ROUGH SEGMENTATION OF BUILDINGS
There are two challenges pop out when we try to
distinguish buildings from other categories:
1) the similarity of color features between buildings, roads and squares;
2) the variety of building's appearance features.
As illustrated in Figure 2, we sample 30 square patches
from the three images in the top row. The 15 patches
inside the red square are extracted from the building areas,
and those in the blue one are sampled in the road areas.
Obviously, it is difficult to recognize buildings by the
local color features independently. The local
neighborhood and the prior knowledge about the shadows
must be considered.
It can be noticed that boundaries between buildings
and roads are often detectable, which indicates that the
regional growth segmentation algorithm possibly works
for this situation. Besides, the existence of shadows may
provide a useful prediction for the locations of buildings.
By combining the two factors, we present a new
algorithm to implement the rough segmentation of
buildings in the complex RSIs. The flow chart of the
algorithm is shown in Figure 3.