CN100559396C - But the multiple bit digital watermark method of a kind of anti printing and scanning and geometric transformation - Google Patents

Abstract

But the present invention discloses the multiple bit digital watermark method of a kind of anti printing and scanning and geometric transformation.The present invention is embed watermark in the discrete Fourier range coefficient of image, which fourier modulus coefficient is each watermark bit embed in according to the discrete logarithm polar coordinates of fourier modulus coefficient determines, suffer geometric transformation such as convergent-divergent rotation when image or when print scanned, can make watermark detection extract step-out; Since the convergent-divergent of image and rotation correspond respectively to Fourier log-polar territory radially with angle direction on translation; during the Detection and Extraction watermark; relevantly between template come heavy synchronizing information watermark according to primary template and embedding; extract meaningful watermark information Bit String again; owing in watermark embedding and testing process, do not need image or its fourier modulus coefficient are carried out the image interpolation computing; thereby can not introduce the interpolation distortion and save the time; the assailant can't remove the template of embedding in addition; the present invention can be applicable to digital picture; the copyright protection of video; file and certificate false proof, and the video broadcasting monitoring etc.

Description

But the multiple bit digital watermark method of a kind of anti printing and scanning and geometric transformation

Technical field

The invention belongs to the image authentication technical field, but specifically be a kind of based on the anti printing and scanning of Fourier discrete logarithm polar coordinate transform and the multiple bit digital watermark method of geometric transformation.

Background technology

Digital watermark technology is a kind of Information Hiding Techniques; its basic thought is to embed secret information in digital products such as digital picture, Voice ﹠ Video; so that the copyright of protection digital product, prove product true and reliable property, follow the tracks of copy right piracy or the additional information of product be provided.Secret information wherein can be copyright marking, user's sequence number or product related information.These significant secret informations constitute the part of digital watermarking, i.e. information watermark.Extraction detection for the synchronizing information watermark also embeds one and is used for synchronous pseudo-random sequence, is called the template watermark.

Digital watermarking image in use suffers various attacks through regular meeting, for example: geometric transformation, JPEG compression, gaussian filtering, cutting etc.These attacks have hindered the detection and the extraction of watermark, even make watermark detection extract failure, and therefore the digital watermarking that need embed will have the robustness of these attacks of antagonism.

Some people has proposed many sane digital watermark methods and has protected digital watermarking.

Kaushal people such as (document [1]) has proposed a kind of blind Detecting digital watermark method of anti printing and scanning.They propose the low frequency coefficient embed watermark at fourier modulus.Because when detecting, need carry out geometric transformation earlier to scan image proofreaies and correct, pixel is carried out the gamma compensation, embed watermark makes image fault bigger at the low frequency place, the algorithm complexity, and the image print scanning device that their method places one's entire reliance upon concrete, versatility is not strong, as their method can not be applied to after the geometric transformation digital picture and to the digital picture occasion of scanning again of making a blueprint.

O ' Ruanaid people such as (document [2]) proposes embed watermark in how much fields of invariants of image, with the opposing geometric attack.They propose image through after the discrete Fourier transform (DFT), again image is carried out Fourier-mellin transform, and watermark just embed in the range coefficient after conversion.Need to carry out various inverse transformations after the embedding, and need image transformation is detected in how much fields of invariants when extracting, this method need be carried out the log-polar coordinate mapping interpolative operation to image coefficient and be caused very big image fault.

List of references is as follows:

[1]Kaushal Solaki，B.S.Manjunath，Ibranhim El-Khalil，”Print andScan’Resilient Data Hiding in Images”，Information Forensics andSecurity，IEEE Transactions on Volume 1，Issue 4，Dec.2006Page(s)：464-478

[2]J.O’Ruanaidh and T.Pun，Rotation，Scale，and Translation InvariantDigital Image Watermarking，Signal Processing，Vol.66，No.3，pp.303-317，1998.

Summary of the invention

But the present invention proposes the multiple bit digital watermark method of a kind of anti printing and scanning and geometric transformation, belong to image, video authentication technical field.The watermark that embeds is divided into template watermark (being designated hereinafter simply as template) and significant information watermark, and the present invention is embed watermark in the fourier modulus coefficient of image, and the watermark data that each coefficient embeds is determined according to the discrete logarithm polar coordinates of this coefficient.Suffer geometric transformation such as convergent-divergent rotation when image or when print scanned, can make watermark detection extract step-out.Since the convergent-divergent of image and rotation correspond respectively to the log-polar territory radially with angle direction on translation, during the Detection and Extraction watermark, relevantly between template come the watermark of weight synchronizing information according to primary template and embedding, extract meaningful watermark information Bit String again.

