Top Banner
Optik 126 (2015) 3755–3760 Contents lists available at ScienceDirect Optik jo ur nal homepage: www.elsevier.de/ijleo Performance and analysis of high capacity Steganography of color images involving Wavelet Transform Siraj Sidhik * , S.K. Sudheer, V.P. Mahadhevan Pillai Department of Optoelectronics, University of Kerala, 695581 Thiruvananthapuram, India a r t i c l e i n f o Article history: Received 3 September 2014 Accepted 27 August 2015 Keywords: Wavelet Transform Haar Wavelet Peak Signal to Noise Ratio Steganography a b s t r a c t This paper proposes a modified and simple method for high capacity Steganography in case of color images. This algorithm is usually based on the wavelet fusion technique which was not attempted for color images and it proves to be a simplest method than other techniques or algorithms for color images. Even though it uses transform domain technique which only provides robustness against attacks and eavesdropping and it is known that spatial domain techniques usually meet the capacity and quality requirements. In this paper the transform domain technique (Wavelet Transform) is used to attain high capacity along with security and maintains the quality of the cover image which acts as the key feature of this work. This improvement in the capacity–quality tradeoff is analyzed and proved experimentally in this paper and various image quality parameters have been evaluated to prove the above. © 2015 Elsevier GmbH. All rights reserved. 1. Introduction The advancement in modern technology, software and net- working has created major threats in obtaining secured data communication through communication networks [11–13]. This led to the research in the field of data security communication. One method of providing more security to data is information hiding [1]. There are three most important parameters for an information hiding system, known as capacity, imperceptibility and robust- ness. Robustness refers to protection against any manipulation or attack sought out by the hackers or the eavesdropper in the transmitted information. Imperceptibility refers to the fact that the original cover image and the hidden information are indistinguish- able. Capacity refers to the amount of data or in case of image the size of image that can be hidden the cover image. The informa- tion hiding schemes are principally classified into Steganography and Watermarking [2–5], depending on the application. The term Steganography is derived from the Greek word, steganos mean- ing “covered” and graphia meaning “writing” hence called covered writing. In Steganography systems, our main aim is to provide more capacity, security and maintain the quality of the cover image in which the payload has been introduced. It usually deals with trans- mission of such hidden data along the communication network such that the hidden messages appear to be undetectable to the eavesdropper who will be putting all his efforts to get through it. * Corresponding author. E-mail address: [email protected] (S. Sidhik). The hidden messages include plain text, cipher text, images and videos. It is known that the capacity requirements are met by the spatial domain techniques, that we call as the time domain tech- niques, e.g. gray scale manipulation and Histogram equalization, whereas transform domain techniques, e.g. Wavelet Transform and Discrete Cosine Transform (DCT) are meant for obtaining high robustness against attacks and eavesdroppers. As a result, most of the watermarking algorithms use this transforms domain tech- niques because of the high robustness, while spatial domain hiding methods are more attractive toward the Steganography schemes because of the capacity concerns. Despite of the basic trend due to increased number of compressed images in the internet and mul- timedia communication systems researchers are trying to include the capacity requirement in the transform domain, thus attaining both capacity and robustness against attacks at the same time itself. In this paper we are using the wavelet domain techniques, i.e. Discrete Wavelet Transform (DWT) because it is having lots of advantages compared to other transform techniques like DCT such as progressive and low bit-rate transmission, quality scalability and region-of-interest (ROI) coding demand more efficient and versa- tile image coding that can be exploited for both image compression and watermarking applications and also it is compatible to Human Visual System (HVS) that provides proper perception quality. As a reference the recent compression technique JPEG2000 is also based on this Discrete Wavelet Transform technique. Hence it is best to consider this DWT technique for Steganography process. This paper has been organized as follows. Section 2 entails the proposed algorithm. Section 3 presents the analysis and design. http://dx.doi.org/10.1016/j.ijleo.2015.08.208 0030-4026/© 2015 Elsevier GmbH. All rights reserved.
6

color steganography

Feb 20, 2016

Download

Documents

BRIJESH KUMAR

novice technique for color steganography
Welcome message from author
This document is posted to help you gain knowledge. Please leave a comment to let me know what you think about it! Share it to your friends and learn new things together.
Transcript
Page 1: color steganography

