×

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

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

International Archives of Photogrammetry and Remote Sensing, Vol. 32, Part 7-4-3 W6, Valladolid, Spain, 3-4 June, 1999
126
Due to the frequent cloud cover, visual imagery for the area of
interest is often not available. In addition, the re-visit time of
the individual satellites alone is not sufficient to monitor fast
changes, e.g. as they occur in a harbour such as Rotterdam, the
Netherlands. Using an image of the relatively high resolution
Indian Satellite IRS-1C PAN (~ 6 m), the interpretation of an
oil storage facility containing tanks with floating roofs has been
performed. The cloud-free image depicts the level of fluid
available in the different tanks. In case of clouds, this
information cannot be obtained. Since SAR sensors are active
instruments with a wavelength that can penetrate clouds, they
can acquire images at any time of the day or year. For
comparison, a multi-temporal composite of three different
Radarsat images has been visually analysed. It leads to the
assumption that the backscatter returning to the sensor is
stronger, the lower the fluid level in the tank. The explanation
lies in the increased comer reflector effect with decreasing fluid
level (see Fig. 2). Using more imagery and measurements of the
point targets in the SAR images could lead to the establishment
of a relationship between strength of backscatter and level of oil
storage. The available knowledge of the location of the tanks,
based on one cloud-free optical image, facilitates the approach.
In another case, a SAR image has been used to study the
difference of tank covers. The tanks had been identified from
optical imagery, which was not sufficient to distinguish floating
from conical roofs. The SAR on the other hand contributed this
information due to the differences in backscatter from the
different types of roofs.
Fig. 2. Backscatter from filled (left) and almost empty (right) oil
tank.
The experience at the WEUSC using visual image interpretation
as major exploitation element has proven that image fusion
plays a vital role in the facilitation of feature detection,
recognition and identification. The main criterion for the choice
of images for image fusion is the contribution of complementary
information contained in each individual image. If the
operational process of image fusion succeeds to maintain the
information contribution of the individual images in the fused
product, it is not necessary to evaluate the individual images
alone. The image analyst can understand the feature and its
context much faster in the fused product than looking at the
individual images. This speeds up the exploitation process and
improves the reliability of the obtained interpretation results.
Image fusion does not produce information that is not already
contained in the original images. But it forms a mean to
enhance certain features and their environment in order to draw
the attention of the human interpreter to relevant aspects.
The most time consuming part in pixel-based image fusion is
the image registration, i.e. the identification of GCPs. This is
especially true for spatially very different data or VER/SAR
registration. A special registration tool has been developed to
address this part of the processing chain. This tool decreases the
time and accuracy needed for registration, because it allows not
only GCPs, but also linear or area features, in the establishment
of the geometric model. Furthermore, it introduces a standard in
the processing chain, which is essential in an operational
environment. In parallel, the WEUSC allocates resources to
research and technical development focusing on an automation
of this process. A first prototype exists integrating sensor
models and a priori information provided with the image data.
At the moment the systems can process SPOT, IRS-1, MK-4
and Radarsat imagery.
A major element of the operational implementation of image
fusion with respect to visual image interpretation is the
interactive component. The operator needs to be capable of
empirically tuning individual parameters involved in the fusion
process or in the enhancement of input data as well as the end
product.
4. RESULTS
Operational image exploitation means the achievement of speed
and quality, as well as reliability of results. However, quality
and accuracy should be seen in the context of requirements.
Depending on the application and time constraints resulting
from the operational environment, images will not always be
processed to the highest accuracy level. The approach and
complexity of image processing is defined on the basis of the
needs expressed. This is very important for an optimization of
resources.
A problem that often reduces the speed of processing is the lack
of availability of ancillary data describing the imagery as well as
ground truth. The image analyst uses the fusion approach with
care to fully understand the nature of the fused product in order
to draw proper conclusions.
5. CONCLUSIONS
The experiences gained show very clearly that a major element
of the operational implementation of image fusion with respect
to visual image interpretation is the interactive component. The
operator needs to be capable of empirically tuning individual
parameters involved in the fusion process or in the enhancement
of input data as well as the end product. The fine tuning of the
image enhancement parameters, i.e. histogram value
distribution, filter, assignment of colours etc., influences the
success of the fusion itself. A small interactive window,
containing a representative subset of the image to be processed,
has proven to be an excellent aid to support the determination of
the ’right’ values for the parameters. This window shows in real
time the effect of changes in the parameters on the input data or
the fused product, depending on the selection of the data to be