A Novel Robust Watermarking Technique Based on Nonsubsampled Contourlet Transform and SVD

872019 A Novel Robust Watermarking Technique Based on Nonsubsampled Contourlet Transform and SVD

httpslidepdfcomreaderfulla-novel-robust-watermarking-technique-based-on-nonsubsampled-contourlet-transform 117

The International Journal of Multimedia amp Its Applications (IJMA) Vol3 No1 February 2011

DOI 105121ijma20113104 37

A Novel Robust Watermarking Technique Basedon Nonsubsampled Contourlet Transform and

SVDCVenkata Narasimhulu 1 amp KSatya Prasad 2

1 Professor Hasvita Institute of Engineering amp TechnologyHyderabad Indianarasimhulucvgmailcom faxtelephone + 91 8418284884

2 Professor Dept of ECE JNTU Kakinada Indiaprasad_kodatiyahoocoin faxtelephone

+91 884 230 0912

ABSTRACT The paper proposes a novel robust watermarking technique based on newly introduced Nonsubsampled

contourlet transform(NSCT) and singular value decomposition(SVD) for multimedia copyright protectionThe NSCT can give the asymptotic optimal representation of the edges and contours in image by virtue of the characteristics of good multi resolution shift invariance and multi directionality After decomposing thehost image into sub bands we choose the low frequency directional sub band and apply singular valuedecomposition The singular values of the original image are then modified by the singular values of nonsubsampled contourlet transformed visual grayscale logo watermark image This hybrid approachimproves the performance of the watermarking technique compared to earlier techniques Experimentalresults shows that the hybrid technique is resilient to various linear and non linear filtering JPEGcompression JPEG2000 compression Histogram equalization Grayscale inversion Contrast adjustmentgamma correction alpha mean cropping Gaussian noise scaling etc

KEYWORDS

Image watermarking nonsubsampled contourlet transform SVD visual watermark logo

1 INTRODUCTIONRapid growth in digital technique and internet usage has created a new set of challengingproblems such as copyright protection authentication and content integrity verification of thedigitized properties Over the last few years watermarking is popularly known as a potentialsolution to address these problems through invisible insertion of auxiliary message (logosymbol)called watermark in digital data [1] This insertion data can be later extracted from or detected inthe multimedia to make an assertion about the data Digital watermarks remain intact undertransmissiontransformation allowing us to protect our ownership rights in digital form Absenceof watermark in a previously watermarked image would lead to the conclusion that the datacontent has been modified A watermarking algorithm consists of watermark structure anembedding algorithm and extraction or detection algorithm In multimedia applicationsembedded watermark should be invisible robust and have a high capacity Invisibility refers todegree of distortion introduced by the watermark and its affect on the viewers and listenersRobustness is the resistance of an embedded watermark against intentional attack and normalsignal processing operations such as noise filtering rotation scaling cropping and losseycompression etc Capacity is the amount of data can be represented by embedded watermark [2][3]




38

Watermarking techniques may be classified in different ways The classification may be based onthe type of watermark being used ie the watermark may be a visually recognizable logo orsequence of random numbers A second classification is based on whether the watermark isapplied in the spatial domain or the transform domain In spatial domain the simplest method isbased on embedding the watermark in the least significant bits (LSB) of image pixels Howeverspatial domain techniques are not resistant enough to image compression and other imageprocessing operations

Transform domain watermarking schemes such as those based on the discrete cosine transform(DCT) the discrete wavelet transform (DWT) contourlet transforms along with numericaltransformations such as Singular value Decomposition (SVD) and Principle component analysis(PCA) typically provide higher image fidelity and are much robust to image manipulations[4]Of the so far proposed algorithms wavelet domain algorithms perform better than other transformdomain algorithms since DWT has a number of advantages over other transforms including timefrequency localization multi resolution representation superior HVS modeling and linearcomplexity and adaptively and it has been proved that wavelets are good at representing pointwise discontinuities in one dimensional signal However in higher dimensions eg image thereexists line or curve-shaped discontinuities Since 2D wavelets are produced by tensor products of 1D wavelets they can only identify horizontal vertical diagonal discontinuities (edges) inimages ignoring smoothness along contours and curves Curvelet transform was defined torepresent two dimensional discontinuities more efficiently with least square error in a fixed termapproximation Curvelet transform was proposed in continuous domain and its discretisation wasa challenge when critical sampling is desired Contourlet transform was then proposed by DO andVetterli as an improvement of Curvelet transform The Contourlet transform is a directional multiresolution expansion which can represents images contains contours efficiently The CT employsLaplacian pyramids to achieve multi resolution decomposition and directional filter banks toachieve directional decomposition [4][5][6] Due to down sampling and up sampling theContourlet transform is Shift variant However shift invariance is desirable in image analysisapplications such as edge detection Contour characterization image enhancement [7] and image

watermarking Here we present a NonSubsampled Contourlet transform (NSCT) [8] which isshift invariant version of the contourlet transform The NSCT is built upon iteratednonsubsampled filter banks to obtain a shift invariant image representation

