A modified pso based graph cut algorithm for the selection of optimal regularizing

International Journal of Advanced Research in Engineering and Technology (IJARET), ISSN
0976 – 6480(Print), ISSN 0976 – 6499(Online) Volume 4, Issue 3, April (2013), © IAEME
273
A MODIFIED PSO BASED GRAPH CUT ALGORITHM FOR THE
SELECTION OF OPTIMAL REGULARIZING PARAMETER IN
IMAGE SEGMENTATION
Shameem Akthar1
, Dr. D Rajaylakshmi2
, Dr. Syed Abdul Sattar3
1
(Computer Science & Engg., KBNCE, Karnataka, India)
2
(IT Department, JNTU University, Andhra Pradesh, India)
3
(EC Dept., Royal Institute of Tech., Andhra Pradesh, India)
ABSTRACT
Image segmentation is an important stage from the image processing to image
analysis. According to the segmentation of image only, the target expression of original
image will be transformed into more abstract and compact manner, which will lead to high-
level analysis of the image and the understandability of image. In our paper, a modified PSO
based Graph cut algorithm is proposed for the selection of the optimal regularizing parameter
in the image. The raw image is pre-processed using filtering and applied to graph-cut after
regularizing the parameters using modified PSO. A proper selection of the parameter remains
a critical problem for practical image segmentation. Based on this optimized parameter value
the important region and the boundary are detected in the given input image. The proposed
method is implemented in MATLAB with various images.
Keywords: Gaussian Filter, Graph cut Algorithm, Min-cut / Max-Flow Algorithm, Particle
Swarm Optimization, Segmentation.
1. INTRODUCTION
Image segmentation is a complex and challenging task due to the intrinsically
imprecise nature of the images [1]. In areas such as computer vision and Image Processing,
image segmentation has been and still is a relevant research area due to its wide spread usage
and application [7]. The segmentation of the target areas is an important aspect in image
segmentation [5]. Generally, segmentation is a first step for a variety of image analysis and
visualization tasks. The steps or processes after segmentation rely on the segmentation
INTERNATIONAL JOURNAL OF ADVANCED RESEARCH IN
ENGINEERING AND TECHNOLOGY (IJARET)
ISSN 0976 - 6480 (Print)
ISSN 0976 - 6499 (Online)
Volume 4, Issue 3, April 2013, pp. 273-279
© IAEME: www.iaeme.com/ijaret.asp
Journal Impact Factor (2013): 5.8376 (Calculated by GISI)
www.jifactor.com
IJARET
© I A E M E

274
quality [4]. Segmentation is a process of partitioning an image space into some non-
overlapping meaningful homogeneous regions. In general, these regions will have a strong
correlation with the objects in the image. The success of an image analysis system depends
on the quality of segmentation [1]. Segmentation is one of the popular methods used to detect
flaws in weldmesh. Generally the flaws that occur are wormholes, inclusion, lack of fusion,
porosity, incomplete penetrations, slag line and cracks [8].
Image segmentation is used to locate and find objects and boundaries(lines,curves
etc.) of image. To perform it there are many ways[11].
Region growing algorithms have been used mostly in the analysis of grayscale images
[3]. The general procedure is to compare a specific feature of one pixel to its neighbor(s)
feature. If a criterion of homogeneity is satisfied, the pixel is classified to the same class as
one or more of its neighbors. The choice of the homogeneity criterion is critical for even
moderate success and in all instances the results are upset by noise [2] [6]. Moreover,
computational cost of segmentation algorithms increases while algorithmic robustness
tends to decrease with increasing feature space sparseness and solution space
complexity [4].
2. PROPOSED METHODOLOGY
Our proposed system comprises of three phases.
1) Pre-processing
2) Optimal parameter selection using PSO
3) Segmentation using Graph Cut algorithm
2.1 Pre-Processing
Initially pre-processing is done through Gaussian filter, we apply a stepper Gaussian
field with less deviation value to remove the unwanted portions in the image such as noise,
blur, reflections. A 5 x 5 Gaussian Filter is used with 4.1=σ .
2.2 Optimal Parameter Selection Using Modified Pso
The regularizing parameters are selected and applied for Graph cut Algorithm. The
parameters that are given as the input for the graph cut are smallest size of area and smallest
threshold cut value.
2.2.1 Modified Particle Swarm Optimization (PSO)
Particle swarm optimization (PSO) is a population-based optimization algorithm
modeled after the simulation of social behavior of birds in a flock. The algorithm of
PSO is initialized with a group of random particles and then searches for optima by
updating generations. Each of the particles are flown through the search space having its
position adjusted based on its distance from its own personal best position and the distance
from the best particle of the swarm[10]. The performance of each particle, i.e. how
close the particles is from the global optimum, is measured using a fitness function which
depends on the optimization problem. There are two position pbest and gbest . Also two
best values pbest value and gbest values [9].
In our modified PSO, in addition with the pbest value, the personal worst location
of the particle is denoted as pworst is used. The previously seen worst locations of the

