y be 
ected 
urate 
ee 
mage 
finite 
finite 
falls 
these 
based 
issign 
ng to 
„The 
or be 
specific 
d Rı) 
s region 
adtree 
ed on 
ibutes 
matches the set of the attributes for the particular 
object class. This work has to be done under the 
guidance of DGs. For example,if the image shown in 
Figure 4 is of an agricultural area containing a 
number of fields and there are three categories ,such 
as arid land,irrigation land ,and bare land , which 
have the same attributes size 60 in the image, the 
quadtrees may be related with the same abstract object 
class, Arid/Irrigation/Bare land. As more attributes 
are discorved in the image, each interpretation class 
may become more specific. R, in Figure 4, for 
instance,can be interpreted as an Arid land according 
to the attributes; 
SizeZ260 and 
90<Hue<120 and 
Saturationzz0. 8 
4.2 Spatial relation extraction based on quadtree 
  
In the last process , we see that quadtrees 
interpretation and image processing are mainly 
concerned with statistical and geometric attributes of 
the image, and classifying shapes that appear in the 
image. However, it is not always unambiguous to 
interpret quadtrees using these attributes. As a 
continuation of the interpretation , we consider further 
solution by using some spatial relations in the 
homegeneous region quadtree. 
An simple example in Figure 5 will illustrate the use 
of spatial relations. In accordance with the attributes 
of the regions R;, Ry, we can not discern if they are 
willow lands or poplar lands. If we consider the 
relation between river (R3) and willow lands (R, and 
R;), the region R, and R; may be interpreted as 
willow lands.So the relations between objects is 
important to exclude the ambiguity of the image 
interpretations. 
IK] Ri, R;—willow/Poplar 
E s 
  
River 
  
Figure 5. An example of spatial relation 
561 
In this paper ,spatial relation extraction is perform on 
the control quadtree and feature quadtree. 
Control quadtree is a enclosure quadtree that is applied 
for testing the relation between homegenous region 
quadtrees or class interpretation quadtrees. Detecting 
the relation. between quad rant c and b can be 
simplified to shrinking and expanding a enclose 
polygon (see Figure 6 a). 
Enclose polygon r---_4 
CC I 1 
’ | | 
! mum 
l--4 C] | [T domesenents ! | | 
| CT ien quadtree |! rim 
I 
L. PER 1 I I 
+ 
—— 0 
a b 
pe] 
icum eie] 
Le : el 
| | 
p «l 
c 
Figure 6. An example of spatial relation extraction base on quadtree. 
Feature quadtree is a specific quadtree which extract 
two kinds of essential components of control 
quadtree; (a) feature node , which is used as label to 
represent the intersecting node caused by the meeting 
of several line quadtrees of a control quadtree (see 
Figure 6 c). (b) feature line quadtree , which is a part 
of control quadtree used for representing the relations 
between homogeneous region quadtrees (see Figure 
6b). Thus the spatial relation extraction is performed 
as the following: 
1) Construct control quadtree by shrinking and 
expanding enclosure polygon. 
2) Quadtree node that is a part of a control quadtree 
is labeled in terms of the relation between 
homogeneous region quadtrees. 
3) Extract feature node and feature line quadtree. 
4) Extract spatial relation between homogeneous 
region quadtrees by using feature node and feature 
line quadtree,and a spatial relation quadtree is made. 
5. CONCLUSIONS