IOP Conference Series: Materials Science and Engineering

PAPER • OPEN ACCESS

Image clustering using genetic algorithm with tournament selection and

uniform crossover
To cite this article: Gevin Valerian et al 2020 IOP Conf. Ser.: Mater. Sci. Eng. 852 012043

View the article online for updates and enhancements.

This content was downloaded from IP address 185.30.169.17 on 21/07/2020 at 18:22

2nd TICATE 2019 IOP Publishing
IOP Conf. Series: Materials Science and Engineering 852 (2020) 012043 doi:10.1088/1757-899X/852/1/012043

Image clustering using genetic algorithm with tournament

selection and uniform crossover

Gevin Valerian*, Tri Sutrisno and Dyah Erny Herwindiati

Informatics Department, Faculty of Information Technology
Universitas Tarumanagara, Jakarta 11440, Indonesia

*[email protected]

Abstract. This study aims to test genetic algorithm with tournament selection and
uniform crossover for clustering complicated images. Genetic algorithm can be used for
searching the optimum centroid for clustering images. Images that used in this study is
beach images, city images, traditional market images, and garden images. Based on the
research result genetic algorithm with tournament selection and uniform crossover can
be used for clustering images but there is some outlier in formed cluster. Based on trial
the best parameters for image clustering with genetic algorithm with tournament
selection and uniform crossover is population=200, iteration=200, and number of
cluster=2. Fitness value on genetic algorithm is increase when population and iteration
value are higher. The result of this study can be used as a reference in the development
of images clustering.

1. Introduction
This paper aims to develop application to clustering images using genetic algorithm with tournament
selection and uniform crossover. clustering analysis is a method, where a few groups are so
automatically established based on the metrics of correlation between samples that the samples within
the same group are similar to each other while the ones from different groups are different from each
other [1]. This paper is made because genetic algorithm is a simple algorithm but powerfull enough to
used as a solution on complex problem like seperating images. Genetic algorithm is used to identify the
best center of the cluster that formed [2].
The most common clustering algorithm is k-means algorithm which has the advantages of simple
and easy to use but there are some short comings with k-means algorithm which is it might fall into local
optimum [3]. Aiming at these shortcomings, this paper proposed a new clustering method based on
genetic algorithm. The focus of this research is to test the genetic algorithm with tournament selecion
and uniform crossover for clustering complex images. Genetic algorithm can solve the local optimum
problem due to improper selection of the intial center , with multivariable global optimization[4].
The images that used in this research is images of beach, images of traditional market, images of city
and images of garden. These images choosen because each images have a different characteristics, e.g.,
beach images have ocean surface and sand surface on it, city images have road and skycrappers on it,
traditional market images have lot of vegetables and people on it and garden images have plant and
footpath on it. The images will clustered based on the colors similiarity between each images and based
on similiarity based on texture of each images. In order to get the texture from images researcher use

Content from this work may be used under the terms of the Creative Commons Attribution 3.0 licence. Any further distribution
of this work must maintain attribution to the author(s) and the title of the work, journal citation and DOI.
Published under licence by IOP Publishing Ltd 1
2nd TICATE 2019 IOP Publishing
IOP Conf. Series: Materials Science and Engineering 852 (2020) 012043 doi:10.1088/1757-899X/852/1/012043

Gray Level Coocurrence Matrix (GLCM) method, and in order to get color value from images researcher
use RGB color in each pixel of images. After the images input into the application , the genetic algorithm
with tournament selection and uniform crossover is start it process to find the best cluster center for each
category of images, when the best cluster center founded the images clustered based on the distance
between the average images color and texture with the center that founded.
The output of this application is folder of formed cluster, scatter plot distribution of cluster, and
graphic plot of increase of fitness value. Based on test conducted on the application of images clusering
using genetic algorithm with tournament selection and uniform crossover the application can clustered
images using genetic algorithm with few outlier in cluster that formed.

