You are on page 1of 6

International Journal of Research and Reviews in Computing Engineering Vol. 1, No.

1, March 2011 Copyright Science Academy Publisher, United Kingdom www.sciacademypublisher.com


Science Academy Publisher

A Fuzzy Watermarking System Using the Wavelet Technique for Medical Images
Sameh Oueslati1, Adnane Cherif 2, and Bassel Solaiman3
1 2

University of Sciences of Tunis, Department of Physics, Laboratory of Signal Processing, Tunisia. Higher National School of Telecommunication of Bretagne, Department of Image and Information Processing, Brest, France.

Correspondence should be addressed to Sameh Oueslati, sameh.oueslati@telecom-bretagne.eu

Abstract In medical imaging, it has been shown that watermarking can improve data protection and content enrichment. In this paper, an adaptive watermarking algorithm performed in the wavelet domain is proposed which exploits a human visual system (HVS) and a Fuzzy Inference System (FIS). Human visual system (HVS) is adopted to further ensure the watermark invisibility. The FIS is utilized to compute the optimum watermark weighting function that would enable the embedding of the maximum-energy and imperceptible watermark. For the purpose of security and robustness, a watermark sequence is embedded by selectively modifying the middle-frequency parts of the image. The experimental results achieved demonstrate that the proposed algorithm is robust against both, signal processing and geometric attacks.

1.

Introduction

Lack of proper controls, procedures, and policies may tempt unauthorized users to access and use patient information in an inadequate fashion, weakening the credibility of healthcare information systems, and putting accuracy in diagnosis, treatment, and research into risk. A degraded or tampered image is a potential source for mistakes in diagnosis, treatment, or research. The fidelity of information in a medical record is essential; this naturally demands means to identify the provenance and authorship of the information [2]. Despite the importance of integrity and authenticity for medical images, only recently this field of research has been addressed. Arguably, one of the first works was published in 1995 [3], but it was only after 1999 [4] that medical image integrity and authenticity finally got their deserved attention. There is a need for an infrastructure to ensure correct storage, processing, and visualization of medical images, a requisite that is also stressed by the Code of Ethics for Health Information Professionals [10]. Many works in this field rely on the use of watermarking that embeds information into its own data. For image security, usually the digital signature is the metadata, which is stored along with the medical image. It is also important to remember that watermarking has not been accepted by the DICOM standard yet, though there is a number of works involving watermarking DICOM images [8]. Some years ago to improve medical image security [5]-[6] and it has received great attention since then. Zhou and Huang [7] presented a method employing digital envelopes, i.e., encrypted digital signatures to add integrity and authenticity. This was expanded in 2003, by storing the digital envelope in the non-

relevant portions of the image and presenting ways to make it consistent with the DICOM standard [8]. One common approach is to define a region of interest (ROI) and store the watermark outside of it, so that the security information would be stored in clinically irrelevant regions. Use of patient data in watermarking is suggested in works of Acharya et al. [9] and Miaou et al. [1], hiding data in the LSBs of the image. Reversible watermarking techniques have also been proposed [11], but with a very limited scope. Recent works advocate the use of wavelets to digitally watermark the image [12]-[13]. A newer work employs wavelets to introduce multiple watermarking to improve image security [14]. This is due to the wavelets' excellent spatial localization, frequency spread, and multi-resolution characteristics, which are similar to the theoretical models of the human visual system. In this paper, Discrete Wavelet Transform (DWT) domain is used and the watermark is embedded in the mid-frequency region, in order to achieve perceptual invisibility as well as robustness to attacks [9]. The challenge here is to introduce a digital watermark that does not alter the perceived quality of the electronic content, while being extremely robust to attack. In other words, one would like to insert the watermark with maximum strength before it becomes visible to the human visual system (HVS). As a result the way the strength of the added watermark is chosen is of highest importance. The rest of this paper is organized as follows: Section 2 we presented a brief introduction of the DWT and the technique of fuzzy model used to create maximum watermark. Section 3 describes the method for embedding and extraction watermark. The proposed algorithm will be evaluated based on its robustness against various Stirmark

International Journal of Research and Reviews in Computing Engineering [10] attacks. And then the experimental results of output images and the analysis will be shown. The conclusion as well as the future work will be discussed in the final section. Finally the reference list will be provided at the end of the paper.

