ISSN: 2277-3754

ISO 9001:2008 Certified

International Journal of Engineering and Innovative Technology (IJEIT)
Volume 4, Issue 1, July 2014

A Study on the Different Image Segmentation

Technique
Rozy Kumari, Narinder Sharma

Abstract:-This paper presents an overview of some well- diverse regarding the attributes. It should be noticed that
known image segmentation techniques. The segmentation no general technique has been developed yet to segment an
process divides a given image into different regions and objects.
Image Segmentation has become popular due to its many vision image precisely, so different techniques are taking floor to
applications. The main goal of this survey is to explore various perform segmentation [3]. In the next section, some of the
algorithms of image segmentation. The short-comings of the image processing methods have been explained.
survey on image segmentation algorithms have also been
evaluated. The main focus of this paper is on the clustering
based segmentation techniques. This paper concludes with
certain limitations of available techniques and also the possible
solutions for the same for future use.

Index terms: Edge detection, region growing segmentation,

Watershed, Clustering, Thresholding.

I. INTRODUCTION
Image Segmentation is a vital procedure of processing
and understanding an image. It is the fundamental
necessity of any computer based vision application on the
grounds that individuals are mostly interested just in
specific parts of the picture. Essentially it is characterized
as the procedure of isolating the picture into distinctive
parts of homogeneity. An image might be defined as a
two-dimensional function, f(x, y), where x and y are the
spatial coordinates, and the amplitude of at any pair of Fig 1 Segmentation results
coordinates is known as the intensity of the image at
specific point. When the intensity of f and the value of x,y II. IMAGE SEGMENTATION TECHNIQUES
are finite then the image is called digital image.[1] A. Edge Detection Methods
The objective of image segmentation is to simplify the Edge detection techniques converts images to edge
representation of a picture into something that is more images thus having benefit from the change of grey tones
genuine and simpler to understand. It is essentially used to in the images. Edges are the sign of lack of continuity, and
discover the location of objects, boundaries, lines and so ending, as a result of this transformation, edge image is
on in the digital images. More precisely, image obtained without encountering any changes in physical
segmentation is the process of assigning a label to every qualities of the main image[4][5]. Objects have various
pixel in the image such that pixels with the same label parts of different color levels. In an image with different
share certain visual characteristics or features. [2] grey levels, despite an obvious change in the grey levels of
Essentially Image segmentation entails partitioning an the object, the shape of the image can be distinguished in
image by grouping similar pixels together. Ideally these Figure 2 [5].
groups should represent the components that the image is
comprised of. The image can then instead be interpreted
using a small number of well-defined components as
opposed to a large array of unrelated pixels. Wikipedia
provides a nice definition: Image segmentation is the
process of assigning a label to every pixel in an image
such that pixels with the same label share certain visual
characteristics.
The output of image segmentation is a set of regions that
together cover the entire image. The consequence of
picture division is a situated of areas that together blanket Fig 2: Type of Edges (a) Step Edge (b) Ramp Edge (c) Line
the whole picture. Each of the pixels in a area are Edge (d) Roof Edge
comparative regarding some property, for example, color, An Edge of the image is an important local change in
texture and intensity. Adjoining regions are fundamentally the intensity of the image, usually related with a

