563 574

ISSN: 2278 – 1323
International Journal of Advanced Research in Computer Engineering & Technology
Volume 1, Issue 4, June 2012

Color Image Segmentationusing Clustering
Technique
Patel Janak kumar Baldevbhai, R.S. Anand

 the literature, it is observed that different transforms are used
Abstract—This work presents image segmentation technique to extract desired information from remote-sensing images or
based on colour features with K-means clustering algorithm. In biomedical images (Mehmet Nadir Kurnaz et al; 2005).
this we did not used any training data. In this paper, we present Segmentation evaluation techniques can be generally divided
a simple and efficient implementation of k-means clustering
algorithm. The regions are grouped into a set of classes using
into two categories (supervised and unsupervised). The first
K-means clustering algorithm. Results are grouped into clusters category is not applicable to remote sensing because an
so avoiding feature calculation for every pixel in the image. optimum segmentation (ground truth segmentation) is
Although the colour is not frequently used for image difficult to obtain. Moreover, available segmentation
segmentation, it gives a high discriminative power of regions evaluation techniques have not been thoroughly tested for
present in the image. Here clusters are grouped & segmentation remotely sensed data. Therefore, for comparison purposes, it
is obtained in form of colors through which important objects
are segmented, extracted or recognized.
is possible to proceed with the classification process and then
indirectly assess the segmentation process through the
Index Terms—color Image segmentation, K-means, clusters, produced classification accuracies. (Ahmed Darwish, et al;
unsupervised classification. 2003).Clustering is a mathematical tool that attempts to
discover structures or certain patterns in a data set, where the
objects inside each cluster show a certain degree of
I. INTRODUCTION similarity.
he process of image segmentation is defined as: “the
T search for homogenous regions in an image and later the
classification of these regions”. It also means the partitioning
For image segment based classification, the images that
need to be classified are segmented into many
homogeneous areas with similar spectrum information
of an image into meaningful regions based on homogeneity firstly, and the image segments‟ features are extracted based
or heterogeneity criteria (Haralick et al; 1992). Image on the specific requirements of ground features classification.
segmentation techniques can be differentiated into the The colour homogeneity is based on the standard deviation of
following basic concepts: pixel oriented, Contour-oriented, the spectral colours, while the shape homogeneity is based on
region-oriented, model- oriented, colour oriented and hybrid. the compactness and smoothness of shape. There are two
Colour segmentation of image is a crucial operation in image principles in the iteration of parameters:1) In addition to
analysis and in many computer vision, image interpretation, necessary fineness, we should choose a scale value as large as
and pattern recognition system, with applications in scientific possible to distinguish different regions; 2) we should use the
and industrial field(s) such as medicine, Remote Sensing, colour criterion where possible. Because the spectral
Microscopy, content- based image and video retrieval, information is the most important in imagery data, the quality
document analysis, industrial automation and quality control of segmentation would be reduced in high weightiness of
(Ricardo Dutra, et al;2008). The performance of colour shape criterion.
segmentation may significantly affect the quality of an image This work presents a novel image segmentation based on
understanding system (H.S.Chen et al; 2006).The most colour features from the images. In this we did not used any
common features used in image segmentation include training data and the work is divided into two stages. First
texture, shape, grey level intensity, and colour. The enhancing color separation of satellite image using decor
constitution of the right data space is a common problem in relation stretching is carried out and then the regions are
connection with segmentation/classification. In order to grouped into a set of five classes using K-means clustering
construct realistic classifiers, the features that are sufficiently algorithm. Using this two-step process, it is possible to
representative of the physical process must be searched. In reduce the computational cost avoiding feature calculation
for every pixel in the image. Although the colour is not
Manuscript received June 19, 2012. frequently used for image segmentation, it gives a high
Patel Janakkumar Baldevbhai is with the Image and Signal Processing discriminative power of regions present in the image.
Lab., Electrical Engineering Department, Research Scholar, EED, Indian
Institute of Technology Roorkee, Uttarakhand, India on duty leave under Colour segmentation is an essential issue with regard to
QIP scheme of AICTE from the L.D.R.P. Institute of Technology & Research, vision applications, such as object detection and navigation
Gandhinagar, and Gujarat, India. (Corresponding author phone: (Bosch et al., 2007; Lin, 2007). The process of color
09458121095; 079-23221371(R) e-mail: janakbpatel71@gmail.com).
R.S. Anand is with the Electrical Engineering Department, Professor,
segmentation consists of color representation, color feature
EED, Indian Institute of Technology Roorkee, Uttarakhand, India extraction, similarity measurement and classification. In

