WSEAS TRANSACTIONS on BIOLOGY

and BIOMEDICINE Aimi Salihah Abdul-Nasir, Mohd Yusoff Mashor, Zeehaida Mohamed

Colour Image Segmentation Approach for Detection of Malaria

Parasites Using Various Colour Models and k-Means Clustering
AIMI SALIHAH ABDUL-NASIR1, MOHD YUSOFF MASHOR2, ZEEHAIDA MOHAMED3
1,2
Electronic & Biomedical Intelligent Systems (EBItS) Research Group
School of Mechatronic Engineering
Universiti Malaysia Perlis
Campus Pauh Putra, 02600 Pauh, Perlis
MALAYSIA
3
Department of Medical Microbiology & Parasitology, School of Medical Sciences
Health Campus, Universiti Sains Malaysia
16150 Kubang Kerian, Kelantan
MALAYSIA
Email: [email protected], [email protected], [email protected]

Abstract: - Malaria is a serious global health problem that is responsible for nearly one million deaths each
year. With the large number of cases diagnosed over the year, rapid detection and accurate diagnosis of malaria
infection which facilitates prompt treatment are essential to control malaria. This paper presents a colour image
segmentation approach for detection of malaria parasites that has been applied on malaria images of P. vivax
species. In order to obtain the segmented red blood cells infected with malaria parasites, the images are first
enhanced by using partial contrast stretching. Then, an unsupervised segmentation technique namely k-means
clustering has been used to segment the infected cell from the background. Different colour components of
RGB, HSI and C-Y colour models have been analysed to identify colour component that can give significant
segmentation performance. Finally, median filter and seeded region growing area extraction algorithms have
been applied for smoothing the image and remove any unwanted regions from the image, respectively. The
proposed segmentation method has been evaluated on 100 malaria images. Overall, segmentation using S
component of C-Y colour model has proven to be the best in segmenting the malaria image with segmentation
accuracy and F-score of 99.46% and 0.9370, respectively.

Key-Words: - Malaria, Colour Segmentation, Colour Models, k-Means Clustering, Seeded Region Growing
Area Extraction.

1 Introduction standard for laboratory diagnosis of malaria [3]-[5].

Malaria is a serious global health problem, causing In general, detection of the presence of malaria
widespread sufferings and deaths particularly in parasites in the examined blood slide is one of the
Africa and south Asia. In 2010, about 3.3 billion most important tasks in malaria diagnosis [6]. The
people which are half of the world populations are procedure is performed manually by expert
at risk of malaria. In addition, this disease has microbiologists by searching for the parasites in
caused the death of an estimation of 655,000 people blood slide using a light microscope [4], [5].
in 2010, with 86% of the victims are children under During malaria diagnosis, the presence of the
five years of age [1]. Malaria is caused by a parasites is recognizable by their physical features
peripheral blood parasite of the genus Plasmodium. as well as the appearance of the red blood cells
The genus Plasmodium has five species that can (RBCs) that they have infected [7]. Here, visual
cause human infection namely P. falciparum, P. detection of the parasites becomes efficient by using
vivax, P. ovale, P. malariae and P. knowlesi [2]. the Giemsa stain. The staining process slightly
Prompt and accurate diagnoses of malaria colorizes the RBCs, but highlights the parasites,
infection are the main keys to control and cure this white blood cells (WBCs), platelets and artefacts
disease effectively. Currently, the most economic [7]. Thus, the whole process requires an ability to
and reliable diagnosis which is based on differentiate between the malaria parasites or
microscopic examination of blood slide, especially infected RBCs with these non-parasitic stained
based on the thin blood smear, still remains the gold components (normal RBCs, WBCs, platelets and

E-ISSN: 2224-2902 41 Issue 1, Volume 10, January 2013

 WSEAS TRANSACTIONS on BIOLOGY
and BIOMEDICINE Aimi Salihah Abdul-Nasir, Mohd Yusoff Mashor, Zeehaida Mohamed

artefacts) using visual information. However, the colour image segmentation of malaria have been
manual recognition method is time consuming and published in literature. A typical malaria image
effortful especially in situation where large number consists of three main regions namely parasites or
of samples require reliable analysis [4]. Therefore, infected cells, normal RBCs and background
fast and efficient methods are required for detection regions as shown in Fig.2. In order to get an
of malaria parasites in order to prevent the false accurate diagnosis, the proposed segmentation
diagnosis of malaria. method must be capable of differentiating between
In order to detect the malaria parasites, one of the the malaria parasites or infected cells with the non-
main tasks that need to be performed during image parasitic stained components. However, majority of
processing is the segmentation of malaria image. It existing methods for segmentation of malaria image
is performed before the parasite recognition to do not addressed this requirement effectively except
segment the parasite or infected cell from its in [12], [17]. In [17], a Bayesian pixel classifier has
complicated blood cells background. Many current been employed in order to differentiate between the
research efforts have been focused on new stained and non-stained pixels. Then, the detected
approaches in segmenting the malaria image by stained pixels have been processed to form labelled
using various image processing techniques such as connected components of the parasite.
thresholding [8]-[10], watershed [11]-[13], In further studies, segmentation of malaria image
morphological [14], normalized cut [15] and fuzzy using thresholding [8]-[10] and morphological [14]
divergence [16]. have given promising results. However, these
Mandal et al. [15] have proposed a segmentation techniques are very sensitive to image quality. The
method based on optimized normalized cut (NCut) differences in smear preparation can also cause
algorithm for segmenting the RBCs that have been variations as often as the imaging conditions. For
infected with malaria parasites in peripheral blood example, acidity (pH) of the buffer solution can
smears. The NCut algorithm is based on a global significantly affect the appearance of the parasites
criterion and it maximizes both the total and RBCs [7]. In addition, the non-standard
dissimilarity between the different groups and the preparation of the slide can also lead on producing
total similarity within the groups. Here, the NCut the under or over-staining conditions of the slide.
has been applied in four colour models which are Based on these arguments, the current study will
RGB, HSV, YCbCr and NTSC. By using this utilize the potential of colour image segmentation
method, the segmented trophozoite and schizont approach using various colour models and k-means
have been obtained and the results indicate that the clustering algorithm in order to obtain the fully
performance of the NCut is best in HSV colour segmented RBCs infected with malaria parasites
model. However, the results shown that the artefacts based on the thin blood smear images.
are still appeared on the segmented image. As this
algorithm is based on global criterion, any
unintended noises can significantly reduce the 2 Methodology
segmentation accuracy. The proposed procedures to develop new image
Anggraini et al. [9] have proposed a histogram- processing approach for segmentation of malaria
based thresholding method to identify the presence parasites are summarized as follows:
of malaria parasites in thin blood smears of P.
falciparum species. The grayscale malaria images Step 1: Capture the malaria slide images.
have been segmented using global thresholding to Step 2: Apply the contrast enhancement technique
obtain the RBCs and other blood cells components namely partial contrast stretching technique
in each image. Then, the parasite and infected cell on original malaria image.
components are obtained by applying multiple Step 3: Extract the colour components of RGB
thresholds on the segmented image. This step is (red, green, blue), HSI (hue, saturation,
based on the knowledge that cytoplasm of the intensity) and C-Y colour models from the
parasite appears lighter, while the nucleus of the enhanced image.
parasite appears darker compared to the cytoplasm Step 4: Apply the unsupervised segmentation
of the RBC. Even though the threshold levels have technique namely k-means clustering
been selected automatically, this method is heavily algorithm.
depends on image quality and fails when the Step 5: Apply the 7×7 pixels median filter.
histogram does not have distinct valleys. Step 6: Apply the seeded region growing area
Since the actual malaria diagnosis is performed extraction algorithm.
based on stained slide, different approaches for

