• No results found

Texture Synthesis Oriented Steganography using DFT

N/A
N/A
Protected

Academic year: 2022

Share "Texture Synthesis Oriented Steganography using DFT"

Copied!
8
0
0

Loading.... (view fulltext now)

Full text

(1)

96 International Journal for Modern Trends in Science and Technology

Texture Synthesis Oriented Steganography using DFT

Sk.Karthika

Assistant Professor, Department of CSE, TRP Engineering College, Tamilnadu.

To Cite this Article

Sk.Karthika, “Texture Synthesis Oriented Steganography using DFT”, International Journal for Modern Trends in Science and Technology, Vol. 04, Issue 01, January 2018, pp: 96-103.

Steganography is an art of conveying messages in a secret way such that only the receiver knows the existence of the message and texture synthesis is the process of re sampling a smaller source texture into a texture of bigger size that has texture property same as that of the source texture. In the proposed work, taking advantage of texture synthesis, it is employed in Steganography to generate Stego texture images carrying the secret message. In this work, texture is synthesized in transformed domain, Discrete Fourier Transform (DFT). The Stego texture images are of arbitrary size which adds flexibility to the method. This proposed method also offers a good embedding capability and security. Unlike other existing methods which make use of input image to hide the secret message, this method generates the Stego image. The experimental results prove that the proposed method is produces Stego images ofgood visual imapct and consumes less computation time.

Keywords: Steganography, texture synthesis, DCT, embedding, stego image..

Copyright © 2018 International Journal for Modern Trends in Science and Technology All rights reserved.

I. INTRODUCTION

Steganography is defined as the technique of concealing information within carriers. It is the art and science of writing secret messages in a way that no one, apart from the sender and intended recipient can suspect the existence of the message.

The cover medium can be an image or an audio or a video or a text. The ultimate aim of Steganography is imperceptibility, that is, the embedded message should not be discernible to the human eye. There are two other goals: Embedding capacity and security. Image Steganography can be implemented either in spatial domain or in frequency domain.

Spatial domain based Steganography techniques are vulnerable to attacks and are thus less secure.

Frequency domain based Steganography techniques add an additional layer of security.

Transformations like Discrete Cosine Transform

(DCT), Discrete Fourier Transform (DFT) and Discrete Wavelet Transform (DWT) can be utilized for Steganography.

Texture is a ubiquitous visual occurrence. It can describe a wide assortment of surface characteristics such as plants, terrain, minerals, skin and fur. Textures are commonly engaged when rendering synthetic inTexture synthesis is an alternative way to create textures. Synthetic textures can be made of any size, without visual repetition. There are good methods available for synthesizing textures and some of the important methods include tiling, stochastic texture synthesis, single purpose structured texture synthesis, patch based texture synthesis and pixel based texture synthesis.

In this paper, a patch based texture synthesis based Steganography is proposed. The texture is synthesized in frequency domain. The texture synthesis process weaved into Steganography to ABSTRACT

Available online at: http://www.ijmtst.com/vol4issue1.html

International Journal for Modern Trends in Science and Technology

ISSN: 2455-3778 :: Volume: 04, Issue No: 01, January 2018

(2)

97 International Journal for Modern Trends in Science and Technology conceal secret messages as well as the source

texture.This process also allows us to extract the secret messagesand the source texture from a stegosynthetic texture. The proposed approach offers four advantages. First, payload which our scheme offers is proportionalto the size of the stego texture image. Secondly, a steganalyticalgorithm is not likely to defeat this proposed method. Third,the reversible capability allows easy recovery of secret message and source texture.Finally, the computation time consumed by our proposed Steganography system is comparatively less since less number of transformed coefficients are considered for synthesizing texture.

The rest of this paper is organized as follows: In Section II, literature survey is presented.In Section III, we have discussedthe proposed embedding and extracting architectures in detail. The performance measures are described in section IV. We have summarized the experimental resultsSection V, followed by conclusions in the final section.

