A Discrete Wavelet Transform Based Technique For Image Data Hiding
A Discrete Wavelet Transform Based Technique For Image Data Hiding
A Discrete Wavelet Transform Based Technique For Image Data Hiding
C01 March1820,2008,FacultyofEngineering,TantaUniv.,Egypt
ABSTRACT
In this paper, a new image data hiding technique based on Discrete Wavelet Transform (DWT) is proposed. The new technique will be used for hiding a secret image S inside a cover image C using two secret keys to obtain a stego-image G. It shows high robustness against many of image processing operations such as lossy compression, blurring, cropping, median filter, sharpen, and addition of noise. The embedded secret image can be extracted with high visual quality. The stego-image is perceptually similar to the original cover image. The proposed technique does not require the original cover image to extract the embedded secret image. The comparative analysis between the proposed technique and the other existing techniques has shown the superiority of the proposed technique.
Keywords: Information hiding, Steganography, image data hiding, and discrete wavelet transform.
INTRODUCTION
Information hiding has become the focus of research now. This is the art of hiding a message signal in a host signal, such as audio, video, still images and text document without any imperceptible distortion of the host signal. An important subdiscipline of information hiding is steganography. It comes from Greek words literally means 'covered writing'. While cryptography is about protecting the content of messages, steganography is about concealing their very existence [1]. Two other technologies that are closely related to steganography are watermarking and fingerprinting. These technologies are mainly concerned with the protection of intellectual property, so the algorithms have different requirements than steganography [2]. By using steganography, you can embed a secret message inside a piece of unsuspicious information and send it without noticing the existence of the secret message. In image steganography [3-5], the information is hidden exclusively in an image which is called the cover image. After embedding the secret message, the cover image is called the stego-image. To be a useful steganography system, it must provide a method to embed data imperceptibly, and the secret message must be able to convey the meaning after extraction. The basic idea of image data hiding is to hide the secret image under the camouflage of the cover-image. There are, in general, two approaches that can be used for image data hiding. One approach is the spatial domain techniques and the second approach is the transform domain techniques. Spatial domain techniques usually embed the bits of the message directly into the least significant bits (LSBs) of the pixels of the cover image [6-8]. LSB encoding is the simplest steganographic techniques, but the stego-image is sensitive, and not robust to operations such as blurring, cropping, lossy compression, and addition of noise.
25thNATIONALRADIOSCIENCECONFERENCE(NRSC2008)
2
C01 March1820,2008,FacultyofEngineering,TantaUniv.,Egypt
In transform domain techniques [9-12], first transform both of the cover image and the secret image into a set of frequency domain coefficients. The transformation may be either Discrete Cosine Transform (DCT), or Discrete Wavelet Transform (DWT). The significant frequency coefficients of the secret information are then embedded in some chosen less significant frequency coefficients of the cover image. Hence, the hidden information is less invisible and more robust to different image processing operations. Finally the modified frequency coefficients are inverse transformed to construct the stego-image. In this paper, a new wavelet-based image data hiding technique is presented. The proposed technique show high robustness against many of image processing operations such as JPEG (Joint Photographic Experts Group) compression, cropping, blurring, sharpen, median filter, and addition of noise to the stego-image. The embedded secret image can be recognized when extracted. The stego-image has high Peak Signal to Noise Ratio (PSNR) value and it is perceptually similar to the original cover image, which keep the existence of embedded secret image unnoticeable. The proposed technique does not require the original cover image to extract the embedded secret image. The rest of the paper is organized as follows: In the next section, a brief introduction to DWT is presented. The proposed technique is described in section 3. Section 4 shows experimental results and discussion. Finally, the conclusions of this paper are given in section 5.
The transform of a signal is just another form of representing this signal. It does not change the information content present in it. In most Digital Signal Processing (DSP) applications, the frequency content of the signal is very important. The Fourier Transform (FT) is probably the most popular transform used to obtain the frequency spectrum of a signal. But the Fourier Transform is only suitable for stationary signals, i.e., signals whose frequency content does not change with time. The Fourier Transform, while it tells how much of each frequency exists in the signal, it does not tell at which time these frequency components occur. Signals such as image or speech have different characteristics at different space or time, i.e., they are non-stationary. Most of the biological signals too, such as, Electrocardiogram, Electromyography, etc., are non-stationary. To analyze these signals, both frequency and time information are needed simultaneously, i.e., a time-frequency representation of the signal is needed. The Wavelet Transform provides a time-frequency representation of the signal. It uses multi-resolution technique by which different frequencies are analyzed with different resolutions. Discrete Fourier Transform (DFT) and Discrete Cosine Transform (DCT) are full frame transform, and hence any change in the transform coefficients affects the entire image except if DCT is implemented using a block based approach. However DWT has spatial frequency locality, which means if signal is embedded it will affect the image locally [13]. Hence a wavelet transform provides both frequency and spatial description for an image. The forward 2-D discrete wavelet transform can be implemented using a set of up-samplers, down-samplers, and recursive two-channel digital filter banks as shown in Figure 1. There are many available filters, although the most commonly used filters are Haar Wavelet Filters and Daubechies Filters. However in this paper, a linear phase two-channel integer filter bank proposed in [14] is used in the implementation of the proposed technique. Each of these filters decomposes the image into several frequencies. When applying discrete wavelet transform on an image, four different sub-images are obtained as follows: (1) LL: A coarser approximation to the original image containing the overall information about the whole image. It is obtained by applying the low-pass filter on both x and y coordinates. (2) HL and LH: They are obtained by applying the high-pass filter on one coordinate and the low-pass filter on the other coordinate. (3) HH: Shows the high frequency component of the image in the diagonal direction. It is obtained by applying the high-pass filter on both x and y coordinates [15]. Since human eyes are much sensitive to the low frequency part (LL sub-image), LL is the most important component in the reconstruction process.
