(IJCSIS)
International
Journal
of
Computer
Science
and
Information
Security,
Vol.
9,
No.
3,
March
2011
Robust Color Image Watermarking UsingNonsubsampled Contourlet Transform
C.Venkata Narasimhulu K.Satya Prasad
Professor, Dept of ECE Professor, Dept of ECE,HIET, Hyderabad, India JNTU Kakinada, Indianarasimhulucv@gmail.com prasad_kodati@yahoo.co.in,
Abstract-
In this paper, we propose a novel hybrid spreadspectrum watermarking scheme for authentication of color images using nonsubsampled contourlet transformand singular value decomposition. The host color imageand color watermark images are decomposed intodirectional sub- bands using Nonsubsampled contourlettransform and then applied Singular value decompositionto mid frequency sub-band coefficients. The singularvalues of mid frequency sub-band coefficients of colorwatermark image are embedded into singular values of mid frequency sub-band coefficients of host color image inRed, Green and Blue color spaces simultaneously based onspread spectrum technique. The experimental resultsshows that the proposed hybrid watermarking scheme isrobust against common image processing operations suchas, JPEG, JPEG 2000 compression, cropping, Rotation,histogram equalization, low pass filtering ,medianfiltering, sharpening, shearing ,salt & Pepper noise,Gaussian noise, grayscale conversion etc. It has also beenshown the variation of visual quality of watermarkedimage for different scaling factors. The comparativeanalysis reveals that the proposed watermarking schemeout performs the color image watermarking schemesreported recently.
Keywords: Color image watermarking, Nonsubsampled Contourlet Transform, Singular value decomposition, Peaksignal to noise ratio, normalized Correlation coefficient.
1. INTRODUCTION:In recent years, multimedia products were rapidlydistributed over the fast communication systems suchas Internet, so there exist strong requirement to protectthe ownership and authentication of the multimediadata. Digital watermarking is a method of securing thedigital data by embedding additional information calledwater mark into the digital multimedia content. Thisembedding information can be later extracted from or detected in the multimedia to make an assertion aboutthe data authenticity. Digital watermarks remain intactunder transmission/transformation, allowing us toprotect our ownership rights in digital form. Absence of watermark in a previously watermarked image wouldlead to the conclusion that the data content has beenmodified. A watermarking algorithm consists of watermark structure, an embedding algorithm andextraction or detection algorithm. In multimediaapplications, embedded watermark should be invisible,robust and have a high capacity. Invisibility refers todegree of distortion introduced by the watermark and itsaffect on the viewers and listeners. Robustness is theresistance of an embedded watermark againstintentional attack and normal signal processingoperations such as noise, filtering, rotation, scaling,cropping and lossey compression etc. Capacity is theamount of data can be represented by embeddedwatermark.[1]Watermarking techniques may be classified indifferent ways. The classification may be based on thetype of watermark being used, i.e., the watermark maybe a visually recognizable logo or sequence of randomnumbers. A second classification is based on whether the watermark is applied in the spatial domain or thetransform domain. In spatial domain, the simplestmethod is based on embedding the watermark in theleast significant bits (LSB) of image pixels. However,spatial domain techniques are not resistant enough toimage compression and other image processingoperations.Transform domain watermarking schemes such asthose based on the discrete cosine transform (DCT), thediscrete wavelet transform (DWT), contourlettransforms along with numerical transformations suchas Singular value Decomposition (SVD) and Principlecomponent analysis (PCA) typically provide higher image fidelity and are much robust to imagemanipulations.[2]Of the so far proposed algorithms,wavelet domain algorithms perform better than other transform domain algorithms since DWT has a number of advantages over other transforms including timefrequency localization, multi resolution representation,superior HVS modeling, and linear complexity andadaptively and it has been proved that wavelets aregood at representing point wise discontinuities in onedimensional signal. However, in higher dimensions,e.g. image, there exists line or curve-shapeddiscontinuities. Since, 2D wavelets are produced bytensor products of 1D wavelets; they can only identifyhorizontal, vertical, diagonal discontinuities (edges) inimages, ignoring smoothness along contours andcurves. Curvelet transform was defined to represent two
100 http://sites.google.com/site/ijcsis/ISSN 1947-5500
(IJCSIS)
International
Journal
of
Computer
Science
and
Information
Security,
Vol.
9,
No.
