CN103793874A - Novel digital watermarking method capable of maintaining fidelity and robustness - Google Patents

Abstract

The invention discloses a novel fidelity and robust digital watermarking method. The algorithm takes an embedding algorithm as the main content. First, the original image is divided into 8x8 sub-blocks that do not cover each other, and then discrete cosine transform is performed on each sub-block; finally, the The goal is to maximize the integrity of the watermark information extracted from the watermarked graphics after being attacked. Under the constraint that the peak signal-to-noise ratio (PSNR) is greater than 25dB, the watermark weight value of each sub-block is optimized, and the optimal solution is obtained as The final watermark embedding scheme; under the premise of ensuring the fidelity of the digital watermark, the present invention uses an evolutionary optimization algorithm to optimize the robustness of the watermark, comprehensively considers the fidelity and robustness, improves the quality of the digital watermark, and the image after embedding the watermark Compared with the original image, it is difficult to detect image degradation, and it has good robustness to attacks such as compression, filtering, scaling, rotation, and shearing.

Description

A kind of novel fidelity robust digital watermark method

Technical field

The invention belongs to digital watermark technology field, relate in particular to a kind of novel fidelity robust digital watermark method.

Background technology

Aspect raising digital watermarking fidelity and robustness, for fidelity, be generally to find vision in raw data by optimized algorithm to destroy minimum watermark embedding scheme at present; For robustness, be also to find suitable frequency band embed watermark by optimized algorithm, make it time under attack, still can not lose watermark information.Fidelity and robustness are two factors of restriction mutually, both can not give up, and need to seek the unification between them, make watermaking system reach optimum under the constraint of this unified point.

Existing digital watermark technology generally improves fidelity by optimized algorithm, but adopt optimized algorithm to improve to consider aspect robustness less.

In addition, choosing of Digital Watermark Scheme parameter is the problem of finding optimum solution or suboptimal solution in a very large solution space, and has interaction relationship consumingly between each factor of solution space.For this complicated optimization problem, common optimization method can not meet the demands.In recent years, a lot of researchists use genetic algorithm to solve the Parametric optimization problem in this watermaking system, have reached good effect.Genetic algorithm is a kind of optimisation technique based on outstanding candidate result is selected and recombinated, but the chain problem that it exists tectonic block to destroy, namely for existing and influence each other while being related between each factor of solution space, genetic manipulation can destroy the good pattern having found, thereby can not obtain the optimum results of global optimum.

Summary of the invention

The object of the present invention is to provide a kind of novel fidelity robust digital watermark method, be intended to solve existing digital watermark technology and generally improve fidelity by optimized algorithm, but consider less problem adopting optimized algorithm to improve aspect robustness, guaranteeing under digital watermarking fidelity precondition, utilize evolutionary optimization algorithm to be optimized watermark robustness, consider fidelity and robustness, improve the quality of digital watermarking.

The present invention is achieved in that a kind of novel fidelity robust digital watermark method is to embed algorithm as main contents, and detailed process is as follows:

Step 1, to cut apart original image be 8 × 8 sub-blocks that do not cover mutually; To the image of a width N*N, carry out 8*8 and divide block operations, obtain the individual fritter of (N/8) * (N/8);

Step 2, each sub-block is carried out to discrete cosine transform;

Discrete cosine transform formula is as follows:

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

Wherein, f (x, y) is the original image of size for N*N,

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

Step 3, turn to target with the watermark information integrity degree maximum extracting in the moisture impression shape under fire, be greater than under 25dB constraint with Y-PSNR (PSNR), watermark weighted value to each sub-block is optimized, and the optimum solution obtaining embeds scheme as final watermark;

Concrete grammar is as follows:

Formula (2) is watermark embedding criterion, wherein W={w (i) } be watermarking images, X={x ₀(i) } be original image, X _w={ x _w(i) } for containing watermarking images, H={h (i) } be watermark weighted value;

x _w(i)=x ₀(i)×(1+h(i)W(i)) (2)

Formula (3) represents watermark information integrity degree, and wherein W represents original watermark, and We represents the watermark extracting, and Nw is watermark size;

