Using k means cluster and fuzzy c means for defect segmentation in fruits

Proceedings of the 2nd International Conference on Current Trends in Engineering and Management ICCTEM -2014
INTERNATIONAL JOURNAL OF COMPUTER ENGINEERING &
17 – 19, July 2014, Mysore, Karnataka, India
TECHNOLOGY (IJCET)
ISSN 0976 – 6367(Print)
ISSN 0976 – 6375(Online)
Volume 5, Issue 9, September (2014), pp. 11-19
© IAEME: www.iaeme.com/IJCET.asp
Journal Impact Factor (2014): 8.5328 (Calculated by GISI)
www.jifactor.com
11

IJCET

© I A E M E

USING K-MEANS CLUSTER AND FUZZY C MEANS FOR DEFECT
SEGMENTATION IN FRUITS
VaniAshok1, Dr. D.S. Vinod2

1(Computer Science Engineering, Sri Jayachamarajendra College of Engineering, Mysore,
Karnataka, India)
2(Information Science Engineering, Sri Jayachamarajendra College of Engineering., Mysore,
Karnataka, India)

ABSTRACT
Quality and safety are the key factors in modern food industries. The quality of fruits and
vegetables is a common combination of characteristics, attributes and properties that have
significance and make for acceptability. One of the most popular applications of computer vision is
to inspect qualities of food products based on form, color and presence of defects. Appearance
factors such as size or dimension, shape, surface texture, surface color, and external or surface
defects define external quality and directly influence consumers in purchasing a product, and they
can be evaluated by means of computer vision techniques. The proposed paper presents defect
segmentation of fruits based on surface color features with unsupervised K-Means clustering and
Fuzzy C-Means algorithms. As the first step, the digital color images of defective apples are pre-processed
using Gaussian low-pass filter (GLPF) smoothing operator to remove noise. The images
are then segmented with the purpose of separating the defects from the edible regions using proposed
clustering algorithms. A comparison analysis is also performed among the two methods.
Keywords: Image segmentation, Clustering, Histogram, Filtering, K-means, Fuzzy C-Means (FCM)
1. INTRODUCTION
Quality and safety are the key factors in modern food industries. Increasing consciousness of
quality, particularly in the food and health sector, strongly demands research activities regarding the
production of defined quality, the preservation of quality during marketing, and thus also the
possibilities of evaluating quality parameters and of integrating this into production processes. The
pallet of possible damage to fruit and vegetables is extremely extensive and is often a criterion of

quality determination methods, for example, malformations, rust fungi, formation of cork, splits,
bitter pit, insect damage, rots, scalds, temperature damage, and glassiness or other physiological
diseases of the storage phase. In addition, there are various risks for injuries during harvest and
transport [1]. With the increasing quality awareness among consumers, the expectation for improved
quality in agricultural and food products has increased the need for enhanced quality monitoring.
There is a need to develop various image analysis techniques to meet the demand of the growing
population. Moreover there is growing demand for automation in food industries due to the fact that
the traditional labor intensive manual inspection processes are inefficient, inaccurate and ineffective.
The application of computer vision in agriculture has increased considerably in recent years due to
the fact that it provides substantial amounts of information about the nature and attributes of the
objects present in a scene.
12