II. LITERATURESURVEY

Numerous Steganographic methods have been developed in frequency domain. In frequency domain, the pixels of the image are represented as transformed coefficients such that there is no correlation between them. The transformed coefficients of the frequency domain can be obtained using transformations like Discrete Cosine Transformation (DCT), Discrete Fourier Transformation (DFT) and Discrete Wavelet Transformation (DWT).

Chin-Chen Chang et al., [1], suggested a reversible Steganography method using side match vector quantization focusing on high capacity of embedding. In [2], the authors have employed Modified Matrix Encoding (MME) for Steganography with minimal distortion. J. Fridrich et al., in [3], have discussed the dead ends, challenges and opportunities of Steganography with JPEG images.

In [4], a data hiding method combining spread spectrum technique and frequency domain technique is reported. This method is computationally complex. A fine coded and coarsely coded data hiding scheme is presented in [5] and it provides good security when compared to the previously discussed techniques. In [6], the authors have outlined a Steganography system for JPEG2000 compressed images. This system addresses capacity issues.

C.C. Lee et al., in [7], have reported a Steganographic scheme based on Search Order Coding (SOC) and Vector Quantization (VQ). In this

method the capacity was increased to 6-bit per pair of blocks with a good quality Stego image. To solve the issues in large embedding rate with minimal distortion, Weiming Zhang et al., in [8], have suggested a technique that generates binary stego codes for embedding. Wei-Jen Wang et al., in [9], have made a detailed study on various data hiding methods for VQ based images. T. Filler et al., [10], have used Syndrome Trellis Codes (STC) for minimizing the additive distortion. In [11], uniform embedding in all coefficients of transformed domain is adopted to increase embedding rate. On combining the concepts of STC and uniform embedding, LinjieGuo et al., in [12], have suggested a high capacity, high imperceptibility Steganography system.

Texture synthesis has received a lot of concentration recently incomputer vision and graphics [13]. The most recentwork has focused on texture synthesis by example, wherein a source texture image is re sampled using either pixel-based orpatch-based algorithm to produce a new synthesized textureimage.Pixel-based algorithms [14]–[16] generate the synthesizedimage, pixel by pixel and makes use of spatial neighborhood comparisonsto choose the most similar pixel in a sample textureas the output pixel. Patch-based algorithms [17], [18] paste patches from asource texture instead of a pixel to synthesize textures.

Thisapproach of Cohen et al. and Xu et al. improves the imagequality as texture structuresinside the patches are maintained. However, as patchesare pasted with a small overlapped region during the syntheticprocess, an effort must be took to ensure that the patchesagree with their corresponding neighbors.Liang et al. [19] have reporteda patch-based samplingstrategy and have used the feathering approach for the overlappedareas of adjacent patches.

Ni et al. [20] reported an image reversible data hidingalgorithm that has a capability to recover the cover image without anydistortion from the stego image after extracting the hidden data. Histogram shifting is a preferred techniqueamong existing approaches of reversible image data hidingsince it controls the modification on pixels, thereforereducingthe embedding distortion.

Recently Kuo – Chen Wu et al., [21] have reported a texture synthesis based Steganography, where message oriented texture synthesis is performed.

The authors have used the original pixels to perform patch based message oriented texture synthesis.

This method is reversible and is secure against certain geometric attacks. It also produce varying

(3)

98 International Journal for Modern Trends in Science and Technology sized stego synthetic textures, but this method is

computationally complex as it involves computation of each pixel to obtain the stego synthetic texture.

Motivated by the fact that the existing methodologies do not focus on all steganographic issues and the fact that they are vulnerable to steganalytic algorithms, in this paper we propose a DFT based texture synthesis oriented Steganography that focuses on all Steganography issues and resistant to steganalytic algorithms.

Here, we present our work thattakes advantage of the patch-based methods to embed asecret message during the synthesizing procedure thus allowingthe source texture to be recovered in the message extractionprocedure, providing reversibility.The detailed description of the proposed method is given in the next section.

