CN102880998B - The extracting method of watermarking images - Google Patents

Abstract

The present invention relates to and be a kind ofly suitable for solving the transparency in watermarking algorithm and contradict the extracting method of the Gray-level Watermarking image of problem with robustness, it comprises: to initial carrier image and the image block do not covered mutually being divided into corresponding number containing watermarking images to be detected, dct transform is carried out to each image block, finds out the low-frequency data matrix A in each image block according to template and svd carried out to each matrix A and isolates the information in singular value, energy Ratios according to each image block presses the principle of adaptive feed-forward network, recovers Gray-level Watermarking image.

Description

The extracting method of watermarking images

The application is divisional application, the application number of original application: 201110139206.9, the applying date: 2011-5-26, invention and created name: the embedding of watermarking images and extracting method.

Technical field

The present invention relates to the technical field of image procossing, specifically a kind of extracting method of watermarking images.

Background technology

Along with the develop rapidly of Digital object identifier and technique of internet, the safety problem of media product becomes the previous quite important and challenging research topic of order.Digital watermark technology is a kind of potential effective ways realizing copyright protection of digital product; become a study hotspot of information security field at present; also be a kind of important branch of Information hiding research field; digital watermarking is arisen at the historic moment just in this context, and popular direction, the forward position becoming academic circles at present research.

In view of current most of trade mark or copyright marking all represent with gray level image or coloured image, the research therefore how embedding Gray-level Watermarking image has very important practical value, also more meaningful, more realistic demand.Due to the restriction of Information Hiding Capacity, embedding difficulty is high to adopt gray level image to carry out embedding not only as image watermark, and is difficult to realize Blind extracting, while the robustness improving embed watermark, also reduce its not sentience, so the research of this respect is not also a lot.

The transparency and robustness are digital watermark technology two fundamental characteristics.If general embedding is containing the image transparency relatively good (namely PSNR is larger) of watermark, robustness poor (the related coefficient NC of detection is smaller), they are conflicting, the transparency and robustness all with the embedded location of watermark and embedment strength closely bound up.The present invention is exactly that combining image energy spectrometer embeds watermark information strong as far as possible in place in order to solve this technical barrier, improves again and detect related coefficient NC while namely improving Y-PSNR under same embedment strength.

Digital watermarking can be divided into robust watermarking, fragile watermark and semi-fragile watermarking by the effect according to digital watermarking.

(1) fundamental purpose of robust watermarking is the copyright protecting copyright; it requires that the watermark after embedding can stand various conventional signal processing operations; comprise unintentionally or the process of malice, as lossy compression method, filtering, level and smooth, reduce, geometry deformation etc.Robust watermarking is after various process, as long as host image is not corrupted to not spendable degree.All should be able to detect.Therefore the robustness of such watermark requires higher.

(2) fragile watermark is mainly for the protection of the integrality of copyright, the true and false of discriminating digit works.

(3) semi-fragile watermarking requires the useful signal processing operations that can resist to a certain degree, and as JPEG compression etc., than fragile watermark, slightly robust is a little in this watermark.

Summary of the invention

The technical matters that the present invention first will solve is to provide a kind of extracting method of Gray-level Watermarking image.

The technical matters of its secondary solution of the present invention solves the transparency in watermarking algorithm to contradict problem with robustness.

For solving the problems of the technologies described above, the extracting method of watermarking images provided by the invention, comprises the steps:

A () is I (M to matrix, N) initial carrier image and matrix are I ' (M, N) the watermarking images that contains to be checked is divided into 8 × 8 image blocks respectively, and dct transform is carried out to each 8 × 8 image blocks, from each 8 × 8 image blocks, corresponding low frequency and high-frequency data is taken out according to 8 × 8 templates, and the matrix A of 3 × 3 and the matrix B of 6 × 6 are formed respectively to the low frequency in each 8 × 8 image blocks in initial carrier image and high-frequency data, to the low frequency in each 8 × 8 image blocks in watermarking images and high-frequency data form respectively the matrix A of matrix 3 × 3 ' with 6 × 6 matrix B ', to described matrix A, A ' carries out svd respectively: A=USV ^t, A '=U ' S ' V ' ^t,

B () solves the watermark information changing value Δ W in each 8 × 8 image blocks _j=S ' _j(1,1)-S _j(1,1), j=1 Λ m, m are the total block data that original image is divided into described 8 × 8 image blocks;

