Biomedical Engineering Reference
In-Depth Information
Undoubtedly, local thresholding performs better than Global thresholding in
tackling the problem of uneven illumination. However, there are difficulties in
applying the technique effectively in hand bone segmentation due to the problems
as follow:
1. The problem arises from making the assumption that no intensity overlapping
between targeted object and background.
2. The size of each sub-image is difficult to determine. If the size is smaller or
larger than it should be, then the result might be even more inferior to using
Global thresholding.
3. The size of the sub-images is globally set and is fixed throughout the entire
image. Some regions need smaller sub-image whereas some regions need larger
sub-image in local thresholding to optimize the segmentation and the computa-
tional efficiency.
4. The number of thresholds needed in each sub-image is difficult to determine.
5. The computational cost is higher in comparison with Global thresholding.
The threshold values are difficult to be set manually as the number of sub-
images increases [
10
]. In Global thresholding as well, the threshold value need to
be correctly set in order to optimize the result. Single global threshold using is set
human inspection. However, comparing with multiple thresholding or local thresh-
olding, automated thresholding is more suitable to decrease repetitive threshold
setting by human which is subjective and yet time-consuming.
2.3.3 Dynamic Thresholding
In Global thresholding, each pixel is compared with the global threshold; in local
thresholding, each pixel in sub-image is compared with each local threshold which
is computed from each sub-image; in dynamic thresholding, each pixel is com-
pared with each dynamic threshold which is computed from sliding a kernel over
the input image [
43
]. One of the popular of dynamic thresholding methods is
Nilback method [
45
].
Niblack method: A
WXW
kernel moves in the input image from pixel to pixel.
Both mean,
m
(
x
,
y
)
and standard deviation,
s
(
x
,
y
)
in each position of kernel are
defined as follow:
i
=
x
+
2
j
=
y
+
2
1
w
2
m
(
x
,
y
) =
f
(
x
,
y
)
(2.3)
i
=
x
−
2
j
=
y
−
2
i
=
x
+
2
j
=
y
+
2
1
w
2
s
(
x
,
y
) =
(
f
(
x
,
y
) −
m
(
x
,
y
))
2
(2.4)
i
=
x
−
2
j
=
y
−
2
Search WWH ::
Custom Search