CN100357971C - Wavelet-based geometric attack resistant digital watermark method - Google Patents

Abstract

The present invention discloses a digital watermarking technique based on wavelet transformation, which belongs to the processing field of multimedia signals. The digital watermarking technique comprises a watermark embedding part and a watermark extracting part, wherein the watermark embedding part comprises the following steps that the wavelet transformation for an original image is carried out, a watermark is embedded into a coefficient of an approximate subgraph, and the global DCT transformation for the approximate subgraph is carried out, so that an image characteristic vector which is robust to geometrical attack is found; the watermark extracting part comprises the following steps that the wavelet transformation for a measured image is carried out, and the global DCT transformation for the approximate subgraph is carried out, so that a characteristic vector of an image to be measured is found; thereby, the watermark of the image to be measured is extracted by the feature vector. The digital watermarking technology of the present invention is based on the wavelet transformation, and is used for extracting the watermark by the characteristic of the image. Tests show that the digital watermarking technique has the strong geometric attack performance for resisting rotation, reduction, amplification, translation and distortion.

Description

A kind of digital watermark method based on the small echo resist geometric attacks

Technical field

The present invention relates to a kind of digital watermark technology based on wavelet transform (DWT) and Image Visual Feature, is a kind of multimedia data protection method, belongs to field of multimedia signal processing.

Technical background

Digital watermarking at present is one of effective ways of protection Digital Media copyright.Present most of watermark can only be resisted conventional attacks such as compression, filtering, and is all powerless for common or even small geometric attack.In recent years, abroad some scholar's Against Geometrical Attacks watermarking algorithms have been done some theoretical researches, but effect is unsatisfactory.Pereira etc. have used the template insertion, but the embedding of template and extracting method are complicated, and the information capacity of embedding is little, and the template that embeds is under attack easily; O ' Ruanaidh etc. are based on the Fourier-Mellin conversion, and (Transtation) unchangeability realizes the resist geometric attacks ability of watermark for Rotate, Scalling, but this method makes the watermarking images quality become relatively poor to utilize this conversion to have RST; Propositions such as Pitas embed watermark information on DFT territory annular region or the circumference, but it can only provide less anti-rotation performance and embed less information.

Therefore to be a challenging difficult problem, particularly Voloshynovskiy etc. at present go out at the article middle finger that local distortion arbitrarily attacks to resist geometric attacks is more complicated a kind of in the geometric attack, almost do not see effective resistance method announcement at present.

And the algorithm of above most of resist geometric attacks is based on the DFT conversion, incompatible with the JPEG compression standard of image compression standard JPEG 2000 of a new generation and use now, the core of JPEG2000 is a wavelet transformation, and wavelet transformation has good space, frequency characteristic.Therefore research has big meaning based on the resist geometric attacks digital watermarking algorithm of wavelet transformation.

Summary of the invention

The objective of the invention is to propose a kind ofly can resist the digital watermarking embedding and the extracting method of rotation, convergent-divergent, translation and distortion geometric attack based on wavelet transformation, it has higher robustness, with the copyright of protection Digital Media.

To achieve these goals, the present invention is performed such:

By wavelet transformation and dct transform, carrying out the embedding of watermark and the visual feature vector of watermarking images extracts, it combines digital watermark technology and visual feature of image vector, cryptography, has realized anti-rotation, convergent-divergent, translation (RST) and the distortion geometric attack of digital watermarking.The method applied in the present invention comprises watermark embedding and watermark extracting two large divisions, and first is that watermark embeds, and comprising: (1) carries out wavelet transformation to host image F, obtains low frequency ll channel FA _L, (2) are at FA _LMiddle embed watermark W (j) forms FA _L', (3) are to containing the ll channel FA of watermark information _L' carry out the full figure dct transform; in DCT Low Medium Frequency coefficient; find a visual feature vector V (j) to geometry attack robust; (4) generate a two-valued function sequence Key (j) according to watermark W (j) and visual feature of image vector V (j) by XOR; Key (j) is applied for to the third party as key; to obtain the entitlement of original works, reach the purpose of copyright protection.Second portion is a watermark extracting, comprise: (5) carry out wavelet transformation to testing image Test_F (brief note is T_F), obtain its low frequency ll channel T_FAL, (6) ll channel T_FAL is carried out the full figure dct transform, the visual feature vector V ' that finds out testing image in DCT Low Medium Frequency coefficient (j), (7) utilize two-valued function sequence Key (j) and testing image visual feature vector V ' (j),, extract watermark W ' (j) by the XOR computing.

