Application of Particle Swarm Optimization
Algorithm to Color Space Image Segmentation and its Analysis
Dr. S. Mary Joans,
1Poornima.J
21Professor, Department of Electronics and communication Engineering , Velammal Engineering College Anna University, Chennai 600 066, India
2Electronics and communication Engineering, Velammal Engineering college, Anna university, Chennai 600 066, India
ABSTRACT
In this paper, a new segmentation method for multicolor images based on Particle Swarm Optimization (PSO) is being proposed.
The algorithm performs segmentation of an image with respect to swarm. The proposed algorithm randomly initializes each particle in the swarm for segmentation of images. PSO-based approach is proposed to tackle the color image quantization and spectral immixing problems. PSO algorithm works well in color spacing images to identify the reference image. The proposed work deals with the conversion of the captured color image into various color spaces and gray scale images. Then segmentation based on PSO algorithm is done at four different levels for each color spaces and gray scale image. The attributes of these color spaces are validated. This method can be used in medical image segmentation and image compression. The optimization method called Particle Swarm Optimization is applied to the field of pattern recognition and image processing.
Keywords: color spaces- RGB and YCbCr, Image segmentation, PSO (Particle Swarm Optimization).
1. INTRODUCTION
Detecting and tracking the moving entities (e.g. people, vehicles) are the primary goal of a video surveillance system.
The surveillance system involves the following three steps: i) segmentation of the video sequence to detect the objects of interest 2) extraction of features (e.g. position, motion, and shape) 3) features tracking. Image segmentation is one of the important stage in various image processing task that aims at partitioning an image into homogenous regions. Image segmentation serves as the key of image analysis and pattern recognition. It is a process of dividing an image into different regions such that each region is homogeneous.
Color of an image retains more information than gray level. In many pattern recognition and computer vision applications, the additional information provided by color can help the image
analysis process to yield better results than approaches using only gray scale information. More research has focused on color image segmentation due to its demanding need. At present, color image segmentation methods are mainly extended from monochrome segmentation approaches by being implemented in different color spaces. Gray level segmentation methods are directly applied to each component of a color space, then the results are combined to obtain the final segmentation result.
2. RELATED WORK
This section presents few relevant works published in the field of image processing, since it takes the foundation for the proposed work. A wide categories of detection algorithms have been proposed, and that can be broadly classified into the following classes: i) statistical approaches ii) non-statistical approaches iii) spatio-temporal approaches.
1) Statistical Based Approaches: In this class of approach, the background is modeled using a normal pdf where each pixel is modeled as a Gaussian distribution. Another type of statistical approach is that the frame differencing method.
2) Non-Statistical Based Approaches: The approach deals with characterizing an admissible interval for each pixel of the background image as well as the maximum rate of change in consecutive images. Background subtraction is a very simple approach to detect moving objects in video sequences. The basic idea is to subtract the current frame from a background image and to classify each pixel as foreground or background by comparing the difference with a threshold.
3) Spatio-Temporal Based Approaches: Another class of algorithm is based on spatio-temporal segmentation of the video signal. Here segmentation is performed in a 3D region of image-time space that considers the temporal evaluation of neighbor pixels. This approach leads to an improvement of the systems performance, compared with traditional frame difference methods.
Image segmentation methods have been divided into five categories: Pixel based segmentation, region based segmentation, edge based segmentation, clustering based segmentation.
2.1. Pixel based segmentation:
This is the simplest method of image segmentation called Thresholding method. An image is divided into an array of overlapping sub images and then optimum threshold is applied for each sub image [1]. The threshold for each single pixel is found by interpolating the results of the sub images. The drawback of this method is that it is computational expensive and therefore, is not appropriate for real-time applications.
2.2. Region-based segmentation:
An image is partitioned into a set of connected regions. The homogenous regions are found according to a specific criterion such as intensity value and texture. Then Thresholding is performed independently in each region. Region based segmentation [1] aims to characterize the detected object by parameter certain parameter analysis (shape, position, size).
Region growing is one of the methodologies in this type of segmentation.
2.3. Edge based segmentation:
In this method, segmentation relies on discontinuities in the image data in order to locate the boundaries [5]. Edge profile like shading or texture can vary due to which this method is not that reliable. Sobel, canny, laplacian of Gaussian are some of the edge based object detection method.
2.4. Clustering based segmentation:
Clustering- based segmentation methods have been widely used in segmentation of grey level images. Clustering methods are directly applicable or easily extendable to higher dimensional data. K-means and fuzzy k-means are the two popular clustering methods.
3. PROPOSED APPROACH
We first perform capturing of live video and taking an image for segmentation. Then we convert the image into color spacing images. PSO (Particle Swarm Optimization) algorithm [2] is then applied to segment the first level of multicolor images. The flow of proposed approach is shown in fig.1. The objective of applying PSO algorithm is that it works well in multicolour images to identify the reference image.
