lue
cify
mating
ing
ng
his
effectively with & Gaussian transformation of user specified width
(standard deviation). There will be correspondingly less emphasis in the
central brightness zone. A Gaussian enhancement can be particularly valuable
in dealing with a biased non-symmetric input histogram, such as a log
normal distribution. Figure 4 illustrates a portion of a Landsat scene of
Chile and Bolivia unstretched (a), after a linear 2 percent saturation
enhancement (b), after a Gaussian transformation (c) and after a uniform
distribution contrast enhancement (d).
(b) Filtering is another common enhancement procedure. Filtering can be
thought of as any process which differentially
tending to emphasize desireable features while
able ones (compare section 3.2).
The enhancement of edges or lines can be a valuable
an image particularly when the contrast is moderate.
x
Figure lh: Section of Landsat-scene of Chile and
modifies image content,
suppressing less desire-
tool for crispening
=
Bolivia border (a) un-
stretched; (b) with linear stretch, 4% saturation;(c) Gaussian standard
deviation 2.7; (d) uniform distribution.
3.4 Pattern Recognition
One may find in the literature euphoristic statements like the following:
"It is possible to produce working systems for most pattern recognition
problems" (Aleksander, 1978). However, the computer recognition of terrain
patterns in single images for automation of geo-science photo-interpreta-
tion presently hardly exists. Although pattern recognition is an extensive
field of research its potential in phto-interpretation is largely un-
explored.
Pattern recognition is a tool for automation
for image pre-processing. It may, however, fulfi
of image interpretation, not
ll à support function.
Generally, the line between pre-processing and analysis cannot always be
clearly drawn. Density slicing and texture class
ification are some of the
few pattern recognition techniques of use or under investigation for image
interpretation. Montoto (1977) reported of an investigation to detect
linear features in Landsat images to be used for
Dune and glacier crevasses pattern have been ana
coherent laser light (Verstappen, 1977).
drainage interpretation.
lysed by optical filtering of
