A chaotic system based fragile watermarking scheme for image tamper detection

Page 1

A

SD

a

ARA

KTCAD

1

odhmnttisscabbiwopebt

1d

Int. J. Electron. Commun. (AEÜ) 65 (2011) 840– 847

Contents lists available at ScienceDirect

International Journal of Electronics andCommunications (AEÜ)

j our na l ho mepage: www.elsev ier .de /a eue

chaotic system based fragile watermarking scheme for image tamper detection

anjay Rawat ∗, Balasubramanian Raman1

epartment of Mathematics, Indian Institute of Technology Roorkee, Roorkee 247 667, India

r t i c l e i n f o

rticle history:eceived 23 September 2010ccepted 24 January 2011

eywords:amper detectionhaotic map

a b s t r a c t

In the past few years, various fragile watermarking techniques have been proposed for image authen-tication and tamper detection. In this paper, a novel chaos based watermarking scheme for imageauthentication and tamper detection is proposed. Tamper localization and detection accuracy are twoimportant aspects of the authentication watermarking schemes. Our scheme can detect any modifica-tion made to the image and can also indicate the specific locations that have been modified. To improvethe security of the proposed scheme two chaotic maps are employed. Since chaotic maps are sensitive

uthenticationigital watermarking

to initial values, the corresponding position relation between pixels in the watermarked image and thewatermark get disturbed, which helps the watermarking scheme to withstand counterfeiting attacks.Initial values of the chaotic maps are used as secret keys in our scheme. The effectiveness of the pro-posed scheme is checked through a series of attacks. Experimental results demonstrate that the proposedscheme is not only secure but also achieves superior tamper detection and localization accuracy underdifferent attacks. For instance in copy-and-paste attack and collage attack.

. Introduction

With the rapid growth of internet technologies, a large amountf digital data is easily accessible to everyone these days. This digitalata can be easily manipulated, tampered and distributed with theelp of powerful image processing tools. The ease and extent of suchanipulations emphasize the need for image authentication tech-

iques in applications where verification of integrity and authen-icity of the image content is essential. Therefore, various authen-ication schemes have recently been proposed for verifying thentegrity and authenticity of the image content. The authenticationchemes can be divided into two categories: digital signature basedchemes and digital watermark based schemes. A digital signaturean be either an encrypted or a signed hash value of image contentsnd/or image characteristics. The major drawback of signatureased schemes is that they can detect if an image has been modified,ut they cannot locate the regions where the image has been mod-

fied [1–3]. To solve this problem, many researchers have proposedatermarking based schemes for image authentication [4–7]. One

f the first watermarking-based authentication schemes was pro-osed by Walton [9]. He divided the image into 8 × 8 blocks and

mbedded the checksum in the LSB of each block. The main draw-ack of Walton’s scheme is that there is a possibility of exchanginghe blocks with the same position in two different authenticated

∗ Corresponding author. Tel.: +91 01332 285852; fax: +91 01332 285852.E-mail addresses: [email protected] (S. Rawat), [email protected] (B. Raman).

1 Tel.: +91 1332 285852.

434-8411/$ – see front matter © 2011 Elsevier GmbH. All rights reserved.oi:10.1016/j.aeue.2011.01.016

© 2011 Elsevier GmbH. All rights reserved.

images without affecting the checksum of the image. Yeung andMintzer [10] proposed a watermarking scheme for image authen-tication that uses a pseudo random sequence and a modified errordiffusion method to embed a binary watermark into an image,so that any change in pixel values of the image can be detected.Fridrich et al. [11] analyzed the security issue in the scheme pro-posed by Yeung and Mintzer [10] and proposed an improvedscheme with localization capability, where a block cipher definedon a local neighborhood rather than on a single pixel is used toreplace the binary look-up tables. Thus, attacker could not deducethe binary look-up table. At the same time, authors embedded animage index into all non-overlapping sub-blocks of each image toprevent the collage attack proposed in [12,13]. Wong [14] proposeda public key fragile watermarking scheme for image authentication.He divided the image into non-overlapping blocks and inserted adigital signature for authentication. In his scheme, a key is used togenerate a signature using the seven most significant bits of the pix-els in each image block together with a logo to form a watermark,and embed the watermark into the least significant bits of the cor-responding blocks. The blockwise independence of the authentica-tion schemes, proposed in the literature was exploited by Hollimanand Memon [15]. They proved that these scheme are vulnerableto vector quantization attack. According to them, a counterfeitimage can be constructed using a vector quantization codebookgenerated from a set of watermarked images. Since each block is

authenticated by itself, the counterfeit image appears authentic tothe watermarking scheme. To withstand the vector quantizationattack, a number of schemes has been proposed. Wong and Memon[16] proposed an improved blockwise authentication scheme by

