International Journal of Computer Applications (0975 – 8887) Volume 21– No.
7, May 2011
Image based Secret Communication using Double Compression
Chaitanya Kommini Kamalesh Ellanti Srinivasulu Asadi Dept of IT, Dept of IT, Dept
of IT, SVEC, Tirupati SVEC, Tirupati SVEC, Tirupati J.N.T.Univeristy, Anantapur J.N.T.Univeristy, Anantapur J.N.T.Univeristy, Anantapur A.P – 517 102, India A.P – 517 102, India A.P – 517 102, India 6 4 4 In the later sections we describe the JPEG quantization process 6 and the process of embedding encrypted data inside an image. 10 Then we give the experimental results. 16 20 2. INFORMATION EMBEDDING AND 24 EXTRACTING SCHEME 5 5 2.1 JPEG image compression 6 8 The first step in image compression is to convert the image from 10 the chrominance RGB format into YCbCr format. Later, 23 channels (Cb and Cr) are subsampled to the rate half of the Y. 24 Now, the partition of each channel into 8x8 non-overlap blocks 22 takes place and the shifting of pixel values in the channel from 6 [0-255] to [-128,127] is done to move to next step i.e., Discrete 5 Cosine Transformation (DCT) . In 6 DCT step, the lowthe frequency and the high frequency coefficients are separated and 10 it allows the truncating of high-frequency coefficients by applying specific quantization tables. 16larger coefficients set A 23 will lead to high compression ratio and reduced image quality. 28 There is a chance of different camera models having same image quality but different QTs. Examples of such QTs can be seen in 22 . 6 7 The QT of Photoshop and Matlab in fig 1 and fig 2 are different 9 but they have same quality factor 80. 12 20 35 32 25 7 9 15 22 27 44 41 31 10 14 22 26 32 42 45 37 20 26 31 35 41 48 48 40 29 37
The Information exchange via any media needs privacy and secrecy. Cryptography is widely used for providing privacy and secrecy between the sender and receiver. But, now, along with Cryptography, we are using Steganography to have more protection to our hidden data. In this paper, we show how a JPEG can be used as an embedding space for a message by adjusting the values in the JPEG Quantization tables (QTs). This scheme also uses some permutation algorithms and it can be widely used for secret communication. This JPEG double compression will give satisfactory decoded results.
GENERAL TERMS INDEX TERMS
Image Processing, Signal Processing, Steganography.
Quantization Tables, JPEG Compression, Steganography, Double Compression.
There is a great demand for speed and integrity of the information transfer on the Internet. In addition to the speed and integrity, there is a great need for secrecy and privacy in the information exchange. Many research works have used cryptography to provide secure communication. But besides cryptography, Steganography is also a widely used technique for providing more protection to the hidden data. Using this technique the data hiding is done in such a way that it avoids the people to even think that there exists information in the carrier media. Data hiding inside an image can be done using many domains where Steganography algorithms exploit like spatial domain or DCT. Images are also of various types but we take the JPEG format because it is the commonly used standard with 30:1 compression ratio and very less loss in image quality. Other reasons for JPEG standard are that we can adjust the degree of compression. In this paper, data is protected in two stages. The first stage is encryption stage where the message6 to be embedded is 4 encrypted using certain encryption algorithms. In this stage, permutation algorithms are used and 3 they require a pair of 6 numbers as a key. After this stage, the encrypted message is embedded in JPEG image by managing the values in the JPEG 9 Quantization tables (QTs). Thus from15 hiding stage we get this the JPEG image. This image has some regions with different 19 image quality. This whole process is done at the sender side. At 22 the receiver end, all the above steps are performed in the reverse 4 order to extract the original data i.e., first we extract the scrambled message in the image and 4 later using the key we 5 perform the decryption.
7 9 21 22 20 5 4 6
15 21 25 21 5 6 8 10 19 32 30 23 6 8 13 21 25 40 38 28 9 13 20 24 30 29 42 34 18 24 29 32 38 45 45 37 27 34 35 36 42 37 38 37
39 45 40 41 40
Fig 1: QT of Photoshop
D e cr y p ti o n
2. each DCT coefficient is quantized to corresponding value in QT.2 The main principle in the proposed Message hiding scheme
Suppose consider the case where a block of 8x8 DCT coefficients (let cij be c11 to c88) be mapped to a QT of same size.
Cij= round (Cij/qij) (1)
. Here. The quantized value is rounded to an integer which is nearest to the value.
