CN105335924A - Wavelet domain color image watermark encryption algorithm based on differential evolution - Google Patents

Abstract

The invention discloses a wavelet domain color image watermark encryption algorithm based on differential evolution. The operating method comprises the following steps: performing color space transformation on an original image to obtain a YIQ color format suitable for a human vision system and extracting a brightness component Y; adopting Arnold scrambling on a watermark image; performing three-stage discrete wavelet decomposition on the component Y of the original image to generate four sub-bands of different frequencies by utilizing discrete wavelet transform, thereby obtaining a sub-band image; representing a sub-band image matrix by using A, and performing singular value decomposition on the image subjected to three-stage wavelet decomposition; performing primary discrete wavelet decomposition on the watermark image subjected to scrambling encryption processing; performing watermark embedding; adaptively optimizing and selecting embedding strength by adopting differential evolution, and adding watermark information into the original image subjected to singular value decomposition in a multiplicative addition form. The invention has the beneficial effects that the watermark algorithm has high invisibility and is high in robustness.

Description

Based on the wavelet field Color digital watermarking cryptographic algorithm of differential evolution

Technical field

The invention belongs to image algorithm technical field, relate to the wavelet field Watermarking of Color based on differential evolution.

Background technology

Along with the continuous progress of computer technology, the communication technology and multimedia technology, various digital product obtains wide-scale distribution in a network with the form of Digital Media, make the mode of production and life of the mankind there occurs profound change, promote the development course of social informatization.But, enjoy while digitizing technique offers convenience people, also occurred a lot of negative, passive impact.Such as, the leakage of digital product information, distort and forge, even sought unlawful interests by as commercial use, the grievous injury legitimate interests of product copyright owner.At present, mainly contain two kinds of settling modes for these phenomenons, one is strengthen the construction of national digital copyright protection laws and regulations, increases the punishment dynamics to illegal infringer ^[1,2]; Two is improve constantly information security technology level, passes through technological means ^[3-5]realize the copyright protection of digital product resource.At present, the problem utilizing digital watermark technology to realize multimedia resource copyright protection is subject to people always and pays close attention to, and is widely applied in a lot of fields.

The invisibility of most important two characteristics of digital watermarking algorithm and algorithm and robustness.But there is certain contradiction in both, on the one hand, little embedment strength is used to be conducive to the sightless of watermark, but bad to the robustness of several frequently seen attack, on the other hand, the excessive robustness that improve watermark of embedment strength, but poor to the invisibility of watermark.Therefore, suitable embedment strength and then the relation both balance how is selected to be improve the key issue of algorithm performance.In recent years, some optimized algorithms, such as simulated annealing (SimulatedAnnealing, SA), genetic algorithm (GeneticAlgorithm, GA), Particle Swarm Optimization (ParticleSwarmoptimization, PSO), ant group algorithm (AntColonyOptimization, ACO) etc. by simulating or disclose some spontaneous phenomenon or process and being developed gradually, provide new thinking and means for solving optimization problem.PereiraS, VoloshynoskiyS etc. ^[6]propose a kind of method utilizing best mode to consider spatial domain restriction, using watermark embedment as a linear programming problem, lose its linear restriction in a series of pixel under, expect the intensity of watermark to maximize.The people such as E.Vahedi, R.A.Zoroofi ^[7]utilize bionic optimization principle to propose a kind of Color Images Watermarking Algorithm based on wavelet transformation newly, optimize digital watermarking algorithm in conjunction with bionics principle.

Differential evolution algorithm (DifferentialEvolutionAlgorithm, DEA) be a kind of evolution algorithmic based on population difference, there is stronger ability of searching optimum and rate of convergence, solving in complicated Global Optimal Problem, be proved to be a kind of searching algorithm of effective globally optimal solution ^[8].2009, V.Aslantas ^[9]propose the Robust Digital Watermarking adopting differential evolution and svd first, this proof of algorithm differential evolution can balance the relation between digital watermarking invisibility and robustness well, and improves Algorithm robustness to a certain extent.Afterwards, differential evolution is widely used in digital watermarking field ^[10-15].Based on this, digital image watermarking algorithm in wavelet domain based on differential evolution is proposed herein, and the coloured image adopting applicability stronger is as the carrier image of algorithm, in conjunction with features such as the many resolutions of wavelet transform, signal partial analysis on the basis of svd, traditional Watermarking of Color is optimized, ensure that watermaking system preferably improves the applicability of algorithm under invisibility and robustness prerequisite, before embed watermark, pre-service is carried out to watermark information in addition, also improve the security of algorithm simultaneously.