The technical scheme of the inventive method is as follows: this method comprises watermark embedding and watermark extracting two big processes, it is characterized in that as shown in Figure 1, the concrete steps of described watermark embed process are:

A. according to the multi-bit watermark of required embedding, calculate watermark matrix to be embedded; Watermark comprises the meaningful information sequence and the template sequence of many bits;

B. original image is done two dimension discrete fourier transform, and translation flip-flop embeds the watermark matrix among the step a in the fourier coefficient amplitude spectrum of gained then to the amplitude spectrum center;

C. the fourier coefficient behind the embed watermark is carried out inverse Fourier transform, obtain containing watermarking images, watermark embed process finishes;

Described watermark extracting is the inverse process that watermark embeds, and concrete steps are as follows:

D. testing image is carried out the two dimension discrete fourier transform of d ' * d ', and translation flip-flop is to the amplitude spectrum center;

E. average to having the polar fourier coefficient of identical discrete logarithm, obtain the fourier coefficient magnitude matrix of a two dimension;

F. calculate the magnitude matrix synchronous according to primary template and magnitude matrix with embed watermark;

G. with original pseudorandom modulation sequence the magnitude matrix that obtains among the step f is separated the spread spectrum modulation, obtain significant multi-bit watermark information.

As shown in Figure 2, the detailed way of described watermark embed process is:

1. definite watermark matrix that will embed; Watermark comprises the meaningful information sequence m{m (i) of L bit, i=0 ..., L-1, m (i) ∈ 0, and 1}} and template sequence T, generating length with key key is N _pAmbipolar pseudorandom modulation sequence p={p _jJ=0 ..., N _p-1} (p _j∈ 1, be N 1}) with length _TTemplate sequence T{T _nN=0...N _T-1, } (T _n∈ 1,1}), each bit of meaningful bit information is carried out band spectrum modulation; If meaningful information bit m (i) is " 1 ", be its band spectrum modulation that the same facies-suite of p obtains frequency expansion sequence W then _i{ w _Ijw _Ij∈ 1 ,+1}, 0≤j＜N _p1 * the p of }=+ if meaningful information bit m (i) is " 0 ", then is its band spectrum modulation the anti-phase sequence of p, i.e. W _i=-1 * p obtains binary message watermark data W{W to be embedded _i0≤i＜L}; Frequency expansion sequence Wi and template sequence T column major order line up the capable two-dimentional watermark matrix W T{wt of N row M (n, m); 0≤m＜M, 0≤n＜N} (as shown in Figure 3), the frequency expansion sequence that generates of each meaningful information bit modulation takies adjacent some row of matrix W T like this, and template sequence T also column major order leaves in adjacent several row of matrix W T and constitutes primary template T _m, (n m) is made of bipolarity bit " 1 " and " 1 " the element wt of matrix W T;

2. get d=max (dy, dx), dy and dx are respectively the length of host image and wide, and image is done the two dimension discrete fourier transform of d * d, flip-flop is moved on to the central authorities of fourier modulus spectrum, and the amplitude spectrum center are as the rectangular coordinate system initial point; Embed watermark in the poincare half plane of fourier coefficient amplitude spectrum.Embedding the regional normalized frequency value that is positioned at the fourier coefficient amplitude spectrum is f _nIntermediate frequency near; The rectangular coordinate of the fourier coefficient that embeds the zone (x, y) or polar coordinates (r, θ) by formula (1) be transformed into the discrete logarithm polar coordinates (l ρ, ta);

$\mathrm{l\ρ}=\mathrm{floor}\left({\log }_a\frac{r}{R}\right)+D_{\mathrm{offset}}$

ta＝floor(N×θ/pi)

(1)

R＝f _n×d

$a^{-M/2}\&le;\frac{r}{R}<a^{M/2}$

Wherein r is that this amplitude is put the distance corresponding to fourier modulus spectrum central point (flip-flop point), $r=\sqrt{x^2+{\mathrm{<figure-callout\; id="2"\; label="y"\; filenames="C20071002878000213.png"\; state="\{\{state\}\}">y</figure-callout>}}^2},$ θ is the angle of this fourier coefficient in polar coordinates, and θ=arctan (y/x), a are one and approach 1 constant greater than 1, as desirable a=2 ^1/M, a=2 ^2/MPerhaps a=2 ^4/MDeng; R is a log-polar

Initial point, generally get the normalized frequency value f of R correspondence _nNear the position of intermediate frequency, as get f _n=0.23, R=f like this _n* d=0.23 * d, D _OffsetBe a deviation constant that guarantees l ρ 〉=0, as desirable M/2; Floor () function representation base plate function in the formula;

3. the fourier coefficient after the adding watermark is carried out inverse Fourier transform, obtain adding the image of watermark, watermark embed process finishes.