25thNATIONALRADIOSCIENCECONFERENCE(NRSC2008)
C01 March1820,2008,FacultyofEngineering,TantaUniv.,Egypt
3.1
Let C and S be the cover-image and the secret-image respectively. The stego-image G can be obtained by the following steps: Step 1: Decompose by one level of 2D-DWT each of the cover image C and the secret image S using a linear phase two-channel integer filter bank proposed in [14] for obtaining the four sub-images (CLL1, CLH1, CHL1, and CHH1) and the four sub-images (SLL1, SLH1, SHL1, and SHH1) respectively. Step 2: Each of SLL1, CLL1, and CHL1 are partitioned into blocks of 4x4 pixels and can be represented by: SLL1= {BSi; 1 <= i < ns} CLL1= {BCk1; 1 <= k1 < nc} CHL1= {BHk2; 1 <= k2 < nc} (1) (2) (3)
where BSi, BCk1, and BHk2 represent the ith block in SLL1, the k1th block in CLL1, and the k2th block in CHL1 respectively, ns is the total number of the 4x4 blocks in SLL1 and nc is the total number of the 4x4 blocks in each of CLL1 and CHL1. Step 3: For each block BSi in SLL1, the best matched block BCk1 of minimum error in CLL1 is searched for by using the root mean squared error (RMSE) criteria. The first secret key K1 consists of the addresses k1 of the best matched blocks in CLL1. Step 4: Calculate the error block EBi between BCk1 and BSi as follows: EBi = BCk1 BSi (4) Step 5: For each error block EBi, the best matched block BHk2 in CHL1 is searched for using the RMSE criteria as before and that BHk2 is replaced with the error block EBi. The second secret key K2 consists of the addresses k2 of the best matched blocks in CHL1. Step 6: Repeat the steps 3-5 until all the produced error blocks are embedded in CHL1. Step 7: Apply the 2D-IDWT to the CLL1, CLH1, CHH1 and the modified sub-image CHL1 to obtain the stegoimage. The block diagram of the embedding procedure is shown in Figure 2.
3.2
In our technique, by receiving the stego-image G and the secret keys from the sender, the embedded secret image can be extracted by the following steps: Step 1: Decompose by one level of 2D-DWT the stego-image G using the same filter bank proposed in [14] to obtain the four sub-images (GLL1, GLH1, GHL1, and GHH1).
25thNATIONALRADIOSCIENCECONFERENCE(NRSC2008)
4
C01 March1820,2008,FacultyofEngineering,TantaUniv.,Egypt
Step 2: Extract the block BCk1 from the sub-image GLL1 by using the first secret key K1. Use the second secret key K2 to extract the error block EBi. The secret block BSi can be obtained by: BSi = BCk1 EBi (5) Step 3: Repeat step 2 until all secret blocks are extracted and form the sub-image SLL1. Step 4: Assign each of sub-images SHL1, SLH1, and SHH1 as zeros, and apply the 2D-IDWT to obtain the embedded secret image which is visually similar to the original secret image. Figure 3 shows the block diagram of the extracting procedure.