III.PROPOSEDSTEGANOGRAPHYMETHOD 3.1 Architecture

The source texture image serves as the input to our Steganography method. The basic unit of the proposed steganographic texture synthesis is referred to as a “patch.” A patch is an image block of a source texture whose size is user-specified. Once the source texture image is split into non overlapping blocks called patches, they are weaved into an empty image called as pre – stego image. The size of the pre – stego image is chosen based on the

size of the secret message to be embedded. The location of each patch in the pre stego image is recorded in an index table which will serve as the key for the receiver to extract the secret message in proper order. Next, candidate patches are generated from the source texture image. Candidate patches are those which are generated by scanning the source texture image with a fixed window size along the scan line by moving through each pixel after which DFT is applied on the source patches and candidate patches. Euclidean distance is computed between the diagonal coefficient values of each patch and diagonal coefficient values of all the candidate patches. Followed by calculating the Euclidean distances, we rank them according to their distance such that the candidate patch with minimum distance will be ranked first. The final task is to synthesize the empty area in the pre stego image around the source patches that are already weaved. Message oriented texture synthesis is employed to fill the empty areas around the source patches. The candidate patch having a rank that is equivalent to the decimal value of the characters in the secret message is embedded in the area that is to be synthesized. The steps are repeated until all the empty blocks in pre stego image are filled. The final output will be a stego synthetic texture image carrying the secret message. The architecture of the proposed DCT based texture synthesis oriented Steganography is presented in Fig 3.1.

Figure 3.1 Architecture for embedding in the proposed Steganography system

Source texture image

Non overlapping source patches

Candidate patches

Pre stego image Index table

DFT Euclidean distance of diagonal values

Rank the distances

Texture synthesis Stego texture image Secret message

(4)

99 International Journal for Modern Trends in Science and Technology For extracting the secret message, first the source

patches are retrieved from the stego synthetic texture image with the help of the index table that is shared between the sender and receiver. Then, the source patches are grouped to form the source texture image. Once the source patches are removed from the stego texture image, the resultant image will be the pre stego image. Next candidate patches are generated from the source texture after which DFT is applied on the source patches, candidate patches and the patches on the pre stego image. Following this, Euclidean distance is

computed between diagonal coefficient values of source patches and diagonal coefficient values of all the candidate patches. Now the distances are ranked. The next step is to compare diagonal values of the patches in the stego image with the diagonal values of candidate patches. The rank of the matching candidate patch is the decimal equivalent of the secret message. This process is repeated for all the remaining patches in the pre stego image. Fig 3.2 shows the architecture for extraction in the proposed method.

Figure 3.2 Architecture for extraction in the proposed Steganography system

Stego texture image Source patches

Source texture Candidate patch

Pre stego image

Index table

DCT

Euclidean distance Ranking

Comparison and matching

Secret message

(5)

100 International Journal for Modern Trends in Science and Technology 3.2 Message Embedding

The detailed description of message embedding process of the proposed method is given in this section. The secret message and the source texture image are the input to the embedding process. The first step is to split the source texture into non overlapping blocks of fixed size and the blocks are the source patches. Let STw, STh denote the width and height of source texture and SPw, SPhdenote the width and height of the source patches respectively. The number of source patches obtained from the source texture image is computed from equation (1).

Number of source patches = 𝑆𝑇𝑆𝑃𝑤𝑋𝑆𝑇

𝑤𝑋𝑆𝑃(1)

Third an empty image called as pre stego image is generated. The size of the pre stego image is chosen depending on the size of the secret message. The number of embeddable blocks in the pre stego image is determined from equation (2).

Number of embeddable blocks = Number of blocks in pre stego image – Number of source patches (2) where number of blocks in pre stego image is, Number of blocks in pre stego image = Number of source patches + Size of the secret message (3) The index table generation is the third step in the embedding process. Index table contains details of locations where the source patches are to be weaved.The source patches are randomly seeded in sparse arrangement so as to increase the security. The fourth step is generation of candidate patches. An arbitrary window size is taken to scan the source texture in scan line order and the resulting patches through scanning are called as candidate patches. The number of candidate patches is computed as,

Number of candidate patches = (STw – WIw + 1) X (STh – WIh +1)(4)

whereWIw and WIh are the width and height of the window taken to scan the source texture.

