Particle Swarm Optimization Based Parameter ... - IEEE Xplore

10 downloads 746 Views 757KB Size Report
Page 1. Particle Swarm Optimization Based Parameter Optimization Technique in. Medical ... Patient Record (EPR), where the embedding factors/scaling factors ...
Particle Swarm Optimization Based Parameter Optimization Technique in Medical Information Hiding Sayan Chakraborty1, Sourav Samanta2, Debalina Biswas1, Nilanjan Dey3, Sheli Sinha Chaudhuri3 1

Dept. of CSE, JIS College of Engineering, Kalyani, West Bengal, India Dept. of CSE, University Institute of Technology, BU, Burdwan, India 3 Dept. of ETCE, Jadavpur University, Kolkata, West Bengal, India [email protected], [email protected], [email protected], [email protected], [email protected] 2

Abstract— In this era of globalization, use of technology has influenced medical science as well. Now-a-days, exchanging medical information using communication technologies like network devices or telecommunication to provide health care services for medical case studies amongst various diagnostic centers or hospitals is a very common practice. In this paper, a Discrete Wavelet Transformation (DWT) based method is proposed for embedding a Hospital Logo or Electronic Patient Record (EPR), where the embedding factors/scaling factors are optimized by Particle Swarm Optimization (PSO). Keywords: Watermarking; Discrete Wavelet Transformation; Particle Swarm Optimization, Biomedical image; Electronic Patient Record;

I. INTRODUCTION Fast improvement and growth of technology and telecommunication has influenced the medical field immensely. Doctors, medical professionals and practitioners around the globe are now using wireless media to exchange medical data for mutual availability of therapeutic case studies. Many hospitals and diagnostic centres around the world also transmit medical information to/from other hospitals or diagnostic centers to improve the diagnostic results and communicate to study a case. With the rise in use of internet and multimedia the stealing of information from biomedical images, has become a major concern for healthcare professionals. Data Hiding/Watermark are added ownership to increase the level of security and to verify authenticity. Patients’ information (Electronic Patient Record), logo of the hospitals or diagnostic centers can be added in the bio-medical data as watermark to prove the intellectual property rights. Extraction is a major factor, as any kind of distortion during retrieval of the watermarked image will be unacceptable. After extraction of the watermark, it claims the original object and thus data can be successfully hidden in an image. In other words, it can be said that watermark is actually a data, which is embedded within an object or image, and later it is extracted to recollect the required information. PSO is an adaptive meta-heuristic search algorithm based on the evolutionary ideas of particle’s movement in a swarm. Particle swarm optimization (PSO) is one of most popular meta-heuristic algorithms. Even though PSO is very similar to some evolutionary algorithms, but the

978-1-4799-1597-2/13/$31.00 ©2013 IEEE

standard PSO does not use evolution operators like crossover and mutation. In 2001, S Zinger, Z. Jin, H. Maitre and B. Sankur [1] et al. proposed a method for optimization of watermarking performances using error correcting codes and repetition. In this paper, they have investigated different ways of applying error correction codes, repetition and some combinations of the two given different capacities of a fixed image for different error rates of the watermarking channel, in order to obtain optimal selection for a given length of signature. In December 2005, a new approach for optimization in image watermarking by using genetic algorithms was proposed by P. Kumsawat [2] et al. In this paper, he proposed the spread spectrum image-watermarking algorithm using the Discrete Multiwavelet Transform. In 2010, Hai Tao, Jasni Mohamad Zain, Ahmed N. Abd Alla and Qin Hongwu [3] et al. proposed implementation of digital image watermarking based on Particle Swarm Optimization. In 2012, D. Venkatesan, K. Kannan and S. Raja Balachandar [4] et al. introduced optimization of fidelity in digital image watermarking using a new genetic algorithm. Therefore, the paper is structured as follows: Section II presents a brief introduction of DWT based watermarking, and the brief introduction of PSO. Section 3 presents the proposed method in detail. Section 4 presents main results and discussions. Finally, the conclusions are drawn in Section 5. II. DWT BASED WATERMARKING AND PSO In DWT based embedding approach, cover biomedical image is decomposed up to 2nd level and watermark [7, 8, 9] is embedded in LH2, HL2, HH2 sub bands. These subbands are modified using generated session key based pseudo random sequences and the content of the corresponding binary watermark image bits using three manually selected scaling factors (k1, k2, k3). Inverse stationary wavelet transform (IDWT) is applied twice to generate the watermarked image. In the recipient end, watermark can be easily recovered by using correlation function and knowing the size of the authenticating image. In this technique, the strength of watermarking is measured using a set of three embedding factors namely, k1, k2 and k3. Manually selecting an optimal set of these