Optik 126 (2015) 3755–3760

Contents lists available at ScienceDirect

Optik

jo ur nal homepage: www.elsev ier .de / i j leo

Performance and analysis of high capacity Steganography of colorimages involving Wavelet Transform

Siraj Sidhik ∗, S.K. Sudheer, V.P. Mahadhevan PillaiDepartment of Optoelectronics, University of Kerala, 695581 Thiruvananthapuram, India

a r t i c l e i n f o

Article history:Received 3 September 2014Accepted 27 August 2015

Keywords:Wavelet TransformHaar WaveletPeak Signal to Noise RatioSteganography

a b s t r a c t

This paper proposes a modified and simple method for high capacity Steganography in case of colorimages. This algorithm is usually based on the wavelet fusion technique which was not attempted forcolor images and it proves to be a simplest method than other techniques or algorithms for color images.Even though it uses transform domain technique which only provides robustness against attacks andeavesdropping and it is known that spatial domain techniques usually meet the capacity and qualityrequirements. In this paper the transform domain technique (Wavelet Transform) is used to attain highcapacity along with security and maintains the quality of the cover image which acts as the key featureof this work. This improvement in the capacity–quality tradeoff is analyzed and proved experimentallyin this paper and various image quality parameters have been evaluated to prove the above.

© 2015 Elsevier GmbH. All rights reserved.

1. Introduction

The advancement in modern technology, software and net-working has created major threats in obtaining secured datacommunication through communication networks [11–13]. Thisled to the research in the field of data security communication. Onemethod of providing more security to data is information hiding[1]. There are three most important parameters for an informationhiding system, known as capacity, imperceptibility and robust-ness. Robustness refers to protection against any manipulationor attack sought out by the hackers or the eavesdropper in thetransmitted information. Imperceptibility refers to the fact that theoriginal cover image and the hidden information are indistinguish-able. Capacity refers to the amount of data or in case of image thesize of image that can be hidden the cover image. The informa-tion hiding schemes are principally classified into Steganographyand Watermarking [2–5], depending on the application. The termSteganography is derived from the Greek word, steganos mean-ing “covered” and graphia meaning “writing” hence called coveredwriting. In Steganography systems, our main aim is to provide morecapacity, security and maintain the quality of the cover image inwhich the payload has been introduced. It usually deals with trans-mission of such hidden data along the communication networksuch that the hidden messages appear to be undetectable to theeavesdropper who will be putting all his efforts to get through it.

∗ Corresponding author.E-mail address: [email protected] (S. Sidhik).

The hidden messages include plain text, cipher text, images andvideos.

It is known that the capacity requirements are met by thespatial domain techniques, that we call as the time domain tech-niques, e.g. gray scale manipulation and Histogram equalization,whereas transform domain techniques, e.g. Wavelet Transformand Discrete Cosine Transform (DCT) are meant for obtaining highrobustness against attacks and eavesdroppers. As a result, mostof the watermarking algorithms use this transforms domain tech-niques because of the high robustness, while spatial domain hidingmethods are more attractive toward the Steganography schemesbecause of the capacity concerns. Despite of the basic trend due toincreased number of compressed images in the internet and mul-timedia communication systems researchers are trying to includethe capacity requirement in the transform domain, thus attainingboth capacity and robustness against attacks at the same time itself.

In this paper we are using the wavelet domain techniques, i.e.Discrete Wavelet Transform (DWT) because it is having lots ofadvantages compared to other transform techniques like DCT suchas progressive and low bit-rate transmission, quality scalability andregion-of-interest (ROI) coding demand more efficient and versa-tile image coding that can be exploited for both image compressionand watermarking applications and also it is compatible to HumanVisual System (HVS) that provides proper perception quality. As areference the recent compression technique JPEG2000 is also basedon this Discrete Wavelet Transform technique. Hence it is best toconsider this DWT technique for Steganography process.

This paper has been organized as follows. Section 2 entails theproposed algorithm. Section 3 presents the analysis and design.

http://dx.doi.org/10.1016/j.ijleo.2015.08.2080030-4026/© 2015 Elsevier GmbH. All rights reserved.

Page 2: color steganography

3756 S. Sidhik et al. / Optik 126 (2015) 3755–3760

Table 1Wavelet families in MATLAB.