Summary of the invention

The object of the present invention is to provide the wavelet field Color digital watermarking optimized algorithm based on differential evolution, solve the robustness of existing digital watermarking algorithm to several frequently seen attack bad, the problem poor to the invisibility of watermark.

The technical solution adopted in the present invention is carried out according to following steps:

Step 1: first carry out color space conversion to original image, converts the YIQ color format of applicable human visual system to and extract light intensity level Y by rgb format;

Step 2: Arnold scramble is adopted to watermarking images;

Step 3: utilize wavelet transform to carry out four different frequency sub-bands of three grades of discrete wavelet transformation generations to the Y-component of original image, obtain sub-band images;

Step 4: represent sub-band images matrix with A, uses svd to the image after three grades of wavelet decomposition;

Step 5: one-level discrete wavelet transformation is carried out to the watermarking images after scramble encryption;

Step 6: watermark embedment: add watermark information in the original image of the form adopting differential evolution adaptive optimization to select embedment strength to be added with multiplicative after svd;

Further, in described step 6, the step of watermark embedment is as follows:

1) color space conversion is carried out to original image A, rgb format is converted to YIQ form, extract the luminance component Y of image.Utilize wavelet transform to carry out three grades of wavelet decomposition to Y-component, obtain the subband of four different frequencies, be i.e. low frequency ll channel LL and three high frequency detail subgraph LH, HL, a HH.

2) respectively svd is carried out to four subbands obtained in the previous step, try to achieve corresponding singular value S.

Y _k＝U _kS _kV _k ^Tk∈(LL、LH、HL、HH)

3) carry out Arnlod scramble to watermarking images W and obtain the watermarking images W' after scramble, scramble number of times gets τ, and adopts one-level wavelet decomposition to obtain a low frequency ll channel LL and three high frequency detail subgraph LH, HL, HH.

4) matrix corresponding with watermarking images respectively for singular value corresponding for Y-component is carried out multiplicative addition, its embedment strength is q, and this value is obtained by differential evolution algorithm.

S _k+q _kW' _k＝C _kk∈(LL、LH、HL、HH)

5) svd is carried out to Matrix C and obtain matrix U _w, S _w, V _w ^t.

C _k＝U _wkS _wkV _wk ^Tk∈(LL、LH、HL、HH)

6) newly singular value S will be obtained _wwith unitary matrix U and V in step (2) ^tcarry out the inverse decomposition of singular value.

U _kS _wkV _k ^T＝Y _wkk∈(LL、LH、HL、HH)

Recycle wavelet transform afterwards to carry out three grades of wavelet inverse transformations (3-IDWT) the luminance component Y adding image after watermark can be obtained _w'.

7) luminance component Y _w' together with I component in the YIQ form of original image A and Q component combine, and be rgb format by YIQ format conversion, coloured image A with watermarked information can be obtained _w.

Further, if A _w ^*represent the coloured image with watermarked information through attacking, the watermarking images W so extracted ^*then by main following steps:

1) step 1 in watermark embed process is used) and 2) to A _w ^*carry out processing and obtain three corresponding matrix U respectively ^*, S _w ^*, V ^{* T}.

Y _wk ^*＝U _k ^*S _wk ^*V _k ^*Tk∈(LL、LH、HL、HH)

2) by singular value corresponding for original image A luminance component Y respectively with Matrix C after attacking ^*carry out except property subtract each other, factor q remains unchanged, obtain matrix W ' ^*.

W _k' ^*＝(C _k ^*-S _k)/q _kk∈(LL、LH、HL、HH)

3) by matrix W ' ^*carry out one-level wavelet inverse transformation, just extract the W' of the scramble watermarking images after attack ^*.

4) to W' ^*carry out (T-τ) secondary Arnold conversion, T is shuffling cycle, can obtain the watermarking images W extracted ^*.

The invention has the beneficial effects as follows that watermarking algorithm has certain security, and obtain good invisibility, can resist common image processing operations, can extract with original watermark image similarity higher containing watermarking images after under attack, the robustness embodying watermarking algorithm is stronger.

Accompanying drawing explanation

Fig. 1 is the inventive method steps flow chart schematic diagram;

Fig. 2 is Arnold conversion different scramble number of times design sketch;

Fig. 3 is three grades of discrete wavelet transformation schematic diagram;

Fig. 3 (a) original image;

Fig. 3 (b) is decomposable process figure;

Fig. 3 (c) is the image after decomposing;

Fig. 4 is differential evolution algorithm implementation process flow diagram;

Fig. 5 is that the NC value of algorithm of the present invention and document [16] contrasts broken line graph.

