ibul 2004
e is the
ssed by
to the
sensitive
1. So the
as:
lgorithm
structure
duced by
peration.
a lighter
| is taken
| belongs
in a
he same
between
to soften
iting (11)
where o is a threshold value.
3. Results and discussion
In this paper the IKONOS image in Shenzhen of China with the
spatial resolution of 4*4 (Fig.1) was selected. The objects in the
image are labeled with different morphology features which are
the sizes of the structure elements. The image was segmented
and coded at the same time according to morphology feature.
The proposed segmentation algorithm was implemented by
VC++ programming. The primary objects in an image include
buildings, streets, green belts, rivers and ponds.
In this method, the selection of structure elements is very
important. The structure elements with different shapes and
sizes should generate different results. The result obtained by
octagonal structure elements is better than that by square
structure elements which have been proved by experiments by
others. In this paper, a structure element sequence formed by
two octagonal structure elements denoted as N, and N,,
where
N, = N, © N, (13)
The diameters of the two structure elements are 7 and 13
pixels and O is 5. Five resultant images could be obtained after
Fig.1 IKONOS image in Shenzhen, China
49
International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Vol XXXV, Part B-YF. Istanbul 2004
segmentation (figures 2-6), and the black part includes pixels
with same morphological features. Figs. 2 and 3 show the peaks
with different features. In Fig.2, the light objects smaller
than N, mainly consist of buildings, while the light objects
g.3. Some
greater than. N, but smaller than N, appear in Fig
pixels belonging to plains, which have a lower response to the
structure elements in a given size, can be found in Fig.4. The
objects included in Fig.4 are related to the selection of threshold
value G , a greater 0 means more objects will fall into this
image. As rivers and ponds are homogeneous in a large area that
are bigger than the structure elements in given sizes, they have a
lower response to the structure elements. In general, if the size
of the given structure element is great enough (larger than these
objects), the objects will belong to valley images because these
objects are very dark (local minimum) in the image. Figs. 5 & 6
consist of valleys with different morphological features. Some
small and dark objects are mainly included in Fig.5, such as
grass plots and shadows between buildings and some streets
within residential areas. Grass land and greenbelts located on
both sides of roads in bigger areas are shown in Fig. 6.
i te
QU Du RY
UE AS vai
I dea SEA
Fig.2 Light objects sensible to structure element N,
#