CN101383041A - Normalized quantisation watermarking method - Google Patents

Abstract

The invention relates to a normalized quantization-based watermarking method which pertains to the digital watermarking technical field. Although the watermarking capacity of a traditional quantization-based watermarking method is quite high, the defect of the traditional method lies in quite weak against the attacks of the scaling of an amplitude value and the modification of a constant value. The invention firstly carries out certain transformation to a carrier signal to be embedded with a watermark to form a signal which has invariability to the operation of scaling of the amplitude value and modification of the constant value, and then the signal is processed by adopting a quantization-based watermarking embedded device, and the signal obtained is converted into a carrier signal with a watermark. When checking watermarking information, the signal conformation method used when embedding the watermark is firstly used for operating the carrier signal of the watermark waiting to be checked, and then a quantization-based watermark detector is adopted to extract watermarking information from the signal obtained after conversion. By adopting the method, the robustness of a quantization-based watermark against the attacks of the scaling of the amplitude value and the modification of the constant value is greatly increased.

Description

A kind of normalized quantization-based watermarking

Technical field

The invention belongs to the digital watermark technology field, be specifically related to a kind of normalized quantization-based watermarking.

Background technology

Past 10 years, the use of digital media information and distribution explosive increase.By the internet, people can issue and obtain various numerical informations easily, comprise image, audio frequency, video and text etc., and many online services.But simultaneously, piracy also becomes and is more prone to, and unrestricted duplicating makes digital content have no copyright with uncontrollable transmission can to say.Management and protection to digital content become the problem that industry presses for solution.

Digital watermarking is a kind of emerging copyright protection technology.How its research hide a certain amount of additional information in raw data, the sign of the entitlement of works, the right to use or company etc. for example, and the purpose by the extraction that is hidden information and identification are reached the checking copyright.Digital watermarking has features such as fidelity, robustness, information capacity, reliability, security, blind Detecting and complexity.Except copyright protection, this technology also can be used for digital signature, digital finger-print, broadcasting supervision, content authentication, copy control and secret communication etc.

At present, academia has proposed a large amount of water mark methods.Basically, they can be divided into two classes: a class is based on the water mark method of spread spectrum (SS:spread spectrum) modulation technique; Another kind of is quantisation watermarking method, for example quantization index modulation watermark (QIM:quantization index modulation).Though the SS water mark method has good anti-interference and compression performance, it can not remove the interference of carrier signal, and it is blind to detect right and wrong.In contrast, quantisation watermarking method can be rejected the interference of carrier signal, has blind Detecting, and can obtain higher information capacity, thereby has received concern more and more widely.

QIM is the quantisation watermarking of a quasi-representative, and Chen etc. have proposed QIM water mark method (seeing patent CN02115174.1).If A _oThe indicate original contents of embed watermark, it can be various types of signals, for example image, audio frequency or video etc.Through a signal extraction process, from original contents A _oIn carrier signal x that to extract a length be N,

, be used for embed watermark information m, wherein

The expression set of real numbers.X can be one by original contents A _oThe vector formed of pixel value, perhaps by original contents A _oThe vector formed of any coefficient in transform domain etc.The basic thought of QIM watermark is to define plurality of sub-regions in the space of carrier signal x, and the corresponding different watermark informations of each subregion are modified in the corresponding subregion according to the value of the watermark information that will embed with x then; Watermark detection is the extracted region watermark information according to given carrier signal place.

Jitter quantisation and watermark modulation (DM:dither modulation) is basic (the B.Chen et al..Quantization index modulation:a class of provably good methodsfordigital watermarking and information embedding.IEEE Trans.Inform.Theory of realization of of QIM water mark method, vol.47, no.4, pp.1423-1443,2001.).Its basic thought is to utilize jitter quantisation tolerance carrier signal x to realize the embedding of watermark information.Particularly, the information m that DM will embed is encoded into one and longly is the G metasequence b of p, its each element b _jSatisfy b _j∈ Ω, Ω=0,1 ..., G-1}; And for embedding the multidigit watermark information, carrier signal x is divided into p subvector, i.e. x=x ₁‖ x ₂‖ ... ‖ x _p, j sub-vector x wherein _jLength be L _j,

For the DM of G unit, need to set up G

The jitter quantisation device , b _j∈ Ω.If use the uniform quantizer of quantization step, for the input vector of L dimension, jitter quantisation device as Δ Central point may be defined as

, wherein,

The expression set of integers, I ^LBe that each element all is 1 L dimensional vector,

It is the security that a random vector is used to promote watermark.The Information Selection corresponding quantitative device that basis will embed when watermark embedded is for b _j, j=1 ..., p selects quantizer

