Volume 8, Issue 7, July – 2023 International Journal of Innovative Science and Research Technology

ISSN No:-2456-2165

Fuzzy C-Means Clustering Using Principal

Component Analysis for Image Segmentation
Anju Bala1, Aman Kumar Sharma2
1
Pursuing a Doctoral degree program in Computer Science, Himachal Pradesh University, Shimla (HP)
2
Chairman of Computer Science Department, Himachal Pradesh University, Shimla (HP)

Abstract:- Nowadays, Image segmentation is the area in The FCM uses the membership function and an
which most of the research is carried out. It is considered objective function to perform cluster classification. This
as one of the most crucial fields in image analysis. It is classification is based on some features, and an image has
used to divide an image into meaningful regions and thus lots of features. So, which feature should be considered for
extract the region of interests. These regions are classification is an issue? This issue is resolved by replacing
considered as objects. Fuzzy c-means (FCM) clustering is all the features with a standard set of features. That will
one of the best clustering method used for image reproduce the inter-correlations of the features [6]. Then
segmentation, but have a drawback of unknown cluster according to the features selected the number of clusters is
number. This paper focuses on this drawback of FCM calculated. This paper uses PCA to identify the number of
and to overcome it, the Principal component analysis clusters for FCM.
(PCA) is used. PCA is used for detection of cluster
numbers for FCM because of its dimension reduction The remaining paper organised as follows: Section 2
capability. The cluster number is the important factor on explains FCM and PCA, section 3 includes the proposed
which the clustering result depends. Experimental results algorithm, section 4 gives the experimental results, and
show that the proposed method efficiently calculate the section 5 concludes the study.
cluster number for different test images and gives
effective results. II. FUZZY C-MEANS

Keywords:- Clustering, Cluster Number, Fuzzy C-Means, Clustering involves partition of image pixels into
Image Compression, Image Segmentation, Principal different clusters such as the pixels in the same clusters are
Component Analysis. similar features, whereas pixels of different clusters are as
dissimilar as possible. The clusters are identified through
I. INTRODUCTION similarity measures. FCM clustering allows a single pixel to
belong to two or more clusters. It is achieved by assigning
Image segmentation segments the image into membership to each pixel corresponding to each cluster
meaningful objects or region of interests. It makes the feature based on the distance between the cluster center and the
extraction easy and lowers the computational complexity of pixel. Minimum the distance between pixel and the center
image by compressing it while maintaining the quality of the more the membership of pixel towards that specific cluster
image. Thus, it simplifies the image into something more center. The sum of membership of each pixel should be one
useful and meaningful [1]. [7]. The membership of each pixel is updated in each
iteration until there is no further updation. The membership
Clustering term was first given by Tryon in 1939 [2]. It is of each pixel is calculated by the membership function,
a process of classifying an image into regions based on which is as follows:
homogeneity or heterogeneity of some feature criterion. The
2
classified regions of clustering process are called clusters. ( −1)
𝜇𝑖𝑗 = 1⁄∑𝑐𝑘=1(𝑑𝑖𝑗⁄𝑑𝑖𝑘) 𝑚 (1)
These clusters are regions which have minimum intra-cluster
distance and maximum inter-cluster distance [3]. Clustering Where uij is membership of ith pixel to the jth cluster
algorithms are categorized into hard and soft clustering. Hard
center, dij represents Euclidean distance between the ith pixel
clustering refers to K-means clustering, where each pixel
and jth cluster center, m is the fuzziness index, and c
belongs specifically to a single cluster. It has a drawback that represents cluster number. Along with the membership
it cannot converge optimally in the general case. Soft function, the cluster center is also updated in each iteration.
clustering refers to Fuzzy c-means clustering, which was
proposed by Dunn in 1972 [4]. It is the extended version of (∑𝑛 𝑚
𝑖=1(𝑢𝑖𝑗) 𝑥𝑖)
k-means in which there is a membership function which 𝑣𝑗 = , ∀𝑗 = 1,2, … 𝑐 (2)
(∑𝑛 𝑚
𝑖=1(𝑢𝑖𝑗) )
states the membership of each pixel to the different clusters.
That means in it a pixel can belong to two or more clusters. Where vj represents the jth cluster center, n is the
Although the computation cost of FCM is higher than the k- number of pixels and xi is the ith pixel.
means, however, it gives optimal results [5].

IJISRT23JUL571 www.ijisrt.com 1139

Volume 8, Issue 7, July – 2023 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165
Fuzzy clustering is carried out through an iterative
updation of membership function and cluster center and
optimization of the objective function.