After generating the candidate patches and source patches, apply DCT on them. The resultant coefficients are highly uncorrelated.

Next, the Euclidean distance between the diagonal values of source patches and all the candidate patches are computed and ranked. Let P1, P2,..,Pn be the diagonal values of source patch and Q1, Q2, ..., Qn be the diagonal values of candidate patch. The formula to calculate

Euclidean distance between a point Pn in source patch and a point Qn in candidate patch is given as,

D(P, Q) =( (𝑄1− 𝑃1)2+ (𝑄2− 𝑃2)2+ ⋯ + (𝑄𝑛− 𝑃𝑛)2)1/2

The candidate patch with minimum distance from the source patch will be ranked first. Followed by this, the candidate patch having a rank equivalent to decimal value of characters in secret message is taken and weaved into the pre stego image. The steps are repeated until all empty areas in pre stego image are synthesized. The final image is the Stego texture image that is ready to be sent to the receiver.

3.3 Message Extraction

To extract secret message from the stego texture image, the first step is to extract the source texture. The source patches are retrieved from the stego texture image with the help of index table that is shared between the sender and receiver.

Since the index table contains information about details of source patches in stego image, it helps in retrieval of source texture. Once the source patches are obtained from the stego image, they are grouped in order to generate the source texture. The image that remains after removing the source patches is the pre stego image that contains the secret message. Next candidate patches are generated from the source texture as described in the previous section. After generating the candidate patches, DFT is applied on the source patches, candidate patches and the patches in pre stego image. As specified in section 3.2, Euclidean distance between the diagonal values of source patches and candidate patches are computed and then ranked. Now, the patches in the pre stego image are taken and their diagonal values are compared with the candidate patches. The rank of the matching candidate patch is the decimal equivalent of the characters in the secret message. These steps are repeated till retrieving the entire message.

IV.PERFORMANCEMEASURE 4.1 Payload Determination

Payload is the amount of bits that can be embedded in the image. The payload capacity of the proposed method can be determined from the formula,

(6)

101 International Journal for Modern Trends in Science and Technology Payload capacity = Bits embedded per patch

x Number of embeddable patches in the stego image (6)

4.2 Quality of the Stego image

As the stego image is generated by the user, its quality can be determined from a measure called as structural similarity (SSIM) index. This measure quantifies the similarity between the pure and Stego textures. The measure between two windows x and y of common size N×N is:

(7) where, is the average of , is the average of , is the variance of , is

the variance of ,

is the covariance of and ,

, are the two variables to stabilize the division with weak denominator, is the dynamic range of the pixel-values, k1 = 0.01and K2 = 0.03 by default.

V.EXPERIMENTSANDRESULTS

The proposed DCT based texture synthesis oriented Steganography system has been texted with more than 200 images. Three such images viz. Grass, rope net and metal of size (64x64) with pixel values in the range 0-255 are presented in Fig. 5.1.

(a) (b) (c) Figure 5.1 Source textures (a) Grass (b) Rope net

(c) Metal

The source texture is partitioned into non overlapping blocks called source patches of size (8x8). Next, an empty image called pre stego image is generated with number of embeddable patches equal to the size of the secret message.The number of embeddable patches can be calculated from equation (2), given in section 3.2. Now, the source patches are weaved into the pre stego image at sparse locations and it is made a note in the index table which serves as a key to the receiver. The index table holds the details of locations of each source patch in the pre stego image.Next candidate patches are generated from the source texture as described in section 3.1.

The number of candidate patches generated from a source texture of specific size can be computed from equation (4) as mentioned in section 3.2.

Followed by this, DFT is applied on the source patches and candidate patches after which the Euclidean distance between the diagonal coefficients of source patch and diagonal coefficients of candidate patch is computed using the formula (5) as mentioned in section 3.2. Then, the distances are ranked such that the candidate patch with minimum distance from the source patch is ranked first. These ranks are now used for message oriented texture synthesis. The candidate patch with a rank equal to the decimal value of the characters of secret message is weaved into the pre stego image. The steps are repeated till all characters of the secret message are embedded. The output image is the stego texture image that has been synthesized. The stego texture images of size (192x192) for the images given in Fig 5.1 are given in Fig 5.2.