E-ISSN: 2224-2902 42 Issue 1, Volume 10, January 2013

 WSEAS TRANSACTIONS on BIOLOGY
and BIOMEDICINE Aimi Salihah Abdul-Nasir, Mohd Yusoff Mashor, Zeehaida Mohamed

Further details for the image acquisition, contrast

enhancement, image segmentation as well as the
evaluation of segmentation performance are
discussed in the following sections.

2.1 Image Acquisition (a) (b)

The first step is to acquire the images of malaria
samples. In this study, the malaria images of ring,
trophozoite and gametocyte stages have been
captured from the thin blood smears of P. vivax
samples. The malaria slides are prepared by Medical
Microbiology & Parasitology Department, Hospital
University Science Malaysia (HUSM). Each slide (c) (d)
has been stained by using the Giemsa staining. The Fig.2 Samples of the captured malaria images
malaria slides are examined using 100X oil
immersion objective of Leica DLMA microscope.
The images are then captured using Infinity-2 digital 2.2 Contrast Enhancement Using Partial
camera at a resolution setting of 800×600 pixels and Contrast Stretching
saved in BMP format. The captured images are The malaria images captured through the
studied under the supervision of microbiologists in microscope may have their own weaknesses such as
order to recognize and differentiate between the blurred or low contrast. Thus, a contrast
three life-cycle stages of P. vivax species. enhancement technique namely partial contrast
Fig.1 shows a set of Leica DLMA microscope, stretching (PCS) is utilized to improve the image
Infinity-2 digital camera and personal computer quality and contrast of malaria image. This
interfaced together to acquire the malaria images. In technique is based on a linear mapping function that
order to assess the proposed segmentation method, is used to increase the contrast and brightness levels
100 malaria images with various conditions have of the image. The detail descriptions of PCS
been captured from five malaria slides and will be technique can be referred in [18], [19].
processed using the proposed procedure. Fig.2(a) Fig.3 briefly illustrates the stretching and
and (b) show the samples of the captured malaria compression processes for PCS technique. By
images with uniform distribution of RBCs and applying this technique, the pixels within the range
stacking RBCs, respectively. Meanwhile, Fig.2(c) of lower threshold value, minTH and upper
and (d) show the samples of the captured malaria threshold value, maxTH will be mapped to a new
images with the presence of platelet (red circle) and range and stretched linearly to a wider range of
artefacts, respectively. Based on these malaria pixels within new lower stretching value, NminTH
images, it can be seen that the colour of the parasites and new upper stretching value, NmaxTH so that the
and normal RBCs regions varies in each slide due to dynamic range of the histogram is fulfilled. On the
the non-standard preparation of the blood slides. other hand, the remaining pixels will experience
compression.

minTH maxTH
Compression
0
Compression
255
process process

Stretching process

NminTH NmaxTH
0 255

Fig.1 A set of Leica DLMA microscope, Infinity-2 Fig.3 Partial contrast stretching process
digital camera and personal computer interfaced
together to acquire the malaria images

E-ISSN: 2224-2902 43 Issue 1, Volume 10, January 2013

 WSEAS TRANSACTIONS on BIOLOGY
and BIOMEDICINE Aimi Salihah Abdul-Nasir, Mohd Yusoff Mashor, Zeehaida Mohamed

2.3 Detection of Malaria Parasites Based on Y  0.299 0.587 0.114   R 

RGB, HSI and C-Y colour Models  R  Y   0.701  
Since the differences in smear preparation as well as     0.587  0.114 G
 
the imaging condition can cause variations in  B  Y    0.299  0.587 0.886   B 
malaria images, selection of colour component is
very important as this step may ease the parasite (5)
detection and segmentation process. Thus, the Here, both R-Y and B-Y represent the chromaticity
current study investigates three types of colour of a colour. In C-Y colour model, saturation (S) and
models which are RGB, HSI and C-Y colour hue, θ which represent the colour property, can also
models. The RGB is the best known colour model be derived from the R-Y and B-Y components as
and is widely used for acquiring and displaying follows [21]:
colour digital images. Each colour pixel is
represented by its three components which are red S  (R  Y )2  (B  Y )2 (6)
(R), green (G) and blue (B). As for the HSI and C-Y
colour models, the two colour models have been
chosen because both are very important and  1 R  Y  for S  0
tan  ,
attractive colour models for image processing θ  B Y  (7)
applications as they can represent colour similarly undefined,
as how the human eye senses colour [20], [21].  for S  0
The HSI colour model represents every colour
with three components which are hue (H), saturation
(S) and intensity (I). Hue is a colour attribute that
describes a pure colour, whereas saturation gives a
measure to which the white light is mixed with the
pure colour [22]. The intensity expresses the (a) Malaria image (b) Red (c) Green
brightness of the hue and saturation. The conversion
from RGB to HSI colour model can be computed
using the following equation [20]:

θ if B  G
Hue   (1) (d) Blue (e) Hue of HSI (f) Saturation of HSI
360θ if B  G



1
1
2
R  G   R  B  
θ  cos  1 
(2)