Revise x _j, promptly $y_j=Q_{b_j}\left(x_j\right)$ , y wherein _jExpression contains j the subvector of the carrier signal y of watermark.Use y to replace x, utilize the content A after the contrary leaching process of carrier signal obtains adding watermark then _wData content A for a watermark to be detected _u, the DM watermark detector use with watermark embed identical carrier signal leaching process obtain one may distortion signal z to be detected, and equally during with embedding be divided into p subvector, re-use minimum distance criterion and from each subvector, extract and hide Info, promptly ${\hat{b}}_j=\arg \underset{b_j\∈\mathrm{\Ω}}{\min }\left|\right|z_j-Q_{b_j}\left(z_j\right)\left|\right|,$ J=1 ..., p, wherein ‖ ‖ represents Euler's norm.

As previously mentioned, DM has many superior performances, but a main weakness of quantisation watermarking is that the amplitude convergent-divergent is attacked very sensitivity.This is because when the amplitude of signal to be detected was scaled, the quantization step that is used to detect was not done corresponding convergent-divergent, thereby has produced matching error between water mark embedding device and detecting device, greatly reduces the performance of watermark.

J.J. dust Gus waits the method for estimation (seeing patent CN 02824865.1) that has proposed a kind of quantization step.This method is used to extract watermark information according to the quantization step that the histogram of signal sampling estimates after attacking through convergent-divergent.But, accomplish to estimate that accurately the data sample amount that needs is very big, and may introduce safe fragility.

A.A.C.M. card restrains the method for estimation (seeing patent CN02824950.X) of having invented another kind of quantization step.This method has been introduced a characteristic parameter, has and the carrier signal feature of the identical multiple of convergent-divergent simultaneously, utilizes the quantization step after characteristic parameter is estimated to attack through convergent-divergent then.Yet, even this method is also non-vanishing in the error rate that does not have to detect under the condition of attacking.

A kind of water mark method (F.P é rez-Gonz à lez et al..Rational dither modulation:A high-rate data-hiding methodinvariant to gain attacks.IEEE Trans.Signal Processing that is called Rational Dithered Modulation (RDM), vol.53, no.10, pp.3960-3975 Oct..2005.) has adopted a quantization step that the amplitude convergent-divergent is had unchangeability at embedding device and detector end.Though RDM has obtained the robustness to the convergent-divergent attack, its performance depends on the memory span of system, and approaches the performance of DM along with the increase of memory span, and at this moment, the system-computed amount has also increased.

Q. Lee waits resistivity (the Q.Li et al..Using perceptual models to improve fidelityand provideresistance to valumetric scaling for quantization index modulation watermarking.IEEE Transactions on Information Forensics and Security that the Watson vision mode that has proposed application enhancements provides QIM that the amplitude convergent-divergent is attacked, vol.2, no.2, pp.127-139, Jun.2007.).The performance of this method depends on human sensor model, thereby underaction, and the quality of carrier signal has been lost in the modification of sensor model than situation about not revising.

To sum up, in known quantisation watermarking technology, amplitude convergent-divergent attack problem is not well solved, and all there is corresponding deficiency in the several method that has had.Revise the attack problem for constant, more do not see in the present document to relate to.

Summary of the invention

At the defective that exists in the prior art, the purpose of this invention is to provide a kind of normalized quantization-based watermarking, this method can be resisted the attack of amplitude convergent-divergent simultaneously and constant is revised attack, the two is all had extremely strong robustness, can obtain in other respects and the approximately uniform performance of original quantisation watermarking simultaneously.

For reaching above purpose, the technical solution used in the present invention is: a kind of normalized quantization-based watermarking, when embed watermark, the carrier signal x that at first treats embed watermark carries out a kind of signal that amplitude convergent-divergent and constant retouching operation are had unchangeability of certain transition structure

, adopt quantisation watermarking to embed device then and handle this signal, again with the signal that obtains

Be converted into the carrier signal y that contains watermark; When detecting watermark information, the carrier signal z that the signal configuration method of using when utilizing watermark to embed is earlier operated watermark to be detected obtains signal

, adopt the signal of quantisation watermarking detecting device then from gained The middle watermark information that extracts.It is characterized in that watermark embeds and detects and all comprises a signal configuration step that amplitude convergent-divergent and constant modification attack is had unchangeability.

Further, a kind of signal that amplitude scale transformation and constant retouching operation are had unchangeability of structure

The method that adopts is first conversion carrier signal x, making its arithmetic mean is a fixing constant, the span of this constant is unrestricted, and then with element corresponding be divided by of the signal after the conversion with a characteristic sequence s, characteristic sequence s satisfies following character: when the amplitude of carrier signal is scaled, and the scaled together thereupon identical multiple of s.