$f_{\mathrm{Int}}(W,W_e)=\frac{{\mathrm{\Σ}}_{x,y}|W(x,y)-W_e(x,y)|}{N_w}---\left(3\right)$

Formula (4) represents Y-PSNR, and wherein I represents original image, and Iw represents that N is original image size containing watermarking images;

$f_{\mathrm{PSNR}}(I,I_w)=10\×{\log }_{10}\left(\frac{N^2\×\max I^2(x,y)}{{\mathrm{\Σ}}_{x,y}[I(x,y)-I_w(x,y){]}^2}\right)---\left(4\right)$

The Optimized model that watermark embeds scheme is:

Optimized variable: watermark weighted value H

Objective function: f _int(W, W _e)

Constraint condition: s.t.f _pSNR(I, I _w) > 25.

Further, the algorithm model of described a kind of novel fidelity robust digital watermark method is:

x _w(i)＝x ₀(i)×(1+h(i)W(i)) (1)

$f_{\mathrm{Int}}(W,W_e)=\frac{{\mathrm{\Σ}}_{x,y}|W(x,y)-W_e(x,y)|}{N_w}---\left(2\right)$

$f_{\mathrm{PSNR}}(I,I_w)=10\×{\log }_{10}\left(\frac{N^2\×\max I^2(x,y)}{{\mathrm{\Σ}}_{x,y}[I(x,y)-I_w(x,y){]}^2}\right)---\left(3\right)$

Formula (1) is watermark embedding criterion, wherein W={w (i) } be watermarking images, X={x ₀(i) } be original image, X _w={ xw (i) } is for containing watermarking images, H={h (i) } be watermark weighted value.

Formula (2) represents watermark information integrity degree, and wherein W represents original watermark, and We represents the watermark extracting, and Nw is watermark size.

Formula (3) represents Y-PSNR, and wherein I represents original image, and Iw represents that N is original image size containing watermarking images.

The Optimized model that watermark embeds scheme is:

Optimized variable: watermark weighted value H

Objective function: f _int(W, W _e)

Constraint condition: s.t.f _pSNR(I, I _w) > 25.

Further, described novel fidelity robust digital watermark method is used the Estimation of Distribution Algorithm (PFEDA based on sequential important sampling particle filter and Cholesky decomposition ₂) Optimized model is optimized, comprise the sampling algorithm that sample distribution is estimated, correlation of variables is not considered in neighborhood sampling, sampling algorithm, the PFEDA that correlation of variables is considered in field sampling ₂.

Further, the concrete steps that described sample distribution is estimated are as follows:

Step 1, preferably concentrate each sample as moment of particle filter t=0, i.e. the initial prior distribution particle obtaining of sampling, and individual particle weights is set to 1/m,

$w_0^i=1/m,i=\mathrm{1,2},...,m;$

Step 2, t=1, with formula (4) renewal

i=1,2,3 ..., m, utilizes formula (5) normalized weight,

for the probability distribution of sample obedience.

$w_t^i=w_{t-1}^{i}p\left(y_t\right|x_t^i)---\left(4\right)$

$w_t^i=\frac{w_t^i}{{\mathrm{\Σ}}_{i=1}^N{w}_t^i}---\left(5\right)$

Further, the sampling of described neighborhood does not consider that the concrete steps of sampling algorithm of correlation of variables are as follows:

In step 1, the distribution that represents at cum rights particle, put back to use roulette method according to particle weight sampling determine a particle;

Step 2, the selected each component value of particle are as average μ _i, utilize formula (6) to calculate the σ that each component is corresponding _i;

Step 3, from N(μ _i, σ _i), i=1,2 ..., d each component of sampling;

Step 4, a combination component obtain a sample;

Step 5, algorithm finish.

