As described in chapter 3 an initial D-DSM was computed from 
the input data. A threshold of 2.5m for the absolute difference 
was then used for each pixel individually, which as expected 
significantly reduced the amount of potential change. The 
resulting binary D-DSM is depicted in Figure 5. 
A first visual check revealed that while most building changes 
are contained in the D-DSM, there are also a number of false 
alarms. Some refer to potential building activity and represent 
excavations, others refer to matching errors, probably due to 
poor contrast. 
An example of errors in the D-DSM is shown in Figure 6. The 
green boxes in the superimposed to the GeoEye images are 
correct building changes, but the red boxes are false alarms: the 
upper one is a group of parked trucks and the lower one belongs 
to a dump caused by a building under construction. 
  
  
  
Figure 6. Two samples of the D-DSM - left: IKONOS image 
(epoch 1), centre: GeoEye image (epoch 2), right: D-DSM. See 
text for further explanation. 
4.2 D-DSM refinement 
For the reasons explained above, candidates for building change 
from the D-DSM had to pass two more tests as described in 
chapter 3: first, in at least one epoch a blob must be detectable 
in the nDSM; and second, in epoch 1 detected building blobs 
needed confirmation from the GIS building layer: a blob must 
cover at least 7596 of the GIS building object, otherwise it was 
deleted. 
Figure 7 shows the nDSM of the IKONOS and GeoEye images, 
Figure 8 depicts the results of comparing the nDSM of epoch 1 
with the GIS building layer. The resulting binary change map 
after all height checks can be seen in Figure 9. 
  
Figure 7. nDSM of epoch 1(left, red) and epoch 2 (right, green) 
as binary maps after applying a height threshold of 2.5m. 
  
International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Volume XXXIX-B7, 2012 
XXII ISPRS Congress, 25 August — 01 September 2012, Melbourne, Australia 
   
  
Figure 8. (a) IKONOS DSM of the whole scene, (b) IKONOS 
DSM superimposed with height blobs from nDSM in red, 
(c) GIS objects covered (yellow) and not covered (red) by 
height blobs, the latter being candidates for demolished 
buildings, (d) blobs (red) not covered by GIS objects, being 
candidates for new buildings. 
  
  
  
  
  
Figure 9. D-DSM as binary map after refinement by height 
information 
As shown in figure 7, there are still a number of change 
candidates with rather strange shape and size, making it 
necessary to further reduce false alarms. As described in 
chapter 3 we use morphological filtering for deleting small and 
thin areas. The size of the filter mask used was set to four 
meters, this value was found empirically. The result is depicted 
in Figure 10, the areas in red are those which are finally 
considered as building change. Note that box fitting or any 
other means to consider the fact that buildings are normally 
rectangular to further reduce potential false alarms has not been 
applied yet.