563
All Rights Reserved © 2012 IJARCET

ISSN: 2278 – 1323
color representation, the RGB (Red, Green and Blue) model, used to estimate the clustering index (Al Aghbari and Al-Haj,
which expresses color as a mixture of red, green and blue 2006). The idea of a „histon‟, which is an encrustation of a
three color components, is often used to depict the color histogram such that the elements in the histon are the set of all
information of an image (Bascle et al., 2007; Weng et al., the pixels that can be classified as possibly belonging to the
2007). By using a transformation, the secondary colors, same segment, was introduced for color segmentation by
which are CMY (Cyan, Magenta and Yellow) or Murshrif and Ray (2008), and the total computation time this
RG–GB–BR, can be obtained and used as an alternative color approach requires for a 179X122 image is 2.41 s. Neural
model (Wang et al., 2007). The HSI model, which transforms networks (Bascle et al., 2007) have recently been used as a
RGB into Hue, Saturation and Intensity, is also a popular clustering kernel for color segmentation, where components
color model at present, and its good performance has been of the RGB space and the intensity are used as inputs and
shown in many works (Kim et al., 2007, 2008; Wangenheim three calibrated colour components are considered as outputs
et al., 2007). HSV (Value) and HSL (Luminance) are very of the modified multi-layer perceptron (MLP). After the
similar to the HSI model due to the transformation formulas training procedure, good segmentation performance is
applied. Using the HSI color model, a specific color is able to achieved. Furthermore, the look-up tables (LUT) of the
be recognized regardless of variations in saturation and modified MLP can be applied for real-time applications, so
intensity. CIE Luv, CIE Lab and YCbCr (Wang and Huang, that the execution time for a 320X 240 image is only 0.00375
2006; He et al., 2007) are color spaces which represent a s. However, a huge database needs to be created for this
color by its lightness (L), luminance (Y) and chromaticity system to work, and if an input image is very different from
(uv, ab and CbCr). The idea of color ratio was first those in the database, the network should be re-trained to
introduced by Barnard and Finlayson in 2000 to identify the improve the results. The well-known K-means method
„„shadow‟‟ and „„non-shadow‟‟ regions to be robust under (Lloyd) is one of the most commonly used techniques in the
changes in luminance. In 2002, the RGB ratio of the pixel clustering-based segmentation field for industrial
value to the local sum (R/Rsum, G/Gsum, B/Bsum) was applications and machine learning (Berkhin, 2002; Mignotte,
proposed by Finlayson et al. to deal with the influences of 2008). The fuzzy c-means theory (the fuzzy version of
shadows produced by variations in illumination. In addition, K-means) is applied as the clustering method (Kuo et al.,
Finlayson et al. (2005) presented an alternative RGB ratio 2008), and similarity measurement is based on Euclidean
definition, which is the ratio of the intensity of a pixel to the distance (Luis-Garcia et al., 2008). Bosch et al. (2007)
local average (R/Rave, G/Gave, B/Bave), and this formula is presented an approach that can recognize grass, sky, snow
used due to its invariance to luminance and device changes. and road using fuzzy logic with predefined classes, for which
In this paper, we propose a new RGB ratio model, which is the average processing time for an image size of 180X120 to
based on the fact that a change in the intensity of a reference 250X250 is 60 s. Efficient fuzzy c-means clustering (qFCM)
color will lead to a change in the RGB color components, but is also applied to speed up the clustering process by splitting
their ratios to the reference color (R/Rref, G/Gref, B/Bref) a target image into several small sub-images (Chen et al.,
will be linear to an intensity change (Benedek and Sziranyi, 2005). The computation time that qFCM requires for a
2007; Mikic et al., 2000). With this property, a specific color, 128X128 gray-level image is 0.1–1.2 s. The use of a template
such as the reference Colour, can be described as a linear image is another fast segmentation method. For instance, an
color model, so that it is invariant to intensity variation. image database of eyes can be established, and a skin colour
Moreover, information about the three color components database can be obtained from a colour conversion matrix
(RGB) is used to describe the chromaticity by the proposed with color of the sclera. Consequently, fixed thresholds of the
RGB ratio space. Therefore, while inheriting the HSV space are introduced to detect the skin area in an input
characteristics of HSI and RGB models, the RGB ratio has image (Do et al., 2007). However, the use of template images
several advantages with regard object recognition under is restricted to specific objects, and may require a large image
variations in intensity. database. In this paper, a dynamic fuzzy variable range is
There exist many complex and state-of-the-art techniques for proposed to achieve a high quality segmentation result.
colour segmentation which are excellent at partitioning an Firstly, the linearity between the RGB ratio and intensity is
input image. For example, the global color statistics can be estimated by a linear progressive method and parameter
represented by a set of overlapping regions and modeled by a estimation. Secondly, upper and lower boundaries are
mixture of Gaussians (GMM), and a local mixture model is obtained statistically for each colour ratio. These boundaries
described by Markov Random Fields (Kato, 2008). By are used to define the fuzzy membership functions ofcolor
optimizing parameters of the global and local models, the ratio clusters, which dynamically vary corresponding to
maximum likelihood is estimated and then a pixel can be intensity changes. The proposed fuzzy system‟s parameter
classified. Although this approach has good segmentation optimization, undertaken using a back propagation neural
results, a large number of iterations are necessary to network, makes the fuzzy decision more adaptive and more
determine the optimal parameters. As a result, 16 s of effective. Meijer (1992) used sine-wave sounds to transform
computation time is needed for an image with a 256X256 image information without any image pre-processing, while a
resolution (Tai, 2007). Hill manipulation of the colour multi-resolution approach was introduced to image-to-sound
histogram is another widely used approach to achieve colour mapping by Capelle et al. (1998).
segmentation. A three-dimensional histogram can be The present work is organized as follows: Section 2
obtained by accumulating three colour components of pixels. describes the data resources and software used. Section 3
Dominant hill detection and minor hill dismantling are then describes the enhancing colour separation of image using