C (), from m 8 × 8 image blocks, finding out with energy Ratios is corresponding 8 × 8 image blocks of a front n maximal value, namely finds out individual 8 × 8 image blocks of n;

D the energy Ratios of one of the energy Ratios of one of individual 8 × 8 image blocks of the n described in () and 8 × 8 image blocks of described m is designated as g respectively _n(i) and g _m(j), i=1 Λ n, j=1 Λ m; According to:

If g _n(i) ≡ g _m(j), then image intensity value draw n image information, K is embedment strength;

E (), by the energy Ratios size sequential of a described n image information according to 8 × 8 image blocks at each image information place, draws 1 row n row image sequence;

F above-mentioned 1 row n row image sequence is recovered watermarking images according to the inverse operation of Chaotic Scrambling algorithm by ().

Present invention also offers embedding and the extracting method of watermarking images, wherein, the embedding grammar of described watermarking images, comprises the steps:

(1) pre-service of watermarking images: watermarking images is carried out chaos sequence scramble, and the image sequence this watermarking images being deformed into 1 row n row, i.e. n image information;

(2) selection of embedded location: be I (M by matrix by initial carrier image, N) 8 × 8 image blocks (namely length and width are all the rectangular image block of 8 pixels) do not covered mutually of corresponding number are divided into, DCT(discrete cosine transform is carried out to each 8 × 8 image blocks) conversion, low frequency corresponding in each 8 × 8 image blocks and high-frequency data is taken out according to 8 × 8 templates (namely length and width are all the rectangular image template of 8 pixels), and described low frequency and high-frequency data are formed the matrix A of 3 × 3 and the matrix B of 6 × 6 respectively, to matrix A, B carries out svd and asks its matrix norm, solve the energy Ratios g of each 8 × 8 image blocks, energy Ratios wherein: s _highfor the first singular value of described matrix B, s _lowfor the first singular value of described matrix A, finding out with energy Ratios is corresponding 8 × 8 image blocks of a front n maximal value,

(3) data selection of watermark embedment: watermark information and a described n image information, select correspondingly successively to embed adaptively in the first value of the svd of the described matrix A at the n arranged from big to small by described energy Ratios g 8 × 8 image blocks, embedment strength regulates automatically with the energy Ratios of image block;

(4) DCT inverse transformation is carried out to each 8 × 8 image blocks of embed watermark information, obtain image with watermarked information.

When carrying out the data selection of watermark embedment of above-mentioned steps (3): to the method that the matrix A of n 8 × 8 images carries out svd be one by one: A=USV ^t, U ∈ R in formula ^{m × M}, S ∈ R ^{m × N}, V ∈ R ^{n × N}, U and V is orthogonal matrix, and S is diagonal matrix; Take out the first singular value of S diagonal matrix and embed watermark, embedding formula i=1 Λ n; Embedment strength K, with block energy adjustment more automatic than g, embeds adaptively described in realizing, and its formula is K _i=α+g _i, i=1 Λ m, in formula, α is element factor, and its value condition need ensure that embedding image Y-PSNR (PSNR) with watermarked information is greater than 36db; M is the total block data that original image is divided into described 8 × 8 image blocks; New singular value is substituted into A '=US ' V ^t, obtain low-frequency data A ' with watermarked information, and data sent back to the correspondence position in corresponding 8 × 8 image blocks.

The extracting method of described watermarking images, comprises the steps:

A () is I (M to matrix, N) initial carrier image and matrix are I ' (M, N) the watermarking images that contains to be checked is divided into 8 × 8 image blocks respectively, and dct transform is carried out to each 8 × 8 image blocks, from each 8 × 8 image blocks, corresponding low frequency and high-frequency data is taken out according to 8 × 8 templates, and the matrix A of 3 × 3 and the matrix B of 6 × 6 are formed respectively to the low frequency in each 8 × 8 image blocks in initial carrier image and high-frequency data, to the low frequency in each 8 × 8 image blocks in watermarking images and high-frequency data form respectively the matrix A of matrix 3 × 3 ' with 6 × 6 matrix B ', to described matrix A, A ' carries out svd respectively: A=USV ^t, A '=U ' S ' V ' ^t,

B () solves the watermark information changing value Δ W in each 8 × 8 image blocks _j=S ' _j(1,1)-S _j(1,1), j=1 Λ m, m are the total block data that original image is divided into described 8 × 8 image blocks;

C (), from m 8 × 8 image blocks, finding out with energy Ratios is corresponding 8 × 8 image blocks of a front n maximal value, namely finds out individual 8 × 8 image blocks of n;

D the energy Ratios of one of the energy Ratios of one of individual 8 × 8 image blocks of the n described in () and 8 × 8 image blocks of described m is designated as g respectively _n(i) and g _m(j), i=1 Λ n, j=1 Λ m; According to:

If g _n(i) ≡ g _m(j), then image intensity value draw n image information, K is embedment strength;

E (), by the energy Ratios size sequential of a described n image information according to 8 × 8 image blocks at each image information place, draws 1 row n row image sequence;

F above-mentioned 1 row n row image sequence is recovered watermarking images according to the inverse operation of Chaotic Scrambling algorithm by ().

