Graphics Reference
In-Depth Information
3
Suspected Target Extraction
Targets mostly hide in the lawn or forest. In remote sensing image, the distribution of
the gray value in those regions is uneven, but the gray value distribution of target area
is relatively homogeneous. This feature can be used to remove the useless back-
ground. Furthermore, the number of pixels in the area of suspected target is much
larger than the number of pixels in other useless objects' area; for this reason, those
areas can be removed along with background.
3.1
Binarization
Binarization is defined as a means to transform the color information into binary im-
age. Assume that the threshold value is Threshold , the input image is
f
(
x
,
y
)
and
g
(
x
,
y
)
the output image is
; therefore the binarization can be expressed by the fol-
lowing formula:
1
f
(
x
,
y
)
Threshold
g
(
x
,
y
)
=
( 5 )
0
f
(
x
,
y
)
<
Threshold
The key function of the binarization is to select an appropriate threshold since a good
threshold can reserve useful image information and eliminate the distracting informa-
tion as much as possible. The fixed threshold method and dynamic threshold method
are two common ways to obtain threshold. The fixed threshold method uses the mean
value of all pixels in the image. Although this method is simple and the execution
speed is fast, the mistake rate is rather high, since it only considers the overall image
and ignores the local information. Here is a kind of dynamic threshold method based
on double window.
We suppose that the input image is
f
(
x
,
y
)
g
(
x
,
y
)
and the output image is
. We
W
W (
W
>
W
set that two windows' size respectively are
). For a pix-
1
2
af , the average gray value of the pixels for the two windows, the central
point of which is
(
,
b
)
el
Ave . The smaller value is thus considered
as the threshold. The formula is as the following:
f
(
a
,
b
)
Ave and
, are
1
2
W
(
x
,
y
)
(
x
,
y
)
Ave
=
( 6 )
1
1
W
×
W
1
1
W
(
x
,
y
)
(
x
,
y
)
Ave
=
( 7 )
2
2
W
×
W
2
2
 
Search WWH ::




Custom Search