2. Genetic Algorithm
Genetic algorithm is a self-organizing and adaptive AI technology. In the algorithm, the initial solution
is called initial population, and usually the population is generated randomly according to certain
constraints. The population can be further divided into independent individuals, also known as
chromosomes, which is one solution to the problem, and all individuals constitute an understanding
space. Individuals are usually represented by code strings consisting of binary character sets {0, 1},
which are used to describe various parameters in practical problems. In the iterative process of
algorithm, the evolution of each generation is called heredity. The genetic process consists of three
operation operators, namely, selection operator, crossover operator and mutation operator. In addition,
the individual's good and bad in the genetic process is evaluated by the fitness, and the fitness is
calculated by the adaptive function to everyone in the population. The greater the fitness of the
individual, the more the solution it represents is closer to the optimal solution, the greater the probability
of being selected to the next generation. The probability of elimination is greater, which is consistent
with the idea of "survival of the fittest" in evolution theory [4]. Genetic algorithm is principally
composed of the following five steps initialize population, fitness value evaluation, selection, crossover,
and mutation [5].

2.1. Initialize population

The first step in genetic algorithm generated an initial population. Each member of this population
served as a possible solution to a problem. Each individual is evaluated and assigned a fitness value
according to the fitness function. Population is generated using random initialization. Random initial
population seeding technique is the simplest and the most common technique that generates the initial
population to genetic algorithm. This technique is preferred when the prior information about the
problem or the optimal solution is trivial [6].

2.2. Fitness evaluation

The critical thing in genetic algorithm is the choices of fitness function where it can influence the results
of the population. For this paper the fitness function is based on distance between data and centroid .
The distance is measure using euclidean distance. Fitness function[7].
1
𝐹 = 𝑑+1 (1)
Information:
F= Fitnesss value
d= distance

The Euclidean Distance method is used to calculate the distance between data and centroid. This
measurement is based on the value of the object in each dimension in learning. Euclidean Distance can
calculate the distance between data as much as two dimensions and more. Euclidean Distance formula
[8]:

𝑑 = √∑𝑛𝑖=1(𝑋𝑖 − 𝑌𝑖)2 (2)

2
2nd TICATE 2019 IOP Publishing
IOP Conf. Series: Materials Science and Engineering 852 (2020) 012043 doi:10.1088/1757-899X/852/1/012043

Information :
d= distance
X=the dimension of the first object
Y=the dimension of the second object
i= iteration
n=amount of data

2.3. Selection
Selection is a random process to select a parent’s chromosome from a population based on fitness value
everychromosome [9]. Selection process by a tournament is more efficient than other selection
methods and leads to optimal solutions [9]. Before the selection process by tournament begins, it must
define the number of participants . Than, tournament selection randomly selects candidate from
the population. The winner of the tournament is the chromosome with the highest fitness
value. This selection process is repeated until 2 chromosomes selected as a pair of the parent.

2.4. Crossover
Crossover is parents chromosomes combining process to get new individual. Used method is
Uniform Crossover. This method defines biner zero and one. when the mask of locus has value 1,
the first spring has gene compositions as same as the first parent, the second spring has gene
compositions as same as the second parent. If the mask of locus has value 0, the first spring has
the second parent gene compositions and the second spring has the first gene compositions[9].

2.5. Mutation
Mutation is a genetic operator used to maintain genetic diversity from one generation of a population of
genetic algorithm chromosomes to the next gen. The mutation method that used in this paper is random
mutation. In random mutation the gen that mutated defines by mutation rate and the gen that mutated is
choice randomly[10].

3. Object classification
Object that used in this paper is 100 images of beach, 100 images of garden, 100 images of traditional
market, 100 images of city. These images is selected because each images have a different
characteristics, e.g., beach images have ocean surface and sand surface on it, city images have road and
skycrappers on it, traditional market images have lot of vegetables and people on it and garden images
have plant and footpath on it. Example of these images can be seen in figure 1 ,figure 2, figure 3 and
figure 4.

