B49010511

A Singh Bhandari et al Int. Journal of Engineering Research and Applications www.ijera.com
ISSN : 2248-9622, Vol. 4, Issue 9( Version 1), September 2014, pp.05-11
www.ijera.com 5 | P a g e
SAR Image Segmentation Based On Hybrid PSOGSA Optimization Algorithm Amandeep Kaur*, Charanjit Singh**, Amandeep Singh Bhandari*** *(Student, Department of Electronics and Communication Engg., Punjabi University, Patiala) ** (Assistant Professor, Department of Electronics and Communication Engg., Punjabi University, Patiala) *** (Assistant Professor, Department of Electronics and Communication Engg., Punjabi University, Patiala) ABSTRACT Image segmentation is useful in many applications. It can identify the regions of interest in a scene or annotate the data. It categorizes the existing segmentation algorithm into region-based segmentation, data clustering, and edge-base segmentation. Region-based segmentation includes the seeded and unseeded region growing algorithms, the JSEG, and the fast scanning algorithm. Due to the presence of speckle noise, segmentation of Synthetic Aperture Radar (SAR) images is still a challenging problem. We proposed a fast SAR image segmentation method based on Particle Swarm Optimization-Gravitational Search Algorithm (PSO-GSA). In this method, threshold estimation is regarded as a search procedure that examinations for an appropriate value in a continuous grayscale interval. Hence, PSO-GSA algorithm is familiarized to search for the optimal threshold. Experimental results indicate that our method is superior to GA based, AFS based and ABC based methods in terms of segmentation accuracy, segmentation time, and Thresholding.
I. INTRODUCTION
An image is an artifact that depicts visual perception similar in appearance to some physical object or a person. Images may be two-dimensional (2D), such as screen display, a photograph, etc., or three- dimensional (3D), such as a hologram or a statue. A digital image is a numeric representation of a two- dimensional image. An image can be defined as a two-dimensional function 푓 푥,푦 , where 푥 and 푦 are three-dimensional coordinates, and the amplitude off at any pair of coordinates 푥,푦 is called the intensity of the digital image at that point.
Image processing is a method of converting an image into digital form and performs some operations on it, to get an enhanced image or to extract some useful information from it. It is a form of signal processing in which the input is an image, like photograph frame, and the output may be either an image or a set of characteristics or parameters associated with the image.
Segmentation is a process that divides an image into its basic parts or objects. In general, independent segmentation is one of the toughest tasks in digital image processing. Image segmentation is useful in many applications. It can identify the regions of interest in a scene or annotate the data. It categorizes the existing segmentation algorithm into region-based segmentation, data clustering, and edge-base segmentation. Region-based segmentation includes the seeded and unseeded region growing algorithms, the JSEG, and the fast scanning algorithm.
RESEARCH ARTICLE OPEN ACCESS