Embodiment

Below in conjunction with embodiment, the present invention is described in detail.

The global convergence ability that the present invention utilizes differential evolution algorithm stronger and robustness, a kind of wavelet field Watermarking of Color based on differential evolution is proposed, first color space conversion is carried out to original image, rgb format is converted to the YIQ color format extract light intensity level Y that are more suitable for human visual system, then wavelet transform is utilized to carry out four different frequency sub-bands of three grades of wavelet decomposition generations to Y-component, use svd, secondly one-level discrete wavelet transformation is carried out to the watermarking images after scramble encryption, the form finally adopting differential evolution adaptive optimization to select embedment strength to be added with multiplicative adds watermark information.Emulation experiment shows, this algorithm has good invisibility, algorithm embodies stronger robustness to common image process operations such as noise, rotation, shearing, translation and compressions, differential evolution algorithm is utilized to balance the contradiction of general watermarking algorithm between invisibility and robustness, and this algorithm is applicable to adding watermark information to coloured image, therefore has good practicality.The inventive method step is as shown in Figure 1:

Step 1: first carry out color space conversion to original image, converts the YIQ color format of applicable human visual system to and extract light intensity level Y by rgb format.Human visual system (HumanVisualSystem, HVS), primarily of human eye and visual central nervous system system composition, is a kind of senior intelligent information handling system.In digital watermark technology; more subjective assessment is pressed close in order to enable the effect of method for objectively evaluating; usual meeting introduces human visual system in the process of image quality evaluation, and the many aspects such as accuracy, monotonicity done like this evaluating all improve a lot.Wherein, brightness is the most basic a kind of characteristic in vision system, and during design digital watermarking algorithm, at the highlight regions embed watermark information of image, the invisibility of algorithm is better.Color space mainly comprises RGB, the forms such as YIQ, CMY, and wherein rgb format is mainly used in Computer display, and YIQ is used for Image Communication.Therefore, when to coloured image embed watermark information, according to the light characteristic of HVS, by the conversion of color space, RGB mode image is converted to YIQ color space.Due to relative to color difference components I and Q, human eye is insensitive for luminance component Y, so embed watermark information can improve the invisibility of watermark in Y-component.Mutual transformation definition between RGB and YIQ form is

$\begin{array}{c}\left[\begin{array}{c}Y\\ I\\ Q\end{array}\right]=\left[\begin{array}{ccc}0.299& 0.587& 0.114\\ 0.596& -0.275& -0.321\\ 0.212& -0.523& 0.311\end{array}\right]\·\left[\begin{array}{c}R\\ G\\ B\end{array}\right]\\ \left[\begin{array}{c}R\\ G\\ B\end{array}\right]=\left[\begin{array}{ccc}1.0& 0.956& 0.620\\ 1.0& -0.272& -0.647\\ 1.0& -1.108& 1.703\end{array}\right]\·\left[\begin{array}{c}Y\\ I\\ Q\end{array}\right]\end{array}---\left(1\right)$

Step 2: Arnold scramble is adopted to watermarking images;

Image scrambling belongs to image encryption technology, refers to and out of order process is carried out in the pixel space position of piece image, destroy the correlativity between original image pixels, to ensure the security of image in transmitting procedure.Disorder method conventional at present mainly contains Arnold conversion, Fibonacci conversion, the conversion of Hilbert curve, affined transformation, magic scrambling, Gray code, orthogonal Latin square conversion etc.The Arnold conversion one conversion that to be V.J.Arnold propose in research ergodic theory, is commonly called as cat face conversion (Arnold ' scatmap) ^[19].Owing to calculating simple, easy realization and having periodically, therefore Arnold conversion is widely used in the disorder processing of watermarking images.Its form of Definition is as follows.

$\left[\begin{array}{c}{x}^{\′}\\ y^{\′}\end{array}\right]=\left[\begin{array}{cc}1& 1\\ 1& 2\end{array}\right]\left[\begin{array}{c}x\\ y\end{array}\right]\left(\mathrm{mod}N\right)x,y\∈\{0,1,\mathrm{...},N-1\}---\left(2\right)$

Wherein (x, y) pixel that is original image, (x ', y ') is the pixel value of new figure image after conversion.General for square, the size of N image.The design sketch obtained after adopting Arnold conversion to carry out pre-service to watermarking images as shown in Figure 2.Can find out, at original watermarking images of getting back after T (T=96) secondary process through transformation period, therefore piece image obtains the image identical with original watermark surely through Arnold conversion one repeatedly.