In all above transform domain watermarking techniques including NSCT the watermarking bitswould be directly embedded in the locations of sub band coefficients Though here the visual of perception of original image is preserved the watermarked image when subjected to someintentional attacks like compression the watermark bits will get damaged Coming to the spatialdomain watermarking using numerical transformation like SVD (Gorodetski [9] liu et al [10])they provide good security against tampering and common manipulations for protecting rightfulownership But these schemes are non adaptive thus unable to offer consistent perceptualtransparency of watermarking of different images [11]To provide adaptive transparency

robustness to the compressions and insensitivity to malicious manipulations we propose a novelimage hybrid watermarking scheme using NSCT and SVD

In this paper proposed method is compared with another which is based on discrete wavelettransform and singular value decomposition (DWT-SVD)The peak signal noise ratio (PSNR)between the original image and watermarked image and the normalized correlation coefficients(NCC) after different attacks were calculated The results show high improvement detectionreliability using proposed methodThe rest of this paper is organized as follows Section2




39

describes the Nonsubsampled contourlet transform section 3 describes singular valuedecomposition section 4 illustrates the details of proposed method in section 5 experimentalresults are discussed without and with attacks conclusion and future scope are given in section 6

2 NONSUBSAMPLED CONTOURLET TRANSFORM The nonsubsampled contourlet transform is a new image decomposition scheme introduced byArthur LCunha Jianping Zhou and Minh NDo [12] NSCT is more effective in representingsmooth contours in different directions of in an image than contourlet transform and discretewavelet transform The NSCT is fully shift invariant Multi scale and multi direction expansionthat has a fast implementation The NSCT exhibits similar sub band decomposition as that of contourlets but without down samplers and up samplers in it Because of its redundancy thefilter design problem of nonsubsampled contourlet is much less constrained than that of contourlet [12][13][14] The NSCT is constructed by combining nonsubsampled pyramids andnonsubsampled directional filter bank as shown in Figureure 1The nonsubsampled pyramidstructure results the multi scale property and nonsubsampled directional filter bank results thedirectional property

(a) (b)

Figure 1 The nonsubsampled contourlet transform (a) nonsubsampled filter bank structure thatimplements the NSCT (b) Idealized frequency partitioning obtained with NSCT

21 Nonsubsampled pyramids

The nonsubsampled pyramid is a two channel nonsubsampled filter bank as shown in Figureure2(a)The H 0(z) is the low pass filter and one then sets H 1(z) =1-H 0(z) and corresponding synthesisfilters G 0(z) =G 1(z)=1The perfect reconstruction condition is given by Bezout identity [7][8][12]




40

H0 (z) G 0 (z) +H 1 (Z) G 1 (Z) =1helliphelliphelliphelliphelliphellip 1

(a) (b)Figure 2 Nonsubsampled pyramidal filters (a) Ideal frequency response of nonsubsampled

pyramidal filter (b)The cascading analysis of three stages nonsubsampled pyramid by iteration of two channels nonsubsampled filter banks

Multi scale decomposition is achieved from nonsubsampled pyramids by iterating thenonsubsampled filter banks The next level decomposition is achieved by up sampling all filtersby 2 in both dimensions The complexity of filtering is constant whether the filtering is with H(z)or an up sampled filter H(z m ) computed using lsquo a trous rsquo algorithm [15] The cascading of threestage analysis part is shown in Figureure2 (b)

22 Nonsubsampled directional Filter Banks

The directional filter bank (DFB) [16] is constructed from the combination of critically-sampledtwo-channel fan filter banks and resampling operations The outcome of this DFB is a tree-structured filter bank splitting the 2-D frequency plane into wedges The nonsubsampleddirectional filter bank which is shift invariant is constructed by eliminating the down and upsamplers in the DFB[13]The ideal frequency response of nonsubsampled filter banks is shown inFigureure3 (a)




41

(a) (b)

Figure 3 Nonsubsampled directional filter bank (a) idealized frequency response of nonsubsampled directional filter bank(b) The analysis part of an iterated nonsubsampled

directional bank

To obtain multi directional decomposition the nonsubsampled DFBs are iterated To obtain thenext level decomposition all filters are up sampled by a quincunx matrix given by [7][8]

helliphelliphelliphelliphellip 2

The analysis part of an iterated nonsubsampled filter bank is shown in Figure 3(b)

3 SINGULAR VALUE DECOMPOSITIONSingular value decomposition (SVD) is a popular technique in linear algebra and it hasapplications in matrix inversion obtaining low dimensional representation for high dimensionaldata for data compression and data denoising If A is any N x N matrix it is possible to find adecomposition of the form

A=USV T

A = [ u 1 u2 hellip u n] [v1 v2hellip v n]T

λ 1

λ

2

λ n




42

Where U and V are orthogonal matrices of order N x N and N x N such that U TU=IV TV=I andthe diagonal matrix S of order N x N has elements λ i (i=123n) I is an identity matrix of orderN x N

The diagonal entries are called singular values of matrix A the columns of U matrix are called

the left singular values of A and the columns of V are called as the right singular values of A [4]The general properties of SVD are [2] [4] [10]

a)

Transpose A and its transpose AT have the same non-zero singular values

b)

Flip A row-flipped Arf and column-flipped Acf have the same non-zero singular values

c)

Rotation A and Ar (A rotated by an arbitrary degree) have the same non-zero singularvalues

d)