Wavelet families Wavelets (MATLAB notation)

Daubechies [6,7] db1 or haar,db2,. . .,db10,. . .,db45Coiflets coif1,. . .,coif5Symlets sym2,. . .,sym8,. . .,sym45Discrete Meyer DmeyBiorthogonal bior1.1,bior1.5,bior2.2,bior2.4,bior2.6Reverse Biorthogonal rbio1.1,rbio1.3,rbio1.5,rbio2.2,rbio2.4

Hp_

Hp_

1 2

1 2

1 2

1 2

Lp_

Lp_

Hp_

Lp_

2 1

2 1

LL

HL

LH

HH

Fig. 1. Filter bank algorithm.

Section 4 covers the results and discussion. Section 5 provides uswith conclusion.

2. Proposed algorithm

2.1. Wavelet Transform

Wavelet Transform is a technique that is used to convert thespatial domain into frequency domain and thus separates the highfrequency and low frequency component of an image. In this casewe usually use the filter bank algorithm in which first of all theimage is convolved with the High Pass Filter and a Low Pass Filterthat gives the high pass component and the low pass componentpixel by pixel, further, each of the low frequency and the high fre-quency component is again convolved with the Low Pass Filter andHigh Pass Filters which provides us with the 2D Wavelet Transformof the image.

Of all the wavelets used Haar Wavelet is considered to be thesimplest to implement the filter bank algorithm used for separatingdifferent frequency components which is given by,

˚(t) =

⎧⎨⎩

1 if 0 ≤ t ≤ 1/2

1 if 1/2 ≤ t ≤ 1

0, otherwise

By using Haar Wavelet, low frequency components are obtainedby taking the average of the pixel values in the image providedwhereas the high frequency coefficients are obtained by taking halfof the difference of the pixel values of the image. Researchers haveshown that in human perception, the retina of the eye splits theimage into number of frequency channels having equal bandwidthwhich is similar to that of the multilevel decomposition and it isusually sensitive to only the low frequency components and notto the high frequency components. Because of this reason onlyWavelet Transforms are used in this case for further operations.Other wavelets that are also considered here for comparison areshown in Table 1.

Here in this scheme Wavelet Transform of the cover imageand payload image is taken using the Haar Wavelet and the pay-load is concealed into the cover image using wavelet fusion. Asshown in Fig. 1 the two-dimensional Wavelet Transform decom-poses the color images into four bands the LL, HL, LH and HHband which represents the low pass approximation, vertical,

Fig. 2. 2D Wavelet Transform.

STEGO IMAGE

Color Cover Image Color Payload Image

Separate RGB Separate RG B

Normalize RGB Normalize RG B

DWT on RGB DWT on RGB

Wavelet Fusion of RGB

X

Combine the RGB

IDWT

Fig. 3. Encoder block diagram.

horizontal and diagonal detail coefficients of the color cover image.The LL band gives the soft approximation of the image which is verymuch sensitive to Human Visual System whereas the HL, LH andHH bands provide us with the details of the image. This process canbe repeated for LL band. Fig. 1 shows the 2D wavelet decomposi-tion of an image. Here we are using the Integer Wavelet Transformintroduced in [5] (Fig. 2).

2.2. High capacity and security Steganography using DWT

The idea behind the proposed algorithm is the wavelet basedaddition or fusion. It involves adding of the wavelet decompositionmatrix of the normalized version of all the three (RGB) components(Red, Blue and Green) of the cover image and the payload into asingle fused result and then combining them to form the originalcolor image for analysis.

2.2.1. HCSSD encoderThe block diagram is shown in Fig. 3. In this first of all, we are

separating the Red, Blue and Green components of both the coverand the payload image that is provided with. After that each ofthe Red, Blue and Green components are normalized to obtain thepixel ranging from (0,1) instead of (0,255). After that we are usuallyperforming the Wavelet Transform using Haar Wavelet on each ofthe RGB components of the cover image and the payload image.Hence we are getting the two matrices book keeping matrix andthe decomposition matrix. We are considering the decomposition

Page 3: color steganography

S. Sidhik et al. / Optik 126 (2015) 3755–3760 3757

PAYLOAD IMAG E

Color Cover ImageStego Image

Separate RGB Components

Separate RG B Components

Normalize RGB Components

Normalize RGB Components

DWT on RGB Components

DWT on RGB Components

Wavelet Diff erence of RGB Components

Combine t he RGB Components of Payload

Fig. 4. Decoder block diagram.