Differentiate the entitlement of testing image by comparing W (j) and W ' correlativity (j).

Now be elaborated as follows to method of the present invention:

First:, obtain two-valued function sequence Key (j) by the embedding operation of watermark

At first with one group of binary pseudo-random W that can represent copyright information, W={w (j) | w (j)=0,1; 1≤j≤L} is as digital watermarking, original image be designated as F={f (i, j) | f (i, j) ∈ R; 1≤i≤N1,1≤j≤N2) }, wherein, w (j) and f (i j) represents the grey scale pixel value of watermark sequence and original image respectively, establishes N1=N2=N, and the embedding of watermark is as follows:

1) original image F is carried out the decomposition of small echo L level and obtains ll channel coefficient FA _L

After original image F carried out L level wavelet decomposition, can obtain a plurality of details subgraph coefficient FD _j ^k, (k=1,2,3; J=1,2,3...L) with a ll channel coefficient FA _L, L is a wavelet decomposition progression, and decomposed class L is high more, and conventional attack abilities such as the anti-JPEG compression of wavelet coefficient, filtering are strong more, but the corresponding lengthening of wavelet decomposition and reconstitution time, the present invention gets L=1 at the trial.

2) at ll channel coefficient FA _LIn, embed digital watermark generates FA ' _L

Adopt method of superposition commonly used, watermark is embedded ll channel wavelet coefficient FA _LIn, concrete formula is: FA ' _L=FA _L+ α W, wherein α is a watermark embed strength, its span generally is 0＜α≤0.5, if α gets 0 value, that is a kind of embedded mode of zero watermark.

3) to being embedded with the ll channel FA ' of watermark _L, carry out the full figure dct transform, obtain the visual feature vector V (j) of anti-how much resistances of image

Earlier to ll channel FA ' _LCarry out the full figure dct transform, obtain DCT matrix of coefficients FD (i, j), (i j), carries out Zig-Zag scanning, obtains frequency DCT coefficient sequence Y (j) from low to high to DCT matrix of coefficients FD again; In the coefficient sequence of Low Medium Frequency, find out J the point good to geometry attack robust, promptly when watermarking images is carried out geometric attack, the symbol of these low frequency coefficients does not change, or do not change substantially, by the symbol (binary sequence) of these coefficients, form a visual feature of image vector V (j), V={v (j) | v (j)=0,1; 1≤j≤J}, the element of the value representation visual feature vector V (j) of J is how much, and the value size of J will decide according to the length of disposable embed watermark and the power that robustness is required, in this paper test, J value 32.

4) utilize XOR, generate two-valued function sequence Key (j)

Key(j)＝V(j)W(j)

In the following formula, V (j) is the visual feature of image vector; W (j) is a watermark that will embed, and by cryptography XOR commonly used, generates a two-valued function sequence Key (j).

Preserve Key (j), will use when this extracts watermark below.By Key (j) is applied for to the third party as key, can obtain the entitlement of original works, reach the purpose of protection copyright.Second portion: the extraction of watermark

5) obtain the ll channel T_FA of testing image by wavelet transformation _L(i, j)

Identical L level wavelet decomposition when carrying out image Test_F to be measured (brief note for T_F) with embedding is obtained ll channel T_FA _L(i, j);

6) to ll channel T_FA _L(i j) carries out dct transform, and the proper vector V ' that obtains testing image (j);

In DCT Low Medium Frequency coefficient, try to achieve proper vector V ' (j), the method for asking with and each element in the position of DCT coefficient the proper vector V (j) when embedding consistent;

7) the proper vector V ' according to logic sequence Key that generates in the watermark embed process (j) and testing image (j) obtains watermark W ' (j), W ' (j)=(j)  Key (j) of V ';

Whether differentiate according to the degree of correlation of W and W ' has watermark to embed again.Thereby determine the owner of testing image.This algorithm extracts watermark does not need original image, therefore can the better protection original image.