Further, by conversion

The signal that behind the watermark embedding operation, obtains

Acquisition contains the carrier signal y of watermark, and its method is with subvector s _jWith

Respective components multiply each other, and then multiply by factor lambda _j, a constant γ at last again superposes each component of gained vector _j, just $y_j={\mathrm{\λ}}_j{s}_j\·{\tilde{y}}_j+{\mathrm{\γ}}_{j}I$ , here, y _jBe j the subvector of y, wherein embedded j the element b of watermark information m _j, s _jWith

Be respectively s and

Subvector and y _jCorresponding, I is that an all elements is all 1 vector.

Further again, by making y _jVariance equal x _jVariance determine λ _j

Further again, by minimizing y _jWith x _jEuler distance determine γ _j

Further again, need before the watermark detection earlier carrier signal z to be detected to be done following pre-service, at first, it is a fixing constant that conversion carrier signal z makes its arithmetic mean; And then element corresponding being divided by of the signal after the conversion and a characteristic sequence s ' obtained signal

Here, corresponding identical to the employed method of carrier signal x when the method for designing of characteristic sequence s ' all embedded with watermark during the transform method that is adopted in the first step and second went on foot, related constant the two also need consistent.

Further, use certain statistic of carrier signal x or its subvector, for example standard deviation value, Minkowski norm value etc. are formed a sequence isometric with carrier signal, and the method for design feature sequence s is to make itself and aforementioned sequence in direct ratio.

Further, signal that amplitude scale transformation and constant retouching operation is had unchangeability of structure

The time be not limited to only use the information of carrier signal self, all be suitable for irrelevant other signal of watermark embed process in the carrier content.

Further again, for promoting the performance of watermark, design a weight sequence v isometric with carrier signal, each element of v is value between 0 and 1 all, uses the 1-v weighted error signal

, and with the error signal after the weighting

Signal again is added to On obtain

, promptly ${\tilde{y}}_c=(1-v)\·(\tilde{x}-\tilde{y})+\tilde{y}$ Then, use

Replace

Generate the carrier signal y that adds watermark, its method and utilization

Generate the method unanimity of y.Here,

The signal that amplitude scale transformation and constant retouching operation are had unchangeability that expression carrier signal x obtains through certain conversion,

Expression adopts quantisation watermarking to embed the device processing signals

The signal of gained.

Further again, design weight sequence v satisfies the carrier signal that the signal configuration method operation of using contains watermark when utilizing watermark to embed

Shi Ze obtains one and signal

Identical or approach

Signal.Here,

The expression signal

Embed device through quantisation watermarking and handle the signal that passes through gained after the error compensation again,

The signal that amplitude scale transformation and constant retouching operation is had unchangeability that expression carrier signal x obtains through certain conversion.

Effect of the present invention is: adopt method of the present invention, do not having under the situation of attacking, watermark detection can obtain the zero error rate; Can resist the amplitude convergent-divergent of any degree in theory and attack and constant modification attack, the two is had unchangeability; The acquisition of this kind unchangeability does not rely on sensor model, thereby more flexible; Simultaneously, attack, can obtain the performance close with original quantisation watermarking for other.

Description of drawings

Fig. 1 is that watermark of the present invention embeds schematic flow sheet;

Fig. 2 is a watermark detection schematic flow sheet of the present invention;

Fig. 3 is that the amplitude convergent-divergent of the embodiment of the invention is attacked experiment;

Fig. 4 is that the modification DC component of the embodiment of the invention is attacked experiment;

Fig. 5 is that the Gaussian noise of the embodiment of the invention is attacked experiment;

Fig. 6 is the JPEG compression attack experiment of the embodiment of the invention.

Embodiment

Describe a specific embodiment of the present invention below in conjunction with accompanying drawing, and further specify effect of the present invention.

With watermark carrier A _oBe a picture signal, realize a kind of normalized quantization-based watermarking, watermark embed process may further comprise the steps as shown in Figure 1:

1) from A _oSpatial domain extract carrier signal x.Earlier with A _oBe divided into a plurality of nonoverlapping sub-pieces, every contains 8 * 8 picture elements, chooses 750 sub-pieces of local variance maximum then.Sub-piece for each selection forms a line the pixel that it comprises again by row.Next, 750 column vectors of gained are formed a line sequentially forms a big column vector as carrier signal x, and its length is N=750 * 64.This step is realized by the carrier signal draw-out device.