matrix for the wavelet fusion process. Finally, the single combinedresultant matrix of the stego image is obtained by the additionof decomposition matrix of the respective sub-bands, i.e. LL, HL,HH and LH of respective components RGB of the cover image andpayload which is given by the following equation:

M(x, y) = C(x, y) + ˛P(x, y) (1)

where M is modified DWT coefficients of each of the componentsRGB, C is the original DWT coefficients of each component of thecover image and P is the DWT coefficients of each component ofthe payload image and ̨ is chosen in such a way that the payloadimage is not visible in the stego image.

2.2.2. HCSSD decoderThe stego image is separated into three components RGB, nor-

malized each of the components, and then DWT is taken for eachof them. The extraction process involves subtracting the DWTcoefficients of the original components (RGB) of the cover imagefrom the DWT coefficients of the RGB components of the stegoimage. Then IDWT on these components is applied to get the Red,Blue and Green components, which are then fused together to formthe payload image. In this work different wavelets are compared(Fig. 4).

2.3. Algorithm

The embedding and extraction algorithms have been givenbelow in detail:

Embedding

1. Separate Red, Blue and Green components of both the cover andpayload color images.

2. Normalize all the three components in both cases.3. Preprocessing is done on both the group of image components.4. Wavelet fusion of DWT coefficients of consequent components

of cover image and payload image.5. Compute IDWT of all the fused Red, Blue and Green components.

6. Combine the Red, Blue and Green components thus obtained toform the fused color image.

Extraction

1. Separate Red, Blue and Green components of both the cover andstego color images.

2. Normalize all the three components in both cases.3. Preprocessing is done on both the group of image components.4. Compute 2D DWT of RGB components of cover and stego image

using Haar Wavelet.5. Subtract the DWT coefficients of the consequent components

(RGB) of the stego and the cover image.6. Compute IDWT of all the subtracted Red, Blue and Green com-

ponents.7. Combine the Red, Blue and Green components thus obtained to

form the fused color image.

3. Analysis and design

For finding out the quality of the stego image generated, weare employing certain image quality parameters which are givenbelow.

3.1. Signal to Noise Ratio (SNR)

This value gives the quality of reconstructed signal. Higher thevalue, better is the stego image. It is given by:

SNR = 10 log10

(�2

x

�2ε

)(1)

where �x and �ε represent the mean square value of the inputimage and the means square difference between the cover andstego image.

3.2. Peak Signal-to-Noise Ratio (PSNR)

The Peak Signal-to-Noise Ratio [8,9] (PSNR) usually depends onthe Mean Square Error (MSE) which is given by:

MSE = 1mn

m−1∑i=0

n−1∑j=0

[I(i, j) − K(i, j)]2 (2)

where I indicates the input image and K represents the stego image,m and n indicate the number of pixels in the image.

The PSNR is defined as:

PSNR = 10 log10

(MAX2

I

MSE

)(3)

Here MAXI indicates the maximum of the pixel value of theimage and MSE represents the Mean Square Error. Typical valuesfor the PSNR in lossy image and video compression are between 30and 50 dB, where higher is better.

3.3. Weighted Peak Signal-To-Noise Ratio (WPSNR)

The formula for WPSNR is shown below:

WPSNR = 20 log10

(255√

MSE × NVF

)(4)

The formula to calculate this factor as a simplified function is:

NVF = NORM

(1

1 + ı2block

)(5)

Page 4: color steganography

3758 S. Sidhik et al. / Optik 126 (2015) 3755–3760

Table 2Image quality parameters for ̨ = 0.2.

Wavelet MSE PSNR (dB) WPSNR (dB) SC

db1 848.5710 17.6853 17.1024 0.9074coif1 843.47 18.8434 17.2238 0.9047sym2 871.8289 18.7265 18.7229 0.9119Dmey 328.9939 22.9589 21.2058 0.8833bior1.1 1.108e + 003 17.6853 17.1024 0.9074rbio1.1 1.108e + 003 17.6853 17.1024 0.9074Haar 848.5710 17.6853 77.1024 0.9074

where ı represents the luminance variance for the 8 × 8 block ofthe image and NORM represents the normalization function value.Typical values for the WPSNR in lossy image and video compressionshould be greater than 40 which indicate high quality.

