hul 2004 
n (1999) 
ions and 
original 
sures and 
1 LIDAR 
ges was 
types of 
nages 
se range 
nation of 
he NDVI 
and NIR 
  
International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Vol XXXV, Part B7. Istanbul 2004 
channels of multispectral image data a range based NDDI is 
defined by 
NDDI - 1 —P. (9) 
Ip - lp 
where fp and Ip indicate the first-pulse and last-pulse range 
image data, respectively. 
       
          
A: DRM MP 
Figure 4: TopHat filtering of last-pulse laser range image 
The term V — and B — classes are chosen because we expect that 
clustering based on the NDDI band and the TopHat band will 
directly point towards vegetation areas with significant 3D 
extend and building areas. But please note that this has no direct 
relation to supervised classification where training sets are 
selected and used for classification. 
  
Figure 3: Normalized difference distance image derived from 
first-pulse and the last-pulse laser range images 
Figure 3 shows that the NDDI image enhances vegetations areas 
with a significant 3D extend. In addition, it can be noticed in 
figure 3 that power lines show up in the NDDI image (Upper 
right region of Figure 3). 
The morphology TopHat operator is utilized to filter elevation 
space. The TopHat transformation with a flat structuring 
element eliminates the trend surface of the terrain. A certain 
problem is to define the proper size of the structuring element 
which should be big enough to cover all 3D objects which can 
be found on the terrain surface. The TopHat operation is 
defined by: 
TopHat = DSM - (DSM ese) (10) 
  
where DSM is the input surface for filtering, se is the 
structuring element function, and © indicates the operator for Figure 5: K-means clustering result (B class regions highlighted 
grey scale opening morphology. The TopHat filtered last pulse in yellow) 
range is shown in Figure 4. It enhances the 3D objects relative 
to the ground surface in the last pulse range image. 3.1 K-means clustering: 
The K-means clustering results is shown in Figures (5) and (6). 
B class regions are highlighted in yellow in Figure 5 and V 
class regions in green colour in Figure 6. Visual inspections 
shows that V-class is directly associated with vegetation, in 
particular trees, bushes or forest and the B-class is mainly 
associated with building regions. 
Input to the clustering processes is the NDDI ratio between first 
and last pulse range images (NDDI band) and the TopHat 
filtered last pulse range image (TopHat band). The three 
processes K-means clustering, fuzzy C-means clustering and 
competitive learning networks are employed and the results are 
shown in the following. 
For all three clustering techniques we will 3.2 Fuzzy c-means clustering 
e restrict to four classes: a V-class, a B-class, a 
Backeround class and a Null class Similarly utilizing the fuzzy C-means algorithm provides the - 
results shown in Figures (7) and (8). 
41