CN101308566B - Contourlet transform based digital image watermarking method resisting geometrical attack - Google Patents

Abstract

The invention discloses a contourlet-transform-based anti-geometric attack digital image watermarking method, mainly solving the problem of poor robustness of the existing similar method. The invention is technically characterized in that the method completes watermark synchronization through a geometric moment method to realize watermark embedding and extraction in a contourlet transform domain,that is, to calculate a first-order origin moment containing a watermark image I when the watermark is embedded, and reserve the first-order origin moment for the estimation of the geometric attack parameters during extraction; during watermark extraction, to estimate the geometric transformation parameters through the attacked first-order origin moment and a second-order center moment containingwatermarks, and to recover the attacked watermarks through contourlet transform. The method described by the invention has the advantages of strong anti-geometric attack ability, good watermark extracting effect and self-positioning watermark embedded location, which can be used for safety protection of digital multimedia product copyright.

Description

Based on contourlet conversion resist geometric attacks digital image watermarking method

Technical field

The invention belongs to the Image Information Processing field, relate to digital image watermarking method, particularly relate to digital watermark method against geometrical attack, can be used for providing technique guarantee the security of digital media product copyright based on the contourlet conversion.

Background technology

The digital watermark technology that with the image is carrier is one of emphasis of current digital watermark research.In recent years, digital watermark technology research has been made significant headway, and is especially various at the water mark method of view data, is subjected to extensive studies, but the robustness of watermaking system is the obstacle of its application of restriction always.Wherein, can the watermark synchronization under geometric transformation is attacked becomes digital watermark move towards commercial determinative, thereby also becomes the emphasis of current digital watermarking area research gradually.In order to address this problem, numerous scholars have carried out research and have proposed certain methods to this, and for example: (1) is embedded into watermark in how much fields of invariants, carry out watermark at the field of invariants of image and embed.In the amplitude space that watermark is embedded in Fourier transform, this space has rotation, convergent-divergent, translation invariance.But this method complexity is higher, and the ability of signal Processing such as the anti-lossy compression method of watermark, filtering is also relatively poor.(2) in image, except that embed watermark, also embed a template and be used to resist geometric attack.Template can be made up of the extreme point in the DFT amplitude spectrum.But the deficiency of these class methods is to embed under the certain situation of capacity at image, known template be embedded in the robustness that can reduce watermark information to a certain extent; In case and template victim destruction, the testing process of watermark just can't be carried out.(3) watermark is embedded in the carrier data with a discernible structure.The major defect of these class methods is easy influences by compression.

In sum, existing its implementation procedure of digital watermark method that can resist geometric transformation operation is more complicated mostly, and poor robustness, and people have proposed based on contourlet changed digital water mark method for this reason.

The Contourlet conversion is a kind of new expansion of wavelet transformation, is a kind of multiple dimensioned, localization, multidirectional graphical representation method, has developed into the 2D signal method for expressing that can catch geometry of a kind of " really ".Catch advantages such as effective, antinoise attack, conversion rate are fast because the contourlet conversion has details, begun to be applied to digital watermark and carried out certain research both at home and abroad at present.That delivers at present mainly contains based on contourlet changed digital water mark method:

(1). Li Haifeng [1] proposes a kind of method that directly is superimposed upon watermark on the big coefficient in contourlet territory and adopts blind signal separation technology realization watermark extracting, and the embedding effect of this method is better, but robustness is not good enough;

(2) .Zhao Xu[2] on this basis embedding is encrypted in watermark,

(3) .Nadia Baaziz[3] the wavelet field vision mode is introduced the contourlet territory, through revising a kind of contourlet territory watermarking algorithm based on the HVS characteristic is proposed, make the embedding of contourlet territory watermark realize self-adaptation;

(4) .Jayalakshmi M.[4] wavelet field watermark and the watermark of contourlet territory are compared, prove that the contourlet territory has effect preferably aspect sheltering and the robust shape really.

More than these relate to the method for contourlet territory watermark, having unified defective is to resist geometric attack effect heterodyne, promptly can not resist the attack than wide-angle rotation, large scale convergent-divergent and affined transformation.

[1] Li Haifeng, Song Weiwei, Wang Shuxun, based on the robustness image watermark algorithm of Contourlet conversion, communication journal, 2007 the 4th volumes, 87-94

[2]Zhao?Xu，Ke?Wang?and?Xiao-hua?Qiao，Novel?Watermarking?Scheme?in?ContourletDomain?Based?on?Independent?Component?Analysis，computer?society，2006

[3]Nadia?Baaziz，Adaptive?Watermarking?Schemes?Based?On?A?Redundant?ContourletTransform，IEEE，2005

[4]Jayalakshmi?M.，S.N.Merchant?and?U.B.Desai，Blind?Watermarking?in?ContourletDomain?with?Improved?Detection，computer?society，2006

The content of invention

The present invention seeks to deficiency at above-mentioned prior art, propose a kind of based on contourlet conversion resist geometric attacks digital image watermarking method, to realize reliably protecting to the digital product copyright.

