Paper Title (use style: paper title)

Subtitle as needed (paper subtitle)

Authors Name/s per 1st Affiliation (Author)

Authors Name/s per 2nd Affiliation (Author)

line 1 (of Affiliation): dept. name of organization

line 2: name of organization, acronyms acceptable
line 3: City, Country
line 4: e-mail address if desired

line 1 (of Affiliation): dept. name of organization

line 2: name of organization, acronyms acceptable
line 3: City, Country
line 4: e-mail address if desired

Abstract Counting of White blood cells (WBC) & characterizing

their nucleus can provide valuable information to the doctors in
order to identify different diseases or stage of a particular disease.
The manual method is a tiresome process and has a lot of
inaccuracy. On the other hand, the machine (hematological
analyzer) based method are very expensive. Digital image
processing can be a less time consuming and cost effective method
for counting and characterizing the WBC. In order to Count or
characterize WBC by image processing, segmentation is a major
part. In this paper we proposed an algorithm to segment WBC
nucleus from microscopic images of stained peripheral blood film
during leukemia and normal condition. The proposed algorithm
involves different steps such as color space conversion, color
thresholding, filtering, marker controlled watershed and different
morphological operations. The accuracy of the result obtained is
91.85%.
Keywords- White blood cells (WBC), segmentation, leukemia,
morphological operations, color space.

I.

INTRODUCTION

Digital image processing provides us the opportunity to

extract some meaningful and useful information from an image.
In case of counting WBC and characterizing their nucleus digital
image processing provides the advantage on required time and
cost over the traditional methods. To accomplish that the key
step is the proper segmentation of WBC. Previously many
researches worked on blood cell segmentation. Krishna kumar
jha, Biplob kanti das and Himadrisekhar dutta proposed a WBC
nucleus segmentation technique depending on thresholding, edge
detection and morphological operation [1]. L. Gopinath, K.
Sivasubramaniam, M. Raguramajan and N.R. Raajan describe
WBC sectionalisation by thresholding and power law
transformation [2]. Hemant Tulsani, Rashmi Gupta and Rajiv
Kapoorshows a technique for WBC, RBC and Platelets counting
by applying two different color space conversion, morphological
operation and watershed segmentation [3]. LayzaBaldo Dorini,
Rodrigo minetto and NeucimarJeronimoLeiteproposed a
semiautomatic technique for WBC nucleus and cytoplasm
segmentation. In this work Self-Dual Multiscale Morphological
Toggle (SMMT) method has been applied for nucleus
segmentation as well as granulometric analysis and
morphological transformation has been applied for cytoplasm
segmentation [4]. Mostafa Mohamed and Behrouz Far
proposedtwo different automatic techniques for WBC nucleus
segmentation. One is based on the green component of the
image, Outs algorithm and morphological operations [5]