Further, in described step (a): take out the data A(9 of 1 × 9 from the low frequency and high frequency region coefficient of 8 × 8 template-setups), the data B(36 of 1 × 36) array, this array is deformed into square formation A, the A ' of 3 × 3, again SVD decomposition operation is carried out to square formation A, A ', take out the first value in singular value diagonal matrix.

In described step (c): be I (M to initial carrier image array, N), to be checked is I ' (M containing watermarking images, N), be divided into 8 × 8 image blocks do not covered mutually of corresponding number respectively, and carry out dct transform to each piece, from the low frequency and high frequency region coefficient of 8 × 8 template-setups, take out the data A(9 of 1 × 9), the data B(36 of 1 × 36) array, this array is deformed into the square formation A of 3 × 3,6 × 6 square formation B, again SVD decomposition operation is carried out to square formation A, B, determine energy Ratios finding out with energy Ratios is corresponding 8 × 8 image blocks of a front n maximal value.

The present invention has positive effect: the embedding grammar of (1) self-adaptation Gray-level Watermarking of the present invention image, utilizes Chaotic Scrambling to remove the correlativity of original watermark and scramble watermark, improves the security of watermark.By making dct transform to initial carrier image I piecemeal and template partitioned matrix SVD decomposes, matrix norm, computing block energy Ratios, looks for the best embedding position of piecemeal, improves the transparency of watermark.In telescopiny, each piecemeal embedment strength regulates automatically with energy Ratios, substantially increases the robustness of watermark.In view of most of trade mark or copyright information are all represent with gray level image or coloured image, utilize the present invention to embed significant Gray-level Watermarking information in initial carrier image I, there is important practical value, more realistic demand.(2) the present invention makes svd in the DCT domain of 8 × 8 image blocks and represents image energy by its matrix norm, utilizes block image energy spectrometer to find method, determines the piecemeal of applicable embed watermark, improve the invisibility of watermark.(3) embedment strength of the present invention's each piecemeal in telescopiny regulates automatically with the energy Ratios g of piecemeal, greatly strengthen the robustness of watermark.(4) piecemeal is divided into basic, normal, high three parts according to 8 × 8 templates by the present invention, by the first value of watermark embedment 3 × 3 Singular Value Decomposition Using, and the robustness of this watermark also improved.

Accompanying drawing explanation

In order to make content of the present invention be more likely to be clearly understood, below basis specific embodiment and by reference to the accompanying drawings, the present invention is further detailed explanation, wherein

Fig. 1 is the watermarking images in embodiment.

Fig. 2 is 8 × 8 templates.

Fig. 3 is the embedding process flow diagram of watermarking images;

Fig. 4 is the extraction process flow diagram of watermarking images;

Image containing watermark is carried out the result schematic diagram that quality factor is the compression of 90,75,50,30,10 by Fig. 5 respectively;

Fig. 6 be make an uproar to adding respectively containing the original image of watermark, the result schematic diagram of Filtering Attacks experiment;

Fig. 7 is to the result schematic diagram sheared respectively containing the original image of watermark, rotate, convergent-divergent attacks experiment.

Embodiment

(embodiment 1)

See Fig. 1-3, the embedded object of the watermarking images of the present embodiment is a matrix is I (M, N) initial carrier image (M=N=512, namely length and width is respectively the image of 512 pixels), its form is unpressed more than 8 high accuracy grey scale bitmaps (bmp or tiff), also can choose the image of JPG form.Gray-level Watermarking image generally chooses trade mark or copyright marking, and the total n of pixel requires less than carrier image (n<M/8*N/8), and in embodiment, selected pixels size is the Gray-level Watermarking of 40 × 49.

Carrier image embedded location is determined and to embed Gray-level Watermarking information Step as follows:

(1) pre-service of watermarking images.For ensureing the security of watermark, watermarking images is carried out some algorithms (as chaos sequence scramble method etc.) scramble, after scramble, image is designated as W, and image 40 × 49 is deformed into 1 row 1960 row image sequence one-dimension array.