The key problem in technology of realizing the object of the invention is to calculate the first moment about the origin that contains watermarking images I when watermark embeds, and to the estimation of suffered geometric attack parameter, realizes the ability of resist geometric attacks when staying to extraction.By the first moment about the origin and the second-order moment around mean that contain watermarking images after attacking are estimated the geometric transformation parameter, and utilize the contourlet transfer pair to be attacked the back watermark and restore when watermark extracting, concrete side is by as follows:

One, watermark embed process

(1) host image I is carried out contourlet and decompose, obtain the subband B (i) of low frequency, medium and low frequency, medium-high frequency and high frequency level and the conversion coefficient x of each subband _i

(2) watermarking images W is carried out contourlet and decompose, obtain the subband B of the frequency level of low frequency, intermediate frequency and high frequency _w(i) and the conversion coefficient y of each subband _i

(3) according to the conversion coefficient x of host image _i, the energy of every layer of each subband of calculating host image I: E _i=∑ (x _i) ², i=1,2,3...n;

(4) to the ENERGY E of every straton band _i, according to sequence notation from big to small, the subband of selecting the energy maximum is that watermark embeds target subband B _Max

(5) with the conversion coefficient y of every layer of watermark _iArrange from big to small according to absolute value, every layer the matrix that puts in order is designated as the first key R of watermark extracting;

(6) with the maximum subband B of every layer of energy _MaxIn conversion coefficient arrange from big to small by absolute value, and S coefficient before selecting, according to each conversion coefficient y of every layer of watermark _iThe relation stack that size is corresponding is about to watermark and is embedded in the host image, and the value of this S is watermarking images W to coefficient number that should the straton band;

(7) the host image conversion coefficient with embed watermark carries out the contourlet inverse transformation, reconstruct the image I that contains watermark ';

(8) calculate the first moment about the origin that contains watermarking images I ': m _1,0, m _0,1And m _0,0, and deposit them in the 1*3 matrix, be designated as the second key X, stay when extracting parameter estimation to use.

Two, watermark extraction process

1) calculate after attacking and contain watermarking images I " first moment about the origin m _1,0', m _0,1', m _0,0' and second-order moment around mean μ _1,1', μ _2,0', μ _0,2', and utilize following formula to estimate the geometric parameter of the suffered geometric attack of image, promptly

Angle parameter ${\mathrm{\θ}}_c=\frac{1}{2}{\tan }^{-1}\frac{2{\mathrm{\μ}}_{\mathrm{1,1}}^{\′}}{{\mathrm{\μ}}_{\mathrm{2,0}}^{\′}-{\mathrm{\μ}}_{\mathrm{0,2}}^{\′}},$

The horizontal scaling parameter $a=\frac{m_{\mathrm{0,0}}{m}_{\mathrm{1,0}}^{\′}}{m_{\mathrm{0,0}}^{\′}{m}_{\mathrm{1,0}}},$ The vertically scale parameter $b=\frac{{m_{\mathrm{0,0}}^{\′}}^2{m}_{\mathrm{1,0}}}{m_{\mathrm{0,0}}^2{m}_{\mathrm{1,0}}^{\′}}$

The horizontal translation parameter $p=\frac{m_{\mathrm{1,0}}^{\′}-m_{\mathrm{1,0}}}{m_{0,0}},$ The vertical translation parameter $q=\frac{m_{\mathrm{0,1}}^{\′}-m_{\mathrm{0,1}}}{m_{0,0}}$

2) according to the suffered geometric attack parameter of image image I under fire " is carried out reverse operating, obtained attacking the reduction image I _r

3) will attack the reduction image I _rCarry out contourlet and decompose, obtain the subband B of low frequency, medium and low frequency, medium-high frequency and high frequency level _r(i) and the conversion coefficient x of each subband _i';

4) calculate attack reduction image I _rThe ENERGY E of every layer of each subband _i'=∑ (x _i') ², then by descending series arrangement subband B _r(i), i=1,2,3...n

5) with the maximum subband B of every layer of energy _RmaxIn coefficient arrange from big to small by absolute value, S coefficient before selecting is watermarking images W to coefficient number that should the straton band, the value of S is identical during with embedding;

6) with every layer of maximum subband B of energy among each coefficient of selected S and the former host image I _MaxSubtract each other after the middle coefficient ordering, obtain extracting every layer of conversion coefficient of back watermark under fire

y _i’＝(I _r-I)/α

It is identical when α is with embedding in the formula;

7) by every layer of conversion coefficient y of watermark after the first key R reduction under fire _i' position;

8) incite somebody to action back watermark conversion coefficient y under fire _i' carry out contourlet inverse transformation, the watermark that obtains extracting.