564

ISSN: 2278 – 1323
decor relation stretching. Section 4 describes the K-means clusters to be located in the data. The algorithm then
clustering method. In section 5 the proposed method of arbitrarily seeds or locates, that number of cluster centers in
segmentation of image based on colour with K-means multidimensional measurement space. Each pixel in the
clustering is presented and discussed. Experimental results image is then assigned to the cluster whose arbitrary mean
obtained with suggested method are shown in section 6. vector is closest. The procedure continues until there is no
Finally, section 7 concludes with some final remarks. significant change in the location of class mean vectors
between successive iterations of the algorithms (Lille sand
Mean shift-based clustering and Keiffer, 2000). As K-means approach is iterative, it is
A clustering algorithm based on mean shift was proposed computationally intensive and hence applied only to image
subareas rather than to full scenes and can be treated as
in [13]. Unfortunately, it becomes impractical in the
unsupervised training areas (Lillesand & Keiffer, 2000).
context of texture segmentation due to the expensive
computation required in order to find the nearest neighbours
K-means-based clustering
of a point in a highdimensional space. Hence, in this work, an Due to its simplicity and good convergence properties, the
approximate version has been utilized. It starts by initializing iterative k-means algorithm is probably the most widely used
the mean shift procedure on a given point and then iterates as clustering algorithm. However, it suffers from important
usual until a stationary point is reached. However, at each drawbacks, such as the requirement of specifying the number
iteration, all points involved in the mean shift computation of clusters and the non-deterministic results produced if
are marked as “already visited”. Therefore, they are not taken random initialization is used (which is often the case).
as initial points anymore. These points are also assigned a In order to overcome the aforementioned problems, a
vote regarding their membership to the cluster associated wrapper for k-means, which is a variation of the
with the mode being detected. The algorithm repeats this resolution-driven clustering algorithm proposed in [11], has
procedure with the remaining “not visited” points. been applied. It has two main stages: split and refinement.
Once all mode candidates have been found, mode merging Regarding the split stage, let us assume that the data points
is performed by means of the same approximate mean shift have been split into
algorithm by considering the found modes as data points. If C disjoint clusters (initially C = 1). The mean distance
two modes are merged, their membership votes are also between the centroid and its associated points (intra-cluster
merged, thus keeping track of the new cluster structure. The mean distance) is computed for each cluster and the global
mode merging step is repeated until no modes are merged. mean distance (mean of intra-cluster mean distances) is
obtained for the whole partition. If this global mean distance
Membership of each point is finally determined by majority
exceeds a threshold, the largest cluster in terms of
voting.
intra-cluster mean distance is split into two. The split is done
by finding the main principal component ρ of the cluster and
Graph clustering based on the normalized cut initializing two new child centroids at c ±d, where c is the
centroid of the cluster to be split and d = ρ√2λ/π, with λ being
The graph clustering algorithm based on the normalized the eigenvalue associated with the main principal component
cut proposed in [14] has become popular in the last years. ρ. After the split stage, the refinement stage consists of
However, the main drawback of this approach is that the applying k-means using the (C + 1) available centroids as
computational technique for minimizing the normalized cut initial seeds. Both split and refinement are iterated until no
is based on eigenvectors. Thus, it suffers from scalability new clusters are generated.
problems, since in cases where the number of data points is The proposed wrapper has two main advantages over the
very large, eigenvector computation becomes prohibitive. classical k-means. First, instead of the desired number of
Recently, Dhillon et al. [15] proposed a more efficient clusters, the mean distance threshold controls the output of
technique referred to as GRACLUS, which embeds a the algorithm.Such a threshold is more intuitive and closely
weighted kernel k-means algorithm into a multilevel related to perceptual properties than the number of clusters.
approach in order to optimize locally the normalized cut. Second, the algorithm always behaves in the same way given
However, before applying GRACLUS to the pattern the same input. Therefore, there is no need for running
discovery stage, the problem of specifying the number of different trials and keeping the best set of clusters according
clusters must be addressed such as with k-means. Usually, to some criterion as it is the case when the initialization step
the alternative is to first bipartition the whole graph and then of k-means has a random component.
repartitions the already segmented parts if the normalized cut
is below a specified value [14]. Colour-Based Segmentation Using K-Means Clustering
Thebasicaimistosegmentcolorsinanautomatedfashionusingth
II. K-MEANS CLUSTERING eL*a*b*colorspaceandK-means
There are many methods of clustering developed for a wide clustering.Theentireprocesscanbesummarizedinfollowingste
variety of purposes. Clustering algorithms used for ps.
unsupervised classification of remote sensing image data Step1:Readtheimage
vary according to the efficiency with which clustering takes Readtheimagefrommother source whichisin.JPEGformat.
place (John R Jenson, 1986).K-means is the clustering Step2:ForcolorseparationofanimageapplytheDecor
algorithm used to determine the natural spectral groupings relationstretching.
present in a data set. This accepts from analyst the number of Step3:ConvertImagefromRGBColorSpacetoL*a*b*ColorSpace.