whereas other one is based on Gram-Schmidt vector

manipulation, minimum filter, outs algorithm and
morphological
operations[6].NeelamSinha
and
A.G.
Ramakrishnanproposed WBC segmentation depending on the
HSV equivalent of the image and K-means clustering followed
by EM-algorithm [7]. J. Hari, A. Sai Prasad and
S.KoteswaraRaoproposed their segmentation technique based on
canny edge detection, morphological operation and watershed
transformation [8]. J.M. Sharif, M.F. Miswan, M.A. Ngadi,
MdSahHj and Muhammad Mahadi bin AbdulJamilproposed an
RBC segmentation technique, involve YCbCr color space
conversion, masking, morphological operations and marker
controlled watershed algorithm [9]. B. Kumar and T. Sreenivas
segment nucleus by using teager vitality operator whereas
segment cytoplasm by using structural hustler approach [10].
P.S. Hiremath, P. Bannigidad and S. Geeta segment nucleus and
cytoplasm depending on Gradient vector flow snake algorithm
and Zack thresholding [11]. Emad A. Mohammed, Mastafa M.A.
Mohamed, Christopher Naugler and Behrouz. Far proposed a
dyad segmentation technique for nucleus and cytoplasm [12].
First they segment the WBC nucleus, then the whole WBC. And
subtraction between these two determine the cytoplasm. RGB to
binary conversion using Otsus algorithm, canny edge detection
and morphological operations are common in both segmentation
cases. After that to identify the nucleus the isolated pixels are
removed and to identify the whole WBC chessboard distance
transform, local minima elimination and watershed
transformation has been used. Min Wang and Rong Chu descrive
another WBC segmentation technique based on boundary
support vector [13]. Huey Nee Lim, MohdYusoffMashor,
NadiaturZawiyahSupardi and Rosline Hassan proposeda WBC
segmentation technique [14]. SubrajeetMohapatra, DiptiPatra
and Kundan Kumar proposed a mixed breed of rough sets and
K-means to segment WBC nucleus and cytoplasm [15]. In this
research we have proposed a robust WBC nucleus segmentation
technique from the microscopic image of stained peripheral
blood film. The segmentation technique is able to segment
nucleus during leukemia and normal condition. The main
difference between leukemia and normal condition is the no. of
WBC, during leukemia no. of WBC is increased due to presence
of blast cells. This result the WBC becomes connected with each
other. The proposed segmentation technique comprise RGB to
L*a*b* color space conversion as RGB is perceptually
nonlinear, color thresholding, average filtering to remove noise,
watershed segmentation and different kinds of morphological
operations.

II.

PROPOSED ALGORITHM

The main goal of the work is to segment the WBC from the
microscopic image of stained peripheral blood film during
leukemia and normal condition. The main difference between
these two conditions is the connectivity between WBC. During
normal condition the WBC are separated from each other and
during leukemia the WBC are connected with each other due to
the presents of blast cell (immature WBC). So the proposed
algorithm has been developed in such a manner so that the
WBC can be separated during leukemia and the separation
process does not hamper during normal condition. The
microscopic images of stained peripheral blood film has been
collected from the university degliStudi di Milano, Italy
provided by Fabio Scotti. All the images have been captured by
an optical laboratory microscope coupled with a canon power
shot G5 camera. The images are24 bit depth and resolution of
2592*1944.Like any other image processing algorithm this also
requires some processing step like color space conversion, color
thresholding, morphological operation. The entire programming
work is done by using MATLAB. Figure 1 shows the Algorithm
for WBC Nucleus segmentation.

RGB
IMAGE

RGB TO
L*a*b*
COLOR
SPACE

AVERAGE
FILTER

BINARY
IMAGE

MARKER
CONTROLLED
WATERSHED

BROKEN
WBC

COLOR
THRESHOLDING

RGB
MASK

PLATELETS
ELIMINATION

INTACT
WBC

BROKEN WBC
RECONSTRUCTION

SEGMENTED
WBC NEUCLUS

ELIMINATE BORDER
TOUCHING
MEGAKARYOCYTES TO
GET BORDER TOUCHING
WBC

ELIMINATE BORDER
NON TOUCHING
MEGAKARYOCYTES TO
GET NON BORDER
TOUCHING WBC

Figure 1Block diagram of proposed algorithm.

First, to demonstrate the algorithm, some random images are
chosen from the data set shown in figure 2. 2(a) and 2(b) are the
images of leukemia condition. Due to the leukemia they
contains high amount of WBC and WBC are connected with
each other. 2(c), 2(d), 2(e) and 2(f) are the images of normal
condition. During normal condition no of WBC are normal and
speared from each other. But 2(d), 2(e) and 2(f) are the images

which contains megakaryocytes. Megakaryocytes are the

immature stage of the platelets. Color is same as WBC and
morphological texture is bigger than mature platelets but
smaller than WBC.All the input images are in the RGB color
space. The mathematical equation of the RGB chromaticity
coordinates is[16]

R
G
B
; g
; b
RG B
RG B
RG B

(1)

Figure 2 Input images in RGB format.