(a) (b) (c)

Figure 5.2 Stego texture images generated by the proposed method for the images presented in Fig. 5.1 (a) Grass (b) Rope net (c) Metal

Stego texture images of various sizes viz.

(120x120), (192x192), (240x240), (256x256),

(512x512), (1024x1024)are generated by the proposed DFT based texture synthesis oriented

(7)

102 International Journal for Modern Trends in Science and Technology Steganography with input source texture image of

size (64x64). The payload capacity is same for all stego textures of same size and is computed as described in section 4.1. These various sizes are tested with payload capacities, 5 bits per patch (BPP) and 8 bits per patch. Payload capacities of

805 BPP, 2560 BPP, 4180 BPP, 4800 BPP, 20160 BPP, 81600 BPP with 5 bits being embedded per patch are obtained for stego texture images of size (120x120), (192x192), (240x240), (256x256), (512x512) and (1024x1024). These results are tabulated in Table 5.1.

Table 5.1 Total Payload Capacity with the proposed Steganography method

Input texture size=(64x64), Source patches =64, Candidate patches = 3249 Size of Stego

texture

Count of patches in Stego synthetic texture,

SSTn

Embeddable Patch Count, EPC

Total capacity

5 BPP 8 BPP

120x120 225 161 805 1288

192x192 576 512 2560 4096

240x240 900 836 4180 6688

256x256 1024 960 4800 7680

512x512 4096 4032 20160 32256

1024x1024 16384 16320 81600 130560

The performance of the proposed Steganography method is evaluated by computing quality of the stego texture image that is synthesized. The quality of the stego image is computed as mentioned in section 3.2. The similarity measure will be in the range [-1, 1]. A result of 1 indicates excellent quality, a result in the range 0.5 to 1 indicates a good quality and -1 refers to poor

quality. Structural similarity factor of 0.814, 0.823 and 0.911 are obtained for Grass, Rope net and Metal images respectively with 5 bits per patch. For 8 bits per patch, structural similarity factor of 0.945, 0.839 and 0.910 are observed for Grass, Rope net and Metal images respectively.

Table 5.2 shows the observed results.

Table 5.2 Structural similarity obtained with the proposed Steganography technique

Texture Source Vs. Stego texture with 5 BPP Source Vs. Stego texture with 8 BPP

Grass 0.814 0.945

Rope net 0.823 0.839

Metal 0.911 0.910

(8)

103 International Journal for Modern Trends in Science and Technology It is evident from the results that high payload and

good quality stego texture image is produced by the proposed Steganography system. A maximum payload of 130560 bits is obtained for stego texture image of size (1024x1024) with 8 BPP. Also on an average the quality of the stego image is in the range 0.8 to 0.9 which is a good quality.

VI.CONCLUSION

In this paper, a DCT based texture synthesis oriented Steganography is proposed. This method employs DFT and patch based texture synthesis to generate the stego texture image. Taking a small source texture image as input, the proposed Steganography system produces stego texture images of bigger size carrying the secret message.

This Steganography system addresses important Steganography issues like imperceptibility, payload, security and to some extent robustness.

The quality of the constructed stego texture image is also good with the proposed technique.

REFERENCES

[1] Chin-Chen Chang, Wei-Liang Tai and Chia-Chen Lin,

“A reversible data hiding scheme based on side match vector quantization”, IEEE Transactions on Circuits and Systems for Video Technology, Vol. 14, No. 10, pp. 1301-1308, 2006.

[2] Y. Kim, Z. Duric and D. Richards, “Modified matrix encoding technique for minimal distortion Steganography”, Proceedings 8th Information Hiding Conference, Vol. 4437, pp. 314-327, 2006.

[3] J. Fridrich, T. Penvy and J. Kodovsky, “Statistically undetectable JPEG Steganography: Dead ends challenges and oppurtunities”, Proceedings of 9th ACM Workshop Multimedia Security, pp.3-14, 2007.