565

ISSN: 2278 – 1323
Howmanycolorsdoweseeintheimage ifweignorevariations unsupervised problem into a supervised one.
inbrightness? Therearethree colors:white,blue,andpink. As its name suggests, a pixel-based classifier aims at
Wecaneasilyvisuallydistinguish thesecolorsfromoneanother. determining the class to which every pixel of an input image
TheL*a*b*colorspace(alsoknownasCIELAB belongs, which leads to the segmentation of the image as a
orCIEL*a*b*)enablesustoquantifythese visualdifferences. The collateral effect.
L*a*b*colorspaceisderivedfromtheCIEXYZtristimulusvalues. In order to achieve this objective, several measures are
computed for each image pixel by applying a number of
The
texture feature extraction methods as described in Section 3.1.
L*a*b*spaceconsistsofaluminositylayer'L*',chromaticity-layer
'a*'indicatingwherecolor falls alongthered-greenaxis,and
Classification with multiple evaluation window sizes
chromaticity-layer'b*'indicatingwherethecolorfallsalongthe
Although previous works on supervised pixel-based
blue-yellow axis.Allofthecolorinformation
classification have already shown the benefits of utilizing
isinthe'a*'and'b*'layers.Wecanmeasurethe difference multiple evaluation window sizes [10, 11], which approach is
betweentwocolorsusingtheEuclideandistancemetric.Convertthe the best for combining these different sources of information is
imagetoL*a*b* colorspace. still an open issue.
Step4:ClassifytheColorsin'a*b*'SpaceUsingK-MeansClustering For instance, in [10], different window sizes were integrated
. by assigning a weight to their corresponding probabilities
Clusteringisa way according to how well each window size separates a given
toseparategroupsofobjects.K-meansclusteringtreatseach training pattern from the others. However, since the training
objectashaving alocationinspace. Itfindspartitions patterns are single-textured images, the assigned weight is not
suchthatobjectswithineachclusterareasclosetoeach representative of the structure of the test image, which in turn
is composed of multiple texture patterns. Furthermore, this
otheraspossible,andas farfromobjectsinotherclustersas
method may be biased to the largest window, as it captures
possible.K-meansclusteringrequires more information and, hence, has better capabilities of
thatyouspecifythenumberofclusters tobepartitioned distinguishing between texture classes. Later, in [11],
andadistancemetrictoquantifyhow improved classification rates were obtained by directly fusing
closetwoobjectsaretoeachother.Sincethecolorinformation the outcome of multiple evaluation window sizes using the
existsinthe'a*b*'space,your KNN rule. The main problem with this approach is that it does
objectsarepixelswith'a*'and'b*'values. UseK-meanstocluster not guarantee that the most appropriate window size will
theobjectsintothreeclusters usingtheEuclideandistancemetric. always receive the majority of votes.
Step5:LabelEveryPixelinthe Ideally, the strategy for classifying a test image using
ImageUsingtheResultsfromK-MEANS multiple evaluation window sizes should apply large windows
Foreveryobjectinourinput,K-meansreturnsanindexcorrespon inside regions of homogeneous texture in order to avoid noisy
classified pixels and small windows near the boundaries
ding toacluster. Labelevery pixelin
between those regions in order to define them precisely.
theimagewithitsclusterindex. Unfortunately, that kind of knowledge about the structure of
Step6:CreateImagesthatSegmenttheImagebyColor. the image is only available after it has been segmented.
Usingpixellabels,wehavetoseparateobjectsinimagebycolor, Notwithstanding, an a priori approximation of that strategy can
whichwillresultinfiveimages. be devised through the following steps:
Step 7: Segment the Nuclei into a Separate Image Step 1: Select the largest available evaluation window and
Then programmatically determine the index of the cluster classify the test image pixels labelled as unknown (initially, all
containing the blue objects because K means will not return the pixels are labelled as unknown).
same cluster idx value every time. We can do this using the Step 2: In the classified image, locate the pixels that belong
cluster center value, which contains the mean 'a*' and 'b*' to boundaries between regions of different textures and mark
value for each cluster. them as unknown, as well as their neighbourhoods.
The size of the neighbourhood corresponds to the size of the
1. Select k -seeds s.t. d ( ki , k j ) > d min window used to classify the image.
Step 3: Discard the current evaluation window.
2. Assign points to clusters by min dist. Step4: Repeat steps 1 to 3 until the smallest evaluation
Cluster ( pi ) = Arg min ( d ( pi , s j )) window has been utilized.
This scheme, which can be thought of as a top-down
s j { s1 ,…, sk } approach, has been used during the supervised classification
3. Compute new cluster centroids: stage of the proposed segmentation technique. In addition to

closely approximating the previously described ideal strategy


Cj  1  pi
for using multiple evaluation window sizes, this approach
avoids the classification of every image pixel with all the
n pi  jthcluster available windows. Hence, it leads to a lower computation
4. Reassign points to clusters (as in 2 above) time than previous approaches.
5. Iterate until no points change clusters

Supervised pixel-based classification III. RESULTS AND DISCUSSION
At this stage, the set of texture patterns found by the previous
stage are used as texture models for a supervised pixel based We implemented proposed algorithm and tested its
classifier, thus effectively transforming the original performance on a number of standard images of Mat Lab

