You are on page 1of 5

2016 International Conference on Circuit, Power and Computing Technologies [ICCPCT]

Reversible Data Hiding Using Texture Synthesis


Approach
P.R.Vignesh Kumar
Department of Computer Science and Engineering
College of Engineering (CEG) Campus, Anna University
Chennai, Tamilnadu, India
prvigneshkumar92@gmail.com

AbstractReversible data hiding refers to embedding secret II. RELATED WORKS


message into a cover and the cover image can be recovered
exactly. In the texture synthesis approach texture image which is A. Pixel Based Algorithms
smaller in size is resampled in order to form a texture image Pixel-based algorithms [1,2,3] compare the pixel which are
which is comparatively of larger size but similar to the original at spatial neighborhood in order to select the pixel which is
one. It helps in concealing the secret information in the image.
most similar in a sample texture as the output pixel. Since the
Patch based texture synthesis pattern preserves the image
quality. By our approach first a lookup table is created denoting
pixels which are synthesized already are used to compute
the location of source patches. Next based on the lookup-table output pixel, any wrong computation will affect eh rest of the
secret messages are placed on a blank workbench. The secret result causing errors.
messages are embedded according to our proposed embedding Otori and Kuriyama [4, 5] proposed the secret messages
algorithm which is robust to any steganalysis attack. The which are to be embedded can be encoded in the form of dotted
reversible capacity inherited from our approach allows the pattern and they can be painted on to a blank workbench. The
receiver to perform second round of steganalysis by using the
remaining pixel values can be coated using pixel based
same source texture which the sender has sent.
approach thus disguising the presence of colored dotted pattern.
Keywordsreversible data hiding; lookup table; stego image; In order to extract the secret message at the receiver side before
applying data-detecting mechanism printout of the stego
I. INTRODUCTION texture image is taken. The embedding capacity provided by
the method of Otori and Kuriyama is based on the number of
The cover medium which is used in most of image
the colored dotted patterns.
reversible data hiding algorithm is an image. Cover image can
hold message only of specific quantity since if more B. Patch-Based Algorithms
information is embedded then it would lead to more distortion.
In this paper, we propose an efficient approach for reversible Patch-based algorithms [6, 7] use patches instead of pixels
data hiding using texture synthesis. in order to synthesize the texture. Since the structure of texture
inside the patches are maintained by Cohen et al. and Xu et al.
Proposed texture synthesis process preserves the image approach the quality of image is improved.
quality as source texture can be recovered exactly as original in
the receiver side and can be employed for second round of Liang et al. [8] used the feathering approach for the
steganalysis. Our proposal has three distinct benefits. First, overlapped pixel region by taking the average of the
approach offers embedding capacity proportional to size of overlapped pixels.
stego synthetic texture. Secondly, since stego image is Efros and Freeman [9] used image quilting approach by
composed of source texture it can resist any steganalytic attack. stitching the overlapped patches. He devised an approach by
Third, source texture can be recovered exactly an original since which the approach finds the source texture and candidate
the proposed approach is reversible. Experimental results have patch which has minimum error tolerance. Thus a boundary
proved that our proposed approach can provide efficient which is optimal between the synthesized and candidate patch
embedding capacities compared to the existing methods. is produced.
The paper is devised in the following manner: in Section II, C. Reversible data hiding Algorithm
we elaborate the techniques which are used to synthesize a
texture. In Section III, we detail algorithms which are used for Ni et al. [10] devised an algorithm which can recover the
embedding and extraction. Experimental findings is stego image without any distortion. Since the texture synthesis
demonstrated in Section IV, followed by brief overview and technique can control the pixel modification, it is an efficient
future enhancements presented in the last section. technique among existing approaches.To the best of our
knowledge, there is hardly any literature that relates reversible
data hiding and patch based texture synthesis.

