ISPRS, Vol.34, Part 2W2, “Dynamic and Multi-Dimensional GIS’’, Bangkok, May 23-25, 2001 
281 
ie impact of 
detection is 
s to remove 
; anticipated 
current and 
ipproach of 
jmporal and 
ie elements 
:ontrol point 
parameter 
tsed image 
t it includes 
procedure, 
;dge about 
tained from 
} difference 
3 detection 
election of 
Rested by 
ficulty and 
in the very 
areas with 
i And those 
) uncertain 
Dcedure is 
tem 
especially 
them is to 
features in 
ant factors 
in images 
srences in 
condition, 
eatures in 
)rocessing 
:tion, area 
)e system, 
e features 
3n the two 
ide image 
pdate the 
uncertain 
changed area they are subdivided into smaller changed and 
unchanged area. And for those certain changed one this 
comparison is a confirmation procedure. When finding some 
conflict results this area will be subdivided. It can be seen that 
this subsystem is based on feature level. 
The algorithm is one of important problems about 
comparison of features. Take one image as reference image, 
features on another image are selected for searching the 
match results in reference image with maximum similarity one 
by one. How to define similarity? For different features (for 
example, edges in the reference image and area in another 
image )it is a easy task. But indeed it is a confused problem for 
same features. Obviously accurate comparison is not a good 
choice because the features themselves contain uncertainties. 
So using tolerance is a advisable method. Here one factor 
called ‘buffer distance’ is 
Data Input (Multi-sensor imagery 
,map and other data ) 
1 
r. 
Information 
visualization 
update and 
Automated image-image 
and image-map registration 
Automated 
change 
identification 
Change detection 
accuracy assessment 
Intelligent change 
features recognition 
& interpretation 
Automated change 
features extracting 
Fig. 1 A framework for automated change detection 
introduced .The so-called buffer distance is a tolerance 
distance in which two same features are identified 
equivalence. Commonly the buffer distance is 2 ~ 4 pixels in 
images. 
3.4 Intelligent change feature recognition subsystem 
After finding changed area and identifying change 
feature, 
another important step is to recognize changed feature and 
classify them to different attribute classes. In this procedure 
many classification algorithms can be combined to get the right 
results. However, because these algorithms always bring 
unreliable information some knowledge-based feature and 
attribute extraction and recognition systems would have to be 
developed. The used knowledge can obtain from database A