Figure 1: Fundamental steps in digital image processing.
Particle Swarm Optimization (PSO)
Particle Swarm Optimization (PSO) is one of the more recently developed evolutionary techniques, and it is based on a suitable model of social interaction between independent agents (particles) and it uses social knowledge in order to find the global maximum or minimum of a generic function. In the PSO the so called swarm intelligence (i.e. the experience accumulated during the evolution) is used to search the parameter space by controlling the trajectories of a set of particles according to a swarm- like set of rules. The position of each particle is used to compute the value of the function to be optimized. Particles are moved in the domain of the problem with variable speeds and every position they reach represents a particular configuration of the variables set, which is then evaluated in order to get a score [19].
Gravitational Search Algorithm (GSA)
In this section, we introduce our optimization algorithm based on the law of gravity [19]. In the proposed algorithm, agents are considered as objects and their performance is measured by their masses. All these objects attract each other by the gravity force, and this force causes a global movement of all objects towards the objects with heavier masses. Hence, masses cooperate using a direct form of communication, through gravitational force. The heavy masses which correspond to good solutions – move more slowly than lighter ones, this guarantees the exploitation step of the algorithm.
In GSA, each mass (agent) has four specifications: position, inertial mass, active gravitational mass, and passive gravitational mass. The position of the mass corresponds to a solution of the problem, and its gravitational and inertial masses are determined using a fitness function. In other words, each mass presents a solution, and the algorithm is navigated by properly adjusting the gravitational and inertia masses.
SAR Images
Ecological monitoring, earth-resource mapping, and military schemes require broad-area imaging at high resolutions. Many times the imagery must be developed in inclement weather or during night as well as day. Synthetic Aperture Radar (SAR) delivers such a capability. SAR systems take benefit of the long-range propagation characteristics of radar signals and the complex material processing capability of modern digital electronics to provide high resolution imagery. Synthetic aperture radar complements photographic and other optical imaging competences because of the minimum constraints on time-of-day and atmospheric circumstances and because of the unique responses of terrain and cultural targets to radar frequencies [20]. Synthetic aperture radar skill has provided terrain structural information to geologists for mineral examination, oil spill boundaries on water to environmentalists, sea state and ice hazard maps to navigators, and reconnaissance and targeting info to military processes.
II. LITERATURE SURVEY:
S. C. Zhu et al (1996), [12] presented a novel statistical and variational approach to image segmentation based on a new algorithm named region
Knowledge Base
Image restoration
Image enhancement
Image acquisition
Segmentation
Representation
& description
Object recognition
Color Image Processing
Wavelets and multi- resolution Processing
Morphological processing
Compression

competition. This algorithm was derived by minimizing a generalized Bayes/MDL criterion using the variational principle. They provided theoretical analysis of region competition including accuracy of boundary location, criteria for initial conditions, and the relationship to edge detection using filters. A. Tsai et al (2001), [15] presented first address the problem of simultaneous image segmentation and smoothing by approaching the Mumford–Shah paradigm from a curve evolution perspective. Various implementations of this algorithm are introduced to increase its speed of convergence. This more general model leads us to a novel PDE-based approach for simultaneous image magnification, segmentation, and smoothing, thereby extending the traditional applications of the Mumford–Shah functional which only considers simultaneous segmentation and smoothing. W. Tao et al (2007), [18] developed a novel approach that provides effective and robust segmentation of color images. By incorporating the advantages of the mean shift (MS) segmentation and the normalized cut (Ncut) partitioning methods, the proposed method requires low computational complexity and is therefore very feasible for real- time image segmentation processing. Because the number of the segmented regions is much smaller than that of the image pixels, the proposed method allows a low-dimensional image clustering with significant reduction of the complexity compared to conventional graph-partitioning methods that are directly applied to the image pixels. The superiority of that method was examined and demonstrated through a large number of experiments using color natural scene images. S. Mirjalili et al (2010), [19] proposed a hybrid population-based algorithm (PSOGSA) is with the combination of Particle Swarm Optimization (PSO) and Gravitational Search Algorithm (GSA). The main idea is to integrate the ability of exploitation in PSO with the ability of exploration in GSA to synthesize both algorithms strength. The results show the hybrid algorithm possesses a better capability to escape from local optimums with faster convergence than the standard PSO and GSA.
M. Miao et.al. (2011), [20] proposed a fast SAR image segmentation method based on Artificial Bee Colony (ABC) algorithm using threshold estimation as a search procedure that searches for an appropriate value in a continuous grayscale interval.ABC algorithm is used to search for the optimal threshold. Experimental results indicate that the proposed method is superior to Genetic Algorithm (GA) based and Artificial Fish Swarm (AFS) based segmentation methods in terms of segmentation accuracy and segmentation time.
C. Liu et al(2013), [22] presented a novel variational framework for multiphase synthetic aperture radar (SAR)image segmentation based on the fuzzy region competition method. A new energy functional is proposed to integrate the Gamma model and the edge detector based on the ratio of exponentially weighted averages (ROEWA) operator within the optimization process.
III. PROPOSED WORK
In the proposed work, Segmentation subdivides an image into its constituent regions or objects. The level to which the subdivision is carried depends on the problem being solved. That is, segmentation should stop when the objects of interest in an application have been isolated. To compare the efficiency of our method with others, segmentation methods based on PSO-GSA algorithm are used to segment some typical images, covering a noise-free optical image, an optical image polluted by synthetic noise (composed of Gaussian noise with mean 0 and variance 0.01, speckle noise with variance 0.005, and salt and pepper noise with density 0.02), and a real SAR image. In this experiment, for PSO-GSA algorithm, the population size is 10, the fixed maximum number of iterations is 10, and the limit times for abandonment is 10, the lower and upper bounds are 0 and 255 respectively, because in grey scale each pixel value is between 0 to 255, there are 265 levels in grey scale images so we take the bounds between 0 to 255
PSO Algorithm
[Step 1] Initialization.
Initialize positions and associate velocity to all particles (potential solutions) in the population randomly in the D-dimension space.
[Step 2] Evolution:
Evaluate the fitness value of all particles.
[Step 3] Compute the personal best:
Compare the personal best (푝푏푒푠푡) of every particle with its current fitness value. If the current fitness value is better, then assign the current fitness value to 푝푏푒푠푡and assign the current coordinates to 푝푏푒푠푡coordinates.
[Step 4] Determine the current best fitness value:
Determine the current best fitness value in the whole population and its coordinates. If the current best fitness value is better than global best (푔푏푒푠푡), then assign the current best fitness value to 푔푏푒푠푡and assign the current coordinates to 푔푏푒푠푡coordinates.