${\widetilde{\mathrm{\σ}}}_i=\frac{b-a}{6}{w}_i---\left(6\right)$

Further, the concrete steps of the sampling algorithm of described field sampling consideration correlation of variables are as follows:

In step 1, the distribution that represents at cum rights particle, put back to use roulette method according to particle weight sampling determine a particle;

Step 2, utilize formula (7) calculate k, go Selected Particles as mean vector μ;

$k=\frac{b-a}{6\·\max \left({\mathrm{\σ}}_i\right)}{w}_i---\left(7\right)$

The covariance matrix ∑ of step 3, calculating selected works, if ∑ is null matrix, uses neighborhood sampling not consider that the sampling algorithm of correlation of variables produces sample, and goes to step seven, otherwise continue to carry out step below;

Step 4, use Cholesky decomposition method are obtained A, make k ∑=AA ^t;

Step 5, from N(0,1) sampling d separate standards normal variate, y ₁, y ₂...., y _d;

The each component of sample that step 6, sampling obtain is

(j=1,2 ..., d), d is dimension;

Step 7, algorithm finishes.

Further, the concrete steps of described PFEDA are as follows:

Step 1, initialization first generation colony;

Step 2, to i=1,2 ..., n _p, sampling

wherein p(x) for being uniformly distributed, n _pfor group size;

Step 3, calculating sample adaptive value;

Step 4, block the preferred sample set that selected population adaptive value is good;

Step 5, the probability distribution that uses sample distribution estimation preferably to collect;

Step 6, the historical optimum of reservation, as a sample for population of future generation;

The sampling algorithm sampling n of correlation of variables is considered in step 7, the sampling of execution field _p-1inferiorly obtain population of future generation;

A sample adaptive value in step 8, calculating population;

If step 9, algorithm reach termination condition, finish algorithm, otherwise, go to step four.

effect gathers

The present invention is guaranteeing under digital watermarking fidelity precondition, utilize evolutionary optimization algorithm to be optimized watermark robustness, consider fidelity and robustness, improve the quality of digital watermarking, image after embed watermark is difficult to discover image deterioration compared with original image, and to compression, filtering, the attacks such as convergent-divergent rotational shear have good robustness.

Accompanying drawing explanation

Fig. 1 is the process flow diagram of the embedding algorithm that provides of the embodiment of the present invention.

Embodiment

In order to make object of the present invention, technical scheme and advantage clearer, below in conjunction with drawings and Examples, the present invention is further elaborated.Should be appreciated that specific embodiment described herein, only in order to explain the present invention, is not intended to limit the present invention.

As shown in Figure 1, the present invention is achieved in that a kind of novel fidelity robust digital watermark method is to embed algorithm as main contents, and detailed process is as follows:

Step 1: cutting apart original image is 8 × 8 sub-blocks that do not cover mutually; To the image of a width N*N, carry out 8*8 and divide block operations, obtain the individual fritter of (N/8) * (N/8);

Step 2: each sub-block is carried out to discrete cosine transform;

Discrete cosine transform formula is as follows:

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

Wherein, f (x, y) is the original image of size for N*N,

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

Step 3: turn to target with the watermark information integrity degree maximum extracting in the moisture impression shape under fire, be greater than under 25dB constraint with Y-PSNR (PSNR), watermark weighted value to each sub-block is optimized, and the optimum solution obtaining embeds scheme as final watermark;

Concrete grammar is as follows:

Formula (2) is watermark embedding criterion, wherein W={w (i) } be watermarking images, X={x ₀(i) } be original image, X _w={ x _w(i) } for containing watermarking images, H={h (i) } be watermark weighted value;

x _w(i)＝x ₀(i)×(1+h(i)W(i)) (2)

Formula (3) represents watermark information integrity degree, and wherein W represents original watermark, and We represents the watermark extracting, and Nw is watermark size;

$f_{\mathrm{Int}}(W,W_e)=\frac{{\mathrm{\Σ}}_{x,y}|W(x,y)-W_e(x,y)|}{N_w}---\left(3\right)$

Formula (4) represents Y-PSNR, and wherein I represents original image, and Iw represents that N is original image size containing watermarking images;

$f_{\mathrm{PSNR}}(I,I_w)=10\×{\log }_{10}\left(\frac{N^2\×\max I^2(x,y)}{{\mathrm{\Σ}}_{x,y}[I(x,y)-I_w(x,y){]}^2}\right)---\left(4\right)$

The Optimized model that watermark embeds scheme is:

Optimized variable: watermark weighted value H

Objective function: f _int(W, W _e)

Constraint condition: s.t.f _pSNR(I, I _w) > 25.