 R  G   R  B G  B  2 

2
  (g) Intensity of HSI (h) R-Y of C-Y (i) B-Y of C-Y

3
Saturation  1  minR, G , B  (3)
RG B
(j) Luminance of C-Y (k) Hue of C-Y (l) Saturation of C-Y

1 Fig.4 Different colour components of RGB, HSI and

Intensity  R  G  B  (4) C-Y colour models that have been extracted from
3 malaria image
The C-Y colour model consists of three colour Fig.4 shows a sample of malaria image and its
components which are R-Y, G-Y and B-Y, and one colour components based on RGB, HSI and C-Y
luminance (Y) component. Here, R-Y, G-Y and B-Y colour models. Based on the observation of several
are the subtraction of Y from R, G and B malaria images, it is found that the appearance of
components, respectively. However, only two of the infected cell is highlighted in most colour
three colour components are needed to define a component images except in the H component
colour. The conversion from RGB to C-Y colour images of both HSI and C-Y colour models. Due to
model can be computed using the following the similar appearance between the infected cell and
equation [21]: normal RBCs regions as shown in Fig.4(e) and (k),

E-ISSN: 2224-2902 44 Issue 1, Volume 10, January 2013

 WSEAS TRANSACTIONS on BIOLOGY
and BIOMEDICINE Aimi Salihah Abdul-Nasir, Mohd Yusoff Mashor, Zeehaida Mohamed

it would be difficult to segment the infected cell 3. Assign all pixels to the nearest centre based on
from the normal RBCs in case of using the H d.
component image. Thus, the rest colour components 4. When all pixels have been assigned, recalculate
have been chosen to be fed as the input images to k- the new position of the centres using:
means clustering for further segmentation process.
1
cj 
nj
  p ( x, y )
yc j xc j
(10)
2.4 Image Segmentation Using k-Means
Clustering
After transforming the RGB into HSI and C-Y 5. Repeat steps 2 to 4 until there is no significant
colour models, the next and important step in image change in the centre positions.
segmentation is to extract the meaningful region
from malaria image. The malaria slides are usually
stained to highlight the region of interest (ROI) 2.5 Image Filtering Using Median Filter
which is referred to the parasite or infected cell. Algorithm
However, segmenting the parasite or infected cell in After the segmented infected cell has been obtained
an image is not an easy task due to the inconsistency using k-means clustering, there might be some
intensity of these two regions as it may appear unwanted regions or noise that are still encountered
lighter or darker depending in the pH of the buffer in the image. Thus, median filter is used as a noise
used. removal in order to obtain a noise-free image. Due
In order to reduce the tedious task of manual to its good smoothing effect, it can also be used to
segmentation, an unsupervised pixel segmentation fill the small holes that might appear on the
based on k-means clustering algorithm [23] is segmented infected cell. Here, the neighbourhood of
applied for easily segmenting the infected cell from n × n (n = 7) pixels is used because large
its complicated blood cells background. The k- neighbourhoods produce more severe smoothing.
means is a clustering method which is one of the
most popular unsupervised learning algorithms due
to its simplicity. In this study, each colour 2.6 Seeded Region Growing Area Extraction
component image of the RGB, HSI and C-Y colour Algorithm
models that has been extracted from the enhanced In this study, a modified version of conventional
RGB image will be fed as input to k-means seed based region growing algorithm namely seeded
clustering for further segmentation process. region growing area extraction (SRGAE) algorithm
Consider a malaria image with resolution of X × [24] has been applied on the segmented image. This
Y pixels to be clustered into nc regions. Let p(x,y) as algorithm is chosen due to its capability to label the
an input pixel to be clustered and cj is the j-th centre ROI according to their order in the image as well as
(cluster) (x = 1, 2, …, X, y = 1, 2, …, Y and j = 1, 2, extracting the size of the segmented region. Since
…, nc). For segmentation of malaria image, the the segmentation using k-means clustering is based
number of clusters, j are set to 3. The k-means only on colour information of the pixels in the
clustering algorithm for image segmentation can be image, some artefacts and unwanted regions which
implemented as follows: share the same colour as the infected cell are still
1. Initialize the centres using: appeared on the segmented image. Thus, the
SRGAE algorithm is applied for the two main
 max p ( x, y )  min p ( x, y )  purposes. First is to calculate the total area in pixels
c j  min p(x,y)  2 j  1  (8) for the ROI. Secondly is to remove any unwanted
 2nc  regions that are bigger in size in which cannot be
cleaned by using the 7×7 pixels median filter.
where minp(x,y) and maxp(x,y) are the minimum In order to apply the SRGAE algorithm, the
and maximum pixel levels in the image. segmented malaria image will first be converted into
2. For each pixel of an image, calculate the binary image, where the ROI and background
Euclidean distance, d using: regions will be assigned to 0 and 255, respectively.
Then, the SRGAE algorithm can be implemented as
d  p( x, y)  c j (9) follows [24]:

E-ISSN: 2224-2902 45 Issue 1, Volume 10, January 2013

 WSEAS TRANSACTIONS on BIOLOGY
and BIOMEDICINE Aimi Salihah Abdul-Nasir, Mohd Yusoff Mashor, Zeehaida Mohamed

1. Initialize Area[k] = 0 and set the value of k = 0, 2.7 Analysis of Segmentation Performance
where Area is the total pixels for the ROI and k After the proposed colour image segmentation
is the number of current ROI. approach using various colour models and k-means
2. Search for the seed with intensity of pixel, I = 0. clustering have been performed, a common
If the seed is found, increase k to k + 1 and quantitative analysis is conducted in order to assess
Area[k] = 1; else go to step 7. the overall performance of the proposed
3. Search for the neighbourhood of 8 surrounding segmentation method. The performances of the
pixels, grow if I = 0 and increase Area[k] = proposed segmentation method are evaluated by
Area[k] + 1 for each pixel that satisfies the using six objective indices. These indices are
growing condition. accuracy, sensitivity, specificity, precision, recall
4. Grow from the neighbour pixels in step 3 and and F-score. The quality of segmented image is
increase Area[k] = Area[k] + 1 for each pixel determined based on pixels similarity of the
that satisfies the growing condition. resultant segmented image against the manual
5. Repeat steps 3 to 4 until all pixels have been segmented image. The accuracy, sensitivity and
considered to be grown or the region cannot be specificity are defined based on Equation 11, 12 and
grown anymore. 13, respectively.
6. Repeat steps 2 to 5 for the new seed which is
not belong to the previous ROI(s). TP  TN
7. End. Accuracy  100 (11)
TP  TN  FP  FN
Afterwards, the selection of ROI is determined
based on its size. Fig.5 shows a single infected cell
TP
in malaria image with its area. After performing Sensitivity  100 (12)
analyses on several malaria images, it has been TP  FN
found that a typical infected cell may have the area
which is greater than 5000 pixels. Thus, any regions TN
Specificity  100 (13)
which are less than 5000 pixels are considered as TN  FP
non-parasite and will be eliminated from the image
during region growing process. This value is chosen TP, TN, FP and FN are the true positive, true
by considering the size of the infected cell from the negative, false positive and false negative,
three malaria stages which are ring, trophozoite and respectively. Based on validation from
gametocyte stages. Therefore, the intensity of pixels microbiologists, TP is refer as the positive region
which are included as ROI will be set to RGB (infected cell) that has been correctly segmented as
colour of the enhanced malaria image, while the positive region, while TN is refer as the negative
intensity of pixels which are not included as ROI region (normal RBCs and background) that has been
will be set to 255. correctly segmented as negative region. Based on
the above equations, segmentation accuracy can be
obtained by calculating the percentage of pixels that
are correctly segmented as infected cell or
background in the image. The sensitivity is defined
as the percentage of pixels that are correctly
(a) Ring. (b) Ring. (c) Ring. segmented as positive region, while the specificity is
Area=11061 Area=13965 Area=14768 defined as the percentage of pixels that are correctly
segmented as negative region.
Afterwards, the performances of the proposed
segmentation method are further evaluated by using
the precision, recall and F-score measures as defined
(d) Trophozoite. (e) Trophozoite. (f) Trophozoite. in Equation 14, 15 and 16, respectively.
Area=15371 Area=17461 Area=19438

