Essay: Digital Image Watermarking

Digital Image Watermarking in High Frequency Band Using Median Filter based on DWT-SVD under Various Attacks

Abstract’ This paper describes digital image watermarking technique based on discrete wavelet transform and singular value decomposition. Median filter function is applied for the removal of noise from the original image before embedding the watermark in it. Watermark is embedded in the original image by modifying the singular value of second level high frequency sub band of discrete wavelet transform. A reliable watermark extraction technique is developed for the watermark extraction from watermarked image. Performance evaluation of proposed technique is done on 8-bit gray scale images using qualitative metrics i.e. Peak-signal-to-noise ratio (PSNR) and Correlation Coefficient (CC). High values of PSNR and CC indicate that the proposed technique is excellent improvement in the quality of image after extraction of watermark as compared to simple DWT-SVD technique. Further, the experimental results show better visual imperceptibility and resiliency of the proposed scheme against intentional or un-intentional variety of attacks.

Keywords’ Watermarking; Discrete Wavelet Transform; Singular Value Decomposition; Median Filter; Peak Signal-to-Noise Ratio; Correlation Coefficient.

I. INTRODUCTION
Digital watermarking is the process to hide the ownership data called watermark into a host signal (multi-media data) such that the watermark can be extracted later to make an assertion about the object. Various techniques are developed [1], [2] for data protection with the consideration that some data must be available to one who is interested in the ownership of the intellectual property.
Digital watermarking can be classified as visible and invisible watermarking. In visible watermarking, watermark after embedding visible in the multimedia data and in invisible watermarking embedded watermark is not visible in host data. It can be further classified into robust, semi fragile and fragile watermarking. Robust watermarking is used for the copyright protection of data and also shows resistant against various attacks. Fragile watermarking is use to protect the data from tampering, which get destroy on small interruption and semi fragile is a combination of both fragile and robust watermarking which is used for authentication of data as shown in figure 1.

Fig.1. Types of watermarks
According to Digital watermarking domain, it can be classified into spatial and transform domain. In spatial domain technique, the watermark is embedded directly by modifying the pixel values of host image (e.g. LSB, Patch work scheme) [3], [4], [5] where as in transform domain modification of frequency coefficient of image is done for embedding watermark. Generally, transform domain based watermarking scheme have more information embedding capacity and more robustness than their spatial counterpart. Discrete Cosine Transform (DCT), Discrete Wavelet Transform (DWT), Discrete Fourier Transform (DFT), [10], [11], [14], [15] and many other frequencies modifying transform variants are commonly used for transform domain watermarking. In last few years the Singular Value Decomposition (SVD) based transform has been popularly used in watermarking [6], [7], [12], [13].
Watermarking schemes based on SVD alone can’t guarantee high robustness against attacks; therefore it is applied in combination with DCT, DFT and DWT [8], [9], [19], [20], [21], [22], [23] due to their multi resolution capabilities.
In this paper we develop a hybrid digital image watermarking scheme using filter function. Hybrid digital watermarking developed is a combination of discrete wavelet transform and singular value decomposition. Embedding is done in 2nd level high frequency sub-band of the image obtains after the decomposition of the image into four sub band using HAAR filter.
The rest part of the paper is organized as follows. Section II explains the pre-processing process. Section III explains preliminaries terms includes Discrete Wavelet Transform and Singular Value Decomposition. Section IV illustrates the proposed new digital watermarking technique. Section V provides experimental results, where the PSNR and CC values of some attacks are compared with previous Work [11] for showing the robustness of the proposed scheme. Section VI concludes the paper.
II. PREPROCESSING
In preprocessing original image is preprocessed using median filter for noise reduction to do a better job of preserving edges than simple smoothing filter. In median filtering, the neighbouring pixels are ranked according to brightness (intensity) and the median value becomes the new value for the central pixelis a non-linear digital filtering technique use to remove the noise from an image. The main idea of the median filter is to run through the signal entry by entry, replacing each entry with the median of neighboring entries. In our proposed work we utilize median filtering function with block proc function to improve the results. Figure 2a and 2b shows the host image and reference image (preprocessed image) obtained after median filter.

Fig. 2. a) Original Image b) Reference Image
III. PRELIMNARIES
A. Discrete Wavelet Transform
In two-dimensional DWT, each level of decomposition produces four bands of data denoted by LL, HL, LH, and HH [11]. This process is continued until we get desired number of levels determined by the application is reached. Figure 3 shows one levels of decomposition.
LL HL
LH LL1 HL1
LH1 HH1

