International Journal of Computer Information Systems and Industrial Management Applications (IJCISIM)

ISSN: 2150-7988 Vol.3 (2011), pp.026-033

http://www.mirlabs.org/ijcisim

Segmentation of Infrared Images and Objectives Detection Using Maximum

Entropy Method Based on the Bee Algorithm

Milad Azarbad1 Attaollah Ebrahimzade 2 Vahid Izadian3

Babol University of Technology Babol University of Technology Babol University of Technology
azarbadmilad@yahoo.com e_zadeh@nit.ac.ir v.eizadiyan@yahoo.com

Abstract that are below or equal to a certain threshold, the other

includes those with gray values above the threshold. Many
Thresholding is a popular image segmentation method thresholding methods have been reported in the literature
that converts a gray-level image into a binary image. [6-10]. An early review of thresholding methods was
Many thresholding techniques have been proposed in the reported in [6]. A comparative performance study of
recent years. Among them, the maximum entropy global thresholding techniques was reported by Lee et al.
thresholding has been widely applied. Image entropy [7]. Another comparative analysis of the performance of
thresholding approach has drawn the attentions in image eleven histogram-based thresholding algorithms was
segmentation. In this paper, the image thresholding carried out by Glasbely [11]. Among early works on
approach with the index of entropy maximization of the thresholding, Prewitt and Mendelson [12] suggested to
grayscale histogram based on a novel optimization select the threshold at the valleys of the histogram, while
algorithm, namely, the bee algorithm is proposed to deal Doyle [13] advocated the choice of the median. Image
with infrared images. The bee algorithm is realized segmentation is the first and key step of target recognition,
successfully in the process of solving the maximum which separates the objects from the background and
entropy problem. The proposed algorithm uses the bee makes preparation for the post processions. Of particular
algorithm which proved to be the most powerful unbiased interest is an information theoretic approach that is based
optimization technique for sampling a large solution on the concept of entropy introduced by Shannon in
space. Because of its unbiased stochastic sampling, it was information theory [14]. There are many methods of
quickly adapted in image processing and thus for infrared image segmentation such as Hough transformation,
image segmentation as well.  The experiments of Template matching, Neural Network, and etc. [15-16],
segmenting of the infrared images are illustrated to prove among which threshold segmentation is the most
that the proposed method can get ideal segmentation important one. Image thresholding is a basic component of
result with less computation cost.  The proposed algorithm many computer vision systems. While humans can easily
is also applied to the segmentation of standard images differentiable an object from complex background, image
with very promising results. thresholding is a difficult task for separate them. The
gray-level histogram of an image is usually considered as
Keywords- Image Segmentation; Bee Algorithm; Infrared efficient tools for development of image thresholding
Images; Maximum Entropy. algorithms.
Li and Lee’s method [17] minimizes relative cross
entropy or Kullback-Leibler (KL) distance between the
1. Introduction probability distribution functions of the original and the
thresholded image. Kitler and Illingworth [18] developed
Image Thresholding is an important technique in a thresholding method that minimizes the notion of a
image segmentation, enhancement and object detection. segmentation error derived using an information-theoretic
However, the segmentation results depend heavily on the approach, while Dunn et al.’s method [19] is based on a
used image thresholding methods. The image thresholding uniform error criterion. Recently, Leung and Lam [20]
is widely used in halftone reproduction [1], infrared image developed a method that maximizes segmented image
segmentation [2], automatic target recognition [3], color information derived using information-theoretic
image segmentation [4] and mixed-type document approaches and demonstrated that their method is better
analysis [5]. The main goal of image segmentation is to than the methods based on minimum error [18] and
determine a threshold for bi-level thresholding or several uniform error [19] criteria. [6] As well as Abutaleb’s 2-D
thresholds for multi-level thresholding for image histogrambased approaches were discussed in the work of
segmentations. While one include pixels with gray values Yang et al. [21], there is no need to repeat their work here.
Segmentation of Infrared Images and Objectives Detection Using Maximum Entropy Method Based on the Bee Algorithm 027