Scaling B is a row-scaled version of A by repeating every row for L1 times For each non-zero singular value λ of A B has radicL1λ C is a column-scaled version of A by repeating everycolumn for L2 times For each nonzero singular value λ of A C has radicL2λ If D is row-scaled byL1 times and column-scaled by L2 times for each non-zero singular value λ of A D has radicL1L2λ

e)

Translation A is expanded by adding rows and columns of black pixels The resultingmatrix Ae has the same Non-zero singular values as A

The important properties of SVD from the view point of image processing applications are

1

The singular values of an image have very good stability ie When a small perturbation isadded to an image their singular values do not change significantly

2

Singular value represents intrinsic algebraic image properties [2][3][4][10][17][18]

Due to these properties of SVD in the last few years several watermarking algorithms have beenproposed based on this technique The main idea of this approach is to find the SVD of a original

image and then modify its singular values to embedded the watermark Some SVD basedalgorithms are purely SVD based in a sense that only SVD domain is used to embed watermark into original image Recently some hybrid SVD based algorithms have been proposed wheredifferent types of transform domains including discrete cosine transform (DCT) discrete wavelettransform (DWT) Contourlet transform (CT) etc are used to embed watermark into originalimage Here the proposed scheme uses nonsubsampled contourlet transform (NSCT) along withSVD for watermarking to obtain better performance compared to existing hybrid algorithms

4 PROPOSED ALGORITHM

In this paper NSCT and SVD based hybrid technique is proposed for watermarking of gray scalewatermark image on gray scale original image The robustness and perceptuality of watermarkedimage is tested with two quantifiers such as PSNR and NCC It is investigated whether the

NSCT-SVD advantages over DWT-SVD with their extra features would provide any significantin terms of watermark robustness and invisibility41 42 explain the watermark embedding andextraction algorithm [2][3][4][19]

41 Watermark Embedding Algorithm

The proposed watermark embedding algorithm is shown in Figure 4 The steps of watermark embedding algorithm are as follows




43

Figure 4 Watermark Embedding Algorithm

Step1 Apply NSCT to the original image to decompose into sub bands

Step2 Apply SVD to low frequency sub band of NSCT of original image

Step3 Apply NSCT to gray scale logo watermark to decompose into sub bands

Step4 Apply SVD to low frequency sub band of NSCT of gray scale logo watermark image

Step5 Modify the singular values of original image with the singular values of gray scale image

watermark ie λ Irsquo = λ I + α λ W

Where α is scaling factor[4] λ I is singular value of original image λ W is singular value of gray scale logo watermark and λ Irsquo

becomes singular value of watermarked image

Step6 Apply inverse SVD of transformed original image with modified singular values in step5

Step7 Apply inverse NSCT using the modified coefficients of the low frequency bands to obtainthe watermarked image




44

42 Watermark Extraction Algorithm

The watermark extraction algorithm is shown in Figureure 5 The Steps of watermark extraction

algorithm are as follows

Figure 5 Watermark Extraction Algorithm

Step1 Apply NSCT to the watermarked image to decompose into sub bands

Step2 Apply SVD to low frequency sub band of transformed watermarked image

Step3 Extract the singular values from low frequency sub band of watermarked and originalimage i e λ W = ( λ Irsquo - λ I ) α Where λ I is singular value of watermarked image

Step4 Apply inverse SVD to obtain low frequency coefficients of transformed watermark image

using Step 3

Step5 Apply inverse NSCT using the coefficients of the low frequency sub band to obtain thegray scale Watermark image




45

5 EXPERIMENTAL RESULTS

In the experiments we use the standard grayscale ldquoLenajpgrdquo of size 512 X 512 as original imageas shown in the Figure 6 and grayscale ldquoCameramanjpgrdquo of size 256 X 256 as watermark asshown in Figure 7 The results show that there are no perceptibly visual degradations on the

watermarked image shown in Figure 8 with a PSNR of 376102dB Extracted watermark withoutattack is shown in Figure 9 with NCC unity MATLAB 76 version is used for testing therobustness of the proposed method The proposed algorithm is also applied for different originalimages such as ldquoPeppersjpgrdquordquoBaboonjpgrdquordquoricejpgrdquordquoBarbarajpgrdquo and ldquoZoneplatejpgrdquo as inTable 1 and it is observed that there are no visual degradations on the respected watermarkedimages For all the different original test images the watermark is effectively extracted with unityNCC Various intentional and non-intentional attacks are tested for robustness of the proposedwatermark algorithm includes JPEGJPEG2000compressionsLow pass filtering RotationHistogram Equalization Median Filtering Alpha Mean Gray Scale Inversion Salt ampPepperNoise Soft Thresholding Weiner Filtering Gamma Correction Gaussian Noise RescalingSharpening Blurring Contrast Adjustment Automatic and Manual cropping Int ThresholdingDilation Mosaic Bit Plane Removal and Row Colum Copying

Figure 6Originalimage- Lenardquo

Figure 7Watermark image-Cameramanrdquo

Figure8WatermarkedldquoLenardquo PSNR=

376102

Figure9ExtractedWatermark Ncc=1

The proposed algorithm is compared with Emir Ganic and Ahmet MEskicioglursquos paper [2] inwhich the watermarking is done by using DWT-SVD hybrid algorithm and the PSNR is reportedas 3442dB and the No of attacks tested are only 12 In our proposed scheme the PSNR obtainedis 376102dB and watermark image can survive up to 24 attacks compared to Emir Ganic andAhmet MEskiciogluas shown in Table2 and Table 3