The present invention is owing to adopt the digital watermark technology be based on the contourlet conversion, watermark embedded have that computing velocity is fast, the advantage of good visual effect; Owing to control the conversion coefficient of selecting to be fit to embed watermark, solved the selection problem of watermark embedded location simultaneously, can independently select suitable coefficient, guaranteed the hidden of watermark in security by energy and coefficient magnitude; In addition since by to original contain watermarking images with under fire after contain the relatively calculating of geometric moment between the watermarking images, go back the suffered attack of original image, the realization image synchronization is finished watermark extracting, can effectively resist rotation, convergent-divergent, the various geometric attacks of translation.

Description of drawings

Fig. 1 is a watermark embed process block diagram of the present invention;

Fig. 2 is a watermark extraction process block diagram of the present invention;

The watermark result figure that Fig. 3 never under fire extracts in the image;

The watermark result figure that Fig. 4 extracts from compression and attacked by noise image;

The watermark result figure that Fig. 5 extracts from rotation attack graph picture;

The watermark result figure that Fig. 6 extracts from amplify 3 times of attack graph pictures;

The watermark result figure that Fig. 7 extracts from dwindle 0.8 times of attack graph picture;

The watermark result figure that Fig. 8 extracts from asymmetric amplification attack image;

The watermark result figure that Fig. 9 extracts from the complex map of strike picture.

Embodiment

One. the basic theory introduction

1, Contourlet conversion

The Contourlet conversion is a kind of new expansion of wavelet transformation, is a kind of multiple dimensioned, localization, multidirectional graphical representation method, and the 2D signal that can catch geometry that has developed into a kind of " really " is represented.Double-smoothing device group structure is adopted in the Contourlet conversion, and it is unusual with capture point at first to adopt the tower decomposition of Laplce that input picture is carried out multiple dimensioned decomposition.Decompose to generate a resolution be half low frequency sub-band and the high-frequency sub-band identical with original image resolution of original image to LP each time, and this high-frequency sub-band is original image and the filtered difference signal of low frequency sub-band up-sampling.Continuing to use the LP conversion to carry out iteration to low frequency sub-band decomposes, just original image can be decomposed into low frequency and high-frequency sub-band on a series of different scales.Subsequently, LP is decomposed the analysis of resulting high-frequency sub-band service orientation bank of filters DFB travel direction.The effect of this DFB is the directivity high-frequency information of catching image, and the singular point that is distributed on equidirectional is synthesized a coefficient.When calculating, adopt 1 layer tree construction to decompose, fan frequency cutting on the type direction every layer of elder generation by fan mode filter group QFB, subsequently with rotation re-sampling operations appropriate combination to realize image high-frequency information directional analysis, catch line, face singularity in the image.The high-frequency information that the net result of DFB can be regarded image as is divided into 21 wedge zones with frequency domain.

2, geometric moment

Geometric moment comprises moment of the orign and central moment, for the two-dimensional function f (x, y) the ∈ L (R that are defined on the o-xy plane ²), its (p+q) rank mixing moment of the orign is defined as:

$m_{\mathrm{pq}}=\underset{-\∞}{\overset{+\∞}{\∫}}\underset{-\∞}{\overset{+\∞}{\∫}}{x}^p{y}^{q}f(x,y)\mathrm{dxdy}$

P wherein, q=0,1,2 ....

For digital picture f (x, y), its (p+q) rank moment of the orign is defined as:

$m_{\mathrm{pq}}=\underset{i}{\mathrm{\Σ}}\underset{j}{\mathrm{\Σ}}{i}^p{j}^{q}f(i,j)$

The central moment of image is defined as:

${\mathrm{\μ}}_{\mathrm{pq}}=\underset{x}{\mathrm{\Σ}}\underset{y}{\mathrm{\Σ}}{(x-\stackrel{\‾}{x})}^p{(y-\stackrel{\‾}{y})}^{q}I(x,y)$

Wherein $\stackrel{\‾}{x}=\frac{m_{\mathrm{1,0}}}{m_{\mathrm{0,0}}},$ $\stackrel{\‾}{y}=\frac{m_{\mathrm{0,1}}}{m_{\mathrm{0,0}}}$

Utilize the affine unchangeability of original image geometric moment, before watermark detection, utilize the one or more geometric moments of original image estimate watermarking images the geometric transformation of process, according to the parameter of estimating watermarking images is carried out corresponding correction and watermark detection.This method can realize in arbitrarily-shaped domain, comprises spatial domain and various frequency domain, is to resist the fairly simple effective a kind of method of geometric attack at present.

