In: Wagner W., Székely, B. (eds.): ISPRS TC VII Symposium - 100 Years ISPRS, Vienna, Austria, July 5-7, 2010, IAPRS, Vol. XXXVIII, Part 7B
453
over-determination in the spatial point intersection.
Stereo matching of a TerraSAR-X image pair is improved by
including geometric constraints. First, one image is pseudo
epipolar registered based on an affine polynomial transformation
using both sensor models. Second, in image matching a starting
location for each pixel is predicted, again using sensor models
and a coarse DSM (SRTM or ASTER model).
The presented approach yields an areal digital surface model.
When subtracting a reference digital terrain model (DTM),
e.g. available from airborne laser scanning, a canopy height
model (CHM) can be extracted (cf. Figure 2 and Eq. (1)). Such
CHMs serve as an important information for the retrieval
of forest parameters. As mentioned before, the canopy
height underestimation can be quantified using laser scanner
ground truth data. Such comparison enables to determine the
underestimation factor r in percent. In regions of forest the
TerraSAR-X DSM is then corrected by multiplication with the
factor 1/(1 — r/100). The forest segmentation presented in the
next section is then used to correct the canopy height bias (see
Figure 1). It should be noted that this problem is not straight
forward, as such underlying image segmentation often is just not
available.
Figure 2: Explanation of relation between digital surface model
(DSM), digital terrain model (DTM) and canopy height model
(CHM).
CHM = DSM - DTM (1)
2.2 Forest Segmentation
The proposed forest segmentation should allow separating
regions of forest from non-forest areas. Recently, first results on
this topic were published in (Breidenbach et al., 2009). They
perform the classification on TerraSAR-X backscatter mean and
standard deviation statistics alone. We extend their method by
including backscatter intensity and texture information, a 3D
canopy height model and interferometric coherence information.
For classification a supervised approach is chosen by selecting
multiple regions together with their ground truth class labels
(forest / non-forest) and training a maximum likelihood
classifier. This classifier is then applied to the whole spatial
extent of given images. The resulting classification is
constructed with a GSD of 5 meters. Next, very small areas are
rejected based on a region labeling approach.
Texture Description. As observed in (Breidenbach et al., 2009)
regions of vegetation are less textured, i.e. more homogenous,
than regions of settlements or agricultural areas. (Haack et al.,
2000) suggest to describe this texture information by a variance
filter. However, our tests showed that such simple parameter is
not working satisfactorily on TerraSAR-X data. Therefore, we
choose the Texture-transform (Tavakoli Targhi et al., 2006) which
is invariant to illumination, computationally simple and easy to
parameterize so that it also performs reasonably on high resolu
tion radar data. This transform can be seen as a spatial frequency
analysis, where the key idea is to investigate the singular values of
matrices formed directly from gray values of local image patches
(the backscatter information in our case). More specifically, the
gray values of a square patch around a pixel are put into a ma
trix of the same size as the original patch. The texture descriptor
is computed as the sum of some singular values of this matrix.
The largest singular value encodes the average brightness of the
patch and is thus not useful as a texture description. However, the
smaller singular values encode high frequency variations charac
teristics of visual texture. Therefore, the singular values of this
matrix are sorted in decreasing order. Then the Texture-transform
at each pixel is defined as the sum of the smallest singular val
ues. For the tests several window sizes and singular values ranges
were chosen, where a window of size 33 x 33 and a range of 20
to 33 smallest singular values performed best.
Canopy Height Model. Obviously, vegetation heights are a use
ful information to segment regions of forest. The canopy height
model is extracted employing the methodology described in Sec
tion 2.1.
InSAR Coherence. For forest segmentation the interferometric
coherence, which is a measure of the interferogram’s quality,
can be of great value since regions of vegetation suffer
from temporal decorrelation (see also the detailed study
on interferometric decorrelation (Zebker and Villasenor,
1992)). The standard coherence estimation is based on a local
complex cross-correlation and is known to over-estimate
the real coherence value. In general, a larger window within
cross-correlation provides a better coherence estimate. At the
time of radar sensors like ERS the standard procedure was
to estimate the coherence over the same window used for
multi looking. As the multi looking sizes become smaller for
TerraSAR-X imagery the coherence was highly over-estimated
resulting in a noisy coherence image. Therefore, a decoupling
of the window size of multi looking and cross-correlation is
introduced. The resulting coherence estimate uses a correlation
window of 10 x 10 pixel and a multi looking window of 2 x 3
pixel (azimuth x range). Regions of very low coherence
correspond mainly to vegetation (forests and agricultural areas).
Thus, such coherence information is used in the classification
process as one feature.
3 TEST DATA
Within the AT-X project the proposed algorithms have been ap
plied to several test sites. For the presented study only a single
test site called “Burgau” is chosen to keep the results clearly ar
ranged. The test site of interest spans an area of 12 x 12 km 2
in Austria. This rural test area covers agricultural as well as for
est areas and shows flat to slightly hilly terrain, the ellipsoidal
heights ranging from 270 to 445 meters above sea level (cf. Fig
ure 3). The forested regions in the area mainly consists of dense
stands of deciduous trees.
Multi-Image DSM Generation. For multi-image DSM
derivation the test data consist of multiple TerraSAR-X
multi-look ground range detected (MGD) Spotlight products
from ascending, respectively descending, orbit. All images were
ordered as single-polarization products (HH) with science orbit
accuracy and were acquired in the period of July and August
2009. Table 1 reports the major parameters of the “Burgau” test
site. It should be noted that the images acquired at steep look
angles (i.e. MGD_ascl and MGD_dscl) have a lower GSD than
all other products.