distortions, correlation procedures may react by an
enlargement of the processed image area. In many
cases this has to be proven as successful in that failures
might be avoided.
As these precautions take place individually for each
object point the considered image areas of adjacent
points may overlap thus producing certain interrelations
between the computed heights. To avoid these depen-
dencies the size of adjacent image areas has to be
harmonized. This might easily be done defining the
windows in a common reference system, as the object
space for example.
Furthermore, almost all other informations supporting
the point determinations (surface shape, object types,
exluded areas etc.) are defined in the object space too.
It therefore seems to be straightforward, likewise to
manage the density values within this environment. In
addition, such a common definition allows the combina-
tion of multiple image sources, if occlusions or other
problems within the calculation makes it necessary.
Point definition and description of the surface shape
In manual driven evaluation procedures the location of
points might be done very individually due to the inter-
pretative capability of human operators. Points are
defined where necessary to garantee a correct registra-
tion of the surface morphology. For computer controlled
evaluations this is not feasible and instead, a regular,
equidistant point grid of sufficient density has to be
selected (cf. Fig. 1).
Y VERS)
N
ADMI
MUN
NM Ÿ) 0 70)
COOPER
Y Ss A
ANM
Fig.1 Surface registration by a regular and dense
point grid
Supposed, the surface morpholgy is adequately reflec-
ted therein, a way has to be found to use this informati-
on for the rectification of the images.
To achieve this, the grid is seperated into regions of
homogeneous morphology as flat terrain, smooth areas,
steep zones, highly varying areas, regions with few or
129
many discontinuities for example. For each region an
adequate functional set up has to be selected reflecting
the typical surface shape. This might be a plane, a
polynomial or seperated functionals as necessary to
model discontinuities.
The information needed for the determination of those
functional parameters will be extracted from the points
lying in the region in consideration. This has to be done
iteratively, because there is no or only raw a priori
knowledge available.
All points within a region will be determined in parallel.
This assures, that at the end of each iteration all points
covering the surface in registration are known, allowing
for a new calculation of the functional parameters and
thus improving the correctness of the surface descripti-
on.
The dimensions of the regions are strongly interrelated
with the functional complexity. Within flat terrain a great
number of points may be determined in parallel, where-
as with increasing complexity the extension of the region
has to be limited in order to describe even frequent
surface variations correctly (cf. Fig.2).
Although the actual concept uses closed parameterized
functions other functionals might be possible if necessi-
tated to improve the flexibility.
eX EORR
? REN AN
AN) ARR
ARSON ERE
HR) WX) AAA) 0 " N ORNS AN
PARU SAO TS
OH
Un
À "
P (2
LR ZA
EO AC) MN 0
1,2,3 = regions of different morphology
Fig.2 Definition of regions with homogen morphology
Rectification and point determination Between abja-
cent grid points, a few surface elements will be defined.
Within each of them the surface albedo will be calcula-
ted as registrated in the images in concern (cf. Fig.3).
For that purpose, the location of these surface elements
within the images has to be computed. This might be
done directly or by interpolation from adjacent anchor
points in the images. The latter attempt supposes, that
the surface geometry is smooth enough to allow this
simplification. Finally, the albedo value is approximated