Two, related symbol explanation

The original host image of I

I subband of each frequency level of the original host image of B (i)

x _iOriginal host image conversion coefficient

The W watermark signal

B _w(i) i subband of each frequency level of watermarking images

y _iThe watermarking images conversion coefficient

E _iThe energy of i subband of each frequency layer order of original host image

B _MaxThe maximum subband of energy in each frequency level of original host image

S selects the coefficient number of embed watermark

Z _iContain the watermark conversion coefficient

Each information of u watermark

The α embedment strength

The location matrix of R watermark conversion coefficient ordering

m _{P, q}P+q rank moment of the orign

μ _{P, q}P+q rank central moment

X geometric moment parameter key

I ' contains watermarking images

I " contains watermarking images after attacking

θ _cAngle parameter

A horizontal scaling parameter

B vertically scale parameter

P horizontal translation parameter

Q vertical translation parameter

I _rOriginal image is gone back in attack

B _r(i) attack i subband going back each frequency level of original image

B _RmaxThe maximum subband of energy in each frequency level of original image also

x _i' attack reduction image transformation coefficient

y _i' under fire back watermarking images conversion coefficient

Three, embed based on the digital watermarking of contourlet conversion resist geometric attacks

With reference to Fig. 1, digital watermark embed process of the present invention is as follows:

Step 1 is carried out contourlet to host image I and is decomposed.

Host image I is carried out 4 layers of contourlet decompose, obtain the subband B (i) of low frequency, medium and low frequency, medium-high frequency and high frequency level and the conversion coefficient x of each subband _iBecause the HFS resistance to compression is poor, and human eye is relatively more responsive to low frequency part, after therefore selecting earlier image to be carried out 4 layers of decomposition, watermark is embedded 2,3,4 layers process respectively again.Multilayer embeds and can well improve the robustness of watermark like this, even and still can preserve watermark energy after disperseing to make image one deck to be attacked the watermark energy, it is recovered.

Step 2 is carried out contourlet to watermarking images W and is decomposed.

Watermarking images W is carried out 3 layers of contourlet to be decomposed, obtain the subband Bw (i) of frequency level of low frequency, intermediate frequency and high frequency and the conversion coefficient yi of each subband, corresponding respectively being embedded in the coefficient of 2,3,4 layers of host images of coefficient with every layer of watermarking images W, make that containing watermarking images has better visual effect.

Step 3 is calculated host image I sub belt energy.

The conversion coefficient of host image I sued for peace can obtain the ENERGY E i of every layer of each subband of host image I, i.e. Ei=∑ (xi) ²Energy has been represented the complexity of this subband texture, can be used for judging whether this subband is fit to embed watermark.

Step 4 is pressed the big minispread subband of energy.

Because the maximum representative image of energy grain details on this direction is the abundantest, it is wherein disguised best just to have hinted that also watermark embeds, thereby can go out the maximum subband Bmax of every layer energy, and the target subband that maximum subband Bmax is embedded as watermark by descending sequence notation.

Step 5 is selected the maximum sub-band coefficients of energy.

Coefficient among the maximum subband Bmax of every layer of energy is arranged from big to small by absolute value, S coefficient before selecting, the value of this S is watermarking images W to coefficient number that should the straton band, the significant coefficient of contourlet conversion also has characteristics, it is exactly the small echo significant coefficient that random noise can produce similar true edge, therefore but can not produce the contourlet significant coefficient, be chosen on this coefficient embed watermark and can effectively resist attacked by noise.

Step 6 is handled watermark coefficient.

The coefficient of watermark is arranged from big to small according to absolute value equally, and to write down this matrix R that puts in order be first key.It is in watermark embedding additive process that watermark coefficient is handled, and makes big coefficient and big coefficient stack, and little coefficient and the stack of little coefficient are with effective enhancing watermark robustness;

Step 7, watermark embeds.

Utilize Superposition Formula each conversion coefficient y with every layer of watermark _iEmbed target subband B with watermark _MaxIn S conversion coefficient according to the corresponding stack of size, be about to watermark and be embedded in host image I, Superposition Formula is:

z _i＝x _i+αu (u＞0) (1)

z _i＝x _i-αu (u＜0) (2)

Wherein, the xi representative is former to be the host image conversion coefficient, and the zi representative contains the watermark conversion coefficient, and u represents each information of watermark.Embedment strength α is selected by test, and the visual effect of going back original image after the watermark stack does not have obvious noise, can accept this embedment strength, and α chooses [0 0.1 0.2 0.3] in the experiment.Because low frequency belongs to smooth region, embedment strength need be selected lower, and high frequency belongs to the texture complex region, and embedment strength can suitably increase.

Step 8, reconstruct contains watermarking images.

The coefficient of embed watermark is carried out the contourlet inverse transformation, and then restructural goes out to contain the image of watermark, and output contains the image I of watermark ';

Step 9, the record geometric moment.

Calculating contains the first moment about the origin of watermarking images I ': m _1,0, m _0,1, m _0,0, depositing them in the 1*3 matrix and be designated as the second key X, parameter estimation is used when staying to extraction.Concrete computing formula is:

$m_{\mathrm{1,0}}=\underset{i}{\mathrm{\Σ}}\underset{j}{\mathrm{\Σ}}i{\×I}^{\′}(i,j)---\left(3\right)$

$m_{0,1}=\underset{i}{\mathrm{\Σ}}\underset{j}{\mathrm{\Σ}}j{\×I}^{\′}(i,j)---\left(4\right)$

$m_{\mathrm{0,0}}=\underset{i}{\mathrm{\Σ}}\underset{j}{\mathrm{\Σ}}{I}^{\′}(i,j)---\left(5\right)$

Wherein, i, j representative contains the coordinate of watermarking images I ' pixel, and (i j) represents the pixel value of this location drawing picture to I '.