Instead, this paper is primarily focused on a comparative the maximum entropy based on the bee algorithm. The
study and analysis among Kittler and Illingworth’s MET, rest of the paper is organized as follows. Section (2) the
the three co-occurrence matrix-based entropy thresholding theory of maximum entropy is presented. Section (3) the
techniques and three relative entropy thresholding bee algorithm is explained. Section (4) the simulation
methods plus Otsu’s [22] method. While the methods of results are presented. Section (5) Conclusions are
Wong and Sahoo [23] and Pal and Pal [24], [25] presented.
incorporate some spatial image information in their
methods, others are mostly histogram-based techniques. 2. The Theory of Maximum Entropy
Although researchers have used spatial image information
in several non thresholding image segmentation methods
2.1. The Shannon Entropy
[26-28], thresholding is a fundamentally different and
simple operation. Further, final segmentations using these
The most generally accepted form of entropy was
methods depend on initial segmentations which are not
derived by Shannon (Shannon and Weaver, 1949) [36] in
needed by thresholding. One common trait of all
connection with information theory. Given a discrete
histogram based approaches is that they do not utilize the
considerable amount of information that is captured in the probability distribution Pi , i = 1,2,..., N , the entropy is
spatial distribution of intensities and in image morphology. given by
It is obvious that, in real-life imaging applications, it is n n
very difficult to select a threshold from the histogram only H = −∑ pi log( pi ) , ∑ pi = 1 , (1)
without seeing the image, while the latter has a clear i =0 i =0
object morphology. Yin [29] developed a recursive 0 ≤ pi ≤ 1
programming techniques to reduce the order of magnitude
of computing the multilevel thresholds and further used
Where i = 1,2,..., N , are a set of possible outcomes or
the PSO algorithm to minimize the cross entropy.
Since the theory of entropy was brought into the states of a discrete information source modelled as a
thresholding selection technology, many methods have Markov process. Shannon's measure is used as a measure
been proposed to deal with the problem of image of information gain, choice and uncertainty.
segmentation. The maximum entropy method is proved to Shannon points out a number of interesting properties of
do good results for the infrared image segmentation. But (1), including:
it needs to compute the entropy of every gray value. Over i) H=0 if and only if pi = 0 ∀i ≠ j , Pj = 1, where j
the last decade, modeling the behavior of social insects, can indicate any position in the distribution. Otherwise H
such as ants and bees, for the purpose of search and is positive. This makes intuitive sense, since
problems solving has been the context of the emerging
area of swarm intelligence. Ant colony algorithm [30] and Pj = 1 indicates certainty of the outcome, and the
particle swarm optimization [31] are two most popular information gained by the occurrence of event j is thus
approaches in swarm intelligence. The honey bee mating zero.
optimization [32] may also be considered as a typical ii) For a given number of discrete states N, H is a
swarm-based approach for optimization, in which the maximum when all the Pi are equal, i.e. Pi = 1 N ,
search algorithm is inspired by the process of mating in
real honey bees. In the literature, the honey bee mating i = 1,2,..., N . Intuitively this is the most uncertain
optimization algorithm had been adopted to search for the situation: all outcomes are equally likely, making accurate
optimal solution in many applications such as clustering prediction impossible.
[33], market segmentation [34] and benchmark
mathematical problems [35]. Taking the consideration the 2.2. Image Thresholding Based on the Maximum
complexity of its computation, we proposed a new Entropy
heuristic optimization algorithm, called the bee algorithm
to search the result for infrared image segmentation. In general, an image can be described by a discrete
The bee algorithm may also be considered as a typical function. For discrete values we deal with probabilities
swarm-based approach for optimization, in which the and summations. The probability of occurrence of gray
search algorithm is inspired by the real bees. The behavior level i in an image is approximated by
of bees is the interaction of their (1) genetic potentiality,
ni
(2) ecological and physiological environments, and (3) the Pi = , i = 0,1,2,..., L − 1. (2)
social conditions of the colony, as well as various prior n
and ongoing interactions between these three parameters.
This paper introduces a new approach for optimization of
28 Azarbad et al.

