/ pixels 
naterial 
g to the 
already 
is to be 
le from 
s those 
iles are 
ay tiles 
elected 
itumen, 
ten are 
naining 
ses and 
centre 
to map 
ne, the 
able in 
ed by 
features 
raction. 
lysis of 
lowing 
identify 
assified 
)DATA 
ut map, 
A class 
red that 
lass for 
CESS. 
a set of 
gorithm 
:d. This 
| space 
ing the 
iminant 
analysis feature extraction (determines a feature subspace that is 
optimal for discriminating between defined classes). The output 
of the extraction is a linear combination of the 125 original bands 
to form new bands (features) that automatically occur in 
descending order of their value for producing an effective 
discrimination. Twenty two (22) features are obtained from the 
feature extraction process. However, only the 11 features obtained 
in the final feature extraction transformation matrix (DAFE) are 
used to form a new data set since these provide most of the 
available separability and this is confirmed by the magnitude of 
the corresponding eigenvalues (high values). The new data is 
classified using the ECHO classifier. The output classification 
map is overlaid with an orthophoto covering the same area as 
shown in Figure 2. 
Bitumen 
Roof material 2 
Roof material 3 
Zinc plated sheet 
Kaufland roof material 
Roof material 5 
Roof material 1 
Red roof chipping 
Roof material 4 
Vegetation 
Background 
  
Roof material 6 
  
Figure 2: Overlay of classification map and orthoimage. 
The classification map fits well with the orthophoto and this gives 
an indication of the accuracy of the classification in terms of 
geometry. In order to identify areas in the classification map 
which require improvement, the corresponding classification 
probability map is inspected (see Figure 3). The pixels 
represented by yellow to red colours in the probability map 
indicate a high probability of being correct. These pixels are very 
close to the training pixels for the classified pixels. Dark blue 
colours represent a low probability of being correct. The pixels 
represented by these colours are very far from the training pixels 
for all the classes and are candidates for definition of additional 
training regions. 
  
Figure 3: Classification probability map. 
Defining additional training regions for areas with a low 
probability helps to improve the result. Most of the roofs in the 
probability map with a low likelihood of being correct consist of 
heterogeneous surface materials. For instance, the material of the 
roof in a white circle is not homogeneous. Therefore, additional 
training regions are required for areas where a surface material 
varies in terms of spectral properties. Defining training regions 
for areas requiring improvement is sufficient for achieving a 
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 
classification result that represents ground features accurately. 
However, the required number of additional training regions 
depends on the scene, the material classes of interest and the 
accuracy requirements. The  discriminant analysis feature 
extraction and the ECHO classifier are applied to the whole 
research area. The processing and analysis is done for each strip. 
The result obtained for each strip is shown in Figure 4. 
  
(a) Stripl (b) Strip 2 
  
(c) Strip 3 
Figure 4: Classification results of the strips covering the research 
area. 
The output classification maps (Figure 4) fit well with 
orthophotos covering the research area in terms of geometry. 
Inspection of the corresponding classification probability maps 
shown in Figure 5 indicates that most of the classified building