The International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences. Vol. XXXVII. Part Bl. Beijing 2008 
Fig.2 Surface and section-view of DSM 
Fig.3 Surface and section-view of DTM 
Fig.4 is a binary image of initial segmented surface S 0 acquired 
by height difference. Fig.5 is a binary image of refined initial 
segmented surface So' being added with edge information. 
The white represents building while black represents other 
objects. Comparing Fig.4 with aerial image data (in Fig-9), we 
Fig.4 Initial building segment S 0 
Fig.6 and 7 are segmentation respectively by quadratic 
gradient and area. We set the quadratic gradient threshold 0.03 
and area threshold 9m 2 . In Fig.6, most vegetation was removed 
by quadratic gradient threshold except few vegetation points 
which were easily removed by area threshold. It shows that 
can see that the ground points are easy to be segmented from 
non-ground points through height difference threshold only few 
left (such as the red region in Fig.4). These points could be 
removed after adding edge information (Fig.5). 
Fig.5 Initial building segment S 0 ’ 
combinative applying the two parameters by this order can 
accurately segment building from vegetation. But there have 
some absence on building roof and incomplete edge 
information which needs further refining. 
Fig.6. Building segment by quadratic gradient threshold Fig.7. Building segment by area threshold 
Fig.8 is the accurate building information acquired by local 
refilling and iterative processing, in which a means binary 
286 
image and b represents 3D model. From the two images we can 
see that, by neighbor iterative approaching, it is better to