Where, R, G and B denote red, green and blue. Then the
input image of RGB color space is converted into L*a*b* color
space image. It is a two-step operation. First the RGB color
space image needs to be converted into XYZwhich can be done
by the following simple matrix, after normalizing each RGB
value [17]-

X
0.412453
Y 0.212671

Z
0.019334

0.357580
0.715160
0.119193

0.180423 R
0.072169 G
0.950227 B

(2)

Where X, Y and Z denotes the response of the mix of

cones, the luminance and the response of S cone which is
similar to blue simulation. Then the XYZ color space image is
converted into L*a*b* using equation 3.0, 4.0 and 5.0 [17]-

Y
Yn

L* 116 f

16

(3)

Y

Yn

(4)

Y
Z
f

b* 200 f
Y
Z
n
n

(5)

X
a* 500 f
Xn

Where,

and

f (s) s

for s 0.008856
16
f ( s ) 7.787 s
for s 0.008856
116
1/ 3

Where L denoted the lightness, a indicates red or green and

b indicates blue or yellow. After the proper conversion of color
space a threshold value is chosen for channel L*, a* and
b*.Now the matrix of L*a*b*color space image is repeated and
all the pixel value set to zero which doesnt fall in the limit of
chosen thresholding values for channel L*, a* and b*. This

process return a L*a*b*color space image containing only the

WBC nucleus, platelets and megakaryocytes (if present).
Resultant L*a*b* color space is shown in the Figure 3.

kind of random noise. The mathematical equation of the average

filter is [9]-

w(s, t ) f ( x s, y t )
g ( x, y )
w(s, t )
a

s a

t b
a

s a

(9)

t b

Where, x=0, 1, 2,,M-1 and y=0, 1, 2,,N-1 and also a=(m1)/2 and b=(n-1)/2. In this work a 15X15 average filter is
applied. Resultant image is shown in figure 5.

Figure 3 WBC mask image in L*a*b* color space.

Now, the matrix of RGB input image is repeated and the pixel
values are set to zero which is zero in the L*a*b* color space
image mask image. Figure 4 shows the resultant RGB color
space mask image.

Figure 5 Images after applying average filter.

Then marker controlled watershed is applied to separate
connected WBC during leukemia. The marker controlled
watershed requires two images one is marker and another is
mask. The marker image is created by distance transform.
Distance transform function determines the distance from every
zero pixel to its nearest nonzero pixel. The operation is done by
the Euclidean equation [8]-

Figure 4 WBC mask image in RGB color space.

Then the RRB mask image is converted into gray scale by the
following mathematical equation-

0.2928 * R 0.5870 * G 0.1140 * B

(6)

Then the Otsus global thresholding technique is applied to

identify a global thresholding value of the gray scale image. The
mathematical equation of Otsus global thresholding technique
is [18]-

w2 (T ) v1 (T ) 12 (T ) v2 (T ) 22 (T )

DEuclidean (( x1 , y1 ), ( x2 , y 2 ))

( x1 x 2 ) 2 ( y1 y 2 ) 2
(10)

Where, x1 and x2 are the coordinates point in x-axis, y1 and y2

are the coordinates point in y-axis. The mask image is created
by computing extended minima transform. This operation filters
out small local minima and produce tiny element in the center
of cells [19]. Figure 6 shows the marker and mask image
together.

(7)

Where w is the within class variance and i is the class variance

of foreground or background and class probabilities are
estimated from histogram. And using the global thresholding
value the gray scale image is converted into binary image by the
following technique [18]-

1,
0,

I bin ( x, y )

I gray ( x, y ) T
otherwise

(8)

Where, T is the sated thresholding value for the entire image.

An average filter is applied on the binary image to eliminate any

Figure 6 Marker and mask image.