566

ISSN: 2278 – 1323
software. We have used Peppers, Planet, Lena images from
Mat Lab software as a standard image. Addition to these
images we have implemented this proposed algorithm on
heart image also & obtain segmentation results. Figure 1(a)
shows original image of Peppers.png image and figure
1(b)-1(g) show various segmented objects from original
image. Here various color clusters and segmented objects are
clearly visible. Table 1 shows parameter values of
Peppers.png image like Min, Max, mean, median,
mode, Standard Deviation and range. Figure 1(h) shows
the scatter plot of original image Peppers.png. Figure 1 (i)
shows Scatter plot with Bar and values of Peppers.png image.
Figure 1 (j) shows Graph of Parameter values of Peppers.png
image. Figure 1 (k) shows Radar Graph of Parameter values
of Peppers.png image. Figure 2 (a) shows the second image
of our test data image of original Planets standard image from
mat lab software. Figure 2(b) and 2(c) shows Object Figure 1 (c) Object Segmentation from Peppers image having
Segmentation from Planets image. Table 2 shows Parameter light green color
Values of Planets image. Figure 2(d) shows Scatter plot of
Planets image. Figure 2 (e) represents Graph of Parameter
values of Planets image and Figure 2 (f) represents Radar
Graph of Parameter values of Planets image. Similarly Figure
3 shows results for Lena Image. Figure 4 shows segmentation
results of Heart image.

Figure 1 (d) Object Segmentation from Peppers image having
red color

Figure 1 (a) Original Peppers standard image from matlab

Figure 1 (e) Object Segmentation from Peppers image

Figure 1 (b) Object Segmentation from Peppers image having
orange color

567

ISSN: 2278 – 1323

220

60
200

50
180

Black
40
160 105 x min
150 x max
127 x mean 30
140 128 x median
136 x mode
120 9.173x std 20
Red
Green
100 Violet 10
Magenta
Yellow
80
100 120 140 160 180 200 220

Figure 1 (f) Object Segmentation from Peppers image Figure 1 (i) Scatter plot with Bar and values of Peppers.png
image

Figure 1 (j) Graph of Parameter values of Peppers.png image

Figure 1 (g) Object Segmentation from Peppers image

Black X Min
Scatterplot of the segmented pixels in 'a*b*' space Yellow Y250 Black Y
220 200 Max
Yellow X 150 Red X
200
100 mean
Magenta 50
0 Red Y median
180 Y
Magenta mode
Green X
X
'b*' values

160
std
Violet Y Green Y
140
Violet X range
120

100

80 Figure 1 (k) Radar Graph of Parameter values of Peppers.png
100 120 140 160 180 200 220
'a*' values
image
Figure 1 (h) Scatter plot of Peppers.png image

568

ISSN: 2278 – 1323

Table 1: Parameter Values of Peppers.png image

Peppers.png Min Max mean med mode STD ran
ge
Black X 105 150 127.21 128 136 9.1729 45
Black Y 126 160 147.13 148 148 6.6843 34
Red X 106 156 122.8 120 115 9.467 50
Red Y 152 176 165 165 167 4.936 24
Green X 156 201 183.05 185 187 7.9559 45
Green Y 133 201 169.10 168 173 12.5043 68
Violet X 128 179 155.5 156 168 12.93 51
Violet Y 176 214 202.4 204 204 7.818 38
Magenta X 110 156 126.3 123 121 9.347 37
Magenta Y 163 200 181.3 182 185 6.347 37
Yellow X 126 184 147.6 147 147 4.66 58
Yellow Y 92 153 115.5 115 115 6.838 61
Figure 2(c) Object Segmentation from Planets image

Scatterplot of the segmented pixels in 'a*b*' space
200

180

160
'b*' values

140

120

100

80

60
120 130 140 150 160 170 180 190 200
Figure 2 (a) Original Planets standard image from matlab 'a*' values

Figure 2(d) Scatter plot of Planets image
Table 2: Parameter Values of Planets image

250
Planets.jpg Min Max mean med mode std range
Red X 120 161 134.2 133 132 4.6 41 200
Red Y 61 121 97.84 96 95 11.52 60
Violet X 120 199 134.7 131 128 12.01 79
150 Red X
Violet Y 118 192 130.8 127 128 12.1 74
Red Y
100
Violet X
50
Violet Y
0

Figure 2 (e) Graph of Parameter values of Planets image

Figure 2(b) Object Segmentation from Planets image

569

ISSN: 2278 – 1323

Min
200
range 150 Max
100 Red X
50 Red Y
0
std mean Violet X
Violet Y

mode median

Figure 2 (f) Radar Graph of Parameter values of Planets
image Figure 3(b) Object Segmentation from Lena image