In Table 2 the normalized correlation coefficient values for different attacks are shown withextracted watermark Y and attacked watermarked image Irsquo The quality and imperceptibility of watermarked image I 1 is measured by using PSNR which can be obtained using eq 3 [20] withrespect to original image I The similarity of extracted watermark(Y) with original watermark (X)

embedded is measured using NCC which is given in eq (4) [21]

helliphelliphellip (3)

Normalized Correlation Coefficient




46

helliphelliphellip(4)

Table 1 watermarked and Extracted watermark with PSNR and NCC for different originalimages

Original image-Baboonjpgrdquo

Watermark image-Cameramanjpgrdquo

watermarked BaboonPSNR= 376289

ExtractedWatermark

Ncc=1

Original image-Peppersjpgrdquo


watermarked pepprsPSNR= 376478

ExtractedWatermark

Ncc=1

Original image-Ricejpgrdquo


Watermarked RicePSNR= 376336

ExtractedWatermark

Ncc=1




47

Original image-Zoneplatejpgrdquo


watermarked Zoneplate

PSNR= 377705

ExtractedWatermark

Ncc=1

Original image-Barbarajpgrdquo


watermarked BarbaraPSNR= 376493

ExtractedWatermark

Ncc=1

+

Table 2 Extracted watermarks with NCC for different attacks along with attacked watermarkedimage

JPEG compressionNcc=09992

JPEG 2000 compressionNcc=09793

Low pass filtering

Ncc= 09793Rotation

Ncc= -04239




48

Histogram equalisation

Ncc= 09722Median filtering

Ncc= 08636

Alpha mean

Ncc= 09619Gray scale inversion

Ncc= 10000

Salt and pepper Noise

Ncc=10000Soft Thresholding

Ncc= 09982

Weiner filter

Ncc= 09907Gamma correction

Ncc= 04734




49

Gaussian NoiseNcc= 10000

RescalingNcc= 10000

Sharpening

Ncc= 05352Blurring

Ncc=09229

Contrast adjustment

Ncc= 09920Automatic cropping

Ncc=-09798

Int thresholding

Ncc= 04572Manual croppingNcc= -09843




50

DilationNcc= 05505

MosaicNcc=09602

Bit plane removal

Ncc=-09873Row Colum copying

Ncc=10000

Table 3 Comparison of proposed method with Emir Ganic and Ahmet MEskicioglursquos algorithm

Characteristic Proposed method Emir Ganic and AhmetMEskicioglu

PSNR in DB 376102 3442No of attacks tested 24 12

We also tested and compared the robustness to various attacks of the proposed method withsimple singular value decomposition and with hybrid algorithms includes discrete wavelettransform and singular value decomposition Contourlet transform and singular valuedecomposition as given in table 4 by taking gray scale ldquolenajpgrdquo of size 512 x 512 as originalimage and gray scale ldquocameramanjpgrdquo of size 256 x256 as watermark The table 4 shows thatproposed algorithm performs better for 16 attacks than that of other algorithms




51

Table 4 Comparison of NCC of various attacks for different algorithms

SNo Attack SVD DWT+SVD CT+SVD NSCT+SVD1 Jpeg compression 08772 09992 09992 099922 Jpeg2000

compression08853 09501 09492 09793

3 Low pass filtering 06197 09743 09681 100004 Rotation(5deg) 02510 02208 01792 028195 Auto cropping -09508 -09471 05975 -098176 Histogram

equalization09505 09537 08238 09722

7 Median filtering 05557 09602 09545 086368 Alpha mean 08949 09458 09566 096199 Gray scale inversion 09868 09868 09874 1000010 Salt and pepper noise 02131 09458 09507 1000011 Int-thresholding 04422 04456 04466 04572

12 Soft thresholding 09982 09982 09981 0998213 Weiner filtering 00185 -05727 07163 0779414 Gamma correction 05004 05030 00118 0473415 Gaussian noise 02590 09755 08414 1000016 Rescaling 10000 10000 10000 1000017 Sharpening 02440 06172 06137 0535218 Blurring 06306 09763 09693 0922919 Contrast adjustment 09997 09997 09864 0992020 Mosaic 09188 09702 09704 0960221 Manual cropping -07885 -09530 05101 -0980622 Dilation 05384 04058 00300 0550523 Bit plane removal -09689 -09648 -09742 -0987324 Row column copying 09987 09997 10000 10000

6 CONCLUSION

In this paper a novel yet simple hybrid nonsubsampled contourlet domain SVD basedwatermarking scheme for image copyright protection is proposed where the singular values of low frequency sub band coefficients of watermark image are embedded on the singular values of low frequency sub band coefficients of original image with an appropriate scaling factor Theproposed algorithm preserves high perceptual quality of the watermarked image and shows anexcellent robustness to attacks like JPEG JPEG2000 compressions Low pass filteringHistogram equalization Gray scale inversion Salt and Pepper Noise Soft Thresholding Weiner