Described formula (1) shows that also the fourier coefficient utmost point footpath r that embeds the zone satisfies a ^-M/2* R＜r≤a ^M/2* R, embedding the zone corresponding to watermark in the rectangular coordinate system is an annular, the discrete logarithm polar coordinates scope that is obtained by formula (1) is like this, 0≤l ρ＜M, 0≤ta＜N; By additivity embedding formula (2) or the property taken advantage of embedding formula (3) embed watermark:

c(y，x)＝c(y，x)+α×W(lρ，ta) (2)

c(y，x)＝c(y，x)×(1+α×W(lρ，ta)) (3)

Formula (2) and (3) show, the embedding of watermark does not need the DFT coefficient of image is carried out interpolative operation, so eliminated the distortion of image interpolation interpolation, at the lower half-plane of fourier modulus spectrum according to the symmetry symmetry embedding identical watermark of Fourier transform about amplitude spectrum center flip-flop point.

The detailed way of described watermark extraction process is:

1. get d '=max (dy ', dx '), dy ' and dx ' are respectively the length of testing image and wide, and testing image is carried out the two dimension discrete fourier transform of d ' * d ', and translation flip-flop is to the amplitude spectrum center, and the amplitude spectrum center is as the rectangular coordinate system initial point; Is the poincare half plane normalized frequency of fourier coefficient amplitude spectrum f _nNear the coordinate of intermediate frequency fourier coefficient from rectangular coordinates transformation to the discrete logarithm polar coordinates (l ρ, ta), as shown in Equation (4):

$\mathrm{l\&rho;}=\mathrm{floor}\left({\log }_a\frac{r}{R}\right)+D_{\mathrm{offset}}$

ta＝floor(N×θ/pi)

(4)

R′＝f _n×d′

$a^{-M^{\&prime;}/2}\&le;\frac{r}{R^{\&prime;}}<a^{M^{\&prime;}/2}$

D _OffsetBe a deviation constant that guarantees l ρ 〉=0, as desirable M '/2; Generally get M '＞M, as desirable M '=2M;

2. to have identical discrete logarithm polar coordinates (l ρ, fourier coefficient ta) (these coefficients are positioned at a sector under rectangular coordinate system) is averaged, as fourier coefficient magnitude matrix amp (lp, element ta) of two dimension; Can obtain a two dimension fourier coefficient magnitude matrix amp{amp (m, n) | 0≤m＜M ', 0≤n＜N};

3. because the rotational transform of image shows as the circulation translation of angle (ta direction) direction in the log-polar territory of the fourier modulus spectrum of image, the scale transformation of image shows as the translation on (the being the lp direction) direction of logarithm utmost point footpath; Therefore with primary template T _mCarry out the relevant or relevant rapid matching computation of phase place of amplitude with magnitude matrix amp according to correlation theorem, and determine the position of embed watermark WT in magnitude matrix amp, thereby obtain the magnitude matrix synchronous with embed watermark WT according to maximum related value $\hat{a}\mathrm{mp}\left\{\hat{a}\mathrm{mp}(m,n)\right|0\&le;m<M,0\&le;n<N\};$

With primary template T _mThe matrix g that benefit 0 forms and magnitude matrix Amp is onesize (m, n), the translation correlation between them is:

$=\underset{k=0}{\overset{M^{\&prime;}-1}{\mathrm{\&Sigma;}}}\underset{l=0}{\overset{N-1}{\mathrm{\&Sigma;}}}\mathrm{amp}(m,n)g((m+k)\mathrm{mod}{M}^{\&prime;},(n+l)\mathrm{mod}N)---\left(6\right)$

0≤k＜M′，0≤l＜N

Again according to the two-dimensional correlation theorem, the product of the relevant two-dimentional DFT frequency-domain function with them of two two-dimensional functions in spatial domain Fourier transform and inverse transformation each other is right, so we have:

r(k，l)＝amp(m，n)оg(m，n)＝IDFT(AMP(u，v)G ^*(u，v)) (7)

Wherein " * " represents complex conjugate, and AMP (u, v)=DFT (amp (m, n)), and G (u, v)=DFT (g (m, n)), be AMP (u, v), (u v) is respectively amp (m to G, n), g (m, n) two-dimension fourier transform coefficient, thus can adopt following formula (8) calculate fast correlation r (k, l):

r(k，l)＝IDFT[AMP(u，v)G ^＊(u，v)] (8)