[Step 5] Calculate accelerations of the agents:
Update velocity (푉푖푑 푡) and position (푋푖푑 푡) of the d- th dimension of the i-th particle using the following equations: 푉푖푑 푡=휔 푡 ∗푉푖푑 푡−1+푐1 푡 ×푟푎푛푑1푖푑 푡×(푝푏푒푠푡푖푑 푡−1−푋푖푑 푡−1+푐2 푡 × 1−푟푎푛푑1푖푑 푡 ×(푔푏푒푠푡푑푡 −1−푋푖푑 푡−1) 푉푖푑 푡>푉푚푎푥 푑표푟푉푖푑 푡<푉푚푖푛 푑,푡푕푒푛푉푖푑 푡=푈(푉푚푖푛 푑,푈푚푎푥 푑) 푋푖푑 푡=푟푎푛푑2푖푑 푡×푋푖푑 푡−1+ 1−푟푎푛푑2푖푑 푡 ×푉푖푑 푡 푐1 푡 , 푐2 푡 = time-varying acceleration coefficients with 푐1 푡 decreasing linearly from 2.5 to 0.5 and 푐2 푡 increasing linearly from 0.5 to 2.5 over the full range of the search, and w(t) = time-varying inertia weight changing randomly between U(0:4 ; 0:9) with iterations, 푟푎푛푑1, 푟푎푛푑2, are uniform random numbers between 0 and 1, having different values in different dimension, t is the current generation number. The above equation has been introduced to clamp the velocity along each dimension to uniformly distributed random value between 푉푚푖푛 푑and 푉푚푎푥 푑if they try to cross the desired domain of interest. These clipping techniques are sometimes necessary to prevent particles from explosion. The maximum velocity is set to the upper limit of the dynamic range of the search (푉푚푎푥 푑=푋푚푎푥 푑) and the minimum velocity (푉푚푖푛 푑) is set to (푋푚푖푛 푑). However, position-clipping technique is avoided in modified PSO algorithm. Moreover, the fitness function evaluations of errant particles (positions outside the domain of interest) are skipped to improve the speed of the algorithm.
[Step 6] Repeat from Steps 2 to 5 until iterations reaches their maximum limit:
Repeat Steps 2 to 5 until a stop criterion is satisfied or a pre-specified number of iteration is completed, usually when there is no further update of best fitness value.
GSA Algorithm
[Step 1] Initialize of the agents.
Initialize the positions of the 푁 number of agents randomly within the given search interval as below: 푋푖=푥푖 1,…..푥푖 푑,……푥푖 푛 푓표푟 푖=1,2,…….,푁 Where,푥푖 푑 represents the positions of the i-th agent in the d-th dimension and 푁 is the space dimension.
[Step 2] Fitness evolution and best fitness computation for each agents:
Perform the fitness evolution for all agents at each iteration and also compute the best and worst fitness at each iteration defined as below (for minimization problems): 푏푒푠푡 푡 =푚푖푛 푗휖 1,…,푁 푓푖푡푗(푡) 푤표푟푠푡 푡 =푚푎푥 푗휖 1,..,푁 푓푖푡푗(푡) Where, 푓푖푡푗(푡) represents the fitness of the j-th agent at 푖푡푒푟푎푡푖표푛 푡,푏푒푠푡(푡) 푎푛푑 푤표푟푠푡(푡) represents the best and worst fitness at generation 푡.
[Step 3] Compute gravitational constant G:
Compute gravitational constant G at iteration t using the following equation: 퐺 푡 =퐺표 (−훼푡/푇) In this problem, 퐺0 is set to 100, 훼 푖s set to 20 and 푇 is the total number of iterations.
[Step 4] Calculate the mass of the agents:
Calculate gravitational and inertia masses for each agents at iteration 푡 by the following equations: 푀푎푖=푀푝푖=푀푖푖=푀푖, 푖=1,2,…,푁 푚푖 푡 = 푓푖푡푖 푡 −푤표푟푠푡푖 푡 푏푒푠푡 푡 −푤표푟푠푡(푡) 푀푖 푡 = 푚푖 푡 푚푗(푡)푁푗 =1 Where, 푀푎푖 is the active gravitational mass of the i-th agent,푀푝푖 is the passive gravitational mass of the i-th agent,푀푖푖 is the inertia mass of the i-th agent.
[Step 5] Calculate accelerations of the agents:
Compute the acceleration of the i-th agents at iteration t as below: 푎푖 푑 푡 = 퐹푖 푑 푡 푀푖푖(푡) Where, 퐹푖 푑(푡) 푖s the total force acting on i-th agent calculated as, 퐹푖 푑 푡 = 푟푎푛푑푗퐹푖푗푑 (푡) 푗휖퐾푏푒푠푡,푗≠푖 퐾푏푒푠푡 is the set of first 퐾 agents with the best fitness value and biggest mass.퐾푏푒푠푡 is computed in such a manner that it decreases linearly with time and at last iteration the value of 퐾푏푒푠푡 becomes 2% of the initial number of agents.퐹푖푗푑 (푡) is the force acting on agent 'i' from agent 'j' at d-th dimension and t-th iteration is computed as below: 퐹푖푗푑 푡 = 퐺 푡 푀푝푖(푡)×푀푎푗(푡) 푅푖푗 푡 +휀 푥푗 푑 푡 −푥푖 푑(푡) Where,푅푖푗 푡 is the Euclidian distance between two agents 'i' and 'j' at iteration t and 퐺(푡) is the computed gravitational constant at the same iteration. 휀 is a small constant
[Step 6] Update velocity and positions of the agents:
Compute velocity and the position of the agents at next iteration (t + 1) using the following equations:

