International Journal of Engineering and Technical Research (IJETR)

ISSN: 2321-0869 (O) 2454-4698 (P), Volume-4, Issue-1, January 2016

Analysis of Image Segmentation Techniques: A

Survey
Apurv Vashisht, Shiv Kumar

Cooperative Segmentation Methods. The author came to a

Abstract This paper presents an overview of some conclusion that the Edge-based techniques cannot be
well-known image segmentation techniques. The segmentation implemented by themselves and are better in serving as a step
process divides a given image into different regions and objects. in other image segmentation algorithms because they provide
Image Segmentation has become popular due to its many vision useful information about the region boundaries.
applications. The main goal of this survey is to explore various
Rajeshwar et al. 2012 [2] Image Segmentation Techniques
algorithms of image segmentation. The short-comings of the
survey on image segmentation algorithms have also been
has done a very broad classification of various image
evaluated. The main focus of this paper is on the clustering segmentation techniques into six categories called
based segmentation techniques. This paper concludes with Segmentation Based on Edge Detection, Thresholding
certain limitations of available techniques and also the possible Method, Region Based Segmentation Methods, Segmentation
solutions for the same for future use. Based on PDE (Partial Differential Equation), Segmentation
Based on Artificial Neural Networks, Segmentation Based on
Index Terms Segmentation, Thresholding, Clustering, Clustering, and Multi-objective Image Segmentation,
Edge Detection, Image Processing. although being a comprehensive paper, it does not appear to
show as to what algorithms are suitable for what type of
images.
I. INTRODUCTION Basavaprasad et al. 2013 [4] A comparative study on
Image Segmentation is a vital procedure of processing and classification of image segmentation methods with a focus on
understanding an image. It is the fundamental necessity of any graph based techniques gives three major categories that
computer based vision application on the grounds that consist of various sub-categories to withhold the various
individuals are mostly interested just in specific parts of the image segmentation techniques, as Pixel-based Segmentation
picture. Essentially it is characterized as the procedure of Methods, Edge-based Segmentation Methods and
isolating the picture into distinctive parts of homogeneity. An Region-based Segmentation Methods. In the paper the author
image might be defined as a two-dimensional function, f(x, y), has rate the image segmentation techniques surveyed on the
where x and y are the spatial coordinates, and the amplitude of basis of Good, Bad and Normal.
at any pair of coordinates is known as the intensity of the While Yogamangalam et al. 2013 [5] Segmentation
image at specific point. When the intensity of f and the value Technique Comparision in Image Processing has classified
of x, y are finite then the image is called digital image [1]. the same techniques into five categories viz., Region Based,
In this paper our main focus will be on Thresholding-based, Edge Based, Threshold, Feature Based Clustering, and Model
Clustering-based and the Edge-based Methods. Segmentation Based Techniques in [5]. A completely different structure of
methods can be classified into three main categories [2]: classification of segmentation techniques has been given in
Classical Methods that include Thresholding, region [5] and it provides experimental results on the Lena image. It
growing, clustering, and edge detection approaches, compares algorithms from different techniques and concludes
etc. that Thresholding is the simplest of all kind of segmentation
Artificial Intelligence Techniques that include techniques, it also states that the same technique is
Neural Network Based image segmentation computationally fastest.
techniques, like, Supervised-KNN, Unsupervised
Hopfield ANN, etc. III. PROPOSED WORK
Hybrid Techniques that either crossover or fall into In this paper, we will lay focus on three main types of
none of the first two categories. For example, Graph segmentation methods, namely, Thresholding based,
Cut. Clustering based and Edge based.
Under the thresholding-base segmentation method we have
picked the basic Global Thresholding (iterative), Otsus
II. LITERATURE REVIEW Method and Local Thresholding.
There exists a large body of research from the last two decade. In the Clustering-based segmentation method we have laid
A lot of surveys have been done, with various classifications focus on K-Means and Mean Shift clustering algorithms.
of image segmentation methods and the segmentation While in the Edge-based segmentation method, we touched
techniques themselves. the basics of the Canny, Sobel and Prewitt operators and their
Patil et al. 2013 [3], Overview of Colour Image use.
Segmentation Techniques classifies all the techniques into We have implemented these algorithms in MATLAB and
three major categories namely, Edge-based Segmentation produced results on three types of images, a Leukemia cell
Methods, Region-based Segmentation Methods and image, a scan of a paper image and a green outdoor image.
Applying the above mentioned techniques, we have taken the
outputs and presented them. We have measure the execution

42 www.erpublication.org
 Analysis of Image Segmentation Techniques: A Survey

time for each segmentation technique and also rated the

segmentation between good, bad and average.