The present invention and existing digital watermark relatively have following advantage:

Because the present invention is based on the digital watermark technology of wavelet transformation, has compatible mutually advantage in JPEG2000 of new generation; When extracting characteristics of image, used main flow conversion DCT simultaneously, also compatible with the JPEG of current trend;

The not only anti-rotation of this invention, convergent-divergent and translation geometric attack ability also have stronger anti-twist geometric attack ability;

Compare with the watermarking algorithm of embedded template, the extraction of this watermark mainly is the proper vector according to image, so this method is not worried " embedded template is attacked " these problems, better robustness is arranged, and the embedding of watermark and extract simple.

Preferably resolve the invisibility of watermark and the contradiction of robustness; The power of embed watermark is mainly by the α decision, and the robustness of watermark is mainly decided by the characteristics of image vector of selecting.

Below we from the explanation of theoretical foundation and test figure:

1) wavelet transformation of image

The wavelet transformation of image

The wavelet transformation that S.Mallat proposed in 1988 is a new signal analysis theory that rose in recent years, its " time-frequently " analytical approach that is a kind of, and its basic thought is with wavelet function ψ _{A, b}(t) be substrate, signal f (t) is decomposed.

$W_{f(a,b)}={\∫}_{R}f\left(t\right){\stackrel{\‾}{\mathrm{\ψ}}}_{a,b}\left(t\right)\mathrm{dt}$

Wavelet function ψ wherein _{A, b}(t) be through translation, flexible and one group of function obtaining by same basis function ψ.

ψ _{A, b}(t)=| a| ^-1/2(a ≠ 0 ψ is called basic small echo to ψ for (t-b)/a) a, b ∈ R, and a is a contraction-expansion factor, and b is a shift factor.

Mallat algorithm decomposition formula is:

$c_{j+1,k}=\underset{n\∈z}{\mathrm{\Σ}}{c}_{j,n}{\stackrel{\‾}{h}}_{n-2k}k\∈z$

$d_{j+1,k}=\underset{n\∈z}{\mathrm{\Σ}}{c}_{j,n}{\stackrel{\‾}{g}}_{n-2k},k\∈z$

Reconstruction formula is:

$c_{j,k}=\underset{n\∈z}{\mathrm{\Σ}}{c}_{j+1,n}{h}_{k-2n}+\underset{n\∈z}{\mathrm{\Σ}}{d}_{j+1,n}{g}_{k-2n},k\∈z$

After the 2D signal image carried out grade wavelet decomposition, former figure is divided into high frequency details subgraph (level, vertical and diagonal) and low frequency ll channel, high frequency details subgraph has mainly comprised the marginal information of former figure, but be subjected to the influence of external disturbance easily, and the low frequency ll channel comprises the essential information (low frequency part) of image, be subjected to external action little, therefore watermark be added in the ll channel, visual characteristic is preferably arranged.

New compression of images JPEG2000 and video compression standard MPE-4 have adopted wavelet transformation.Therefore select for use small echo to carry out the making of watermark, have compatible preferably with new image and video compression standard.

2) cosine transform (DCT)

DCT is used for the standard that picture coding is present widely used JPEG compression and MPEG-1/2.Its fast operation, the precision height is celebrated with the ability of extracting characteristic component and the optimum balance between the arithmetic speed.

2-D discrete cosine direct transform (DCT) formula is as follows:

$F(u,v)=c\left(u\right)c\left(v\right)\underset{x=0}{\overset{M-1}{\mathrm{\Σ}}}\underset{y=0}{\overset{N-1}{\mathrm{\Σ}}}f(x,y)\cos \frac{\mathrm{\π}(2x+1)u}{2M}\cos \frac{\mathrm{\π}(2y+1)v}{2N}$

u＝0，1，Λ，M-1；v＝0，1，Λ，N-1；

In the formula $c\left(u\right)=\left\{\begin{array}{cc}\sqrt{1/M}& u=0\\ \sqrt{2/M}& u=\mathrm{1,2},\mathrm{\Λ},M-1\end{array}\right.$