The use of this technology has spread rapidly in inspecting agri-food commodities, including
meat quality assessment, automated poultry carcass inspection, quality evaluation of fish,
visualization of sugar distribution of melons, measuring ripening of tomatoes, defect detection of
pickling cucumber, and classification of wheat kernels [2]. The computer vision is particularly used
more in automatic inspection of fruits and vegetables, since it is more reliable and objective than
human inspection [3].
Quality assessment of fruits and vegetables is done based on the analysis of external features
like color, size, shape, texture and presence of damage. As consumers are mostly influenced to
choose or reject a particular fruit by its color, it is the most important attribute for assessing the
quality of fruits. The most widely used color spaces in computers and digital images are RGB, HSI
and L*a*b*. In RGB the color of a pixel in image is expressed as three coordinates of primary colors
red, green and blue in a color space. HIS is the color space which is closer to the human perception
of color, like the hue, saturation and intensity. However, RGB and HSI are non-uniform color spaces
and hence uniform color space like L*a*b* is used to implement the proposed algorithms.
The defect segmentation of fruits based on surface color feature can be considered as an
instance of image segmentation where we are segmenting only the defective portion of the fruit.
Image segmentation is the process of partitioning the image into several constituent components. It
partitions the digital image into disjoint (non-overlapping) regions. Segmentation is an essential step
in computer vision and automatic pattern recognition processes based on image analysis of foods as
subsequent extracted data are highly dependent on the accuracy of this operation. Food image
segmentation is still an unsolved problem because of its complex and under constrained attributes
[4].
Image segmentation methods are generally based on one of two fundamental properties of the
intensity values of image pixels: similarity, where the image is partitioned into regions that are
similar according to a set of predefined criteria, and discontinuity, where the image is partitioned
based on sharp changes in intensity values. Based on the discontinuity or similarity criteria, many
segmentation methods have been introduced which can be broadly classified into six categories: (1)
Histogram based method, (2) Edge Detection, (3) Neural Network based segmentation methods, (4)
Physical Model based approach, (5) Region based methods (Region splitting, Region growing and
merging), (6) Clustering (Fuzzy C-means clustering and K-Means clustering) [5]. The unsupervised
clustering algorithms are particularly appropriate for the exploration of interrelationship among the
data points to make an assessment of the structure where there is little a priori information available
about the data [6].
2. RELATED WORKS
Physical appearances of food extremely vary causing difficulties for computer vision
systems. Fruits, in particular, have numerous kinds of defects and highly varying skin color. Hence,

they pose even more problems for computer vision-based quality inspection systems. Paul Martinsen
and Peter Schaare [7] have proposed a method for predicting chemical component distribution in
kiwifruit using imaging spectroscopy techniques. The author used partial least- squares method for
data modeling. J Blasco et al. [8] proposed the segmentation procedure, based on a Bayesian
discriminant analysis, to allow fruits to be precisely distinguished from the background. The machine
vision system thus developed detected the external defects of the apples with 86% accuracy. The
feasibility of NIR spectroscopy in combination with powerful multivariate calibration techniques,
such as Partial Least Squares regression (PLS), to measure quality attributes of fruit and vegetables
has been demonstrated by B.M. Nicolai et al. [9]. J. Tan [10] suggested filtering, background
removal, segmentation of fat from muscles, isolation of the LD muscle and segmentation of marbling
from the LD muscle to assess the quality attributes of meat images. K.Vijayarekha [11] discusses the
multivariate image analysis technique applied to the defect segmentation of apple fruit in
multispectral range.
13

An intensive study on apple quality inspection is carried out by Unay [12]. The apple images
were captured through color/monochrome camera in diffusely illuminated tunnel with two different
light sources (fluorescent tubes and incandescent spots). To improve the image quality a noise
removal operation was performed before applying the image segmentation operation to detect the
defect type. The image intensity and texture based shape features were extracted from each
segmented portion of the image. The performance of several classification methods (Linear
Discriminant Classifier (LDC), k-Nearest Neighbors (k-NN), Fuzzy k-NN, Support Vector Machine
(SVM), Decision Tree and Multi-layer Perceptrons (MLP) were studied for defect segmentation and
detection. They identified bruise, flesh damage, frost damage, hail, hail with perforation, limb rub,
scar tissue, rots, russet and scald defects.
Gabriel Leiva et al. [13] devised methodology to classify blueberries with fungal decay,
shrivelling and mechanical damage using statistical pattern recognition techniques: extracting the
most possible features, selecting the best ones, training the best classification algorithm. The feature
extraction strategies used were Sequential Forward Selection (SFS), with objective functions: Fisher
discriminant, K-Nearest Neighbour (KNN), Linear and Quadratic Discriminant Analysis (LDA and
QDA); and other feature extractors without objective function used were: Forward orthogonal search
algorithm by maximizing the overall dependency, Least Squares Ellipse Fitting (LSEF) and Rank
key features by class separability criteria. With the selected features, decision lines, planes or hyper
planes classifiers were implemented using LDA, QDA, minimal distance, Mahalanobis Distance
(MD), KNN (with 4 to 30 nearest neighbours), Support Vector Machine (SVM) and different Neural
Networks techniques (NN).
Apart from fruit quality assessment researchers are trying to investigate fruit characteristics
that can be used for fruit grading process. V. Leemans et al. [14] have achieved fruit grading is six
steps: image acquisition; ground colour classification; defect segmentation; calyx and stem
recognition; defects characterization and finally the fruit classification into quality classes. They
realized color grading using a simple neural network with no hidden layer, using the three
luminances (red, green and blue) of the considered pixel as input. The calyx and stem ends, which
appear on an image as defects, were detected using a correlation pattern recognition technique and
defect segmentation was done using Gaussian model of the fruit color, measuring the Mahalanobis
distance separating the mean color of the fruit and of each pixel.
Yousef Al Ohali [15] has designed computer mediated date fruit quality assessment and
sorting system. He has used the color intensity distribution in the image as an estimate of flabbiness
of date fruit. Back Propagation Neural Network (BPNN) is used to classify the dates into three
groups, Grade-1: fruits having good shape, high flabbiness, Grade 2: fruits with distorted shape, low
flabbiness and Grade 3: fruits having defects.