2013 IEEE International Conference on Computational Intelligence and Computing Research

factors is a complicated task. In the current work, selection of optimal embedding factors is performed using particle swarm optimization (PSO). Particle swarm optimization (PSO) [5 ,6] is a metaheuristic search method, which is inspired by particle's movement in a swarm. PSO technique proposes optimization method of a problem by iteratively trying to improve a candidate solution by looking at the particle's previous position and best position. PSO starts with a set of possible solutions or particles as a swarm, which is created randomly. A new swarm position is generated by updating particle's velocity and position iteratively and then comparing with their old position. This is inspired by the fact, that newer positions of the particles' in the swarm will be better than the older ones. The particles iteratively update their position and velocity by using some mathematical operations that include pbest (local best position), gbest (Global best), weighting function etc. After a number of iterations, the swarm position contains better fitness values. The main advantage of PSO is that it can take any values, even outside their side constraints, based on their current position in the swarm and calculate the velocity vector. For a given problem, PSO will use these distinct components to optimize the problem: 1) pbest or Local best, 2) gbest or Global best, 3) Weighting Factor, 4) Velocity of Particles and 5) Weighting Function. PSO based optimization works with an objective function which undertakes a population based search. In this algorithm particles are considered as solutions. Initialization of these particles is done randomly, as they become ready to move freely and fly across the multidimensional search space. Each particle updates its velocity during moving freely in the search space and also changes their position according to their and the population's best experience. The velocity of all particles in each iteration is updated using the following equation:

JJG JJG JJJJJG JJG JJJJJG JJG vi = wvi + c1R 1 (pi,best − pi ) + c2 R 2 (g i,best − pi ) JJG JJG where pi and vi are the position and velocity of particle JJJJJG JJJJJG i, respectively; similarly pi,best and g i,best is the ‘best’ position (position with the ‘best’ fitness value) discovered particle i and the entire population respectively. Here w is a parameter controller which is controlling the flying dynamics. R1 and R2 are two random variables from range [0, 1]. The factors which are controlling the weight of corresponding terms are C1 and C2.

JJG vi should be

checked after updation so that it lies within the predefined range. This step is necessary to avoid violent random walking. The positions of every particle between successive iterations are updated according to the following equation

JJG JJG JJG p1 = pi +vi

This updation of velocity and position drives the particle towards the region with better objective function value. In each iteration, local best and global best value both are updated according to the following conditions:

JJJJJG JJG JJG JJJJG pi,best = pi if f (pi ) > f (pi,best ) JJJJJG JJG JJG JJJJJG gi,best = gi if f ( g i ) > f ( gi,best ) JJJJG Where, f ( x ) is the objective function subject to maximization. After some iterations, algorithm stops, based on some predefined conditions (no. of iterations). After termination of the process, all particles come together at the point having highest objective value. Hence, the values of

JJJJG JJJJG g best and f ( g best ) can be termed

as the problem’s solution. Start

Initialize particles with random position and velocity vectors

For each particle’s position (p) evaluate fitness

If fitness (p) better than fitness (pBest) then pBest=p

Set best of pbests as gBest

Update particles Velocity and position

Stop Given gBest optimal Solution Figure 1. Flow Chart for General PSO Algorithm

III. PROPOSED METHOD OF PSO BASED OPTIMIZATION In this proposed method, we are using particle swarm optimization for embedding factors, as PSO. At first PSO randomly generates a no. of particles, which forms the initial swarm. Each swarm has their own individual values of embedding factors. By using these values, a watermark [10, 11, 12] image is embedded and the PSNR value of watermarked biomedical image is calculated. Using PSO, the highest fitness amongst all solutions can be obtained. PSNR+100*correlation is chosen as the best position and the corresponding values of the embedding factors are stored as the optimized embedding factor used for embedding watermark.

2013 IEEE International Conference on Computational Intelligence and Computing Research

A. Optimal Scaling Factors Generation using PSO Step 1. N no. of initial particle swarm positions are generated randomly within the range of 0 to 1. Each particle is represented by its x, y, z positions where x, y and z is the optimization factors, k1, k2 and k3 respectively. All the particles are initialized with initial velocity which is,

JG JG JJG vi (t) = xi (t) - xi ( t – 1 ).