(2) embedded location is selected.By initial carrier image I (512,512) 8 × 8 image blocks do not covered mutually of corresponding number are divided into, amount to m=4096 block, dct transform is carried out to each image block, take out low frequency corresponding in each 8 × 8 image blocks and high-frequency data according to 8 × 8 templates, and described low frequency and high-frequency data are formed respectively the B matrix of the A and 6 × 6 of 3 × 3.

One width digital picture can regard the matrix be made up of many real number items as from linear algebra angle.With A ∈ R ^{m × N}represent an image array, its svd is the product of 3 matrixes.

Namely $U^T\mathrm{AV}=S=\left(\begin{array}{cc}{S}_r& 0\\ 0& 0\end{array}\right)$ Or A=USV ^t(1)

U ∈ R in formula ^{m × M}, S ∈ R ^{m × N}, V ∈ R ^{n × N}, U and V is orthogonal matrix, and S is diagonal matrix.

S _r=diag(s ₁,s ₂,Λ,s _r) （2）

On S diagonal line, element meets s ₁>=s ₂>=Λ>=s _r>s _r+1=Λ=s _n=0, the order of element to be 0, r be matrix A on off-diagonal, formula (1) is exactly the svd of matrix A.

Norm according to matrix norm definition A is ${\left|\right|A\left|\right|}_F={\left(\begin{array}{cc}\underset{i=1}{\overset{M}{\mathrm{\&Sigma;}}}& \underset{j=1}{\overset{N}{\mathrm{\&Sigma;}}}{\left|a_{\mathrm{ij}}\right|}^2\end{array}\right)}^{\frac{1}{2}}={\left(\mathrm{tr}\left(\begin{array}{cc}{A}^T& A\end{array}\right)\right)}^{\frac{1}{2}},$

By A=USV ^tsubstitution above formula obtains ${\left|\right|A\left|\right|}_F={\left(\mathrm{tr}\left(\begin{array}{cc}{A}^T& A\end{array}\right)\right)}^{\frac{1}{2}}=\sqrt{\mathrm{tr}({\left({\mathrm{USV}}^T\right)}^T\&times;\left({\mathrm{USV}}^T\right))}$ $=\sqrt{\mathrm{tr}(\left({\left({\mathrm{SV}}^T\right)}^T{U}^T\right)\&times;\left({\mathrm{USV}}^T\right))}=\sqrt{\mathrm{tr}(\left({\left({\mathrm{SV}}^T\right)}^T{U}^T\right)\&times;\left({\mathrm{USV}}^T\right))}=\sqrt{\mathrm{tr}\left({\mathrm{VS}}^T{U}^T{\mathrm{USV}}^T\right)}$ $=\sqrt{\mathrm{tr}\left({\mathrm{VS}}^T{\mathrm{SV}}^T\right)}=\sqrt{\mathrm{tr}\left(V{\left|\right|S\left|\right|}_F^2{V}^T\right)}=\sqrt{\mathrm{tr}\left({\left|\right|S\left|\right|}_F^2{\mathrm{VV}}^T\right)}={\left|\right|S\left|\right|}_F\sqrt{\mathrm{tr}\left({\mathrm{VV}}^T\right)}$

Because U and V is orthogonal matrix, its inner product is 1, so above formula has again ${\left|\right|A\left|\right|}_F={\left|\right|S\left|\right|}_F=\sqrt{\underset{i=1}{\overset{r}{\mathrm{\&Sigma;}}}{s_i}^2}.$

Represent image energy by image norm, from above-mentioned theory is derived, known image energy is the evolution of singular value quadratic sum, again because that play a decisive role after digital picture svd is s ₁(s ₁much larger than other singular value), that is other singular value is become 0, only retain s ₁the image of rear reconstruct and original image do not have too large degeneration, and this is a feature of digital picture svd.Can be reduced to again accordingly derive according to above-mentioned theory, when the energy Ratios g of calculating 8 × 8 image block, as long as carry out svd to A, B matrix, solve each image block energy Ratios (s _highfor the first singular value of described matrix B, s _lowfirst singular value for described matrix A), finding out with energy Ratios is corresponding 8 × 8 image blocks of front 1960 maximal values, can go out which image block and be suitable for embed watermark information by quick position.

