—S 
99). The 
Weidner 
Surface, 
from the 
puildings 
given by 
Order t; 
ngs wil 
h height 
; texture 
| Used to 
he gray 
g pixels 
Shapiro, 
ice with 
second 
scribed 
s kappa 
Sander Oude Elberink 
  
NH NO 
CONTRAST= V S 1-05 10157 
i=0 j=0 
The knowledge that the orientation and shape of man-made objects result in orientated height derivatives allows for the 
use of the contrast texture measure to separate buildings and trees. Trees are supposed to have contrast in horizontal, 
diagonal, as well in vertical directions. Buildings will show contrast at the edges in only one direction. Therefore, one 
can use an anisotropic operation to discriminate between orientated and non-orientated features. The minimum of the 
contrast measures for all directions will have an high value at trees and will show low values at buildings, except for 
corner pixels (figure 5). In the figure only the horizontal and vertical directions of the contrast measure have been 
  
Fig. 5: Flow chart of the anisotropic contrast measure; a) building and tree in height model; b) co-occurrence contrast 
measure in vertical direction; ¢) co-occurrence contrast measure in horizontal direction; d) minimum of two contrast 
measures; e) trees extraction, result after morphological 2x2 opening: f) building extraction. 
showed. These corner pixels of buildings can be removed by a 2x2 morphological gray value opening. Once the trees 
are detected and removed from the height model, buildings will be leftover together with small objects like cars and 
bushes (Oude Elberink, 2000). 
Further options of isotropic texture measures would be the first and second derivatives of height data. On gable roofs 
for example, the first derivative should be constant and close to one, while the second derivative should be zero (Maas, 
1999). A general disadvantage of ‘| em ‘; 
these measures, however, is their 
noise sensitivity (especially in very 
dense data sets) and the fact that 
they take very high values at 
building edges. 
3.3 Practical results of the 
anisotropic height texture Fig. 6: Left: ariginal DSM; center: trees extraction as a result of the anisolropic 
measures contrast measure; eut: pates extraction. 
  
  
  
The following examples show the 
practical results of the minimum of 
contrast measures in four 
directions. The density of the laser 
Scanner data has a great influence 
on the success of the contrast 
measure. In cases of 0.5 meter grid 
resolution the contrast measure 
will give a sufficient extraction of 
  
Fi ig 7: Left: first pulse DSM; center trees extraction as a result of the anisotropic 
contrast measure; right: building extraction. 
  
International Archives of Photogrammetry and Remote Sensing. Vol. XXXIII, Part B3. Amsterdam 2000. 681