978-1-5090-1277-0/16/$31.00 2016 IEEE


2016 International Conference on Circuit, Power and Computing Technologies [ICCPCT]

Fig. 1. Reversible Data Hiding using Texture Synthesis

In this proposal, we present our findings that can take that synthetic texture can be accessed and source texture can be
advantage of the patch-based methods in order to disguise easily recovered with the help of lookup table.
secret data during the synthesizing procedure. This allows us to
extract the source texture exactly and thus providing the The main concern of the secret message embedding scheme
reversibility capacity. Our approach is detailed in the next is the embedding capacity. Proposed approach offers relatively
section. higher embedding capacity compared to the existing
approaches.
III. PROPOSED WORK
The overall system that will be developed is shown in Fig1.
Patch is the basic unit which is used in our reversible data The phases in which the system will be developed and
hiding texture synthesis approach. A patch is formed from a individual modules is also shown. The input to the system is
source texture with the size of the patch determined by the the source texture into which the message is to be embedded.
user. Size of the patch is usually determined by its width and Source patch is formed from the source texture by expanding
height. A patch is composed of a core part and an outer part the border pixel values. Lookup Table is generated by
where the core part is termed as the kernel region and the part distributing the source patch id at random positions so that it is
which surrounds the kernel region is termed as the boundary not possible for the intruder to intercept the text. For
region. Initial process is the lookup table generation where embedding the message, input message is xor with the key to
lookup table is produced so that the location of source patch form the cipher text. The cipher text is embedded by replacing
can be recorded. The main purpose of using lookup table is the pixel values as spatial domain is used. If any overlapping of
patches occurs then feathering approach is employed on the
2016 International Conference on Circuit, Power and Computing Technologies [ICCPCT]
overlapped region. The stego-synthetic texture encompassing Step4. Replace randomly chosen lookup table values with
the message is sent to the receiver. The receiver on receiving source patch id
the stego-synthetic texture finds the location of the lookup
table. Then the lookup table is extracted to know the location Step5. Remaining lookup table values will be filled by the
of the source patch-id in the stego synthetic texture. After message embedding procedure.
extracting the source patch from the stego-synthetic texture the C. Message Embedding Procedure
final decrypted message is obtained.
The sender and receiver of our system must agree that
A. Source Patch Formation boththe length of the message and key must be same. If the
At the first step, length of the message is computed and the length of the key is less than that of the message then we add
size of each kernel block is determined. Then source texture is somebogus character to make the length of the key equal to the
divided into fixed number of patches. length of the message. At the very first step in encryption
algorithm, both the message and key are converted into its
Input: Source texture, Message, Source Patch width corresponding ASCII values. The ASCII values of small
characters range from 97 to 122 whereas the ASCII values of
Output: Kernel Blocks capital letters range from 65 to 90. For numbers the ASCII
Algorithm values range from 48 to 57. From the ASCII we obtain the
corresponding binary values of each character. Bitxor of
Step1. Message length is computed. message binary and key binary is done to obtain cipher text in
Step2. Size of each kernel block is determined(Size of binary form. Then normalization of cipher values is done by
source texture/Length of input message). dividing the cipher values by 123 to obtain cipher text in the
range [0, 1]
Step3. The source texture is divided into fixed number of
source patches. Input: Secret Key, Message
Output: Encrypted message
Source patch is formed by expanding the kernel block to a
specified patch width. For that purpose we have to select the Algorithm
kernel blocks to be formed into a source patch. Expand the
kernel blocks to a specified patch width by replicating the Step1. Key length and message length is computed
border pixel values. Step2. If Key length is less than message length then add
Input: Kernel Blocks some bogus characters to key length equal to the
message length
Output: Source Patch
Step3. Key and message is converted into its corresponding
Algorithm ASCII values.
Step1. Select each of the kernel blocks to be formed into a Step4. Convert ASCII values into its corresponding binary
source patch values.
Step2. Expand the kernel blocks to the specified patch Step5. Bitxor operation (Secret Key bits, Message Bits) is
width by replicating the pixel values. performed
Step3. Discard the kernel blocks if any of the row or Step6. Convert cipher back to decimal by taking 8 bits at a
column index value is zero time
B. Lookup Table Generation Step7. Normalize the cipher values to a range [0, 1] by
dividing the values by 123.
In lookup table generation we produce a lookup table so
that source patch location in synthetic texture can be recorded. Input Source Patch is taken and applied transposition
By using the lookup table synthetic texture can be easily function to take transpose of the image. Then the source patch
accessed and source texture can be completely recovered. is converted from RGB to HSI. The I-plane of the image is
Such a reversible embedding style reveals one of the major taken and the message is embedded.
benefits our proposed algorithm offers.
Input: Source Patch, Secret data, Secret key
Input: Source Patch
Output: Stego Source Patch
Output: Lookup table with source patch id
Algorithm
Algorithm
Step1. Secret data is encrypted using Encryption algorithm
Step1. Lookup table dimension based on the number of to get cipher text normalized in [0, 1].
source patches is determined
Step2. Apply the transposition function to transpose and get
Step2. Initialize the lookup table to some constant value. the transposed image
Step3. Some random lookup table values is chosen.
2016 International Conference on Circuit, Power and Computing Technologies [ICCPCT]
Step3. Transform the image from RGB to HSI and Variance of modified image
separate the I-plane. Covariance of reference image*modified image
Step4. Since I-plane is achromatic plane secret message c1, c2two variables to stabilize the division with weak
will be embedded in that plane denominator
Step5. Using lookup table, the block which holds the
message is embedded into its corresponding
image block with the help of lookup table. 0.72
Structural
Step6. whilecounter <=size of cipher do 0.7 SIMilarity
a. Consider a pixel Ij (x, y) 0.68 Index(SSIM)
(Existing
b.Replace the pixel with the cipher text normalized 0.66 Method)
value
0.64
c. counter=counter +1;
0.62
end
0 10 20 30 40
Step7. Re-rotate the sub-images and combine it to form
stego I-plane.
Step8. Combine Hplane, Splane, and Iplane Fig. 2. Computed Resemblace values

Step9. Resulting HSI image is converted back to RGB,


From Fig 2. it is found that proposed method has better
and transposition function is applied again in order
similarity values when compared to the existing method.
toget the final stego image.
When the length of input message increases, then similarity
D. Patch Composition Process values will naturally tend to decrease.
Source patch holding the message is combined to form a
stego source patch using the Lookup Table. Feathering B. Mean Square Error (MSE)
approach is employed in case of overlapping.
Mean Square Error is computed by taking average of the pixel
Input: Stego source patch, Lookup Table value differences between an original and modified image.