284
 ISSN: 2277-3754
ISO 9001:2008 Certified
International Journal of Engineering and Innovative Technology (IJEIT)
Volume 4, Issue 1, July 2014
discontinuity in either the first derivative of the image segmentation techniques are known as region-based
intensity or the image intensity. Discontinuities in the segmentation and edge detection, respectively [11].
image intensity can be either Step edge, where the image The splitting and merging techniques of the region starts
intensity abruptly changes from one value on one side of with splitting of an image into number of small regions
the discontinuity to a different value on the opposite side, and this proves continues till regions with mandatory
or Line Edges, where the image intensity abruptly changes degree of homogeneity are formed. The splitting of the
value but then returns to the starting value within some phase has an impact on the overall segmentation of the
short distance [6]. Though Step and Line edges are rare in image. This results in over segmentation of the image
real images. Due to low frequency components or the which is further followed by the merging phase. Thus
smoothing introduced by most sensing devices, sharp region splitting and merging techniques are complex and
discontinuities do not exist in real signals. Step edges consumes a lot of time. The main objective of region
become Ramp Edges and Line Edges become Roof edges, growing is to map individual pixels called seeds in input
where intensity changes are not instantaneous but occur image to a set of pixels called region and region growing
over a finite distance [7]. Illustrations of these edge shapes method starts with initial seeds and grows with
are shown in Fig.2. neighboring homogenous elements where seed may be
1. Steps in Edge Detection pixel or region [3]. The technique of the region growing is
Edge detection consists of three steps namely Filtering, an iterative process.
Enhancement and Detection. The overview of the steps in
edge detection are as follows. 1. Steps in Region Based Method
a)Filtering: Images are regularly corrupted by arbitrary a) The entire image is segmented into pattern cells
variations in intensity values, this is called noise. Salt and b) Comparison of every pattern cell is done with its
pepper noise, impulse noise and Gaussian noise are some neighboring cells to determine whether they are similar,
of the type of noise. Salt and pepper noise holds arbitrary using a similarity measure. In case they are similar, merge
events of both dark and white intensity values. the cells to form a fragment and the property used in the
Notwithstanding, there is an exchange off between edge comparison is updated.
strength and noise lessening.. More filtering to reduce c) After comparison, continue growing the fragment by
noise results in a loss of edge strength [8]. examining all its neighbors until no joinable regions
b) Enhancement: To facilitate the detection of edges, it is remain. Then label the fragment as a completed region.
necessary to find changes in intensity in the neighborhood d) Move to the next uncompleted cell, and repeat these
of a point. Enhancement emphasizes pixels where there is steps until all cells are labeled [11].
a significant change in local intensity values and is usually
performed by computing the gradient magnitude [9].
c) Detection: A lot of points in an image have a nonzero
value for the gradient, and all of these points are not edges
for a particular application. Therefore, some method
should be used to determine which points are edge points.
Frequently, thresholding provides the criterion used for
detection [10].

B. Region Based Methods

The process of segmentation is one of the very first
steps in the various remote sensing image analyses. Fig 3: a) Input Image b) Segmented Image [12]
Generally the image is divided into regions which