Figure 1. City image Figure 2. Beach image Figure 3. Garden image Figure 4. Traditional
Market images

The images color value taken using RGB of each pixel in the image.The texture value taken using
GLCM feature. The images value later used to become the population in genetic algorithm

3
2nd TICATE 2019 IOP Publishing
IOP Conf. Series: Materials Science and Engineering 852 (2020) 012043 doi:10.1088/1757-899X/852/1/012043

3.1. RGB images

In RGB color model, each colour appears in its primary spectral componets of red,green,and blue. The
colour of a pixel is made up of three components; red, grenn, and blue(RGB), described by there
corresponding intensities. Colour componenta are also known as colour channels or colour planes
(components). In the RGB colour model, a colour image can be represented by the intensity function.
I RGB = (FR , FG , FB) (3)
Where FR(x,y) is the intensity of the pixel (x,y) in the red channel, fG(x,y) is the intensitty of pixel (x,y)
in the greenchannel, and fB(x,y) is the intensity of pixel (x,y) in the blue channel [11].

3.2. Gray-Level Co-Occurrence Matrix (GLCM)

The texture analysis approach has been widely developed in recent decades, and can be classified into
four categories such as statistical, geometrical, model-based and signal processing. Gray Level Co-
occurrence Matrix (GLCM) has been shown to be one of the most efficient approaches to texture
analysis among other statistical methods[12]. Texture analysis using Gray Level Co-occurence Matrix
(GLCM) was introduced by Haralick[13]. GLCM is a matrix sized NxN where N is the grayscale
intensity level. Each of matrix element value located on (i,j) is calculated from the number pattern from
original grayscale image that contain a specified pattern. The specified pattern contains the gray pixel
ineighboring with gray paxel j for distance d and angle θ. The parameter d represents the distance of two
neigboring pixel in the grayscale image and θ is the size of the discrete angle (0º , 45º , 90º, 135º ). In
this paper the feature of GLCM that used is contrast ,correlation,energy, and homogenenity.
Contrast is a measure of intensity or gray level variations between the reference pixel and its
neighbor. Large contrast reflects large intensity differences in GLCM:
𝑐𝑜𝑛𝑡𝑟𝑎𝑠𝑡 = ∑𝑖 ∑𝑗(𝑖 − 𝑗)2 𝑃𝑑(𝑖, 𝑗) (4)

Homogeneity measures how close the distribution of elements in the GLCM is to the diagonal of
GLCM. As homogeneity increases, the contrast, typically, decreases:
1
ℎ𝑜𝑚𝑜𝑔𝑒𝑛𝑒𝑖𝑡𝑦 = ∑𝑖 ∑𝑗 2 𝑃𝑑(𝑖, 𝑗) (5)
1+(𝑖−𝑗)

Correlation feature shows the linear dependency of gray level values in the cooccurrence matrix:
(𝑖−𝜇𝑥)(𝑗−𝜇𝑦)
𝑐𝑜𝑟𝑟𝑒𝑙𝑎𝑡𝑖𝑜𝑛 = ∑𝑖 ∑𝑗 𝜎𝑥𝜎𝑦
𝑃𝑑(𝑖, 𝑗) (6)

Energy is derived from the Angular Second Moment (ASM). The ASM measures the local uniformity
of the gray levels. When pixels are very similar, the ASM value will be large :
𝐴𝑆𝑀 = ∑𝑖 ∑𝑗(𝑃𝑑(𝑖, 𝑗))2 (7)
𝑒𝑛𝑒𝑟𝑔𝑦 = √𝐴𝑆𝑀 (8)

4. Classification& Evaluation
The Classification prosses begin when the application take the input of images that want to be classified,
then the application will get the color and texture value of each images, the values of each images
become the population that start in genetic algorithm, then measure the distance each value using
euclidean distance that later used for fitness value, then the tournament selection begin to get the best
parents and the crossover between the best parrent to get the better offspring, then repeat until get the
best center of cluster. The flowchart of the classification prosses can be seen in Figure 5.