TP
Precision = (14)
TP  FP

(g) Gametocyte. (h) Gametocyte. (i) Gametocyte. TP

Area=20321 Area=19681 Area=22170
Recall = (15)
TP  FN
Fig.5 A single infected cell in malaria image

E-ISSN: 2224-2902 46 Issue 1, Volume 10, January 2013

 WSEAS TRANSACTIONS on BIOLOGY
and BIOMEDICINE Aimi Salihah Abdul-Nasir, Mohd Yusoff Mashor, Zeehaida Mohamed

precision  recall
F-score = 2  (16)
precision  recall

The precision and recall represent the quality of

segmented image either the image is over- (a) Ring_1 image (b) Trophozoite_1 (c) Gametocyte_1
image image
segmented or under-segmented. Low precision
value will relate to over-segmentation, while low Fig.7 Results of images of partial contrast stretching
recall value will relate to under-segmentation. Here, technique
recall is similar to the sensitivity. F-score is a
combination of precision and recall to provide a In order to perform segmentation on malaria
single statistical measure for the segmented image, image, different colour components of RGB, HSI
which will be 1 for a perfect segmentation. and C-Y colour models have been extracted from
the enhanced image. Fig.8, 9 and 10 represent the
different colour components of RGB, HSI and C-Y
colour models that have been extracted from partial
3 Results and Discussions contrast stretching images of Ring_1, Trophozoite_1
In this study, the proposed colour image and Gametocyte_1, respectively. Based on the
segmentation approach using various colour models resultant red, blue, R-Y and B-Y components
and k-means clustering have been applied on 100 images as shown in image (a), (c), (f) and (g),
malaria images of P. vivax species. Comparisons respectively, the infected cell and RBCs regions
between different colour components of RGB, HSI tend to have the similar pixel values for each of
and C-Y colour models that have been used as input these four colour components. Here, both regions
images to k-means clustering have been made in tend to appear as the darker part of the image.
order to recognize the significance of each colour However, this situation does not appear for the
component on image segmentation. In order to green, S and I components of HSI colour model, as
assess the proposed work, the captured malaria well as the Y and S components of C-Y colour
images with various conditions have been processed model as shown in image (b), (d), (e), (h) and (i),
using the proposed procedure. The qualities of respectively. Based on the green, I and Y
segmented images have been determined based on components images, the infected cell appeared as
both qualitative and quantitative evaluations. the darker part of the image, while the RBCs and
background regions appeared as the medium and
brighter parts of the image, respectively. As for the
3.1 Qualitative Analysis S component images of both HSI and C-Y colour
The original malaria images of ring, trophozoite and models, the infected cell appeared as the brighter
gametocyte stages named as Ring_1, Trophozoite_1 part of the image, while the RBCs and background
and Gametocyte_1 are shown in Fig.6(a)-(c), regions appeared as the darker part of the image.
respectively. Based on these malaria images, the
morphologies of infected cell are hardly seen due to
the low image contrast. The results of applying the
partial contrast stretching technique on Ring_1,
Trophozoite_1 and Gametocyte_1 images are shown
in Fig.7(a)-(c), respectively. Based on these
resultant images, the contrast of the infected cell, (a) Red (b) Green (c) Blue

RBCs and background regions has been improved

significantly compared to the original images.

(d) Saturation of HSI (e) Intensity of HSI (f) R-Y of C-Y

(a) Ring_1 image (b) Trophozoite_1 (c) Gametocyte_1

image image
Fig.6 Original malaria images (g) B-Y of C-Y (h) Luminance of C-Y (i) Saturation of C-Y

E-ISSN: 2224-2902 47 Issue 1, Volume 10, January 2013

 WSEAS TRANSACTIONS on BIOLOGY
and BIOMEDICINE Aimi Salihah Abdul-Nasir, Mohd Yusoff Mashor, Zeehaida Mohamed

Fig.8 Different colour components of RGB, HSI and Based on these resultant images, the images have
C-Y colour models that have been extracted from been clustered into three groups which are infected
PCS image of Ring_1 cell (black), RBCs (gray) and background (white)
regions except for image (d) where the infected cell
is represented by black and gray colour.

(a) Red (b) Green (c) Blue

(a) Red (b) Green (c) Blue

(d) Saturation of HSI (e) Intensity of HSI (f) R-Y of C-Y

(d) Saturation of HSI (e) Intensity of HSI (f) R-Y of C-Y

(g) B-Y of C-Y (h) Luminance of C-Y (i) Saturation of C-Y

Fig.9 Different colour components of RGB, HSI and
C-Y colour models that have been extracted from (g) B-Y of C-Y (h) Luminance of C-Y (i) Saturation of C-Y
PCS image of Trophozoite_1 Fig.11 Results of images for Ring_1 after applying
k-means clustering on different colour components
images

(a) Red (b) Green (c) Blue

(a) Red (b) Green (c) Blue

(d) Saturation of HSI (e) Intensity of HSI (f) R-Y of C-Y

(d) Saturation of HSI (e) Intensity of HSI (f) R-Y of C-Y

(g) B-Y of C-Y (h) Luminance of C-Y (i) Saturation of C-Y

Fig.10 Different colour components of RGB, HSI
and C-Y colour models that have been extracted (g) B-Y of C-Y (h) Luminance of C-Y (i) Saturation of C-Y
from PCS image of Gametocyte_1 Fig.12 Results of images for Trophozoite_1 after
applying k-means clustering on different colour
Due to the similar appearance between the components images
infected cell and RBCs regions as shown in red,
blue, R-Y and B-Y components images, it would be
difficult to determine the threshold value in case of
applying the thresholding technique for
segmentation of malaria image. Thus, the
unsupervised k-means clustering algorithm has been
applied on each colour component image for easily
segmenting the infected cell from the RBCs and
background regions. Fig.11, 12 and 13 show the
resultant images after applying k-means clustering.

