International Journal of Engineering and Applied Sciences (IJEAS)

ISSN: 2394-3661, Volume-3, Issue-3, March 2016

Optimization of Fuzzy C Means with Darwinian

Particle Swarm Optimization on MRI Image
A.Murugan, M.Leelavathi, A.P.Shivadharshini, P.Kousalya, B.Gayathridevi

into clusters and can be hard or soft. Soft clustering is

Abstract Image segmentation is one of the most important supreme as every pixel can be assigned to all clusters with
and most difficult low-level image analysis tasks. Automatic different membership values. The most popular soft
target recognition (ATR) often uses segmentation to separate the clustering methods applied to MR images are fuzzy -means
desired target from the background. Fuzzy c-means (FCM) (FCM) clustering, mixture modeling, and hybrid of both the
algorithm is one of the most popular fuzzy clustering techniques
methods.
because it is efficient, and easy to implement. Fuzzy clustering is
a main problem which is the subject of dynamic research in
several real world applications. However, FCM is sensitive to II. FUZZY C-MEANS ALGORITHM
initialization and is easily trapped in local optima. In this paper, K-means is one of the most accepted hard clustering
DPSO is used to escape from local optima and to determine the algorithms which separates the data objects into k clusters
global optima which are calculated on comparing with single
where the number of clusters, k, is determined in advance
swarm and similar set of swarms, operating on the test problem
obtained for PSO according to the given image. This model is unsuitable for
real data sets in which there are no distinct boundaries
Index Terms Darwinian Particle Swarm Optimization, between the clusters.
Fuzzy clustering, Particle Swarm Optimization. After the fuzzy theory launched by LotfiZadeh [1], who
situates the fuzzy theory into clustering. Fuzzy algorithms can
allocate data object partially to multiple clusters. The degree
I. INTRODUCTION of membership in the fuzzy clusters depends on the nearness
Image segmentation is the method of separating the main of the data object to the cluster centers.
objects from background objects and other objects with The most accepted fuzzy clustering algorithm is fuzzy
respect to one or more characteristics. It is mandatory and c-means (FCM) which was introduced by Bezdek[2] in 1974
most difficult low-level image analysis tasks. and now it is universally used. Fuzzy c-means clustering is an
Segmentation plays a very important role in medical efficient algorithm; but the random selection in center points
image processing and it is used in many applications, for makes iterative process diminishing into the local optimal
instance detection of tumors, detection of the coronary solution easily.
border, measuring tumor volume and its volumetric response Fuzzy c-means separations set of n objects p = {p1,p2 pn} in
to therapy, classification of blood cells, detection of micro Rd dimensional space into c (1 <c <n) fuzzy clusters with o =
calcification on mammograms, surgical planning, heart image {o1, o2...,oc} cluster centers or centroids. The fuzzy clustering
extraction from cardiac cine angiograms, etc. of objects is described by a fuzzy matrix with n rows and c
In recent years, many algorithms have been columns in which n is the number of data objects and c is the
incorporated in MRI segmentation. The most accepted number of clusters. ij, the element in the ith row and jth
methods are thresholding, region-growing and clustering, column in , indicates the degree of association or
Edge detection, and model-based methods. Thresholding is membership function of the ith object with the jth cluster. The
one of the most popular segmentation approaches because of characters of are as follows:
its simplicity.
Clustering analysis is frequently used as a vital tool to sort
collection of objects into homogeneous groups, which allows
discovering similarities and dissimilarities. Clustering is an
unsupervised learning strategy that groups related patterns
0
A.Murugan, Professor, Department of Electronics And Communication
Engineering, Excel College Of Engineering And Technology, The objective function of FCM algorithm is to minimize the
Komarapalayam, Namakkal Eq. (4):
M.Leelavathi, UG Student, Department of Electronics And
Communication Engineering, Excel College Of Engineering And
Technology, Komarapalayam, Namakkal
A.P.Shivadharshini, UG Student, Department of Electronics And
Communication Engineering, Excel College Of Engineering And Where,
Technology, Komarapalayam, Namakkal
P.Kousalya, UG Student, Department of Electronics And Communication
Engineering, Excel College Of Engineering And Technology,
Komarapalayam, Namakkal in which, m (m>1) is a scalar termed the weighting proponent
B.Gayathridevi, UG Student, Department of Electronics And and controls the fuzziness of the ensuing clusters and dij is the
Communication Engineering, Excel College Of Engineering And Euclidian distance from object pi to the cluster center Oj. The
Technology, Komarapalayam, Namakkal
Oj, centroid of the jth cluster, is obtained using Eq. (6).