Four, extract based on the digital watermarking of contourlet conversion resist geometric attacks

With reference to Fig. 2, digital watermarking leaching process of the present invention is as follows

Step 1 is estimated attack parameter.

Utilize first moment about the origin formula (3), (4), (5) earlier to after attacking, containing watermarking images I " first moment about the origin m _1,0, m _0,1, m _0,0Calculate;

Utilize following second-order moment around mean formula to calculate second-order moment around mean μ again _1,1', μ _2,0', μ _0,2',

${\mathrm{\μ}}_{1,1}^{\′}=\underset{i}{\mathrm{\Σ}}\underset{j}{\mathrm{\Σ}}(i-\stackrel{\‾}{i})\×(j-\stackrel{\‾}{j})\×I^{\′\′}(i,j)---\left(6\right)$

${\mathrm{\μ}}_{2,0}^{\′}=\underset{i}{\mathrm{\Σ}}\underset{j}{\mathrm{\Σ}}{(i-\stackrel{\‾}{i})}^2\×I^{\′\′}(i,j)---\left(7\right)$

${\mathrm{\μ}}_{0,2}^{\′}=\underset{i}{\mathrm{\Σ}}\underset{j}{\mathrm{\Σ}}{(j-\stackrel{\‾}{j})}^2\×I^{\′\′}(i,j)---\left(8\right)$

Wherein, i, j represent image I under fire " coordinate of pixel, I " (i j) represents the pixel value of this location drawing picture, $\stackrel{\‾}{i}=\frac{m_{\mathrm{1,0}}^{\′}}{m_{\mathrm{0,0}}^{\′}},$

$\stackrel{\‾}{j}=\frac{m_{\mathrm{0,1}}^{\′}}{m_{\mathrm{0,0}}^{\′}};$

Utilize the first moment about the origin among the second key X: m at last _1,0, m _0,1, m _0,0With the parameter estimation formula, estimate the suffered geometric attack parameter of image, wherein:

Angle parameter is: ${\mathrm{\θ}}_c=\frac{1}{2}{\tan }^{-1}\frac{2{\mathrm{\μ}}_{\mathrm{1,1}}^{\′}}{{\mathrm{\μ}}_{\mathrm{2,0}}^{\′}-{\mathrm{\μ}}_{\mathrm{0,2}}^{\′}}---\left(9\right)$

The horizontal scaling parameter is: $a=\frac{m_{\mathrm{0,0}}{m}_{\mathrm{1,0}}^{\′}}{m_{\mathrm{0,0}}^{\′}{m}_{\mathrm{1,0}}},---\left(10\right)$

The vertically scale parameter is: $b=\frac{{m_{\mathrm{0,0}}^{\′}}^2{m}_{\mathrm{1,0}}}{m_{\mathrm{0,0}}^2{m}_{\mathrm{1,0}}^{\′}}---\left(11\right)$

The horizontal translation parameter is: $p=\frac{m_{\mathrm{1,0}}^{\′}-m_{\mathrm{1,0}}}{m_{0,0}}---\left(12\right)$

The vertical translation parameter is: $q=\frac{m_{\mathrm{0,1}}^{\′}-m_{\mathrm{0,1}}}{m_{0,0}}---\left(13\right)$

Step 2 is reduced image under fire.

The image under fire of reducing is exactly the reverse operating that image is under fire carried out following steps:

A) " the rotation angle-of-attack θ that is suffered that estimates image I under fire by estimation formulas (9) _c, image I under fire " is carried out reverse rotation θ _cAngle, the suffered rotation of promptly reducible image is attacked;

B) estimate horizontal scaling parameter a and the vertically scale parameter b that under fire image I suffered by estimation formulas (10) and (11), image I is under fire carried out the horizontal scaling that ratio is 1/a, ratio is the vertically scale of 1/b, and the suffered convergent-divergent of promptly reducible image is attacked;

C) estimate horizontal translation parameter p and the vertical translation parameter q that under fire image I suffered by estimation formulas (12) and (13), image I under fire is of a size of-horizontal translation of p, be of a size of-vertical translation of q, the suffered translation of promptly reducible image is attacked.

Step 3 will be gone back original image and be decomposed

To reduce image I _rCarry out 4 layers of contourlet and decompose, obtain the subband B of low frequency, medium and low frequency, medium-high frequency and high frequency level _r(i) and the conversion coefficient x of each subband _i'.