$c\left(v\right)=\left\{\begin{array}{cc}\sqrt{1/N}& v=0\\ \sqrt{2/N}& v=\mathrm{1,2},\mathrm{\Λ},N-1\end{array}\right.$

2-D discrete cosine inverse transformation (IDCT) formula is as follows:

$f(x,y)=\underset{u=0}{\overset{M-1}{\mathrm{\Σ}}}\underset{v=0}{\overset{N-1}{\mathrm{\Σ}}}c\left(u\right)c\left(v\right)F(u,v)\cos \frac{\mathrm{\π}(2x+1)u}{2M}\cos \frac{\mathrm{\π}(2y+1)v}{2N}$

x＝0，1，Λ，M-1；y＝0，1，Λ，N-1

X wherein, y is the spatial domain sampled value; U, v are the frequency field sampled value, and usually digital picture is represented with the pixel square formation, i.e. M=N, and the Low Medium Frequency coefficient of image DCT transform has been shown the principal character of image.

3) a kind of choosing method of image principal character vector:

Major part based on the relatively poor main cause of wavelet transformation watermarking algorithm resist geometric attacks ability is merely at present: attack when watermarking images is subjected to geometric attack, bigger change can take place in wavelet coefficient.If but we can find the proper vector of image, and this proper vector has stronger anti-attack ability to the geometric transformation of image, can utilize this proper vector to carry out the extraction of watermark so.Show by great number tested data, an image, through wavelet transformation, we are embedded in watermarking images in its ll channel of representing low-frequency information, then this ll channel are carried out the full figure dct transform earlier, the DCT coefficient is carried out the Zig_zag ordering again, preceding ten coefficients before listing in the table 1 after the process ordering, C1, C2, C3...C10 observes these coefficients.Find a phenomenon, when an image is rotated, during geometric transformation such as convergent-divergent, translation and distortion, some generally can take place and change in the size of medium and low frequency coefficient value, but the symbol of coefficient remains unchanged substantially.In the table 1, be Fig. 1 (a) as the former figure that tests, (lena512).The image that is subjected to behind the various common geometric attacks is seen Fig. 1 (b)-(f).Be located in the C1-C10 coefficient, positive coefficient is used " 1 " expression, negative or zero coefficient is used " 0 " expression, the coefficient symbols sequence of C1-C10 coefficient correspondence sees Table 1 " coefficient symbols sequence " so.Observe this row and find, can be consistent for the series of symbols of this symbol sebolic addressing of geometric attacks different in the table 1 and former figure, their corresponding symbol series normalized correlation coefficient value is big (seeing " related coefficient " these row) all;

The low frequency part coefficient is influenced by geometric attack after the table 1 image process DWT-DCT conversion

	PSNR (dB)	C1	C2	C3	C4	C5	C6	C7	C8	C9	C10	The coefficient symbols sequence	Related coefficient
														Former figure Lena512		5.086	-0.661	0.311	-0.042	0.359	-0.093	0.503	-0.414	0.218	-0.121	10 1010 1010	1.0
Rotation 20 degree (up time)	10.94	4.493	-0.553	0.393	-0.818	0.248	-0.484	-0.068	-0.659	0.221	-0.218	10 1010 0010	0.8
														Zoom factor 0.5		2.553	-0.333	0.157	-0.022	0.181	-0.047	0.251	-0.203	0.112	-0.059	10 1010 1010	1.0
Horizontal left 20 pixels	12.18	4.988	-0.285	0.259	-0.072	0.339	-0.473	0.707	-0.357	0.169	-0.128	10 1010 1010	1.0
														Rotation distortion 50 degree	16.18	5.135	-0.579	0.331	-0.051	0.378	-0.081	0.362	-0.483	0.167	-0.146	10 1010 1010	1.0
Ripple distortion quantity 200	22.87	5.136	-0.641	0.289	-0.041	0.334	-0.079	0.481	-0.401	0.206	-0.119	10 1010 1010	1.0

^*Coefficient unit 1.0e+004

For whether the further series of symbols of the DWT-DCT coefficient of authentication image has represented the geometrical characteristic of image, we are again common different standard testing image (seeing Fig. 2 (a)-(f)), carry out identical DWT-DCT conversion and Zig-Zag ordering, obtain corresponding coefficient and see Table 2 C1-C10, see its coefficient symbols sequence see the 12nd row, can find different test patterns, the related coefficient of symbol sebolic addressing is very little, and concrete correlation is seen " related coefficient " row; This explanation watermarking images is through wavelet transformation and dct transform, and wherein the low frequency coefficient symbol sebolic addressing can reflect the main visual signature (for simplicity only having got preceding 10 coefficients here observes) of this image.