Step 3: utilize wavelet transform to carry out four different frequency sub-bands of three grades of wavelet decomposition generations to the Y-component of original image, obtain sub-band images;

Adopt wavelet transform (DiscreteWaveletTransform, DWT).By DWT, original image is resolved into four sub-band images of different spaces, different frequency, low frequency sub-band LL and three high-frequency sub-band HL, LH, HH, wherein LL and original image are the most similar, be called approximation subband, HL, LH, LL represent the horizontal direction details of image, vertical direction details and to angular direction details respectively.If decompose further low frequency sub-band, four different subbands can be obtained again.As shown in Figure 2, three grades of wavelet decomposition processes of image and schematic diagram thereof.

Its low frequency components contains most information of original image, and high fdrequency component represents the edge of original image, profile and Texture eigenvalue.What Fig. 3 (c) represented is the different frequency sub-bands design sketch obtained after carrying out three grades of wavelet transforms to the original image Lena shown in Fig. 3 (a).Fig. 3 (b) is decomposable process figure.Step 4: represent sub-band images matrix with A, uses svd to the image after three grades of wavelet decomposition;

Svd (SingularValueDecomposition, SVD) is a kind of by the efficient algorithm of diagonalization of matrix in numerical linear algebra, is the intrinsic characteristic of matrix, plays an important role in matrix theory and matrix computations.In image procossing, SVD has following characteristic ^[21]:

1) singular value of piece image has good stability, and when image is subject to slight disturbance, its singular value significant change can not occur;

2) in the singular value sequence that obtains through SVD process of image, first singular value is more much larger than remaining, if ignore these less singular value items, the picture quality reconstructed also larger degeneration can not occur;

3) singular value can show the algebraic property of image inherence.The extraction of watermark often can be subject to the impact of image geometry operation, the especially extraction of blind watermatking.According to the characteristic of SVD, if in the embedding being applied in watermark and leaching process, carrier image just can be made to bear certain geometric distortion.

Represent sub-band images matrix with A, then svd is carried out to A and can obtain three matrix: unitary matrix U, the associate matrix V of diagonal matrix S and unitary matrix V ^t.If A ∈ is R _{n × n}be a square formation, can be broken down into according to SVD:

A＝USV ^T(3)

Wherein, UU ^t=I _n, VV ^t=I _n, S is diagonal matrix, and the element on its diagonal line is called the singular value of matrix A, adopts σ ₁>=σ ₂>=...>=σ _n>=0 represents, and meets formula (4).If r≤n is the order of matrix A, then A also can be decomposed into (5) formula.

σ ₁≥σ ₂≥...≥σ _r≥σ _r+1≥...≥σ _n＝0(4)

$A=\underset{i=1}{\overset{r}{\Σ}}{\mathrm{\σ}}_i{u}_i{v}_i^T---\left(5\right)$

Wherein, U _i, V _imatrix U respectively, the k rank proper vector of V.

Step 5: one-level discrete wavelet transformation is carried out to the watermarking images after scramble encryption;

Step 6: add watermark information in the original image of the form adopting differential evolution algorithm adaptive optimization to select embedment strength to be added with multiplicative after svd;

(1) differential evolution:

First the individuality that Stochastic choice two is different from population does difference, generates a differential vector, then selects the 3rd individuality and this differential vector to sue for peace thus produces a new individuality; Then according to certain rule, interlace operation is carried out to the parent individuality of new individuality and current population, generate new offspring individual; Last offspring individual and parent individuality competes, and the most the superior of general is saved in colony of future generation.Algorithm, through constantly iterative computation, improves population quality gradually, thus guides population to approach to optimum solution position.As shown in Figure 4, its specific descriptions are as follows for the implementation of differential evolution algorithm:

1) initialization population

Determine population scale NP, population dimension D, produces initial population at random.

X _i(0)＝(x _i,1,x _i,2,...,x _i,D)i＝1,2,...NP(6)

x _i,j＝a _j+rand·(b _j-a _j)j＝1,2,...D(7)

Wherein X _i(0) i-th individuality of initial population is represented, x _i,jrepresent i-th individual jth component, rand is the uniform random number between 0 and 1, a _j≤ x _i,j≤ b _j.

2) mutation operation

The mutual different individual x of Stochastic choice three from current population _r1 ^g, x _r2 ^g, x _r3 ^g, r1 ≠ r2 ≠ r3, G represents the algebraically of current population.With wherein two individual x _r1 ^g, x _r2 ^gdo difference, generate differential vector D _r1,2.

D _r1,2＝x _r1-x _r2(8)

Then new individual V is generated according to formula (9) _i ^g+1, F is a real constant on [0,2], is mutation operator.