𝐽(𝑈, 𝑉) = ∑𝑛𝑖=1 ∑𝑐𝑗=1(𝑢𝑖𝑗)𝑚 ‖𝑥𝑖 − 𝑣𝑗‖2 (3)

'||xi – vj||' is the Euclidean distance between ith data and

jth cluster center.

'β' is the termination criterion between [0, 1].

 Algorithm:

 Initialize U=[uij ] matrix.

 At k- step: for given cluster numbers c, calculate cluster
centers using equation 2.
 Update U using equation 1.
 If ||U(k+1) - U(k)|| < β, then stop; otherwise go to step 2.
 The lower the value of B, the good the quality of result
but at the cost of computational complexity. It gives better
results than k-means but has some disadvantages such as
cluster number should be known prior [8].
 This paper focuses on the cluster number problem and to
solve this; the principal component analysis is used.

III. PRINCIPAL COMPONENT ANALYSIS

The principal component analysis is a process of reducing

the broad set of variables or features to small set that still
contains most of the information. It transforms a large number of
correlated variables into a smaller number of uncorrelated
variables. These smaller number of variables are called principal
components. The first principal component consists of most of
the variability in data and succeeding components consists of
remaining. This paper uses the covariance matrix instead of the Fig1 Flowchart of the Proposed Algorithm
correlation matrix as the variances of a pixel are not much [9].
 The covariance matrix of regions is calculated to obtain the
IV. PROPOSED METHOD relationship between the pixels. A covariance matrix is a
matrix whose x and y elements is in covariance position
There are many extended versions of FCM available in the between the xth and yth elements of a random vector. A grey
literature [10-12] but, the basic FCM is one of the most useful level covariance matrix is generated.
general-purpose clustering schemes [13]. The basic FCM  Then the principal component analysis is applied on the grey
requires determination of several parameters such as cluster level covariance matrix. The latent variables are extracted
number, fuzzy index, stopping criterion, cluster centers etc. the from the results. The latent variables are also called hidden
initial cluster centers are randomly chosen. In contrast, the fuzzy variables. The latent variables are the true variables which do
index value is hypothetically considered good between 1.1 and not considers all the features of an image but considers a few
5; there is no theoretical basis for it. The main problem is to of them from which all others can be derived [15]. These few
decide the number of cluster in FCM [14]. The proposed method latent values can be used instead of other features of an
tries to solve this problem by using the reduction of n image.
dimensions variables to k dimensions variables of PCA. The  Half of the mean of the latent variable values are calculated.
proposed algorithm has the following steps: This value is used to decide the number of clusters present in
an image.
 The first step in image segmentation using the FCM is to  The number of latent variables having a value higher than
read the input image and convert it into a grey image. The the value calculated in the above step is counted. This
FCM is used on the grey images in the proposed method. number is taken as cluster number.
 Now, FCM is applied to the pre-processed image having
cluster number calculated from the above steps.
 These are the steps of the proposed algorithm. The proposed
algorithm is applied to the number of test images of
Berkeley's database.

IJISRT23JUL571 www.ijisrt.com 1140

Volume 8, Issue 7, July – 2023 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165
V. EXPERIMENTAL RESULTS covariance matrix is generated. After that, PCA is applied to get
the number of clusters, which is used as an input parameter to
To assess the performance and efficiency of the proposed FCM, and the results are obtained. The results show the
algorithm, several test images are used from Berkeley's efficiency of the proposed method.
database. The test images are converted to grey images, and a

Table 1 Shows the (a) Original Image, (b) Segmented Image and (c) Cluster Number Calculated from the Proposed Method
Test images

Segmented
images

Cluster
numbers 4 4 3 2

The table 1 shows that the test images and their segmented images through Fuzzy c-means whose cluster numbers are calculated from
the proposed algorithm. The figure also shows the calculated cluster number.
To quantitatively analyse the results of the proposed method, the size of a compressed or segmented image and the original image is
given in Table 2

Table 2 Result Analysis Shows the Compressed Size of Images after Segmentation.
Result analysis Image 1 Image 2 Image 3 Image 4
Original size (kb) 52.9 73.2 68.8 68.3
Calculated 22.5 44.7 36.3 33.0
Cluster-1
Segmented Calculated 20.5 25.6 23.2 23.0
image size (kb) Cluster
Calculated 50.8 52.2 55.2 46.3
Cluster+1
Calculated Cluster-1 57.5 38.9 47.2 51.7
Calculated 61.2 65.0 66.3 66.4
Compression Cluster
percentage Calculated 3.96 28.6 31.4 32.2
Cluster+1

