338 
(between lines or planes), etc. Constraints and criteria concern 
extreme values of variables and parameters, continuity of lines 
and/or surfaces, thresholds, etc. Control and check data comprise 
individual points, test fields, check profiles, etc. 
External information should be adequately conditioned (i.e. 
structured and formatted) for use in the subsequent process stages. 
12. Predicting conjugacy 
For a lay-out of target segments (vide III.5), the corresponding 
search segments can be located approximately. A good aproximation 
increases reliability and efficiency of matching. 
At pre-processing, prediction can be based on known terrain model 
geometry and/or coarse pre-measurement of the maximum parallax. 
Approximate conjugacy can also be assessed during matching, i.e., 
from the already known neighbouring parallaxes whereby the neigh- 
bourhood can be in space (for static scenes) or in time (for dynamic 
Scenes). 
Prediction of approximate conjugacy can be regarded as part of the 
matching process. 
Trends in pre-processing 
The trends can be summarized as follows: Pre-processing and matching 
strategy tend to be optimised together, and external (a-priori) 
information is being integrated in the process. Moreover, interac- 
tive (man-machine) operation for exclusion of regions, supervision, 
and interference in critical situations is gaining in importance. 
IV IMAGE MATCHING 
By matching the conjugate data of a target segment are identified in 
corresponding search segment. Matching strategies and techniques can 
differ; table 1 indicates four basic variants. 
| 
  
  
  
  
     
      
Lay-out of e n seme 
segments pa um 
single multiple -- Identifying Assessing +, 
Algorithms | conjugate -————» similarity 
  
features of segments 
Single x X rr 
‘Parallaxes 
Multiple X X | 
LPostrpracesssinge 
T 
Table 1: Variants of matching Fig. 4: Interactions at 
strategies matching 
| 
Parallaxes 
  
     
  
  
  
  
  
Figure 4 shows a two-stage strategy of matching, including various 
interactions. The two stages concern distinct image features (or 
primitives) and non-distinct image segments. The two stages tend to 
reenforce each other. Parallaxes (of segments) are byproducts of 
matching. 
Matching strategy, algorithms, techniques and corresponding hardware 
are strongly interrelated (figure 5).