2) the information m that will embed is expressed as one and longly is the binary sequence b of p, wherein each element b _j∈ Ω, Ω=0, and 1}, j=1 ..., p.This step is realized by information representation and code device.

3) for the sake of simplicity, present embodiment is divided into p isometric subsequence with carrier signal x, is designated as x=x ₁‖ x ₂‖ ... ‖ x _p, wherein

, L is the length of each subsequence,

Be that casting out rounds operational character.Here, suppose that it is R=1/L that watermark embeds bit rate.

4) calculated characteristics sequence s.Present embodiment adopts each subvector x _jStandard deviation value construction feature sequence s.If function Var{} represents biased operation of vector.Particularly, j the subvector s of characteristic sequence s _jBe taken as

$s_j=\sqrt{\mathrm{Var}\left\{x_j\right\}}{I}^L,j=1,\·\·\·,p---\left(1\right)$

Here, I represents one and x _jIsometric, and all elements is all 1 vector.This step is realized by the characteristic sequence extraction element.

5) the carrier signal x that treats embed watermark carries out signal that amplitude scale transformation and constant retouching operation is had unchangeability of transition structure

Earlier with each subvector x _jArithmetic mean transform to zero point, the method that present embodiment adopts is with x _jEach component deduct E{x _j, wherein, function E{} represents arithmetic mean Value Operations of vector.Then, again with the signal x of gained _j-E{x _jAnd characteristic sequence s _jCorresponding element be divided by and obtain

, be designated as

${\tilde{x}}_j=(x_j-E\left\{x_j\right\})./s_j,j=1,\·\·\·,p---\left(2\right)$

This step is realized by the carrier signal converting means.

6) utilize the DM method that each element of sequence b is embedded into signal

Corresponding subvector in obtain adding the signal of watermark

, be expressed as

${\tilde{y}}_j=Q_{b_j}\left({\tilde{x}}_j\right),j=1,\·\·\·,p---\left(3\right)$

This step is realized by information embedding device.

7) pass through signal Implement the carrier signal y that a series of conversion obtain embed watermark.The method that present embodiment adopts is with subvector s _jWith

Respective components multiply each other, and then multiply by factor lambda _j, a constant γ at last again superposes each component of gained vector _j, promptly

$y_j={\mathrm{\λ}}_j{s}_j\·{\tilde{y}}_j+{\mathrm{\γ}}_{j}I,j=1,\·\·\·,p---\left(4\right)$

Here, y _j, s _jWith Be respectively y, s and

J subvector, factor lambda _jBy condition Var{y _j}=Var{x _jDetermine constant γ _jBy minimized distance ‖ y _j-x _j‖ determines.This step is realized by the carrier signal inverter.

8) signal calculated

With

Between error signal, be designated as $q_e=\tilde{x}-\tilde{y}$ 。Design a weight sequence v, its each element is value between 0 and 1 all.With 1-v weighted error signal q _e, and the weighted signal of gained is added to obtains a new signal on the carrier signal y that adds watermark

, promptly

${\tilde{y}}_c=\tilde{y}+q_e\·(1-v)---\left(5\right)$

Select v to make

λ _j＝1

，j＝1，...，p(6)

γ _j＝E{x _j}

Set up or approximate the establishment.This step is realized by error adjustment and compensation system.Then, use

Replace Generate the carrier signal y that another contains watermark _c, be designated as

$y_{c_j}={\mathrm{\λ}}_j{s}_j\·{\tilde{y}}_{c_j}+{\mathrm{\γ}}_{j}I,j=1,\·\·\·,p---\left(7\right)$

9) generate the picture material A that adds watermark by the carrier signal that adds watermark _wPresent embodiment will add the carrier signal y of watermark earlier _cOr each component of y is converted into integer by the method that rounds up, order is 255 greater than 255 component of signal value wherein then, order is 0 less than 0 component value, other component value remains unchanged, at last replace initial carrier signal x, and its each component is placed among the original signal x on the position of corresponding picture element in spatial domain with the signal after the conversion.This step is realized by the contrary draw-out device of carrier signal.

A given picture signal A to be detected _u, watermark extraction process may further comprise the steps as shown in Figure 2:

1) from A _uSpatial domain extract carrier signal z.This step is identical with the step 1 of watermark embed process, is realized by the carrier signal draw-out device.

2) step 3 with watermark embed process is identical, and carrier signal z is divided into p isometric subsequence, is designated as z=z ₁‖ z ₂‖ ... ‖ z _p

3) calculated characteristics sequence s '.Adopt the method identical to extract characteristic sequence with the step 4 of watermark embed process.Particularly, j the subvector of characteristic sequence s ' Be taken as

$s_j^{\′}=\sqrt{\mathrm{Var}\left\{z_j\right\}}{I}^L,j=1,\·\·\·,p---\left(8\right)$

This step is realized by the characteristic sequence extraction element.