Output: Stego source patch  (2)

Algorithm
where,
Step1. Select the stego source patch according to the lookup r (x, y) reference image
table. m (x, y) modified image
Step2. Place the stego source patch on the workbench. i and j are the pixel position of the MN image.
Step3. If any overlapping occurs, feathering approach is
employed by taking the average of the boundary 85
value pixels.
Mean Square Error(%)

Step4. Finally a relatively large size stego source patch is Mean Square
formed 84.5 Error (Existing
Method) (MSE)
The same procedure is reversed in the receiver side in order to
(%)
obtain the secret message. 84
Mean Square
IV. EXPERIMENTAL RESULTS AND DISCUSSIONS Error (Proposed
A. Structural Similarity (SSIM) 83.5 Method)
0 20 40 (MSE)(%)
To compute the resemblance between two images Structural Input Length
similarity is used
a b c c
 (1)
a b c c Fig. 3. Average Squared Error between refernced and modified image
a Mean of reference image
Variance of reference image
b Mean of modified image
2016 International Conference on Circuit, Power and Computing Technologies [ICCPCT]
From Fig.3. it is found that proposed method has smaller [4] H. Otori and S. Kuriyama, Data-embeddable texture synthesis, in
Proc. of the 8th International Symposium on Smart Graphics, Kyoto,
error values when compared to the existing method. Japan,
[5] H. Otori and S. Kuriyama, Texture synthesis for mobile data
C Peak Signal to Noise Ratio (PSNR) communications, IEEE Comput. Graph. Appl., vol. 29, no. 6, pp. 74-
81
PSNR is the ratio between distorting signal power and [6] A. Efros and T. K. Leung, Texture synthesis by non-parametric
possible value of signal (maximum). sampling, in Proceedings of 7th IEEE International Conference on
Computer Vision, pp. 10331038, Sep. 1999.
[7] Efros and T. K. Leung, Texture synthesis by non-parametric
 (3)
sampling, in Proceedings of 7th IEEE International Conference on
Computer Vision, pp. 10331038, Sep. 1999.
where MSE is the Mean Square Error
[8] L. Liang, C. Liu, Y.-Q. Xu, B. Guo, and H.-Y. Shum, Real-time
MAXf is the Maximum frequency texture synthesis by patch-based sampling, ACM Transactions on
Graph Theory, vol. 20, no. 3, pp. 127150, May 2001.
[9] Efros and W. T. Freeman, Image quilting for texture synthesis and
31 transfer, in Proceedings of 28th Annual Conference on Computer
Graph. Interaction Technology, pp. 341346, Sep. 2011.
PSNR(PSNR) (dB)
PSNR(dB)

30.5 [10] Z. Ni, Y.-Q. Shi, N. Ansari, and W. Su, Reversible data hiding,
Existing System IEEETrans. Circuits Syst. Video Technol., vol. 16, no. 3, pp. 354-
30 362, 2006.

PSNR(PSNR) (dB)
29.5
Proposed
0 20 40 System
Input Length

Fig. 4. Ratio between distorting signal power and maximum signal value

From Fig.4.it is found that proposed method has better


values when compared to the existing method.

V. CONCLUSION AND FUTURE WORK


This paper proposes a reversible data hiding using texture
synthesis. Our proposed method is novel and original source
texture can be retrieved exactly of the original one and can
be employed for second round of steganalysis. One of the
possible future works is to expand our proposed scheme to
transform domain by employing transform domain methods.

ACKNOWLEDGEMENT
The authors would like to thank the reviewers for their
valuable comments and suggestions. We would like to
express gratitude towards Dr. K. S. Easwarakumar for his
assistance in improving the clarity of this article.

REFERENCES

[1] N. F. Johnson and S. Jajodia, Exploring steganography: Seeing


theunseen, Computer, vol. 31, no. 2, pp. 26-34, 1998.
[2] N. Provos and P. Honeyman, Hide and seek: an introduction
tosteganography, Security & Privacy, IEEE, vol. 1, no. 3, pp. 32-44,
2003.
[3] F. A. P. Petitcolas, R. J. Anderson, and M. G. Kuhn, Information
hiding-a survey, Proceedings of the IEEE, vol. 87, no. 7, pp. 1062-
1078, 1999

You might also like