[4] Chun-Hsiang Huang, Shang-Chih Chuang and Ja-Ling Wu, “Digital invisible ink data hiding based on Spread spectrum and Quantization techniques”, IEEE Transactions on Multimedia, Vol. 10, No. 4, pp.

557-583, 2008.

[5] Zhiyuan Zhang, Ce Zhu and Yao Zhao, “Two description image coding with Steganography”, IEEE Signal Processing Letters, Vol. 15, pp. 887-890, 2008.

[6] Liang Zhang, Haili Wang and Renbiao Wu, “A high capacity Steganography scheme for JPEG2000 Baseline system”, IEEE Transactions on image Processing, Vol. 18, No. 8, pp. 1797-1803, 2009.

[7] C.C. Lee, W.H. Ku and S.Y. Huang, “A new Steganographic scheme based on vector quantisation and search order coding”, IET Image Processing, Vol.

3, Iss. 4, pp. 243-248, 2009.

[8] Weiming Zhang, Xinpeng Zhang and Shuozhong Wang, “Near optimal codes for information embedding in gray scale signals”, IEEE Transactions

on Information Theory, Vol. 56, No. 3, pp. 1262-1270, 2010.

[9] Wei-Jen Wang, Cheng-Ta Huang and Shiuh-Jengwang, “VQ applications in Steganographic data hiding upon multimedia images”, IEEE Systems Journal, Vol. 5, No. 4, pp.

528-537, 2011.

[10] Thomas Filler, Jan Judas and Jessica Fridrich,

“Minimizing additive distortion in Steganography using syndrome trellis codes”, IEEE Transactions on Information Forensics and Security, Vol. 6, No. 3, pp.

920-935, 2011.

[11] L. Guo, J. Ni and Y.Q. Shi, “An efficient JPEG Steganographic scheme using Uniform embedding”, Proceedings of 4th IEEE International Workshop on Information Forensics Security, pp. 169-174, 2012.

[12] LinjieGuo, Jiangqun Ni and Yun Qing Shi, “Uniform embedding for efficient JPEG Steganography”, IEEE Transactions on Information Forensics and Security, Vol. 9, No. 5, pp. 814-825, 2014.

[13] RegunathanRadhakrishnan, Mehdi Khauazi and NazirMemon, “Data masking: A new approach for steganography”, Journal of VLSI Signal Processing, Springer, Vol. 41, pp. 293-303, 2005.

[14] YangJie, “Algorithm of Image Information Hiding Based on New Anti-Arnold transform and Blending in DCT Domain”, 12 IEEE conference on communication technology, pp. 312-315, 2005.

[15] Jessica Fridrich, “Minimizing the embedding impact in steganography”, Proceedings of ACM multimedia security’06, pp. 89-95, 2006.

[16] T. Pevny´ and J. Fridrich, “Determining the stego algorithm for JPEG images”, IEE Proc.-Inf. Secur., Vol. 153, No. 3, pp. 77-86, 2006.

[17] Zhiyuan Zhang, Ce zhu and Yao zhao, “Two description image coding with steganography”, IEEE Signal Processing Letters, Vol. 15, No. 5, pp.

887-890, 2008.

[18] Liang Zhang, Haili Wang, and Renbiao Wu, “A High-Capacity Steganography Scheme for JPEG2000 Baseline System”, IEEE Transactions on Image Processing, Vol. 18, No. 8, pp.1797-1803, 2009.

[19] A.Nag, S. Biswas, D. Sarkar and P.P. Sarkar, “A novel technique for image steganography based on block DCT and Huffman Encoding”, International Journal Computer Science and Information Technology, Vol.

2, No. 3, pp. 103-112, 2010.

[20] Fangjun Huang, Jiwu Huang and Yun-Qing Shi,

“New Channel Selection Rule for JPEG Steganography”, IEEE Transactions on Information Forensics and Security, Vol. 7, No. 4, pp.1181-1191, 2012.

[21] Kuo – Chen Wu and Chung – Ming Wang,

“Steganography using reversible texture synthesis”, IEEE Transactions on Image Processing, Vol. 24, no.

1, pp. 130-139, 2015.

References

Related documents