where, as noted at the beginning of this section, n is the i = 1,2,..., N . Effectively, by maximizing the entropy, we
total number of pixels in the image, ni is the number of are attempting to maximize our information gain or,
pixels that have gray level i , and L is the total number of equivalently, arrive at a solution which fits our prior
possible gray levels in the image (L=256). If we consider a knowledge but makes no assumptions beyond what is
threshold that is named T (0<T<L-1) we have two known. In the absence of any prior information, the
maximum entropy distribution is simply the uniform
regions CO , C B . Where, CO is the object region and distribution, as indicated by Shannon's point (ii) above. It
C B is the background region. Shannon defines the entropy can thus be aptly viewed as an application of Laplace's
of a system, which has n status, as principle of insufficient reason, which states that the
uniform distribution is the most unbiased when one has no
n prior knowledge regarding a probabilistic event. We solve
H = −∑ pi log( pi ) (3) equation (6) to get the optimized threshold (T = t )
*
i =0
which can make H (t ) maximum. The threshold T is

Where pi is the probability of the occurrence of the selected as the one which maximizes H (t ) .

event i , and the entropy H is the measuring of the system T = t * = Arg (max H (t )) , (7)
information. The value of the information obtained from an
Where 0 ≤ T = t ≤ 255 , directly to solve the
*
event is the inverse probability of the occurrence. Now
formula (2) is analyzed from the point of an image. The equation (7) will cost lots of time. In order to enhance the
event i can be regarded as a gray value of an image while speed and accuracy of the proposed algorithm, the bee
pi is the probability of the pixel being i . The theory of algorithm is used to extract the optimized threshold.
maximum entropy is to select i which makes H be 3. The Bee Algorithm
maximum value [37]. pO , p B are denoted as the
probabilities of grey levels of the object and background A colony of honey bees can extend itself over long
regions respectively. distances (more than 10 km) and in multiple directions
simultaneously to exploit a large number of food sources
pO = ∑p
i∈CO
i
, Where i = 0,1,2,..., T − 1; (4-1) [38-39]. A colony prospers by deploying its foragers to
good fields. In principle, flower patches with plentiful
amounts of nectar or pollen that can be collected with less
pB = ∑p
i∈C B
i
, (4-2) effort should be visited by more bees, whereas patches
with less nectar or pollen should receive fewer bees [40-
41]. The foraging process begins in a colony by scout bees
Where i = [T , T + 1, T + 2,..., 255] and the entropy being sent to search for promising flower patches. Scout
of the object region and the background region can be bees move randomly from one patch to another. During the
denoted as follows: harvesting season, a colony continues its exploration,
T −1 keeping a percentage of the population as scout bees [39].
H O (T ) = −∑ [( pi pO ) log( pi pO )], (5-1) When they return to the hive, those scout bees that found a
i =0 patch which is rated above a certain quality threshold
(measured as a combination of some constituents, such as
255 sugar content) deposit their nectar or pollen and go to the
H B (T ) = −∑ [( pi p B ) log( pi p B )], (5−2 ) “dance floor” to perform a dance known as the “waggle
i =T dance” [38].
The bee algorithm is an optimization algorithm inspired
by the natural foraging behavior of honey bees to find the
Thus, the function of the entropy is: optimal solution [42]. Table 1 shows the pseudocode for
the algorithm in its simplest form. The algorithm requires a
number of parameters to be set, namely: number of scout
Fitness = H (t ) = H O (t ) + H B (t ) , (6) bees (n), number of sites selected out of n visited sites (m),
number of best sites out of m selected sites (e), number of
The maximum entropy formalism attempts to bees recruited for the other (m-e) selected sites (n1),
maximize the equation (6), subject to known constraints number of bees recruited for best e sites (n2), initial size of
about the parameters we are trying to estimate Pi , patches (ngh) which includes site and its neighborhood and
Segmentation of Infrared Images and Objectives Detection Using Maximum Entropy Method Based on the Bee Algorithm 029

stopping criterion. The algorithm starts with the n scout Table 1. Parameters of the Bee Algorithm
bees being placed randomly in the search space.
The Bee Algorithm Parameters Symbol Value
According to the fitnesses of the sites visited by the
scout bees are evaluated. In step “select m sites for
neighborhood search”, bees that have the highest fitnesses Number of scout bees n 30
are chosen as “selected bees” and sites visited by them are
chosen for neighborhood search. Then, in steps “recruit Number of best selected sites m 25
bees for selected sites” and “select the fittest bee from each
patch”, the algorithm conducts searches in the
neighborhood of the selected sites, assigning more bees to Site radius for neighbourhood search ngh 1
search near to the best e sites. The bees can be chosen
directly according to the fitnesses associated with the sites Number of elite sites out of m selected e 20
they are visiting. Alternatively, the fitness values are used sites
to determine the probability of the bees being selected.
Searches in the neighborhood of the best e sites which Number recruited bees around best n1 15
represent more promising solutions are made more detailed selected sites
by recruiting more bees to follow them than the other
Number of recruited bees around elite n2 20
selected bees. Together with scouting, this differential selected sites
recruitment is a key operation of the bee algorithm.
However, in step “select the fittest bee from each patch”,
for each patch only the bee with the highest fitness will be
selected to form the next bee population. In nature, there is I n pu t I m a g e