Page 2

S. Rawat, B. Raman / Int. J. Electron. Commun. (AEÜ) 65 (2011) 840– 847 841

enom

afvlhsopssaiueiaatT

Fig. 1. Periodic ph

dding an image index and a block index to the inputs of the hashunction. However, this idea works at the expense of requiring theerifier to have a priori knowledge about the image index, whichimits its applicability to some extent. Celik et al. [17] proposed aierarchically structured watermarking scheme based on Wong’scheme [14] which provides a blockwise authentication with highlyverlapping blocks. In Celik et al.’s scheme, the original image isartitioned into blocks in a multi-level hierarchy and then blockignatures in this hierarchy are calculated. Based on this hierarchytructure, the scheme can effectively thwart vector quantizationttack. Suthaharan [18] proposed a fragile watermarking schemen which the security against vector quantization attack is achievedsing a gradient image and its bits distribution properties to gen-rate a large key space. Chang et al. [19] proposed a block-basedmage authentication scheme which can withstand counterfeiting

ttacks by combining the local and global features to obtain theuthentication data. Chen et al. [20] proposed a fuzzy c-means clus-ering based watermarking scheme to resist counterfeiting attacks.o break the block wise independency, they applied the fuzzy c-

ImageI

ScrambledImage Iscr

Logisticmap

Chaoticsequence

Reshape and

round off

Binarywatermark

cat mapk times

Waterma Imag

Fig. 2. Block diagram of e

enon in cat map.

means clustering technique to cluster all the image blocks, so thatthe relationship between blocks can be created. The authenticationdata is embedded into two least significant bits of each image block.

In this paper, a novel watermarking scheme based on chaoticmaps is proposed. The pixels of the cover image are disturbed withthe help of Arnold’s cat map. The image is further divided into 8-bitplanes and the least significant bit (LSB) plane is used for watermarkembedding. A binary logo is used as watermark in our scheme. Achaotic image pattern is generated by using logistic map. A scram-bled watermark is obtained by using exclusive-or (XOR) operationbetween chaotic image pattern obtained by using logistic map andthe binary watermark. The scrambled watermark is then embeddedin the least significant bit (LSB) plane of the image. Watermarkedimage is obtained by performing an inverse cat map.

The rest of the paper is organized as follows. In Section 2,

Arnold’s cat map and logistic map are briefly described. In Section3, the proposed watermarking scheme is explained. Experimen-tal results are given in Section 4. Conclusions are drawn inSection 5.

ChaoticImage

XOR Chaoticwatermark

Replace the LSB bitplane of I scr

by chaotic watermark

Modified Iscrrkede

cat map

(T- k ) times

mbedding process.

Page 3

8 n. Commun. (AEÜ) 65 (2011) 840– 847

2

mcdsw

2

dsd[

Tfae.N[

w

atWrpibcti

2

x

wssg

WatermarkedImage

ScrambledImage

Logisticmap

Chaoticsequence

ChaoticImage

Reshape and

round off

Originalwatermark

XOR

LSB bitplane of scrambled Image

cat map

Extractedwatermark

TakeAbsolute

Difference

Locatemodified regions

cat map(T- k ) times

k times

42 S. Rawat, B. Raman / Int. J. Electro

. Chaotic maps

In recent years, chaotic maps have been used for digital water-arking, to increase the security. The most attractive features of

haos in information hiding are its extreme sensitivity to initial con-itions and the outspreading of orbits over the entire space. Thesepecial characteristics make chaotic maps excellent candidates foratermarking and encryption [8].

