Thresholding of Medical Images - Medical Image Processing: Advanced Fuzzy Set Theoretic Techniques - page 110

Thresholding is a transformation of an input image A into a segmented out-
put image B as follows:

b = 1

0

for
for

a T
aT

≥

ij ij

= ≤

ij

where
T is the threshold
b ij = 1, for the image pixels that belong to the object class
b ij = 0, for the image pixels that belong to the background class

However, the selection of threshold is very crucial in segmenting an image.
If the threshold is not selected properly, proper segmentation is not obtained.
A threshold may be global or local. In global thresholding, the image is con-
sidered as a whole and the threshold value is constant throughout the image.
Depending on the modality of the histogram, threshold levels may be single,
double or multiple. In single threshold, the image contains an object against
a background. The histogram of such type of image contains two peaks -
one for the object and the other for the background - and the threshold value
lies in the valley of the two peaks. In double threshold, the image consists of
two objects of different grey levels and the histogram contains three peaks -
two for the objects and one for the background. The thresholds lie in the two
valleys of the three peaks.
In local threshold, the threshold varies throughout the image. In some
images, different objects of different grey levels may be presented, and in
that, global threshold will not work. The image is divided into subregions,
and the threshold is computed for each region. This is useful in processing
medical images where the grey levels are not uniformly distributed.

6.2 Threshold Detection Methods
There are six methods of thresholding, namely, (1) global thresholding,
(2) iterative thresholding, (3) optimal thresholding, (4) local thresholding,
(5) the Chow and Kaneko approach for adaptive threshold and (6) multispec-
tral thresholding.

6.2.1 Global Thresholding

1. P-tile thresholding : On a printed sheet text, if the area covered by the
text is 1/ p of the area of the image, then the threshold, T , chosen is
such that 1/ p of the image area has grey level less than the thresh-
old, T, T , and the rest of the values are larger than the threshold, T .