i x −μ (2)
μ : (3)
= = (5)
14
3. CLUSTERING ALGORITHMS

In this section, the basic sets of definitions are presented to provide the preliminaries of the
clustering methods. First we define the K-means clustering method. The second part discusses about
Fuzzy C- means (FCM) clustering.
3.1 K-Means Clustering Algorithms
K-means method is an unsupervised clustering method that classifies the input data objects
into multiple classes on the basis of their inherent distance from each other [5]. Clustering algorithm
assumes that a vector space is formed from the data features and tries to identify natural clustering in
them. The objects are clustered around the centroids i = 1 i μ
…..k which are computed by
minimizing the following objective
−μ
= Î
V = 2
j (x )
1 x S
j i
i
k
i
(1)
Where k is the number of clusters i.e. Si, i = 1, 2…., k and i μ is the mean point or centroid of all the
points j i x ÎS .
The algorithm of an iterative version K-means clustering is as follows:
Step 1. Compute the distribution of the intensity values.
Step 2. Using k random intensities initialize the centroids.
Step 3. Cluster the image points based on the distance of their intensity values from the centroid
intensity values.
(i ) c := arg minj
2
( )
j
Step 4. Compute new centroid for each cluster.

1{ }

=
=
c j x
=
=
= m
i
i
m
i
i
i
i
c j
1
( )
1
( )
( )
1{ }
where k is the number of clusters, i iterates over all the intensity values, j iterates over all the
centroids.
Step 5. Repeat Step 3 and Step 4 until the labels of the cluster do not change any more.
3.2 Fuzzy C-Means Clustering Algorithms
Fuzzy C-means (FCM) is a clustering technique which differs from hard K-means that
employs hard partitioning [16]. The FCM employs fuzzy partitioning such that a data point can
belong to all groups with different membership grades between 0 and 1. FCM is an iterative
algorithm. The aim of FCM is to find cluster centers (centroids) that minimize a dissimilarity
function. To accommodate the introduction of fuzzy partitioning, the membership matrix (U) is
randomly initialized according to,
= =
=
u j n
c
i
ij 1, 1,....,
1
(4)
The dissimilarity function used in FCM is given by,
2
c
c i J U c c c J u d
1 2 ( , , ,...., ) ij
1 i
1 1
n
j
m
ij
c
i
= = =
Uij is between 0 and 1;
Cij is the centroid of cluster i;

dij is the Euclidean distance between ith centroid(ci) and jth data point;
mÎ [1,] is a weighting exponent or fuzziness parameter
1 (6)
S S U U u u (8)
15

To reach a minimum of dissimilarity function there are two conditions which are given in
Equation (6) and Equation (7).

u x
=
= = n
j
m
ij
j
n m
j ij
i
u
c
1
=
−

=
c
k
m
ij
kj
ij
d
d
u
1
2 /( 1)
1
(7)
Precisely speaking, initially the uij and the centers of the clusters are assigned randomly and
the uij is updated in each iteration. The iterative process stops when
( ) ( 1) ( ) ( 1) p e max − − − = − s
ij
S
ij
The FCM that iteratively updates the cluster centers and the membership grades for each data
point is as follows:
Step 1. Randomly initialize the membership matrix (U) that has constraints as given in Equation (4)
Step 2. Calculate centroids (ci) by using Equation (6).
Step 3. Compute dissimilarity between centroids and data points using equation (5). Stop if its
improvement over previous iteration is below a threshold.
Step 4. Compute a new U using Equation (7). If condition given in Equation (8) is False, go to
Step 2.
FCM iteratively moves the cluster centers to the right location within a data set.
4. DEFECT SEGMENTATION
In this section, the implementation of K-means and Fuzzy C-means clustering algorithms for
defect segmentation in apple fruits is presented.
4.1 Using K-Means Clustering
Step 1. Read the input image of defective fruit.
Step 2. In order to remove the image noise and reduce detail levels the Gaussian low-pass filter
(GLPF) smoothing operator is applied.
Step 3. Transform the image from RGB to L*a*b* color space as all of the color information is
present in the a* and b* layers only.
Step 4. Calculate the histograms of the image to decide the number of clusters.
Step 5. Classify colors using K-means clustering in a*b* space, with Euclidean distance to measure
the distance between two colors.