The table 2 shows that the proposed method gives good VI. CONCLUSIONS
results in terms of segmented image as well as a compressed
image. The comparative results are shown in the table which The cluster number of an image is calculated using PCA
includes the size of segmented image using calculated cluster applied to the covariance matrix. PCA is a dimension reducing
number (represented in bold characters), one less than the algorithm which done the cluster number calculation task very
calculated cluster number and one greater than the calculated efficiently. The image obtained after segmentation using the
cluster number. The results show that the size of the segmented calculated cluster number is more compressed and small in size
image using cluster number obtained from proposed algorithm is than the other segmented images. Hence, the results show that
less than the other cluster numbers. It is also found that the the cluster number obtained from the proposed method gives
compression percentage of all the segmented test images using more accurate results than the other cluster numbers. The cluster
the proposed method is above 50 percent, which is considered to number computation is a significant problem in the automation
be useful to reduce the size of the image. Hence, it is found that of the FCM algorithm, which is fulfilled by the proposed
the proposed method gives good segmented image and has less algorithm. There are other problems in FCM which can be
size than the original image. considered for further research.

From the results, it is evident that the proposed method is

capable of finding good cluster number values for various test
images.

IJISRT23JUL571 www.ijisrt.com 1141

Volume 8, Issue 7, July – 2023 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165
REFERENCES

[1]. R. M. Haralick, "Image segmentation survey,"

Fundamentals in computer vision, OD Faugeras, ed.,
Cambridge Univ. Press, Cambridge, pp. 209-224, 1983.
[2]. R. C. Tryon, Cluster analysis: Correlation profile and
orthometric (factor) analysis for the isolation of unities
in mind and personality. Edwards brother, Incorporated,
litho printers and publishers, 1939.
[3]. J. C. Bezdek, "Cluster validity with fuzzy sets," 1973.
[4]. J. C. Dunn, "A fuzzy relative of the ISODATA process
and its use in detecting compact well-separated
clusters," J.Cybern., vol. 3, pp. 32-57, 1973.
[5]. J. C. Bezdek, Pattern recognition with fuzzy objective
function algorithms. New York, Plenum, 1981.
[6]. R. C. Tryon and D. E. Bailey, "The BC TRY computer
system of cluster and factor analysis," Multivariate
Behavioral Research, vol. 1, no. 1, pp. 95-111, 1966.
[7]. W. Cai, S. Chen, and D. Zhang, "Fast and robust fuzzy
c-means clustering algorithms incorporating local
information for image segmentation," Pattern
recognition, vol. 40, no. 3, pp. 825-838, 2007.
[8]. J. A. Hartigan, Clustering algorithms. 1975.
[9]. I. Jolliffe, "Principal component analysis," in
International encyclopedia of statistical science:
Springer, 2011, pp. 1094-1096.
[10]. M.-C. Hung and D.-L. Yang, "An efficient fuzzy c-
means clustering algorithm," in Data Mining, 2001.
ICDM 2001, Proceedings IEEE International
Conference on, 2001, pp. 225-232: IEEE.
[11]. N. Forghani, M. Forouzanfar, and E. Forouzanfar, "MRI
fuzzy segmentation of brain tissue using IFCM
algorithm with particle swarm optimization," in
Computer and information sciences, 2007. iscis 2007.
22nd international symposium on, 2007, pp. 1-4: IEEE.
[12]. C. I. Gonzalez, P. Melin, J. R. Castro, O. Castillo, and
O. Mendoza, "Optimization of interval type-2 fuzzy
systems for image edge detection," Applied Soft
Computing, vol. 47, pp. 631-643, 2016.
[13]. J. C. Bezdek, R. Ehrlich, and W. Full, "FCM: The fuzzy
c-means clustering algorithm," Computers &
Geosciences, vol. 10, no. 2-3, pp. 191-203, January
1984.
[14]. R. L. Cannon, J. V. Dave, and J. C. Bezdek, "Efficient
implementation of the fuzzy c-means clustering
algorithms," IEEE transactions on pattern analysis and
machine intelligence, no. 2, pp. 248-255, March 1986.
[15]. S. Wold, K. Esbensen, and P. Geladi, "Principal
component analysis," Chemometrics and intelligent
laboratory systems, vol. 2, no. 1-3, pp. 37-52, 1987.

IJISRT23JUL571 www.ijisrt.com 1142