IV. SEGMENTATION METHODS AND TECHNIQUES

4.1 Thresholding
Threshold technique is one of the most used techniques in
image segmentation. This technique can be expressed as [6]:
T=T[x , y, p(x, y), f(x, y)]
where:
T is the threshold value.
x, y are the coordinates of the threshold value point.
p(x, y) and f(x, y) are points in the gray level image pixels.

Thresholding in an image can be defined as:

Some of the thresholding techniques are:

4.1.1 Basic Global Thresholding
In basic global thresholding, a single pixel is used for all the Fig. 2: Otsus Thresholding flow chart.
image pixel. . When the pixel values of the components and
that of background are fairly consistent in their respective Algorithm:
values over the entire image, global thresholding could be It more or less works in the same way as the Global
used. The iterative global thresholding flow chart is presented Thresholding.
below:
4.1.3 Local Thresholding
Algorithm: Local thresholding is an image thresholding techniques that
Step 1: Select an initial estimate for T. segments an image on the basis of local threshold, i.e. it
Step 2: Segment the image using T. This will produce two compares threshold for every pixel in an image. The main idea
groups of pixels. G1 consisting of all pixels with grey level is that each pixel is compared to an average of the surrounding
values >T and G2 consisting of pixels with values <=T. pixels. Specifically, an approximate average of the last s s
Step 3: Compute the average grey level values mean1 and window of pixelscenteredaround each pixel is calculated
mean2 for the pixels in regions G1 and G2. while throughout the image which means, it considers
Step 4: Compute a new threshold value T=(1/2)(mean1 neighbouring pixels on all sides in the region. If the value of
+mean2) the current pixel is t percent lower than the average then it is
Step 5: Repeat steps 2 through 4 until difference in T in set to black, otherwise it is set to white [9]. The flow chart of
successive iterations is smaller than a predefined parameter local thresholding is given here:
T0, i.e., if |T-Tnew| > T, back to step 2, otherwise stop.[7]

Fig. 1 Global Thresholding flow chart.

Fig.3 Local thresholding flow chart.

4.1.2 Otsus Threshodling

Algorithm:
This method was proposed by N. Otsu in 1975 and has been in
procedure AdaptiveThreshold(in, out, w, h)
fashion ever since [8]. It is aimed at reducing the within-class
Step1: for i = 0 to w do
variance while maximizing the inter-class variance. The flow
Step2: sum 0
chart for Otsu;s thresholding is show here:
Step3: for j = 0 to h do

43 www.erpublication.org
 International Journal of Engineering and Technical Research (IJETR)
ISSN: 2321-0869 (O) 2454-4698 (P), Volume-4, Issue-1, January 2016
Step4: sum sum + in [i, j]
Step5: if i = 0 then 4.2.2 Mean Shift Clustering:
Step6: intImg[i, j] sum This is an advanced technique for clustering based-method.It
Step7: else starts by taking a mean for each pixel, using which a kernel
Step8: intImg[i, j] intImg[i1, j] +sum desnity estimator is built. This is shifted throughout the
Step9: end if image. Typical kernel density estimators are Gaussian. By
Step10: end for iteratively shifting the mean based on the kernel, all the pixels
Step11: end for get drawn to a number of local points of convergence.

4.2 Clustering-based Methods Algorithm:

Clustering method is a procedure in which a data set or say Step1: For each i = 1n run the mean shift procedure for xi
pixels are exchanged by cluster, pixels may belong together and store the convergence point in zi..
because of the same color, texture etc. There are two types of Step2: Identify clusters {Cp.} p=1m of convergence
clustering that is hierarchical clustering and partional points by linking together all zi which are closer than 0,5 from
clustering [10]. each other in the joint domain.
Some of the clustering technique under implemntation are: Step3: For each i = 1n assign Li = {p | zi Cp }.
4.2.1 K-Means Clustering:
This is an iterative technique used to partition an image into
lot of clusters in which there is option of k cluster centers;
randomly each pixel in the image is assign to the every cluster.
The cluster centers is again computed by taking mean of all
the pixels in the center. The quality of explanation obtained
from this method depends upon the initial set of clusters and
the value of k. The flow chart of k-Means Clustering is given
here:

Fig. 5: Mean Shift flow chart.

