| 2004 
e the 
s and 
puting 
d scale 
classes 
f. 
es are 
inherit 
ritance 
ce the 
'S to a 
group 
ire that 
and all 
part of 
rchies, 
chy of 
groups 
is tool 
ions to 
ects we 
| means 
>oolean 
| where 
(class) 
evel of 
ting of 
solved 
S9. We 
S1. 93, 
and S9 
al with 
(3) 
International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Vol XXXV, Part B4. Istanbul 2004 
The inheritance hierarchy and the group's hierarchy 
complement each other in the following sense. While the 
inheritance hierarchy is used to subsequently separate and 
differentiate classes of the objects in the feature space, the 
group's hierarchy permits the meaningful grouping of the 
resulting classes. 
The local government, land management and regional 
development solve many tasks where the above described 
approach is applicable on the different levels in the field of 
monitoring, mapping, inventory, change detection or parameters 
estimation. 
The class hierarchy. and related restrictions and assumptions, 
which can be continually changed during the processing, it is 
the base of knowledge for the image object classification. In 
many applications the desired geo-information and the objects 
of interest are than extracted step by step, by iterative process of 
classifying and processing. It is very similar to human image 
understanding processes. 
This kind of circular processing results in a sequence of partial 
states, with an increasing differentiation of the classification 
result and the increasing abstraction of the original image 
information. 
On the each step of the abstraction new information and new 
knowledge is generated and can be used beneficial for the next 
analysis step. High beneficial is the fact, that after successful 
analysis, a lot of interesting, additional information can be 
derived. 
There are many reasons for the knowledge-based spatial data 
network building and sharing. The pixel-oriented classification 
can be accepted as the pre-processing phase that is followed by 
object-oriented contextual classification applied through 
decision tree design. 
2.2 Decision-making process 
The average resolution of image objects can be adapted to the 
scale of interest and resulting information can be represented in 
the scale based on the average size of image objects. This fact is 
coherent with the hierarchical networking and representation of 
image objects. In the hierarchical structure each object knows 
its neighbours, sub objects and super objects. 
— 
BÀ d ta 
    
Figure 2. Part of cadastral plan. 
We distinguish three types of elemental object features: 
Q Attribute (physical properties of objects that follow 
from real world or image or different information 
layers related to the object). 
109 
OQ Topological features (describe the geometric 
relationships between the objects or the whole scene). 
C) Context features (represent the objects’ semantic 
relationships. The gardens are inside the urban area; 
the island is surrounded by the water, and so on. It 
means between class (object) relationships). 
  
Figure 3. Identification of outdoor pools inside the gardens. 
The different segmentation techniques can be used to construct 
a hierarchical network of image objects and each level in this 
hierarchical network is produced by a single segmentation run. 
The hierarchical structure represents and contains the 
information of the image data at different resolutions 
simultaneously. Fine objects are sub-objects of the coarser 
structures. It means, that each object knows its context, its 
neighbourhood and its sub-objects. 
3. FUZZY DESIGN 
3.4 Membership function design 
Each of the decision-making process is associated with some 
level of overall uncertainty. Human experience is often 
expressed using the expressions like: a little bit more, more than 
usual, less than last time, etc. 
This uncertainty can rise from the facts that: 
QO We are not able to define exactly the problem, the 
class of objects. 
Q We utilize not quite correct spatial data with respect 
to define task. 
Q We are not able to verify all assumptions, and 
estimate and validate all restrictions we need to build 
out the hierarchical decision model. 
The automation of the classification and interpretation process 
gives better results when the fuzzy elements are implemented 
into the decision structure. 
One of the well-known approaches to decision making based on 
uncertain information is Fuzzy Logic. This framework is 
frequently used for decision making in case when the expert 
knowledge and rules are to be implemented into the decision 
[Saeco tae tanien tici or 
A 
ee 
Le 
= 
ERT 
à 
ir 
LE 
us