ng 
e, 
by 
np 
]se 
on 
ch 
ler 
ee 
he 
list 
very useful for remotely sensed image. 
3.2 Region relation formation 
  
On the synthetic land unit map, each region is the 
Structure of a matrix or a rectangular raster and the 
relations between regions cann" t easily be judged, so 
we introduce the region relation representation. 
At first, we defined an image or digital map as 
complex with a label assigned to each complex element 
. The label originate from the gray values measured 
during the scanning of an image. The region then is 
formed in accordance with the corresponding notion of 
the subcomplex (V * A * Kovalevsky ,1988). 
Because the complex labing image is still represented 
in raster form ,to obtain topological information , we 
transform the complex into a data structure called the 
correlation list . It consists of o — dimensional, ] — 
dimensional, and 2 — dimensional topological sublists 
(Guan, 1990). 
With the description of correlation list, appropriate 
region relation graph (see Figure 5a) and region 
relation list. (see Figure 5 b)can be made to represent 
the relations between regions. 
In the relation list ,a; ,a; are the labels, identifying the 
relations between regions such as adjacency ,parallel 
  
  
  
  
  
  
  
  
  
etos, 
€ A|B|C|D/|--- 
A Alla |0]0]- 
B Bjai|1][laila 
| [Em 0 E clofaf1]o0 
Dj01lal0!11]--. 
£j 
  
  
  
  
  
  
  
  
a Region relation graph 
b Region relation list 
Figure 5. Region relation representation 
3.9 Image interpretation 
The organization flowchart in Figure 6 gives an 
overview of the procedure to use a computer system to 
interpret an image based on multlevel representation 
tree. Reference to Figure 7,the object interpretation 
881 
precedure is briefly described as follows. 
1) Form segmentation image and digital maps. 
2) Construct synthetic land unit map. 
3) Construct correlation list , region relation list ,land 
unit relatability list and attribute list. 
4) Construct a knowledge base which consists of 
different types of knowledge such as geometric, 
topologic ,spatial related knowledge etc. 
5) If only one interpretation is found by matching the 
image features and rule hypotheses, the interpretation 
will be assigned to the image region ,or the knowledge 
base is revised or more information extracted from 
image and auxiliary data must be acquired. 
6) If there is no information and knowledge to specify 
the interpretation further, assign the image a 
ambiguous interpretation class. 
  
Acquisition of image 
  
and auxiliary 
data 
  
  
  
es 
Segmentation 
  
  
  
Synthetic land 
image unit map 
| [4 
| 
Construction of 
  
  
  
  
  
  
  
  
  
. ; à Construction of 
correlation list Construction of > 
[^] "land unit 
and region relation attribute list 
relatability list 
  
  
  
  
  
list 
  
  
n 
  
Construction of 
  
knowledge base 
  
  
  
  
  
    
Revise rules nterpretation 
  
   
  
  
  
  
js determinate 
   
  
Input interpretation 
class 
  
  
  
Figure 6. A procedure to interpret an image based 
on multilevel representation tree. 
For example ,in Figure 7 region R; of segmentation