4) adopt the method conversion to be detected carrier signal z identical to obtain a signal that amplitude scale transformation and constant retouching operation is had unchangeability with the step 5 of watermark embed process

, be designated as

${\tilde{z}}_j=(z_j-E\left\{z_j\right\})./s_j^{\′},j=1,\·\·\·,p---\left(9\right)$

This step is realized by the carrier signal converting means.

5) utilize the watermark detector of DM method from signal

Each subvector in watermark information that extract to hide

, be expressed as

${\hat{b}}_j=\arg \underset{b_j\∈\mathrm{\Ω}}{\min }\left|\right|{\tilde{z}}_j-Q_{b_j}\left({\tilde{z}}_j\right)\left|\right|,j=1,\·\·\·,p---\left(10\right)$

This step is realized by information extracting device.

6) with the watermark sequence that extracts

Obtain watermark information through decoding This step is realized by information decoding device.

Be illustrated from principle below by the performance of present embodiment the water mark method of the present invention's proposition.For convenience, adjust when adding the carrier signal of watermark not introducing weight sequence v, claim that method of the present invention is NQW, otherwise be called the NQW that is with error compensation, note by abridging and be DC-NQW.

Get by (1)-(4)

$\mathrm{Var}\left\{y_j\right\}={\mathrm{\&lambda;}}_{\mathrm{<figure-callout\; id="2"\; label="j"\; filenames="A200810155544E00161.png"\; state="\{\{state\}\}">j</figure-callout>}}^2\mathrm{Var}\left\{x_j\right\}\mathrm{Var}\{{\tilde{x}}_j+q_{e_j}\}$

Therefore, as condition Var{y _j}=Var{x _jWhen setting up, must have

${\mathrm{\&lambda;}}_j=1/\sqrt{\mathrm{Var}\{{\tilde{x}}_j+q_{e_j}\}},j=1,\&CenterDot;\&CenterDot;\&CenterDot;,p---\left(11\right)$

For making y apart from ‖ _j-x _jThe ‖ minimum, can push away by (1)-(4)

${\mathrm{\&gamma;}}_j=E\left\{x_j\right\}-{\mathrm{\&lambda;}}_j\sqrt{\mathrm{Var}\left\{x_j\right\}}E\left\{q_{e_j}\right\},j=1,\&CenterDot;\&CenterDot;\&CenterDot;,p---\left(12\right)$

Wherein,

Be error vector q _eJ subvector.In (11) and (12) substitutions (4), can push away

$\frac{y_j-E\left\{y_j\right\}I}{\sqrt{\mathrm{Var}\left\{y_j\right\}}}={\mathrm{\&lambda;}}_j(Q_{b_j}\left({\tilde{x}}_j\right)-E\left\{q_{e_j}\right\}I),j=1,\&CenterDot;\&CenterDot;\&CenterDot;,p---\left(13\right)$

Following formula shows not to be had under the situation of attacking, and NQW has introduced following two kinds of additional operations can influence watermark detection: parameter is

Constant retouching operation and zoom factor be λ _jThe amplitude zoom operations.

In order to estimate

And λ _jTo the NQW Effect on Performance, introduce the document watermark than (thedocument-to-watermark ratio DWR), may be defined as j subvector

${\mathrm{\&zeta;}}_j\stackrel{\mathrm{\&Delta;}}{=}\mathrm{LVar}\left\{x_j\right\}/{\left|\right|y_j-x_j\left|\right|}^2$

Wherein, L is subvector x _jLength.Utilize this definition, and associating (1)-(4), (11) and (12) can push away

${\mathrm{\&lambda;}}_j=(1-\frac{1}{2{\mathrm{\&zeta;}}_j})/(1+\frac{q_{e_j}^T{\tilde{x}}_j}{L}),j=1,\&CenterDot;\&CenterDot;\&CenterDot;,p---\left(14\right)$

Studies show that

With

Be to add up independently, so can release λ by (14) _j≈ 1-0.5/ ζ _jThis shows, work as ζ _jDuring 〉=20dB, in actual applications, this condition can satisfy usually, at this moment | and λ _j-1| is quite little, does not detect wrong the appearance so that do not have.For

As long as satisfy condition $\left|E\left\{q_{e_j}\right\}\right|<\mathrm{\&Delta;}/4$ , just can not detect mistake.When Δ less, when L is enough big, It is approximate that to satisfy average be that 0 variance is a Δ ²/ (12L) Gaussian distribution.Therefore, $\left|E\left\{q_{e_j}\right\}\right|<\mathrm{\&Delta;}/4$ The probability of setting up is