The proposed DWT-based image data hiding technique is implemented using Matlab. In order to test the proposed technique, two original cover images Lena and Goldhill both of size 512x512 pixels, and two original secret images Airplane, and Einstein both of size 256x256 pixels shown in Figure 4 are used. The image Airplane was embedded into the cover image Lena. Likewise, we embed the image Einstein into the cover image Goldhill. The stego-images and the extracted secret images are shown in Figure 4. The overall experimental results are shown in Table 1 and Table 2. In previous work, Chang et.al [9], proposed an approach for hiding a secret image in the Discrete Cosine Transform (DCT) domain of the cover image. The Peak Signal to Noise Ratio (PSNR) of the stego-image without any attack was about 37.50 dB for secret images of size (256x256), and about 36.20 dB for secret images of size (512x512). In [12], Lin and Ching proposed a wavelet-based image hiding method. The PSNR define value of the stego-image without any attack was about 32.51 dB for secret images of the same size 512x512 as the cover image. In the proposed technique, after the embedding process, the PSNR of the stego-image without any attack is about 54.90 dB for secret images of size (256x256), and about 46.55 dB for secret images of size (512x512). The stego-images in both cases are indistinguishable from the original cover images. The extracted secret images have good visual quality, and perceptually similar to the original secret images. The PSNR of the extracted secret image of size (256x256) is about 31.10 dB which is better than 25.82 dB in [11]. And it is about 34.71 dB for secret images of size (512x512) which is also better than 22.27dB in [12]. Table 3 tabulates the comparison results of the stego-images. Notice that, our PSNR values are better than those in [9], and in [12] for secret images of both sizes. In order to test the robustness of the proposed technique, some image processing operations are applied to the stego-images such as JPEG compression with different Quality Factors (QF), cropping a quarter of the stegoimage, Gaussian blur, smart blur, radial blur, sharpen, median filter, and addition of noise. The PSNR values of the stego-images and the extracted secret images at different image processing operations are shown in Table 4 and Table 5. Figure 5 shows the extracted secret images Airplane, and Einstein at different quality factors (QF) of JPEG compression applied to the stego images Lena and Goldhill respectively. Notice that, the extracted secret images can be extracted with good visual quality even at QF= 0%. These results show that, after applying different image processing operations to the stego-image, a perceptible secret image can still be extracted.
CONCLUSIONS
In this paper, a new image data hiding technique based on discrete wavelet transform (DWT) has been proposed. The stego-image is looking perfectly intact and has high peak signal to noise ratio (PSNR) value. Hence, an unintended observer will not be aware of the very existence of the secret-image. The extracted secret image is perceptually similar to the original secret image. In addition, it will survive if some image processing operations such as JPEG compression, blurring, cropping, median filter, sharpen, and addition of noise are applied to the stego-image. The proposed technique does not require the original cover image to extract the embedded secret image. The comparative analysis between the proposed technique and the other existing techniques has shown the superiority of the proposed technique.
25thNATIONALRADIOSCIENCECONFERENCE(NRSC2008)
C01 March1820,2008,FacultyofEngineering,TantaUniv.,Egypt
REFERENCES
[1] F.A.P. Petitcolas, R. J. Anderson, and M. G. Kuhn, Information Hiding - A Survey, Proceeding of the IEEE, special issue on protection of multimedia content, Vol. 87, No. 7, pp. 1062-1078, July 1999. [2] R. J. Anderson and F.A.P. Petitcolas, "On the Limits of Steganography", IEEE Journal on Selected Areas in Communications, Vol. 16, No. 4, pp. 474-481, 1998. [3] Y.K. Lee and L.H. Chen, High Capacity Image Steganographic Model, Proceeding of IEEE Vision, Image Signal Process, Vol. 147, No. 3, pp. 288-294, 2000. [4] V. Potdar and E. Chang, Grey Level Modification Steganography for Secret Communication, Proceeding of second IEEE, International Conf. Ind. Inform (INDIN2004), Berlin, Germany, 2004. [5] R. Hovank, P. Fori, and D. Levick, Steganography Based on DWT Transform, Available on: www.urel.feec.vutbr.cz/ra2007/archive/ra2006/abstracts/038.pdf [6] R. Z. Wang, C. F. Lin, and J. C. Lin, Image Hiding by Optimal LSB Substitution and Generic Algorithm, Pattern Recognition, Vol. 34, pp. 671-683, 2001. [7] Chin-Chen Chang, , Ju-Yuan Hsiao, and Chi-Shiang Chan, Finding Optimal Least-Significant bit Substitution in Image hiding by Programming Strategy, Pattern Recognition, Vol. 36, pp. 1583-1595, 2003. [8] Chi-Kwong Chan and L.M. Cheng, Hiding Data in Images by Simple LSB substitution, Pattern Recognition, Vol. 37, pp. 469-474, 2004. [9] Chin-Chen Chang, Chi-Lung Chiang, and Ju-Yuan Hsiao, "A DCT-domain System for Hiding Fractal Compressed Images", In AINA'05, Proceeding of the 19th International Conf. on Advanced Information Networking and Applications, 2005. [10] J.J. Chae and B.S. Manjunath,"A Robust Embedded Data from Wavelet Coefficients", In SPIE'98, Proceedings of the SPIE International Conference on Storage and Retrieval for Image and Video Databases VI, San Jose, CA, 1998. [11] Y.Y. Chung and Y. Sun, "High Capacity Image Steganographic System using wavelet zerotree", Transactions on Engineering, Computing and Technology, pp. 1305-5313, 2004. [12] Chih-Yang Lin and Yu-Tai Ching, A Robust Image Hiding Method Using Wavelet Technique, Journal of Information Science and Engineering, Vol. 22, pp. 163-174, 2006. [13] V.M. Potdar, S. Han, and E. Chang, A Survey of Digital Image Watermarking Techniques, 3rd IEEE International Conference on Industrial Informatics (INDIN), 2005. [14] A. Abdelwahab, E. Saad, and S. Gaber, A New Image Compression Embedded Vector Quantizer, Proceeding of the IEEE European conference on Circuit Theory and Design (ECCTD03), September 2003. [15] K. Wang, Region-Based Three-Dimensional Wavelet Transform Coding, Thesis submitted to the Faculty of Graduate Studies and Research in partial fulfillment of the requirements for the degree of Master of Applied Science, Carleton University, Ottawa, Canada, May 2005.
