ISPRS Commission III, Vol.34, Part 3A „Photogrammetric Computer Vision“, Graz, 2002 
  
  
streets VIS 
region of interest 
houses settlement 
  
garden 
Figure 5: Example 1 
The objects extracted as houses have a high quality fac- 
tor, because the area of these objects lies between 60m? 
and 500m? and the orthogonality of the house corners is 
fulfilled. For the other existing class garden the expected 
area has realistic values. The ratio of the area filled with 
houses and the garden area is used for the interpretation of 
the examined region, also the quality values of the objects 
itself. 
The second example is more complex, it describes the sep- 
aration of a region into several classes with use of the 
quality values for the concurrent interpretations during the 
analysis. In figure 6 a part of the lower left side of fig- 
ure 4 is shown separated by streets, which was extracted 
by application of a road detection operator. 
        
i 2 U 
buildings| parking area garden houses 
acreage 
Figure 6: Example 2 
On the left side of figure 6 objects that permit the inter- 
pretation industry (e.g. buildings or parking areas) are il- 
lustrated with different colors, on the right side the same 
region is shown with illustrated objects that permit the in- 
terpretation settlement (e.g. houses or garden). Black re- 
gions illustrate areas that are not compatible to the sup- 
posed class. The decision for the region between the two 
competing interpretations is motivated by the hypothesis 
industry. The areas that are interpreted as buildings have 
attributes with high quality values for this interpretation. 
The areas of the different buildings have values between 
120m? and 1000m? and the orthogonality of the objects is 
fulfilled. The other objects, like parking area and group of 
trees have expected realistic values for the class industry. 
For the classification of the examined region the quality 
values of the objects and the ratio of the whole area of the 
region and the sum of the building area is used. For the hy- 
pothesis settlement the analysis is similar, but the objects 
and object types are taken from class settlement. The anal- 
ysis has the following results: the hypothesis industry has a 
quality value of 0, 92 the competing hypothesis settlement 
has a value of 0, 42. The analysis selects due to the quality 
values the interpretation industry. The sum of all objects 
of the class industry build the result area. 
The big regions on the lower left and lower right side of 
the image in figure 6 are not included to the interpreta- 
tion industry, because this acreage and forest areas do not 
match the class industry. These areas are interpreted as 
own classes, so that a separation of a region into three sub- 
regions is realized. 
The third example should explain the use of spatial rela- 
tionships between the extracted objects for grouping the 
objects. The classes forest and settlement implicit that the 
included objects have a special spatial relationship among 
each other. The houses or trees of these classes have dis- 
tances less a normal value. To illustrate this issue in fig- 
ure 7 the result of extracted trees, extracted with use of 
laser-scan data, is given. In addition to the extracted sin- 
  
   
   
  
Single. tree_1 | 
1 
| 
| 
  
i tree 
[ Si nale_ 2 
i ] L Sings tee. k | 
Forest. 1 [2m . Forest 3 BEN 
[- Single tree. 1 Bottom-Up L— Single tree 1 | 
"Boltom-Up, 
p ae tree_2 | Operator : em tree 2 | | 
i i | 
L single. tree. n | Single tree | || 
| j 
  
  
Figure 7: Example 3 
gle trees a graph that connects all trees is plotted. The sum 
of all distances of the connections is minimized. The dis- 
connection of the graph at the marked positions A and B 
generates two groups of trees, that have enough trees to be 
interpreted as forest. The two found regions of forest are 
marked with dark color. The two interpretation steps in 
figure 7 show the change of the instance net. This method