e
nt
se
1e
re
e-
an
18
us
be
a
tie
be
he
en
ad
he
ht
>
be
on
ge
cal
the
ont
on,
atmosphere etc.). For this purpose a software package
has been developed which removes brightness and
contrast differences and converts the data of several
scenes into one gray scale system (Kahler, 1989).
The procedure makes again use of the multiple infor-
mation within the overlapping areas of adjacent
scenes. The program starts with the interactive defi-
nition of the overlapping areas to be used for mosa-
icking. Now median integral histograms are calcu-
lated for each overlapping region. From this data base
correction tables for each scene and each spectral band
are derived in an iterative process. Once this is
achieved the corrected single images have to be
merged to one image file per spectral band. All-
together the sophisticated software package (described
comprehensively by Kahler, 1989) offers great flexibi-
lity and yields excellent results.
The combination of multisensor data is a task of great
practical importance since high resolution panchro-
matic data from the SPOT satellite became available in
1986. It is evident that by merging multispectral data
with panchromatic data of high geometrical resolu-
tion excellent results can be achieved.
A well known application is the combination of the
TM system with its seven spectral bands and SPOT-
HRV data with a spatial resolution of 10 m in the
panchromatic mode leading to maps at a scale of
1:50,000 (e.g. Albertz et al., 1990).
In order to preserve both advantages by combining the
data comprehensive investigations using the IHS
color transformation (Tauch et al., 1988, Albertz et al.,
1990) have been undertaken. The principle idea is to
transform RGB colors into the IHS color domain. By
substitution of the intensity component through high
resolution panchromatic SPOT data and subsequent
retransformation into the RGB color system an
enhanced image is obtained. Because of different
radiometric histograms it is often useful to carry out a
radiometrical adjustment of the SPOT data onto the
TM intensity component and to calculate the new
intensity as a weighted average of both data.
Although very successful, the merging of multisensor
data sometimes contains problems due to the different
spectral characteristics, the geometrical resolution of
the sensors and due to differences in the dates of
acquisition.
Different Spectral Resolution in Multisensor Data:
The high spectral resolution of the Thematic Mapper
data makes it possible to distinguish various landuse
classes which can not be differentiated in panchro-
matic SPOT data (e.g. deciduous and coniferous
forest). Forest information, extracted by multispectral
classification of TM data, undergoes a special contrast
enhancement and is preserved after adding to the
initial data again.
Different Geometrical Resolution of Multisensor Data:
Due to the different geometrical resolution of multi-
sensor data the determination of ground control
points is more difficult and has to be carried out care-
311
fully. As known from experience during the produc-
tion of several Satellite Image Maps the relative rectifi-
cation of TM data onto SPOT data can be performed
with a relative accuracy of + 1.0-1.5 pixel (+ 10-15 m)
and an absolute accuracy of + 2.0-3.0 pixel (+ 20-30 m)
what is less than one TM pixel.
Another effect caused by the differences in geometrical
resolution is the appearence of color lines especially
along edges with high contrasts. In spite of the high
accuracy achieved in geometrical rectification such
color lines can not be avoided, because of the differen-
ces in geometrical resolution. But in most cases they
do not appear as a degradation effect.
Seasonal Effects: After merging data sets of different
seasons many unnatural colors may come into being.
These failures have effects on large areas and can be
removed only by a lot of additional mostly interactive
operations.
Land Use Changes: Often it can not be avoided to use
data sets which are acquired in different years. If in the
meantime the landuse of particular fields has changed
this leads to unnatural colors of these areas after the
combination of both data sets. These are local and
irregular failures, but fortunately they occur just
sporadically if the data used differ only a few years.
Tidal changes: In coastal areas it can happen, that the
data to be mosaicked or merged are acquired under
significantly different tidal situations. For mosaicking
there might be chances to select the dividing line at
places where the tidal influences are small. Otherwise
there is no method known to avoid some irritating
effects in the mosaic.
Usually some postprocessing techniques are applied
after mosaicking and merging multisensor data in
order to optimize the data for the particular purpose.
Depending on the structure of the scene and the
quality of the data as well as the particular map scale
different types of edge enhancing filters can be applied
in order to improve the visibility of details in the final
product. Postprocessing comprises the interactive se-
lection of the color rendition of the map to be printed.
So far all operations were carried out in the digital data
files for the three colors red, green and blue, as it is
common in digital image processing. However, in
order to achieve a high quality image map, four colors
must by applied as it is usually done in offset printing.
This requires the calculation of a black color data set
out of the three other data files. It will be obvious from
the section below, that this operation has to be carried
out before the integration of graphical elements.
The result of all the previous operations are the digital
data files for each color plate. In order to generate the
originals for offset printing these data sets must be
printed on films by means of a large format high re-
solution raster plotter, e.g. a SCITEX or HELL system.
Through this process the gray values of the image files
are converted to screens, considering also the screen-
ing angles for each color.