A morphological reconstruction based operation is applied
between marker and mask image which modify the marker
image such that it has only the regional minima when the mask

image is non zero. The mathematical equation of the

morphological reconstruction is [20]-

hk 1 ( hk B ) g

delete all the broken part of WBC. The resultant image is shown
in the figure 9.

(11)

Where, g is the mask image, B is the structuring element, h 1 will

be initialize as marker image and the operation will be iterate
until the hk+1 =hk. Then watershed function is applied which
identify the catchment basins on the image [21]. And returns a
label matrix where 0 represent the background and the
watershed line, 1 represent the 1 st watershed region, 2 represent
the 2nd watershed region and so on The resultant image of
watershed line is shown in figure7.

Figure 9 WBC that wasnt broken due to watershed

To return the broken WBC set based subtraction operation is
done between image No. 8 and image No. 9. This result all the
broken cells shown in figure 10.

Figure 7 Watershed lines separate the connected WBC nucleus

The connected objects are separated by setting all the pixel
values of binary image to zero which is on the label matrix of
the watershed function. But this technique breaks down the
nucleus of neutrophil, eosinophil and monocyte. For proper
segmentation reconstruction of the broken nucleus is essential.
But before that the platelets are removed by area removal
technique. Platelets are very small in size compare to WBC so a
thresholding value of area is chosen and all the objects are
deleted which has less area than the chosen thresholding value.
Result, all the platelets are eliminated from the image. But as
we know the megakaryocytes are bigger than the mature
platelets so the images still contains the megakaryocytes.
Resultant image is shown in figure 8.

Figure 10 WBC that is broken due to Watershed transformation.

In order to reconstruct the broken WBC morphological closing
operation is applied on the image No. 10. To do the closing
operation a disk shape structuring element is chosen with radius
5. The mathematical equation of morphological closing is [20]-

A B ( A B)B

(12)

The resultant is image shown in figure 11.

Figure 11 Broken WBC is reconstructed.

Figure 8 Images after eliminating Platelets.
To reconstruct the broken WBC we need to distinguish between
intact WBC (doesnt break down by applying marker controlled
watershed) and broken WBC. The broken parts of the WBC are
smaller in size so a thresholding value of area is chosen to

The image contains megakaryocytes. The megakaryocytes are

smaller in size compare to WBC. So those can be eliminated
from the image by area removal technique. We cannot eliminate
megakaryocytes by choosing a single thresholding value
because that will eliminate some of the border touching WBC.
To overcome this problem first the image (figure 11) is
segmented into two sub images. One contains all the border

touching cells another contains all the border non touching cells.
Than two different thresholding values of area is chosen to
eliminate the border touching megakaryocytes and border non
touching megakaryocytes. Figure 12 and figure 13 shows the
resultant image after eliminating border touching and border
non touching megakaryocytes.

Figure 12 Images after eliminating border touching

megakaryocytes

Figure 13 Images after eliminating border non touching

megakaryocytes
So, all the WBC present among figure 9, figure 12 and figure
13. So combination of these three images produces an identical
binary image of the input image containing only the WBC.
Resultant image is shown in figure 14.

Figure 14 OR operations between image no. 9, 12 and 13.

III.

RESULTS AND DISCUSSION

These chapters discuss the implementation of the proposed

algorithm to get significant result for WBC nucleus
segmentation. As mention in the previous chapter the proposed
algorithm involves different types of color space conversion,
color thresholding, marker controlled watershed, filtering and
morphological operations. First the input RGB color space image
is converted into L*a*b* color space image and then