25thNATIONALRADIOSCIENCECONFERENCE(NRSC2008)
6
C01 March1820,2008,FacultyofEngineering,TantaUniv.,Egypt
CHL1
CHH1
EBi
IDWT
Secret Keys
SLL1
CLL1
CHL1
DWT
SLL1
SHL1
SLH1
SHH1
Secret Image
Stego-Image
25thNATIONALRADIOSCIENCECONFERENCE(NRSC2008)
C01 March1820,2008,FacultyofEngineering,TantaUniv.,Egypt
Stego-Image
K1 BCk1
+
GLL1
-
BSi
K2 EBi
SLL1
Secret Keys
GHL1
SLL1
SHL1 IDWT
SLH1
Table 1: The image quality (PSNR) in dB of the cover images after embedding. Secret images (256x256) Airplane Scene Einstein Cameraman Lena Cover images (512x512) Lena 51.29 50.14 54.45 50.24 54.90 Peppers 51.70 50.36 54.55 51.64 54.09 Boat 51.87 50.93 54.21 52.14 54.00 Goldhill 50.93 50.50 52.49 50.63 52.24
25thNATIONALRADIOSCIENCECONFERENCE(NRSC2008)
8
C01 March1820,2008,FacultyofEngineering,TantaUniv.,Egypt
(a)
(b)
(c)
(d)
(f) (e)
(g)
(h)
Figure 4 (a) Original cover image 1: Lena 512x512. (b) Original secret image 1: Airplane 256x256. (c) Original cover image 2: Goldhill 512x512. (d). Original secret image 2: Einstein 256x256. (e) Stego-image Lena PSNR= 51.29 dB. (f) Extracted secret image Airplane PSNR= 28.95 dB. (g) Stego-image Goldhill PSNR= 52.49 dB. (h) Extracted secret image Einstein PSNR= 29.99 dB.
Table 2: The PSNR in dB of the extracted secret images. Cover images (512x512) Lena Peppers Boat Goldhill Extracted secret images (256x256) Scene Einstein Cameraman 28.11 28.13 28.14 28.08 30.00 30.00 30.01 29.99 26.87 26.93 26.90 26.89
Table 3: The comparison results in dB of the stego images. Lena Chang Proposed et.al [9] 53.04 54.90 42.97 37.25 N/A N/A 33.20 35.33 The cover images (512x512) Baboon Airplane Chang Lin & Chang Proposed et.al Ching Proposed et.al [9] [12] [9] 52.67 53.22 33.77 41.58 37.50 37.28 N/A N/A N/A N/A N/A N/A 52.62 53.21 42.27 43.81 35.42 35.22 33.19 35.22
Peppers (256x256) Lena (256x256) Baboon (512x512) Lena (512x512) N/A: Not Available.
25thNATIONALRADIOSCIENCECONFERENCE(NRSC2008)
C01 March1820,2008,FacultyofEngineering,TantaUniv.,Egypt
Table 4: The PSNR in dB of the stego-image Lena and extracted Airplane at different image processing operations. Image Processing Operations Images after attack Cropping + Gaussian Blur
11.57
14.08
Table 5: The PSNR in dB of the stego-image Goldhill and extracted Einstein at different image processing operations. Image Processing Operations JPEG compression with QF=40% + Sharpen 29.15 24.56
Radial Blur
22.83
21.33
(a) QF = 75%
(b) QF = 35%
(c) QF = 0%
(d) QF = 75%
(e) QF = 35%
(f) QF = 0%
Figure 5. The extracted secret images Airplane, and Einstein at different quality factors (QF) of JPEG compression applied to the stego images Lena and Goldhill respectively. (a) PSNR=26.9734 dB. (b) PSNR=24.60 dB. (c) PSNR= 22.55 dB. (d) PSNR=27.71 dB. (e) PSNR=25.52 dB. (f) PSNR=24.13 dB.