Figure 3(c) Object Segmentation from Lena image

Figure 3 (a) Original standard image of Lena from matlab

570

ISSN: 2278 – 1323

Lena.tif Min Ma mea media mod std rang
f x n n e e
Black X 168 190 173.7 174 174 2.91 22
3
Black Y 140 187 151.3 151 149 3.99 47
1
Red X 166 182 171 171 172 2.40 16
2
Red Y 127 148 142 142 143 3.29 21
5
Green X 147 176 161 163 165 5.96 29
2
Green Y 124 143 133.6 134 141 5.32 19
2
Violet X 132 178 161.1 162 163 4.51 46
9
Violet Y 90 125 116.2 117 120 5.71 35
4
Magenta 125 148 139.5 139 138 3.63 23
Figure 3(d) Object Segmentation from Lena image X
Magenta 109 182 143.4 141 139 8.52 73
Y 3
Yellow 133 169 157.9 158 156 6.08 36
X 1
Yellow 142 210 152.1 151 146 6.85 68
Y 5
Table 3: Parameter Values of Lena image

220

200

180
'b*' values

160

140

120

100

Figure 3(e) Object Segmentation from Lena image
80
120 130 140 150 160 170 180 190
'a*' values

Figure 3(f) Scatter plot of Lena image

571

ISSN: 2278 – 1323

250

200
Min
150 Max
mean
100
median
50 mode
std Figure 4 (b) Segmented object1 of Heart image
0
range
Magenta X
Magenta Y
Violet X
Violet Y

Yellow Y
Red Y

Yellow X
Red X

Green Y
Green X
Black Y
Black X

Figure 3(g) Graph of Lena image parameter values

Figure 4 (c) Segmented object2 of Heart image

Min
Black X
Yellow 250
Y
200 Black Y Max
Yellow X 150 Red X mean
100
Magenta 50
0 Red Y median
Y
Magenta mode
Green X
X
Violet Y Green Y std
Violet X
range
Figure 4 (d) Segmented object3 of Heart image
200