.1. Arnold’s cat map

In order to shuffle the pixel positions of the host image, twoimensional Arnold cat map is employed in our scheme. The clas-ical Arnold cat map is a two-dimensional invertible chaotic mapescribed by

xn+1yn+1

]=

[1 11 2

] [xn

yn

]mod 1 (1)

he map is area-preserving since the determinant of its linear trans-ormation matrix is equal to 1. The map is chaotic in real spacend in addition it is one to one map. The above 2D cat map can bextended as follows. First, the phase space is generalized to [0, 1, 2,

. ., N − 1] × [0, 1, 2, . . ., N − 1], i.e. only positive integers from 0 to − 1 are taken, then above map is generalized as:

xn+1yn+1

]=

[1 ab ab + 1

] [xn

yn

]mod N = A

[xn

yn

]mod N (2)

here a and b are positive integers and det(A) = 1.The generalized cat map given by Eq. (2) is also chaotic and

rea preserving. Since we have restricted the phase space to posi-ive integers, the generalized cat map becomes periodic in nature.

e say that the equation has period T if the pixel at location (x, y)eturns to its original position after being transformed T times. Theeriod T depends on the parameters a, b and size N of the original

mage. Thus parameters a, b and the number of iterations k, all cane used as secret keys. Fig. 1 shows the periodicity property of theat map where the parameters of Eq. (2) are a = 1, b = 1 and size ofhe image is N = 128. It shows that for these parameters the periods equal to 96.

.2. Logistic map

Logistic map is one of the simplest chaotic maps, described by

k+1 = �xk(1 − xk) (3)

here 0 < � ≤ 4. When 3.5699456 < � ≤ 4, the map is in chaotictate. All the sequences generated by the logistic map are very sen-itive to initial conditions, in the sense that two logistic sequencesenerated from different initial conditions are statistically uncor-

Fig. 4. (a) Original sailboat image, (b) binary w

Fig. 3. Block diagram of extraction process.

related. Moreover, all the orbits of the logistic map are dense in therange of the map [0, 1].

3. The proposed scheme

In this section, we explain the proposed watermarking scheme.Let us consider, I is the host image of size M × N and W is the binarywatermark of size m × n.

3.1. Watermark embedding

The watermark is embedded as follows:

1. Scramble the original image I, using Arnold cat map k times,where k is the encryption key of the chaotic mixing process forI. Let us denote the result by Iscr.

2. Divide the scrambled image Iscr into 8-bit planes.3. Generate a chaotic sequence S of length m × n using logistic map.

Round off the chaotic sequence and rearrange to get the chaoticimage pattern Scp.

4. Obtain a binary chaotic watermark Wp using exclusive-or (XOR)operation between the watermark W and Scp as follows:

Wp = Scp ⊕ W

5. Replace the least significant bit plane of Iscr by Wp.

6. Apply Arnold cat map (T − k) times on modified Iscr to get the

watermarked image, where T is the period of cat map.

The block diagram of the embedding process is shown in Fig. 2.

atermark and (c) watermarked image.

Page 4

S. Rawat, B. Raman / Int. J. Electron. Commun. (AEÜ) 65 (2011) 840– 847 843

ered

3

1

2

Fig. 5. (a) Original sailboat image, (b) watermarked image, (c) tamp

.2. Watermark extraction

The watermark is extracted as follows:

. Scramble the Watermarked image IW, using Arnold cat map ktimes. Let us denote the result by IWscr .

. Divide the scrambled watermarked image IWscr into 8-bit planes.

Fig. 6. (a) Original sailboat image, (b) watermarked image, (c) tampered

image, (d) extracted watermark and (e) detected tampered region.

3. Obtain the same chaotic image pattern Scp as in step 3 of embed-ding algorithm.

4. Apply exclusive-or (XOR) operation between the least signifi-cant bit plane of IWscr and chaotic image pattern Scp to get the

ext

extracted watermark W .5. Take the absolute difference of extracted watermark Wext and

original watermark W. Apply Arnold cat map (T − k) times tolocate the tampered areas of the watermarked image.

image, (d) extracted watermark and (e) detected tampered region.

Page 5

844 S. Rawat, B. Raman / Int. J. Electron. Commun. (AEÜ) 65 (2011) 840– 847

ered

4

t2e

Fig. 7. (a) Original sailboat image, (b) watermarked image, (c) tamp

The block diagram of the extraction process is shown in Fig. 3.

. Experimental results

Various experiments are carried out in this section, to assesshe performance of the proposed algorithm. A binary logo of size56 × 256 is used as watermark in all the experiments. The param-ters of Arnold cat map used in our scheme are, a = 1, b = 1, and

Fig. 8. (a) Original sailboat image, (b) watermarked image, (c) tampered

image, (d) extracted watermark and (e) detected tampered region.

k = 75. The parameters of logistic map are chosen as � = 3.854 andx(0) = 0.654. PSNR (peak signal-to-noise ratio), is used in this paperto analyze the visual quality of the watermarked image Î in com-parison with the original image I. PSNR is defined as:

PSNR = 10log10

(2552

MSE

)dB (4)

image, (d) extracted watermark and (e) detected tampered region.