Fig.1. Block diagram
PSO Algorithm:
Particle Swarm Optimization (PSO) is a kind of evolutionary computational techniques developed by Kennedy and Elberhart. PSO is a simple but powerful search technique that can be applied successfully to a wide variety of optimization problems, including image processing problems such as image segmentation [3].First, some particles are initialized which are known as initial population and the number of particles chosen are referred as swarm size. Each particles are initialized with initial assumed solutions and then gradually all the particles
best so far and the group or global best in current time as shown in equation 4 and equation 5. The best values ( pBest or gBest ) are obtained using a function to be optimize (i.e.
maximize or minimize) . That function is known as fitness function in swarm intelligence terminologies. The flowchart of PSO algorithm is shown in fig.2.
Vit+1=Vit + K1 x rand() x (Pi-Xit)+K2 x rand() x (Gt-Xit) -- (1)
Xit+1 = Xit + Vit+1 --- (2)
where V
i t and X
i
t are the velocity and position of ith particle in
tth iteration. P
i is the pBest of ith particle. Gt is the gBest at tth iteration. K
1 and K
2are the speed factors generally taken as 2 and rand() is the random function in the interval [0,1]. we can define the fitness function as given in equation (3). The threshold value for which this function gives maximized fitness value is preferred and the fitness function is given by,
f(t)=F0+F1 --- (3)
PSO steps:
1. Initialize each particle .
2. For each particle calculate the fitness value and personal best (pBest).
3. calculate Global Best = best among all particles.
4. Update new velocity and positions.
5. Repeat 2 to 4 till termination criteria reach.
PSO flowchart:
Fig.2. Flowchart
The resolution of the image in the proposed system is found to be 1280x720 pixels. The color spacing images are extracted from the captured image as shown below from fig. 3 to 5.. The segmentation algorithm is then applied to certain color spaces like RGB, YCbCr and gray scale and segmented as shown from fig.6 to 14 at various levels of wavelength such as 700- 635nm, 560-490nm and 490-450nm. In Table1, it is seen that the brightness and contrast is comparably good in the case of YCbCr when compared to grey scale image.
4. EVALUATION RESULTS
In Table1, it is seen that the brightness and contrast is comparably good in the case of YCbCr when compared to grey scale image.
Fig.3. YCbCr image
Fig.4. RGB image
Fig.5. Gray scale image
Fig.6. RGB I level segmentation
Fig.7. RGB II level segmentation
Fig.8. RGB II level segmentation
Fig.9.YCbCr I level segmentation
Fig.10.YCbCr II level segmentation
Fig.11.YCbCr II level segmentation
Fig.12.GrayscaleI level segmentation
Fig.13.GrayscaleII level segmentation
Fig.14.GrayscaleIII level segmetnation Table1. Attributes of gray scale and color image.
Fig.15. Comparison Graph
Though the histogram values of the gray scale attributes are higher than the color images as shown in fig.15, color spaces are advantageous. The reason is that the here the captured colored images are converted to gray scale and other color spaces. Where, the existing one performs gray scale to gray scale image segmentation. Fitness value of color space images are approximately nearer to the gray scale values as tabulated in table 2. Then fig.16 shows a smooth optimization in the fitness angle.
Table 2. Image Parameters analysis table.
5. CONCLUSION
In this paper, the proposed method of segmentation using PSO provides optimized parameter values for color images with better results when compared to grayscale images. We conclude that a PSO-based algorithm can serve as a suitable method for color image segmentation. The fitness angle of color spaces is optimized to 97%. The intensities of the color images are high. The future work consists of IR image segmentation and combined with point matching approach to determine that the object of interest in the captured image is the one which is in the reference image. Thus this project deals with the surveillance image processing which is beneficial for military purposes and several other surveillance purposes.
ACKNOWLEDGEMENT
I wish to acknowledge my sincere gratitude for the support from my guide Dr. S. Mary Joans, PhD, Professor for her technical supports throughout this project conduct.
REFERENCES
[1] Manya V. Afonso, Jacinto C. Nascimento, Jorge S.
Marques, “Automatic Estimation of Multiple Motion Fields From Video Sequences Using a Region Matching Based Approach”, IEEE transaction (2013).
[2] Fahd mohsen, Mohiy Hadhoud, Kamal Mostafa, “A new image segmentation method based on particle swarm optimization”, The international arab journal of Information Technology, (2012).
[3] Amanpreet Kaur, M.D. Singh , Department of Electrical and Instrumentation Engineering, Thapar University, Patiala, “An Overview of PSO- Based Approaches in Image Segmentation”, International Journal of Engineering and Technology, Volume -2 (2012).
[4] Fahd M. A. Mohsen,Mohiy M. Hadhoud , Khalid Amin, “ A new Optimization-Based Image Segmentation method By Particle Swarm Optimization”, IJACSA, (2011).
[5] Marko Heikkila, Matti Pietikainen, and Cordelia Schmid,
“Description of interest regions with local binary patterns,” Pattern Recognition, Vol. 42, no. 3, pp. 425–
436 (2009).
[6] Mahamed G. H. Omran, “Particle Swarm Optimization Methods for Pattern Recognition and Image Processing”, University of Pretoria, (2004).
[7] Chui, Haili; Rangarajan, Anand, “A new point matching algorithm for non-rigid registration”, Computer Vision and Image Understanding, 89 (2), pp. 114–141, (2003).
[8] L.Wang, W.Hu, and T. Tan,“Recent developments in human motion analysis”, Pattern Recognition, Vol. 36 (3), pp. 585–601, (2003).