Also can calculate translation phase place correlation r _φ(k, l):

$r_{\mathrm{\&phi;}}(k,l)=\mathrm{IFFT}\left[{\mathrm{AMP}}_{\mathrm{\&phi;}}(u,v)G^{*}(u,v)\right],{\mathrm{AMP}}_{\mathrm{\&phi;}}(u,v)=e^{j{\mathrm{\&phi;}}_{\mathrm{AMP}}(u,v)}---\left(9\right)$

φ wherein _AMP(u v) is AMP (u, a phase angle v);

Because the primary template matrix has correlativity with the magnitude matrix of embedded template, therefore by the correlation matrix r (k, l) or r _φ(k, maximal value l) can be determined the embed watermark position, obtains the magnitude matrix synchronous with embed watermark WT $\hat{a}\mathrm{mp}\left\{\hat{a}\mathrm{mp}(n,m)\right|0\&le;m<M,0\&le;n<N\};$

4. use original pseudorandom modulation sequence p right Matrix is separated the spread spectrum modulation, obtains significant many bit informations; By embedding W _iThe time the position and the order from

The middle taking-up and W _iCorresponding N _pIndividual fourier coefficient, and form one section sequence, every section and N _pThe original PN sign indicating number p of bits is correlated with, if correlation is greater than 0, then to embed information bit be " 1 " in judgement, is " 0 " otherwise judgement embeds information bit, and despreading just obtains the binary sequence m that recovers afterwards ^*

The beneficial effect of this method is: owing in watermark embedding and testing process, do not need image or its fourier coefficient are carried out interpolative operation, thereby can not introduce the interpolation distortion and save the time; The assailant can't remove the template of embedding in addition, and the present invention can be applicable to the copyright protection of digital picture, video, file and certificate false proof, and the video broadcasting monitoring etc.

Description of drawings

Fig. 1 is the process flow diagram of the inventive method watermark embed process.

Fig. 2 is the process flow diagram of the inventive method watermark extraction process.

Fig. 3 is two-dimentional watermark WT arranged figure.

Fig. 4 generates magnitude matrix amp synoptic diagram; Wherein (a) is the amplitude spectrum annular region corresponding with amp, (b) for being averaged respectively and obtaining two-dimentional magnitude matrix amp having the polar coefficient of identical discrete logarithm.

(a) is original image Lena among Fig. 5; (b) be the watermarking images that adds of Lena, PSNR is 42.1dB.

(a) is that Lena adds the image of watermarking images after printing scanning among Fig. 6; (b) be that Lena adds watermarking images through the image of cutting (long and wide gone 25% by sanction respectively).

Fig. 7 is that Lena adds watermarking images earlier through rotation 45 degree, amplifies back original image size and shape through cutting again.

Fig. 8 is the r that image detection shown in Figure 7 is obtained _φ(k, l) distribution plan.

Embodiment

Below in conjunction with specific embodiment the inventive method is further elaborated.

Shown in Fig. 5,6,7,8, in this example, utilize this method in the gray level image of Lena 512 * 512, to embed the meaningful information of 60 bits.Here can be with the direct embedded images of the meaningful information of 60 bits, but for attack such as antinoise pollution better, also can carry out Error Correction of Coding to this 60 bit information, encode as turbo, the sequence that obtains behind the coding replaces in front that said meaningful information sequence embeds in the summary of the invention, correspondingly when watermark detection, also need error correction decoding to recover 60 bit meaningful informations.

Fig. 8 is the r that image detection shown in Figure 7 is obtained _φ(k, l) distribution plan.According to r among the figure _φ(k, maximal value correspondence l) (k, l) translational coordination can synchronization watermarking detects and can determine the geometric transformation parameter that image suffers, as the anglec of rotation and zoom factor.Therefore, the template watermark of this method can also be used for other occasions that needs the geometric transformation parameter that detected image suffers.

As shown in Figure 1, the way of watermark embed process is:

(1) prepare 60 bit meaningful informations, as by character string " SunYatSenU " the 60 bit meaningful informations formed of 6 bit ASCIIs of each character be " 11,001,010,101,101,110,110,011 0100001001011110010101001011101101010 ".60 bit meaningful informations are carried out the turbo coding, and the code length of editorial afterword is 124 bits.