180
Figure 3(h) Radar Graph plot of Lena image parameter
values 160
'b*' values

140

120

100

80

60
110 120 130 140 150 160 170 180 190 200
'a*' values

Figure 4 (e) Scatter plot of Heart image

Figure 4 (a) Original image of Heart

572

ISSN: 2278 – 1323

Figure 5 Quantitative Comparison of Segmentation Methods

Table 4: Methods

573

ISSN: 2278 – 1323
[4] Jean-Christophe Devaux et al; Aerial colour image segmentation by
Karhunen-Loeve transform, 0-7695-0750-6, IEEE 2000, pp 309-
312.
[5] Jun Tang, A color image segmentation algorithm based on region
growing, 978-1-4244-6349-7, IEEE, vol 6, 2010, pp. 634-637
[6] Mehmet Nadir Kurnaz,et al; Segmentation of remote-sensing images
by incremental neural network, Pattern Recognition Letters 26
(2005) 1096–1104, pp 1096-1103.
[7] N Bartneck et al; Colour segmentation with polynomial
classification,0-8186-2915-0/92, 1992, pp. 635-638.
[8] Nae-Joung Kwak et al; color image segmentation using edge and
adaptive threshold value based on the image characteristics,IEEE
proceeding0-7803-8639-6,2004, pp 555-558.
[9] Lingkui Meng, et al, Study on Image Segment Based Land Use
Classification and Mapping, 2009 IEEE, pp
[10] Pal N.R.et al; A review on image segmentation techniques, Pattern
Recognition 26(9), 1993, pp1277-1294.
[11] Ricardo Dutra da Silva et al; Satellite image segmentation using
wavelet transforms based on color and texture features, ISVC 2008,
part II, LNCS 5359, 2008, pp 113-122
[12] Robert A Schowengerdt, Remote sensing- models and Methods for
Image Processing, IIIrd edition, Elsevier Inc.
[13] T. W. Chen, Y. L. Chen and S. Y. Chien. “Fast image segmentation
based on K-Means clustering with histograms in HSV color space”. In
Proceedings of 10th IEEE Workshop on Multimedia Signal
Processing, 2008.
[14] C.W. Chen, J. Luo, K.J. Parker, “Image segmentation via adaptive
K-mean clustering and knowledge based morphological operations
with biomedical applications”, IEEE Transactions on Image
Processing, Vol.7 (12), 1998, pp 1673-1683.
[15] B. Sowmya, B. Sheelarani, “Colour Image Segmentation Using Soft
Computing Techniques”.
International Journal of Soft Computing Applications, 4:69-80, 2009.
[16] M. Mirmehdi, M. Petrou, Segmentation of color textures, IEEE Trans.
Pattern Anal. 22 (2000) 142-159.
[17] S.C. Kim, T.J. Kang, Texture classification and segmentation using
wavelet packet frame and Gaussian mixture model, Pattern Recogn. 40
(2007) 1207-1221.
[18] D. Puig, M.A. Garcia, Automatic texture feature selection for image
pixel Classification, Pattern Recogn. 39 (2006) 1996-2009.
[19] J. Melendez, M.A. Garcia, D. Puig, Efficient distance-based
per-pixeltexture classification with Gabor wavelet filters, Pattern
Anal. Appl. 11
(2008) 365-372.
[20] M. Omran, A. Engelbrecht, A. Salman, An overview of clustering
