Symposium on Remote Sensing for Resources Development and Environmental Management / Enschede /August 1986
Image optimization versus classification - An application oriented
comparison of different methods by use of Thematic Mapper data
Hermann Kaufmann & Berthold Pfeiffer
Institut für Photogrammetrie und Topographie, Universität Karlsruhe, FR Germany
ABSTRACT: Image optimization and classification procedures are comparatively analyzed and discussed relative
to their possibilities and limitations for distinct applications. To meet the objectives LANDSAT Thematic
Mapper, data of two test-sites in different climatic areas (Saudi-Arabia and Southwest-Germany) have been
choosen.
For optimized presentation of TM-imagery, a basic concept was developed through which structural as well as
spectral image contents can be enhanced within one product. By choice of bands and processing parameters the
concept can be modified and fitted to various applications. Supervised and unsupervised classification methods
are applied to separate surface phenomena by use of original-, enhanced-, and preprocessed data. Comparative
evaluations have shown, that both methods give best results using not more than three information bands for
one evaluation step. By qualitative and quantitative analysis it could be demonstrated, that classifying with
original, enhanced or preprocessed components leads to the same results. For geologic applications in arid
areas a standard band combination could be defined, which offers optimized conditions for rock discrimination
and possibilities for separating diagnostic features. Concerning landuse applications, the various spectral
behaviours of different surface phenomena, cannot be meaningfully represented within one product. The lack of
structural information on classification results however could be removed by special merging techniques under
use of image optimization products.
Keywords: Image optimization, classification, Thematic Mapper, comparison of methods.
1 INTRODUCTION
In the past years a great variety and amount of remo
tely sensed data has been offered from operational
sensors. Forthcoming systems, characterized by ad
vanced technologies with possibilities of multiband
and stereorecordings are already launched. For data
evaluation, different methods for enhancement and
display are applied.
This paper compares the advantages and disadvan
tages of image optimization and classification me
thods for several tasks with regard to distinct
climates. Besides this optimal, transferable results
for operational processing and meaningful combina
tions of both methods can be demonstrated.
To meet the objectives, two test-sites of different
climatic character were selected (Fig. 1a, b). One
is located in the Rhine-graben area and shows the
city of Karlsruhe in its center. The second area is
situated in SW-Saudi Arabia at the border to S.-
Jemen and characterized by manifold lithological
and structural varieties.
2 STATE OF THE ART
2.1 Image Optimization
Image optimization or enhancement is one method app
lied to digitally recorded od digitized data. By
employing various techniques, image contents already
present in the raw data are made visible for the use
of interpreters. In general, enhancement algorithms
can be divided into two major groups. One group is
especially used to enhance or suppress structural
information. Such filtering techniques offer a wide
range of applications e.g. for structural geology.
The second group is best suited to enhance spectral
image contents of multiband recordings. Ratio- and
principle component transformations are applied to
two ore more bands of multispectral information of
the same target area, to emphasize spectral diffe
rences, whereby structural (albedo-information) is
simultaniously suppressed.
Common color composites represent to a certain
degree a combination of enhanced structural and
spectral information based on contrast stretch of
each single band used for coding. However, if high-
pass filtering is applied and added to the data,
spectral variations (saturation) decrease. On the
other hand, color composites calculated by use of
Pc's or ratioproducts show a bad signal to noise ratio,
which makes delineation of textural patterns, or
subtle structural features, inpracticable.
Considering that image products calculated by
enhancement techniques are used to support geo
scientists in planning, execution and completion of
field work, requirements for basic concepts became
obvious. Such a concept has to include the choice of
bands and algorithms used. It is a matter of fact,
that not all information given by a seven band sen
sor (e.g. TM) can be presented in one single product.
But contrary to vegetated areas, where more than
a three band combination is needed to utilize the
full spectrum of applications, experiences with
many arid areas have shown, that a combination of
TM-bands 1,4,7, covers most requirements for rock
discrimination. This additionally includes the
possibility to map hydrothermally altered areas
through the presence of diagnostic absorption
bands for Fe^+ and clay minerals (Kahle et a 1,1982).
The'image enhancement now follows a concept which
was developed to satisfy the needs of geologists
for lithological and structural data evaluation.
In particular
- data should be processed in a way, which can be
understood and interpreted by different user
groups.
- both structural and spectral (lithological) in
formation should be emphasized and well diffe
rentiated.
- only those algorithms should be used, which do
not reduce the signal to noise ratio.