The different test patterns of table 2 are through DWT-DCT conversion Low Medium Frequency part coefficient

^*Coefficient unit: 1.0e+004

5) characteristics of image is to the screening of measuring element

According to human visual system (HVS), the Low Medium Frequency signal is bigger to people's visual impact, the principal character of representative's image.For good robustness is arranged, our embed watermark is in the ll channel of wavelet transformation, and selected visual feature of image vector is the symbol sebolic addressing of DCT Low Medium Frequency coefficient.But we find, not all Low Medium Frequency coefficient is all fine to geometry attack robust, when specifically choosing which Low Medium Frequency coefficient symbols as unique point, can be earlier by RST (or it is main to select one of them to be transformed to according to demand) geometric operation, find out the better coefficient of those robustnesss (its coefficient symbols is constant substantially), remember the position of these coefficients, the coefficient symbols that is positioned at these positions is formed the proper vector of a sequence as this image, as for the number (element number of proper vector) of selecting coefficient, to decide according to the length of disposable embed watermark and the power that robustness is required, in general, the contained element number of proper vector is few more, the robustness of watermark is good more, but the watermark information of disposable embedding is also few more, takes all factors into consideration, in this paper test, the element number J of characteristics of image vector, value 32.

In sum, we utilize the symbol sebolic addressing of some Low Medium Frequency coefficient to obtain a kind of method that obtains the Image Visual Feature vector by host image being carried out the analysis of the coefficient behind wavelet transformation and the dct transform; Invented the digital watermark method that anti-RST and distortion are attacked.

Description of drawings

Fig. 1 (a) is original image (Lena512).

Fig. 1 (b) is through the image of rotation (20 degree turn clockwise).

Fig. 1 (c) is the image (zoom factor is 0.5) through convergent-divergent.

Fig. 1 (d) is the image (horizontal left 20pixel) through moving horizontally.

Fig. 1 (e) is through the image of rotation distortion (rotation distortion 50 degree).

Fig. 1 (f) is through the image of ripple distortion (ripple distortion quantity 200).

Fig. 2 (a) standardized test chart Lena512.

Fig. 2 (b) standardized test chart Baboo512.

Fig. 2 (c) standardized test chart Peppers512.

Fig. 2 (d) standardized test chart Crowd512.

Fig. 2 (e) standardized test chart Harbour512.

Fig. 2 (f) standardized test chart Woman512.

The watermarking images of Fig. 3 (a) when not disturbing.

The watermark detection of Fig. 3 (b) when not disturbing.

Watermarking images behind Fig. 4 (a) rotation 35 degree.

Watermark detection behind Fig. 4 (b) rotation 35 degree.

Fig. 5 (a) moves horizontally the image behind the 40pixel.

Watermark detection after Fig. 5 (b) moves horizontally.

Fig. 6 (a) original image.

Fig. 6 (b) zoom factor is 0.5 watermarking images.

Fig. 6 (c) zoom factor is 0.5 watermark detection.

Fig. 7 (a) rotation distortion is the watermarking images of 80 degree.

Fig. 7 (b) rotation distortion is the watermark detection of 80 degree.

Fig. 8 (a) ripple distortion quantity is 400 watermarking images.

Fig. 8 (b) ripple distortion quantity is 400 watermark detection.