Filtering Gaussian Noise Rescaling and Contrast adjustment This algorithm is quite resilient toRotation Median filtering Alpha mean Gamma correction Sharpening Blurring CroppingDilation Int thresholding Mosaic and bit plane removal attacks The proposed algorithm achieves higher PSNR when compared with Emir Ganic and Ahmet MEskicioglursquos paper Itdemonstrates that nonsubsampled contourlet transform domain performs better than waveletdomain The proposed algorithm is also tested for different original images and respectivewatermarked images are obtained without any visual degradation




52

7 REFERENCES

[1] Santi PMaity Prasun Nandy Tirtha SDas and Maly KKundu ldquoRobust image watermarking usingmultiresolution analysisrdquo IEEE INDIA Annual Conference 2004INDICON 2004

[2] Emir Ganic and ahmet M Eskicioglu ldquo Robust embedding of visual watermarks using discretewavelet transform and singular value decomposition Journal Of Electron Imaging Vol 14043004 (2005) doi10111712137650 Published 12 December 2005

[3] Alexander Sverdlov Scott Dexter and Ahmet MEskicioglu ldquoRobust DCT_SVD domain imageWatermarking for copyright protection embedding data in all frequenciesrdquo

[4] CVenkata Narasimhulu and KSatya Prasad ldquoA hybrid watermarking scheme using contourletTransform and Singular value decompositionrdquo IJCSNS International Journal of Computer Science andNetwork Security Vol10No9 September 2010

[5] Minh N Do and Martin Vetterli ldquoThe Contourlet Transform An Efficient DirectionalMultiresolution Image Representationrdquo IEEE transaction on image processingvol 14issue no 12pp2091-2106Dec 2005

[6] Elham salahi MShahram Moin and Ahmad salahi ldquoA new Visually Imperceptible and Robust ImageWater marking Scheme in contourlet Domainrdquo International conference on intelligent information hidingand multimedia signal processing2008

[7] Jianping Zhou Cunha AL MNDo ldquoNonsubsampled contourlet transform construction andapplication in enhancementrdquo IEEE Trans Image Proc Sept 2005

[8] Arthur L Cunha J Zhou and M N Do ldquoNonsubsampled contourlet transform filter design andapplications in denoisingrdquo IEEE international conference on image processing September 2005

[9] VIGorodetski LJPopyack and VSamoilov ldquoSVD-based approach to transparent embedding datainto digital imagesrdquo in proc int workshop MMM-ACNS StPeterburg Russia May 2001pp263-27410 RLiu and TTan ldquoAn SVD-Based Watermarking scheme for protecting rightful ownershiprdquo IEEE TransMultimedia vol4 no1 pp121-128 Mar2002

[11] Paul Bao and Xiaohu Ma ldquoImage adaptive watermarking using wavelet domain singular valuedecompositionrdquo IEEE Transaction on circuit and system for video technology vol15 no1 January 2005

[12] A L Cunha J Zhou and M N Do ldquoThe Nonsubsampled contourlet transform theory design andapplicationsrdquo IEEE Trans Image Proc May 2005

[13] A L Cunha J Zhou and M N Do ldquoThe Nonsubsampled contourlet transform theory design andapplicationsrdquo IEEE Trans Image Proc vol15 no10 October 2006

[14] Xiang Yang Wang Yi-Ping Yang and Hong-Ying Yang ldquoA novel nonsubsampled contourlet domainImage watermarking using Support Vector Regressionrdquo Journal of Optics A Pure and Applied OpticsSeptember 2009

[15] MJShenshardquoThe discrete Wavelet Transform Wedding the A Trous and Mallat algorithmsrdquo IEEETrans vol 40 no 10 Pp2464-2482 Oct 1992

[16] RHBamberger and MJTSith ldquoAfilter bank for the directional decomposition of images Theoryand Designrdquo IEEE Trans Signal Processing vol40no4pp882-893 Apr 1992

[17] BChandra Mohan and SSrinivas Kumar ldquoA Robust Image watermarking scheme using Singularvalue decompositionrdquo Journal of MultimediaVol3NO1May 2008

[18] Ke-Feng HeJun GaoLiang-Mei Hu ldquoWatermarking for images using the HVS and SVD in theWavelet Domainrdquo Procedings fo 2006 IEEE International on Mechatronics and Automation June 25 to282006 LuoyangChaina




53

[19] Salwa AKMostafa ASTolba FMAbdelkader Hisham MElhindy ldquoVideo WatermarkingScheme based on Principal Component Analysis and Wavelet Transformrdquo IJCSNS International Journalof Computer Science and Network Security Volume9 No 8 August 2009

[20] Ashraf K Helmy and GHSEl-Taweel ldquoAuthentication Scheme Based on Principal ComponentAnalysis for Satellite Imagesrdquo International Journal of Signal Processing Image Processing and PatternRecognition Vol 2 No3 September 2009

[21] Matlab 76 version Image Processing Tool Box

Authors

CV Narasimhulu

He received his Bachelor degree in Electronics and Communication Engineering fromSV University Tirupati India in 1995 and Master of technology in Instruments andControl Systems from Regional Engineering College Calicut India in 2000He iscurrently pursuing the PhD degree in the department of Electronics and CommunicationEngineering from Jawaharlal Nehru Technological University Kakinada India He has

more than 15 years experience of teaching under graduate and post graduate level He isinterested in the areas of signal processing and multimedia security

