ISPRS Commission III, Vol.34, Part 3A ,,Photogrammetric Computer Vision", Graz, 2002 
  
With these conditions a feature analysis operator was created. 
The sensitivity and the point, from which the operator shows a 
positive reaction, were calibrated by human operators by using 
test textures. For every such texture they decided on the 
operator's reaction. Using these data the operators were 
calibrated. 
Table 1 shows exemplary results of the verification of three 
textures with the three image processing operators “parallel 
lines”, “dismembered structure” and “preferred direction”. The 
operator for “parallel lines” shows a positive reaction for 
texture 1 and a negative reaction for texture 3. The operator 
"dismembered structure" shows positive reaction for texture 2 
and negative for texture 1. The third operator “preferred 
direction" found this feature in textures 1 and 3, but only a 
small quantity in texture 2. These results correspond to the 
results of human operators. 
  
  
  
    
  
  
Operator, Parallel Dismembered | Preferred 
Lines Structure Direction 
Text 
1,00 0,01 1,00 
0,34 1,00 0,26 
0,69 0,16 0,90 
  
  
  
  
  
  
Table 1. Reaction of some operators for different textures 
S. STRATEGY OF MULTITEMPORAL 
INTERPRETATION 
Literature discusses different approaches for multitemporal 
interpretation of remote sensing data (Lunetta & Elvidge, 
1999). The first group of approaches is known as pixel-wise 
comparison. These approaches compare pixels of different 
epochs (e.g. Peled et. al., 1998). One possible way is to subtract 
the grey values of the pixel, thus detecting changes. Also, 
different vectors can be subtracted, based on multispectral 
images. The disadvantage of these approaches is the necessity 
of precise spatial rectification. 
The second group of approaches are known as postclassification 
approaches. These approaches start with a separate 
multispectral classification of the images of every epoch (e.g. 
Weismiller et. al., 1977). Then the classification results of the 
different epochs are compared to each other. Here it is of 
disadvantage that these approaches depend on the classification 
methods. It is difficult to decide whether the differences 
between the classifications of the various epochs result from 
inaccuracies of the classification method or from real changes 
in the scene. These approaches also need precise spatial 
rectification. 
The third group of approaches compares the images of different 
epochs on the semantic level. Different conditions are 
formulated for the possible changes of different objects from 
A - 236 
one epoch to another. Therefore, the objects in the scene are 
interpreted by using the knowledge of possible temporal 
changes. The approach used in this work is assigned to this 
group. 
The approach discretely describes temporal conditions of 
regions, and it transfers the most probable temporal changes of 
the given conditions as temporal knowledge into a state 
transition diagram. This is used for multitemporal 
interpretation, which means, that the temporal part of the prior 
knowledge is implemented into a state transition diagram. 
Figure 1 shows the state transition diagram, which in this 
approach was used for the interpretation of industrially used 
moorland. 
Area of KL AN 
Degeneration N 
| \ 
SES N 
N 
NN d ji 
= 7 (. Na | SW 
Milled Peat 
1 
| A ov 1 2; di ENTaeion Aras d Area Extraction 
\ | IN | | 
X N 
Forest 
Soil with Spores | / | 
| before [ [S5 % | > 
Peat Extraction | Y IN s Y 
[| 
| fra dod Tay |. Soil with Spores 
| | 
| after | N Inactive Area of 
| | a 
A Peat Extraction Peat Extraction 
  
  
  
| Milled Peat IN | 
\ | Strip Extraction ~~~ x | e 
\ E e 
A a Swe 
5 — 
\ A at > | 
yir id ^is Ned | 
“ x 
a ge A 
N Area of | esl 
\ 2 4 
J NS 
Wet Area oos. - : 
| fF in « 
f, 
A Birch State 
{ 
PNE NM E 
  
  
  
i 
Nl Area of | 
N Regeneration in | 
| Is 
; T Heather State | 
\ 
TET 
Figure 1. State Transition Diagram 
Although many more state transitions are theoretically possible, 
there are restrictions by law and by nature, and we can use these 
restrictions to improve the interpretation. The state transition 
diagram contains twelve different states (in multitemporal 
context, in the following classes are also designated as states). 
The first state, upland moor, is implemented only to complete 
the diagram. Distinction of so many states is only possible by 
using for every segment the knowledge of temporal history. It 
is, for example, very difficult to distinguish the “area of 
degeneration” from “area of regeneration” in aerial images 
taken at one epoch. But this distinction is possible by using 
temporal knowledge. For example: If the state “area of milled 
peat extraction” is given, the system will know that this 
segment has passed the state “area of degeneration”, and it will 
not try to find features and structures for the state “area of 
degeneration” for the interpretation of the next epochs. The 
search space as well as the possible errors will be reduced. This 
use of temporal knowledge also allows distinction of the two 
regeneration states “heather” and “birch”. 
Based on the concepts described above the system used was 
extended by temporal relations (see Growe, 2001). They realize 
the use of temporal knowledge. For each temporal relation a 
priority can be defined for sorting the possible successor states 
by decreasing probability. During scene analysis the state 
transition diagram is used for generating hypotheses for the next 
observation epoch. For each of these possible state transitions a 
hypothesis is generated. All hypotheses are treated as 
competing alternatives.