V _i ^G+1＝x _r3 ^G+F(x _r1 ^G-x _r2 ^G)(9)

Mutation operation is a most important step in differential evolution algorithm process, and the origin of differential evolution title is also in this.3) interlace operation

The new individual V produced in mutation operation process _i ^g+1with i-th individual x of population _i ^gintersect, generate new experimental subjects U _i,j ^g+1.

U _i,j ^G+1＝(u _i,1 ^G,u _i,2 ^G,...,u _i,j ^G)j＝1,2,...D(10)

${u_{i,j}}^{G+1}=\left\{\begin{array}{c}{v_{i,j}}^{G+1},ifrand\≤CRorj=j_{rand}\\ {x_{i,j}}^{G+1},otherwise\end{array}\right.---\left(11\right)$

Wherein CR ∈ (0,1) is the crossover probability factor, and rand represents the uniform random number between interval (0,1), j _randfor the random integers in interval [1, D], U _i,j ^g+1represent that G+1 is for new population i-th individuality.Interlace operation adds the diversity of population.

4) operation is selected

The experimental subjects u generated after variation and interlace operation _i ^g+1with the individual x of original seed group _i ^gbe at war with, select the equation of operation as follows.

${x_i}^{G+1}=\left\{\begin{array}{cc}{u_i}^{G+1}& iff\left({u_i}^{G+1}\right)\≤f\left({x_i}^G\right)\\ {x_i}^G& else\end{array}\right.---\left(12\right)$

In formula, f is fitness function, x _i ^g+1for i-th individuality of population of future generation.

(2) based on the digital watermarking algorithm of differential evolution

Embed the watermark information of proper strength respectively to ensure the optimization of algorithm at four subbands of original image, wherein embedment strength value is obtained by differential evolution algorithm, the optimum balance of implementation algorithm invisibility and robustness.

Embedment strength

Embedment strength value (q) of different frequency sub-bands corresponding for original image is carried out respectively embedding and the extraction of watermark information.After watermarking images embeds original image, attack carrying out common algorithm containing watermarking images, according to various attack function, distortion is in various degree caused to image, in conjunction with the optimization of the objective function implementation algorithm related in DE algorithm, invisibility and the robustness of objective function and watermarking algorithm have substantial connection, and its expression formula is as shown in (13) formula.

Maximizef＝NC(W,W*)+NC(I,I*)(13)

Wherein, W and W* represents the watermarking images of watermarking images and extraction respectively, and I, I* represent original image and the image containing watermark respectively.NC (normalizedcorrelation) i.e. normalized correlation coefficient, reflects the similarity of two width images, is also the objective standard evaluating watermark extracting effect.NC value is larger, and the similarity between image is higher, and NC (I, I*) value is larger, and represent that original image is more similar to containing watermarking images, namely the invisibility of watermarking algorithm is better.NC (W, W*) value is larger, and illustrate that watermarking images is more similar to the watermarking images of extraction, namely the robustness of algorithm is stronger.Shown in NC function is defined as follows.

$NC(X,\hat{X})=\frac{{\Σ}_i{\Σ}_{j}X(i,j)\hat{X}(i,j)}{\sqrt{{\Σ}_i{\Σ}_{j}X{(i,j)}^2}\sqrt{{\Σ}_i{\Σ}_j\hat{X}{(i,j)}^2}}---\left(14\right)$

Wherein, represent original image (watermarking images) respectively and contain watermarking images (watermarking images of extraction).Watermark embed process

1) color space conversion is carried out to original image A, rgb format is converted to YIQ form, extract the luminance component Y of image.Utilize wavelet transform to carry out three grades of wavelet decomposition to Y-component, obtain the subband of four different frequencies, be i.e. low frequency ll channel LL and three high frequency detail subgraph LH, HL, a HH.

2) respectively svd is carried out to four subbands obtained in the previous step, try to achieve corresponding singular value S.

Y _k＝U _kS _kV _k ^Tk∈(LL、LH、HL、HH)(15)

3) carry out Arnlod scramble to watermarking images W and obtain the watermarking images W' after scramble, scramble number of times gets τ, and adopts one-level wavelet decomposition to obtain a low frequency ll channel LL and three high frequency detail subgraph LH, HL, HH.

4) matrix corresponding with watermarking images respectively for singular value corresponding for Y-component is carried out multiplicative addition, its embedment strength is q, and this value is obtained by differential evolution algorithm.

S _k+q _kW' _k＝C _kk∈(LL、LH、HL、HH)(16)

5) svd is carried out to Matrix C and obtain matrix U _w, S _w, V _w ^t.