44

2.

The Proposed Watermarking Technique

In this section, we explain the chosen method for embedding the watermark. The discrete wavelet transform decomposes an image in subbands having a bandwidth approximately equal on a logarithmic scale. To achieve imperceptibility, the lowest band of the image is left unmodified. To select the perceptually significant coefficients for each subband, we generate a watermarking key. Then inverse transforming the marked coefficients forms the watermarked image. We test strength of managing watermarked images then the information of image changes. Then we extract the watermark. After then, we evaluate the similarity between the original watermark of database and the extracted watermark. The embedding process is summarized in Figure 1.
Host Image

other transform such as progressive and low bit-rate transmission, quality scalability and region-of-interest (ROI) coding demand more efficient and versatile image coding that can be exploited for both, image compression and watermarking applications. The DWT is very suitable to identify the areas in the host image where a watermark can be embedded effectively. Wavelets are also being used in several emerging image and video compression standards such as JPEG2000 and MPEG 4 [19]. Therefore, we think it is imperative to consider the wavelet transform domain for watermarking applications. 2.2. Watermarking using Fuzzy Logic The principle of fuzzy logic approaches the human approach in the sense that the variables treated are not binary but of variable linguistic relatives of human language as high contrast, a lot more clearly, very textured, very homogeneous .... etc Moreover, these linguistic variables are processed with rules that refer to some knowledge of system. The fuzzifier and the defuzzifier have the role of converting external information in fuzzy quantities and vice versa, the core of a FIS is its knowledge base, which is expressed in terms of fuzzy rules and allows for approximate reasoning. In consequence, the approximations of the inferred values are optimized and are used to generate the adaptive strength of the watermark. In this work, the FIS is used since it can integrate the fuzzy and nonlinear aspect of human vision. The FIS is utilized to compute the optimum watermark weighting function. The determination of the membership function is done by using the help of ANFIS Toolbox in MATLAB. This technique enabled excellent model development for non-linear process in which the rules were automatically generated under ANFIS environment.

Embedding Watermarked image Communication Channel Distorted Watermarked image Extraction

Watermark image

Attacks

3.

Watermark Embedding and Detection

Secret Key

Recovered Watermark

Figure 1. Digital watermarking system.

2.1. Watermarking in the Wavelet Domain Among the transform domain watermarking techniques discrete wavelet transform (DWT) based watermarking techniques are gaining more popularity. Wavelet based watermarking techniques have multi-resolution hierarchical characteristics. Furthermore, its ability to decompose an image into bands that vary in both spatial frequency and orientation (vertical, horizontal and diagonal) has made it of great relevance when modeling the anisotropic properties of a Human Visual System (HVS). This mimics the human visual perception and allows the independent processing of the resulting components [18]. The high frequency subbands of the wavelet transform include the edges and textures of the image and the human eye is not generally very sensitive to changes in such bands. Also, watermark detection may be achieved at lower image resolutions, which saves computational load. DWT has a number of advantages over

This research applies wavelet transform by using Haar wavelet. When embedding watermark, a secret key is used to determine the watermark embedding beginning location. The coefficients, whose magnitudes are larger are selected to embed watermark in middle-frequency subband in particular, in the components of the first level Haar DWT. With the DWT, the edges and textures usually exist in high frequency sub bands, such as HH, HL, and LH. The large coefficients in these bands usually indicate edges and texture in the image [15]-[16]-[17]. Therefore, embedding the watermark into the maximum coefficients of the detail subbands is difficult for the human vision system to perceive. The detailed image embedded watermark insertion and extraction algorithm are discussed in this section. 3.1. Algorithm for Embedding Watermark in Wavelet Transform In our experiment, the host image is gray-scale image, and the watermarking image is binary image. The embedding process is described in the follow steps: Step1: Input the host image and the watermark image. Step2: Convert the watermark into a stream of bits (zeroes, and ones). Step3: Decompose the host image by using Haar wavelet transform.