Step 2. Each particle is applied to watermark the image and the original image is recovered and the local best solution is found. Step 3. Until maximum iterations Step 4. Particle velocity is changed. Step 5. The best solution is found and the global best position is updated if present position is better than previous global best. Step 6. The particle position is changed according to global best. Step 7. Go to step 3

Original Image Watermarked Image Watermark Image

Watermark Embedding

Compute Correlation

Watermark Extraction Extracted Watermark

PSNR Computation

Particle Velocity Updating

Select Best Position

Update Global Best (gbest)

Update Particle Position

Figure 2. Optimal Scaling Factors Generation using PSO

In this proposed method, the randomly generated n number of particles forms the initial swarm. Each of the particles has individual values of embedding factors k1, k2 and k3. Then, by using these values, a watermark image is embedded and the PSNR value of the watermarked biomedical image is calculated. PSNR measures the quality of watermarked image. This performance metric is used to determine perceptual transparency of the watermarked image with respect to original image. Thus, high PSNR value implies better invisibility of the watermark. The PSNR value of the watermarked biomedical image is calculated afterwards.

PSNR =

XY max Px2, y

∑(P

x, y

x, y

x, y

− Px , y )2

) .... (1)

After secret image embedding process, the similarity between the original image x and modified image x' is measured by the standard correlation coefficient as follows:

C=

∑ ∑ (x m

n

mn

− x ')( y mn − y ')

⎛ 2 ⎞⎛ 2 ⎞ ⎜ ∑ ∑ (x mn − x') ⎟⎜ ∑ ∑ ( y mn − y ') ⎟ ⎝ m n ⎠⎝ m n ⎠ .... (2)

Where y and y' are the transforms of x and x'. Then velocity of the swarm and particles is updated. After updating velocity, positions of particles are also updated. The above procedure is repeated for no. of iterations (n) defined. After n iterations, a set of (n+1) best positions is obtained. From these (n+1) positions, the position with the best fitness value is chosen as the best position. The entire procedure is repeated further to generate new sets of positions to produce the gbest. Out of these n no. of gbests, the position with the highest fitness value where highest fitness is PSNR+100*correlation, is chosen as the best position and the corresponding values of the embedding factors are stored as the optimized embedding factor used for embedding watermark.

2013 IEEE International Conference on Computational Intelligence and Computing Research

IV. RESULTS MATLAB 2012a software is extensively used for the study of PSO based watermarking technique on biomedical images. Sample biomedical image is taken

under consideration. Concerned images obtained in the result are shown below. The proposed algorithm is tested on DRIVE database[13] of retinal images [14].

TABLE I Sl. No.

Total No. of Swarm

Total Number of Iteration

Best Fitness

Best PSNR

Best Correlation

Optimal k1,k2,k3

1

15

5

85.3174

24.9071

0.6041

0.1270, 0.5697, 0.9572

2

15

6

85.4121

25.5874

0.5982

0.1270, 0.9157, 0.2162

3

15

7

85.9731

25.6031

0.6037

0.1270,0.9157,

0.2769

4

15

8

86.4010

25.6750

0.6073

0.5109,0.6668,

0.5440

5

15

9

86.4025

25.4229

0.6098

0.5204,0.4507,

0.7982

6

15

10

86.4717

25.5616

0.6091

0.5109,0.6668,

0.5440

7

15

15

87.1786

25.7253

0.6145

0.1270 ,0.9157, 0.2769

8

15

20

87.1786

25.7253

0.6145

0.1270 ,0.9157, 0.2769

9

15

25

87.1786

25.7253

0.6145

0.1270 ,0.9157, 0.2769

10

15

30

87.1786

25.7253

0.6145

0.1270 ,0.9157, 0.2769

11

15

35

87.1786

25.7253

0.6145

0.1270 ,0.9157, 0.2769

12

15

40

87.1786

25.7253

0.6145

0.1270 ,0.9157, 0.2769

13

15

45

87.1786

25.7253

0.6145

0.1270 ,0.9157, 0.2769

14

15

50

87.1786

25.7253

0.6145

0.1270 ,0.9157, 0.2769

(Ranges between 0-1)

Table I and figure 3 reports that in this current case study, 15 no. of swarms and 15th iteration gives optimal k1, k2 and k3. It is clearly visible from the above mentioned table, that after 15th iteration the best fitness and the corresponding embedding factors (k1, k2 and k3) remained unaltered. Hence, it can be said, that the convergence of results is obtained after 15th iteration. For this particular case study of retinal images, the embedding factors are k1 = 0.1270, k2 = 0.9157 and k3 = 0.2769 respectively. The PSNR value obtained after convergence is quiet satisfactory(25.7253), which claims the robustness and the efficacy of the proposed method. Figure 3. No. of iteration vs Fitness value.

