International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Vol XXXV, Part B7. Istanbul 2004
2. METHODOLOGY
2.1 Residential area image in high-resolution image
Residential area is a spatial entity consists of buildings, road
inside the buildings, green ground, activity place and factory
and it has some structure, function and spatial state. There are
two different kinds of residential area: urban residential area
and country residential area. The urban residential area mainly
includes buildings, road network, green and vacant ground. The
country residential area mainly includes buildings, green and
vacant ground, just big country residential area has street. And
the density and size of buildings is smaller than the urban
residential area. In the high-resolution image, the buildings in
the residential area are mixed up a single or cluster light pixels
and a few dark pixels. Objects such as the roads, ditch, dykes,
dames and wet ground along the river and lake in the residential
area also are light pixels while the water body, vegetation and
nudity ground are dark pixels. The residential area belongs to
the area object with complex structure and there are some
difficulties to describe the residential area image feature
through the texture analytical method. The texture character of
different residential area has great difference with the change of
background and it’s very difficult to obtain the self-adaptive
segmentation threshold of the residential area and background.
And the interior composing of the residential area is very
complex and the texture character changes are huge for
example the urban residential area may include wide range
virescence square with extrusive feature but obvious difference
of the background structure which make it very difficult to
extract the residential area through one time region increasing.
In the other side, the texture characters of road and residential
area in the image are quite similar and promiscuous thus it’s
very difficult to distinguish the road and residential area
efficiently. Another difficulty is that the change of the
residential area boundary is not obvious and quite similar to the
background, which directly affects the precision of residential
area extraction. The difference of maximize and minimize grey
value is large and the grey distribution range is overlapped to
the other object in the high-resolution image. In summary, most
of residential area pixels are mixed pixels and less are pure
pixels, as Figure 1 shown.
Histgram & Statis
~ Histgram Info — —
2.2 Texture Feature Analysis of Residential Images
For the image segmentation based on texture analysis, the
selected texture feature should make some image statistic
values such as deviation change obviously after the image
texture transformation. There are some classic texture feature
description methods such as the deviation texture analysis,
Fourier texture description and grey symbiotic matrix feature
sessi as var (x (2) and (3).
w/2
v, (i, == lg jm, (i, Mea
w/21=-w/2
k=
w/2
2M
k--w/2
w/2
2.80 +k, j+1)
21=—-w/
m. (1, j) = (2)
Where vg(i,j)is the deviation feature of pixel (i,j), and the grey
value of pixel (i,j) is g(i,j). The calculation window is Wow
and mg is the average grey value of the calculation window.
Plu,v)= |F (us vl (3)
p(r)= oS Pr,o0 (4)
0-0
Where p(r)} is the sum of energy in the loop region in the
Fourier frequency domain and can be taken as the Fourier
texture feature and (u,v) is the Fourier transformation of
image f G. JY
W,, +3 s PAH
yy
Wy is the grey symbiotic matrix feature and P(i,j) À is the grey
symbiotic matrix.
Figure 2 shows the high-resolution original image with
residential area and corresponding deviation, Fourier and grey
symbiotic matrix texture feature images.
(5)
Histgram & Statistie
-Histgram Info — — — -
tic
Max PS
— Min 0
Fig 1(a)
Fig | Character Analyze of Resident Area in High Resolution Image
Figure 1(b) and 1(c) are the histogram of the original image 2 and
residential image selected from the test image manually. They
have the similar distribution. And the statistic values are almost
the same( the difference of average is 2.169, the difference of
deviation is 4.02 and the difference of entropy is only 0.026). It
is very difficult to extract residential area from the test image
based on histogram feature in spatial domain. Thus, the
automatic extraction of the residential area has many difficulties
related to the current research progress in this field.
Entropy yi 7.42522 |
| |
| i
| | à diio.
Std Dew 45.8773 | Max 255 Std Dev 49.8967
Mean 127.103 Min 0 Mean 124.934
Entropy H[(x]- 7.45187
Fig 1b)
Fig 1(c)
2b Deviation Texture Image
2a Original Image
1122