275
particle are represented by this pworst value. While updating the velocity, pworst value is
also taken into consideration along with the difference between the personal best position of
the particle and the current location of the particle. By including pworst value, the particle
can detour its previous worst location and try to select the better position.
2.2.1.1. Modified PSO Algorithm Steps
Step 1: Initialize a population of i particles with each particle’s position ix and velocity iv
on a problem space n
R of dimension n .
Step 2: Compute the fitness function for each particle i in d variables.
Step 3: Make comparison between the particle’s fitness value, fitnessx and particle’s pbest
fitness value, fitnessp . If the current fitness value of particle is better than the particle’s
pbest fitness value, then set the pbest value into current position in the d th
dimension.
Step 4: Check out all of the particle’s pbest fitness value, fitnessp with value of gbest . If
the current value, pbest is better than the gbest value means, then set the gbest value into
current particle’s array index and value.
Step 5: Update the velocity and position of the particles given as in equations (1) and (2).
)()(
)(
3221
11
idididididb
idididaidid
xgbestrpworstpworstxr
pbestxpbestrvv
−××+×−××
+×−××+×=
ϕϕ
ϕω
(1)
ididid vxx += (2)
Step 6: Repeat step 2, until a better fitness or maximum number of iterations are met.
Process of Merging: After getting the values of gbest , the Merging Process is used to merge
the regions. The input for the Merging Process is all these obtained gbest values. At each and
every step, the adjusted regions are merged one by one, with the gbest values. As a result,
the boundary of the objects is obtained from the images which are not smooth. In order to
obtain the refined boundary, boundary refinement technique is used.
Refinement of Boundaries: If an image pixel presents on the boundary of at least two
distinct regions means, then a discrete disk with the radius 3 will be placed on it. For the
refinement of jagged boundaries, the similarity between these two regions is evaluated
individually.
2.3. Segmentation Using Minimum-Cut/Maximum-Flow Graph Cut Algorithm
A graph is represented as },,{ WEVG = , where V represents finite set of nodes
(vertices); E is a set of unordered pairs edges from V ; and W denotes the affinity matrix that
associates a weight to each edge in E . There are two special nodes called as terminals as
source, Vs∈ and sink, Vt∈ . Other nodes are non-terminal nodes. There are two types of
edges linkt − and linkn − . linkn − connect the non-terminal nodes and linkt − connects a
non-terminal node with a terminal node. linkn − indicates a neighborhood system in the
image. Each of the edges E is associated with weight or cost W .

276
The Minimum-Cut of a graph G is the cut that partitions the graph into disjoint
subsets, such that the sum of the weights W associated with the edges E of the graph G
between the different subsets X andY is minimized. In graph theoretic language, the cut is
represented as,
∑
∈∈
=
YvXu
vuWYXCut
,
),(),( (3)
The parameters of energy in our work is smallest size of area and smallest threshold cut
value, the edge weights are appropriately set. If the weights of edge are set, then a minimum
cut will correspond to a labeling with the minimum value of this energy.
3 RESULTS AND DISCUSSIONS
Our proposed work is implemented in MATLAB platform. The images are collected
from various databases and given for the implementation. The results for the segmented
images are shown in figure 1, 2 and 3.
Image
s
Original image
Segmented non-tumor
region
Segmented tumor
region
1
2
Fig.1: Segmented output for normal MRI brain images using our proposed method

277
Image
s
Original image Segmented WM region Segmented GM region
1
Fig. 2 : Segmented output for abnormal brain MRI images using our proposed method
Original
image
Segmented
Flower
region
Segmented
Leaf
Region
Fig. 3: Segmented output for Flower images using our proposed method
In our implementation work, some of the samples are given in our discussion. In
figure 1,2,3 first column shows original images, the second and third column shows the
segmented non-tumor & tumor region for figure 1, WM & GM region for figure 2 and
flower and leaf regions, for figure 3 respectively. From the columns 2 and 3 in figures 1, 2
and 3, it is noted that our proposed method effectively segment the given input images.

278
Performance Evaluation
To evaluate our work, Jaccard Similarity evaluation, Dice Co-efficient evaluation and
Accuracy are considered as measure of metric.
Jaccard similarity
(JS)
Dice coefficient (DS) Accuracy
ao
ao
RR
RR
JS
∪
∩
=
FNFPTP
TP
RR
RR
DC
ao
ao
++×
×
=
+
∩
=
)2(
22
FNFPTNTP
TNTP
Acc
+++
+
=
Table 1: JC,DC and Accuracy equations
Using table 1 equations of JS, DS and Accuracy the performance for our proposed method is
evaluated. In these equations, the values are True Positive (TP), True Negative (TN), False
Positive (FP), and False Negative (FN) for tumor part correctly, non-tumor part correctly,
none—tumor part incorrectly, and tumor part incorrectly for figure 1.
Likewise, we have taken all these values according to the segmentation regions for figure 2
and figure 3. In table 2, the JS and DC values are tabulated for the conventional method and
for the proposed method.
Images Proposed method Existing method
JS DC JS DC
1 0.919612 0.958123 0.930397 0.963944
2 0.853747 0.921104 0.76172 0.864746
3 0.847663 0.917551 0.747444 0.855471
4 0.726466 0.841564 0.691094 0.822275
Table 2: JC and DC values for existing and proposed method
Table 3, gives the accuracy measure for the proposed PSO and Graph Cut algorithm and
existing conventional.
Images Proposed method Conventional PSO
Conventional Graph
cut
1 95.3 92.13 93.2
2 95.12 92 93.4
3 95.2 92.61 93
4 94.99 92.3 93.55
Table 3: Accuracy measure comparison for both Proposed and conventional methods (in %)
4 CONCLUSION
In this paper, a modified PSO based graph cut segmentation, was presented. We have
used smallest size area and smallest threshold cut value as regularizing parameter. Along with
this, we found the worst position of the particle so that the particle can move away from that
position to the best position, which reduces the time taken for convergence of the search
space, with better achievement of an optimal solution when compared with the conventional
PSO and Graph Cut approaches. Thus our modified PSO with Graph Cut provides better
accuracy value than the conventional method with 95.1525% .