Embodiment

The invention will be further described below in conjunction with accompanying drawing:

Use 1000 groups of independently binary pseudo-random (value is+1 or-1), every group of sequence length is 32bit, and in these 1000 groups of data, we appoint one group (we select the 500th group) of extraction here, as the watermark sequence that embeds.Original image is seen Fig. 3 (a), former figure is expressed as F (i, j), 1≤i≤512 wherein, 1≤j≤512, earlier through one deck wavelet decomposition, embed watermark in its low frequency ll channel carries out the full figure dct transform to the ll channel that is embedded with watermark more then, and its corresponding DCT matrix of coefficients is FD (i, j), wherein 1≤i≤256,1≤j≤256 are Y (j) to the DCT ordinal number that obtains sorting after its process Zig_Zag scanning, the DC component of first value Y (1) representative image, frequency order from low to high arranges then.Consider the capacity of robustness and disposable embed watermark, we select 32 unique points stronger to geometry attack robust in preceding 64 low frequency coefficients.After detecting W ', we have judged whether that by calculating normalized correlation coefficient NC (Normalized Cross Correlation) watermark embeds.

NC is as the output of watermark detector, and this value can reflect obviously whether watermark exists.

Fig. 3 (a) is the watermarking images that does not add when disturbing;

Fig. 3 (b) does not add when disturbing, and the output of watermark detector can be seen NC=1.0, obviously detects the existence of watermark.

Below we judge this digital watermark method resist geometric attacks ability by concrete test: (1) rotational transform

Fig. 4 (a) is 35 ° of watermarking images rotations, the PSNR=10.38dB of watermarking images at this moment, and signal to noise ratio (S/N ratio) is very low;

Fig. 4 (b) is the watermarking images of detection, can obviously detect the NC=0.63 that exists of watermark.

Table 3 is the anti-rotation of watermark challenge trial data.Can see in the table that when watermarking images rotates 40 ° NC=0.57 still can detect watermark and exist; The resist geometric attacks algorithm that people such as Pitas propose embeds watermark in the garden ring of DFT amplitude spectrum, can only resist the rotation that is not more than 3 degree.

The anti-rotation of table 3 watermark challenge trial data

The rotation number of degrees	0°	5°	10°	15°	20°	25°	30°	35°	40°
										PSNR(dB)		14.06	11.94	11.18	10.94	10.76	10.51	10.38	10.39
NC	1.00	1.00	0.93	0.87	0.75	0.68	0.81	0.63	0.57

(2) translation transformation

Fig. 5 (a) is the move to left situation of 40pix of image level, PSNR=10.58dB at this moment, and signal to noise ratio (S/N ratio) is very low;

Fig. 5 (b) is watermark detector output, can obviously detect the NC=0.76 that exists of watermark.

Table 4 is the anti-translation challenge trial of watermark data.From table, learn, still can detect the existence of watermark, so this digital watermarking has stronger anti-translation capability as horizontal left 55pixel.

The anti-translation challenge trial of table 4 watermarking images data

Displacement (pixel)	20	25	30	35	40	45	50	55	60
										PSNR(dB)	12.18	11.61	11.15	10.82	10.58	10.38	10.21	10.08	9.95
NC	0.94	0.94	0.82	0.82	0.76	0.76	0.64	0.69	0.51

(3) scale transformation

Fig. 6 (a) is an original image;

Fig. 6 (b) is 0.5 watermarking images for zoom factor;

Fig. 6 (c) is a watermarking detecting results, can obviously detect the NC=0.94 that exists of watermark.

Table 5 is watermark convergent-divergent challenge trial data, as can be seen from Table 5 when the watermarking images zoom factor little to 0.2 the time, facies relationship numerical value is still very high, NC=0.94 can obviously record the existence of watermark.The method of inserting template in DFT of employings such as Pereira can only be resisted zoom factor and be not less than 0.65 convergent-divergent, illustrates that this invention has stronger nonshrink exoergic power.

Table 5 watermark convergent-divergent challenge trial data

Zoom factor	0.2	0.5	0.8	1.0	1.2	1.5	2.0
								NC	0.94	0.94	0.94	1.0	1.0	0,94	0.94

After image carried out wavelet transformation, dct transform, observe its Low Medium Frequency coefficient, find that the coefficient of its conversion is done the amplification of corresponding proportion and dwindled when watermarking images carries out convergent-divergent, but coefficient symbols is constant, so this invention has the very strong nonshrink geometric attack ability of putting.

(4) rotation torsion test

Fig. 7 (a) is the watermarking images of rotation distortion 80 degree, and at this moment image obviously rotates distortion, PSNR=14.71dB.

