(f) Terrain-corrected_ortho_images derive from the raw
images, resampled to the proper terrain geometry. One
will obtain geocoded ortho-rectified radar images from
each of the 3 cycles. An individual surface point may
thus be shown on different images. Due to Magellan’s
quasi-polar orbit, near to the pole a location may have
been imaged tens to hundreds of times.
(g) Shape-from-shading models how a human observer
would interpret brightness and shadows in the radar
image, which roughly appears like an oblique
illumination of the terrain. Unjustified, this process is
assuming uniform backscatter-properties for the whole
surface of the planet and accuracies are much
dependent on the look-direction of the sensor. Still, a
shape-from-shading approach applied to geocoded
images is beneficial to refine details of the digital
terrain elevation beyond what can be extracted from
the geometric disparities in stereo pairs.
(h) An image and data stack may consist of an optimum
digital elevation model and a set of overlapping
geocoded and ortho-rectified images from multiple
cycles. One may organize the material into an image
data base as discussed in Section 3.
5 COPING WITH DISSIMILARITY
5.1 Opposite Side Images
Subsection 4.3 assumed that all the 3 Cycles of Magellan
images can be geocoded using terrain elevations extracted
from stereo or from altimetry. This is only true if the
images from the third non-stereo (,opposite-side‘“) Cycle
can successfully be processed with the elevation data.
5.2 Experiments with Image Simulation
One software tool for this task is simulation. The opposite-
side image can be synthesized using the topographic relief
available from altimetry, stereoscopy or from shape-from-
shading. Figure 4 illustrates the problem of an opposite-
side Magellan pair and relates this to a DEM obtained by
stereo. Simulating the opposite-side image produces an
intermediate product that may be matched with the actual
opposite-side image by conventional methods. So, one can
verify if and where the two images are geometrically
identical. If not, then the digital elevation model needs to
be corrected so that identity between the simulated and
real image is accomplished. This problem domain is the
subject of ongoing work (Kellerer-Pirklbauer, in print).
Figure 4: Opposite-side Magellan images of accentuated
terrain. Look-angles are 43? East and 25? West.
Location at 2°S. Area covered is 30 km x 30 km.
495
6 CONCLUSIONS AND OUTLOOK
We have described geometric processing of Magellan
radar images. The Internet provides access to the images
in the Planetary Data System PDS. A systematic effort
currently converts the full resolution original image strips
into a final image product, the F-MAP. However, those
images still are affected by the errors of the preliminary
ephemerides obtained during the Mission, and they just
corrected using the poor DEM obtained from altimetry.
We have proposed and are actively pursuing the
development of a suite of modules that would represent an
alternative system for reprocessing the raw phase histories.
We work towards a final image data base of ortho-rectified
and geocoded images. The DEM would be produced from
stereo, taking into account also the opposite-side imagery.
We are addressing a number of individual problems such
as the database issues of managing the large 400 Gbyte
images from Venus; the difficulty one faces because of the
excessive layover in topographically accentuated terrain
which occurs because of the very steep look-angles of
Magellan; the matching of opposite-side image using the
topographic elevations obtained from stereo; the use of
shape-from-shading and of image simulation that should
alleviate the concerns resulting from layover and
dissimilarities in opposite-side images. We are also
concerned about computing times and therefore have
embarked on an effort to parallelize key algorithms such
as shape-from-shading, image matching, surface point
gridding, and perspective rendering for quality control
(Goller et al., in print; Goller, in print).
The Magellan mission was terminated in 1994; the
imaging portion of the mission ended already two years
earlier in 1992, after the imaging radar had operated for 3
years. Since the end of the Mission the data analysis has
become a coordinated but modest effort, much reduced
from that which existed during the Mission. It is through
the Planetary Data System and its participating 12
institutions that the data are being processed at the rate
which funding supports. We hope that radargrammetric
processing will evolve into a complete suite of software
elements to produce an accurate DEM and associated
ortho-rectified images. The basic prerequisite is an
improved ephemeris which is outside the scope of the
radargrammetric work. It will be the pivotal element to
make the geometric reprocessing effort worthwhile.
ACKNOWLEDGEMENT
This work has been supported in part by the Austrian
Research Foundation FWF (project S 7001), and by a
grant from the Austrian Academy of Sciences.
REFERENCES
Connor C. (1994) Determing Heights and Shapes of Fault
Scarps and Other Surfaces on Venus Using Magellan
Stereo Radar. Manuscript, Princeton Univ. Dept. of
Geological Sciences.
Ford Pl, G. Pettengill (1992) Venus Topography and
Kilometer-Scale Shapes. J. Geographical Research, Vol.
97, pp.13103-13318.
International Archives of Photogrammetry and Remote Sensing. Vol. XXXI, Part B4. Vienna 1996