International Journal of Research and Reviews in Computing Engineering Step4: Insert the data in the wavelet coefficients (which have the largest values) in middle-frequency subband. Step5: Perform the inverse Haar wavelet transform to obtain the watermarked image. Step6: Display watermarked image. 3.2. Algorithm for Extracting Watermark in Wavelet Transform The original image is not necessary at the watermark extraction stage. This refers to a blind watermarking process. The extracting process is described in the follow steps: Step1: Input watermarked image. Step2: Decompose image by using Haar wavelet transform. Step3: Select the wavelet coefficients (which have the largest values) in middle-frequency subband. Step4: Compare between coefficients in two images depending on locations of them, If coefficient embedding > coefficient original then the data store in it is 1. If coefficient embedding < = coefficient original then the data store in it is 0. Step5: Display Recovered Watermark.

45

where M , N is the size of the image and contains M N pixels, f ( x, y) is the host image and f ' ( x, y) is the watermarked image. This measure gives an indication of how much degradation, values near to zero indicate less degradation. We also found that obtained N C of extracted watermark depends on the host image as each image owns different frequency ranges.
M 1 1 M 2 1

W (i, j)W
i 0 j 0 M 1 1 M 2 1 i 0 j 0

(i, j )
2

NC

(4)

[W (i, j)]

Here, for an M1 M 2 Watermark image, W (i, j ) and


W (i, j ) denote the (i, j ) pixels values in the original image watermark and that in the extracted watermark, respectively. The original image is embedded by watermark and the watermarked image is transmitted via disturbed environment such a Additive Noise, Gaussian Noise, PEG compression, Median Filter, Rotation, Gaussian Blur. After the attack has been carried out, the extraction algorithm is applied. The full list of attacks and their result is given in Table1. Correlation Factor takes values between 0 and 1. A Correlation factor of 0.75 or more is acceptable. Values over 36 dB in PSNR are acceptable in terms of degradation, which means no significant degradation is observed by the human eye. Figure 2 shows the original image and watermarked image, is the result of the proposed technique without any attack. The PSNR is 41.95dB and wPSNR is 43.36dB. There is no perceptual distortion in the original and watermarked image, this proves that the watermarked image has a good image quality. Which means that scheme has satisfied the criteria that an efficient watermark should be unobtrusive, discreet and easily extracted. The results are discussed as follows.
*

4.

Experimental Results

There is always a tradeoff between imperceptibly level and robustness of the system. The bigger strength factor value, the more robustness obtained, but worse in perceptible level. Although, PSNR is often a useful tool to measure perceptibly level, it is not always accurate to human eyes judgment. We use a new version of the Peak Signal-to-Noise Ratio as a distortion metric to estimate image degradation. It takes into consideration the importance of the human visual system and its characteristics and therefore is more suitable for digital watermarking. Results of experiments, carried out on a database of 512 512 pixel-sized medical images, show that the watermarked image has at least 41.95dB in peak signal-to-noise ratio ( PSNR ) and show good imperceptibility and high robustness against common image processing attacks. In our paper, images are extracted from DICOM library and we use Matlab and its library as software to experiment our approaches and to validate our results. The ( wPSNR ) is used to evaluate the image quality by calculating the ( wMSE ) between the images to compare. The equations are as follows: Peak Signal-to-Noise Ratio:
PSNR 10l log10 (
2 X max 2552 ) 10 log10 ( ) MSE MSE

(a)

(b)

Figure 2. (a) Original medical image (b) Medical image watermarked.

(1)
Table 1. Results of the experiments using PSNR , wPSNR and N C metrics for fidelity and robustness. Type of Attack JPEG 80% JPEG 60% JPEG 40% JPEG 20% Median filtering (3x3) Median filtering (5x5) Median filtering (7x7) Gaussian noise 3% Gaussian noise 5% Gaussian Blur filtering (3x3) Gaussian Blur filtering (5x5) Rotation 2 degrees Rotation 3 degrees

where MSE is the mean-square error between the original image and the distorted one. The ( wPSNR ) is used to

PSNR
42.24 39.38 36.00 34.91 35.67 33.00 31.67 35.49 34.12 38.72 36.82 27.95 24.89

wPSNR

evaluate the image quality by calculating the ( wMSE ) between the images to compare. The equations are as follows:
wPSNR 10l log10 (
1 MN
M 1 N 1 x 0 y 0

