A comparison of image segmentation techniques, otsu and watershed for x ray images

IJRET: International Journal of Research in Engineering and Technology eISSN: 2319-1163 | pISSN: 2321-7308
_______________________________________________________________________________________
Volume: 04 Issue: 04 | Apr-2015, Available @ http://www.ijret.org 720
A COMPARISON OF IMAGE SEGMENTATION TECHNIQUES, OTSU
AND WATERSHED FOR X-RAY IMAGES
P. Manju Bhargavi1
, V. Sai Kiran Mayee2
, T. Manaswini3
, S. Manvitha4
1
Department of Biomedical Engineering, GRIET, Hyderabad
2
3
4
Abstract
The most dangerous and rapidly spreading disease in the world is Tuberculosis. In the investigating for suspected tuberculosis
(TB), chest radiography is the only key techniques of diagnosis based on the medical imaging So, Computer aided diagnosis
(CAD) has been popular and many researchers are interested in this research areas and different approaches have been proposed
for the TB detection. Image segmentation plays a great importance in most medical imaging, by extracting the anatomical
structures from images. There exist many image segmentation techniques in the literature, each of them having their own
advantages and disadvantages. The aim of X-ray segmentation is to subdivide the image in different portions, so that it can help
during the study the structure of the bone, for the detection of disorder. The goal of this paper is to review the most important
image segmentation methods starting from a data base composed by real X-ray images.
Keywords— chest radiography, computer aided diagnosis, image segmentation, anatomical structures, real X-rays.
---------------------------------------------------------------------***--------------------------------------------------------------------
1. INTRODUCTION
Tuberculosis (TB) is an infectious disease over the world
population. It is caused by bacteria which is known as
Mycobacterium tuberculosis and mostly affected to the
lungs of the human beings. In order to detect Tuberculosis,
we have set of steps involved [1].
A CAD system usually applies a series of pre-processing
steps to an input image. The pre-processing step is
performed to enhance the image quality so that objects of
interest become more evident. The quality of the pre-
processing step strongly affects the performance of the
subsequent processing steps. The pre-processing steps for
X-ray screening are contrast enhancement and lung
boundary detection. The software implementation of these
steps is shown in this paper.
When analyzing objects in images, it is necessary to
distinguish the objects of interest from the background. This
task can be realized through segmentation. Image
segmentation is one of the most challenging issues in image
processing domain [2] and it has been an active research
area in the last years. Through segmentation we aim to
fragment the image in a series of regions, based on the
attributes of the image that are approximately constant in
each region, but differ significantly from a region to another.
Segmentation aims to extract useful information from
images in medical imaging applications as well Medical
images segmentation [2] has also found applications in
studying the anatomical structure, fractures localizations or
tumors, diagnosis and treatment planning, computer-
integrated surgery, tissue classification, or tumor volume
estimation. They are also affected by noise, artifacts or
spatial aliasing so that the boundaries of the regions of
interest to become indistinct. X-rays are having various
orientations, resolutions, and luminous intensities,
depending on the X-ray equipment, which influences the
quality of the segmentation result [3]. Unlike many other
medical imaging modalities, bone regions often overlap with
other organs or bones and the joints between bones in X-
rays. Several segmentation techniques have been developed
and reported in the literature. However, a perfect method,
which is universally applicable to all kind of images, does
not exist. This paper focuses on the comparison of several
already existing image segmentation techniques in the case
of X-ray images.
2. WORK
2.1 Pre-Processing:
The objective of the image pre-processing is not only to
improve the image quality but also to reduce the undesired
portion from the background of the images. Most pre-
processing methods apply the intensity value of
neighborhood pixel for obtaining the brightness intensity
value of the input images. The reasons for the image pre-
processing phase are:
i. In order to meet the requirements of the physician
for the image quality improvement.
ii. Edge detection –To detect the edges, Sobel
operator is discussed in this paper.
iii. Contrast enhancement - To enhance the contrast at
the region of interest, histogram equalization is
performed on the image. Histogram analysis
increases the contrast in low-contrast regions by
spreading out the most frequent intensity levels.

_______________________________________________________________________________________
iv. Preparing the data optimally for post-processing
steps. The process is described in figure 1.
Fig 1: Flowchart of pro-processing step
Fig 2: Output of pre-processing step
In figure 2, the first image is the input which is a gray image
converted from RGB image. The second is histogram
equalized image and the last is the edge detected image
using a Sobel operator.
2.2 Segmentation Techniques
Image segmentation is very important step to extract good
quality features for classification. The objective of image
segmentation is to cluster the pixels having same intensity
value from the whole image regions, separating regions of
desired part of the original image, hiding the undesired
region. Image segmentation step is applied in many image
pre-processing areas such as editing, or query from image
database, object occlusion, recognition of object, estimation
of the boundary and image compression.
In general, algorithms for image segmentation are based on
two fundamental properties of intensity values which are
discontinuity and similarity. Discontinuity property does the
partition of the image based on the changes in sharp
intensity value. Similarity property does the partition of the
image into regions that are alike to the specified criteria.
In chest radiographs, segmentation process must be done
accurately for the medical image analysis task which is
especially for computer-aided diagnosis. For example, lung
nodules are determined according to the different
measurements of nodules size.
Different kinds of approaches have been discussed in the
lung segmentation step. Various approaches for lung
segmentation are presented which includes:
2.2.1 Watershed Segmentation
Watershed segmentation (watershed transform) is a region-
based method [4] under the classical technique of
segmentation. This method is based on the grayscale
mathematical morphology [5] and it is used for multi
component images. Intuitively, the watershed algorithm can
be thought of as a landscape that is flooded by water. At
each point, the height of the landscape represents the pixel's
intensity. The watershed transform computes the image
regions which represent the basins and region boundaries
(the ridgelines). The image gradient is used as input of the
transform, such that the basin limits are situated at high
gradient points. The steps involved in this method are shown
in figure 3.
A Watershed segmentation process starts at some regional
minima Mi value which locates the lowest points of the area
into where the water flows. By measuring appropriate
distance, the area is separated into regions Ωi which has
growth from the corresponding minimum Mi by adding to
Ωi, repeatedly, unlabeled points on the outer boundary of Ωi.
A point is added to region Ωi If its distance from the region
is smaller than those from other regions. The process is
repeated until no remaining unlabeled points are those of the
watershed line. In implementation, in order to obtain a thin-
watershed line, a point is added to the region Ωi even when
its distance from the region equals those from some other
regions. Hence, there is no point that belongs to the
watershed line.
This type of segmentation method is very simple and
intuitive and has good properties, which make it useful for
many image segmentation applications. However, it has
other disadvantages such as: over-segmentation and poor at
detecting thin structures and structures with low signal-to-
noise ratio.