$P(\left|E\left\{q_{e_j}\right\}\right|<\mathrm{\&Delta;}/4)=1-\mathrm{erfc}\left(\sqrt{3/8L}\right),j=1,\&CenterDot;\&CenterDot;\&CenterDot;,p---\left(15\right)$

Wherein, Function e rfc () is defined as

$\mathrm{erfc}\left(u\right)\stackrel{\mathrm{\&Delta;}}{=}\frac{2}{\sqrt{\mathrm{\&pi;}}}{\&Integral;}_u^{\&infin;}{e}^{-{\mathrm{<figure-callout\; id="2"\; label="t"\; filenames="A200810155544E00161.png"\; state="\{\{state\}\}">t</figure-callout>}}^2}\mathrm{dt}$

By (15) as can be known, probability $P(\left|E\left\{q_{e_j}\right\}\right|<\mathrm{\&Delta;}/4)$ Along with the increase of L approaches 1, that is to say when L is enough big condition $\left|E\left\{q_{e_j}\right\}\right|<\mathrm{\&Delta;}/4$ Must set up, at this moment,

Can not introduce bit error rate to NQW.

Refer now to why NQW is a robust to attack of amplitude convergent-divergent and constant modification attack.As the carrier signal y that adds watermark _jBy flexible ρ _jDoubly, again its each component is added C _j, the signal z after obtaining attacking _j, be designated as z _j=ρ _jy _j+ C _jI.This formula with in (8) and (9) substitutions together (10), be can be observed attack parameter ρ _jAnd C _jBe cancelled in expression formula, just watermark detection process does not rely on ρ _jAnd C _j

Next, analyze the performance of NQW under the situation that noise occurs.y _jThrough obtaining signal z behind the additive noise passage _j, be designated as z _j=y _j+ n _j, n wherein _jThe expression average is 0 unknown noise, and n _jAnd y _jVertical mutually.Because NQW attacks the amplitude convergent-divergent in theory and constant modification attack has unchangeability, therefore considers that above-mentioned situation is sufficient.In order to assess The noise, (thewatermark-to-noise ratio WNR), may be defined as j subvector to introduce the watermark noise ratio

${\mathrm{\&xi;}}_j\stackrel{\mathrm{\&Delta;}}{=}{\left|\right|y_j-x_j\left|\right|}^2/{\left|\right|n_j\left|\right|}^2$

Then, can release by (1)-(4), (8) and (9) and (11) and (12)

${\tilde{z}}_j={\mathrm{\&lambda;}}_j^{\&prime;}(Q_{b_j}\left({\tilde{x}}_j\right)+n_j^{\&prime;}),j=1,\&CenterDot;\&CenterDot;\&CenterDot;,p---\left(16\right)$

Wherein, ${\mathrm{\&lambda;}}_j^{\&prime;}={\mathrm{\&lambda;}}_j\sqrt{\frac{{\mathrm{\&zeta;}}_j{\mathrm{\&xi;}}_j}{{\mathrm{\&zeta;}}_j{\mathrm{\&xi;}}_j+1}},$ $n_j^{\&prime;}=\frac{n_j}{{\mathrm{\&lambda;}}_j\sqrt{\mathrm{Var}\left\{x_j\right\}}}-E\left\{q_{e_j}\right\}I$ 。Usually, in actual applications, ζ _jξ _j" 1, therefore

Quite little.At this moment, the detecting position error rate mainly from

Ignore

With

Influence, NQW can resist the noisiness identical with DM.

For DC-NQW, can push away by (5)

$y_{c_j}={\mathrm{\&lambda;}}_j\sqrt{\mathrm{Var}\left\{x_j\right\}}({\tilde{x}}_j+v_j\&CenterDot;q_{e_j})+{\mathrm{\&gamma;}}_{j}I,j=1,\&CenterDot;\&CenterDot;\&CenterDot;,p---\left(17\right)$

Select v to make $(y_{c_j}-E\left\{y_{c_j}\right\})/\mathrm{Var}\left\{y_{c_j}\right\}={\tilde{y}}_{c_j}$ Set up or approximate the establishment.Can find out that by (12), (17) this condition can be converted into condition (6).Utilize (11), (12) and (17), condition (6) can be expressed as again again

$\mathrm{Var}\{{\tilde{x}}_j+v_j\&CenterDot;q_{e_j}\}=1$

，j＝1，...，p(18)

$E\{v_j\&CenterDot;q_{e_j}\}=0$

Separate this system of equations and can obtain weight vectors v.Usually, v _jLength L satisfy L 2, at this moment, equation (18) has or not array to separate, and can therefrom select one group and separate the best performance that makes watermark.Notice that introducing weight vectors v not only can eliminate (or part is eliminated) λ _jWith

Influence, also can compensate the embedding distortion.Well-known this technology can improve the performance of DM watermark.