3.4. Structural Content (SC)

It is used for measuring the similarity content [9,10] betweentwo images, i.e. the cover image and the stego image. It is formu-lated as

SSIM(x, y) = (2�x�y − c1)(2�xy − c2)

(�2x − �2

y − c1)(�2x − �2

y − c2)(6)

where �x is the average of x, �y is the average of y, �xy is thecovariance of x and y, c1 = (k1L)2, c2 = (k2L)2, k1 = 0.01 and k2 = 0.03by default and L is the dynamic range of pixel values.

3.5. Histogram analysis

Histogram is the graphical representation of the distribution ofdata. It gives the variation of number of pixels with respect to theintensity of the image. Histogram is mainly used for finding outthe threshold for compression and for conversion of color image togray image. Also we can check the variation of the pixel values ofdifferent images obtained from Steganography.

These are standard image parameters that are used for findingout the quality of the stego image to that of the original image.

4. Results and discussion

In this case we are usually choosing the cover image of size(256 × 256) and the payload image of varying size which is greaterthan the cover image is chosen and hiding such a large image (pay-load image) into a smaller image (cover image) without degradingthe quality of the cover image indicates the high capacity of thealgorithm and also the quality–capacity tradeoff.

The MATLAB 7.0 has been used for simulations and demonstra-tion of the proposed idea. To evaluate the quality of the stego image,Mean Square Error, PSNR and WPSNR between the stego imageand the original image have been calculated and satisfactory resultswere obtained indicating the better quality of the stego image gen-erated by the algorithm. After that it was possible to successfullyextract the payload image from the stego image as in Figs. 3 and 4without much difficulty.

Case I – First of all we are taking the value of scaling factor ̨ = 0.2 and we are observing the stego image and comparing the

image quality parameters for different wavelet filters in the MAT-LAB (Table 2).

4.1. Histogram analysis

Here we can see that the perception quality of the stego image ispoor as indicated by the various image quality parameters (Fig. 5).

Case II – Next we are taking the value of ̨ = 0.01 and plotting thestego image (Fig. 6).

Fig. 5. MATLAB simulation (a) cover image; (b) payload image; (c) stego image(embedded image); (d) stego image; (e) cover image; (f) payload image (extracted).

Fig. 6. Histogram plot of cover image.

Fig. 7. Histogram plot of stego image.

Page 5: color steganography

S. Sidhik et al. / Optik 126 (2015) 3755–3760 3759

Fig. 8. MATLAB simulation (a) cover image; (b) payload image; (c) stego image(embedded image); (d) stego image; (e) cover image; (f) payload image (extracted).

Fig. 9. Histogram plot of cover image.

Fig. 10. Histogram of stego image.

Table 3Image quality parameters for ̨ = 0.01.

Wavelet MSE PSNR (dB) WPSNR (dB) SC

db1 12.2523 37.2486 41.0969 0.9941coif1 10.9863 37.7223 41.2287 0.9941sym2 15.7810 36.1495 41.1998 0.9940dmey 1.7278 45.7560 45.9679 0.9934bior1.1 12.2523 37.2486 41.0969 0.9941rbio1.1 12.2523 37.2486 41.0969 0.994haar 12.2523 37.2486 41.0969 0.9941

Table 4Capacity of the image.

Payload size MSE PSNR WPSNR SC

300 × 300 2.5533 44.0597 43.5617 0.9895297 × 596 10.1819 38.0525 38.1605 0.9747500 × 370 20.4494 35.6240 39.5022 0.9903318 × 159 13.8208 36.7255 51.3217 0.9925512 × 512 12.2523 37.2486 41.0969 0.9941500 × 500 10.2623 39.5234 42.1934 0.9904600 × 600 10.2134 37.4587 40.3446 0.9901

Fig. 11. (a) Capacity factor vs MSE; (b) capacity factor vs PSNR; (c) capacity factorvs WPSNR.

Page 6: color steganography

3760 S. Sidhik et al. / Optik 126 (2015) 3755–3760

From the table we can see that the stego image obtained is usu-ally having high quality and capacity requirements are also satisfied(Figs. 7–10, and Table 3).

Case III – Here we are using different sized payload color imagesfor a cover image of 256 × 256 to validate the capacity of the algo-rithm and various quality parameters are also compared for theseimages (Table 4).