NC
0.98 0.93 0.81 0.78 0.80 0.74 0.70 0.87 0.84 0.89 0.78 0.75 0.65

2 X max 2552 ) 10 log10 ( ) wMSE wMSE

(2) (3)

wMSE

(2 *

f ( x, y ) f ' ( x, y ) f ( x , y ) f ( x, y ) '

)2

43.01 40.96 39.32 36.30 37.54 34.91 30.58 36.89 35.43 40.15 38.03 29.76 27.82

International Journal of Research and Reviews in Computing Engineering 4.1. Watermarked Image under JPEG Compression Attack Lossy compression algorithms such as JPEG are commonly used for efficient storage and transmission of images over the Internet. This is also an important test that any watermarking algorithm has to defend itself against. In order to perform this experiment, the watermarked image shown in Figure 3 was compressed using different quality factors at 20%, 40%, 60%, and 80% respectively. The extracted watermarks from the aforementioned compressed images are shown in Figures. 4(b), 4(c), 4(d), and 4(e) respectively. The watermarked image indicates a good perceptual quality and the extracted watermark is similar to original one. As shown in Figure 4, the extracted watermark is still visually acceptable after the watermarked image has been undergone several attacks such as JPEG lossy compression. From the PSNR values indicated in the table 1, it is clear that the quality of the images extracted as well as the embedded is better when the compression is less. This seems normal since JPEG is a lossy compression method, the higher the compression; the more is the loss in information of the image and vice versa.

46

reducing the sharpness of the image. It is similar to an averaging filter in which the value of the output pixel is the mean of the pixel values in the neighborhood of the corresponding input pixel. However, in the present method of median filtering, as one might have guessed, the value of an output pixel is determined by the median of the neighborhood pixels. A median filtering with window size (3x3) (5x5) (7x7) was applied to the various watermarked images are depicted in Figure 5 and the maximum detector response is presented in Table 1. The filtered image was degraded; the detector response still provides the highest correlation value.

(a)

(b)

(a)

(b)

(c) Figure 5. Median filtered watermarked image: (a) with a (3x3) windows, (b) with a (5x5) windows, (c) with a (7x7) windows. 4.3. Watermarked Image under Noise Attack The watermarking scheme was also tested under additive Gaussian noise pollution, with zero mean and distinct variance values. When Gaussian distributed noise is added to the watermarked image we obtain Figure 6. The results demonstrate that this scheme is significantly robust against additive Gaussian noise.

(c)

(d)

Figure 3. The medical watermarked image is manipulated by using a jpeg compression: (a) with quality factor 80%, (b) with quality factor 60%, (c) (with quality factor 40%, (d) with quality factor 20%.

(a) (a) (b) (c)

(b)

(c)

(d)

(c)
Figure 6. The watermarked medical image is corrupted by Gaussian noise (a) 3%, (b) 5% and (c) 9%.

Figure 4. Original and extracted watermarks: (a): original, (b) JPEG 80%, (c)JPEG 60%, (d) JPEG 40%, (e) JPEG 20%.

4.2. Watermarked Image under Median Filter Attack Median filtering is used to remove outliers without

4.4. Watermarked Image under Gaussian Blur Attack The Gaussian blur filter creates an output image after applying a Gaussian weighted average of the input pixels around the location of each corresponding output pixel.

International Journal of Research and Reviews in Computing Engineering Figure 7 is the Gaussian blurred watermarked image. This is achieved by calculating the average of a pixel and all of its eight immediate neighbors. This average is then used instead of the original value of the pixel.

47

0.9

0.8

Guo's scheme Chen's scheme Our scheme

Correllation

0.7

0.6

0.5

(a)

(b) (c)
0.4 20 30 40 50 60 70 JPEG compression quality 80 90

Figure 9. Comparison with algorithms using other techniques.

5.

Conclusion

Figure 7. Watermarked medical image is corrupted by Gaussian Blur: (a) (3x3) windows, (b) (5x5) windows, (c) (7x7) windows.