represent the relevant objects in best method in the scene. III. THRESHOLD BASED SEGMENTATION
Various region properties like area, shape, statistical Threshold technique is one of the most used techniques
parameters and texture can be extracted and used for in image segmentation. This technique can be expressed
additional analysis of the data. The segmentation task can as:
be accomplished in two ways: 1) dividing up the images T=T[x, y, p(x, y), f(x, y]
into a number of homogeneous regions, each having a Where: T is the threshold value [11] .
unique label, 2) determining boundaries between x, y are the coordinates of the threshold value point. p(x,y)
homogeneous regions of different properties and these ,f(x,y) are points the gray level image pixels.
285
 ISSN: 2277-3754
ISO 9001:2008 Certified
International Journal of Engineering and Innovative Technology (IJEIT)
Volume 4, Issue 1, July 2014
Threshold image g(x,y) can be define [13]: 2(T) +P2(T)2
2(T) (4)
Where:
C (T) is the within-group variance.
P1 (T) is the probability for group with values less than T.
P2(T) is the probability for group with values greater than
1. Some of threshold techniques T.
a) Mean Technique
1 (T) is the variance of group of pixels with values less
This procedure utilized the mean estimation of the
pixels as the threshold value and works well in strict than or equal
instances of the images that have roughly half to the pixels T.2 (T) is the variance of group of pixels with values
having a place with the objects and the other half to greater than T.
background. This strategy case seldom happens.
d)EMT Technique
b) P-Tile Technique The threshold image by using this technique is used
The p-tile technique uses knowledge about the area size when there are extra homogenous region in image or when
of the desired object to the threshold an image. And this there is a change on illumination in the object and its
method is one of the earliest threshold methods based on background.
the gray level histogram [13]. This technique supposes that
the objects in an image are brighter than the background, e) Visual Technique
and occupy a particular percentage of the picture area. This These techniques enhance peoples ability to exactly
fixed percentage of picture area is known as P%. The search for target items. These techniques are similar to P-
threshold is defined as the gray level that mostly Tile technique as they all uses the component segments of
corresponds to mapping at least fixed percentage of picture original images in new ways to improve visual search
area of the gray level into the object. performance but is dissimilar from p-tile.
Let n be the maximum gray level value, H (i) be the
histogram of image (i = 0. n), and P be the object area C. Clustering Based Segmentation
ratio. The algorithm of the P-tile method is as follows: Clustering method is a procedure in which a data set or
S=sum (H(i)) (3) say pixels are exchanged by cluster, pixels may belong
Let f=s together because of the same color, texture etc. There are
For k=1 to n two types of clustering that is hierarchical clustering and
f=f-H(k-1) partional clustering [16].
If (f/t) <p then stop
T=k 1. Some of the Clustering methods
Where: S total area of image a) K- means clustering method
f is the initialize all area as object area This is an iterative technique used to partition an image
into lot of clusters in which there is option of k cluster
T is the final threshold value [13]
centers; randomly each pixel in the image is assign to the
This method is simple and suitable for all sizes of objects.
every cluster. The cluster centers is again computed by
It yields good anti-noise capabilities; however, it is taking mean of all the pixels in the centre. The quality of
obviously not applicable if the object area ratio is explanation obtained from this method depends upon the
unknown or varies from picture to picture [14]. initial set of clusters and the value of k.