Case IV – Next we are usually considering another factor calledas the Capacity Factor for analysis purpose that is obtained by divid-ing the cover image size by the payload image size which indicatesthe increase in size of the payload image with respect to the coverimage. This factor is then plotted against various image qualityparameters, i.e. Mean Square Error, PSNR and WPSNR to study thevariation. Various plots are given in Fig. 11.

5. Conclusion

This paper proposes a modified and simple method for highcapacity Steganography in case of color images. The algorithm isusually based on Discrete Wavelet Transform in which waveletfusion is used for Steganography of color images. As Steganogra-phy systems are considered there are two general requirements,one maximum payload size should be incorporated into the coverimage and the second is it should not be easily perceived by theeavesdropper. Robustness is not considered as main criteria sinceif someone is trying to modify the file, and then it will also affect thehidden bits in Steganography making it of no use for the eavesdrop-per. It has been shown that the three compensating parameterscapacity, security and robustness are maintained by this algorithmwhich is the desired aim of this work. From the two color imagestaken payload image is having size greater than the cover image andthe cover image is having the size of 256 × 256 and this large imageis hidden into this small image which indicates the high capacityof the algorithm and also from the literature the work has beendone on gray but this work is mainly aimed at color images whichis fast as well as a simple technique to be used. Computer simula-tion results have verified the validity and efficiency of the proposedtechnique. Mean Square Error, Peak Signal to Noise Ratio, WeightedPeak Signal to Noise Ratio and Structural Content have been

calculated for ̨ = 0.2 and ̨ = 0.01 in order to support this work andis found that there is much improvement in the quality of stegoimage so that no eavesdropper can usually identify it.

The extraction of the hidden image has been shown in the resultand we are following the reverse process of that used in case ofembedding process and are taking the difference of the waveletcoefficient of the stego image and original image. It is knownfrom the figures that the perception quality of the payload imageextracted is high.

Steganography is an emerging field of research and furtherimprovement can be provided by following other algorithms toattain more capacity, security and robustness.

References

[1] D. Artz, Digital steganography: hiding data within data, IEEE Internet Comput.(2001) 75–80, May–June.

[2] L. Sunil, C.D. Yoo, T. Kalker, Reversible image watermarking based on integerto integer wavelet transform, IEEE Trans. Inform. 2 (2007).

[3] A. Bilgin, J. Sementilli, F. Sheng, W. Marcellin, Scalable image coding usingreversible integer wavelet transform, Comput. J. Image Process. IEEE Trans. 9(4) (2000) 1972.

[4] N. Bi, Q. Sun, D. Huang, Z. Yang, J. Huang, Robust image watermaking basedon multiband wavelets and empirical mode decomposition, IEEE Trans. ImageProcess. (2007), August.

[5] T.G. Gao, Q.L. Gu, Reversible watermarking algorithm based on wavelet liftingscheme, in: Wavelet Analysis and Pattern Recognition Conference, November,2007.

[6] I. Daubechies, The Wavelet Transform, time–frequency localization and signalanalysis, IEEE Trans. Inform. Theory 36 (1990).

[7] L. Deand, Wavelet Transformation and Their Application, Birkhauser, Boston,2002.

[8] S.T. Welstead, Fractal and Wavelet Image Compression Techniques, SPIE Pub-lication, 1991, pp. 155–156.

[9] Q. Huynh-Thu, M. Ghanbari, Scope of validity of PSNR in image/video qualityassessment, Electron. Letter 44 (13) (2008) 800–801.

[10] Z. Wang, A.C. Bovik, H.R. Sheik, E.P. Simoncell, Image quality assessment fromerror visibility to structural similarity, IEEE Trans. Image Process. 13 (4) (2004April) 600–612.

[11] Loza, et al., Structural similarity based object tracking in video sequence, in:Proc. of the 9th International Conference on Information Fusion, 2006.

[12] H. Baraka, H.A. El-Manawy, Attiya, An integrated model for internet securityusing prevention and detection techniques, IEEE J. Comput. Commun. 99 (1998)25–33.

[13] M. Arnold, M. Schumucker, S.D. Wolthusen, Techniques and Applications ofDigital watermarking and Content Protection, Artech House, Boston, London,2003.