KSatya Prasad

He r eceived his PhD degree from IIT Madras India He is presently working asprofessor in the department of Electronics and Communication Engineering JNTUcollege of Engineering Kakinada and Rector of Jawaharlal Nehru TechnologicalUniversity Kakinada India He has more than 30 years of teaching and researchexperience He published 30 research papers in international and 20 research papers inNational journals He guided 8 PhD theses and 20 PhD theses are under his guidanceHis area of interests includes Digital Signal and Image Processing CommunicationsAdhoc networks etc




38

Watermarking techniques may be classified in different ways The classification may be based onthe type of watermark being used ie the watermark may be a visually recognizable logo orsequence of random numbers A second classification is based on whether the watermark isapplied in the spatial domain or the transform domain In spatial domain the simplest method isbased on embedding the watermark in the least significant bits (LSB) of image pixels Howeverspatial domain techniques are not resistant enough to image compression and other imageprocessing operations

Transform domain watermarking schemes such as those based on the discrete cosine transform(DCT) the discrete wavelet transform (DWT) contourlet transforms along with numericaltransformations such as Singular value Decomposition (SVD) and Principle component analysis(PCA) typically provide higher image fidelity and are much robust to image manipulations[4]Of the so far proposed algorithms wavelet domain algorithms perform better than other transformdomain algorithms since DWT has a number of advantages over other transforms including timefrequency localization multi resolution representation superior HVS modeling and linearcomplexity and adaptively and it has been proved that wavelets are good at representing pointwise discontinuities in one dimensional signal However in higher dimensions eg image thereexists line or curve-shaped discontinuities Since 2D wavelets are produced by tensor products of 1D wavelets they can only identify horizontal vertical diagonal discontinuities (edges) inimages ignoring smoothness along contours and curves Curvelet transform was defined torepresent two dimensional discontinuities more efficiently with least square error in a fixed termapproximation Curvelet transform was proposed in continuous domain and its discretisation wasa challenge when critical sampling is desired Contourlet transform was then proposed by DO andVetterli as an improvement of Curvelet transform The Contourlet transform is a directional multiresolution expansion which can represents images contains contours efficiently The CT employsLaplacian pyramids to achieve multi resolution decomposition and directional filter banks toachieve directional decomposition [4][5][6] Due to down sampling and up sampling theContourlet transform is Shift variant However shift invariance is desirable in image analysisapplications such as edge detection Contour characterization image enhancement [7] and image

watermarking Here we present a NonSubsampled Contourlet transform (NSCT) [8] which isshift invariant version of the contourlet transform The NSCT is built upon iteratednonsubsampled filter banks to obtain a shift invariant image representation

In all above transform domain watermarking techniques including NSCT the watermarking bitswould be directly embedded in the locations of sub band coefficients Though here the visual of perception of original image is preserved the watermarked image when subjected to someintentional attacks like compression the watermark bits will get damaged Coming to the spatialdomain watermarking using numerical transformation like SVD (Gorodetski [9] liu et al [10])they provide good security against tampering and common manipulations for protecting rightfulownership But these schemes are non adaptive thus unable to offer consistent perceptualtransparency of watermarking of different images [11]To provide adaptive transparency

robustness to the compressions and insensitivity to malicious manipulations we propose a novelimage hybrid watermarking scheme using NSCT and SVD

In this paper proposed method is compared with another which is based on discrete wavelettransform and singular value decomposition (DWT-SVD)The peak signal noise ratio (PSNR)between the original image and watermarked image and the normalized correlation coefficients(NCC) after different attacks were calculated The results show high improvement detectionreliability using proposed methodThe rest of this paper is organized as follows Section2




39



(a) (b)







40










41

(a) (b)


directional bank





A=USV T


λ 1

λ

2

λ n




42




a)


b)


c)


d)


e)



1


2












43















44









using Step 3





45







376102










46






ExtractedWatermark

Ncc=1




ExtractedWatermark

Ncc=1




ExtractedWatermark

Ncc=1




47




PSNR= 377705

ExtractedWatermark

Ncc=1




ExtractedWatermark

Ncc=1

+




Low pass filtering

Ncc= 09793Rotation

Ncc= -04239




48



Ncc= 08636

Alpha mean


Ncc= 10000



Ncc= 09982

Weiner filter


Ncc= 04734




49


RescalingNcc= 10000

Sharpening

Ncc= 05352Blurring

Ncc=09229

Contrast adjustment


Ncc=-09798

Int thresholding





50

DilationNcc= 05505

MosaicNcc=09602

Bit plane removal


Ncc=10000








51



compression08853 09501 09492 09793


equalization09505 09537 08238 09722



6 CONCLUSION






52

7 REFERENCES





















53




Authors

CV Narasimhulu



KSatya Prasad





39



(a) (b)







40










41

(a) (b)


directional bank





A=USV T


λ 1

λ

2

λ n




42




a)


b)


c)


d)


e)



