Before processing SEM data, the data acquisition system has to be calibrated. This 
calibration can be performed, following ELGHAZALI (5), by using a silicon replica 
grid because, among other quantities, the tilt angle cannot be defined without any 
object information. Assuming parallel projection, no tilt about the Y-axis does 
influence the y-parallaxes. There would be just affine distortions of the model with 
respect to X- and Z-direction. 
As the aim of stereoscopically evaluating SEM imagery is to achieve a geometrically 
similar reconstruction of the object, knowledge of absolute coordinates is not 
necessary. 
A statement of BOYDE (3) proved true that practical SEM users (as the BAM) do not 
need the high accuracy provided by rigorous system calibration. Besides, with regard 
to the data of this investigation it was found that rotations about the X- and Z-axis 
were very small. Perspective distortions - i.e. differences between central and 
parallel projection - occurred only with very low magnifications. If the object is 
tilted in such a way that there is no remarkable translation of surface points in the 
micrographs, which means tilting about an axis through a central surface point, 
there will be no influence of perspective distortions on x-parallaxes and little 
influence on x- and y-coordinates, which can easily be taken into account. Other 
distortions which in earlier investigations were found to be of some influence, such 
as spiral or tangential distortion, had no considerable effect on our results. Probably 
the reason for this was the direct digital recording of the data. 
Using SEM imagery, photogrammetric calibration can be rigorous only if the 
calibration procedure is conducted under the operational conditions in which the 
results will be applied. A 'state of statistical control' (5) should be achieved. This 
means the SEM has to perform consistently over the period of time for which 
calibration results are to be applied. It is very important not to re-focus the 
electronical system during that period. 
After the determination of the orientation elements the images are resampled. This 
results in a y-parallax-free model, so that automatic correlation along epipolar lines 
can be carried out. 
STEREO PROCESSING 
'Conventional' plotting of stereoscopic imagery in the DSS is optional. Contour lines 
may be drawn on the color display monitor by means of the rubberband method, but 
the monitor resolution (512x512 pixels) turned out to be unsufficient for precise 
processing. In addition to this, fracture topography is too coarse to evaluate DEM 
data efficiently by just interactive operation. For that reason this way of processing 
the SEM data was not taken into consideration. 
Nevertheless, it is necessary that image matching operations have to be partly 
interactive for verification of the results and intervention in critical situations by 
the operator. Therefore stereoscopic viewing is provided for both the image pair and 
the obtained DEM data, which makes interactive editing of raw DEM data possible. 
Superposition is a prior condition for the revision of these data. The integration of 
correcting and editing features for revision purposes took place, respectively is 
under development. 
134