Further, the algorithm model of described a kind of novel fidelity robust digital watermark method is:

x _w(i)＝x ₀(i)×(1+h(i)W(i)) (1)

$f_{\mathrm{Int}}(W,W_e)=\frac{{\mathrm{\Σ}}_{x,y}|W(x,y)-W_e(x,y)|}{N_w}---\left(2\right)$

$f_{\mathrm{PSNR}}(I,I_w)=10\×{\log }_{10}\left(\frac{N^2\×\max I^2(x,y)}{{\mathrm{\Σ}}_{x,y}[I(x,y)-I_w(x,y){]}^2}\right)---\left(3\right)$

Formula (1) is watermark embedding criterion, wherein W={W (i) } be watermarking images, X={x ₀(i) } be original image, X _w={ x _w(i) } for containing watermarking images, H={h (i) } be watermark weighted value.

Formula (2) represents watermark information integrity degree, and wherein W represents original watermark, and We represents the watermark extracting, and Nw is watermark size.

Formula (3) represents Y-PSNR, and wherein I represents original image, and Iw represents that N is original image size containing watermarking images.

Further, described novel fidelity robust digital watermark method is used the Estimation of Distribution Algorithm (PFEDA based on sequential important sampling particle filter and Cholesky decomposition ₂) Optimized model is optimized, comprise the sampling algorithm that sample distribution is estimated, correlation of variables is not considered in neighborhood sampling, sampling algorithm, the PFEDA that correlation of variables is considered in field sampling ₂;

The Optimized model that watermark embeds scheme is:

Optimized variable: watermark weighted value H

Objective function: f _int(W, W _e)

Constraint condition: s.t.f _pSNR(I, I _w) > 25.

Further, the concrete steps that described sample distribution is estimated are as follows:

Step 1, preferably concentrate each sample as moment of particle filter t=0, i.e. the initial prior distribution particle obtaining of sampling, and individual particle weights is set to 1/m,

$w_0^i=1/m,i=\mathrm{1,2},...,m;$

Step 2, t=1, with formula (4) renewal

i=1,2,3 ..., m, utilizes formula (5) normalized weight,

for the probability distribution of sample obedience.

$w_t^i=w_{t-1}^{i}p\left(y_t\right|x_t^i)---\left(4\right)$

$w_t^i=\frac{w_t^i}{{\mathrm{\Σ}}_{i=1}^N{w}_t^i}---\left(5\right)$

Further, the sampling of described neighborhood does not consider that the concrete steps of sampling algorithm of correlation of variables are as follows:

In step 1, the distribution that represents at cum rights particle, put back to use roulette method according to particle weight sampling determine a particle;

Step 2, the selected each component value of particle are as average μ _i, utilize formula (6) to calculate the σ that each component is corresponding _i;

Step 3, from N(μ _i, σ _i), i=1,2 ..., d each component of sampling;

Step 4, a combination component obtain a sample;

Step 5, algorithm finish.

${\widetilde{\mathrm{\σ}}}_i=\frac{b-a}{6}{w}_i---\left(6\right)$

Further, the concrete steps of the sampling algorithm of described field sampling consideration correlation of variables are as follows:

In step 1, the distribution that represents at cum rights particle, put back to use roulette method according to particle weight sampling determine a particle;

Step 2, utilize formula (7) calculate k, go Selected Particles as mean vector μ;

$k=\frac{b-a}{6\·\max \left({\mathrm{\σ}}_i\right)}{w}_i---\left(7\right)$

The covariance matrix ∑ of step 3, calculating selected works, if ∑ is null matrix, uses neighborhood sampling not consider that the sampling algorithm of correlation of variables produces sample, and goes to step seven, otherwise continue to carry out step below;

Step 4, use Cholesky decomposition method are obtained A, make k ∑=AA ^t;

Step 5, from N(0,1) sampling d separate standards normal variate, y ₁, y ₂...., y _d;

The each component of sample that step 6, sampling obtain is

(j=1,2 ..., d), d is dimension;

Step 7, algorithm finishes.