4.5. Watermarked Image under Rotation Attack Figure 8 shows the result of rotating the watermarked image Tests have been done for rotations by 2 and 3 degrees. After this has been done we crop the four corners of the rotated image in order to keep the same size as the original image. Now when the extraction algorithm is applied, we are able to retrieve more than 50% of the watermark after extraction.

The proposed watermarking scheme has become intelligent by adding an adaptive fuzzy logic interface to judge the strength of watermark and optimize the watermark embedding process. By exploiting the fuzzy inference system, our watermarking system provides the maximum watermark embed strength which is robust to common attacks. By different embedding strengths decided by fuzzy inference system according to different textural features of image, the resulting watermarked image is extremely robust to a wide range of image JPEG compression. Implementation results show that the imperceptibility of the watermarked image is acceptable. The watermarks can be extracted from the most of the common image processing attack with high Normalized correlation values. Our watermarking system does not require the original image for watermark detection. As future work; we will extend the algorithm in order to obtain less degradation in the watermarked image and obtain better accuracy in the recovered watermark.

Acknowledgment
(a) (b)
Figure 8. Rotated watermarked image: (a) 2 degrees, (b) 3 degrees.

The authors express their gratitude to the doctors of the clinic MANAR II for their assistance.

4.6. Comparison with Other Techniques The proposed method is also compared with recent approaches developed in the last years. When compared with [21]-[23], this method proved its efficiency with better robustness against different geometrical attacks. Whereas if comparing it with [24]-[25] we find that the proposed method allows more robustness in term of correlations results with minimal distortions. Figure 9 show a comparison between the proposed method and recent algorithms such as [20]-[22]. It is clear that in term of distortion against intentional attack through medical image manipulation or transmission such as lossy JPEG compression, high robustness is proved and the patient data is conserved.

References
[1] S. Miaou, C. Hsu, Y. Tsai, and H. Chao, A secure data hiding technique with heterogeneous data-combining capability for electronic patient records, presented at the World Congr. Med. Phys. Biomed. Eng., Proc., Session Electron. Healthcare Records (IEEE-BEM), Chicago, IL, 2000. A. L. Rector, W. A. Nolan, and S. Kay, Foundations for an electronic medical record, Methods Inf. Med., vol. 30, 1991, pp. 179186. S. T. C. Wong, M. Abundo, and H. K. Huang, Authenticity techniques for PACS images and records, Proc. SPIE, vol. 2435, 1995, pp. 6879. B. Macq and F. Dewey, Trusted headers for medical images, presented at the DFG VIII-DII Watermarking Workshop, Erlangen, Germany, 1999. G. Coatrieux, H. Maitre, B. Sankur, Y. Rolland, and R. Collorec, Relevance of watermarking in medical imaging, in Proc. 2000 IEEE EMBS Conf. Inf. Technol. Appl. Biomed., pp. 250255. G. Coatrieux, H.Maitre, and B. Sankur, Strict integrity control of biomedical images, presented at the SPIE Conf. 4314: Security Watermarking Multimedia Contents III, San Jose, CA, 2001. X. Q. Zhou and H. K. Huang, Authenticity and integrity of digital mammography images, IEEE Trans. Med. Imag., vol. 20, no. 8, Aug. 2001, pp. 784791.

[2] [3]

[4]

[5]

[6]

[7]