no such a restriction. This restriction is introduced here to

reduce the number of points to be explored. In step “assign E v a lu a t io n o f t h e M a x im u m E n t r o p y
F u n c t io n
the (n-m) remaining bees to random search”, the remaining
bees in the population are assigned randomly around the
I n i t ia li se a P o p u l a t io n o f n S c o u t B e e s
search space scouting for new potential solutions. These
steps are repeated until a stopping criterion is met.
E v a lu a t e t h e F i t n e ss o f t h e P o p u la t io n

3.2. The Determination of the Bee Algorithm

S e le c t m S it e s f o r N e ig h b o u r h o o d S e a rc h
Parameters
D e t e r m in e t h e S iz e o f N e ig h b o u r h o o d
In this section, the proposed algorithm is employed as ( P a t c h S iz e n g h )
an optimized method to extract ideal threshold.
Considering the maximum entropy method and bee R e c r u it B e e s f o r S e le c t e d S it e s
algorithm together, we set the threshold vector (t) as the ( m o r e B e e s f o r t h e B e st e S i t e s )
sites, and H(t) as the fitness to guide the search. After
getting the histogram of the image, we adopt the bee S e le c t t h e F i t t e s t B e e f r o m E a c h S i t e
algorithm procedure to explore the optimum result of (t*)
which can produce the maximum fitness. Then, the A ss ig n t h e ( n – m ) R e m a i n i n g B e e s t o
infrared image can be segmented according to the value of R a n do m S ea r ch
(t*). All programs are designed by using the Matlab
program. In the simulation of the proposed algorithm, it N e w P o p u l a t io n o f S c o u t B e e s
needs many parameters that are predefined. Table.1 shows
predefined values of parameters for the max-entropy & bee
No
algorithm and also Fig. 1 describes the process of the H (t )
proposed algorithm. The values were determined m a x im u m ?
empirically in table. 1.
Yes

T h r e s h o ld in g
&
S eg m e nta ti o n

S e g m en te d I m a g e

Figure 1. Flow chart of the proposed method

30 Azarbad et al.

4. Results images. It is seen from the segmented image that the

computed thresholds are our favorite results.
The manual segmented results of six different images Table 2. Experimental Results
based on the maximum entropy that is supplied by the bee
algorithm are shown in Fig. 2 and Fig. 3. Considering the
segmented images, we think the results obtained by our Image Computed Threshold Number of Iterations
method can be regarded as reasonably good and applicable
in subsequent processing. Experiments on infrared images FLOWERS 128 200
segmentation are used to illustrate the validation of the
proposed algorithm. The proposed algorithm is tested on STEALER 129 600
the six different original images. The original images are
named FLOWERS, STEALER, DOGS, AIRPLANE, DOGS 124 50
SHIP and TANK images. And the images used contains
(500*375), (320*240), (350*155), (320*240), (255*195) AIRPLANE 132 100
and (300*202) pixels respectively, which are shown as Fig.
2 and Fig. 3. SHIP 127 50
Table.2  presents the segmentation results that are
TANK 123 50
obtained by the proposed Algorithm for the six different
It is possible to reveal from the selected thresholds  are illustrated to show that the proposed method can get
in table.2 that the maximum entropy image segmentation ideal segmentation result with less computation cost. The
based on the bee algorithm can find a better threshold and presented detection method is not limited to infrared
because of its selected threshold, can detect and extract images segmentation. On the other hand, this approach
the favorite object from the background of the image. could be extended and modified to segmentation and
Also, the experiments of segmenting the infrared images detection of other types of standard images.