C _k＝U _wkS _wkV _wk ^Tk∈(LL、LH、HL、HH)(17)

6) newly singular value S will be obtained _wwith unitary matrix U and V in step (2) ^tcarry out the inverse decomposition of singular value.

U _kS _wkV _k ^T＝Y _wkk∈(LL、LH、HL、HH)(18)

Recycle wavelet transform afterwards to carry out three grades of wavelet inverse transformations (3-IDWT) the luminance component Y adding image after watermark can be obtained _w'.

7) luminance component Y _w' together with I component in the YIQ form of original image A and Q component combine, and be rgb format by YIQ format conversion, coloured image A with watermarked information can be obtained _w.

Watermark extraction process

If A _w ^*represent the coloured image with watermarked information through attacking, the watermarking images W so extracted ^*then obtained by main following steps.

1) step 1 in watermark embed process is used) and 2) to A _w ^*carry out processing and obtain three corresponding matrix U respectively ^*, S _w ^*, V ^{* T}.

Y _wk ^*＝U _k ^*S _wk ^*V _k ^*Tk∈(LL、LH、HL、HH)(19)

2) by singular value corresponding for original image A luminance component Y respectively with Matrix C after attacking ^*carry out except property subtract each other, factor q remains unchanged, obtain matrix W ' ^*.

W _k' ^*＝(C _k ^*-S _k)/q _kk∈(LL、LH、HL、HH)(20)

3) by matrix W ' ^*carry out one-level wavelet inverse transformation, just extract the W' of the scramble watermarking images after attack ^*.

4) to W' ^*carry out (T-τ) secondary Arnold conversion, T is shuffling cycle, can obtain the watermarking images W extracted ^*.

Experiment simulation is carried out to the inventive method:

The present invention carries out emulation experiment on MatlabR2013a platform, testing original image used is the Lena coloured image of 512 × 512 and the Baboon coloured image of 512 × 512, the required watermarking images embedded is band " Qufu Normal University " printed words, and size is the gray level image of 128 × 128.By the correlation parameter in differential evolution algorithm experimentally effect set, population scale NP is 150, and mutagenic factor F is 0.5, intersect the factor be 0.5, the dimension of problem required by algorithm and D are 4, and the span of embedment strength is [0,1].

In order to the validity of verification algorithm, common image attack process is carried out to containing watermarking images, mainly comprise a. Gaussian noise (GN) b. salt-pepper noise (SPN) c. rotation process (RO) d. shearing manipulation (CR) e. translation (TS) f. squeeze operation (JPEG), various attack pattern and design parameter value select situation as shown in table 1.

The various attack pattern of table 1 and corresponding parameter value selective listing

Invisibility is analyzed

The distortion level caused original image after utilizing Y-PSNR (Peaksignalnoiseratio, PSNR) to evaluate embed watermark, PSNR is larger, and degree of distortion is less, and the invisibility of watermark is better.Generally, according to the visual discrimination rate of people, as PSNR>30, the amendment of human eye to image is invisible, is namely difficult to the difference telling original image and reconstructed image.Y-PSNR function form of Definition is as follows.

$PSNR=10{\log }_{10}\left(\frac{{\left(X_{MAX}\right)}^2}{(1/n\×n){\Σ}_i{\Σ}_j{\left(X\right(i,j)-\hat{X}(i,j\left)\right)}^2}\right)---\left(21\right)$

Wherein, represent original image and reconstructed image respectively, X _mAXfor the pixel maximal value of image X, n is image size.According to the PSNR value that experimental result herein records original image and contains between watermarking images, as shown in table 2.

Table 2 original image and the PSNR value containing watermarking images

The PSNR value that table 2 lists " Lena " and " Baboon " two groups of coloured images are corresponding, can find out, the PSNR value of image is all more than 40, some values then reach about 50, key diagram picture has very high visual quality, prove that this algorithm effectively can ensure the disguise of image embed watermark thus, namely algorithm has good invisibility.

Robust analysis

Robustness embodies the various ability of having a mind to or being not intended to attack of watermaking system opposing, and concerning this algorithm, namely contain watermarking images after attacking, the watermark information of extraction still can be distinguished preferably.For this reason, the 12 kinds of attack patterns related in his-and-hers watches 2, list its correspondence by the watermarking images schematic diagram of attack graph picture and extraction.See intuitively, the watermarking images that image extracts after various attack is more clear, higher with the similarity degree of original watermark image, embodies visual quality preferably, indirectly illustrates that algorithm has good robustness.