Further, the concrete steps of described PFEDA are as follows:

Step 1, initialization first generation colony;

Step 2, to i=1,2 ..., n _p, sampling

wherein p(x) for being uniformly distributed, n _pfor group size;

Step 3, calculating sample adaptive value;

Step 4, block the preferred sample set that selected population adaptive value is good;

Step 5, the probability distribution that uses sample distribution estimation preferably to collect;

Step 6, the historical optimum of reservation, as a sample for population of future generation;

The sampling algorithm sampling n of correlation of variables is considered in step 7, the sampling of execution field _p-1inferiorly obtain population of future generation;

A sample adaptive value in step 8, calculating population;

If step 9, algorithm reach termination condition, finish algorithm, otherwise, go to step four.

By reference to the accompanying drawings the specific embodiment of the present invention is described although above-mentioned; but not limiting the scope of the invention; one of ordinary skill in the art should be understood that; on the basis of technical scheme of the present invention, those skilled in the art do not need to pay various modifications that performing creative labour can make or distortion still within protection scope of the present invention.

Claims (7)

1. a novel fidelity robust digital watermark method, is characterized in that, described novel fidelity robust digital watermark method is to embed algorithm as main contents, and detailed process is as follows:

Step 1, to cut apart original image be 8 × 8 sub-blocks that do not cover mutually; To the image of a width N*N, carry out 8*8 and divide block operations, obtain the individual fritter of (N/8) * (N/8);

Step 2, each sub-block is carried out to discrete cosine transform;

Discrete cosine transform formula is as follows:

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

Wherein, f (x, y) is the original image of size for N*N,

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

Step 3, turn to target with the watermark information integrity degree maximum extracting in the moisture impression shape under fire, be greater than under 25dB constraint at Y-PSNR (PSNR), watermark weighted value to each sub-block is optimized, and the optimum solution obtaining embeds scheme as final watermark;

Concrete grammar is as follows:

Formula (2) is watermark embedding criterion, wherein W={w (i) } be watermarking images, X={x ₀(i) } be original image, X _w={ x _w(i) } for containing watermarking images, H={h (i) } be watermark weighted value;

x _w(i)＝x ₀(i)×(1+h(i)W(i)) (2)

Formula (3) represents watermark information integrity degree, and wherein W represents original watermark, and We represents the watermark extracting, and Nw is watermark size;

$f_{\mathrm{Int}}(W,W_e)=\frac{{\mathrm{\Σ}}_{x,y}|W(x,y)-W_e(x,y)|}{N_w}---\left(3\right)$

Formula (4) represents Y-PSNR, and wherein I represents original image, and Iw represents that N is original image size containing watermarking images;

$f_{\mathrm{PSNR}}(I,I_w)=10\×{\log }_{10}\left(\frac{N^2\×\max I^2(x,y)}{{\mathrm{\Σ}}_{x,y}[I(x,y)-I_w(x,y){]}^2}\right)---\left(4\right)$

The Optimized model that watermark embeds scheme is:

Optimized variable: watermark weighted value H

Objective function: f _int(W, W _e)

Constraint condition: s.t.f _pSNR(I, I _w) > 25.

2. novel fidelity robust digital watermark method as claimed in claim 1, is characterized in that, the algorithm model of described a kind of novel fidelity robust digital watermark method is:

x _w(i)＝x ₀(i)×(1+h(i)W(i)) (5)

$f_{\mathrm{Int}}(W,W_e)=\frac{{\mathrm{\Σ}}_{x,y}|W(x,y)-W_e(x,y)|}{N_w}---\left(6\right)$

$f_{\mathrm{PSNR}}(I,I_w)=10\×{\log }_{10}\left(\frac{N^2\×\max I^2(x,y)}{{\mathrm{\Σ}}_{x,y}[I(x,y)-I_w(x,y){]}^2}\right)---\left(7\right)$

Formula (5) is watermark embedding criterion, wherein W={w (i) } be watermarking images, X={x ₀(i) } be original image, X _w={ x _w(i) } for containing watermarking images, H={h (i) } be watermark weighted value;

Formula (6) represents watermark information integrity degree, and wherein W represents original watermark, and We represents the watermark extracting, and Nw is watermark size;

Formula (7) represents Y-PSNR, and wherein I represents original image, and Iw represents that N is original image size containing watermarking images;

The Optimized model that watermark embeds scheme is:

Optimized variable: watermark weighted value H

Objective function: f _int(W, W _e)

Constraint condition: s.t.f _pSNR(I, I _w) > 25.