4.3 Edge-based Methods

The variation in the image feature usually brightness of image
gives rise to the edges. More objectively, the edges are the
representation of discontinuities of image intensity function.
There could be various reasons such as lightning condition,
bject geometry, type of material, surface texture, etc as well as
their mutual interaction and so on [11]. In edge based
segmentation, gradient is known as the first derivative of an
image f(a,b). The first derivative operator uses gradient
Fig. 4 K-Means Clustering flow chart.
method to find the edges by using the maximum and minimum
value of the gradient. The gradient is a measure of change in a
Algorithm:
function.
Step1: Input image leuk.jpg
The famous edge-based methods are:
Step2: Give the number of cluster value as k. Here we chose
4.3.1 Canny Edge Detection:
the cluster value as 3.
Canny edge operator is asecond order derivative operator for
Step3: Randomly choose the k cluster centers.
edge detection and is considered as superior edge detection
Step4: Calculate the mean or center of clusters.
operator among the available operators based on the
Step4: Calculate the distance between each pixel to each
experimental results as it determines strong and weak edges in
cluster center.
the image. Image is first smoothed by using circular
Step5: If the distance is near to the center then move to that
two-dimension Gaussian function, computing the gradient of
cluster
the result and then using the gradient magnitude and direction
Step6: Otherwise move to next cluster.
to approximate edge strength and direction at every point. The
Step7: Re-estimate the center.

44 www.erpublication.org
 Analysis of Image Segmentation Techniques: A Survey

gradient magnitude array so obtained consists of undesirable

ridges around local maxima and are to be suppressed to get
discrete orientations of the edge normal by the process of
nonmaxima suppression. Then the technique of double
thresholding is employed to reduce false fragments. Two
thresholds are used to solve the purpose T1 and T2 where T2
Fig. 8: An example of Prewitt Operator.
2T1 [11].
Algorithm:
Algorithm:
Step1: Convolve image f(r, c) with a Gaussian function to get
Prewitt Pretty much works in the same way as Sobel operator.
smooth image f ^ (r, c). f ^ (r, c)=f(r,c)*G(r,c,6)
Step2: Apply first difference gradient operator to compute
edge strength then edge magnitude and direction are obtain as
before.
Step3: Apply non-maximal or critical suppression to the
gradient magnitude.
Step4: Apply threshold to the non-maximal suppression
image.

Fig. 9: Prewitt Edge Detection flow chart

In the next section we will show the algorithms of these

image segmentation techniques.

Fig 6: Canny Edge Detection flow chart

V. RESULTS

4.3.2 Sobel edge Detector: The above mentioned algorithms were tested on various
This is widely used first derivative operator to find edges and images. In here we present the results on the medical image
is modification of Prewitts operator, as will be discussed leuk.jpg.
next. It changes the center coefficient by 2. The Sobel
operatiors are given as [12]:

Fig. 7: An example of Sobel Operator.

Algorithm [13]:
Step1: Apply mask Gx, Gy to the input image.
Step2: Apply Sobel edge detection algorithm and the
gradient.
Step3: Masks manipulation of Gx, Gy separately on the input
image.
Step4: Results combined to find the absolute magnitude of the
gradient.

Step5: the absolute magnitude is the output edges.

4.3.3 Prewitt Edge Detector:

This operator uses 3 X 3 mask to find the edges. The mask
used along x and y direction corresponding [12] they are:

45 www.erpublication.org
 International Journal of Engineering and Technical Research (IJETR)
ISSN: 2321-0869 (O) 2454-4698 (P), Volume-4, Issue-1, January 2016
The experimental results of the algorithms on the outdoor
image outdoor.jpg:

Fig. 10: From top-left to bottom: Original Leukemia Image;

Globally thresholded image; Otsus thresholded image;
Locally Thresholded Image; Segmented Image from
K-Means Clustering, with k=2; Segmented Image from Mean
Shift Clustering, with k=4 and bandwidth bw=0.3 ; Canny
Edge Detected Image; Sobel Edge Detected Image; Prewitt
Edge Detected image.

The experimental results of the algorithms on the paper image

paper.jpg:

Fig. 12: From top-left to bottom: Original Outdoor Image;

Globally Thresholded Image; Otsus Thresholded Image;
Locally Thresholded Image; K-Means Segmented Image,
with k=9; Mean Shift segmented image, with k=8 and
bandwidth bw= 0.7 Canny Edge Detected Image; Sobel Edge
Detected Image; Prewitt Edge Detected Image.

Fig. 13: Graph of various segmentation techniques time.