In order to further illustrate the robustness of this algorithm, carry out Experimental comparison with the method in list of references [16], compare the NC value of the watermarking images of two kinds of algorithm watermarking images and extraction, specific experiment data trend figure as shown in Figure 5.As can be seen from experimental result, the NC value containing the watermarking images that watermarking images extracts after various attack in this algorithm is larger, all more than 0.97, the value obtained except Gaussian noise is less than except documents, high all than in [16] of NC value corresponding to other several attack patterns.Prove thus, the method optimizing embedment strength value based on differential evolution algorithm manually adjusts intensity level than general watermarking algorithm and carries out watermark embedment and have better Algorithm robustness.

In addition, for verifying the practicality of this algorithm, the six kinds of attack patterns enumerated in his-and-hers watches 1, difference according to its parameter value carries out Algorithm robustness test respectively, no matter be " Lena " image or " Baboon " image, the NC value after Gaussian noise, salt-pepper noise, rotation, shearing, compression and translation all reaches more than 0.9.When compression factor in compression attack is 100%, namely do not process containing watermarking images, the watermarking images of extraction and the NC value of original watermark image are 1, and namely algorithm can extract original watermark image.

The present invention proposes the wavelet field Watermarking of Color based on differential evolution, according to the singularity of different carriers image, adopt the embedment strength of the adaptively selected watermark information of differential evolution algorithm, effectively balance the contradiction between the invisibility of general watermarking algorithm existence and robustness.By the emulation experiment that watermark information scrambling encryption, watermark information embed and extract, result shows that this watermarking algorithm has certain security, and obtain good invisibility, common image processing operations can be resisted, can extract with original watermark image similarity higher containing watermarking images after under attack, the robustness embodying watermarking algorithm is stronger.This algorithm using coloured image as the carrier image of embed watermark information, and utilizes differential evolution adaptive optimization algorithm performance, avoids manually adjustment and embeds the complicacy of the factor, have actual application value widely.

The above is only to better embodiment of the present invention, not any pro forma restriction is done to the present invention, every any simple modification done above embodiment according to technical spirit of the present invention, equivalent variations and modification, all belong in the scope of technical solution of the present invention.

List of references:

[1] Xiang Liling, Liu Hongli. the legal relation of cloud epoch digital resource open access and accountability mechanism [J]. Theory and Application, 2014,37 (7): 28-32.

[2] Zhang Libo, Ma Haiqun, Zhou Lixia. harbour of refuge principle applied research and legislative proposal [J]. knowledge of library and information science, 2013, (01): 122-127.

[3] Lei Juncheng, Huang Tongcheng, Liu little Wen. a kind of copy-right protection method based on Text character extraction [J]. computer science, 2012,39 (10): 104-107

[4] Jiang Li, Xu Zhengquan, Xu Yanyan, etc. the remote sensing image security algorithm [J] of combining cipher technology and labelling technique. Central China University of Science and Technology's journal, 2014,42 (6): 121-126.

[5] Jiang Mingfang. a kind of file and picture multifunctional watermark scheme [J] based on copyright protection and piracy tracking. Library Information Service, 2013,57 (18): 127-131.

[6]PereiraS,VoloshynoskiyS,PunT.Optimaltransformdomainwatermarkembeddingvialinearprogramming.SignalProcessing,2011,91(6):1251-1260.

[7]E.Vahedi,R.A.Zoroofi,M.Shiva,Towardanewwavelet-basedwatermarkingapproachforcolorimagesusingbio-inspiredoptimizationprinciples,DigitalSignalProcessing22(1)(2012)153-162.

[8] Yang Qiwen, Cai Liang, Xue Yuncan. differential evolution algorithm summary [J]. pattern-recognition and artificial intelligence, 2008,21 (4): 506-513.

[9]V.Aslantas.AnoptimalrobustdigitalimagewatermarkingbasedonSVDusingdifferentialevolutionalgorithm,OpticsCommunications282(5)(2009)769-777.

[10] Wang Wenhui, Xu Rongcong. improve SVD image watermark [J] based on differential evolution algorithm. computer utility and software, 2011,28 (7): 277-280.

[11]BaiyingLei,TianfuWang,SipingChen,etal.RobustwatermarkingschemeformedicalImageusingoptimizationmethod,Information,CommunicationsandSignalProcessing(ICICS)(2013)1-4.

[12]R.LakshmiPriya,T.Belji,Dr.V.Sadasivam.Securityofhealthimageryviareversiblewatermarkingbasedondifferentialevolution,m-HealthandEmergingCommunicationSystems(MedCom)(2014)30-34.