methods, Intell. Data Anal. 11 (2007) 583-605.
[21] D. Comaniciu, P. Meer, Mean shift: A robust approach toward feature
Figure 6 Quantitative Comparisons of Segmentation Space analysis, IEEE Trans. Pattern Anal. 24 (2002) 603-619.
Methods [22] J. Shi, J. Malik, Normalized cuts and image segmentation, IEEE
Trans.
Pattern Anal. 22 (2000) 888-905.
IV. CONCLUSION [23] I.S. Dhillon, Y. Guan, B. Kulis, Weighted graph cuts without
eigenvectors: A multilevel approach, IEEE Trans. Pattern Anal. 29
(2007) 1944-1957.
We have presented an efficient implementation of k-means [24] D. Tsujinishi, Y. Koshiba, S. Abe, Why pairwise is better than
clustering algorithm. The algorithm has been implemented oneagainst-all or all-at-once, in: Proceedings of the IEEE IJCNN,
on standard images from mat lab software. Results are plotted 2004, pp.693-698.
in scatter plots showing the clusters & Radar plot showing [25] W.-Y. Ma, B.S. Manjunath, Edge Flow: A technique for boundary
detection and image segmentation, IEEE Trans. Image Process. 9
the data analysis of clusters. Various segmentation methods (2000) 1375-1388.
are given in form of chart. The plot of segmentation method [26] A.Y. Yang et al., Unsupervised segmentation of natural images via
shows unsupervised k means cluster Method is better as lossydata compression, Comput. Vis. Image Und. 110 (2008) 212-225.
compared to supervised classification segmentation methods.
Janak B. Patel (born in 1971) received B.E.
And the more well separated the clusters, the faster the (Electronics & Communication Engg from L.D.
algorithm runs. This algorithm is significantly more efficient College of Engg. Ahmedabad, and M.E.
than the other methods. (Electronics Communication & System Engg.) in
2000 from DDIT. He is Asst. Prof. & H.O.D. at
REFERENCES L.D.R.P.I.T.R., Gujarat. He is pursuing Ph.D. at
Indian Institute of Technology, Roorkee.
[1] Ahmed Darwish, et al, Image Segmentation for the Purpose Of
Object-Based Classification,, 2003 IEEE pp. 2039-2041 R R.S. Anand received B.E., M.E. and Ph.D. in
[2] Darren MacDonald, et al; Evaluation of colour image segmentation Electrical Engg. from University of Roorkee in
hierarchies, proceeding of the 3rd Canadian conference on 1985, 1987 and 1992, respectively. He is a
Computer and Robot Vision, IEEE, 2006. professor at Indian Institute of Technology,
[3] H C Chen et al, Visible color difference-based quantitative evaluation Roorkee. He has published more than 100
of colour segmentation, IEEE proceeding, Vis image signal process research papers in the area of image processing and
vol.153 No.5 Oct 2006 pp 598-609. signal processing.

574

563 574

More Related Content

What's hot (18)

Viewers also liked (7)

Similar to 563 574 (20)

More from Editor IJARCET (20)

Recently uploaded (20)

563 574