푣푖
푑 푡 + 1 = 푟푎푛푑푖 × 푣푖
푑 푡 + 푎푖
푑 (푡)
푥푖
푑 푡 = 푥푖
푑 푡 + 푣푖
푑 푡 + 1
[Step 7] Repeat from Steps 2 to 6 until iterations
reaches their maximum limit.
Return the best fitness computed at final iteration
as a global fitness of the problem and the positions of
the corresponding agent at specified dimensions as
the global solution of that problem.
IV. SIMULATION RESULTS
Figure 2: Original Image Figure 3 : Segmented Image
We take a general purpose image shown in figure 2
for comparison to base paper images and calculate
the threshold value (167), Time (0.119), and fitness
of an image, and compare the results with the
previous work taken in the Table below show the
comparison of ant colony, Genetic, Artificial Fish
Swarm algorithms. Image Fitness is 1.0 × 103
*2.5775.
Table 1: Comparison over different Algorithms using image (figure 2)
Algorithms Fitness Function Threshold Time (s)
PSO-GSA Variance Grey entropy 167 0.119
ACS Improved two-dimensional grey entropy 205 5.821
GA Two-dimensional entropy 207 14.391
GA Two-dimensional grey entropy 206 17.640
AFS Two-dimensional entropy 187 12.015
Figure 4: Original Image Figure 5: Segmented Image
We take a general purpose image shown in
figure 4 for comparison to base paper images and
calculate the threshold value (165), Time (0.123), and
fitness of an image, and compare the results with the

