ibul 2004 
annot be 
ecause of 
operator 
Itrast in a 
erator in 
le texture 
  
d as an 
on. The 
database 
manual 
number 
with the 
tion can 
28 in the 
lassified 
ees and 
d in this 
contain 
ded that 
t classes 
ATKIS 
| areas, 
jal use. 
that are 
ie of the 
. have a 
iust not 
1g areas 
senting 
selected 
in order 
an. The 
roofs of 
; on an 
and the 
International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Vol XXXV, Part B2. Istanbul 2004 
  
  
  
  
  
  
  
  
  
  
settlement training 
settlement [PF pixels areas 
object above for 
pixel terrain houses 
normalized vegetation 
DHM index 
  
  
  
  
  
  
Figure 4. Calculation of the training areas for houses 
  
  
  
  
Figure 5. Training areas for the landuse class houses 
3.4 Object-based-Classification 
The basic idea of object-based classification is to classify not 
single pixels but groups of pixels that represent already existing 
objects in a GIS database. Each object is described by an n- 
dimensional feature vector and classified to the most likely class 
based on a supervised maximum likelihood classification. The 
n-dimensional feature vector describes the spectral and textural 
appearance of the objects. Again, the trainings areas are derived 
automatically from an existing database. 
3.4.1 Object characteristics: In order to distinguish 
between residential and industrial settlement objects we have 
first to describe the typical appearance of these two object 
classes. The following five characteristics can be used in order 
to decide if a settlement object represents a residential or an 
industrial area (these characteristics are especially valid in 
Germany — in other countries they may differ): 
e average size of the houses: in industrial areas the houses 
are typically very large whereas in residential areas 
houses are typically smaller 
e average roof slope of houses: in industrial areas are 
typically houses with flat roofs whereas in residential 
areas are typically houses with sloped roofs 
e percentage of trees: trees can be found very often in 
residential areas but only rarely in industrial areas 
e percentage of sealed ground: the percentage of scaled 
ground is typically higher in industrial areas as in 
residential areas 
e textural appearance: the textural 
industrial areas is more homogenous as in residential 
areas 
Not all characteristics must be valid for an object. Very often 
only three or four characteristics apply for a specific object but 
appearance of 
this is not a problem because the object-based classification 
classifies the object into the most likely class. This is a very 
robust approach that can handle also fuzzy descriptions of 
objects. Figure 6 shows two typical examples of residential and 
industrial areas. 
a) rgb 
b) LIDAR 
    
   
Figure 6. Comparison of residential (a) & (b) and 
industrial areas (c) & (d) 
3.4.0 Calculation of the object characteristics: In order to 
represent the object characteristics as a feature vector we have 
to transform them into numeric values. Figure 7 shows the 
calculation of the different object characteristic on an example. 
Figure 7 a shows the calculation of the average house size per 
object. All pixels, which were classified as houses, are selected 
from the pixel-based classification result. Than, for each object, 
the average house size is calculated. It can be seen in the 
example that industrial settlement objects are having typically a 
larger average house size per object that is represented with 
brighter grey values and residential settlement objects are 
having typically a smaller average house size that is represented 
by darker grey values. 
The calculation of the average roof slope per object is shown in 
Figure 7 b. All house pixels are selected from the pixel-based 
classification result and intersected with the normalized DHM. 
With that input data, the average roof slope per object can be 
calculated. High average roof slopes stand for areas with sloped 
roofs and are represented with bright grey values and low 
average roof slopes stand for areas with flat roofs and are 
represented with dark grey values. 
Figure 7 c shows the calculation of the percentage of trees per 
object. All tree pixels are selected from the pixel-based 
classification result and for cach object the percentage of trees 
is calculated. Trees can be found only seldom in industrial 
settlement objects. Therefore it can be seen in the example that 
industrial. settlement objects are represented with dark grey 
values which stand for a low percentage of trees and residential 
settlement objects with bright grey values which stand for a 
high percentage of trees. 
uA