(2) the turbo sign indicating number to gained carries out band spectrum modulation.Earlier generating numerical value with key is ± 1 Template Information and pseudo-random sequence p, and the code length of p is 128.Get M=64, N=360, a=2 ^1/MWhen the turbo code element was " 1 ", this code element was become the positive sequence W of p by band spectrum modulation _i=+1 * p; When the turbo code element was " 0 ", this code element by band spectrum modulation was-p.128 bit sequences that band spectrum modulation obtains are lined up two identical row of length by row, the final matrix of forming one 64 * 248 of the sequence that all band spectrum modulation obtain.Template sequence T also lines up 64 * 112 primary template T by row _m, it forms 64 * 360 watermark matrix W T (as shown in Figure 3) with the information watermark matrix of front.(n m) is made of bipolarity bit " 1 " and " 1 " the element wt of matrix W T.

(3) embed watermark in the amplitude spectrum of image fourier modulus spectrum is got f _n=110/512=0.23.

(4) fourier coefficient that adds watermark is carried out inverse transformation, obtain watermarking images, watermarking images PSNR=42.1dB.

As shown in Figure 2, the way of watermark detection extraction is:

(1) testing image is carried out the DFT conversion.

(2) get M '=2M=128, the rectangular coordinate of fourier coefficient is carried out the discrete logarithm polar coordinate transform.Average to having the polar fourier coefficient of identical discrete logarithm, mean value is as element in the two-dimentional magnitude matrix, form one 128 * 360 magnitude matrix amp (n, m).Because the primary template of this example is placed on the 1st of watermark matrix W T and walks to the 64th row, the 1st to the 112nd row, filling 0 formation and amp (n, (n is in the time of m) for m) onesize matrix g, primary template information is placed on the 33rd and walks to the 96th row, and the 1st row are to the submatrix of 112 row.

(3) with pattern matrix g (n, m) with magnitude matrix amp (n m) carries out relevant matches according to correlation theorem and calculates fast, and according to the maximal correlation matching value at magnitude matrix amp (n, m) determine to be embedded with the zone of watermark in, obtain the magnitude matrix synchronous with embed watermark WT $\hat{a}\mathrm{mp}\left\{\hat{a}\mathrm{mp}(n,m)\right|0\&le;m<M,0\&le;n<N\}.$

(4) right with original pseudorandom modulation sequence p

Matrix is separated the spread spectrum modulation, from

The middle taking-up and W _iCorresponding 128 fourier coefficients of adjacent two row carry out relevantly with the original pseudo random sequence code p of 128bits, if correlation is greater than 0, then to embed information bit be " 1 " in judgement, otherwise judgement embedding information bit is " 0 ".124 binary sequences that just obtain recovering after the despreading.

(5) 124 binary sequences are carried out error-correcting decoding, obtain 60 bit meaningful informations, watermark extracting finishes.

It below is the detection of the 512 * 512Lena image (PSNR=42.1dB) that is embedded with the meaningful watermark of 60 bits being carried out various attack.Wherein the printer model of print scanned test is: CANON IR5000, and print resolution is 600dpi, and scanner models is hp scanjet 4570c, and scanning resolution is 82dpi.

Following form table 1～table 5 is testing results that the 512 * 512Lena image (PSNR=42.1dB) that is embedded with the meaningful watermark of 60 bits is carried out various attack.The printer model of print scanned test is in the table 1: CANON IR5000, and print resolution is 600dpi, and scanner models is HP scanjet 4570c, and scanning resolution is 82dpi.Table 2 is to adopt the result of general in the world Stirmark 3.1 testing tools to the water mark method test to table 5.Watermark error bit number accounted for the ratio of the total bit number of watermark when the bit error rate was meant the extraction watermark bit before the decoding; The decoding back bit error rate is meant that the watermark bit that extracts accounts for the ratio of the total bit number of watermark through error-correcting decoding recovery back watermark error bit number.The big little finger of toe image of image is through the size after print scanned in the table 1, and its size changes with comparing before printing as can be seen.The compression quality factor is meant the JPEG compression quality factor in the table 2.The zoom factor of table 3 is meant the ratio of image zooming.The cutting rate of table 4 is meant the length and the wide length ratio that is cropped respectively of image, crops image part all around during cutting, keeps the core of image.Table 5 (a) is the detection data after image is rotated cutting, and the rotation cutting of image is meant image is rotated a certain angle, and it is square keeping image then and be around core crops part and making images cut.Anti-rotation, cutting and the amplification of table 5 (b) gangs up against and is meant on the basis of rotation cutting, and image is amplified back original size.

