dies, The techniques used are similar to those developed for
the rectification of aerial photography (photogrammetry), and
by analogy the term radargrammetry has been coined. Radargram-
metry requires a mathematical model relating an object point
in the Cartesian coordinates to the radar coordinates, i.e,
time and slant range distance. This type of processing requi-
res the use of digital terrain and digital height models. Aq-
ditionally, the position and orientation of the sensor (sensor
ephemeris) during data acquisition should be accurately deter-
mined and used. A simpler rectification procedure uses control
points or control lines as coastlines. Geometric correction
with control points is achieved by correlating image chips
with a library of pre-selected chips from previous passes,
Another preliminary processing to be done once the image
is formed refers to the calibration of radar backscatter to
remove system dependent effects such as antenna gain pattern,
The different features which can be computed for each re-
gion of the segmented image are: texture, shape and adjacency
values. The most common way to define the texture of a region
relates to the statistical description (e.g. mean value, stan-
dard deviation/mean) of the grey level distribution.
At this point of the processing each segment of the image
can be classified. Generally speaking, this high level pro-
cessing can be performed by means of statistical pattern re-
cognition, syntactic pattern recognition or KBS approaches.
In the first method, a set of N-dimensional features are
clustered into a number of classes. Statistical tests are
available which check the points for purely random distribu-
tion and determine the number of distinct classes in the popu-
lation. This is a well advanced technique, but little work
has been carried out with SAR data. Image structure is used in
the syntactic pattern recognition. Primitives describe the
shape of the regions in the image and strings describe collec-
tion of regions. Image classification is obtained by proces-
sing the strings through rules of a suitable grammar. A major
limitation of this image processing procedures is the inabili-
ty to handle contextual information, because contextual infor-
mation tends to be of declarative nature rather than of proce-
dural form (i.e. mathematical modelling and derivations) com-
monly used in image processing algorithms. Information of de-
clarative nature can be incorporated in a KBS. Only recently
this technique has been applied to remote sensing imagery,
neverthless a number of KBS programs have been developed th-
roughout the world for processing multitemporal and multi-
Spectral digital images and for detection changes.
Concerning ARTS-IP, image processing algorithms take place
at the on-line and off-line levels. On-line image processing,
also referred to as simplified KBS (SKBS), is now considered.
Examine the case in which the LRWS mode applies: in. the 1st
look while the HRNS mode applies in the 2nd look. The SKBS is
a matter of segmenting the image (low level image processing)
and checking for relevant changes with previous segmented ima-
354
| >>
A wer vct 84