some regions which To embed 25 times (g) Extracted embedded pattern. we be around 40dB if the difference Q2.m)=1. 95
90 85 80 75 70 65 60 (a) (b) (c) (d) Fig 2: QT of Matlab 55 90
In the fig 5. we have to know factors taken are and secret image with size 64x64. are compressed with lower quality q0. the difference between C2 and C1 is minimum and this feature is made clear in . (f) ng D Difference image after scaling 25 times.7 and s2 is taken as threshold. The quality factor of the saved image must be high i. This process is which determines whether the steganographic content is present and quality Step 1: The embedded image is resaved with lower it also checks whether the hidden information is embedded using the system. (e) Embedded image.40) and stage (2) Embedding (e2. in the fig 3. From the the Host image.30) andregion ‟)=(1. difference between the resaved image and original image is small and hence the difference image will be very dark and blurred.7 41. (g) Extracted e patterns.k1)=(1. Let us take the permutation parameters Q1 Q2 and k=83 and for as e=2 decoding we need to take the parameters as e=2 and k=109. 66 times (k=66) and this between Q1 and Q2 is around 10 to for 42. The resaved message takes place. The key CDR was easily be extracted from Table II. Q2 must be very small and hidden message takes place.7 41.25 dB. the permutation algorithm factor of order is applied hidden We manage the qualityin reverse each pixel in the with image parameters (e2. first the receiver has to detect the 39. the image is saved in JPEG format. After scaling the difference image.2 This embedding has two steps: Step 1: The original image is divided into blocks of size P/L x Q/M.. the difference (Q1-Q2) vs. In the receiver side.
40. we apply an pixels in the secret image gives the CDR value. 43. O(x.5 84. If the difference is very small then the CDR cannot be 100% Formost case consider thatthe poor contrast of blackdifference in example. wavelet coefficients over Step 2: In the second step we have to subtract the above resaved 85 N subbands are used.
95.5 43.8dB and different quality factors so that the for getting the correct withCDR is 94.3 The aspect ratio of secret image should be same as that of host 41.m)=0 for black pixels and C(l. Step 2: If C(l. Stegdetect is the scheme as follows: Farid‟s from the Host image. The output embedded JPEG images are tested in order to evaluate the capability of evade steganalyser The schemes.2 Q1 AND Q2 41. with C(l.5 steghide is very impressive.4 43. The ratio of to assess the performance Fig number of pixels decoded embedding and extractingof total the 3 shows the steps in the correctly and the number of the message in the proposed scheme. then the difference in (2) will be a minimum as q2=q1 (only when we take difference in (2) in terms of (e) (f) (g) function of q2). (a) Original image. Let the original image be of size 1024x1024 and the secret message is of size 128x128. we can observe thewhich with different pairs of Q1 and Q2.e. get and we embed it in )=(2. The Host imagePSNRof size WITH DIFFERENT SETS OF 75 41.1
. at least 95.e. is as follows: Let the recipient‟s received image be I1.0 If C(l.6 41. factor Q2. so that we Let us the expected secret message C be a binary image of size LxM consider our result. The PSNR value will have applied modulo operator with 80 parameter e=2 on the original image15.1 84.y) is compressed withQ1. the PSNR calculated for embedded image is regions decoding process.m)=0. (b) Secret message. The selectable difference between shared with the sender will be used to rescramble the extracted Q1 and Later the reconstruction of range is about 20 to pattern. TABLE I is O (DB) PxQ.k2)=(2.7.4%.8).2 40. So for recovering the image in the receiver side. the automorphism taken are Q1=95 the image where the permutation quality factors algorithm  to andQ2=80 and PSNR value is of pixels in the hiddenimage is scaled 25 times. 40.2 41.2 should o correctly selected while maintaining the quality of the be 78. 100
95 90 85 80 75 70 65 60 95
If we recalculate the DCT set (C1=Cij x qij) and if we quantize 44. n image. because of we have embedded the pixels in lower quality blocks. And that value is the best candidate and the average of these s1 80 92. TABLE II CDR (%) WITH 80.6 84.7 embeddedFig 3: Embedding and extracting message scheme. Here.1 the c1 set again. Subtracting I2 from I1 yields difference image.k).
After embedding.1 81.
y In table III.. May 2011
3.8 gives the image next to the original image in fig 3. Now we calculate s1-s2 value which is the greatest difference in the difference image. EXPERIMENTAL RESULTS
Fig 4 shows the example.3 40. the blocks that (b) (c) (a) were compressed with lower quality factor Q2 will appear as black Q2 Q1 blocks whereas the other blocks will be decoded as white blocks. 42.0
Step 5: We decode the pixel as black if the sum of the pixel value in that block is smaller than the threshold or else we decode it as white.2 are so specific Secret message after decoding. cr
Step 4: We take the greatest sum values as s1 and least sum values as s2 in the difference image.3 40.m)=1 for white pixels. it gives the DCT coefficients set C2. These lower quality regions will carry the secret information. 8x8 blocks. In other words. O(x.International Journal of Computer Applications (0975 – 8887) Volume 21– No. In fig 4. (h)–(i) Decoded secret message.8
Fig 5: Two-round encryption result. For example. By comparing the host image and embedded image and the Correct Decoding Rate (CDR) value.3 condition round(x/(P/L))=l 43. Q1=80 and Q2=55.
Fig 4: Example of embedding/extracting processes: (a) Original image (b) Secret messageImage Encrypted Image (Second Round) Encrypted (c) Secret message after Embedding encryption (d) Embedded image (e) Difference image after scaling the information in JPEG image.4
79.6 DIFFERENT SETS OF Q1 AND Q2 Step 3: The difference image is now divided into the blocks i. „W‟ is the 43.y) is compressed with Q2 70
40. image from the original image and we get the obtained from 90. Now we apply modulo operation to thethe PSNR value withadifferent Q1 and The above table shows image which gives random pattern. 40. In that we have used two pairs of parameters (e1.4 89.