Figure 2. (a) Original Image of FLOWERS (b) Segmented Image of FLOWERS (c) Histogram of FLOWERS Image (d) Original Image of STEALER
(e) Segmented Image of STEALER (f) Histogram of STEALER Image (g) Original Image of DOGS (h) Segmented Image of DOGS (l) Histogram of
DOGS Image
Segmentation of Infrared Images and Objectives Detection Using Maximum Entropy Method Based on the Bee Algorithm 031

(c)
(a)  (b)

(f) 
(d)  (e)

(g)  (h) (l) 

Figure 3. (a) Original Image of AIRPLANE (b) Segmented Image of AIRPLANE (c) Histogram of AIRPLANE Image (d) Original Image of SHIP
(e) Segmented Image of SHIP (f) Histogram of SHIP Image (g) Original Image of TANK (h) Segmented Image of TANK (l) Histogram of TANK Image

Fig. 2 shows the segmentation results obtained on the proposed approach. In the second row, the image
FLOWERS, STEALER and DOGS images by the “STEALER” mainly consists of one ideal object, the
proposed algorithm. For original images in the first stealer man, and we have decided to detect that from the
columns (Fig. 2), the segmentation results are shown in background objects. It is seen from the segmented image
the second columns and the histograms of the original that the proposed algorithm succeeded to detect the stealer
images are shown in the third column in Fig. 2. In the first man as black colour and segmented the background as
row, the flowers of the image were clearly segmented as white colour, as shown as in Fig. 2(e). In the third row, the
black and white colours, as shown in Figs. 2(b). And also image “DOGS” includes two noticable objects, two dogs,
the background information was segmented as black that should be segmented and extracted from the
colour. As seen from the first row of Fig. 2, the main background. According to the segmented image in Fig.
features of the image were clearly detected by the
32 Azarbad et al.