35 www.ijeas.org
 Optimization of Fuzzy C Means with Darwinian Particle Swarm Optimization on MRI Image

improvement over Ant Systems are Decision Rule -

Pseudorandom proportional rule, Local Pheromone Update,
Best only offline Pheromone Update. The Ant Colony
Optimization is applied over many fields such as Routing in
Fuzzy c-means clustering is an effective algorithm, but the telecommunication networks, Traveling Salesman, Graph
random selection in center points makes iterative process Coloring, Scheduling and Constraint Satisfaction. Like Wise,
falling into the local optimal solution easily. For solving this ACO plays a vital role in Optimization Method because of
problem, recently evolutionary algorithms such as genetic following advantages namely, Inherent parallelism, Positive
algorithm (GA), simulated annealing (SA), ant colony Feedback accounts for hurried discovery of good Solutions,
optimization (ACO), and particle swarm optimization (PSO) capable for Traveling Salesman Problem and which can be
have been successfully applied. used in dynamic applications. (Adapts to changes such as new
distances, etc)

III. ANT COLONY ALGORITHM VI. PARTICLE SWARM OPTIMIZATION

Ant colony method is a system for optimization that Ant Colony Optimization method contribute some
was launched in the premature 1990s. Ant (ACO) Colony Disadvantages such as Theoretical analysis is difficult
Optimization is a currently projected meta-heuristic approach compared with other Optimization methods, Sequences of
for solving hard combinatorial optimization problems. The random results (not independent), Probability distribution
inspiring source of ACO is communication medium. During modifies by iteration, Research is heuristic rather than
the algorithms affecting to reflect their search experience, the theoretical, Time to convergence uncertain (but convergence
first model of such an algorithm is Ant System (AS). It had the is Guaranteed!).To overcome these problems the Particle
important role of stimulating further research on algorithmic Swarm Optimization is widely preferred. An Experimental
variants obtain much better computational performance, as results show that the PSO image classifier performs better
well as on applications[4] to a large variety of different operation than state-of-the-art image classifiers (viz,
problems like the quadratic assignment, vehicle routing, K-means, Fuzzy C-means, K-Harmonic means and Genetic
sequential ordering, scheduling, routing in Internet-like Algorithms) in all measured criteria. The influence of
networks, and so on . Motivated by this success, the ACO dissimilar values of PSO manage parameters on concert is
meta-heuristic has been projected as a common framework also established.
for the existing one application and algorithmic variants. OVERVIEW
Algorithms which follow the ACO meta-heuristic will be Particle Swarm Optimization is an evolutionary computation
described in the following ACO algorithms. Recent technique developed by Eberheart & Kennedy [6] in 1995 and
applications of ACO algorithms [5] fall into the two important is based on bird gathering and fish schooling. PSO is a
problem classes of static and dynamic combinatorial meta-heuristic technique as it makes few assumptions about
optimization problems. Static harms are those whose the problem being optimized and can explore very large
topology and cost do not modify while the problems are being spaces of candidate solutions. Its simplicity and faster
solved. Differently, in dynamic problems the topology and convergence make it an attractive algorithm to employ. The
costs can modify while solutions are built. An example of population is called swarm and the individuals are termed as
such a problem is routing in telecommunications networks, in particles. The word swarm is inspired from pointed faction
which traffic patterns change all the time. The ACO of particles [7] in the problem region. The particles are
algorithms for cracking these two classes of problems are very implicit to be mass-less and volume-less.
similar from a high-level perspective, but they differ IMPLEMENTATION
signicantly in implementation details. The dimension of the problem is dogged by the
number of generating units. Then the current position of ith
IV. IMPLEMENTATION OF ACO particle can be represented by Pi = [Pi1, Pi2,Pi3PiN] where Pi
belongs to problem space S. The particle flies with the current
Probabilistic technique, searching for optimal velocity given by V = [Vi1, Vi2,.. ViN] which is generated
path in the graph based on behavior of ants seeking a path randomly in the range of [-Vjmax,Vmax]. The objective function
among their colony and Source of food. In Meta-heuristic morals are calculated and are set as values of the particles.
optimization, Ants navigate the shortest path is exposed via The best rate, based on individual fitness function is denoting
pheromone trails. Each ant moves at random Pheromone is by or the global best value of swarm. New velocity for every
dumped on path more pheromone on path raises probability of dimension in each particle is updated as,
path being followed. This problem might seem trival to us, but
to the certain inferior specious, this considered Hard.
Where w is the weight vector whose value is to be suitably
V. OVER VIEW OF THE SYSTEM chosen, c1&c2 are constants, rand is a uniformly circulated
Virtual trail accumulated on path segments. Path
preferred at random based on amount of "trail" present, on arbitrary value between [0 1], t represents iteration and the
probable paths from initial node. Ant attained next node, current position of the jth dimension in the ith particle. The new
chooses next path that keeps on until reaches starting node. particle position is given by
Entire tour is a solution. Tour is examined for optimality. Pnew = Pold + Vnew [7]
ACO is meta-heuristic three Soft computing techniques for When the stopping criteria are satisfied and there is no further
solving hard discrete optimization problems, First major improvement in the objective function, the position of the

