International Archives of Photogrammetry and Remote Sensing, Vol. 32, Part 7-4-3 W6, Valladolid, Spain, 3-4 June, 1999 
urban environment the classification results can be improved 
considerably, if the normalized DSM representing the local 
height above terrain is introduced as an additional database 
during classification. 
□ 
Buildings 
Roads 
Trees 
Grass-covered 
Fig. 8. Maximum likelihood classification based on CIR 
imagery and laser data. 
3.4. Further Improvement by Shadow Analysis 
One problem, when classifying areas with rough surfaces and 
steep slopes as they can occur in urban areas, is the change of 
spectral reflectance due to the presence of shadow areas. In 
order to avoid these problems, the shadow areas can be derived 
automatically based on the given DSM and used as an 
additional source of information during the classification 
process. The result of the automatic generation of shadow areas 
is given in Figure 9. 
Fig. 9. Shadow areas as derived from DSM. 
For the automatic detection of shadow areas, the local height, 
which is provided by the laser DSM as well as the elevation and 
azimuth of the sun at the time of image acquisition is required. 
The elevation and azimuth of the sun can either be determined 
manually, by an interactive measurement of an edge of a 
shadowed area in the image and the corresponding object height 
in the normalized DSM, or derived automatically from the 
geographical latitude and longitude of the captured area and the 
time of image acquisition. 
The improvement of the classification when using predefined 
shadow areas is demonstrated in Figures 11 and 12, using the 
test area depicted in Figure 10. In Figure 11, the 
misclassifications in shadow areas are clearly visible. These 
misclassifications can be avoided, if the class shadow is 
determined in advance based on the analysis of the given DSM 
and excluded from further classification like it is demonstrated 
in Figure 12. 
Fig. 10. Ortho-image. 
Buildings 
Grass-covered 
Roads 
Trees 
Not classified 
Fig. 11. Classification without using shadow areas. 
In order to avoid the shadow class in the final result, the 
approach can be further refined by splitting each of the landuse 
classes into one separate class for shadow areas and one landuse 
class for non-shadow areas. This leads to a significant 
improvement of the results because the pixels in shadow areas 
have completely different spectral characteristics than the ones