In order to show remarkable result of the present invention, present embodiment provides some experimental results.Select 4000 width of cloth images to be used for test from the Corel database, every width of cloth size of images is 256 * 384.For every width of cloth image, the carrier signal x that can to extract a total length in watermark embed step 1 be N=48000.The binary sequence b that length that produces at random is p=1500 is embedded in each width of cloth image as watermark information m, and wherein, per 32 pixels embed 1 bit information, i.e. L=32, and j=1 ..., p.Measure the picture quality spike signal to noise ratio (S/N ratio) (PSNR, peak signal-to-noise ratio) of embed watermark.The performance of the water mark method that provides by several typical attack experiment test embodiment.Simultaneously, for purpose relatively, provided the test result of DM and several improvement DM, comprise and be with the DM of distortion compensation (to abbreviate DC-DM as, please refer to document B.Chenet al..Quantization index modulation:a class of provably good methods fordigitalwatermarking and information embedding.IEEE Trans.Inform.Theory, vol.47, no.4, pp.1423-1443,2001.), compensating factor is taken as 0.66, the Oostveen method (abbreviates ODM as, J.C.Oostveen et al..Adaptive quantization watermarking.in Proc.of SPIE:Security, Steganography, and Watermarking of Multimedia Contents VI, 2004, vol.5306 is the RDM method of base pp.296-303) and with Euler's norm, and wherein the memory span factor is taken as 50.In all experiments, PSNR is taken as 35dB, and this can realize by adjusting quantization step.The bit error rate that provides in the accompanying drawing is the mean value of test gained result on all images.

Fig. 3 has provided one group of amplitude convergent-divergent attack test result.Clearly, NQW and DC-NQW have showed best performance under convergent-divergent is attacked, even when ρ≤1.2, they have obtained the zero error rate.Curious is, RDM has relatively poor relatively performance in this respect, and this is owing to round off and break-in operation has had a strong impact on the performance of RDM.

Fig. 4 has provided one group and has revised DC component attack test result.DC component is modified by the identical amount of the value increase/minimizing of each picture element of image.As can be seen, to this attack operation, the performance of ODM and RDM all is better than original DM from the figure, and NQW that present embodiment provides and DC-NQW then have optimum performance.

Fig. 5 has provided one group of Gaussian noise attack test result.Fig. 5 shows that NQW and DC-NQW have obtained to be better than the performance, particularly DC-NQW of original DM far away.This is because NQW and the employed quantization step of DC-NQW are proportional to the local variance of image, thereby the quantization step that adopts when having caused watermark to embed is bigger with respect to the quantization step that DM uses, and has promoted the ability that watermark opposing Gaussian noise is attacked.

Fig. 6 has provided one group of JPEG compression attack test result.Fig. 6 shows that the performance of NQW and DC-NQW is better than DM and other three kinds of improved DM methods, and its reason is consistent with the situation that Gaussian noise is attacked.

The experimental result that present embodiment obtains can illustrate, uses method of the present invention not only to realize the robustness that the amplitude convergent-divergent is attacked and the constant modification is attacked, and has obtained the performance more excellent than original DM watermark in other several typical attack operations.

Present embodiment has provided the implementation result of image watermark, handles but the invention is not restricted to image watermark, and the present invention is equally applicable to multimedia digital watermark processing such as digital music, video.

The present invention can implement with other concrete form, and does not break away from its spirit or essential characteristic.It only is illustrative and nonrestrictive that described embodiment is considered in all respects, for example:

1) the carrier signal type is not limited to image, audio frequency and video etc.;

2) embed the territory and be not limited to spatial domain, DCT territory, Fourier transform domain and wavelet field etc.;

3) design of characteristic sequence is not limited to use mean value, standard deviation, and various norms etc.;

4) water mark method that is adopted is not limited to the jitter quantisation and watermark modulation of jitter quantisation and watermark modulation, band distortion compensation and based on the quantification dithering modulation of watermarking of transform expansion etc.;

5) selection of various parameters etc.

Therefore, scope of the present invention by appended claims but not foregoing description indicate.Falling into the meaning of equivalence techniques scheme of claim and all changes of scope is included among its scope.