c) Histogram Dependent Technique (HDT) b) C-means clustering method

These are dependent on the success to estimate the It is fuzzy clustering method where each point has a
threshold value that separate the two homogonous region degree of belonging of clustering as in fuzzy logic, rather
of the object and background of an image. The HDT is than belongs to just one cluster. In this, centroid of a
right for image having large homogonous and will split cluster is the mean of all points, weighted by then degree
regions where all area of the objects and background are of belonging to the cluster. It is also called soft k-means
homogonous apart from the area in between the objects clustering method.
and background.
This technique [14] can be expressed as : c) FELICM
C (T) =P1 (T)1
286
 ISSN: 2277-3754
ISO 9001:2008 Certified
International Journal of Engineering and Innovative Technology (IJEIT)
Volume 4, Issue 1, July 2014
It is abbreviation of Fuzzy C-Means with Edge and Amrita Mohanty et al.(2013) [19] This paper had
Local Information, which introduces the pixels weights described the analysis of color images using cluster based
within local neighbor windows to decrease the edge segmentation techniques. It has discussed the various
degradation. Mainly this method has somehow used to segmentation techniques like thresholding, edge-based
overcome the isolated distribution of pixels inside detection, region based detection and clustering. The main
segments of image. illustration is based upon the color image segmentation
using k-means, fuzzy c-means and clustering and
d ) Mean Shift corresponding results have been compared. R.Ravindraih
This is an advanced technique for clustering based et al. (2013) [20] This is a survey paper which as
segmentation and is a non parametric iterative algorithm. It represented the various image segmentation algorithms
is implemented through kernel density estimation which is based on fuzzy clustering. It had described the use of
a non parametric way to approximate the density of a image segmentation with the medical image processing. It
random variable. It is a popular method for estimating had covered different types of fuzzy c-means algorithms
probability. For every data point, mean shift defines a like fcm, improved fcm, bias corrected fuzzy c-means and
window around it and computes the mean of data point their corresponding use to the image segmentation.
Hamed Shamsi et al (2012) [21] this paper illustrated
IV. LITERATURE SURVEY that the traditional way to segment the MR images is the
Nan Li et al. (2013) [17] have proposed an image fuzzy c-means (fcm) clustering algorithm. This method
clustering method called fuzzy C-means with edge and basically sensitive to noise. So to overcome the traditional
local information(FELICM), which decreases the edge method a new modified fuzzy c-means clustering with
degradation by adding the weights of pixels within local spatial information for image segmentation was proposed.
neighbor windows. They had used the canny edge The proposed method used the spatial information for the
detection for edge extraction, adaptive threshold values. It better results as compared to fcm. The results proved that
had been shown that the proposed method could be the proposed method was more robust to noise than the
directly applied without using the filters. This approach standard fcm algorithm. Santanu Bhowmik et al(2012) [2]
has delivered the better results with respect to fuzzy c This paper basically represented the survey on clustering
means clustering method and mean shift approach. The based image segmentation. It described the various
experiments show that the proposed FELICM method is clustering techniques to achieve image segmentation.
insensitive to the isolated regions and more accurate edges Clustering is done based on different attributes of an image
than FLICM. such as size, color, texture etc. This paper covered the
Monika Xess et al.(2013) [18] This paper presented an types of clustering such as log-based clustering,
overall search over the topic clustering based color image hierarchical clustering. It had described the segmentation
segmentation and novel approaches to FCM algorithm. in relevant to various image processing techniques. S.
This paper described two techniques namely K-means Krinidis et al. (2010) [23] had proposed algorithm named
clustering and fuzzy c-means (FCM) clustering for better fuzzy local information C-means (FLICM) which can
segmentation results. The benefits of Spatial Fuzzy C- remove the shortcomings of the already known fuzzy c-
means (SFCM) which overcomes the limitation of means algorithms. It used fuzzy local information like
conventional FCM towards noisy image have been spatial and gray level which guarantee noise sensitiveness
discussed. Thresholding by Fuzzy C-means (THFCM) and robustness to noisy images.
approach solves the problem of existing method to Y. Tarabalka et al (2009) [24] had introduced a new
determine a threshold for excellent segmentation. Navneet spectral-spatial classification scheme for the hype spectral
Kaur et al.(2013) [16] This paper had presented the use of images which combines the results obtained by the pixel
image segmentation in lesion segmentation which is wise support vector machine classification and the
needed for monitoring and quantifying lesion. It had also segmentation map obtained by partitional clustering. The
illustered the k-means clustering method which is an proposed schemes perform well for the recognition of
iterative technique that is used to partition an image into images having large spatial structures. Li et
clusters in which there is choice of k clusters along with al.(2012)[25]has suggested new approach to lessen over-
the types of clustering that is hierarchical clustering and segmentation using both pre and post processing for
partitioned clustering. It has also given the description watershed segmentation. A new watershed segmentation
over the region growing segmentation which is process that combines pre-processing of the image and
partitioning of an image into homogeneous connected post-processing of image objects to produce the final
pixels. Each pixel in the region is similar due to some segmentation results. In the initial stage of the watershed
property such as color and texture. transform, this not only generates a gradient image from