The detection performance of table 1 anti printing and scanning

The image size	The bit error rate before the decoding	The decoding back bit error rate
			542×592	0.0242	0
553×601	0.0484	0
			550×597	0.0645	0
572×624	0.0403	0

The anti-JPEG compressed detected of table 2 data

The compression quality factor	The bit error rate before the decoding	The decoding back bit error rate
			30％	0.0726	0.05
35％	0.0565	0.0167
			40％	0.0645	0.0167
45％	0.0484	0
			50％	0.0484	0
55％	0.0161	0

The nonshrink detection data of putting of table 3

Zoom factor	The bit error rate before the decoding	The decoding back bit error rate
			0.6	0.2581	0.2167
0.7	0.1371	0.05
			0.8	0.0161	0
0.9	0.0081	0
			1	0	0
1.1	0	0
			1.2	0.0081	0
1.3	0.0323	0
			1.4	0.0081	0
1.5	0.0242	0
			1.6	0.0081	0
1.7	0	0

The anti-cutting of table 4 detects data

Shearing rate	The bit error rate before the decoding	The decoding back bit error rate
			1％	0.0161	0
2％	0.0081	0
			5％	0.0161	0
10％	0.0323	0
			15％	0.0242	0
20％	0.0161	0
			25％	0.0484	0
50％	0.1452	0.1
			75％	0.4919	0.4667

Table 5 (a)

The anglec of rotation	The bit error rate before the decoding	The decoding back bit error rate
			0	0	0
0.5	0.0242	0
			1	0.0161	0
1.5	0.0161	0
			2	0.0323	0
5	0.0161	0
			5.5	0.0242	0
10	0.0323	0
			10.5	0.0081	0
20	0.0161	0
			30	0.0081	0
40	0.0323	0
			45	0.0081	0
50	0.0242	0
			60	0.0323	0
70	0.0323	0
			80	0.0242	0
90	0	0
			90.5	0.0242	0
100	0.0242	0
			110	0.0081	0
120	0.0081	0
			130	0.0081	0
140	0.0403	0
			150	0.0323	0
160	0.0403	0
			170	0.0242	0
180	0	0
			180.5	0.0242	0

Table 5 (b)

The anglec of rotation	The bit error rate before the decoding	The decoding back bit error rate
			0	0	0
0.5	0.0081	0
			1	0.0323	0
1.5	0.0323	0
			2	0.0081	0
5	0.0081	0
			5.5	0.0242	0
10	0.0403	0
			10.5	0.0161	0
20	0.0161	0
			30	0.0645	0
40	0.0242	0
			45	0.0161	0
50	0.0242	0
			60	0.0403	0
70	0.0403	0
			80	0.0484	0
90	0	0
			90.5	0	0
100	0.0323	0
			110	0.0242	0
120	0.0726	0
			130	0.0242	0
140	0.0403	0
			150	0.0565	0
160	0.0403	0
			170	0.0403	0
180	0	0
			180.5	0	0

Claims (4)

1. But 1, the multiple bit digital watermark method of a kind of anti printing and scanning and geometric transformation, this method comprise watermark embedding and watermark extracting two big processes, it is characterized in that the concrete steps of described watermark embed process are:

   A. according to the multi-bit watermark of required embedding, calculate watermark matrix to be embedded; Watermark comprises the meaningful information sequence and the template sequence of many bits;

   B. original image is done two dimension discrete fourier transform, and translation flip-flop embeds the watermark matrix among the step a in the fourier coefficient amplitude spectrum of gained then to the amplitude spectrum center;

   C. the fourier coefficient behind the embed watermark is carried out inverse Fourier transform, obtain containing watermarking images, watermark embed process finishes;

   Described watermark extracting is the inverse process that watermark embeds, and concrete steps are as follows:

   D. testing image is carried out the two dimension discrete fourier transform of d ' * d ', and translation flip-flop is to the amplitude spectrum center;

   E. average to having the polar fourier coefficient of identical discrete logarithm, obtain the fourier coefficient magnitude matrix of a two dimension;

   F. primary template is carried out the relevant or relevant rapid matching computation of phase place of amplitude with magnitude matrix according to correlation theorem, and determine the position of embed watermark in magnitude matrix, thereby obtain the magnitude matrix synchronous with embed watermark according to maximum related value;

   G. with original pseudorandom modulation sequence the magnitude matrix that obtains among the step f is separated the spread spectrum modulation, obtain significant multi-bit watermark information.
2. 2, multiple bit digital watermark method according to claim 1 is characterized in that the detailed way of described watermark embed process is:

   1. definite watermark matrix that will embed; Watermark comprises the meaningful information sequence m={m (i) of L bit, i=0 ..., L-1, m (i) ∈ 0, and 1}} and template sequence T, generating length with key key is N _pAmbipolar pseudorandom modulation sequence p={p _jJ=0 ..., N _p-1}, p _j{ 1,1} and length are N to ∈ _TTemplate sequence T{T _nN=0...N _T-1, },

