Graphics Reference
In-Depth Information
image segmentation, those segmented images should have some parts of the pixels to
overlap. The size of the overlap area requires twice as much as the size of the biggest
extracted target.
The size of the original image is
A
×
B
and the size of the biggest extracted target
bSize
×
bSize
is not bigger than
. If the original image is segmented into 4 parts, the
A
B
(
)
(
)
+
bSize
×
+
bSize
size of images after the segmentation would be
. Parallel
2
2
processing is used to process those segmentation images in order to improve
processing speed.
2.2
Image Enhancement
The purpose of the image enhancement is to highlight useful information of the image
and enlarge the differences between different objects. It makes a good foundation for
the following steps. The simple image enhancement method, which is commonly
used, is the linear gray level transformation, histogram equalization, etc.
The linear gray level transformation is based on linear formula. Suppose that the
gray-scale range of image
f
(
a
,
b
)
[
f
,
f
]
is
, the gray-scale range of output
min
max
g
(
a
,
b
)
extends to
[0,255]
. The transformation formula is:
image
f
(
x
,
y
)
−
f
g
(
x
,
y
)
=
min
×
255
(
2
)
f
−
f
max
min
The histogram equalization is based on a statistical theory. We suppose that an image
has n pixels and l different gray levels. The original image's gray-scale is represented
by r. The gray-scale of image, which is processed by histogram equalization algo-
rithm, is s.
n
equals to the number of pixels which gray-scale is
r
, and then the
appearance frequency of the kth gray-scale is expressed as:
n
P
(
r
)
=
k
(
3
)
r
k
n
0
≤
r
≤
1
k
=
0
,...,
l
−
1
In the expression,
After the equalization, each pixel's gray value is:
k
=
s
=
P
(
r
)
(
4
)
k
r
j
j
0
0
≤
j
r
≤
1
k
=
0
,...,
l
−
1
In the expression,
Search WWH ::
Custom Search