Figure 4. Segmented images 
a-before-event, b-after-event 
Figure 3. Correspondence region matching arithmetic 
Figure 5. Change detetection results 
a-Changed area, b-Omitted area, c-False alarm area 
d-Visual interpretation result, e-Image differencing result 
On comparing the ARs at the threshold of 15, the changed 
regions are labelled on after-event images,delineated by 
EafiSS.Overall accuracy is 89.73%,omission error is 5.79% and 
commission error is 4.48%. 
Visual interpretation is largely dependent on analysts’ 
expertise,which has a higher accuracy ,but is 
time-consuming.As Figure 5. e shows,image differencing 
mainly detects the edge of changed proportion and omitted 
phenomena is very severe.Moreover, the wholeness of changed 
area is bad and “salt and pepper” effect exists in the resulting 
map.After the comparison,it is easy to find that region-based 
change detection method has higher accuracy than pixel-based 
approaches, a little inferior to the visual interpretation result,but 
is time-saving. 
6. CONCULSIONS 
With the popularization of high resolution image 
resources,region-based change detection methods close to 
human visual interpretation characteristic has a bright future.lt 
is proved that the technique proposed in this research detects 
object-specific changes effectively with a accuracy above 
85%.Furthermore,this method compares regionsfnot pixels) 
features so that it is not sensitive to the accuracy of image 
registration.Also,Features are not relatated to the characteristic 
of images radiance and the acquisition way so as to combine