287
 ISSN: 2277-3754
ISO 9001:2008 Certified
International Journal of Engineering and Innovative Technology (IJEIT)
Volume 4, Issue 1, July 2014
the original image. It also presents the texture gradient. effect that colors have on segmentation results has also
The texture gradient can be obtained by a gray-level co- been neglected.
occurrence matrix. So in near future we will propose a new integrated Edge
Ghoshale et al. (2013) [28] has described the several preserving smoothing, region growing and FELICM based
edge sharpening filters and to discover the effect on the image segmentation algorithm to improve the accuracy of
output image using watershed algorithm. A spatial the segmentation procedure further. The motivation behind
sharpening filter on the performance of the segmented the proposed approach is simple and effective. First of all
images and mathematical morphology plays a very edge preserving smoothing will filter the objects available
important role. Morphology is the special type of filtering in digital image so that the complex objects can also be
and structuring elements used for shape smoothing and easily detected. However, the use of HSV has the ability to
elimination of small holes. Rahman et al. (2013)[29] segment the color images in efficient manner. The actual
present, a novel image segmentation process based on segmentation is done by using the integrated region
adaptive threshold and masking operation with watershed growing and FELICM based image segmentation
algorithm. Whose objective is to overcome over- algorithm.
segmentation problem of the traditional watershed
algorithm. REFERENCES
[1] P. S. Liao, T.-S. Chen and P.-C. Chung, A fast algorithm
for multilevel thresholding, J. Inf. Sci. Eng., vol. 17, no. 5,
V. GAPS IN LITERATURE
pp. 713727, Sep. 2001.
Following are the various gaps in earlier work on image
segmentation techniques. [2] Santanu Bhowmik, Viki Datta, A Survey On Clustering
Based Image Segmentation, International Journal Of
1. Isolated regions: Most of the existing techniques Advanced Research In Computer Engineering &
have focused on the complex regions. Not much Technology Volume 1, Issue 5, July 2012.
work done for the images with mixed regions. [3] Verma, Om Prakash, Madasu Hanmandlu, Seba Susan,
2. Principal region extraction: The effect of the regions Muralidhar Kulkarni, and Puneet Kumar Jain. A simple
on the segmentation has been neglected by many single seeded region growing algorithm for color image
segmentation using adaptive thresholding IEEE
researchers.
International Conference on Communication Systems and
3. Effect of color: The effect of the color on the Network Technologies (CSNT), pp. 500-503., June, 2011.
segmentation results has also been neglected by many
[4] Ibrahiem M. M. El Emary, On the Application of Artificial
researchers. Neural Networks in Analyzing and Classifying the Human
4. Accuracy and complexity: The results [base paper] Chromosomes, Journal of Computer Science, vol.2 (1),
has shown that the FELICM has shown better results 2006, pp.72-75.
when the window size 11*11. But 11*11 mask will [5] N. Senthilkumaran and R. Rajesh, A Study on Edge
result in high computational time complexity. Also Detection Methods for Image Segmentation, Proceedings
of the International Conference on Mathematics and
result for standard mask size i.e. 3*3 it has shown Computer Science (ICMCS-2009), 2009, Vol.I, pp.255-259.
poor results than mean shift algorithm.
[6] M. Abdulghafour,Image segmentation using Fuzzy logic
and genetic algorithms, Journal of WSCG, vol. 11, no.1,
VI. CONCLUSION AND FUTURE SCOPE 2003.
Image Segmentation is a process of image processing
[7] Metin Kaya,Image Clustering and Compression Using an
and understanding. It is defined as the process of dividing Annealed Fuzzy Hopfield Neural Network, International
the image into parts based on homogeneity. The purpose of Journal of Signal Processing,2005, pp.80-88.
image segmentation is to make the representation of an
[8] N. Senthilkumaran and R. Rajesh, A Study on Split and
image simpler into something that is more meaningful and Merge for Region based Image Segmentation, Proceedings
easier to understand. The process in which a data set or of UGC Sponsored National Conference Network Security
say pixels are replaced by cluster, pixels may belong (NCNS-08) , 2008, pp.57-61.
together because of the same color, texture etc is known as
[9] Xian Bin Wen, Hua Zhang and Ze Tao Jiang, Multiscale
Clustering based method. The survey has shown that of the Unsupervised Segmentation of SAR Imagery Using the
existing techniques, main focus is on complex regions. Genetic Algorithm, Sensors, vol.8, 2008, pp.1704-1711.
Therefore not much work has been done for the images
[10] Mantas Paulinas and Andrius Usinskas, A Survey of
with mixed regions. The effect of the regions on the Genetic Algorithms Applications for Image Enhancement
segmentation has been neglected by many researchers. The and Segmentation, Information Technology and Control,
Vol.36, No.3, 2007, pp.278-284.