E-ISSN: 2224-2902 48 Issue 1, Volume 10, January 2013

 WSEAS TRANSACTIONS on BIOLOGY
and BIOMEDICINE Aimi Salihah Abdul-Nasir, Mohd Yusoff Mashor, Zeehaida Mohamed

(a) Red (b) Green (c) Blue (a) Red (b) Green (c) Blue

(d) Saturation of HSI (e) Intensity of HSI (f) R-Y of C-Y (d) Saturation of HSI (e) Intensity of HSI (f) R-Y of C-Y

(g) B-Y of C-Y (h) Luminance of C-Y (i) Saturation of C-Y (g) B-Y of C-Y (h) Luminance of C-Y (i) Saturation of C-Y
Fig.13 Results of images for Gametocyte_1 after Fig.14 Results of images for Ring_1 after the
applying k-means clustering on different colour colour of k-means clustering image has been
components images retrieved based on PCS image

Fig.14, 16 and 18 show the resultant images after

the colour of k-means clustering image has been
retrieved based on PCS image. These images
include several unwanted regions such as small
background pixels, segmented RBCs, platelet (red
(a) Red (b) Green (c) Blue
circle) and artefacts. Since these unwanted regions
share the same colour as the infected cell, these
regions have not been eliminated through clustering
process. Thus, the output image of k-means
clustering has been processed using median filter to
produce a cleaner and smoother image. Due to its (d) Saturation of HSI (e) Intensity of HSI (f) R-Y of C-Y
good smoothing effect, it has been used to cover the
small holes that appeared on the segmented image.
However, some unwanted regions such as
segmented RBCs, platelet and artefacts are still
appeared on the image due to their size in which (g) B-Y of C-Y (h) Luminance of C-Y (i) Saturation of C-Y
cannot be cleaned by using the 7×7 pixels median Fig.15 Results of images for Ring_1 after
filter. applying median filter and SRGAE algorithms to k-
In order to remove the large unwanted regions, means clustering image
SRGAE algorithm has been furthered applied to
extract the size of the segmented region. During
applying the SRGAE algorithm, any regions which
are less than 5000 pixels are considered as non-
parasite and will be eliminated from the image.
Fig.15, 17 and 19 represent the final segmented
(a) Red (b) Green (c) Blue
images that have been obtained after applying
median filter and SRGAE algorithms on Ring_1,
Trophozoite_1 and Gametocyte_1 images,
respectively. Although many of the unwanted
regions have been eliminated through region
growing process, several segmented RBCs are still (d) Saturation of HSI (e) Intensity of HSI (f) R-Y of C-Y
appeared on the image because their sizes are
similar to the infected cell.

E-ISSN: 2224-2902 49 Issue 1, Volume 10, January 2013

 WSEAS TRANSACTIONS on BIOLOGY
and BIOMEDICINE Aimi Salihah Abdul-Nasir, Mohd Yusoff Mashor, Zeehaida Mohamed

(g) B-Y of C-Y (h) Luminance of C-Y (i) Saturation of C-Y (a) Red (b) Green (c) Blue
Fig.16 Results of images for Trophozoite_1 after
the colour of k-means clustering image has been
retrieved based on PCS image

(d) Saturation of HSI (e) Intensity of HSI (f) R-Y of C-Y

(a) Red (b) Green (c) Blue

(g) B-Y of C-Y (h) Luminance of C-Y (i) Saturation of C-Y

Fig.19 Results of images for Gametocyte_1 after
applying median filter and SRGAE algorithms to k-
means clustering image
(d) Saturation of HSI (e) Intensity of HSI (f) R-Y of C-Y
The performances of the proposed segmentation
method have also been tested on other three malaria
images named as Ring_2, Trophozoite_2 and
Gametocyte_2 images as shown in Fig.20, 21 and
22, respectively. Based on the final segmented
(g) B-Y of C-Y (h) Luminance of C-Y (i) Saturation of C-Y
Fig.17 Results of images for Trophozoite_1 after images obtained from the six malaria images, the
applying median filter and SRGAE algorithms to k- fully segmented infected cell as well as a clean
means clustering image segmented malaria image has been obtained by
applying k-means clustering on S component image
of HSI colour space. Besides, a clean segmented
malaria image can also be obtained by applying k-
means clustering on R-Y, B-Y and S components
images of C-Y colour model. Based on the final
segmented images provided by the red, green, blue,
(a) Red (b) Green (c) Blue I and Y, it can be seen that segmentation using these
five colour components have result on obtaining
over-segmented images. This is due to pixel
similarity between the infected cell and RBCs
regions provided by these colour components.
(d) Saturation of HSI (e) Intensity of HSI (f) R-Y of C-Y Moreover, the segmented RBCs could not be
eliminated through region growing process due to
their sizes which are similar to the infected cell.
Thus, these five colour components are unsuitable to
be fed as inputs to k-means clustering especially in
the case of overlapping RBCs.
(g) B-Y of C-Y (h) Luminance of C-Y (i) Saturation of C-Y
Fig.18 Results of images for Gametocyte_1 after
the colour of k-means clustering image has been
retrieved based on PCS image

E-ISSN: 2224-2902 50 Issue 1, Volume 10, January 2013

 WSEAS TRANSACTIONS on BIOLOGY
and BIOMEDICINE Aimi Salihah Abdul-Nasir, Mohd Yusoff Mashor, Zeehaida Mohamed

(a) Original image (b) Partial contrast (c) Manual segmented (a) Original image (b) Partial contrast (c) Manual segmented
stretching image image stretching image image

(d) Red (e) Green (f) Blue (d) Red (e) Green (f) Blue

(g) Saturation of HSI (h) Intensity of HSI (i) R-Y of C-Y (g) Saturation of HSI (h) Intensity of HSI (i) R-Y of C-Y

