334 
The impact of these operations on performance and reliability varies 
widely and it depends strongly on the overall process strategy. 
The individual operations include the following. 
1. Merging or. linking 
Two or more related data sets can be merged into a single set or- 
they can be linked mutually. Re-scanned image data or neighbouring 
data (pixels or lines) can be merged, e.g., by averaging or by logi- 
cal "AND" connection of the corresponding intensity values (1). 
Merging implies a simple form of resampling (vide III.4.). 
Linking is applied to a-priori data (key- and attributes), e.g., to 
tie them with the image raster. This implies gridding of data (e.g., 
distinct lines and surfaces of morphometric and/or artificial fea- 
tures) into a raster. Key-features can be linked with different 
attributes by means of pointers or addresses. Examples are classes 
of regions and networks (of chains and points). Such a classifica- 
tion can serve for specifying the parameter values for subsequent 
processes. A-priori data represent an autonomous data set, and 
should therefore be preserved (in original form) for further uses. 
2. Exclusion of regions 
Anomalous and/or non-relevant regions should be excluded before 
further processing. Examples of anomalous regions are areas covered 
  
' by clouds, water, snow or other featureless (homogeneous) areas. 
These can be manually delimited and excluded. Anomalous regions are 
inside (internal) or along boundaries (external) of the area of 
interest. The situation is more complicated when internal regions 
are nested (i.e., lakes with islands with lakes, etc.). 
3. Corrections 
Corrections concern both geometric and pictorial (intensity), do- 
mains. Geometric corrections can be applied for the camera (or sen- 
sor) internal geometry and its external orientation (attitude). 
Both can be differentiated further. 
Intensity corrections concern issues such as reduction to mean le- 
vel, amplitude scaling, compensation for image spread (inverse fil- 
tering), CCD sensor characteristics, etc. The choice and application 
of corrections requires utmost care. 
4. Resampling 
Correcting and resampling can be carried out separately or in combi- 
nation.  Resampling produces a new data set from an existing (old) 
one. It usually involves both geometric and intensity domains. The 
simplest geometric version is to form new pixels composed of 2, 4, 
6.... old pixels (or of 4, 16, 36... old pixels) by averaging their 
intensities. Another simple version is resampling merely in the 
intensity domain, i.e., to produce 4 bit (i.e., 16) intensity levels 
from initially 8 bit (i.e. 256) levels. | 
335