   

   Each bit to meaningful bit information carries out band spectrum modulation; If meaningful information bit m (i) is " 1 ", be its band spectrum modulation that the same facies-suite of p obtains frequency expansion sequence W then _i{ w _Ijw _Ij∈ 1 ,+1}, 0≤j＜N _p1 * the p of }=+ if meaningful information bit m (i) is " 0 ", then is its band spectrum modulation the anti-phase sequence of p, i.e. W _i=-1 * p obtains binary message watermark data W{W to be embedded _i0≤i≤L-1}; Frequency expansion sequence W _iWith template sequence T column major order line up the capable two-dimentional watermark matrix W T{wt of N row M (n, m); 0≤m＜M, 0≤n＜N}, the frequency expansion sequence that generates of each meaningful information bit modulation takies adjacent some row of matrix W T like this, and template sequence T also column major order leaves in adjacent several row of matrix W T and constitutes primary template T _m, (n m) is made of bipolarity bit " 1 " and " 1 " the element wt of matrix W T;

   2. get d=max (dy, dx), dy and dx are respectively the length of host image and wide, and image is done the two dimension discrete fourier transform of d * d, flip-flop is moved on to the central authorities of fourier modulus spectrum, and the amplitude spectrum center are as the rectangular coordinate system initial point; Embed watermark in the poincare half plane of fourier coefficient amplitude spectrum.Embedding the regional normalized frequency value that is positioned at the fourier coefficient amplitude spectrum is f _nIntermediate frequency near; The rectangular coordinate of the fourier coefficient that embeds the zone (x, y) or polar coordinates (r, θ) by formula (1) be transformed into the discrete logarithm polar coordinates (l ρ, ta);

   $\mathrm{l\&rho;}=\mathrm{floor}\left(\mathrm{lo}{g}_a\frac{r}{R}\right)+D_{\mathrm{offset}}$

   $\mathrm{ta}=\mathrm{floor}(N\&times;\mathrm{\&theta;}/\mathrm{pi})---\left(1\right)$

   $R=f_n\&times;d$

   $a^{-M/2}\&le;\frac{r}{R}<a^{M/2}$

   Wherein r is that this amplitude is put the distance corresponding to fourier modulus spectrum central point, $r=\sqrt{x^2+y^2},$ θ is the angle of this fourier coefficient in polar coordinates, and θ=arctan (y/x), a are one and approach 1 constant greater than 1; R is a log-polar

   

   Initial point, get the normalized frequency value f of R correspondence _nBe the position near intermediate frequency, R=f _n* d, D _OffsetIt is a deviation constant that guarantees l ρ 〉=0; Floor () function representation base plate function in the formula;

   3. the fourier coefficient after the adding watermark is carried out inverse Fourier transform, obtain adding the image of watermark, watermark embed process finishes.
3. 3, multiple bit digital watermark method according to claim 2 is characterized in that described formula (1) shows that also the fourier coefficient utmost point footpath r that embeds the zone satisfies a ^-M/2* R＜r≤a ^M/2* R, embedding the zone corresponding to watermark in the rectangular coordinate system is an annular, the discrete logarithm polar coordinates scope that is obtained by formula (1) is like this, 0≤l ρ＜M, 0≤ta＜N; By additivity embedding formula (2) or the property taken advantage of embedding formula (3) embed watermark:

   c(y，x)＝c(y，x)+α×W(lp，ta) (2)

   c(y，x)＝c(y，x)×(1+α×W(lρ，ta)) (3)

   Formula (2) and (3) show, the embedding of watermark does not need the DFT coefficient of image is carried out interpolative operation, so eliminated the distortion of image interpolation interpolation, at the lower half-plane of fourier modulus spectrum according to the symmetry symmetry embedding identical watermark of Fourier transform about amplitude spectrum center flip-flop point.
4. 4, multiple bit digital watermark method according to claim 1 is characterized in that the detailed way of described watermark extraction process is:

   1. get d '=max (dy ', dx '), dy ' and dx ' are respectively the length of testing image and wide, and testing image is carried out the two dimension discrete fourier transform of d ' * d ', and translation flip-flop is to the amplitude spectrum center, and the amplitude spectrum center is as the rectangular coordinate system initial point; Is the poincare half plane normalized frequency of fourier coefficient amplitude spectrum f _nNear the coordinate of intermediate frequency fourier coefficient from rectangular coordinates transformation to the discrete logarithm polar coordinates (l ρ, ta), as shown in Equation (4):