Fig. 3 2nd Level Wavelet decomposition
Applying IDWT to LL, HL, LH, and HH, we can get four different frequency’s images that are low frequency image, middle-low frequency image, middle-high frequency image, high frequency image separately.
B. Singular value Decomposition
Singular Value Decomposition (SVD) is one of the popular mathematical tools for analysis of matrices. SVD is use to decompose the image into three matrices [12].
If A is an image of size mn, then SVD of A is shown as SVD (A) =USVT
where U and V are orthogonal matrices whose columns are called as left and right singular vectors respectively. S is a diagonal matrix of singular values i , i=1, 2’n arranged in decreasing order and elements of S matrix are zero except diagonal elements. The SVD has some inherent properties in it which makes it popular to use stated as follows:
1. Singular Values (SVs) are stable i.e. any change to it doesn’t affect the image quality [16].
2. SVs are able to represent inherent algebra properties of digital image.
3. The size of matrices can be square or rectangular in SVD with ease in hardware implement ability [17], [18].
4. SVs are known to be invariant to some common attacks such as JPEG compression, noise addition, low pass filter (LPF), rotation, scaling and cropping.
Also SVD has shown its usefulness in variety of applications including image processing, compression and watermarking.

IV. A NEW HYBRID DIGITAL IMAGE WATERMARKING TECHNIQUE
A new digital image watermarking technique is proposed in which original image is pre processes using median filter function then embedding is done in singular value of second level high frequency band of discrete wavelet transform which shows the robustness under various attacks when compared with existing technique [11].

A. Watermark Embedding Technique
Let A(i,j) represents the host image of size 256256 and W(i,j) represents the watermark image of size 256256.
1. Firstly Pre-process the image using median filter to make the image smooth by removing the noise as shown in figure 2.
2. Perform 2-level discrete wavelet transform on the preprocessed B(i,j) and the watermark image W(i,j).
3. Apply SVD on 2nd level high frequency sub-band of both reference image and watermark image using singular transform.

4. Modify the singular value of reference image with the singular value of watermark image
Where alpha is scaling factor.
5. Obtain the modified High frequency sub band of reference image as .
6. Perform 2nd level inverse discrete wavelet transform using and other three bands to obtain the watermarked image.
B. Watermarking extraction process
The main aim of this proposed scheme is to obtain watermark image similar to original watermark image without any degradation. Proposed technique is semi-blind scheme as only watermarked and watermark image is required for extraction.
1. Perform 2-level discrete wavelet transform on the watermarked image as LLWI, LHWI, HLWI, HHWI
2. Perform singular value transform on the high frequency band on watermarked image, watermark image and reference image as in embedding process in step 4.
3. Extract the singular value of watermark image from high frequency sub-band as
4. Obtain the modified high frequency band using:

5. Apply 2nd inverse discrete wavelet transform using and other three sub-bands to obtain the watermark image.
V. EXPERIMENTAL RESULTS
The performance of proposed watermarking technique is explored using matlabR2010b and various experiments are calculated using different image such as lena, cameraman, peppers and lake of size 256256 as shown in Figure 4. Mandrill image is used as watermark image as shown in Figure 5 b. The watermarked image quality is measured using PSNR (poison signal to noise ratio).
Where PSNR can be defined as

Where MSE is the cumulative square error among original and watermarked image. MSE formula is as follw

In watermark embedding process, strength factor is set to 0.2. For wavelet decomposition of original image, HAAR filter coefficients and 2-level of decomposition are used. Different measures can be used to show the similarity between the original and the extracted singular values. In the proposed algorithm, correlation coefficient is used to measure the similarity between two images.
The proposed scheme was tested against various image processing attacks: Gaussian blur, Gaussian noise, rotation and sharpening. Table I shows the PSNR and CC value after embedding the watermark image. Table II shows the comparison between the existing and proposed technique after the extraction of watermark. Comparison between the proposed technique and existing technique [11] under various attacks is shown in Table III to Table VI. High values of PSNR and CC for the proposed technique under various attacks shows better imperceptibility and robustness.
TABLE I. SHOWS THE PSNR AND CC VALUE AFTER EMBEDDING THE WATERMARK

TABLE IV. PERFORMANCE EVALUATION AFTER GAUSSIAN NOISE

TABLE IV. PERFORMANCE EVALUATION AFTER GAUSSIAN NOISE

TABLE II. PERFORMANCE EVALUATION AFTER EXTRACTION OF WATERMARK
TABLE 2. PERFORMANCE EVALUATION OF DIFFERENT IMAGES BASED ON PSNR & CC TABLE III. PERFORMANCE EVALUATION UNDER ROTATION ATTACK TABLE V. PERFORMANCE EVALUATION AFTER GAUSSIAN BLUR

TABLE VI. PERFORMANCE EVALUATION AFTER SHARPENING THE IMAGE
VI. CONCLUSION
The proposed watermarking algorithm is developed by embedding watermark in singular value of 2nd level high frequency band. Original image is preprocessed using median filter for removing noise and it also makes it robust under various attacks. As this method is semi blind image watermarking so only original watermark and algorithm is required. The PSNR value after embedding and extraction is above 50 and CC is close to 1 for all the 8-bit grayscale images. Images are subjected to four types of attacks, namely rotation, blur, Gaussian noise and sharpening. PSNR for extracted images after attacks is above 46 and CC is near to 0.999. High robustness and imperceptibility are two trade-offs and proposed algorithm achieve this under various attacks.
In this method only 2nd level high frequency band is used, embedding can be done in any of the other three bands also and different size of the images can be taken for data hiding with different alpha value.