(3) data selection of watermark embedment: determine in which image block embed after, watermarking images information is selected to touch in the first value of the svd of the low frequency range (i.e. matrix A) of plate 8 × 8 to embed adaptively, embedment strength regulates than automatic with Block Energy, object is for the image after embedding has comparatively strong robustness, and can resist certain geometric attack.Concrete implementation step is as follows:

1. svd is carried out to matrix A: A=USV ^t, take out the first singular value embed watermark of S diagonal matrix, embedding formula i=1 Λ n; Embedment strength K is with block energy than automatically regulating, and its formula is K _i=α+g _i, i=1 Λ m, in formula, α is element factor, and its value condition need ensure to embed and be greater than 36db(containing the image PSNR of watermark and be less than 36db human eye and almost can perceive image change), generally get 0.1.G is image block energy Ratios, and m is the sum that original image is divided into 8 × 8 pieces, n=1960, m=4096 in the present embodiment.

2. new singular value is substituted into A '=US ' V ^t, obtain low-frequency data A ' with watermarked information, and send data back in block correspondence position.

(4) DCT inverse transformation is carried out to each 8 × 8 image blocks of the image of embed watermark, obtain image I ' with watermarked information (512,512).

(embodiment 2)

Recovery and the authenticating step of Gray-level Watermarking image are as follows:

A () is I (512 to initial carrier image array, 512), to be checked is I ' 512 containing watermarking images, 512), be divided into 8 × 8 image blocks do not covered mutually of corresponding number respectively, and dct transform is carried out to each image block, from each 8 × 8 image blocks, corresponding low frequency and high-frequency data is taken out according to 8 × 8 templates, and the matrix A of 3 × 3 and the matrix B of 6 × 6 are formed respectively to the low frequency in each 8 × 8 image blocks in initial carrier image and high-frequency data, to form respectively containing the low frequency in each 8 × 8 image blocks of watermarking images and high-frequency data the matrix A of matrix 3 × 3 ' with 6 × 6 matrix B ', to described matrix A, A ' carries out svd respectively: A=USV ^t, A '=U ' S ' V ' ^t,

B () solves the watermark information changing value Δ W in each 8 × 8 image blocks _j=S ' _j(1,1)-S _j(1,1), j=1 Λ m, m are the total block data that original image is divided into described 8 × 8 image blocks;

C initial carrier image array is I (512 by (), 512) 8 × 8 image blocks do not covered mutually of corresponding number are divided into, and carry out dct transform to each piece, the data A (9) of 1 × 9, data B (36) array of 1 × 36 is taken out from the low frequency and high frequency region coefficient of 8 × 8 template-setups, this array is deformed into the square formation A of 3 × 3,6 × 6 square formation B, again SVD decomposition operation is carried out to square formation A, B, determine energy Ratios find out n maximal value corresponding blocks before energy Ratios.

D the energy Ratios of one of the energy Ratios of one of individual 8 × 8 image blocks of the n described in () and 8 × 8 image blocks of described m is designated as g respectively _n(i) and g _m(j), i=1 Λ n, j=1 Λ m.Because of in 8 × 8 image blocks that information insertion n maximal value before with energy Ratios being is corresponding, the recovery of its watermark is carried out according to following formula: $\left\{\begin{array}{cc}\mathrm{message}\left(i\right)=\frac{\mathrm{\&Delta;W}\left(j\right)}{K_j}& g_n\left(i\right)\&equiv;g_m\left(j\right)\\ \mathrm{message}\left(i\right)=0& g_n\left(i\right)\&NotEqual;g_m\left(j\right)\end{array}\right.,$ I=1 Λ n, j=1 Λ m, K in formula _j=α+g _j;

E (), by the energy Ratios from big to small sequential of a described n image information according to 8 × 8 image blocks at each image information place, draws 1 row 1960 row image sequence; This image sequence is deformed into 40 × 49 Gray-level Watermarking images of scramble;

F the inverse operation of () foundation Chaotic Scrambling algorithm goes out the watermarking images of 40 × 49 from 40 × 49 Gray-level Watermarking Postprocessing technique of above-mentioned scramble.