(g) B-Y of C-Y (h) Luminance of C-Y (i) Saturation of C-Y (g) B-Y of C-Y (h) Luminance of C-Y (i) Saturation of C-Y
Fig.20(a) Original Ring_2 image, (b) PCS image, Fig.22(a) Original Gametocyte_2 image, (b) PCS
(c) manual segmented image and (d)-(i) final image, (c) manual segmented image and (d)-(i) final
segmented images after applying the proposed segmented images after applying the proposed
segmentation method segmentation method

3.2 Quantitative Analysis

After the proposed segmentation method has been
applied on malaria images, the performances of the
(a) Original image (b) Partial contrast (c) Manual segmented
proposed segmentation method are further evaluated
stretching image image by comparing the results of segmented images with
manual segmented images. Table 1 tabulates the
segmentation performances based on sensitivity,
specificity and accuracy that have been obtained
from segmented images of Ring_1, Ring_2,
Trophozoite_1, Trophozoite_2, Gametocyte_1,
(d) Red (e) Green (f) Blue
Gametocyte_2 and the overall 100 malaria images.
The best results obtained for analyses using the
overall 100 malaria images are made bold.
Based on segmentation accuracy provided by six
malaria images as given in Table 1, segmentations
(g) Saturation of HSI (h) Intensity of HSI (i) R-Y of C-Y using S component of HSI colour model, as well as
R-Y and S components of C-Y colour model have
provided good segmentation performance in terms
of high segmentation accuracy compared to the
results provided by other colour components. Here,
(g) B-Y of C-Y (h) Luminance of C-Y (i) Saturation of C-Y segmentation using S component of C-Y colour
Fig.21(a) Original Trophozoite_2 image, (b) PCS model has produced the highest segmentation
image, (c) manual segmented image and (d)-(i) final accuracy in Ring_1, Ring_2 and Trophozoite_1
segmented images after applying the proposed images, while segmentation using R-Y component
segmentation method has produced the highest segmentation accuracy in

E-ISSN: 2224-2902 51 Issue 1, Volume 10, January 2013

 WSEAS TRANSACTIONS on BIOLOGY
and BIOMEDICINE Aimi Salihah Abdul-Nasir, Mohd Yusoff Mashor, Zeehaida Mohamed

Trophozoite_2, Gametocyte_1 and Gametocyte_2 each RGB, HSI and C-Y colour models is very
images. important in order to achieve a good segmentation
Based on the average segmentation performance performance. Based on the average segmentation
for the overall 100 images, segmentation using S performance in terms of segmentation accuracy and
component of C-Y colour model has proven to be F-score for the overall 100 images, segmentation
the best in segmenting the entire area in malaria using green component has provided the best
image and background region with segmentation segmentation performance among the colour
accuracy and specificity of 99.46% and 99.95%, components in RGB colour model. Meanwhile, both
respectively. Meanwhile, segmentation using S HSI and C-Y colour models have achieved the best
component of HSI colour model has proven to be segmentation performance by using the S
the best in obtaining a fully segmented infected cell component.
with sensitivity of 93.84%. In order to ascertain the By comparing the results provided by Table 1
segmentation performances provided in Table 1, the and 2, it can be noticed that F-score has the
performances of the proposed segmentation method capability to provide better measurement of
are further evaluated by computing three objective segmentation performance compared to accuracy.
indices namely precision, recall and F-score. This is because accuracy is directly depends on
The performances of segmentation based on pixels similarity, whereas F-score covers the
precision, recall and F-score that have been obtained information of precision and recall. By referring the
from segmented images of Ring_1, Ring_2, segmentation performance for Ring_1 image,
Trophozoite_1, Trophozoite_2, Gametocyte_1, segmentation using R-Y and S components of C-Y
Gametocyte_2 and the overall 100 malaria images colour model have provided the similar result with
are tabulated in Table 2. Similar with Table 1, segmentation accuracy of 99.93%. However, by
segmentations using S component of HSI colour using the F-score measure, segmentation using S
model, as well as R-Y and S components of C-Y component has proven to be slightly better with F-
colour model have provided good segmentation score of 0.9870 compared to segmentation using R-
performance in terms of high F-score value for the Y component with F-score of 0.9869. Thus, F-score
six malaria images compared to the results provided is found suitable for measuring segmentation
by other colour components. Based on the average performance as it is more sensitive compared to
segmentation performance for the overall 100 accuracy. Meanwhile, precision has been found
images, segmentation using S component of C-Y suitable for measuring the over-segmented image
colour model has proven to be the best in compared to specificity.
segmenting the entire area in malaria image with F- By comparing the segmentation results provided
score of 0.9370. by the S component of HSI and C-Y colour models,
By comparing the precision results provided by segmentation using S component of HSI colour
each colour components, there are only three colour model has proven to be the best in obtaining a fully
components that have the capability to avoid over- segmented infected cell with recall of 0.9384.
segmentation in malaria image by providing However, the results for the precision and F-score
precision value more than 0.8. This is due to pixel that have been obtained are quite small with 0.7353
similarity between the infected cell and RBCs and 0.7591, respectively. Thus, segmentation using
regions provided by several colour components S component of C-Y colour model has proven to be
which lead on producing over-segmented images as the best in segmenting the entire area in malaria
shown in Section 3.1. However, segmentation using image and background region with F-score and
S component of C-Y colour model has provided the precision of 99.46% and 99.95%,, respectively.
most acceptable result in segmenting the Overall, the results of segmentation performances
background region with precision of 0.9871. As for provided by Table 1 and 2 have strongly supported
the recall results, it can be seen that each colour the qualitative findings provided in Section 3.1.
component has the capability to avoid under-
segmentation in malaria image by providing recall Table 1: Segmentation performance based on
value more than 0.8. However, segmentation using sensitivity, specificity and accuracy for the
S component of HSI colour model has proven to be segmented images
the best in obtaining a fully segmented infected cell Image Colour Colour Sensitivity Specificity Accuracy
Model Component (%) (%) (%)
with recall of 0.9384 which is similar to the Ring_1 RGB Red 81.98 91.05 90.80
sensitivity. Green 95.50 85.88 86.14
Based on the results provided by Table 1 and 2, Blue 89.47 86.36 86.45
HSI Saturation 95.99 100 99.89
it is shown that selection of colour component from