previous work taken in the Table below show the comparison of ant colony, Genetic, Artificial Fish Swarm algorithms. Image Fitness is 1.0×1003 *2.4103. Table 2: Comparison over different Algorithms using image (figure 4)
Algorithms
Fitness Function
Threshold
Time (s)
PSO-GSA
Variance Grey entropy
165
0.123
ACS
Improved two-dimensional grey entropy
204
6.669
GA
Two-dimensional entropy
163
15.358
GA
Two-dimensional grey entropy
207
19.152
AFS
162
12.546
Figure 6:Original Image# Case 3
Figure 7: Segmented Image# Case 3
We take an SAR image 5.4 and calculate the threshold value (61), Time (0.115), and fitness of an image, and compare the results with the previous work taken in the Table below show the comparison of ant colony, Genetic, Artificial Fish Swarm algorithms. Image Fitness is 883.960. Table 3: Comparison over different Algorithms using image (figure 6)
Algorithms
Fitness Function
Threshold
Time (s)
PSO-GSA
Variance Grey entropy
61
0.115
ACS
Improved two-dimensional grey entropy
95
4.835
GA
131
13.460
GA
Two-dimensional grey entropy
94
17.740
AFS
62
6.441
V. CONCLUSION
We proposed a fast segmentation method on SAR images. The technique regards threshold estimation as a search process and employs PSO- GSA algorithm to optimize it. In order to provide PSO-GSA algorithm with an efficient fitness function, we integrate the concept of grey number in Grey theory, maximum conditional entropy to get improved two-dimensional grey entropy. In essence, the fast segmentation speed of our method owes to PSO-GSA algorithm, which has an outstanding convergence performance. On the other hand, the segmentation quality of our method is benefit from the improved two-dimensional grey entropy, for the fact that noise almost completely disappears. Experimental results indicate that our method is superior to GA based, AFS based and ABC based methods in terms of segmentation accuracy, segmentation time, and Thresholding.
REFERENCES [1] R. Wilson and M. Spann, Image segmentation and uncertainty, John Wiley & Sons, Inc., 1988. [2] A. R. Weeks, Fundamentals of electronic image processing, SPIE Optical Engineering Press Bellingham, 1996. [3] E. R. Dougherty, R. A. Lotufo and T. I. S. for Optical Engineering SPIE, Hands-on morphological image processing, vol. 71, SPIE press Bellingham, 2003. [4] Y.-J. Zhang, Advances in image and video segmentation, IGI Global, 2006.
[5] L. Zhang and Q. Ji, "A Bayesian Network Model for Automatic and Interactive Image Segmentation," Image Processing, IEEE