Proceedings of the 2nd International Conference on
Step 6. Label each pixel in the image from the results of K
labeled with its cluster index.
Step 7. Generate different images for each cluster.
4.2 Using Fuzzy C-Means
ICCTEM -2014
Step 1. Read the input image of defective fruit.
Step 2. In order to remove the image noise and reduce detail levels the Gaussian low
(GLPF) smoothing operator is applied.
Step 3. Calculate the histograms of the image to decide the number of clusters.
Step 4. Classify pixel intensities using FCM algorithm (initial value of
number of clusters as determined in Step 3.
Step 5. Generate image by allocating different intensity levels for each subclass of the image.
5. EXPERIMENTAL RESULTS
To determine the performance of the clustering approaches, we have considered apples as a
case study. The data set consists of various images of apples with defects such as apple scab, rot and
blotch for the purpose of defect segmentation. Fig. 1
apples from the data set.
(a)
Fig. 1:
Fig. 2, Fig. 3 and Fig. 4 show defect segmentation result
means clustering.
(a) (b)
Fig. 2: K-Means
(a) Filtered Image (b) First Cluster
Current Trends in Engineering and Management
16
K-means. Every pixel of the image will be
ead m = 2 and
represents some of the images of defective
(b) (c)
Sample Images from the Dataset
results of sample apple fruits using K
(c) (d)
eans Defect Segmentation of an Apple with Scab
(c) Second Cluster (d) Third Cluster (e) Histogram

low-pass filter
e = 0.01) with
K-
(e)

Proceedings of the 2nd International Conference on
(a) (b)
Fig. 3: K-Means D
600
500
400
300
200
100
0
0 50
Defect Segmentation of an Apple with Blotch
(a) Filtered Image (b) First Cluster
(a) (b)
Fig. 4: K-Means Defect Segmentation
100 150 200 250
egmentation of a Rotten Apple (a) Filtered Image (b) First Cluster
(c) Second C
Current Trends in Engineering and Management
(c) (d)
(c) Second Cluster (d) Third Cluster (e) Histogram
(c) (d)
Cluster (d) Third cluster (e) Histogram
We have segmented the input image into three clusters depending on the data provided by the
histogram shown in Fig. 2(e), Fig. 3(e) and Fig. 4(e). Fig. 2(a), Fig. 3(a) and Fig. 4(a) show the
preprocessed image filtered by Gaussian low
different clusters second cluster correctly segments the defective portion of the image whereas the
first cluster demonstrates the non-defective defect
part of the fruit.
Fig. 5, Fig. 6 and Fig. 7 illustrate the resultant images ob
(a)
low-pass filter smoothing operator. Among the images in
Fig. 5: Fuzzy C-Means Defect Segmentation
(a)
Fig. 6: Fuzzy C-Means Defect Segmentation
(a) Filtered Image (b) FCM Processed Image
17
obtained from FCM.
(b)
of an Apple with Scab (a) Filtered Image (b) FCM
Processed Image
(b)
of an Apple with Blotch
ICCTEM -2014

(e)
(e)

