Page 291 - Digital Analysis of Remotely Sensed Imagery

Spectral Image Analysis 253

to classification. An effective means of eliminating shadow is via spe-
cial image processing such as band ratioing.
Texture refers to the spatial frequency of variation in image tone or
pixel value. This image element is measured on the basis of a group
of pixels. Use of texture in image analysis is problematic owing to its
scale dependency and the difficulty of its precise quantification. This
topic is so complex that it will be covered in depth in Chap. 10.
Pattern refers to the regular and predictable spatial arrangement
of the same object. It differs from texture in that it never occurs at the
pixel level. Instead, pattern implies the repetitive occurrence of the
same artificial objects along a certain direction, such as buildings and
land plots. Thus, it is commonly associated with residential pattern
and land use pattern. Pattern can be a critical element in identifying
an object or an activity if it is manifested with a unique pattern. How-
ever, in the digital environment pattern is of little use in image analy-
sis at the pixel level. It can be made very useful if the spatial relation-
ship of objects is exploited as with intelligent image classification or
pattern recognition. This issue is so complex that it will not be cov-
ered in this book.
Also called location, contexture, and convergence of evidence,
association refers to the inherent linkage between one object and
another in a neighborhood. At present, association is not routinely
used in image classification. Its use to improve image classification
accuracy forms a frontier in digital image analysis, such as rule-based
or knowledge-based classification in which the relationship is explic-
itly coded and multiple statements may be combined logically to
infer the identity of pixels. Use of contexture in image classification
will be covered in Chap. 10.
In visual interpretation it is impossible to rank these elements in
terms of their significance as the value of each element depends on
the area under study and the object of image interpretation. How-
ever, in the digital environment, the most useful element is tone or its
equivalent. As a matter of fact, it is the only image element used in all
per-pixel classifications in the spectral domain. The decision-making
evidence is based solely on the pixel’s DN and its relationship with
that of others. By comparison, use of the other six image elements is
problematic and much more challenging owing to the difficulty of
their definition and representation. How to incorporate some of these
mathematically ambiguous elements into image classification to
enhance the accuracy of image classification is a challenge at present.
This issue will be addressed separately in chapters to follow.

7.1.3 Data versus Information
In digital image analysis, data and information are by no means syn-
onymous with each other. On the contrary, they have quite different
connotations. Data refer to all the remotely sensed images that the