_______________________________________________________________________________________
Fig 3: Flowchart of Watershed segmentation
Fig 4: Output of reconstruction technique
In figure 4, the first image is opening and closing technique
by reconstruction and the next image is the regional maxima
of opening and closing technique by reconstruction.
Fig 5: Output of thresholding technique
In figure 5, the first image is by applying a threshold to the
opening-closing reconstruction technique and the next
image is the watershed ridge line.
Fig 6: Output of watershed segmentation
In figure 6, markers and object boundaries are superimposed
on an original image and the last image is the coloured
watershed label matrix.
2.2.2 Otsu Thresholding
Thresholding is one of the most simple segmentation
techniques [2], involves thresholding the intensity of the
image. In the case of global thresholding, only one threshold
value is selected for the entire image. Global methods are
based on the fact that the image has a bimodal histogram.
The region of interest can be separated from the background
by comparing the intensity of each pixel in the image with a
threshold. Some pixels, whose intensity values are greater
than the threshold, are classified as being part of group A -
object of interest (with an intensity value of 1), and the rest
of the pixels as being part of group B-background (with an
intensity value of 0). The following assumptions are applied
in Otsu method:

_______________________________________________________________________________________
i. The image has a bimodal histogram.
ii. Neither use of spatial coherence nor any other
object structure.
iii. Intensity values are stationary statistics but they can
be changed to locally adaptive.
iv. Intensities are implicitly uniform illumination thus
the brightness behavior of the bimodal arises from
the object appearance differences only.
Global methods are simple and fast, but are suitable only for
images with bimodal intensity distribution (probability
distribution with two different modes). Another factor that
affects the performance of thresholding is the unequal
illumination in the image. In addition, global methods are
not useful for multichannel images, since only two classes
are generated. Figure 7 shows the procedure involved in
otsu thresholding.
Fig 7: Flowchart of Otsu thresholding technique
Fig 8: Output of Otsu thresholding technique
In figure 8, the first image is the original image that is given
as the input which is actually pre-processed and the next
three images are obtained by setting different threshold
values to the input image.
3. CONCLUSION
We described a series of approaches that have been
published in the recent literature, concerning medical
images segmentation. We have provided an overview
regarding the implementation of each segmentation method,
highlighting the advantages and disadvantages of each
method. The evaluation of these segmentation techniques
can be done, in terms of: performance, computational
complexity and sensitivity to noise. The most accurate
techniques are the most complex and time consuming. We
realized the classification starting with the most simple and
fast methods and we increased the computational
complexity and the processing time with each presented
method.
As future work, we intend to combine a classical image
segmentation technique for an initial segmentation and then
to apply a deformable model in order to increase the
segmentation accuracy. We intend to calculate the region of
interest (ROI) on the segmented regions and then apply a
classifier technique in order to detect the Tuberculosis and if
present, we intend to find the intensity of its spread to
different regions in the thoracic cavity.
REFERENCES:
[1] S. V.Zharkova, EGSO, 29 october 2002. [Online].
Available:
https://www.google.co.in/url?sa=t&rct=j&q=&esrc=s
&source=web&cd=2&cad=rja&uact=8&ved=0CCQ
QFjAB&url=https%3A%2F%2Fblue-sea-697d.quartiers047.workers.dev%3A443%2Fhttp%2Fwww.researchgate.ne
t%2Fprofile%2FValentina_Zharkova%2Fpublication
%2F250738597_Survey_of_Image_Processing_Tech
niques%2Flinks%2F5488d3ae0cf268d28. [Accessed
february 2015].
[2] G.Dougherty, "Medical Image Processing
Techniques and Applicatios," Springer2011.
[3] D. W. k. D.Feng, july 2006. [Online]. Available:
http://www.comp.nus.edu.sg/~leowwk/thesis/dingfen
g-proposal.pdf. [Accessed march 2015].
[4] S. V. Kasmir Raja, A. Shaik Abdul Khadir, and S. S.
Riaz Ahamed, "Moving toward region-based image
segmentation techniques: a study", Journal of
Theoretical and Applied Information Technology,
5:81-87, 2009.
[5] S. S.K.Mahendran, "Enhanced automatic X-ray bone
image segmentation using wavelets and
morphological operators," 2011.
[6] M. W.Burgern, "Principles of Digital Image
Processing Fundamental Techniques," Springer,
2009.

A comparison of image segmentation techniques, otsu and watershed for x ray images

More Related Content

What's hot (20)

Viewers also liked (18)

Similar to A comparison of image segmentation techniques, otsu and watershed for x ray images (20)

More from eSAT Journals (20)

Recently uploaded (20)

A comparison of image segmentation techniques, otsu and watershed for x ray images