E-ISSN: 2224-2902 52 Issue 1, Volume 10, January 2013

 WSEAS TRANSACTIONS on BIOLOGY
and BIOMEDICINE Aimi Salihah Abdul-Nasir, Mohd Yusoff Mashor, Zeehaida Mohamed

Intensity 94.22 82.09 82.43 Intensity 0.1309 0.9422 0.2299

C-Y R-Y 97.92 99.98 99.93 C-Y R-Y 0.9947 0.9792 0.9869
B-Y 93.07 100 99.81 B-Y 0.9998 0.9307 0.9640
Luminance 94.22 82.09 82.43 Luminance 0.1309 0.9422 0.2299
Saturation 97.81 99.99 99.93 Saturation 0.9962 0.9781 0.9870
Ring_2 RGB Red 81.18 100 99.41 Ring_2 RGB Red 0.9994 0.8118 0.8959
Green 90.60 100 99.70 Green 0.9994 0.9060 0.9504
Blue 92.44 88.53 88.65 Blue 0.2072 0.9244 0.3385
HSI Saturation 89.90 100 99.68 HSI Saturation 0.9996 0.8990 0.9467
Intensity 87.69 100 99.61 Intensity 0.9996 0.8769 0.9342
C-Y R-Y 90.77 86.36 86.50 C-Y R-Y 0.1775 0.9077 0.2970
B-Y 90.80 99.98 99.70 B-Y 0.9945 0.9080 0.9493
Luminance 88.87 100 99.65 Luminance 0.9996 0.8887 0.9409
Saturation 92.29 99.99 99.75 Saturation 0.9982 0.9229 0.9591
Trophozoite_1 RGB Red 83.20 94.99 94.63 Trophozoite_1 RGB Red 0.3443 0.8320 0.4871
Green 89.61 100 99.68 Green 0.9999 0.8961 0.9451
Blue 91.62 98.65 98.44 Blue 0.6824 0.9162 0.7822
HSI Saturation 91.33 100 99.73 HSI Saturation 0.9998 0.9133 0.9546
Intensity 84.33 100 99.52 Intensity 0.9998 0.8433 0.9150
C-Y R-Y 96.36 99.96 99.85 C-Y R-Y 0.9868 0.9636 0.9751
B-Y 65.58 99.95 98.90 B-Y 0.9770 0.6558 0.7848
Luminance 87.01 100 99.60 Luminance 0.9999 0.8701 0.9305
Saturation 97.01 99.98 99.89 Saturation 0.9942 0.9701 0.9820
Trophozoite_2 RGB Red 88.21 100 99.53 Trophozoite_2 RGB Red 0.9998 0.8821 0.9373
Green 94.81 100 99.79 Green 1.0000 0.9481 0.9734
Blue 91.44 100 99.66 Blue 1.0000 0.9144 0.9553
HSI Saturation 94.92 100 99.80 HSI Saturation 1.0000 0.9492 0.9739
Intensity 93.61 100 99.74 Intensity 1.0000 0.9361 0.9670
C-Y R-Y 96.97 99.99 99.87 C-Y R-Y 0.9982 0.9697 0.9838
B-Y 95.40 99.99 99.81 B-Y 0.9985 0.9540 0.9757
Luminance 94.30 100 99.77 Luminance 1.0000 0.9430 0.9707
Saturation 94.23 100 99.77 Saturation 1.0000 0.9423 0.9703
Gametocyte_1 RGB Red 86.76 84.77 84.85 Gametocyte_1 RGB Red 0.1985 0.8676 0.3230
Green 92.31 100 99.68 Green 0.9999 0.9231 0.9600
Blue 91.56 82.47 82.85 Blue 0.1850 0.9156 0.3078
HSI Saturation 92.20 100 99.67 HSI Saturation 0.9999 0.9220 0.9594
Intensity 95.63 89.41 89.67 Intensity 0.2818 0.9563 0.4353
C-Y R-Y 93.28 99.97 99.69 C-Y R-Y 0.9929 0.9328 0.9619
B-Y 90.25 99.99 99.59 B-Y 0.9977 0.9025 0.9477
Luminance 89.90 100 99.58 Luminance 1.0000 0.8990 0.9468
Saturation 89.03 100 99.54 Saturation 0.9997 0.8903 0.9418
Gametocyte_2 RGB Red 78.76 93.00 92.37 Gametocyte_2 RGB Red 0.3402 0.7876 0.4752
Green 92.22 99.93 99.59 Green 0.9836 0.9222 0.9519
Blue 81.03 98.19 97.44 Blue 0.6728 0.8103 0.7352
HSI Saturation 94.29 99.99 99.74 HSI Saturation 0.9974 0.9429 0.9694
Intensity 84.32 99.91 99.23 Intensity 0.9775 0.8432 0.9054
C-Y R-Y 97.26 99.89 99.77 C-Y R-Y 0.9755 0.9726 0.9740
B-Y 90.94 99.96 99.56 B-Y 0.9896 0.9094 0.9478
Luminance 88.50 99.92 99.42 Luminance 0.9799 0.8850 0.9301
Saturation 88.24 100 99.48 Saturation 1.0000 0.8824 0.9376
Average of RGB Red 80.29 91.55 91.14 Average of RGB Red 0.5009 0.8029 0.5409
100 Images Green 91.18 97.59 97.31 100 Images Green 0.8480 0.9118 0.8339
Blue 83.28 90.15 89.88 Blue 0.4051 0.8328 0.4822
HSI Saturation 93.84 95.81 95.70 HSI Saturation 0.7353 0.9384 0.7591
Intensity 87.56 93.70 93.45 Intensity 0.6367 0.8756 0.6630
C-Y R-Y 86.89 99.87 99.36 C-Y R-Y 0.9647 0.8689 0.9128
B-Y 92.44 95.36 95.23 B-Y 0.7217 0.9244 0.7587
Luminance 89.29 96.28 95.99 Luminance 0.7836 0.8929 0.7723
Saturation 87.60 99.95 99.46 Saturation 0.9871 0.8760 0.9370

Table 2: Segmentation performance based on 4 Conclusion

precision, recall and F-score for the segmented In this paper, the results of applying the proposed
images colour image segmentation approach using various
Image Colour Colour Precision Recall F-score
Model Component colour models and k-means clustering algorithm
Ring_1 RGB Red 0.2077 0.8198 0.3315 have been presented. Comparisons between
Green 0.1622 0.9550 0.2772 different colour components of RGB, HSI and C-Y
Blue 0.1581 0.8947 0.2687
HSI Saturation 0.9988 0.9599 0.9790
colour models that have been used as input images

