318 
5.4 Terrain features modelling 
The terrain features specified for a GI base should be modelled in 
3D space, though 3D representation is not required for all 
applications. Hence, basically the stereo methods of restitution 
should be used; the mono methods can be regarded as a supplement. 
Both methods can be implemented in analytical as veil as digital 
systems. The procedures for preparation, manual restitution and 
post-processing are similar, and both system types can have the 
graph-image overlay capability for single images or stereopairs. 
Analytical systems can be equipped with high-quality observation 
optics, with a resolution greater than 150 lp/mm, and a field of 
view up to 40 mm diameter. High-quality optics and a large field of 
view allow effective manual image interpretation. Interpretation 
requires a simultaneous perception and analysis of the specified 
terrain features and their context. Both the features and their 
context have their specific structures of the primitives and their 
composites, which have to be analyzed and understood for correct 
interpretation. 
The geometric accuracy of analytical stereoplotters is usually very 
high (s=l-10 ym). The positioning and measuring devices have to be 
periodically tested and calibrated to update the parameter values of 
the corresponding correction models. The calibration process can be 
semi- or fully automatic. 
Digi tal systems have the inherent potential to carry out processes 
also in the semantic domain, i.e., preprocessing, analysis and 
understanding of images, and selective feature extraction. There is 
an on-going transition towards more intelligent systems but 
automatic image interpretation is today still in the stage of 
infancy. 
At present, the major limitations are low quality of digital images, 
caused by the relatively large pixel sizes (10-50 ym), and small 
display windows (512 x 512 - 1024 x 1024 pixels). To preserve the 
information content of a high quality image (e.g. 150 lp/mm), a 
pixel size of 2 ym is needed. The image window required for display, 
which would correspond to the field of view of 40 mm diameter in an 
analytical stereoplotter, is represented by a matrix of about 18000 
x 18000 pixels. This is, however, beyond the state of the art. 
For the commonly used photogrammetric images and an average 
stereoplotter, the requirements are less stringent. In this case (50 
lp/mm; 30 mm diameter), a pixel size of 7 ym and thus a display 
matrix of about 3800 x 3800 pixels suffice. This can be attained at 
present, though the price is high. 
The corresponding intensity value (8 bit) or the colour (24 bit) is 
assigned to each pixel; this results in a huge amount of image 
information to be stored, retrieved, transferred, processed and 
displayed.