Step 4 is calculated sub belt energy

To attacking the reduction image I _rConversion coefficient x _i' sue for peace, can obtain attacking the reduction image I _rThe ENERGY E of every layer of each subband _i', i.e. E _i'=∑ (x _i'); Then by descending series arrangement subband B _r(i), i=1,2,3...n;

Step 5 is selected maximum subband and coefficient

To attack the reduction image I _rEvery layer of maximum subband B of energy after the conversion _RmaxIn coefficient arrange from big to small by absolute value, S coefficient before selecting, the S value is watermarking images W to coefficient number that should the straton band;

Step 6 is calculated the watermark conversion coefficient.

With every layer of maximum subband B of energy among each coefficient of selected S and the former host image I _MaxSubtract each other after the middle coefficient ordering, obtaining extracting under fire, watermark every layer of conversion coefficient in back is:

y _i’＝(I _r-I)/α (14)

It is identical when wherein α is with embedding;

Step 7, reduction watermark coefficient positions.

Coefficient positions information by writing down among the first key R is reduced to the original position with watermark coefficient, reconstructs every layer of conversion coefficient yi ' of watermark;

Step 8, the reduction watermark

To back watermark conversion coefficient y under fire _i' carry out contourlet inverse transformation, the watermark that obtains extracting.

Effect of the present invention can further specify by following experiment simulation.

1, simulated conditions

The lena.bmp image of selecting 256*256 for use, is chosen the bianry image of 64*64 and is tested, shown in Fig. 3 b shown in Fig. 3 a as host image.The experiment software environment is Matlab7.0.A series of attack tests have been designed, comprise Gauss's low-pass filtering, Wiener filtering, medium filtering, the spiced salt add make an uproar, Gauss add make an uproar, JPEG attacks, shear, the rotation of various yardsticks, convergent-divergent, translation transformation etc., under maximum attack strength situation, the watermark of extracting is carried out quality assessment by normalized correlation coefficient NC, Y-PSNR PSNR and square error RSM.

2, simulation result

Experimental result respectively as: Fig. 3 c, Fig. 3 d, Fig. 4, Fig. 5, Fig. 6, Fig. 7, Fig. 8 and Fig. 9.

Fig. 3 c is the composograph result behind the embed watermark, and its square error is 16.1703, and PSNR can reach 34.9178, has visual effect and disguise preferably.

Fig. 3 d is not for adding the watermark result of extracting under the attack condition, and visible watermark is excellent.

Fig. 4, Fig. 5, Fig. 6, Fig. 7, Fig. 8 and Fig. 9 have all embodied the ability of digital watermarking opposing various attack.

With reference to Fig. 4, wherein Fig. 4 a is the result when being subjected to the JPEG compression quality factor to be 10 attacks, and visible watermark still can clearly be discerned, and its NC value is 0.9372; Fig. 4 b makes an uproar for added by the spiced salt, extracts watermark result when mean square value is attack in 0.03 o'clock; Fig. 4 c is that added by Gauss to make an uproar mean square value be 0.1 result who extracts when attacking, and by Fig. 4 b and Fig. 4 c as seen, through the attacked by noise mistake, the NC value of watermark all remains on about 0.9, can know identification.This be since the contourlet conversion suppressing to have outstanding performance aspect the noise, for watermark also brings strong antinoise attacking ability.

With reference to Fig. 5, wherein Fig. 5 a, 5b, 5c are after dextrorotation 60 degree are attacked, original image, go back original image and extract watermark result, Fig. 5 d, 5e, 5f are that dextrorotation 45 degree are attacked the back original image, go back original image and extracted watermark result, and Fig. 5 g, 5h, 5i are that dextrorotation 30 degree are attacked the back original image, go back original image and extracted watermark result.As seen from Figure 5, owing to adopted the geometric moment parameter estimation techniques, image can be restored automatically after attacked by rotation, watermark extracting NC value can remain on more than 0.8700, has shown the ability that powerful opposing rotation is attacked.

With reference to Fig. 6, wherein Fig. 6 a is the original watermarking images that contains, and Fig. 6 b contains watermarking images after amplifying 3 times of attacks, and Fig. 6 c is the watermark result through extracting after the geometric attack reduction, as seen realize that through reduction back image the NC value of watermark extracting can reach 0.8798 synchronously from Fig. 6 c.

With reference to Fig. 7, wherein Fig. 7 a is the original watermarking images that contains, and Fig. 7 b contains watermarking images after dwindling 0.8 times of attack, and Fig. 7 c is the watermark result through extracting after the geometric attack reduction, realizes that through reduction back image the NC value of watermark extracting can reach 0.9916 synchronously.From Fig. 7 c as seen, this method is attacked convergent-divergent has good resistivity, and the watermark of extraction all can be known identification.

With reference to Fig. 8, wherein Fig. 8 a, Fig. 8 b, Fig. 8 c are that level is amplified 3 times of attack results, Fig. 8 c, Fig. 8 d, Fig. 8 e are the vertical 2 times of attack results that amplify, and the watermark NC value that extracts after handling through the attack reduction has shown the robustness of right title amplification attack all more than 0.88.