Page 6

S. Rawat, B. Raman / Int. J. Electron. Commun. (AEÜ) 65 (2011) 840– 847 845

F , (d) wd

wI

M

4

4

bsi(ptbwFd

b

ig. 9. (a) Original sky image, (b) original bird image, (c) watermarked sky imageetected tampered region.

here MSE is the mean squared error between the original image and the attacked image Î, given by

SE = 1MN

M−1∑i=0

N−1∑j=0

[I(i, j) − I(i, j)]2

(5)

.1. Performance evaluation

.1.1. Performance under copy and paste attackIn this experiment, ‘Sailboat’ image of size 256 × 256 is used. A

inary logo image of size 256 × 256 is used as watermark. Fig. 4hows the host image, binary watermark and the correspond-ng watermarked image. The PSNR value of watermarked imageFig. 4(c)) is 50.7261 dB. Two kinds of copy and paste attacks areerformed in our scheme. In first kind of copy and paste attackhe watermarked sailboat image is modified by inserting two moreoats in the image, where the boats are copied from the sameatermarked image. The tampered image is shown in Fig. 5(c).

ig. 5(d) shows the extracted watermark from Fig. 5(c). The tamperetection result is shown in Fig. 5(e).

In second kind of copy and paste attack the watermarked sail-oat image is modified by inserting an aeroplane in the image,

atermarked bird image, (e) tampered sky image, (f) extracted watermark and (g)

where the aeroplane is copied from some other watermarkedimage. The tampered image is shown in Fig. 6(c). Fig. 6(d) showsthe extracted watermark from Fig. 6(c). Detected tampered regionis shown in Fig. 6(e).

4.1.2. Performance under text additionIn this experiment, the watermarked image, shown in Fig. 7(b)

is modified by adding the text ‘SAIL BOAT’ at the bottom of theimage. Fig. 7(c) shows the tampered image. Extracted watermarkfrom Fig. 7(c) is shown in Fig. 7(d). Detected tampered region isshown in Fig. 7(e).

4.1.3. Performance under content removalIn this experiment, some content of the watermarked image is

removed without degrading the image quality. We have removedsome portion of the cloud from the watermarked image. The tam-pered image is shown in Fig. 8(c). Fig. 8(d) shows the extractedwatermark from Fig. 8(c). The tamper detection result is shown inFig. 8(e).

4.1.4. Performance under collage attackTo evaluate the performance under collage attack, a counterfeit

image is formed by combining the portions of multiple water-

Page 7

846 S. Rawat, B. Raman / Int. J. Electron. Commun. (AEÜ) 65 (2011) 840– 847

F , (d) wd

mws

•

•

ig. 10. (a) Original sofa image, (b) original doll image, (c) watermarked sofa imageetected tampered region.

arked images while preserving their relative spatial locationsithin the target image. We have performed this attack for two

ets of images. The simulation results are as follows:

Sky and bird: The original sky and bird images are shown inFig. 9(a) and (b). Both the images are of size 256 × 256. The cor-responding watermarked images are shown in Fig. 9(c) and (d),where the PSNR values are 51.1552 dB and 51.0992 dB, respec-tively. The counterfeit image, as show in Fig. 9(e) was constructedby copying the bird from Fig. 9(d) and pasting it in Fig. 9(c). Fig. 9(f)shows the extracted watermark from Fig. 9(e) and Fig. 9(g) showsthe detected tampered region.Sofa and doll: The original sofa and doll images are shownin Fig. 10(a) and (b). Both the images are of size 256 × 256.The corresponding watermarked images are shown in Fig. 10(c)and (d), where the PSNR values are 50.6722 dB and 51.0752 dB,

respectively. The counterfeit image, as show in Fig. 10(e) was con-structed by copying the teddy bear from Fig. 10(d) and pastingit in Fig. 10(c). Fig. 10(f) shows the extracted watermark fromFig. 10(e) and Fig. 10(g) shows the detected tampered region.

atermarked doll image, (e) tampered sofa image, (f) extracted watermark and (g)

5. Conclusion

A novel fragile watermarking scheme for image authenticationand locating tampered regions, is presented in this paper. Chaoticmaps are used in our scheme to make the scheme highly secure.Since chaotic maps are sensitive to initial values, they are used askeys in our scheme. Extracting the right watermark is only pos-sible if someone has correct keys. A person with wrong keys willnot be able to forge the watermark. As, in order to thwart counter-feiting attacks it is essential to break pixel wise independency, theproposed scheme employs chaotic maps to break the correspond-ing position relation between pixels in the watermarked imageand the watermark. Experimental results show that our schemehas high fidelity and is capable of localizing modified regions inwatermarked image.

