
system administrator is needed to update the image processing
system periodically and to back up system data and temporary results
regularly.

Limited Accuracy
The second major limitation of digital image analysis is the lower-
than-expected classification accuracy. Classification accuracy varies
with the detail level and the number of ground covers mapped. In
general, it hovers around 60 to 80 percent. A higher accuracy is not so
easy to achieve because the computer is able to take advantage of
only a small portion of the information inherent in the input image,
while a large portion of it is disregarded. Understandably, the
accuracy is rather limited for ground covers whose spectral response
bears a high resemblance to that of other covers.

Complexity
A digital image system is complex in that the user requires special
training before being able to use it with confidence. Skillful operation
of the system requires many hours of training and practice. As the
system becomes increasingly sophisticated, it becomes more difficult
to navigate to a specific function or to make full use of the system’s
capability.
Limited Choices
All image processing systems are tailored for a certain set of routine
applications. In practice it may be necessary to undertake special
analyses different from what these prescribed functions can offer.
Solutions are difficult to find among the functions available in a given
package. Although this situation has improved with the availability
of a special scripting language in some image analysis systems, it is
still not easy to tackle this scripting job if the user does not have a
background in computer programming.

1.3 Components of Image Analysis
The process of image analysis starts from preparation of remotely
sensed data readable in a given system and feeding them into the
computer to generate the final results in either graphic or numeric
form (Fig. 1.2). Additional preliminary steps, such as scanning, may
also be required, depending on the format in which data are stored.
There is no common agreement as to what kind of postclassification
processing should be contained in the process. In this book, three
postclassification processings are considered: accuracy assessment,
change detection, and integration with non-remote sensing data. The
logical sequence of these processing steps is chronologically presented