2(h), two dogs were evidently segmented as white colour REFERENCES

and the background was detected as black colour.
[1] H.D. Cheng, J.R. Chen, J. Li, Threshold selection based on fuzzy c
Fig. 3 shows the segmentation results obtained on the -partition entropy approach to image thresholding, Pattern
AIRPLANE, SHIP and TANK images by the proposed Recognition 31 (1998) 857– 870.
algorithm. For original images in the first columns (Fig. 3), [2] F. Du, W.K. Shi, L.Z. Chen, Y. Deng, A. Zhu, Infrared image
the segmentation results are shown in the second columns segmentation with 2-D maximum entropy method based on particle
and the histograms of the original images are shown in the swarm optimization (PSO), Pattern Recognition Letters 26 (2005)
597–603.
third column in Fig. 3. In the first row, the original image
[3] V. Boskovitz, H. Guterman, An adaptive neuro-fuzzy system for
“AIRPLANE” consists of two favorite objects, the airplane automatic image segmentation and edge detection, IEEE
and the light bulb. According to the segmented image, the Transactions on Fuzzy Systems 19 (2002) 47–262.
ideal objects were clearly segmented as white color, as [4] F. Kurugollu, B. Sankur, A.E. Harmanci, Color image
shown in Figs. 3(b). And also the background information segmentation using histogram multi-thresholding and fusion, Image
was segmented as black color. As seen from the first row and Vision Computing 19 (2001) 915–928.
of Fig. 3, the main features of the image were evidently [5] C. Strouthopolos, N. Papamarkos, Multi-thresholding of mixed-
detected by the proposed approach. In the second row, the type documents, Engineering Applications of Artificial Intelligence
13 (2000) 323–343.
image “SHIP” mainly consists of one ideal object, the ship,
[6] Sahoo, P.K., Soltani, S., Wong, A.K.C., and Chen, Y.C.: ‘A survey
and we have decided to extract it from the background of thresholding techniques’, Comput. Vis. Graph. Image Process.,
objects. It is seen from the segmented image that the 1988, 41, pp. 233–260
proposed algorithm succeeded to detect the ship as white [7] Lee, S.U., Chung, S.Y., and Park, R.H.: ‘A comparative
color and segmented the background as black color, as performance study of several global thresholding techniques for
shown as in Fig. 3(e). In the third row, the image “TANK” segmentation’, Comput. Vis. Graph. Image Process., 1990, 52, pp.
includes one ideal object, the tank, which should be 171–190
segmented and extracted from background. According to [8] Haralick, R.M., and Shapiro, L.G.: ‘Computer and robot vision, vol.
the segmented image in Fig. 3(h), the tank was evidently 1’ (Addison-Wesley, 1992)
segmented as white color and the background was detected [9] Pal, N.R., and Pal, S.K.: ‘A new definition and its application’,
IEEE Trans. Syst. Man Cybern., 1991, 21, (5), pp. 1260–1270
as black color.
[10] Sahoo, P.K., Slaaf, D.W., and Albert, T.A.: ‘Threshold selection
using a minimal histogram entropy difference’, Opt. Eng., 1997, 36,
5. Conclusions pp. 1976–1981
[11] C.A. Glasbey, “An Analysis of Histogram Based Thresholding
An automatic thresholding of gray-level images using Algorithms,” Proc. Graphical Models and Image Processing, vol.
maximum entropy and bee algorithm is proposed. The 55, pp. 532-537, 1993.
proposed approach is based on the bee algorithm to [12] J.M.S. Prewitt and M.L. Mendelsohn, “The Analysis of Cell
Images,” Ann. New York Academy Science, vol. 128, pp. 1035-
optimization of the maximum entropy method for 1053, 1966.
segmentation of infrared images. The goal of this research [13] W. Doyle, “Operations Useful for Similarity-Invariant Pattern
was to develop a better segmentation algorithm. In order to Recognition,” J. ACM, vol. 9, pp. 259-267, 1962.
achieve this goal, we developed an important thresholding [14] Cover, T., and Thomas, J.: ‘Elements of information theory’ (John
method based on a recent evolutionary algorithm. The Wiley & Sons, Inc., 1991)
proposed algorithm simulates the behavior of honey bees [15] Yue Wang, Ding- gang Shen, and Eam Khwang Teoh, “ Lane
to develop the algorithm to search for the optimum detection and track using B-Snake,” Image Vision Comput.2004,
thresholds for segmentation of images. With the help of the 22(4).
bee algorithm, the favorite threshold can be obtained easily [16] Joel C. McCall, Mohan M. Trivedi, “An Integrated, Robust
with high efficiency. The selected threshold can be Approach to Lane Marking Detection and Lane Tracking,” Proc,
IEEE Intelligent Vehicle Symposium, June 2004.
segmented images and evidently extracted the ideal objects
[17] C.H. Li and C.K. Lee, “Minimum Entropy Thresholding,” Pattern
from background. Experimental results on a large number Recognition, vol. 26, pp. 617-625, 1993.
of different kinds of infrared images show that the
[18] J. Kittler and J. Illingworth, “Minimum Cross Error Thresholding,”
proposed method performs fairly well in term of the Pattern Recognition, vol. 19, pp. 41-47, 1986.
segmentation and the objectives detection quality. The [19] S.M. Dunn, D. Harwood, and L.S. Davis, “Local Estimation of the
proposed segmentation method is not limited to infrared Uniform Error Threshold,” IEEE Trans. Pattern Analysis and
images segmentation. On the other hand, this approach Machine Intelligence, vol. 1, pp. 742-747, 1984.
could be extended and modified to segmentation of other [20] C.K. Leung and F.K. Lam, “Maximum Segmented Image
types of standard images. This result is promising and it Information Thresholding,” Graphical Models and Image
encourages further research for applying the maximum Processing, vol. 60, pp. 57-76, 1998.
entropy and the bee algorithm to develop algorithms for [21] Yang, C.-W., Chung, P.-C., and Chang, C.-I: ‘A hierarchical fast
two-dimensional entropic thresholding algorithm using a histogram
image processing and pattern recognition. pyramid’, Opt. Eng., 1996, 35, (11), pp. 3227–3241
[22] Otsu, N.: ‘A threshold selection method from grey-level
histograms’, IEEE Trans. Syst. Man Cybern., 1979, SMC-9, (1), pp.
62–66
Segmentation of Infrared Images and Objectives Detection Using Maximum Entropy Method Based on the Bee Algorithm 033

