Biomedical Engineering Reference
In-Depth Information
perception towards image quality. Besides, the process itself is extremely time-
consuming if the operation involves multiple thresholds. Therefore, it is not prac-
tical to determine the threshold value of a large number of images. In short, the
manually determined value is not effective.
For automated thresholding method, various methods exist: the simplest
method is to utilize the image statistics such as mean, median (second quartile)
and first quartile, third quartile, to act as threshold value [
46
]: this method per-
forms only relatively well in an image that free of noises; the reason is that the
noises in the image have influenced the statistic of the image. Typically, if the
mean of an image used as threshold value, then it can separate a typical image with
object brighter than background into two components; however, while noises exist,
the noises have altered the nature that the pixels with intensity greater than mean
are belonged to the object. Besides, this kind of thresholding method assumes that
the object and the background are themselves homogenous. In other words, the
object is a group of pixels containing similar pixel intensity; the background is a
group of pixels with similar intensity. This assumption has serious limitation espe-
cially in medical image segmentation where the targeted objects such as organs
and bone are not inherently homogenous. Besides using simple afore-mentioned
statistic in input image, there are other methods to choose the threshold value. In
next paragraph, different type of automated thresholding techniques are explored
and studied. The Global thresholding and dynamic thresholding using automated
threshold selection on hand bone radiographs are shown in Figs.
2.3
and
2.4
.
Fig. 2.3
Segmentation
result using Global threshold
value/values of (
a
) mean
(
b
) median (
c
) between first
quartile and third quartile (
d
)
below first quartile and above
third quartile
(a)
(b)
(c)
(d)
Search WWH ::
Custom Search