36 www.ijeas.org
 International Journal of Engineering and Applied Sciences (IJEAS)
ISSN: 2394-3661, Volume-3, Issue-3, March 2016
particles represented by gives the optimal best G gives the Critical threshold of search counter
optimal dispatch.
DARWINIAN PARTICLE SWARM OPTIMIZATION (DPSO) FLOWCHART:
Some of the particular algorithm like PSO could not able to
work over many problems at the homogeneous time. Such that
the PSO is used to test the solution only in the single swarm, it
doesnt have the capability to compare the set of swarms for
the analysis which leads to failure in time (Time-Out)[8] and
restart the algorithm or delete the information obtained in the
global optimum hopes that it will not return to it. Angelin [9]
executed a type of selection process. Hence, to overcome
these problems DPSO is raised.
This DPSO is used to test the solution by comparing all the
set of swarms at any instant. It carries out its specific function
on each swarm similar to that of PSO which hold on the
respected rules of governing the collection of swarms for
simulating the natural selection. The collection of swarms
[10] is altered constantly which can be done by the selection
of swarms through the implementation of the selection
process.
IMPLEMENTATION OF DPSO:
Some of the assumption adapted in DPSO is
spawning a swarm, favorable adaptation and unfavorable
adaptation. The simple ideas are implemented similar to the
natural selection. The swarm is taken to the prolonged
survive, has more chance of possessing offspring. The swarm
will have its exceeded life time to find more fit state like
favorable adaptation and in contrast, the life time of swarm is
reduced for failing to find fit state like unfavorable
adaptation.
ALGORITHM DETAILS IN DPSO:
i. Initialize the appropriate range of particle array
element and velocities .
ii. Allow the particle to traverse at .
iii. Evolve swarm algorithm, to find new global fitness.
iv. Checks the conditions as follows
If is decreased, the swarm is deleted. A
swarm particle population encircle with