[23] A.K.C. Wong and P.K. Sahoo, “A Gray-Level Threshold Selection [25] N.R. Pal and S.K. Pal, “Entropy: A New Definition and Its
Method Based on Maximum Entropy Principle,” IEEE Trans. Applications,” IEEE Trans. Systems, Man, and Cybernetics, vol. 21,
Systems, Man, and Cybernetics, vol. 19, pp. 866-871, 1989. pp. 1260-1270, 1991.
[24] N.R. Pal and S.K. Pal, “Entropy Thresholding,” Signal Processing, [26] F.C. Cheng and J.W. Woods, “Compound Gauss-Markov Random
vol. 16, pp. 97-108, 1989. Fields for Image Segmentation,” IEEE Trans. Signal Processing,
vol. 39, pp. 683-697, 1991.
[27] A. Sarkar, M.K. Biswas, and K.M.S. Sharma, “A Simple [40] Bonabeau E, Dorigo M, and Theraulaz G. Swarm Intelligence:
Unsupervised MRF Model Based Image Segmentation Approach,” from Natural to Artificial Systems. Oxford University Press, New
IEEE Trans. Image Processing, vol. 9, pp. 801-812, 2000. York, 1999.
[28] M. Kass, A. Witkin, and D. Terzopoulos, “Snakes: Active Contour [41] Camazine S, Deneubourg J, Franks NR, Sneyd J, Theraula G and
Models,” Int’l J. Computer Vision, pp. 321-339, 1988. Bonabeau E. Self-Organization in Biological Systems. Princeton:
[29] P.Y. Yin, Multilevel minimum cross entropy threshold selection Princeton University Press, 2003.
based on particle swarm optimization, Applied Mathematics and [42] Eberhart, R., Y. Shi, and J. Kennedy, Swarm Intelligence. Morgan
Computation 184 (2007) 503–513. Kaufmann, San Francisco, 2001.
[30] Y.F. Han, P.F. Shi, An improved ant colony algorithm for fuzzy
clustering in image segmentation, Neurocomputing 70 (2007) 665–
671.
[31] J. Kennedy, R.C. Eberhart, Particle swarm optimization, Authors’ information
in:Proceeding of IEEE International Conference on Neural
1
Network, vol. IV, 1995, pp. 1942–1948. Faculty of Electrical and Computer Engineering, BUT.
2
[32] H.A. Abbasss, Marriage in honey bee optimization (HBO): a Faculty of Electrical and Computer Engineering, BUT.
3
haplometrosis polygynous swarming approach, in: The Congress Faculty of Electrical and Computer Engineering, BUT.
on Evolutionary Computation (CEC001), 2001, pp. 207–214.
[33] M. Fathian, B. Amiri, A. Maroosi, Application of honey bee mating
optimization algorithm on clustering, Applied Mathematics and Milad Azarbad was born in Amol. He received his B.S.
Computations (2007) 1502–1513. degree in electrical engineering from Mazandaran in 2010.
[34] B. Amiri, M. Fathian, Integration of self organizing feature maps His areas of researches are: General Area of Signal
and honey bee mating optimization algorithm for market Processing, Biomedical Engineering, Biomedical Image
segmentation, Journal of Theoretical and Applied Information Processing, Artificial Intelligence, Statistical Pattern
Technology (2007) 70–86. Recognition, Digital Communications and soft computing.
[35] D. Karaboga, B. Basturk, On the performance of artificial bee
colony algorithm, Applied Soft Computing (2008) 687–697.
[36] Shannon, C.E. and W. Weaver (1949). The Mathematical Theory Ataollah Ebrahimzadeh was born in Babol. He
of Communication. Univ. of Illinois Press, Urbana, IL. received his PhD degree in electrical engineering. Now
he is a professor in the Faculty of Electrical and
[37] Gonzalez, R. C., & Woods, R. E. (1992). Digital image processing
Computer Engineering at University of Mazandaran.
(3nd Edition 2006)
His research interests are: General Area of Signal
[38] Von Frisch K. Bees: Their Vision, Chemical Senses and Language. Processing, Wireless Communications, Biomedical
(Revised edn) Cornell University Press, N.Y., Ithaca, 1976. Engineering, Statistical Pattern Recognition, Artificial
[39] Seeley TD. The Wisdom of the Hive: The Social Physiology of Intelligence, Digital Communications.
Honey Bee Colonies. Massachusetts: Harvard University Press,
Cambridge, 1996.