18

(a) (b)
Fig. 7: Fuzzy C-Means Defect Segmentation of a Rotten Apple
(a) Filtered Image (b) FCM Processed Image
The filtered images are shown in Fig. 5(a), Fig. 6(a) and Fig. 7(a). The FCM segmented
images with three clusters are shown in Fig. 5(b), Fig. 6(b) and Fig. 7(b). The amount of defect in a
given sample using K-Means and FCM clustering is tabulated in Table 1. FCM, being unsupervised
fuzzy clustering algorithm, is motivated by the need to find interesting patterns or groupings in a
given set of data. The cluster allocation in FCM is based on the high membership value and less on
distance.
Table 1: Amount of Defect in Selected Fruit Samples
SI.
No.
Data
Sample
K-Means FCM
1 Figure 1(a) 60% 52%
2 Figure 1(b) 44% 40%
3 Figure 1(c) 10% 9%
6. CONCLUSION
The automated inspection of agricultural products, fruits in particular, is an important process
as it reduces human interaction with the inspected goods, classify generally faster than humans and
tend to be more consistent in classification. The segmentation of defects in fruits is proposed and
evaluated in this paper. The proposed approach used K-Means clustering and Fuzzy C-Means
clustering to segment defects in apple images. Experimental results suggest that the algorithms are
able to segment the defects more accurately. The major drawback of K-Means is that, there may be a
skewed clustering result if the cluster number estimate is incorrect. It is overcome to certain extent in
the proposed method by determining the number of clusters using the histogram of the image. The
image is also pre-processed to remove noise.
7. REFERENCES
1. Peter Butz, Claudia Hofmann, Bernhard Tauscher, Recent Developments in Noninvasive
Techniques for Fresh Fruit and Vegetable Internal Quality Analysis, Journal Of Food Science,
Vol. 70, Nr. 9, 2005.
2. Sun, D.-W, Hyperspectral imaging for food quality:A non-destructive tool for food quality and
safety evaluation and inspectio”, London: Academic, 2010.
3. Sergio Cubero, Nuria Aleixos, Enrique Moltó, Juan Gómez-Sanchis, Jose Blasco, Advances
in Machine Vision Applications for Automatic Inspection and Quality Evaluation of Fruits and
Vegetables, Food Bioprocess Technol 4: 2011, 487–504.
4. Domingo Mery, Franco Pedreschi, Segmentation of colour food images using a robust
algorithm, Journal of Food Engineering 66, 2005, 353–360.

19

5. Shiv Ram Dubey, Pushkar Dixit, Nishant Singh, Jay Prakash Gupta, Infected Fruit Part
Detection using K-Means Clustering Segmentation Technique, International Journal of
Artificial Intelligence and Interactive Multimedia, 2013, Vol. 2, No.2.
6. Jaskirat Kaur, Sunil Agarwal, Venu Vig, Performance analysis of clustering based image
segmentation and optimization methods, CSIT-CSCP, 2012, 245-254.
7. Paul Martinsen, Peter Schaare, Measuring soluble solids distribution in kiwifruit using near-infrared
imaging spectroscopy, Postharvest Biology and Technology 14014, 1998, 271–281.
8. J. Blasco, N. Aleixos, E. Molto, Machine Vision System for Automatic Quality Grading of
Fruit, Biosystems Engineering, 2003, 85 (4), 415–423.
9. Bart M. Nicolai, Katrien Beullens, Els Bobelyn, Ann Peirs, Wouter Saeys, Karen I. Theron,
Jeroen Lammertyna, Nondestructive measurement of fruit and vegetable quality by means of
NIR spectroscopy: A review, Postharvest Biology and Technology, 46, 2007, 99–118.
10. J. Tan, “Meat quality evaluation by computer vision”, Journal of Food Engineering 61, 2004,
27–35.
11. K. Vijayarekha, Multivariate image analysis for defect identificationof apple fruit images,
IEEE Transactions, 2008.
12. Unay, Dervim, Multispectral Image Processing and Pattern Recognition Techniques for
Quality Inspection of Apple Fruit, Ph.D. Thesis, 2006, Facult´e Polytechnique de Mons,
Belgium.
13. Gabriel Leivaa, Germán Mondragónb, Domingo Meryb, José Miguel Aguilera, The automatic
sorting using image processing improves postharvest blueberriesstorage quality, 2011, ICEF.
14. V. Leemans, H. Mageinb, M.-F. Destain, On-line Fruit Grading according to their External
Quality using Machine Vision, Biosystems Engineering, 2002, 83 (4), 397–404.
15. Yousef Al Ohali, Computer vision based date fruit grading system: Designand
implementation, Journal of King Saud University – Computer and Information Sciences, 2011,
23, 29–36.
16. Ebrahim Aghajari, Gharpure Damayanti, Incorporating FCM and Back Propagation Neural
Network for Image Segmentation, International Journal of Computer communication
Technology, 2011, Volume-2, Issue-VIII.

Using k means cluster and fuzzy c means for defect segmentation in fruits

More Related Content

What's hot (17)

Viewers also liked (16)

Similar to Using k means cluster and fuzzy c means for defect segmentation in fruits (20)

More from IAEME Publication (20)

Recently uploaded (20)

Using k means cluster and fuzzy c means for defect segmentation in fruits