References

[1] Dittmann J, Steinmetz A, Steinmetz R. Content-based digital signature formotion pictures authentication and content-fragile watermarking. In: Proceed-ings of the IEEE international conference on multimedia computing systems.1999. p. 209–13.

Page 8

n. Com

[

[

[

[

[

[

[

[

[

[tampering authentication scheme. Pattern Recognition Letters 2006;27(5):439–46.

[20] Chen WC, Wang MS. A fuzzy c-means clustering-based fragile watermark-

S. Rawat, B. Raman / Int. J. Electro

[2] Lin CY, Chang SF. A robust image authentication method surviving JPEG lossycompression. In: Proceedings of SPIE international conference on storage andretrieval of image/video database, vol. 3312. 1998. p. 296–307.

[3] Lu CS, Liao HYM, Sze CJ. Structural digital signature for image authentication:an incidental distortion resistant scheme. In: Proceedings of the multimediasecurity workshop 8th ACM international conference on multimedia. 2000. p.115–8.

[4] Eggers JJ, Girod B. Blind watermarking applied to image authentication. In:Proceedings of IEEE international conference on acoustics, speech and signalprocessing. 2001. p. 1977–80.

[5] Kundur D, Hatzinakos D. Digital watermarking for telltale tamper proofing andauthentication. In: Proceedings of the IEEE special issue on identification andprotection of multimedia information. 1999. p. 1167–80.

[6] Lin ET, Podilchuk CI, Delp EJ. Detection of image alterations using semi-fragilewatermarks. In: Proceedings of SPIE conference on security and watermarkingof multimedia contents II. 2000. p. 152–63.

[7] Wang MS, Chen WC. A majority-voting based watermarking scheme forcolor image tamper detection and recovery. Computer Standards & Interfaces2007;29:561–70.

[8] Dawei Z, Guanrong C, Wenbo L. A chaos-based robust wavelet-domain water-marking algorithm. Chaos Solitons and Fractals 2004;22:47–54.

[9] Walton S. Information authentication for a slippery new age. Dr Dobb’s Journal1995;20(4):18–26.

10] Yeung M, Mintzer F. An invisible watermarking technique for image verifi-

cation. In: Proceedings of IEEE international conference on image processing.1997. p. 680–3.

11] Fridrich J, Goljan M, Baldoza AC. New fragile authentication watermark forimages. In: Proceeding of IEEE international conference on image processing,vol. 1. 2000. p. 446–9.

mun. (AEÜ) 65 (2011) 840– 847 847

12] Memon N, Shende S, Wong P. On the security of the Yeung–Mintzer authenti-cation watermark. In: Proceedings of the IS&T PICS symposium. 1999. p. 301–6.

13] Fridrich J, Goljan M, Memon N. Further attacks on Yeung–Mintzer watermark-ing scheme. In: Proceedings of SPIE electronic imaging, vol. 3971. 2000. p.428–37.

14] Wong PW. A public Key watermark for image verification and authentication.In: Proceedings of IEEE international conference on image processing. 1998. p.455–9.

15] Holliman M, Memon N. Counterfeiting attacks on oblivious block-wiseindependent invisible watermarking schemes. IEEE Transactions on Image Pro-cessing 2000;9(3):432–41.

16] Wong PW, Memon N. Secret and public key image watermarking schemes forimage authentication and ownership verification. IEEE Transactions on ImageProcessing 2001;10(10):1593–601.

17] Celik MU, Sharma G, Saber E, Tekalp AM. Hierarchical watermarking for secureimage authentication with localization. IEEE Transactions on Image Processing2002;11(6):585–95.

18] Suthaharan S. Fragile image watermarking using a gradient image for improvedlocalization and security. Pattern Recognition Letters 2004;25(16):1893–903.

19] Chang CC, Hu YS, Lu TC. A watermarking-based image ownership and

ing scheme for image authentication. Expert Systems with Applications2009;36:1300–7.