288
 ISSN: 2277-3754
ISO 9001:2008 Certified
International Journal of Engineering and Innovative Technology (IJEIT)
Volume 4, Issue 1, July 2014
[11] Bins, L. Santanna, Leila M. Garcia Fonseca, Guaraci Jos [25] Boren Li, Mao Pan, and Zixing Wu, An improved
Erthal, and F. Mitsuo Ii. Satellite imagery segmentation: a segmentation of high spatial resolution remote sensing
region growing approach Simpsio Brasileiro de image using Marker-based Watershed Algorithm, 20th
Sensoriamento Remoto, pp. 677-680, April 1996. IEEE International Conference on Geoinformatics, Year
2012, pp.1-5.
[12] http://www.mathworks.in/discovery/image-
segmentation.html [26] Farheen K. Siddiqui and Vineet Richhariya, An Efficient
Image Segmentation Approach through Enhanced
[13] Du Gen-yuan,Miao Fang,Tian Sheng-li,Guo Xi- Watershed Algorithm, Computer Engineering and
rong.,"Remote Sensing Image Sequence Segmentation Intelligent Systems, Vol.4, No.6, pp 1-7, 2013.
Based on the Modified Fuzzy C-means", Journal of
Software, Vol. 5, No. 1, PP.28-35, 2009. [27] P.P. Acharjya, A. Sinha, S. Sarkar, S. Dey and S. Ghosh,A
New Approach Of Watershed Algorithm Using Distance
[14] A.S. Abutaleb, Automatic Thresholding of Gray-Level Transform Applied To Image Segmenttion,International
Pictures Using Two Dimensional Entropy, Computer Journal of Innovative Research in Computer and
Vision, Graphics, and Image Processing, Vol.47, PP.22-32, Communication Engineering, Vol.1,No.2, pp 185-189, April
1989. 2013.
[15] Salem Saleh Al-amri1, N.V. Kalyankar2 and Khamitkar [28] Dibyendu Ghoshal and Pinaki Pratim Acharjya,Effect of
S.D, Image Segmentation by Using Threshold Various Spatial Sharpening Filters on the Performance of
Techniques, JOURNAL OF COMPUTING, VOLUME 2, the Segmented Images using Watershed Approach based on
ISSUE 5, MAY 2010. Image Gradient Magnitude and Direction, International
[16] Navneet Kaur, Gagan Jindal "A Survey Of K Means Journal of Computer Applications, Vol. 82, No.6, pp 19-25,
Clustering With Modified Gradient Magnitude Region November 2013.
Growing Technique For Lesion Segmentation, [29] Md. Habibur Rahman and Md.RafiqualIslam,
International Journal Of Innovations In Engineering And Segmentation of color image using adaptive thresholding
Technology, April 2013. and masking with watershed algorithm,IEEE International
[17] Nan li, Hong huo, Yu-ming zhao, Xi chen, and Tao fang, A Conference on Informatics, Electronics & Vision (ICIEV),
Spatial Clustering Method with Edge weighting for Image Year 2013, pp.1-6.
segmentation, IEEE Geoscience And Remote Sensing
Letters, Vol. 10, No. 5, September 2013.
AUTHOR BIOGRAPHY
[18] Monika Xess, S. Akila Agnes Survey On Clustering Based
Color Image Segmentation And Novel Approaches To Fcm
Rozy Kumari, Pursuing M.tech in Electronics and communication
Algorithm, International Journal Of Research In engineering from Amritsar college of Engineering and technology,
Engineering And Technology, Vol. Issue: 2013. Amritsar, Punjab, India
[19] Amrita Mohanty, S Rajkumar, Zameer Muzaffar Mir And Narinder Sharma, H.O.D of Electrical and Electronics Engineering in
Puja Bardhan,"Analysis Of Color Images Using Cluster Amritsar college of Engineering and technology, Amritsar, Punjab, India.
Based Segmentation Techniques", International Journal Of
Computer Applications, October 2013.
[20] R.Ravindraiah, K.Tejaswini,"A Survey Of Image
Segmentation Algorithms Based On Fuzzy Clustering",
International Journal Of Computer Science And Mobile
Computing "'July 2013.
[21] Hamed Shamsi And Hadi Seyedarabi, A Modified Fuzzy
C-Means Clustering With Spatial Information For Image
Segmentation, International Journal Of Computer Theory
And Engineering, Vol. 4, No. 5, October 2012.
[22] S. Krinidis And V. Chatzis, A Robust Fuzzy Local
Information C-Meansclustering Algorithm, Ieee Trans.
Image Process., Vol. 19, No. 5,Pp. 13281337, May 2010.
[23] Y. Tarabalka, J. A. Benediktsson, and J. Chanussot,
SpectralSpatial Classification of Hyper spectral Imagery
Based on Partitional Clustering techniques, Ieee Trans.
Geosci. Remote Sens., Vol. 47, No. 8, Pp. 29732987, Aug.
2009.
[24] Z.Xiangrong, L.Jiao, F.Liu, L.Bo and M.Gong'"Spectral
Clustering Ensemble Applied to SAR Image
Segmentation"IEEE Trans, Geosci. Remote Sens, Vol.46,
no.7 Jul 2008.
289