With reference to Fig. 9, wherein Fig. 9 a amplifies twice and rotates pi/4 attack back image, and Fig. 9 b is an also original image of process geometric attack, and Fig. 9 c is for extracting watermark result.See that from Fig. 9 c watermark NC value can reach 0.8815, prove that this method all can better be gone back original image, finishes watermark extracting, realizes protection and identification to the digital picture copyright to any geometric attack combination.

More than all to attack the back as shown in table 1 to watermark extracting result's experimental data:

Watermark extracting result data after table 1 various attack

Attack type NC PSNR RSM

Gaussian filtering 0.9983 28.3421 0.0015

Wiener filtering 0.9929 20.8088 0.0083

Medium filtering 0.9946 21.3524 0.0073

Shearing attack (150*150) 0.9180 11.2240 0.0754

Translation 1.0000 36.1236 0

JPEG attacks (10) 0.9372 10.6459 0.0862

The spiced salt adds makes an uproar (0.03) 0.9065 10.0490 0.0989

Gauss adds and makes an uproar (0.1) 0.8913 9.8499 0.1035

(pi/3) 0.8997 9.2216 0.1196 attacked in rotation

(pi/4) 0.9153 9.2662 0.1184 attacked in rotation

(pi/5) 0.9912 17.9282 0.0161 attacked in rotation

Enlarge 3 times 0.8798 8.4224 0.1438

Dwindle 0.8 times 0.9916 19.9958 0.0100

Asymmetric amplification 0.8815 8.5118 0.1409

Amplify twice rotation pi/4 0.9392 10.4769 0.0896

By table 1 as seen, have the anti-filtering characteristics that frequency domain algorithm itself has based on the watermarking algorithm in contourlet territory, watermark extracting NC value all remains on more than 0.9900 under traditional filtering is attacked.Because adopt the transform domain embedding grammar, embedded location is even, can effectively resist shearing attack, after deducting full figure 150*150 pixel, the NC value that extracts watermark can reach 0.9180.Translation is attacked can not produce any destruction to watermark extracting, and the NC value is 1.0000.Owing to adopted geometric moment reduction geometric attack technology, make this method have powerful resistibility ability in addition to various geometric attacks.The result proves that this is a kind of watermark embedding extraction method of comprehensive robust by experiment.

Claims (5)

1. one kind based on contourlet conversion resist geometric attacks digital image watermark embedding method, comprises following process:

(1) host image I is carried out contourlet and decompose, obtain the subband B (i) of low frequency, medium and low frequency, medium-high frequency and high frequency level and the conversion coefficient x of each subband _i

(2) watermarking images W is carried out contourlet and decompose, obtain the subband B of the frequency level of low frequency, intermediate frequency and high frequency _w(i) and the conversion coefficient y of each subband _i

(3) according to the conversion coefficient x of host image _i, the energy of every layer of each subband of calculating host image I: E _i=∑ (x _i) ², i=1,2,3 ... n;

(4) to the ENERGY E of every straton band _i, according to sequence notation from big to small, the subband of selecting the energy maximum is that watermark embeds target subband B _Max

(5) with the maximum subband B of every layer of energy _MaxIn conversion coefficient arrange from big to small by absolute value, and S coefficient before selecting, the value of this S is watermarking images W to coefficient number that should the straton band;

(6) with the conversion coefficient y of every layer of watermark _iArrange from big to small according to absolute value, every layer the matrix that puts in order is designated as the first key R of watermark extracting;

(7) with each conversion coefficient y of every layer of watermark _iEmbed target subband B with watermark _MaxIn S conversion coefficient according to the corresponding stack of size, be about to watermark and be embedded in the host image;

(8) the host image conversion coefficient with embed watermark carries out the contourlet inverse transformation, reconstruct the image I that contains watermark ';

(9) calculate the first moment about the origin that contains watermarking images I ': m _1,0, m _0,1, m _0,0, depositing them in the 1*3 matrix and be designated as the second key X, parameter estimation is used when staying to extraction.

2. watermark embedding method according to claim 1, wherein step (7) is described is embedded in watermark in the host image, is to carry out according to following formula:

z _i＝x _i+αu(u＞0)

z _i＝x _i-αu(u＜0)

Wherein, x _iRepresent original host image conversion coefficient, z _iRepresentative contains the watermark conversion coefficient, and u represents each information of watermark, and embedment strength α is selected by test.

3. watermark embedding method according to claim 1, wherein step (9) is carried out according to the following procedure:

$m_{\mathrm{1,0}}=\underset{i}{\mathrm{\Σ}}\underset{j}{\mathrm{\Σ}}i\×I^{\′}(i,j)$

$m_{\mathrm{0,1}}=\underset{i}{\mathrm{\Σ}}\underset{j}{\mathrm{\Σ}}j\×I^{\′}(i,j)$

$m_{0,0}=\underset{i}{\mathrm{\Σ}}\underset{j}{\mathrm{\Σ}}{I}^{\′}(i,j)$

Wherein, i, j representative contains the coordinate of watermarking images I ' pixel, and (i j) represents the pixel value of this location drawing picture to I '.