1


2












43















44









using Step 3





45







376102










46






ExtractedWatermark

Ncc=1




ExtractedWatermark

Ncc=1




ExtractedWatermark

Ncc=1




47




PSNR= 377705

ExtractedWatermark

Ncc=1




ExtractedWatermark

Ncc=1

+




Low pass filtering

Ncc= 09793Rotation

Ncc= -04239




48



Ncc= 08636

Alpha mean


Ncc= 10000



Ncc= 09982

Weiner filter


Ncc= 04734




49


RescalingNcc= 10000

Sharpening

Ncc= 05352Blurring

Ncc=09229

Contrast adjustment


Ncc=-09798

Int thresholding





50

DilationNcc= 05505

MosaicNcc=09602

Bit plane removal


Ncc=10000








51



compression08853 09501 09492 09793


equalization09505 09537 08238 09722



6 CONCLUSION






52

7 REFERENCES





















53




Authors

CV Narasimhulu



KSatya Prasad





40










41

(a) (b)


directional bank





A=USV T


λ 1

λ

2

λ n




42




a)


b)


c)


d)


e)



1


2












43















44









using Step 3





45







376102










46






ExtractedWatermark

Ncc=1




ExtractedWatermark

Ncc=1




ExtractedWatermark

Ncc=1




47




PSNR= 377705

ExtractedWatermark

Ncc=1




ExtractedWatermark

Ncc=1

+




Low pass filtering

Ncc= 09793Rotation

Ncc= -04239




48



Ncc= 08636

Alpha mean


Ncc= 10000



Ncc= 09982

Weiner filter


Ncc= 04734




49


RescalingNcc= 10000

Sharpening

Ncc= 05352Blurring

Ncc=09229

Contrast adjustment


Ncc=-09798

Int thresholding





50

DilationNcc= 05505

MosaicNcc=09602

Bit plane removal


Ncc=10000








51



compression08853 09501 09492 09793


equalization09505 09537 08238 09722



6 CONCLUSION






52

7 REFERENCES





















53




Authors

CV Narasimhulu



KSatya Prasad





41

(a) (b)


directional bank





A=USV T


λ 1

λ

2

λ n




42




a)


b)


c)


d)


e)



1


2












43















44









using Step 3





45







376102










46






ExtractedWatermark

Ncc=1




ExtractedWatermark

Ncc=1




ExtractedWatermark

Ncc=1




47




PSNR= 377705

ExtractedWatermark

Ncc=1




ExtractedWatermark

Ncc=1

+




Low pass filtering

Ncc= 09793Rotation

Ncc= -04239




48



Ncc= 08636

Alpha mean


Ncc= 10000



Ncc= 09982

Weiner filter


Ncc= 04734




49


RescalingNcc= 10000

Sharpening

Ncc= 05352Blurring

Ncc=09229

Contrast adjustment


Ncc=-09798

Int thresholding





50

DilationNcc= 05505

MosaicNcc=09602

Bit plane removal


Ncc=10000








51



compression08853 09501 09492 09793


equalization09505 09537 08238 09722



6 CONCLUSION






52

7 REFERENCES





















53




Authors

CV Narasimhulu



KSatya Prasad





42




a)


b)


c)


d)


e)