thresholding value is chosen for channel L*, a* and b*. By using

the thresholding values for the channels a RGB image of WBC
nucleus mask is created. After that by using Outs algorithm the
RGB mask image is converted into binary image. Then an
average filter is applied on the binary image to eliminate any
kind of random noise. The marker controlled watershed is
applied to separate the connected WBC during leukemia
condition. The marker controlled watershed requires a marker
and a mask image, morphological reconstruction and watershed
function. The marker image is created is created by applying
distance transform function and the mask image is created by
applying skeleton and pruning function. Then a morphological
reconstruction operation is applied between marker and mask
image and after that watershed function is applied on the
reconstructed image. The watershed function separated all the
connected WBC during leukemia condition but the disadvantage
of applying the marker controlled watershed is, it breakdown the
nucleus of neutrophil, eosinophil and monocyte into multiple
segment. For the proper segmentation of the WBC nucleus the
broken nucleus must need to be reconstructed. But before going
to the reconstruction part the platelets are eliminated from the
image by applying area removal technique. As the broken WBC
are smaller than the intact WBC so a thresholding value is
chosento distinguish between broken WBC and intact WBC.
Depending on that thresholding value two separated image is
created. One contain all the intact WBC another contains all the
broken WBC. Then morphological closing operation is applied
on the image which contains the broken WBC nucleus with a
disk shape structuring element of radius 5. That reconstructs all
the broken WBC. Then this image is again segmented into two
sub images. One contains all the border touching WBC nucleus
another contains all the border non touching WBC nucleus. Two
thresholding value is chosen to eliminate the border touching and
border non touching megakaryocytes. At last the combination of
the image of intact WBC, border touching WBC and border none
touching produce the properly segmented nucleus of the input
image.
IV.

CONCLUSION

No. of WBC in the blood serum can provide valuable

information about prescient condition whereas the morphology
of WBC nucleus enables us to classify the five basic classes of
WBC and blast cells. But for counting and classification the key
requirement is the proper segmentation of WBC nucleus. The
proposed framework is to segment the WBC nucleus during
leukemia and normal condition by using color space conversion,
color thresholding, watershed segmentation and morphological
operation. RGB to L*a*b* color space conversion and color
thresholding provides the segmented image of WBC nucleus,
platelets and megakaryocytes (if present). Average filter is
applied to remove any kind of random noise. The marker
controlled watershed is applied to separate the connected WBC
nucleus during leukemia condition. The platelets and
megakaryocytes have been removed by area removal technique
and morphological closing operation is applied to reconstruct
the broken WBC due to marker controlled watershed function.
The main goal of the research is to segment the WBC nucleus
from the microscopic image of stained peripheral blood film
and the proposed algorithm shows 91.85% accuracy. The
segmented image further can be used for WBC counting,
nucleus characterization and classify the member of WBC.

There is also opportunity to improve the segmentation technique

during different diseases.

[2]
[3]

ACKNOWLEDGMENT
The authors are would like to thank Fabio Scotti for
providing the dataset and Dr. Abdur razzak to verify the result by
visual inspection.

[4]
[5]
[6]

REFERENCES
[1]

G. Eason, B. Noble, and I. N. Sneddon, On certain integrals of LipschitzHankel type involving products of Bessel functions, Phil. Trans. Roy.
Soc. London, vol. A247, pp. 529551, April 1955. (references

[7]

J. Clerk Maxwell, A Treatise on Electricity and Magnetism, 3rd ed., vol. 2.

Oxford: Clarendon, 1892, pp.6873.
I. S. Jacobs and C. P. Bean, Fine particles, thin films and exchange
anisotropy, in Magnetism, vol. III, G. T. Rado and H. Suhl, Eds. New
York: Academic, 1963, pp. 271350.
K. Elissa, Title of paper if known, unpublished.
R. Nicole, Title of paper with only first word capitalized, J. Name Stand.
Abbrev., in press.
Y. Yorozu, M. Hirano, K. Oka, and Y. Tagawa, Electron spectroscopy
studies on magneto-optical media and plastic substrate interface, IEEE
Transl. J. Magn. Japan, vol. 2, pp. 740741, August 1987 [Digests 9th
Annual Conf. Magnetics Japan, p. 301, 1982].
M. Young, The Technical Writer's Handbook. Mill Valley, CA: University
Science, 1989.