International Journal of Research and Reviews in Computing Engineering


[8] F. Cao, H. K. Huang, and X. Q. Zhou, Medical image security in a HIPAA mandated PACS environment, Comput. Med. Imag. Graph., vol. 27, pp. 185196, 2003. U. R. Acharya, P. S. Bhat, S. Kumar, and L. C. Min, Transmission and storage of medical images with patient information, Comput. Biol. Med.,vol. 33, 2003, pp. 303210. IMIA. IMIA Code of Ethics for Health Information Professionals [Online]. (2002). Available: http://www.imia.org/English_code_of_ethics.html J. M. Zain, L. P. Baldwin, and M. Clarke, Reversible watermarking for authentication of DICOM images, in Proc. 26th Annu. Int. Conf. Eng. Med. Biol. Soc. (EMBC 2004), 2009, vol. 2, pp. 32373240. A. Giakoumaki, S. Pavlopoulos, and D. Koutsouris, A multiple watermarking scheme applied to medical image management, in Proc. 26th Annu. Int. Conf. IEEE (EMBS 2004), pp. 32413244. A. Wakatani, Digital watermarking for ROI medical images by using compressed signature image, in Proc. 35th Hawaii Int. Conf. Syst. Sci., 2002, pp. 20432048. A. Giakoumaki, S. Pavlopoulos, and D. Koutsouris, Secure and efficient health data management through multiple watermarking on medical images, Med. Biol. Eng. Comput., vol. 44, 2006, pp. 619 631. A. lumini, D. Maio "A blind Watermarking System for Digital Images in the Wavelet Domain", University of Bologna, Viale Risorgimento. X. G. Xia, " A Multi Resolution Watermark for Digital Images" proc. IEEE Int. Conf. on Image Processing ,vol.1, oct. 1997, pp.548-551. A. K. Musa, "Watermark Applications in Color Image using Wavelet Transforms ", PH.D, thesis, Iraqi Commission for Computer and Informatics, Baghdad, Iraqi, 2004. S. Srdjan, O. Irena, An application of multidimensional timefrequency analysis as base for the unified watermarking approach, IEEE Transactions on Image Processing, vol. 19, no. 3, 2010, pp. 736 745. K. Nima, M., Seyed, A Robust image watermarking in the ridgelet domain using universally optimum decoder, IEEE Transactions on Circuits and Systems for Video Technology, vol.20, n0.3, 2010, pp. 396-406. L.H. Chen and J.J. Lin, Mean quantization based image watermarking, Image and vision computing vol. 21, No. 8, 1 August 2003, pp. 717-727. Z. Dawei, C. Grong and L. Wenbo, A chaos based robust waveletdomain watermarking algorithm, Science direct journal, March 2004. [22]

48

[9]

[10]

[11]

[12]

H. Guo and N. D. Georganas, Multi-resolution Image Watermarking Scheme in the Spectrum Domain, IEEE Canadian conference on electrical and computes engineering, may 2002, pp125. [23] B.S. Kim, J.G. Choi, C.H. Park, J.U. Won, D.M. Kwak, S.K. Oh, C.R. Koh and K.H. Park, "Robust digital image watermarking method against geometrical attacks, " Real time imaging journal, Vol. 9, No. 2, April 2003, pp. 139-149. [24] F-Y. Shih and S.Y.T Wu, F.Y. Shih, S.Y.T. Wu, "Combinational image watermarking in the spatial and frequency domains", Pattern Recognition, vol. 36, Issue 4, April 2003, pp. 969-975. [25] P-T. Yu, H-H. Tsai and J-S. Lin, "Digital watermarking based on neural networks for color image", Elsevier Trans. On Signal Processing, Vol.81, Issue 3, March 2001, pp. 663-667.

[13]

[14]

[15] [16] [17]

Sameh Oueslati is a researcher at the Image and Information Processing Department Higher National School of Telecommunications of Bretagne she is also in signal processing laboratory at the University of Sciences of Tunis - Tunisia (FST). Degree in electronics and she received a Masters degree in 2006 from the University of Sciences of Tunis. She is currently a PhD student at the Faculty of Sciences of Tunis of where she is a contractual assistant. His research interests include information hiding and image processing, digital watermarking, database security.

[18]

[19]

Adnane Cherif obtained his engineering diploma in 1988 from the Engineering Faculty of Tunis and his Ph.D. in electrical engineering and electronics in 1997. Actually he is a professor at the Science Faculty of Tunis, responsible for the Signal Processing Laboratory. He participated in several research and cooperation projects, and is the author of more than 60 international communications and publications.

[20]

[21]

Bassel Solaiman Ing. Ph.D., Professor, is Telecom Engineer, holds a Ph.D. and HDR in Information Processing, University of Rennes I, is currently Professor and Head of Image and Information Processing from the Higher National School of Telecommunication of Bretagne in Brest, France. His research interests include, among others, on different approaches to treatment and Information Fusion and have been the subject of numerous publications.

You might also like