Fig. 7 (b) is a watermarking detecting results, and NC=0.75 can obviously detect the existence of watermark.

Table 6 is the anti-rotation of watermark torsion test data, can see that from table this invention has stronger anti-rotation distortion ability.

The anti-rotation distortion of table 6 watermarking images

The rotation distortion number of degrees	10°	20°	30°	40°	50°	60°	70°	80°	90°	100°
											PSNR(dB)	20.89	18.45	17.37	16.72	16.185	15.74	15.37	14.71	14.44	14.19
NC	0.94	0.94	0.87	0.874	0.87	0.82	0.82	0.75	0.69	0.56

^*The realization of distortion and parameter adjustment can be used the distortion function in the photoshop filter.

(5) ripple torsion test

Fig. 8 (a) is for carrying out ripple warp image (ripple distortion quantity is 400) to watermarking images;

Fig. 8 (b) is its watermark detection situation, can obviously detect the existence of watermark, NC=0.94.

Table 7 is the anti-ripple torsion test of watermark data, and test figure can be learnt from table, and when ripple distortion quantity was 700, related coefficient was still higher, NC=0.94.Illustrate that distortion also has very strong robustness to this algorithm to ripple.

The anti-ripple torsion test of table 7 watermarking images data

Distortion quantity	50	100	200	300	400	500	600	700
									PSNR(dB)	30.94	26.69	22.87	20.95	19.69	18.79	18.06	17.49
NC	1.0	1.0	1.0	1.0	0.94	0.94	0.94	0.94

^*Specific implementation and parameter adjustment can be by the distortion functional modules in the photoshop filter

Claims (1)

1. But 1, a kind of digital watermark method based on the small echo resist geometric attacks, it is characterized in that: based on wavelet transform and Image Visual Feature vector, and digital watermark and cryptography combined, anti-rotation, convergent-divergent, translation and the distortion geometric attack of digital watermarking have been realized, this digital watermark method amounts to seven steps altogether in two sub-sections:

   First is that watermark embeds: by image being carried out the embedding operation and the feature extraction of watermark, obtain a corresponding two-valued function sequence Key (j);

   1) former figure F is carried out L level wavelet decomposition, obtain low frequency ll channel FA _L

   2) with method of superposition watermark is embedded into ll channel wavelet coefficient FA _LIn, formula is: FA ' _L=FA _L+ α W; Wherein α is a watermark embed strength;

   3) by to being embedded with the ll channel FA ' of watermark _L, carry out overall dct transform, find a watermarking images visual feature vector V (j) to geometry attack robust, this proper vector generates V={v (j) by the symbol sebolic addressing of DCT Low Medium Frequency coefficient | v (j)=0,1; 1≤j≤J}, J represent photopic vision proper vector V (j) element what, the value size of J will decide according to disposable embed watermark length with to the power that robustness requires; The more little watermark robustness of J is good more, but the information of disposable embed watermark is few more;

   4) utilize in the cryptography XOR commonly used, by the visual feature of image vector V (j) and watermark W (j) the generation two-valued function sequence Key (j) that will embed;

   Key(j)＝V(j)W(j)；

   Preserve Key (j), will use when this extracts watermark below,, can obtain the entitlement of original works, reach the purpose of protection copyright by Key (j) is applied for to the third party as key;

   Second portion is a watermark extracting: according to the proper vector V ' of two-valued function sequence Key (j) and testing image (j), extract watermark W ' (j);

   5) to testing image Test_F, note by abridging and be T_F, carry out L level wavelet decomposition, obtain the ll channel T_FA of testing image _L

   6) to ll channel T_FA _LCarry out dct transform, the proper vector V ' that obtains testing image (j); Wherein the position chosen of the length of testing image proper vector and vector element is identical with the proper vector V (j) that the watermark embedded part extracts;

   7) according to the vectorial V ' of the characteristics of image of the logic sequence Key (j) that in watermark embed process, forms and testing image (j), utilize the character of XOR obtain in the testing image watermark W ' (j), W ' (j)=(j)  Key (j) of V ';

   Whether have watermark embed, thereby determine the entitlement of testing image if differentiating according to the degree of correlation of W and W ' again.

Images