Claims (10)

1. normalized quantization-based watermarking, when embed watermark, the carrier signal x that at first treats embed watermark carries out a kind of signal that amplitude convergent-divergent and constant retouching operation are had unchangeability of certain transition structure

Adopt quantisation watermarking to embed device then and handle this signal, again with the signal that obtains

Be converted into the carrier signal y that contains watermark; When detecting watermark information, the carrier signal z that the signal configuration method of using when utilizing watermark to embed is earlier operated watermark to be detected obtains signal Adopt the signal of quantisation watermarking detecting device then from gained

The middle watermark information that extracts is characterized in that, watermark embeds and detects and all comprises a signal configuration step that amplitude convergent-divergent and constant modification attack is had unchangeability.

2. a kind of normalized quantization-based watermarking as claimed in claim 1 is characterized in that: construct a kind of signal that amplitude scale transformation and constant retouching operation are had unchangeability

The method that adopts is first conversion carrier signal x, making its arithmetic mean is a fixing constant, the span of this constant is unrestricted, and then with element corresponding be divided by of the signal after the conversion with a characteristic sequence s, characteristic sequence s satisfies following character: when the amplitude of carrier signal is scaled, and the scaled together thereupon identical multiple of s.

3. a kind of normalized quantization-based watermarking as claimed in claim 1 or 2 is characterized in that: by conversion

The signal that behind the watermark embedding operation, obtains

Acquisition contains the carrier signal y of watermark, and its method is with subvector s _jWith Respective components multiply each other, and then multiply by factor lambda _j, a constant γ at last again superposes each component of gained vector _j, just $y_j={\mathrm{\&lambda;}}_j{s}_j\&CenterDot;{\tilde{y}}_j+{\mathrm{\&gamma;}}_{j}I,$ Here, y _jBe j the subvector of y, wherein embedded j the element b of watermark information m _j, s _jWith

Be respectively s and

Subvector and y _jCorresponding, I is that an all elements is all 1 vector.

4. a kind of normalized quantization-based watermarking as claimed in claim 3 is characterized in that: by making y _jVariance equal x _jVariance determine λ _j

5. a kind of normalized quantization-based watermarking as claimed in claim 3 is characterized in that: by minimizing y _jWith x _jEuler distance determine γ _j

6. a kind of normalized quantization-based watermarking as claimed in claim 2 is characterized in that: need before the watermark detection earlier carrier signal z to be detected to be done following pre-service, at first, it is a fixing constant that conversion carrier signal z makes its arithmetic mean; And then element corresponding being divided by of the signal after the conversion and a characteristic sequence s ' obtained signal

Here, corresponding identical to the employed method of carrier signal x when the method for designing of characteristic sequence s ' all embedded with watermark during the transform method that is adopted in the first step and second went on foot, related constant the two also need consistent.

7. a kind of normalized quantization-based watermarking as claimed in claim 2, it is characterized in that: certain statistic of using carrier signal x or its subvector, for example standard deviation value, Minkowski norm value etc. are formed a sequence isometric with carrier signal, and the method for design feature sequence s is to make itself and aforementioned sequence in direct ratio.

8. a kind of normalized quantization-based watermarking as claimed in claim 2 is characterized in that: construct a signal that amplitude scale transformation and constant retouching operation is had unchangeability

The time be not limited to only use the information of carrier signal self, all be suitable for irrelevant other signal of watermark embed process in the carrier content.

9. a kind of normalized quantization-based watermarking as claimed in claim 1 is characterized in that: for promoting the performance of watermark, design a weight sequence v isometric with carrier signal, each element of v is value between 0 and 1 all, uses the 1-v weighted error signal

And with the error signal after the weighting Signal again is added to

On obtain Promptly ${\tilde{y}}_c=(1-v)\&CenterDot;(\tilde{x}-\tilde{y})+\tilde{y};$ Then, use

Replace

Generate the carrier signal y that adds watermark, its method and utilization

Generate the method unanimity of y; Here,

The signal that amplitude scale transformation and constant retouching operation are had unchangeability that expression carrier signal x obtains through certain conversion,

Expression adopts quantisation watermarking to embed the device processing signals

The signal of gained.

10. a kind of normalized quantization-based watermarking as claimed in claim 9 is characterized in that: design weight sequence v satisfies the carrier signal that the signal configuration method operation of using contains watermark when utilizing watermark to embed

Shi Ze obtains one and signal Identical or approach

Signal, here,

The expression signal

Embed device through quantisation watermarking and handle the signal that passes through gained after the error compensation again,

The signal that amplitude scale transformation and constant retouching operation is had unchangeability that expression carrier signal x obtains through certain conversion.

Images