If maximum critical threshold

exceeds, a particle is detected from the
VII. DISSCUSSION
swarm at scheduled time.
v. Collection of swarm is maintained. The main result of the paper is that Darwinian PSO
vi. Each swarm may spawn a new swarm. helped get around local optima in the selected experiments.
If =0, then a new swarm can able to be produced Since this work is preliminary and the algorithm personalized
only velocity control, it is our opinion that the result could be
with the probability of
improved by mounting the set of possible adaptations. In the
Where, uniform random number on [0,1] upper panel, the algorithm is seen to achieve an on average
Number of swarm steady state number of swarms throughout the experiment. In
To compress number of swarm creation in the lower panel, the Darwinian swarms simply peak and then
existence (at large number of swarm) die. It is possible that the Darwinian swarms were not given
The main function of reduction in tolerance for stagnation is ample time to search the fitness landscape. Since the number
performed to maintain the collection of swarms which of swarms in the population at any given instant is controlled
improves in active manner. In order to chosen the value of SC by, in addition to the fitness landscape, the parameter P
in the reset period at the number of particles got deleted in introduced adaptation of the parameter P would be beneficial.
certain period, the expression is described as follows: Since the computing demands of this algorithm are higher
than a pre-engineered single swarm algorithm, a
high-performance computing platform is desirable. The
Darwinian PSO algorithm can be parallelized at two levels.
Where,
The individual swarms particles could be distributed across a
Amount of the particle deleted from a swarm over a
cluster, which is beneficial when the fitness computation is
period in which there is no improvements in fitness.
lengthy. Second, the swarms could be distributed as well. The

37 www.ijeas.org
 Optimization of Fuzzy C Means with Darwinian Particle Swarm Optimization on MRI Image

high Computational demands of the algorithm motivated our

choice for a lower number of Trials (15 per test function).

SUGGESTION
In this paper we are using ant colony algorithm for
optimization and also how it is differentiated from DPSO.
Instead of using ant colony algorithm, for beneficial purpose
suggested that bee colony algorithm, bacteria colony
algorithm, canalized Fuzzy c- means algorithm. It may give
the efficient optimized results.

REFERENCES
[1] Lotizadeh, 1965Fuzzy sets and system, IN: Fox J, editor. System
theory. Brooklyn, NY: Polytechnic Press, 1965:29-39.
[2] J.Bezdek, Fuzzy Mathematics in pattern classification, Ph.D. Thesis,
Ithaca, NY: Cornell University, (1974)
[3] Aden Jos-Garcia, Wilfrido Gmez-Flores. (2016) automatic
clustering using nature-inspired Metaheuristics: A survey. Applied
Soft Computing 41, 192-213.
[4] M. Dorigo, M. Birattari, T. Sttzle, Ant Colony Optimization
Artificial Ants as a Computational Intelligence Technique, IEEE
Computational Intelligence Magazine,2006
[5] M. Dorigo K. Socha, An Introduction to Ant Colony Optimization,
T. F. Gonzalez, Approximation Algorithms and Metaheuristics, CRC
Press, 2007
[6] M. Dorigo T. Sttzle, The Ant Colony Optimization Metaheuristics:
Algorithms, Applications, and Advances, Handbook of
Metaheuristics, 2002
[7] J. Kennedy, R. Eberhart, Particle swarm optimization, in:
Proceedings IEEE International Conference on Neural Networks, vol.
4, Perth, Australia, 1995, (pp. 19421948).
[8] K.S.Chaung, H.L.Hzeng, C.Chen, J.Wu.T.J.Chen Fuzzy c means
clustering with spatial information for image segmentation,
computerized Medical Imaging and Graphics 30 (2006)
[9] Clerc, M. The Swarm and The Queen: towards a deterministic and
adaptive particle swarm optimization in Congress on Evolutionary
Computation.1999.Washington, DC.
[10] Angeline, P.J. Using selection to Improve particle swarm
optimization in IEEE World Congress on Computational
Intelligence.1988.Anchorage, Alaska and USA.
[11] Kennedy, J and R.C.Eberhart. A discrete binary version of the particle
swarm algorithm in systems, Man and Cybernetics.1997.Piscataway,
NJ: IEEE Service Center.

38 www.ijeas.org