Comparing with the two algorithms, our algorithm
considers both color and shadow in a brand new
framework.
The pre-segmentation based on SLIC remove
the influence of noises and enhance the robustness to the
local textures.
The shadow-based approach is capable of
separating buildings and road spontaneously. Though the
iterations of the boundaries based on the level set model,
all contours of building areas will precisely converge to
the real edges and some false alarmed regions will be
removed.
All these advances assure better performance of
building detection and segmentation.
In some cases, however, roads and
buildings resemble in both color and shape.
Without the
prior knowledge like shadows, it is impossible to
distinguish a building from its neighboring distracters
with the same color and shape, so that the algorithm tends
to recognize roads and squares as buildings.