4
2nd TICATE 2019 IOP Publishing
IOP Conf. Series: Materials Science and Engineering 852 (2020) 012043 doi:10.1088/1757-899X/852/1/012043

Figure 5. Flowchart process clustering

Test conducted on the application of images clustering using genetic algorithm with tournament
seleciton and uniform crossover consist of testing the module and testing the data. Testing of the module
aims to test wheter each module in application is running well. Testing of data is a test to find out the
system runs according to the concept and to figure out the best parameters. Data testing consist of 286
test that combine between parameters that used in genetic algorithm such as combination of beach
images and city images, combine with iteration value 200 , population value 200 and cluster =2. This
method is used to find out the best parameters to make the best cluster as possible. Based on the test the
best cluster formed is between beach images and garden images. The best parameters for image
clustering using genetic algorithm with tournament selection and uniform crossover is cluster=2,
population= 200 , and iteration=200. The higher the iteration and population the fitness value of centroid
that formed is higher too. The best formed cluster in this test is between beach and garden images.The
plot can be seen in figure 6.

Figure 6. Plot 3D formed cluster and centroid between beach and garden

5
2nd TICATE 2019 IOP Publishing
IOP Conf. Series: Materials Science and Engineering 852 (2020) 012043 doi:10.1088/1757-899X/852/1/012043

Figure 7. Cluster 1 beach-garden Figure 8. Cluster 2 beach -garden

We are able to see that the images between beach and garden can be clustered in figure 7 & figure 8.
The formed cluster have a high precision of 98%, altought there are some outlier that can be found in
the formed cluster. In the formed cluster member of cluster-1 is 98 images of beach and 2 images of
garden, member of cluster-2 is 98 images of garden and 2 images of beach. The images that become
oulier placed in wrong cluster because the characteristic of the images more similar with the center of
the cluster that images placed than the center of the cluster that the images should. With genetic
algorithm the centroid that selected can be more optimum in each iteration because in genetic algorithm
the best centroid will selected and combined in each process of iteration .Based on this test we can say
that genetic algorithm can be use as alternative way to cluster a complex images altought there is some
outlier in the formed cluster .

5. References
[1] John McCall, 2005, Journal of Comutational and Applied Mathematics. 205-22
[2] Lewis R, Krawiec M, Confer E, et al. 2014 Data clustering: 50 years beyond K-means, Pattern
Recognition Letters, Vol. 31, No. 8, pp.651-666
[3] Rahman MA, Islam MZ 2014 A hybrid clustering technique combining a novel genetic algorithm
with KMeans, Knowledge-Based Systems, Vol. 71, No. 71, pp. 345-365
[4] EH Ruspini, 1970, Numerical methods for fuzzy clustering. Information Sciences, 2 (3):319-350.
[5] Konar, Amit. 2005. Computational Intelligence Principles, Techniques, and Applications.
Springer: Calcutta, India
[6] Gong J, Haibin M, Yong D, et al. 2005 Redundant elimination of intrusion events with multi
feature association, Journal of Southeast University (Natural Science), Vol.35, No. 3, pp. 366-
371
[7] Lichman, M. 2013. UCI Machine Learning Repository
[8] Rabunal, J.R. & Dorado, J. 2006. Artificial Neural Networks in Real-Life Applications, Ideal
Group Publishing: Hershey, United States of America
[9] Bai, L., J. Liang, & C. Dang. 2011. An Inititalization Method to Simultaneously Find Initial
Cluster Centers and The Number of Clusters for Clustering Categorical Data.
KnowledgeBased Systems24: pp. 785–795
[10] J P Dias and H S Ferreira Automating the Extraction of Static Content and Dynamic Behaviour
from eCommerce Websites 297–304 ANT 2017 Procedia Computer Science 109C
[11] Miyazaki D, Kagesawa M and Ikeuchi K 2004 Transparent surface modeling from a pair of
polarization images IEEE Trans. Pattern Anal. Mach. Intell. 26 73–82.
[12] Haralick, R. M., Shanmugam, K., & Dinstein, I. H. 1973. Textural features for image
classification. Systems, Man and Cybernetics, IEEE Transactions on, (6), 610-621
[13] Hu, Y., Zhao, C. X., & Wang, H. N. (2008, December). Directional analysis of texture images
using gray level co-occurrence matrix. In Computational Intelligence and Industrial
Application, 2008. PACIIA'08. Pacific-Asia Workshop on (Vol. 2, pp. 277-281). IEEE..