E-ISSN: 2224-2902 53 Issue 1, Volume 10, January 2013

 WSEAS TRANSACTIONS on BIOLOGY
and BIOMEDICINE Aimi Salihah Abdul-Nasir, Mohd Yusoff Mashor, Zeehaida Mohamed

to k-means clustering have been made in order to and Image Understanding, Vol.114, No.1,
recognize the significance of applying each colour 2010, pp. 21-32.
component for segmentation of malaria image. The [7] WHO, Basic Malaria Microscopy, Part I.
proposed segmentation method has been tested on Learner’s Guide, World Health Organization,
100 malaria images qualitatively and quantitatively. 2010.
Based on the qualitative findings provided in [8] N. E. Ross, C. J. Pritchard, D. M. Rubin, and
Section 3.1, a fully and smoother infected cell has A. G. Duse, Automated image processing
been obtained by applying k-means clustering on S method for the diagnosis and classification of
component image of HSI colour model. Meanwhile, malaria on thin blood smears, Medical &
a clean segmented image has been obtained by using Biological Engineering & Computing, Vol.44,
the S component image of C-Y colour model. 2006, pp. 427-436.
Quantitatively, the results indicate that segmentation [9] D. Anggraini, A. S. Nugroho, C. Pratama, I. E.
using S component of C-Y colour model has proven Rozi, A. A. Iskandar, and R. N. Hartono,
to be the best in segmenting the entire area in Automated status identification of microscopic
malaria image with F-score and precision of 0.9370 images obtained from malaria thin blood
and 0.9871, respectively. Meanwhile, segmentation smears, 2011 International Conference on
using S component of HSI colour model has proven Electrical Engineering and Informatics, 2011.
to be the best in obtaining a fully segmented [10] V. V. Panchbhai, L. B. Damahe, A. V.
infected cell with recall of 0.9384. Nagpure, and P. N. Chopkar, RBCs and
parasites segmentation from thin smear blood
cell images, I.J. Image, Graphics and Signal
Acknowledgements Processing, Vol.10, 2012, pp. 54-60.
The authors gratefully acknowledges and thanks the [11] K. N. R. Mohana-Rao and A. G. Dempster,
team members of malaria research at Universiti Modification on distance transform to avoid
Malaysia Perlis (UniMAP) for making this research over-segmentation and under-segmentation, 4th
achievable and Universiti Sains Malaysia (USM) for EURASIP - IEEE Region 8 International
providing the malaria blood samples and validate Symposium on Video/Image Processing and
the results. Multimedia Communications, 2002, pp. 295-
301.
[12] F. B. Tek, Computerised diagnosis of malaria,
References: Ph.D. Thesis, School of Life Sciences,
[1] WHO, World Malaria Report 2011, World University of Westminster, London, United
Health Organization, 2011. Kingdom, 2007.
[2] J. Cox-Singh, T. M. E. Davis, K. S. Lee, S. S. [13] D. Das, M. Ghosh, C. Chakraborty, A. K.
G. Shamsul, A. Matusop, S. Ratnam, H. A. Maiti, and M. Pal, Probabilistic prediction of
Rahman, D. J. Conway, and B. Singh, malaria using morphological and textural
Plasmodium knowlesi malaria in humans is information, 2011 International Conference on
widely distributed and potentially life Image Information Processing, 2011.
threatening, Clinical Infectious Disease, [14] C. D. Ruberto, A. Dempster, S. Khan, and B.
Vol.46, No.2, 2008, pp. 165-171. Jarra, Analysis of infected blood cell images
[3] A. Moody, Rapid diagnostic tests for malaria using morphological operators, Image and
parasites, Clinical Microbiology Reviews, Vision Computing, Vol.20, No.2, 2002, pp.
Vol.15, No.1, 2002, pp. 66-78. 133-146.
[4] Y. Purwar, S. L. Shah, G. Clarke, A. [15] S. Mandal, A. Kumar, J. Chatterjee, M.
Almugairi, and A. Muehlenbachs, Automated Manjunatha, and A. K. Ray, Segmentation of
and unsupervised detection of malarial blood smear images using normalized cuts for
parasites in microscopic images, Malaria detection of malarial parasites, 2010 Annual
Journal, Vol.10, 2011, pp. 1-10. IEEE India Conference, 2010.
[5] M. T. Makler, C. J. Palmer, and A. L. Alger, A [16] M. Ghosh, D. Das, C. Chakraborty, and A. K.
review of practical techniques for the diagnosis Ray, Plasmodium vivax segmentation using
of malaria, Annals of Tropical Medicine & modified fuzzy divergence, 2011 International
Parasitology, Vol.92, No.4, 1998, pp. 419-433. Conference on Image Information Processing,
[6] F. B. Tek, A. G. Dempster, and I. Kale, Parasite 2011.
detection and identification for automated thin [17] F. B. Tek, A. G. Dempster, and I. Kale, Malaria
blood film malaria diagnosis, Computer Vision parasite detection in peripheral blood images,

E-ISSN: 2224-2902 54 Issue 1, Volume 10, January 2013

 WSEAS TRANSACTIONS on BIOLOGY
and BIOMEDICINE Aimi Salihah Abdul-Nasir, Mohd Yusoff Mashor, Zeehaida Mohamed

British Machine Vision Conference, 2006, pp.

347-356.
[18] A. R. Weeks, Fundamentals of Electronic
Image Processing, SPIE Optical Engineering
Press, 1996.
[19] A. N. Aimi Salihah, M. Y. Mashor, N. H.
Harun, and H. Rosline, Colour image
enhancement techniques for acute leukaemia
blood cell morphological features, 2010 IEEE
International Conference on Systems, Man, and
Cybernetics, 2010, pp. 3677-3682.
[20] R. C. Gonzalez and R. E. Woods, Digital
Image Processing, Prentice Hall, 2007.
[21] E. R. Dougherty, Electronic Imaging
Technology, SPIE-The International Society for
Optical Engineering, 1999.
[22] B. Jahne, Digital Image Processing, Springer-
Verlag Berlin, Heidelberg, 2005.
[23] J. MacQueen, Some methods for classification
and analysis of multivariate observations,
Proceedings of 5th Berkeley Symposium on
Mathematical Statistics and Probability, 1967,
pp. 281-297.
[24] N. H. Harun, M. Y. Mashor, and H. Rosline,
Calculation of blast area for acute leukemia
blood cells images, Proceedings of the
International Postgraduate Conference on
Engineering, 2010.

E-ISSN: 2224-2902 55 Issue 1, Volume 10, January 2013