3,
March
2011
dimensional discontinuities more efficiently, with leastsquare error in a fixed term approximation. Curvelettransform was proposed in continuous domain and itsdiscretisation was a challenge when critical sampling isdesired. Contourlet transform was then proposed by DOand Vetterli as an improvement of Curvelet transform.The Contourlet transform is a directional multiresolution expansion which can represents imagescontains contours efficiently. The CT employsLaplacian pyramids to achieve multi resolutiondecomposition and directional filter banks to achievedirectional decomposition [3] Due to down samplingand up sampling, the Contourlet transform is Shiftvariant. However shift invariance is desirable in imageanalysis applications such as edge detection, Contour characterization, image enhancement [4] and imagewatermarking. Here, we present a NonSubsampledContourlet transform (NSCT) [5] which is shiftinvariant version of the contourlet transform. TheNSCT is built upon iterated nonsubsampled filter banksto obtain a shift invariant image representation.In all above transform domain watermarking techniquesincluding NSCT the watermarking bits would bedirectly embedded in the locations of sub bandcoefficients. Though here the visual of perception of original image is preserved, the watermarked imagewhen subjected to some intentional attacks likecompression the watermark bits will get damaged.Coming to the spatial domain watermarking usingnumerical transformation like SVD (Gorodetski [6], liuet al [7]) they provide good security against tamperingand common manipulations for protecting rightfulownership. But these schemes are non adaptive, thusunable to offer consistent perceptual transparency of watermarking of different images. To provide adaptivetransparency, robustness to the compressions andinsensitivity to malicious manipulations, we propose anovel image hybrid watermarking scheme using NSCTand SVD.In this paper, proposed method is compared withanother which is based on Contourlet Transform andsingular value decomposition (CT-SVD). The peak signal to noise ratio (PSNR) between the original imageand watermarked image and the normalized correlationcoefficients (NCC) and bit error rate (BER) betweenthe original watermark and extracted were calculatedwith and without attacks. The results show highimprovement detection reliability using proposedmethod. The rest of this paper is organized as follows.Section 2 describes the Nonsubsampled contourlettransform, section 3 describes singular valuedecomposition, section 4 illustrates the details of proposed method, in section 5 experimental results arediscussed without and with attacks, conclusion andfuture scope are given in section 6.2. NONSUBSAMPLED CONTOURLETTRANSFORMThe Nonsubsampled contourlet transform is a newimage decomposition scheme introduced by Arthur L.Cunha, Jianping Zhou and Minh N.Do [8]. NSCT ismore effective in representing smooth contours indifferent directions of in an image than contourlettransform and discrete wavelet transform. The NSCT isfully shift invariant, Multi scale and multi directionexpansion that has a fast implementation. The NSCTexhibits a similar sub band decomposition as that of contourlets, but without down samplers and up samplersin it. Because of its redundancy the filter design problemof nonsubsampled contourlet is much less constrainedthan that of contourlet. The NSCT is constructed bycombining nonsubsampled pyramids andnonsubsampled directional filter bank as shown infigure (1).The nonsubsampled pyramid structure resultsthe multi scale property and nonsubsampled directionalfilter bank results the directional property.
(a) (b)Figure 1 The nonsubsampled contourlet transform (a)nonsubsampled filter bank structure that implements the NSCT.(b) Idealized frequency partitioning obtained with NSCT
2.1 Nonsubsampled Pyramids
The nonsubsampled pyramid is a two channelnonsubsampled filter bank as shown in figure2(a).The H
0
(z) is the low pass filter and one then setsH
1
(z) =1-H
0
(z). the corresponding synthesis filtersG
0
(z) =G
1
(z)=1.the perfect reconstruction condition is given by
Bezout identity
H
0
(z)G
0
(z)+H
1
(Z) G
1
(Z) =1………………(1)
101 http://sites.google.com/site/ijcsis/ISSN 1947-5500
(IJCSIS)
International
Journal
of
Computer
Science
and
Information
Security,
Vol.
9,
No.
3,
March
2011
3
:
(a) (b)Figure (2): Nonsubsampled pyramidal filter (a). Ideal frequency response of nonsubsampled pyramidal filter (b).The cascading analysis of three stages nonsubsampled pyramid by iteration of two channelsNonsubsampled filter banks
.Multi scale decomposition is achieved fromnonsubsampled pyramids by iterating thenonsubsampled filter banks by up sampling all filtersby 2 in both direction the next level decomposition isachieved. The complexity of filtering is constantwhether the filtering is with H(z) or an up sampledfilter H(z
m
) computed a Trous algorithm Thecascading of three stage analysis part is shown infigure 2( b)
2.2 Nonsubsampled directional Filter Banks:
The directional filter bank (DFB) is constructed fromthe combination of critically-sampled two-channelfan filter banks and resampling operations. Theoutcome of this DFB is a tree-structured filter bank splitting the 2-D frequency plane into wedges. Thenonsubsampled directional filter bank which is shiftinvariant is constructed by eliminating the down andup samplers in the DFB.The ideal frequency responseof nonsubsampled filter banks is shown in figure3 (a)To obtain multi directional decomposition, thenonsubsampled DFBs are iterated. To obtain thenext level decomposition, all filters are upsampled by a quincunx matrix given byQ =……………..(2)The analysis part of iterated nonsubsampled filter bank is shown in figure 3 (b)
(a) (b)Figure (3) Nonsubsampled directional filter bank (a) idealized frequency response of nonsubsampled directional filter bank.(b) Theanalysis part of an iterated nonsubsampled directional bank.
3. SINGULAR VALUE DECOMPOSITIONSingular value decomposition (SVD) is apopular technique in linear algebra and it hasapplications in matrix inversion, obtaining lowdimensional 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
1
1
1
‐
1
102 http://sites.google.com/site/ijcsis/ISSN 1947-5500
Search History:
Result 00 of 00
00 results for result for
p.
Robust Color Image Watermarking Using Nonsubsampled Contourlet Transform
In this paper, we propose a novel hybrid spread spectrum watermarking scheme for authentication of color images using nonsubsampled contourlet transform and singular value decomposition. The host color image and color watermark images are decomposed into…
(More)
801
Reads
0
Readcasts
0
Embed Views
More From This User
Notes
Load more