Transactions on, vol. 20, no. 9, pp. 2582- 2593, Sept 2011. [6] J. Tang, "A color image segmentation algorithm based on region growing," in Computer Engineering and Technology (ICCET), 2010 2nd International Conference on, 2010. [7] W. Tao, H. Jin and Y. Zhang, "Color Image Segmentation Based on Mean Shift and Normalized Cuts," Systems, Man, and Cybernetics, Part B: Cybernetics, IEEE Transactions on, vol. 37, no. 5, pp. 1382- 1389, Oct 2007. [8] S. Wang and J. Siskind, "Image segmentation with ratio cut," Pattern Analysis and Machine Intelligence, IEEE Transactions on, vol. 25, no. 6, pp. 675-690, June 2003. [9] H. Zhang, Q. Zhu and X. feng Guan, "Probe into Image Segmentation Based on Sobel Operator and Maximum Entropy Algorithm," in Computer Science Service System (CSSS), 2012 International Conference on, 2012. [10] A. Lahouhou, E. Viennet and A. Beghdadi, "Combining and selecting indicators for image quality assesment," in Information Technology Interfaces, 2009. ITI '09. Proceedings of the ITI 2009 31st International Conference on, 2009. [11] S. Usai, "A least squares database approach for SAR interferometric data," Geoscience and Remote Sensing, IEEE Transactions on, vol. 41, no. 4, pp. 753-760, 2003. [12] S. C. Zhu and A. Yuille, "Region competition: Unifying snakes, region growing, and Bayes/MDL for multiband image segmentation," Pattern Analysis and Machine Intelligence, IEEE Transactions on, vol. 18, no. 9, pp. 884-900, 1996. [13] K. Haris, S. N. Efstratiadis, N. Maglaveras and A. K. Katsaggelos, "Hybrid image segmentation using watersheds and fast region merging," Image Processing, IEEE Transactions on, vol. 7, no. 12, pp. 1684- 1699, 1998. [14] J. Shi and J. Malik, "Normalized cuts and image segmentation," Pattern Analysis and Machine Intelligence, IEEE Transactions on, vol. 22, no. 8, pp. 888-905, 2000. [15] A. Tsai, A. Yezzi Jr and A. S. Willsky, "Curve evolution implementation of the Mumford-Shah functional for image segmentation, denoising, interpolation, and magnification," Image Processing, IEEE Transactions on, vol. 10, no. 8, pp. 1169- 1186, 2001.
[16] S. Wang and J. M. Siskind, "Image segmentation with ratio cut," Pattern Analysis and Machine Intelligence, IEEE Transactions on, vol. 25, no. 6, pp. 675-690, 2003. [17] Y.-C. Lin, Y.-P. Tsai, Y.-P. Hung and Z.-C. Shih, "Comparison between immersion- based and toboggan-based watershed image segmentation," Image Processing, IEEE Transactions on, vol. 15, no. 3, pp. 632-640, 2006. [18] W. Tao, H. Jin and Y. Zhang, "Color image segmentation based on mean shift and normalized cuts," Systems, Man, and Cybernetics, Part B: Cybernetics, IEEE Transactions on, vol. 37, no. 5, pp. 1382- 1389, 2007. [19] S. Mirjalili and S. Z. M. Hashim, "A new hybrid PSOGSA algorithm for function optimization," in Computer and information application (ICCIA), 2010 international conference on, 2010. [20] M. Ma, J. Liang, M. Guo, Y. Fan and Y. Yin, "SAR image segmentation based on Artificial Bee Colony algorithm," Applied Soft Computing, vol. 11, no. 8, pp. 5205- 5214, 2011. [21] H. Zhang, Q. Zhu and X.-f. Guan, "Probe into Image Segmentation Based on Sobel Operator and Maximum Entropy Algorithm," in Computer Science & Service System (CSSS), 2012 International Conference on, 2012. [22] C. Liu, Y. Zheng, Z. Pan, J. Duan and G. Wang, "SAR image segmentation based on fuzzy region competition method and Gamma model," Journal of Software, vol. 8, no. 1, pp. 228-235, 2013.

B49010511

More Related Content

What's hot (20)

Viewers also liked (20)

Similar to B49010511 (20)

B49010511