   $\mathrm{l\&rho;}=\mathrm{floor}\left(\mathrm{lo}{g}_a\frac{r}{R^{\&prime;}}\right)+D_{\mathrm{offset}}$

   $\mathrm{ta}=\mathrm{floor}(N\&times;\mathrm{\&theta;}/\mathrm{pi})---\left(4\right)$

   $R^{\&prime;}=f_n\&times;d^{\&prime;}$

   $a^{-M^{\&prime;}/2}\&le;\frac{r}{R^{\&prime;}}<a^{M^{\&prime;}/2}$

   D _OffsetBe a deviation constant that guarantees l ρ 〉=0, can be taken as M '/2; Get M '＞M;

   2. to have identical discrete logarithm polar coordinates (l ρ, fourier coefficient ta) is averaged, as fourier coefficient magnitude matrix amp (lp, element ta) of two dimension; Can obtain a two dimension fourier coefficient magnitude matrix amp{amp (m, n) | 0≤m＜M ', 0≤n＜N};

   3. because the rotational transform of image shows as the circulation translation of angle direction in the log-polar territory of the fourier modulus spectrum of image, the scale transformation of image shows as the translation on the direction of logarithm utmost point footpath; Therefore with primary template T _mCarry out the relevant or relevant rapid matching computation of phase place of amplitude with magnitude matrix amp according to correlation theorem, and determine the position of embed watermark WT in magnitude matrix amp, thereby obtain the magnitude matrix synchronous with embed watermark WT according to maximum related value $\hat{a}\mathrm{mp}\left\{\hat{a}\mathrm{mp}(m,n)\right|0\&le;m<M,0\&le;n<N\};$

   With primary template T _mThe matrix g that benefit 0 forms and magnitude matrix amp is onesize (m, n), the translation correlation between them is:

   

   $=\underset{k=0}{\overset{M^{\&prime;}-1}{\mathrm{\&Sigma;}}}\underset{l=0}{\overset{N-1}{\mathrm{\&Sigma;}}}\mathrm{amp}(m,n)g((m,k)\mathrm{mod}{M}^{\&prime;},(n+l)\mathrm{mod}N)---\left(6\right)$

   $0\&le;k<M^{\&prime;},0\&le;l<N$

   Again according to the two-dimensional correlation theorem, promptly Fourier transform and inverse transformation are right each other for the product of the relevant two-dimentional DFT frequency-domain function with them of two two-dimensional functions in the spatial domain, so we have:

   r(k，l)＝amp(m,n)оg(m,n)＝IDFT(AMP(u,v)G ^*(u,v))

   Wherein " * " represents complex conjugate, and AMP (u, v)=DFT (amp (m, n)), and G (u, v)=DFT (g (m, n)), be AMP (u, v), (u v) is respectively amp (m to G, n), g (m, n) two-dimension fourier transform coefficient, thus can adopt following formula (8) calculate fast correlation r (k, l):

   r(k，l)＝IDFT[AMP(u，v)G ^*(u，v)] (8)

   Also can calculate translation phase place correlation r _φ(k, l):

   $r_{\mathrm{\&phi;}}(k,l)=\mathrm{IFFT}\left[{\mathrm{AMP}}_{\mathrm{\&phi;}}(u,v)G^{*}(u,v)\right],{\mathrm{AMP}}_{\mathrm{\&phi;}}(u,v)=e^{j{\mathrm{\&phi;}}_{\mathrm{AMP}}(u,v)}---\left(9\right)$

   φ wherein _AMP(u v) is AMP (u, a phase angle v);

   Because the primary template matrix has correlativity with the magnitude matrix of embedded template, therefore by the correlation matrix r (k, l) or r _φ(k, maximal value l) can be determined the embed watermark position, obtains the magnitude matrix synchronous with embed watermark WT $\hat{a}\mathrm{mp}\left\{\hat{a}\mathrm{mp}(m,n)\right|0\&le;m<M,0\&le;n<N\};$

   4. use original pseudorandom modulation sequence p right

   

   Matrix is separated the spread spectrum modulation, obtains significant many bit informations; By embedding W _iThe time the position and the order from

   

   The middle taking-up and W _iCorresponding N _pIndividual fourier coefficient, and form one section sequence, every section and N _pThe original PN sign indicating number p of bits is correlated with, if correlation is greater than 0, then to embed information bit be " 1 " in judgement, is " 0 " otherwise judgement embeds information bit, and despreading just obtains the binary sequence m that recovers afterwards ^*

Images