et al. 1989; Wall et al. 1991), or from a second imaging angle, as available in radar
stereo data or interferometry. Since radar stereo or interferometry data are not
readily available in most cases, DEM information becomes an essential element
for the determination of local distortions in radar imagery.
In general, two methods have been proposed for image rectification of terrain
distortions. One relies mainly upon DEM information and has been used in conjunction with airborne and spaceborne SAR data. The other utilizes the platform
trajectory information with a limited amount of DEM data and has been applied to
correct spaceborne SAR data. The first method depends on the relative registration
of a SAR image to a simulated image generated from the DEM. This technique
employs a ‘rubber-sheeting’ process, after scan- and skew-distortions of the earth’s
surface have been eliminated. A series of common reference points on both the
actual and the simulated image are required in order to estimate the polynomial
coefficients of the warping function. This function is then used to transform the
radar image coordinates into the simulated image coordinates. Once the SAR scene
is co-registered a re-sampling routine may be applied to provide a rectified format.
The accuracy of this format is a function of the density of the selected reference
points, or ground control points (GCPs). This method is often applied for small
SAR sub-scenes.
The second method does not require the generation of a simulated radar image
from a DEM, nor does it rely on a dense grid of GCPs to characterize the image-tomap distortions. Instead, an algorithm is employed which models the inherent
geometric distortions based on the radar ephemeral data, such as platform altitude,
the signal Doppler parameters, and the local terrain elevation. Using this algorithm
an automated registration transformation may be performed. Only very few GCPs
are required to remove residual translational errors between the predicted location
and the actual geodetic location of a target area on a topographic map. In the case of
severe terrain distortion, as in SAR images of mountain areas, DEMs are used in
conjunction with the imaging geometry to generate a transformation ‘map’. This
‘map’ removes the local distortions, or foreshortening, when the slant-range pixels
of an image are spatially mapped into their respective geo-coded pixel location.
Using raw satellite SAR data it may generate an output product with an absolute
location uncertainty of less than 50 m.
14.5 Enhancement of Digital SAR Data
Apart from radiometric and geometric corrections, a number of digital image
processing techniques (Figs. 14.3, 14.4, and 14.5) may be applied to enhance the
image content of a SAR scene. These enhancement techniques are useful, because
they may provide a more suitable image product for subsequent manual interpretation (Trevett 1986). They can also be applied before or after performing digital
classification. Enhancement techniques include filtering for (additional) speckle
reduction, edge enhancement and contrast stretching. In addition the value of a SAR
288
S. Pirasteh et al.
stereo data or interferometry. Since radar stereo or interferometry data are not
readily available in most cases, DEM information becomes an essential element
for the determination of local distortions in radar imagery.
In general, two methods have been proposed for image rectification of terrain
distortions. One relies mainly upon DEM information and has been used in conjunction with airborne and spaceborne SAR data. The other utilizes the platform
trajectory information with a limited amount of DEM data and has been applied to
correct spaceborne SAR data. The first method depends on the relative registration
of a SAR image to a simulated image generated from the DEM. This technique
employs a ‘rubber-sheeting’ process, after scan- and skew-distortions of the earth’s
surface have been eliminated. A series of common reference points on both the
actual and the simulated image are required in order to estimate the polynomial
coefficients of the warping function. This function is then used to transform the
radar image coordinates into the simulated image coordinates. Once the SAR scene
is co-registered a re-sampling routine may be applied to provide a rectified format.
The accuracy of this format is a function of the density of the selected reference
points, or ground control points (GCPs). This method is often applied for small
SAR sub-scenes.
The second method does not require the generation of a simulated radar image
from a DEM, nor does it rely on a dense grid of GCPs to characterize the image-tomap distortions. Instead, an algorithm is employed which models the inherent
geometric distortions based on the radar ephemeral data, such as platform altitude,
the signal Doppler parameters, and the local terrain elevation. Using this algorithm
an automated registration transformation may be performed. Only very few GCPs
are required to remove residual translational errors between the predicted location
and the actual geodetic location of a target area on a topographic map. In the case of
severe terrain distortion, as in SAR images of mountain areas, DEMs are used in
conjunction with the imaging geometry to generate a transformation ‘map’. This
‘map’ removes the local distortions, or foreshortening, when the slant-range pixels
of an image are spatially mapped into their respective geo-coded pixel location.
Using raw satellite SAR data it may generate an output product with an absolute
location uncertainty of less than 50 m.
14.5 Enhancement of Digital SAR Data
Apart from radiometric and geometric corrections, a number of digital image
processing techniques (Figs. 14.3, 14.4, and 14.5) may be applied to enhance the
image content of a SAR scene. These enhancement techniques are useful, because
they may provide a more suitable image product for subsequent manual interpretation (Trevett 1986). They can also be applied before or after performing digital
classification. Enhancement techniques include filtering for (additional) speckle
reduction, edge enhancement and contrast stretching. In addition the value of a SAR
288
S. Pirasteh et al.