1


2












43















44









using Step 3





45







376102










46






ExtractedWatermark

Ncc=1




ExtractedWatermark

Ncc=1




ExtractedWatermark

Ncc=1




47




PSNR= 377705

ExtractedWatermark

Ncc=1




ExtractedWatermark

Ncc=1

+




Low pass filtering

Ncc= 09793Rotation

Ncc= -04239




48



Ncc= 08636

Alpha mean


Ncc= 10000



Ncc= 09982

Weiner filter


Ncc= 04734




49


RescalingNcc= 10000

Sharpening

Ncc= 05352Blurring

Ncc=09229

Contrast adjustment


Ncc=-09798

Int thresholding





50

DilationNcc= 05505

MosaicNcc=09602

Bit plane removal


Ncc=10000








51



compression08853 09501 09492 09793


equalization09505 09537 08238 09722



6 CONCLUSION






52

7 REFERENCES





















53




Authors

CV Narasimhulu



KSatya Prasad





43















44









using Step 3





45







376102










46






ExtractedWatermark

Ncc=1




ExtractedWatermark

Ncc=1




ExtractedWatermark

Ncc=1




47




PSNR= 377705

ExtractedWatermark

Ncc=1




ExtractedWatermark

Ncc=1

+




Low pass filtering

Ncc= 09793Rotation

Ncc= -04239




48



Ncc= 08636

Alpha mean


Ncc= 10000



Ncc= 09982

Weiner filter


Ncc= 04734




49


RescalingNcc= 10000

Sharpening

Ncc= 05352Blurring

Ncc=09229

Contrast adjustment


Ncc=-09798

Int thresholding





50

DilationNcc= 05505

MosaicNcc=09602

Bit plane removal


Ncc=10000








51



compression08853 09501 09492 09793


equalization09505 09537 08238 09722



6 CONCLUSION






52

7 REFERENCES





















53




Authors

CV Narasimhulu



KSatya Prasad





44









using Step 3





45







376102










46






ExtractedWatermark

Ncc=1




ExtractedWatermark

Ncc=1




ExtractedWatermark

Ncc=1




47




PSNR= 377705

ExtractedWatermark

Ncc=1




ExtractedWatermark

Ncc=1

+




Low pass filtering

Ncc= 09793Rotation

Ncc= -04239




48



Ncc= 08636

Alpha mean


Ncc= 10000



Ncc= 09982

Weiner filter


Ncc= 04734




49


RescalingNcc= 10000

Sharpening

Ncc= 05352Blurring

Ncc=09229

Contrast adjustment


Ncc=-09798

Int thresholding





50

DilationNcc= 05505

MosaicNcc=09602

Bit plane removal


Ncc=10000








51



compression08853 09501 09492 09793


equalization09505 09537 08238 09722



6 CONCLUSION






52

7 REFERENCES





















53




Authors

CV Narasimhulu



KSatya Prasad





45







376102










46






ExtractedWatermark

Ncc=1




ExtractedWatermark

Ncc=1




ExtractedWatermark

Ncc=1




47




PSNR= 377705

ExtractedWatermark

Ncc=1




ExtractedWatermark

Ncc=1

+




Low pass filtering

Ncc= 09793Rotation

Ncc= -04239




48



Ncc= 08636

Alpha mean


Ncc= 10000



Ncc= 09982

Weiner filter


Ncc= 04734




49


RescalingNcc= 10000

Sharpening

Ncc= 05352Blurring

Ncc=09229

Contrast adjustment


Ncc=-09798

Int thresholding





50

DilationNcc= 05505

MosaicNcc=09602

Bit plane removal


Ncc=10000








51



compression08853 09501 09492 09793


equalization09505 09537 08238 09722



6 CONCLUSION






52

7 REFERENCES





















53




Authors

CV Narasimhulu



KSatya Prasad





46






ExtractedWatermark

Ncc=1




ExtractedWatermark

Ncc=1




ExtractedWatermark

Ncc=1




47




PSNR= 377705

ExtractedWatermark

Ncc=1




ExtractedWatermark

Ncc=1

+




Low pass filtering

Ncc= 09793Rotation

Ncc= -04239




48



Ncc= 08636

Alpha mean


Ncc= 10000



Ncc= 09982

Weiner filter


Ncc= 04734




49


RescalingNcc= 10000

Sharpening

Ncc= 05352Blurring

Ncc=09229

Contrast adjustment


Ncc=-09798

Int thresholding





50

DilationNcc= 05505

MosaicNcc=09602

Bit plane removal


Ncc=10000








51



compression08853 09501 09492 09793


equalization09505 09537 08238 09722



6 CONCLUSION






52

7 REFERENCES





















53




Authors

CV Narasimhulu



KSatya Prasad





47




PSNR= 377705

ExtractedWatermark

Ncc=1




ExtractedWatermark

Ncc=1

+




Low pass filtering

Ncc= 09793Rotation

Ncc= -04239




48



Ncc= 08636

Alpha mean


Ncc= 10000



Ncc= 09982

Weiner filter


Ncc= 04734




49


RescalingNcc= 10000

Sharpening

Ncc= 05352Blurring

Ncc=09229

Contrast adjustment


Ncc=-09798

Int thresholding





50

DilationNcc= 05505

MosaicNcc=09602

Bit plane removal


Ncc=10000








51



compression08853 09501 09492 09793


equalization09505 09537 08238 09722



6 CONCLUSION






52

7 REFERENCES





















53




Authors

CV Narasimhulu



KSatya Prasad





48



Ncc= 08636

Alpha mean


Ncc= 10000



Ncc= 09982

Weiner filter


Ncc= 04734




49


RescalingNcc= 10000

Sharpening

Ncc= 05352Blurring

Ncc=09229

Contrast adjustment


Ncc=-09798

Int thresholding





50

DilationNcc= 05505

MosaicNcc=09602

Bit plane removal


Ncc=10000








51



compression08853 09501 09492 09793


equalization09505 09537 08238 09722



6 CONCLUSION






52

7 REFERENCES





















53




Authors

CV Narasimhulu



KSatya Prasad





49


RescalingNcc= 10000

Sharpening

Ncc= 05352Blurring

Ncc=09229

Contrast adjustment


Ncc=-09798

Int thresholding





50

DilationNcc= 05505

MosaicNcc=09602

Bit plane removal


Ncc=10000








51



compression08853 09501 09492 09793


equalization09505 09537 08238 09722



6 CONCLUSION






52

7 REFERENCES





















53




Authors

CV Narasimhulu



KSatya Prasad





50

DilationNcc= 05505

MosaicNcc=09602

Bit plane removal


Ncc=10000








51



compression08853 09501 09492 09793


equalization09505 09537 08238 09722



6 CONCLUSION






52

7 REFERENCES





















53




Authors

CV Narasimhulu



KSatya Prasad





51



compression08853 09501 09492 09793


equalization09505 09537 08238 09722



6 CONCLUSION






52

7 REFERENCES





















53




Authors

CV Narasimhulu



KSatya Prasad





52

7 REFERENCES





















53




Authors

CV Narasimhulu



KSatya Prasad





53




Authors

CV Narasimhulu



KSatya Prasad


A Novel Robust Watermarking Technique Based on Nonsubsampled Contourlet Transform and SVD

Documents