3. novel fidelity robust digital watermark method as claimed in claim 1, is characterized in that, described novel fidelity robust digital watermark method is used the Estimation of Distribution Algorithm (PFEDA based on sequential important sampling particle filter and Cholesky decomposition ₂) Optimized model is optimized, comprise the sampling algorithm that sample distribution is estimated, correlation of variables is not considered in neighborhood sampling, sampling algorithm, the PFEDA that correlation of variables is considered in field sampling ₂.

4. novel fidelity robust digital watermark method as claimed in claim 3, is characterized in that, the concrete steps that described sample distribution is estimated are as follows:

Step 1, preferably concentrate each sample as moment of particle filter t=0, i.e. the initial prior distribution particle obtaining of sampling, and individual particle weights is set to 1/m,

$w_0^i=1/m,i=\mathrm{1,2},...,m;$

Step 2, t=1, with formula (4) renewal

i=1,2,3 ..., m, utilizes formula (5) normalized weight, for the probability distribution of sample obedience;

$w_t^i=w_{t-1}^{i}p\left(y_t\right|x_t^i)---\left(8\right)$

$w_t^i=\frac{w_t^i}{{\mathrm{\Σ}}_{i=1}^N{w}_t^i}---\left(9\right).$

5. novel fidelity robust digital watermark method as claimed in claim 3, is characterized in that, the sampling of described neighborhood does not consider that the concrete steps of sampling algorithm of correlation of variables are as follows:

In step 1, the distribution that represents at cum rights particle, put back to use roulette method according to particle weight sampling determine a particle;

Step 2, the selected each component value of particle, as average μ i, utilize formula (10) to calculate the σ that each component is corresponding _i;

Step 3, from N(μ _i, σ _i), i=1,2 ..., d each component of sampling;

Step 4, a combination component obtain a sample;

Step 5, algorithm finish;

${\widetilde{\mathrm{\σ}}}_i=\frac{b-a}{6}{w}_i---\left(10\right).$

6. novel fidelity robust digital watermark method as claimed in claim 3, is characterized in that, the concrete steps of the sampling algorithm of described field sampling consideration correlation of variables are as follows:

In step 1, the distribution that represents at cum rights particle, put back to use roulette method according to particle weight sampling determine a particle;

Step 2, utilize formula (11) calculate k, go Selected Particles as mean vector μ;

$k=\frac{b-a}{6\·\max \left({\mathrm{\σ}}_i\right)}{w}_i---\left(11\right)$

The covariance matrix ∑ of step 3, calculating selected works, if ∑ is null matrix, uses neighborhood sampling not consider that the sampling algorithm of correlation of variables produces sample, and goes to step seven, otherwise continue to carry out step below;

Step 4, use Cholesky decomposition method are obtained A, make k ∑=AA ^t;

Step 5, from N(0,1) sampling d separate standards normal variate, y ₁, y ₂...., y _d;

The each component of sample that step 6, sampling obtain is

(j=1,2 ..., d), d is dimension;

Step 7, algorithm finishes.

7. novel fidelity robust digital watermark method as claimed in claim 3, is characterized in that, the concrete steps of described PFEDA are as follows:

Step 1, initialization first generation colony;

Step 2, to i=1,2 ..., n _p, sampling wherein p(x) for being uniformly distributed, n _pfor group size;

Step 3, calculating sample adaptive value;

Step 4, block the preferred sample set that selected population adaptive value is good;

Step 5, the probability distribution that uses sample distribution estimation preferably to collect;

Step 6, the historical optimum of reservation, as a sample for population of future generation;

The sampling algorithm sampling n of correlation of variables is considered in step 7, the sampling of execution field _p-1inferiorly obtain population of future generation;

A sample adaptive value in step 8, calculating population;

If step 9, algorithm reach termination condition, finish algorithm, otherwise, go to step four.

Images