[13]MusrratAliandChangWookAhn.AnoptimizedwatermarkingtechniquebasedonDEinDWT-SVDdomain,IEEESymposiumonDifferentialEvolution(2013)99-104.

[14]MusrratAli,ChangWookAhn.Anoptimalwatermarkingtechniquebasedonself-adaptiveDEinDWT-SVDtransformdomain,SignalProcessing94(2014)545-556.

[15]GuangHua,JonathanGoh,Vrizlynn.L.Thing.Time-SpreadEcho-basedaudiowatermarkingwithoptimizedimperceptibilityandrobustness,IEEE/ACMtransactiononaudio,speech,andlanguageprocessing(2015)227-239.

[16]FengmeiLIANG,LijiaWANG,"AnImprovedWavelet-BasedColorImageWatermarkAlgorithm,"JournalofComputationalInformationSystems7,vol.6,pp.2013-2020,2011.

Claims (3)

1., based on the wavelet field Watermarking of Color of differential evolution, it is characterized in that carrying out according to following steps:

Step 1: first carry out color space conversion to original image, converts the YIQ color format of applicable human visual system to and extract light intensity level Y by rgb format;

Step 2: Arnold scramble is adopted to watermarking images;

Step 3: utilize wavelet transform to carry out four different frequency sub-bands of three grades of discrete wavelet transformation generations to the Y-component of original image, obtain sub-band images;

Step 4: represent sub-band images matrix with A, uses svd to the image after three grades of wavelet decomposition;

Step 5: one-level discrete wavelet transformation is carried out to the watermarking images after scramble encryption;

Step 6: watermark embedment: add watermark information in the original image of the form adopting differential evolution adaptive optimization to select embedment strength to be added with multiplicative after svd.

2., according to the wavelet field Watermarking of Color based on differential evolution described in claim 1, it is characterized in that:

In described step 6, the step of watermark embedment is as follows:

1) color space conversion is carried out to original image A, rgb format is converted to YIQ form, extract the luminance component Y of image, wavelet transform is utilized to carry out three grades of wavelet decomposition to Y-component, obtain the subband of four different frequencies, i.e. low frequency ll channel LL and three high frequency detail subgraph LH, HL, a HH;

2) respectively svd is carried out to four subbands obtained in the previous step, try to achieve corresponding singular value S

Y _k＝U _kS _kV _k ^Tk∈(LL、LH、HL、HH)

3) carry out Arnlod scramble to watermarking images W and obtain the watermarking images W' after scramble, scramble number of times gets τ, and adopts one-level wavelet decomposition to obtain a low frequency ll channel LL and three high frequency detail subgraph LH, HL, HH;

4) matrix corresponding with watermarking images respectively for singular value corresponding for Y-component is carried out multiplicative addition, its embedment strength is q, and this value is obtained by differential evolution algorithm;

S _k+q _kW' _k＝C _kk∈(LL、LH、HL、HH)

5) svd is carried out to Matrix C and obtain matrix U _w, S _w, V _w ^t;

C _k＝U _wkS _wkV _wk ^Tk∈(LL、LH、HL、HH)

6) newly singular value S will be obtained _wwith unitary matrix U and V ^tcarry out the inverse decomposition of singular value

U _kS _wkV _k ^T＝Y _wkk∈(LL、LH、HL、HH)

Recycle wavelet transform afterwards to carry out three grades of wavelet inverse transformations and obtain the luminance component Y adding image after watermarks _w';

7) luminance component Y _w' together with I component in the YIQ form of original image A and Q component combine, and be rgb format by YIQ format conversion, obtain coloured image A with watermarked information _w.

3. according to the wavelet field Watermarking of Color based on differential evolution described in claim 1, it is characterized in that: if A _w ^*represent the coloured image with watermarked information through attacking, the watermarking images W so extracted ^*then by main following steps:

1) to use in watermark embed process step to A _w ^*carry out processing and obtain three corresponding matrix U respectively ^*, S _w ^*,

Y _wk ^*＝U _k ^*S _wk ^*V _k ^*Tk∈(LL、LH、HL、HH)

2) by singular value corresponding for original image A luminance component Y respectively with Matrix C after attacking ^*carry out except property subtract each other, factor q remains unchanged, obtain matrix W ' ^*;

W _k' ^*＝(C _k ^*-S _k)/q _kk∈(LL、LH、HL、HH)

3) by matrix W ' ^*carry out one-level wavelet inverse transformation, just extract the W' of the scramble watermarking images after attack ^*;

4) to W' ^*carry out (T-τ) secondary Arnold conversion, T is shuffling cycle, can obtain the watermarking images W extracted ^*.