279
REFERENCES
[1] Sriparna Saha, and Sanghamitra Bandyopadhyay, "MRI Brain Image Segmentation by
Fuzzy Symmetry Based Genetic Clustering Technique", In Proceedings of IEEE Conference
on Evolutionary Computation, pp. 4417-4424, September, 2007.
[7] P.Raja Sekhar Reddy, N.Naga Lakshmi, and Thandu Ashalatha, "Image Segmentation
Using A Region Growing Method – A Comparative Study", International Journal of
Engineering and Social Studies, Vol. 2, No. 7, pp. 62-69, 2012.
[5] Dongwei Guo, Bo Zhang, and Yunna Wu, Lisai Cao, "The Man-made Areas
Segmentation Based on Region Growing Method Using Local Fractal Dimension", Journal of
Information and Computational Science, Vol. 10, No. 3, pp. 659–667, 2013.
[4] Nazahah Mustafa, Nor Ashidi Mat Isa and Mohd Yusoff Mashor, "Automated Multicells
Segmentation of ThinPrep Image Using Modified Seed Based Region Growing Algorithm",
Biomedical Soft Computing and Human Sciences, Vol.14, No.2, pp.41-47, 2009.
[8] B.Karthikeyan, V.Vaithiyanathan, B.Venkatraman, and M.Menaka, "Analysis of Image
Segmentation for Radiographic Images", Vol. 5, No. 11, pp. 3660-3664, November 2012.
[2] M. M. Abdelsamea, "An Automatic Seeded Region Growing for 2D Biomedical Image
Segmentation", In Proceedings of International Conference on Environment and Bio-Science,
2011.
[3] Yongming Li, Dongming Lu, Xiqun Lu, and Jianming Liu, "Interactive Color Image
Segmentation by Region Growing Combined with Image Enhancement Based on Bezier
Model", In Proceedings of IEEE Conference on Multi-Agent Security and Survivability, pp.
96-99, 2004.
[4] J.Leela Mahendra Kumar, A.Vasavi , and B.Praveena, "Semantic Region Growing For
Multivariate Image Segmentation Using Adaptiver Edge Penality", International Journal of
Engineering Research & Technology, Vol.1, No. 4, June 2012.
[6] Rolf Adams and Leanne Bischof, “Seeded Region Growing”, IEEE Transactions on
Pattern Analysis and Machine Intelligence, Vol. 16, No. 6, June 1994.
[9] Fahd M. A. Mohsen, Mohiy M. Hadhoud, and Khalid Amin, "A new Optimization-Based
Image Segmentation method By Particle Swarm Optimization", International Journal of
Advanced Computer Science and Applications, Vol. 7, No. 4, pp. 10-18, 2010.
[10] Shameem Akthar, Dr.D.Rajaylakshmi, Dr.Syed Abdul Sattar, “Contour Detection Using
PSO and Graph Cut”,IJARCSEE Volume 1, Issue 1.
[11] Shameem Akthar, Dr.D. Rajaylakshmi, Dr.Syed Abdul Sattar, “A theoretical survey for
edge detection technique and watershed transformation”, IJCTEE Volume. 2, Issue 1.
[12] Gaganpreet Kaur and Dr. Dheerendra Singh, “Pollination Based Optimization for Color
Image Segmentation”, International journal of Computer Engineering & Technology
(IJCET), Volume 3, Issue 2, 2012, pp. 407 - 414, ISSN Print: 0976 – 6367, ISSN Online:
0976 – 6375.
[13] Ankit Vidyarthi and Ankita Kansal, “A Survey Report on Digital Images Segmentation
Algorithms”, International journal of Computer Engineering & Technology (IJCET),
Volume 3, Issue 2, 2012, pp. 85 - 91, ISSN Print: 0976 – 6367, ISSN Online: 0976 – 6375.

A modified pso based graph cut algorithm for the selection of optimal regularizing

More Related Content

What's hot (18)

Similar to A modified pso based graph cut algorithm for the selection of optimal regularizing (20)

More from IAEME Publication (20)

Recently uploaded (20)

A modified pso based graph cut algorithm for the selection of optimal regularizing