Below simulating, verifying experiment and explanation:

On the block position of all multi-watermark embedding methods is selected (i.e. the process of the watermark transparency), all adopt the mode of computing block variance to find, the present invention have employed Energy Analysis for High in the searching of block position, and experimental result is as table 1:

Intensity	0.1	0.15	0.2	0.25
					PSNR1	42.4793	38.9575	36.4587	34.5205
PSNR2	42.5254	39.0036	36.5048	34.5666
					NC1	0.9809	0.9847	0.9857	0.9864
NC2	0.9922	0.9967	0.9983	0.9995

Table 1

(note: the image Y-PSNR that spatial domain variance searching block method embeds containing watermark is PSNR1, and the watermark related coefficient detected is NC2; Frequency domain energy block finding method is designated as PSNR2 and NC2)

Can find out PSNR2>PSNR1, NC2>NC1 each embedment strength from table 1 data, the embedded location that the embedded location that illustration method 2 is selected all wants ratio method 1 to select in invisibility and robustness is more reasonable, superior.

For the checking robustness of algorithm under other common image process is attacked herein, make an uproar to compressing respectively containing watermark original image, adding, filtering, shearing, rotation etc. attack experiment, in experiment, PSNR value be attack the rear calculated value containing watermarking images and original image.

(1) compression attack.Image containing watermark is carried out the compression that quality factor is 90,75,50,30,10, result Fig. 5 respectively.

Algorithm has very strong resistance to compression herein as can be seen from Figure 5, even if the related coefficient detected when quality factor is 10 high compression is all close to 0.9, watermarking images is also clear and legible.

(2) noise, Filtering Attacks.

Algorithm has comparatively strong immunity to gaussian filtering, 3 rank Butterworth low pass (D0=120) attacks herein as can be seen from Figure 6, comparatively responsive to medium filtering.

(3) shearing, rotation, convergent-divergent are attacked.By carry out containing the original image of watermark rotation 45 °, 90 °, amplify 2 times, reduce the attacks such as half, after this kind of attack, image needs to carry out process and returns to and original image same size, and attack algorithm adopts " bilinear interpolation (bilinear) ".Shearing attack is when detecting watermark, and need to do same shearing at original image and detect, experimental result as shown in Figure 7.As long as the one-piece construction of image is not destroyed as can be seen from Figure 7, its watermarking images can detect, and as half-twist, amplification 2 times, reduces half.If picture structure wrecks, the watermark of detection is just more poor, as rotated 45 °, the attack such as shearing.

Above-described embodiment is only for example of the present invention is clearly described, and is not the restriction to embodiments of the present invention.For those of ordinary skill in the field, can also make other changes in different forms on the basis of the above description.Here exhaustive without the need to also giving all embodiments.And these belong to spirit institute's apparent change of extending out of the present invention or change and are still among protection scope of the present invention.

Claims (1)

1. an extracting method for watermarking images, comprises the steps:

A () is I (M to matrix, N) initial carrier image and matrix are I ' (M, N) the watermarking images that contains to be checked is divided into 8 × 8 image blocks respectively, and dct transform is carried out to each 8 × 8 image blocks, from each 8 × 8 image blocks, corresponding low frequency and high-frequency data is taken out according to 8 × 8 templates, and the matrix A of 3 × 3 and the matrix B of 6 × 6 are formed respectively to the low frequency in each 8 × 8 image blocks in initial carrier image and high-frequency data, to the low frequency in each 8 × 8 image blocks in watermarking images and high-frequency data form respectively the matrix A of matrix 3 × 3 ' with 6 × 6 matrix B ', to described matrix A, A ' carries out svd respectively: A=USV ^t, A '=U ' S ' V ' ^t, image array A A ∈ R ^{m × N}represent, U ∈ R in formula ^{m × M}, S ∈ R ^{m × N}, V ∈ R ^{n × N}, U and V is orthogonal matrix, and S is diagonal matrix,

B () solves the watermark information changing value Δ W in each 8 × 8 image blocks _j=S ' _j(1,1)-S _j(1,1), j=1 ... m, m are the total block data that original image is divided into described 8 × 8 image blocks;

C (), from m 8 × 8 image blocks, finding out with energy Ratios is corresponding 8 × 8 image blocks of a front n maximal value, namely finds out individual 8 × 8 image blocks of n;

D the energy Ratios of one of the energy Ratios of one of individual 8 × 8 image blocks of the n described in () and 8 × 8 image blocks of described m is designated as g respectively _n(i) and g _m(j), i=1 ... n, j=1 ... m; According to:

If g _n(i) ≡ g _m(j), then image intensity value draw n image information, K is embedment strength;

E (), by the energy Ratios size sequential of a described n image information according to 8 × 8 image blocks at each image information place, draws 1 row n row image sequence;

F above-mentioned 1 row n row image sequence is recovered watermarking images according to the inverse operation of Chaotic Scrambling algorithm by ().