REFERENCES
[1] Langelaar, G.C., Setyawan, I., Lagendijk, R.I., 2000. Watermarking digital image and video data. IEEE Signal Process. Magazine (September), 20’46.
[2] Potdar et al., ‘A survey of digital image watermarking techniques’, Proc. 3rd IEEE Int. Conf. on Industrial Informatics, Frontier Technologies for the Future of Industry and Business, pp. 709-716, Perth, WA, Aug. 2005.
[3] Bender W., Gruhl D., Morimoto N., ‘Techniques for data hiding’, Proc. SPIE, vol. 2420, page 40, Feb. 1995.
[4] Weng S.W., Zhao Y., Pan J.S., ‘Reversible watermarking based on improved patchwork algorithm and symmetric modulo operation’, Lecture Notes in Computer Science on Knowledge-based Intelligent Information and Engineering Systems, Springer, Berlin/Heidelberg, vol. 3684, 2002.
[5] Yeo K., Kim H.J., ‘Generalized patchwork algorithm for image watermarking’, Multimedia Systems, Springer, Berlin-Heidelberg, vol. 9, pp. 261-265, 2003.
[6] Liu R, Tan T. An SVD-based watermarking scheme for protecting rightful ownership. IEEE Transactions on Multimedia 2002; 4(1): pp 121-128.
[7] Chang CC, Tsai P, Lin CC. SVD-based digital image watermarking scheme. Pattern Recognition Letters 2005 (26) pp1577-86.
[8] Ganic E, Eskicioglu AM. Robust embedding of visual watermarks using DWT-SVD. Journal of Electronic Imaging 2005;14(4).
[9] Li Q, Yuan C, Zong YZ. Adaptive DWT-SVD domain image watermarking using human visual model. In: Proceedings of IEEE International Conference on Advanced Communication Technology (ICACT-2007), vol. 3; 2007. p. 1947-51.
[10] Bhatnagar Gaurav, Raman Balasubramanian. A new robust reference logo watermarking scheme. Multimedia Tools and Applications 2011;52(2): pp 621-40.
[11] K. Bhagyashri and Joshi M.Y., ‘Robust Image Watermarking based on Singular Value Decomposition and Discrete Wavelet Transform’, Nanded 2010 IEEE.
[12] Maliheh Shirvanian. "Adaptive SVD-Based Digital Image Watermarking",Lecture Notes in Computer Science, 2009.
[13] M. Barni, F. Bartolini, A. Piva, Improved wavelet based watermarking through pixel wise masking, IEEE Transcations on Image Processing 10 (2001) 783’791.
[14] Z. Dawei, C. Guanrong, L. Wenbo, A chaos based robust wavelet domain watermarking algorithm, Chaos, Solitons, and Fractals 22 (2004) 47’54.
[15] A.A. Reddy, B.N. Chatterjii, A new wavelet based logo’watermarking scheme, Pattern Recognition Letters 26 (2005) 1019’1027.
[16] R. Liu, T. Tan, An SVD-based watermarking scheme for protecting rightful ownership, IEEE Transactions on Multimedia 4 (1) (2002) 121’128.
[17] H. C. Andrews and C. L. Patterson, ‘Singular Value Decomposition (SVD) Image Coding,’ IEEE Transactions on Communications, 24(4):pp 425-432, April 1976.
[18] A. Sverdlov, S. Dexter, A. M. Eskicioglu, ‘Robust DCT-SVD domain image watermarking for copyright protection: embedding data in all frequencies,’ EUSIPCO2005, Antalya, Turkey, September 2005.
[19] D.S. Chandra, Digital image watermarking using singular value decomposition, Proceedings of the 45th Midwest Symposiumon Circuits and Systems(MWSCAS’02), vol. 3, 2002, pp. 264’267.
[20] Gaurav Bhatnagar and Balasubramanian Raman ‘A new robust reference watermarking scheme based on DWT-SVD’, 2009.
[21] Praful Saxena, Shanon Garg and Arpita Srivastava, ‘DWT-SVD Semi-Blind Image Watermarking Using High Frequency Band’, 2nd International Conference on Computer Science and Information Technology (ICCSIT’2012) Singapore April 28-29, 2012.
[22] Ramandeep Kaur and Sonika jindal, ‘Semi Blind Image Watermarking Using high Frequency Band Based on DWT-SVD’, Sixth International Conference on Emerging Trends in Engineering and Technology (ICETET 2013) Nagpur, December 2013, pp 19 – 24.
[23] Ramandeep Kaur and Sonika jindal, ‘Robust Digital Image Watermarking in High Frequency Band Using median Filter Function based on DWT-SVD’ in Fourth International Conference on Advanced Computing & Communication Technologies (ACCT-2014) Rohtak, February 2014, pp 47 ‘ 52.