Fig. 11: From top-left to bottom: Original Paper Image;

Globally Thresholded Image; Otsus Thresholded Image;
Locally Thresholded Image; K-Means Segmented Image,
with k=9; Mean Shift segmented image, with k=4 and
bandwidth bw= 0.3; Canny Edge Detected Image; Sobel Edge
Detected Image; Prewitt Edge Detected Image. Table 1: Execution Times of various segmentation techniques
on each type of test image.

46 www.erpublication.org
 Analysis of Image Segmentation Techniques: A Survey

VI. CONCLUSION [8] Otsu, Nobuyuki. "A threshold selection method from gray-level
histograms."Automatica 11.285-296 (1975): 23-27.
On the basis of our experimental observations, we were able [9] Bradley, Derek, and Gerhard Roth. "Adaptive thresholding using the
to come to the following conclusions: integral image." Journal of graphics, gpu, and game tools 12.2
i. Otsus Thresholding performs closely well enough to (2007): 13-21.
[10] Navneet Kaur, Gagan Jindal "A Survey Of K Means Clustering With
Canny Edge Detection Technique on the Paper Image. Modified Gradient Magnitude Region Growing Technique For Lesion
ii. Both Mean Shift and Local Thresholding perform really Segmentation, International Journal Of Innovations In Engineering
closely on the outdoor image. And Technology, April 2013.
iii. Mean Shift clustering outperforms its counterparts under [11] Otsu N, Discriminant and Least Square Threshold
Selection. In: Proc 4IJCPR, 1978, pp.592596.
the survey and particularly on the Medical Image and [12] A. M. Khan, Ravi. S Image Segmentation Methods: A Comparative
Outdoor Image, but lags behind on the mostly black Study, International Journal of Soft Computing and Engineering,
and white Paper Image Volume-3, Issue-4, September 2013.
iv. K-Means is computationally exhaustive, though [13] Vincent, O. R., and Olusegun Folorunso. "A descriptive algorithm
for sobel image edge detection." Proceedings of Informing Science
provides suitable results given the correct number of & IT Education Conference (InSITE). Vol. 40. 2009.
clusters (though not ideal).
v. Edge Detection techniques cant be used independently,
though they give useful information that can be used in Apurv Vashisht, M.Tech Research Scholar, Department of Computer
Science & Engineering, Mewar University, Chittorgarh, Rajasthan
some other step of image segmentation along with
India.
some other technique. Mr. Shiv Kumar, Assistant Professor,Department of Computer Science &
Thresholding is computationally inexhaustive, but does not Engineering, Mewar University, Chittorgarh, Rajasthan., India.
take the spatial features of an image into account and even
though it can be used independently it is more suitable as a
step in some image processing application.
While clustering techniques work quite well, they are
susceptible to noise and are highly dependent on the input
parameters.
The second order edge operators (like Canny) give quite
reliable results, while the first order edge operators give
useful information which can be used as a step in some other
edge detection technique. Also, it is quite clear that no single
operator can be applied to all images and sometimes the edges
obtained are not continuous.
Thus, it seems that as sound as each method may sounds
theoretically, they have their limitations. The biggest one
being, that no method can be applied to all kinds of images,
universally.

VII. FUTURE WORK

A survey of new (advanced) image segmentation techniques
can be done, while many methods like Artificial or Hybrid
that could not be included in this survey can be worked upon.

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, pp. 713727, Sep. 2001
[2] Rajesh Dass, Priyanka, Swapna Devi, Image Segmentation
Techniques, IJECT Vol.3, Issue 1, Jan-March 2012
[3] Smita Patel, Prof.Mrs. A. A. Junnarkar. Overview of Colour Image
Segmentation Techniques International Journal of Advanced
Research in Computer Science and Software Engineering 3.09 (2013):
418-423
[4] Basavaprasad, B., and M. Ravi. "A comparative study on
classification of image segmentation methods with a focus on graph
based techniques."International Journal of Research in Engineering
and Technology 3.03 (2014): 310-314.
[5] Yogamangalam, B. Karthikeyan R., and B. Karthikeyan.
"Segmentation Technique Comparision in Image Processing." IJET
(ISSN: 0975-4024) 5.1 (2013): 307-313.
[6] Bins, L. Santanna, Leila M. Garcia Fonseca, Guaraci Jos Erthal, and
F. Mitsuo Ii. Satellite imagery segmentation: a region growing
approach Simpsio Brasileiro de Sensoriamento Remoto, pp. 677-
680, April 1996.
[7] Raju, P. Daniel Ratna, and G. Neelima. "Image segmentation by using
histogram tresholding." International Journal of Computer Science
Engineering and Technology 2 (2012): 776-779.

47 www.erpublication.org