2013 IEEE International Conference on Computational Intelligence and Computing Research

Figure 4. Original Image

Figure 6. Recovered Watermark Image

(a)

Figure 5. Watermarked Image

Figure 7. Recovered Watermark Image

(b)

Figure 8. (a) Initial swarm position, (b) Obtain optimal k1, k2 and k3 for swarm no. 15 and Iteration no. 1

V. CONCLUSION In this paper, a digital image watermarking in biomedical image based on particle swarm optimization is proposed. The watermark scaling factors are optimized using particle swarm optimization. This proposed method is robust enough against various common attacks on biomedical images carrying

valuable information. Best fitness function having high PSNR and correlation values and their convergence, claims the efficacy and robustness of the proposed method. This work can be further extended using hybrid particle swarm intelligence, and multi-objective optimization techniques and can compared with the obtained PSO based results.

2013 IEEE International Conference on Computational Intelligence and Computing Research

REFERENCES [1]. S Zinger, Z. Jin, H. Maitre and B. Sankur "A method for optimization of watermarking performances using error correcting codes and repetition" in CMS’2001: Communications and Multimedia Security. [2]. P. Kumsawat "Optimization in image watermarking by using genetic algorithms" Published in: Signal Processing, IEEE Transactions on (Volume: 53, Issue: 12) on Dec.2005. [3]. H. Tao, J. Md Zain, A. N. Abd Alla and Q. Hongwu "Implementation of digital image watermarking based on Particle Swarm Optimization." published in Second International Conference, NDT 2010, Prague, Czech Republic, July 7-9, 2010. Proceedings, Part I. [4]. D. Venkatesan, K. Kannan and S. R. Balachandar "A optimization of fidelity in Digital Image Watermarking using a new Genetic Algorithm" published in Appl. Math. Sci., Vol. 6, 2012, no. 7376, 3607-3614. [5]. M. Rohani and A. N. Avanaki "A watermarking method based on optimizing SSIM index by using PSO in DCT Domain" published in Computer Conference, 2009. CSICC 2009. 14th International CSI. [6]. H-C Huang, Y-H Chen and A. Abraham "An optimized Watermarking using Swarm-Based Bacterial Foraging" published in Hybrid Intelligent Systems, 2009. HIS 09. Ninth International Conference on 12-14th Aug, 2009. [7]. N. Dey, M. Pal, A. Das,” A Session Based Blind Watermarking Technique within the NROI of Retinal Fundus Images for Authentication Using DWT, Spread Spectrum and Harris Corner Detection”, International Journal of Modern Engineering Research, May, 2012. [8]. N. Dey, P. Das, S. S. Chaudhuri, A. Das,” Feature analysis for the blind-watermarked

electroencephalogram signal in wireless telemonitoring using Alattar's method”, published by ACM in Proceedings of the Fifth International Conference on Security of Information and Networks , P-87-94,October,2012. [9]. N. Dey, S. Mukhopadhyay, A. Das, S. S. Chaudhuri,” Analysis of P-QRS-T Components Modified by Blind Watermarking Technique Within the Electrocardiogram Signal for Authentication in Wireless Telecardiology Using DWT”, published by International Journal of Image, Graphics and Signal Processing (IJIGSP) Vol-4, no.7, pp.33-46, 2012. [10]. N. Dey, P. Das, S. Biswas, A. Das, S. S. Chaudhuri,” Feature analysis for the reversible watermarked electrooculography signal using Low distortion Prediction-error Expansion”, published by IEEE in International Conference on Communications, Devices and Intelligent Systems (CODIS), P-624-627,December,2012. [11]. N. Dey, P. Maji, P. Das, S. Biswas, A. Das, S. S. Chaudhuri,” An Edge Based Blind Watermarking Technique of Medical Images without Devalorizing Diagnostic Parameters”, published by Advances in Technology and Engineering (ICATE), 2013 International Conference on Tribology and Engineering Systems, 2013. [12]. N. Dey, P. Maji, P. Das, S. Biswas, A. Das, S. S. Chaudhuri, “Embedding of Blink Frequency in Electrooculography Signal using Difference Expansion based Reversible Watermarking Technique”, Scientific Bulletin of the Politehnica University of Timisoara - Transactions on Electronics and Communications p-ISSN 15833380, vol. 57(71), no. 2, 2012. [13]. http://www.isi.uu.nl/Research/Databases/DRIVE/ [14]. J.J. Staal, M.D. Abramoff, M. Niemeijer, M.A. Viergever, B. van Ginneken, "Ridge based vessel segmentation in color images of the retina", IEEE Transactions on Medical Imaging, 2004, vol. 23, pp. 501-509.