You are using an outdated browser that does not fully support the intranda viewer.
As a result, some pages may not be displayed correctly.

We recommend you use one of the following browsers:

Full text

Fusion of sensor data, knowledge sources and algorithms for extraction and classification of topographic objects
Baltsavias, Emmanuel P.

International Archives of Photogrammetry and Remote Sensing, Vol. 32, Part 7-4-3 W6, Valladolid, Spain, 3-4 June, 1999
IRS-1C PAN). It has been shown that even the fusion of
spatially very different datasets can result in increased
interpretability; an operational example is the use of space
photography and pushbroom scanners, e.g. Russian imagery &
The challenges faced in the resolution merge are often the large
difference in spatial resolution. Combining images of resolution
differences of more than 1:10 causes a number of problems.
Starting with the difficulty in identifying control points in the
data pair, the images are difficult to be co-registered. In this
case, points or features have to be selected with care, due to the
additional large difference in viewing geometry of the sensors
involved. In the case of large spatial or spectral differences in
images to be co-registered, the identification of features (lines
or areas) leads to more accurate results than the measurement of
points only. For this reason, the WEUSC has implemented a
tool that allows the identification of similar features in two data
sources by on-screen digitizing. Those features are used to
calculate a transformation model to co-register the data. This
approach leads to better results in terms of geometric accuracy
than the application of polynomials based on tie point
Depending on the further use of the fused product, the
resolution difference of 1:10, as it is the case for a combination
of SPOT XS (20 m) and Russian imagery (2 m), still results in a
suitable product. For visual purposes, the resulting fused image
provides valuable information that the human interpreter can
exploit considering eventually occurring errors due to
In spectral terms, the fusion of panchromatic and multispectral
imagery does not invoke particular problems, due to the similar
nature of the images involved. This matter becomes more
critical when fusing VIR and SAR. In a case study conducted in
the area of Madrid, Spain, the aim of introducing spectral
information in high resolution imagery has been achieved. The
fused product has been obtained using a linear combination of
the multispectral SPOT (20 m) image, resampled to the pixel
resolution of the scanned Russian photograph (2 m), and
introducing weighting factors to balance the contribution of
each sensor to the benefit of the interpretability of the result.
Minor problems were faced regarding the shadows in the
Russian photography, due to the appearances of high buildings
in the city centre of Madrid. Nevertheless, the fused product
immediately discloses the different appearance of the new part
of the town (different texture and more red) in contrast to the
old city centre (less red). The difference in the spectral response
lies in the fact that the old city contains buildings with different
roof shapes and materials. The user still has the possibility to
assign different colour bands to the RGB colour composite,
which leads to different representations of the same data. This
becomes relevant when the results are presented to decision
makers, who are generally not familiar with false colour
composites. Therefore, the closer the images get to ’natural’
appearance, the better.
Another effect to be mentioned is the relief effects, which
played an important role in another work performed in the area
of Goma, Congo. The presence of relief introduces geometric
distortions in the imagery which have to be taken into account
in the registration process. Depending on the sensor type and
the type of relief, the distortion varies and can reach
unacceptable ranges so that the use of a digital elevation model
(DEM) in the rectification process becomes vital. In cases of
moderate terrain height variations, the selection of small subsets
using local geometric models for co-registration avoids strong
distortions occurring with global models for the whole scene. In
principle, a co-registration accuracy of < 1 pixel is desirable.
The town of Goma, Congo (population 250,000) is situated on
the northern shore of Lake Kivu, adjacent to the border between
Congo and Rwanda. It became a key element of the refugee
crisis during November 1996 (Pohl et al., 1997). The
appearance of the refugee camps in one of the SPOT images is
marked by their clear-cut boundaries with the surrounding
terrain and by the presence of internal roads. Mugunga, the
refugee camp discussed in this paper, also shows an area which
appears to lack any internal infrastructure, suggesting it is an
overspill area formed after the original capacity of the camp has
been exceeded. Fusion of the SPOT PAN and XS data via the
IHS transformation facilitated the location and description of
the camp. It preserved the spectral contrast with respect to
surrounding terrain, showing the details of the internal layout
seen in the panchromatic data. The water ponds inside the camp
were identifiable in the fused image by their distinct blue
The third example of successfully used resolution merges at the
WEUSC is the city of Mostar, former Yugoslavia. Again the
challenge in a SPOT XS and Russian data merge has been the
large difference in spatial resolution for the identification of
GCPs. The result immediately puts the objects into their correct
context due the presence of high resolution and colour at the
same time. Adding the third dimension to this fused image in
terms of a DEM, a 3D perspective can be created which helps
the human interpreter to understand the environment of this site.
Three-dimensional representations can be used to support the
interpretation itself and to illustrate the results to an end-user
(Prisco et al., 1997).
3.2. Multi-sensor (VIR/SAR) Fusion
Due to the disparate nature of optical and microwave sensors,
their imagery depicts different characteristics of the features
observed. In combining these images, the user obtains a more
complete view of the object of interest, which might lead to
improved understanding of the environment being looked at.
SAR data alone is normally very difficult to be visually
interpreted, since it does not correspond to the human visual
perception as do optical images. However, in combination with
VIR images, SAR data can reveal very valuable information to
the human interpreter. In areas of saturation of optical images,
SAR images can visualize structural components. Typical
comer reflectors allow certain conclusions in optical data as
well (e.g. the detection of power lines, high man-made
structures, fences, etc.).