4. one kind based on contourlet conversion resist geometric attacks digital image watermark extraction method, comprises following process:

(1) calculate after attacking and contain watermarking images I " first moment about the origin m _1,0', m _0,1', m _0,0' and second-order moment around mean μ _1,1', μ _2,0', μ _0,2', utilize the first moment about the origin among the second key X: m _1,0, m _0,1, m _0,0With the parameter estimation formula, estimate the suffered geometric attack parameter of image, wherein:

Angle parameter theta _c ${\mathrm{\θ}}_c=\frac{1}{2}{\tan }^{-1}\frac{{2\mathrm{\μ}}_{\mathrm{1,1}}^{\′}}{{\mathrm{\μ}}_{2,0}^{\′}{-\mathrm{\μ}}_{\mathrm{0,2}}^{\′}}$

The horizontal scaling parameter $a=\frac{m_{\mathrm{0,0}}{m}_{\mathrm{1,0}}^{\′}}{m_{\mathrm{0,0}}^{\′}{m}_{\mathrm{1,0}}},$ The vertically scale parameter $b=\frac{{m_{\mathrm{0,0}}^{\′}}^2{m}_{\mathrm{1,0}}}{m_{\mathrm{0,0}}^2{m}_{\mathrm{1,0}}^{\′}}$

The horizontal translation parameter $p=\frac{m_{\mathrm{1,0}}^{\′}-m_{\mathrm{1,0}}}{m_{\mathrm{0,0}}},$ The vertical translation parameter $q=\frac{m_{0,1}^{\′}-m_{0,1}}{m_{\mathrm{0,0}}};$

(2) at these geometric attack parameters image I under fire " is carried out reverse operating, obtained attacking the reduction image I _r

(3) will attack the reduction image I _rCarry out contourlet and decompose, obtain the subband B of low frequency, medium and low frequency, medium-high frequency and high frequency level _r(i) and the conversion coefficient x of each subband _i';

(4) calculate attack reduction image I _rThe ENERGY E of every layer of each subband _i'=∑ (x _i') ², then by descending series arrangement subband B _r(i), i=1,2,3 ... n;

(5) with the maximum subband B of every layer of energy _RmaxIn coefficient arrange from big to small by absolute value, S coefficient before selecting is watermarking images W to coefficient number that should the straton band, the value of S is identical during with embedding;

(6) with every layer of maximum subband B of energy among each coefficient of selected S and the former host image I _MaxSubtract each other after the middle coefficient ordering, obtain extracting every layer of conversion coefficient of back watermark under fire

y _i’＝(I _r-I)/α

It is identical when wherein α is with embedding;

(7) by every layer of conversion coefficient y of watermark after the first key R reduction under fire _i' position;

(8) incite somebody to action back watermark conversion coefficient y under fire _i' carry out contourlet inverse transformation, the watermark that obtains extracting.

5. watermark extracting method according to claim 4, wherein the described calculating of step (1) contains watermarking images I after attacking " first moment about the origin m _1,0', m _0,1', m _0,0' and second-order moment around mean μ _1,1', μ _2,0', μ _0,2', be calculated as follows:

$m_{\mathrm{1,0}}^{\′}=\underset{i}{\mathrm{\Σ}}\underset{j}{\mathrm{\Σ}}i\×I^{\′\′}(i,j)$

$m_{\mathrm{0,1}}^{\′}=\underset{i}{\mathrm{\Σ}}\underset{j}{\mathrm{\Σ}}j\×I^{\′\′}(i,j)$

$m_{\mathrm{0,0}}^{\′}=\underset{i}{\mathrm{\Σ}}\underset{j}{\mathrm{\Σ}}{I}^{\′\′}(i,j)$

${\mathrm{\μ}}_{\mathrm{1,1}}^{\′}=\underset{i}{\mathrm{\Σ}}\underset{j}{\mathrm{\Σ}}(i-\stackrel{\‾}{i})\×(j-\stackrel{\‾}{j})\×I^{\′\′}(i,j)$

${\mathrm{\μ}}_{\mathrm{2,0}}^{\′}=\underset{i}{\mathrm{\Σ}}\underset{j}{\mathrm{\Σ}}{(i-\stackrel{\‾}{i})}^2\×I^{\′\′}(i,j)$

${\mathrm{\μ}}_{0,2}^{\′}=\underset{i}{\mathrm{\Σ}}\underset{j}{\mathrm{\Σ}}{(j-\stackrel{\‾}{j})}^2\×I^{\′\′}(i,j)$

Wherein, i, j represent image I under fire " coordinate of pixel, I " (i j) represents the pixel value of this location drawing picture, and:

$\stackrel{\‾}{i}=\frac{m_{\mathrm{1,0}}^{\′}}{m_{\mathrm{0,0}}^{\′}}$ $\stackrel{\‾}{j}=\frac{m_{\mathrm{0,1}}^{\′}}{m_{\mathrm{0,0}}^{\′}}.$

Images