88. and (h) 85 43. The recipient will resave the image I1 to I2 with lower quality factor Q2. Whenever q2=q1. (c)–(d) Secret message after the first and Extracted Resavi second Image encryption round. The quality the pair (e. average PSNR and CDR p 90 is shown. The extracting of image from Host image image. In the Farid‟s scheme.k1 of the JPEG image. But generally the result difference image.3 91.2
Where x.y values satisfy the round(y/(Q/M))=m.2 key 5 shows another example with host image of size 512x512 Fig for decryption. In the first stage. Baseddone using two schemes: extract the hidden The testing is on thresholding. Outguess. QTs are responsible scrambled pattern will extraction rate. we can Stegdetect  and messagescheme .7 83.6 image so that the shape of the secret image remains unchanged. if we process the C1 with lower quality. High CDR is very essential and for that Q1 and Q2 ti 85. we get PSNR that is used (1) Encryption stage of the proposed scheme.k2‟ the specific (e1. jphide and jsteg are the systems that stegdetect can easily identify. we can also apply an encryption algorithm with a strong key.66).
4 39.3 is to enhance the security.6 88. The experimental result shows that the proposed scheme has 40 43.8 95.0 W.6 39.. 1996.7 39. 2008.html 38. Forensics Security..7 Data Compression Standard.5 41.3 39 38.cs.. The "Exposing digital DCT domain. Farid.0
Q1-Q2 5 10 55 It returns negative results for different pairs of Q1 and Q2 for all 15 39.4
evaluated the capability of steganalyser schemes.3% 40.9 http://www.5 25 innocent group even with the quality 40. Available: http://www. 93..” in 8th Int.. vol.. 2005. Egypt.5 in the JPEG image.6 39. 154–160. this scheme is very 38.8 38. 237–240.” in SPIE Symp.6 for our secret information which is 38.. downloaded at http://www.6 steganalyser design: “LogSv”. .
4.9 two different quality factors combined with encryption is done 40.9 Computer 84.8 93.. Shi.7% but it 30 39.0 technique using different QTs is adding a more protective layer 38..4 very important in this scheme. Raja and C. Image 84. G. Conf. L. hidden Now-a-days.2 quantization.7 . Image
98.9 .4 91.5 suggested to make use of complex encryption procedures to provide more security and reliability. Conf.2 39.1 92. pp.5 switching between the morphisms in image watermarking. S.9 The proposed scheme is very practical and it does not use any 39.7.” in Int.4 43.com/ photo/jpeg39. Tso et al..3 IEEE Int....0 88. no.9 [Online]..1
43. .. Supporting technical document and the utility can be .3 82.
.4 39. Lyu and H. “A secure image steganography using LSB. Intelligent Sensing and Information Processing. Video Technol.39. Stegdetect gives the output image into 20 39.org ..impulseadventure.8 40. pp. Conf.
39. Kurugollu. G. Intelligent Systems Designs and Application.9 92. pp.8 38. Farid. pp. Pennebaker and J. 1.0 information. Chowdary.3
Mitchell." IEEE Trans.8 increases to the percentage of 13. Voyatzis and I.4 if (Q1-Q2) value is over 35 41.3 60. vol.8 85. New 89.1 96. K.4 Applications 83..1 39.edu/farid/research/
5.umich. 18.citi. 2009. 38. [Online]. 65
38. Electronic Imaging.6 40. Further it is 39. The purpose of embedding 38.6 Farid‟s scheme‟s detection rate performance will be 4..3 Processing (ICIP 2009).2 93.
H.2 94. Proceedings.2 41.. The reason for using encryption procedure 40 38. Conf.0 TABLE III AVERAGE PSNR AND CDR (%) VERSUS 39. Image 39. 170–176. 2004 [Online]. “A novel universal 90. 1. 2008. Cairo.outguess.0 42.1 .7 40. Circuits Syst.. “Steganalysis using color wavelet statistics and one-class support vector machine.. vol.
.. 85. Ni and Y. CONCLUSION
Mar.Available: 87.2 DIFFERENCE BETWEEN Q1 AND Q2 39.1 88. H.0 difference value as 70.. Available: Detection framework can be found at http://www.8 Processing. B. pp.. 497–509.. Z. no. Inform. 4.4 86. 616–619.6 96. Gul and F. “Applications of toral autotraditional space like forgeries from JPEG ghosts. Pitas.” IEEE Trans. 1993. 39.4 39.3 41.7 practical for maintaining secrecy and privacy for sender‟s 39.9 40. 20
in proposed scheme. JPEG: Still York: Springer.. pdf. “A lossless secret image sharing method.5 97.” in87. K.” in 3rd Int. REFERENCES
. “Robust lossless image data hiding designed for semi-fragile image authentication.edu/u/-provos/papers/detecting.dartmouth. The strength of the encryption key is 38.4 the embedded images. May 2011
(0975 – 8887)
. DCT and compression techniques on raw images. 4.1 94.0 PSNR
International Journal of Volume 21– No.