The International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences. Vol. XXXVII. Part B3b. Beijing 2008 
722 
Figure 8: Left, example of Hough Transformation. Right, same 
example with ‘Forced Angles’. 
3.4 Bridging Gaps 
The last advantage that is listed here focuses on handling 
missing/incomplete data. If the grey value differences inside a 
building are bigger than the differences between grey values 
inside and grey values outside the building, the adaptive region 
growing algorithm may fail to find all relevant pixels of a 
building. For example, in Figure 9 various steps of the region 
growing process are depicted. The seed point, that describes the 
found potential building candidate, is marked with a red cross in 
the image. In step 5 the grown region overflows the building 
edge and a big area is accumulated that does not belong to the 
investigated building. Hence the region growing process is 
terminated and the found region of step 4 is used for further 
processing. 
Figure 9: Example showing how the algorithm overcomes 
missing data. 
left vertical edge. Nevertheless, when applying the proposed 
building extraction algorithm this edge is found without 
difficulties. By intersecting it with the upper horizontal edge, 
the top left comer point of the building (which is not visible at 
all in the grown region) is reconstructed without effort. This 
problem is very common, since many times parts of buildings 
are covered by shadows or vegetation, and hence some of the 
comer points are not visible. 
4. DISADVANTAGES OF THE PROPOSED 
APPROACH 
4.1 Heterogeneous Roofs 
Heterogeneity in the imagery can lead to misinterpretations and 
may mainly affect the steps of seed point determination and 
adaptive region growing, in the proposed workflow. 
One of the basic assumptions made for finding seed points 
inside buildings was the homogeneity of roof colour and roof 
texture. In practice, there are cases with roofs of a rather 
heterogeneous, unsymmetrical behaviour. This is not a problem 
as long as the radiometric variations inside the building are less 
than the difference between the interior and exterior of the 
building. 
Furthermore, also the behaviour of the adaptive region growing 
algorithm becomes unpredictable as it is unable to decide where 
to stop the growing process. In these cases additional 
constraints must be taken into consideration, like e.g. 
approximate region shape and size, in order to support the 
region growing algorithm. 
Figure 10: Example illustrating the region growing problem 
when dealing with heterogeneous roofs. 
Such a problematic case is shown in Figure 10. When looking 
at the region growing steps, one can see that the accumulated 
area first flows over the border (from step 4 onwards) and not 
before step 7 the whole building is covered by the grown region. 
Even if the algorithm were able to automatically recognize 
when the whole building is filled, the subsequent processes (i.e. 
edge and comer extraction) would fail to compute a correct 
solution. 
4.2 Shadow Effects 
Since the proposed approach is based on intensive image 
processing, effects leading to alterations in the grey value 
distribution can influence the result negatively. 
Illumination problems or shadow effects can affect the image 
radiometry so that the exploitation techniques fail or produce 
erroneous results. Especially in highly urban areas with high- 
rise buildings and skyscrapers smaller buildings may be totally 
It is obvious that the top left comer is not identifiable at all in 
this region. Even more it is also rather difficult to imagine the