CN100517381C - Methof for embedding and extracting watermark based on statistical model of coefficient in transform domain of digital images - Google Patents

Abstract

A method for embedding and picking up digital watermark based on digital image transform coefficient statistic model includes making desired value of image sub-block DCT transform be positive or be negative by changing distribution of alternating-current coefficient on image sub-block DCT transform, enabling to embed and to pick up digital watermark before and after printing and scanning as said property can be remained still after image is printed and scanned.

Description

Watermark embedding and extracting method based on the coefficient in transform domain of digital images statistical model

Technical field

The invention belongs to Information hiding and digital watermark technology field, be specifically related to a kind of digital watermark method, relate in particular to a kind of digital watermarking embedding and extracting method based on the coefficient in transform domain of digital images statistical model based on statistical model.

Background technology

The develop rapidly of digitalized network technology has proposed new higher requirement to multimedia copyright protection and content discriminating, and digital watermark technology is answered this requirement and progressively grown up.So-called digital watermarking (Digital Watermarking) is exactly to utilize the insensitivity of the human vision or the sense of hearing to add some redundant information in the multi-medium data (as digital signals such as image, audio frequency, videos) reaching the purpose of Information hiding, thereby can carry out intellectual property protection to copyright notice, permission to use condition etc.The motive power of development digital watermark technology is for multi-medium data provides copyright protection, but in fact digital watermarking also has very important application in the confidential corespondence of the implicit mark of the real and fake discrimination of electronic data, digital product and network.

Great amount of images also exists with printing form simultaneously, as trade mark, the packing of product, books and periodicals, certificate, passport or the like, also has the document copyright protection that exists with printing form certainly.Print and scan and become the universal way that present image duplicates and propagates.Along with the fast development of digitizing technique, image changed becoming very easy between electronic digit form and print format, embedding and the extraction algorithm therefore designed and Implemented based on the digital watermarking of printed matter are necessary.At present in the ascendant and obtained a lot of achievements about the research of digital figure watermark algorithm, but these algorithms mainly are at the copyright in the computer network, and most of algorithms can not be resisted attacks of print_scan.This is because through print scanned process, promptly though the image after twice conversion looks very similar with former figure, in fact very big variation has taken place the pixel value of image D-A (digital signal-simulating signal) with A-D (simulating signal-digital signal).This just requires digital watermarking algorithm to have very strong robustness, could resist format conversion twice.Therefore, seek constant numerical characteristic in printing and scanning process, design has very strong robustness algorithm, becomes a research difficult point of digital watermark technology.

Digital watermarking can be divided into spatial domain watermark and transform domain watermark by embedding grammar.The spatial domain method is hidden watermark by the spatial domain characteristic that changes carrier information; The transform domain method is hidden watermark by some coefficients that change the data conversion territory.Two typical algorithm at the spatial domain embed watermark are least significant bit (LSB) algorithm (LSB:least significant bits) and Patchwork algorithm.The LSB algorithm is that people such as R.van Schyndel are at paper " A Digital Watermark " (R.van Schyndel, A.Z.Tirkel, C.F.Osborne, 1st IEEE International Conference on Image Processing, Austin Texas USA, 1994, Vol.II, pp.86-90) propose in, authors propose watermark information is according to pixels put the lowest order that is inserted into the original image pixels value one by one in this piece paper, and this can guarantee that the watermark that embeds is sightless.But the poor robustness of this algorithm, watermark information are easy to be low pass filtering or basic image manipulation such as lossy compression method destroys.The Patchwork algorithm is that people such as W.Bender are at paper " Techniques for DataHiding " (W.Bender, D.Gruhl, N.Morimoto, A.Lu, IBM Systems Journal, 1996,35 (3﹠amp; 4): what propose 313-336) is a kind of based on the watermarking algorithm that changes the view data statistical property.This algorithm is selected the same number of two pixel set A and B at first at random, increases the pixel value of each pixel in the set A then, reduces the pixel value of each pixel in the set B simultaneously; The average pixel value of set A mid point and the average pixel value of set B mid point can be compared during watermark detection, if the former thinks that more then watermark exists.The Patchwork algorithm has certain resistibility to lossy compression method, filtering and image cutting, but the watermark of this method embeds finite capacity, can only embed the information of a bit.

Watermarking algorithm based on transform domain can embed a large amount of Bit datas and can not cause visual perceptibility, and this class algorithm often takes the technology of similar spread spectrum to hide digital watermark information.This class technology is generally based on image transformation commonly used, such as, comprise discrete cosine transform (DCT:Discrete Cosine Transform), wavelet transform (DWT:Discrete WaveletTransform), discrete Fourier transformation (DFT:Discrete Fourier Transform) and Hadamard transform (Hadamard transform) or the like based on the orthogonal transformation of block image.Wherein an algorithm of tool typical meaning is that people such as I.J.Cox are at paper " Secure Spread Spectrum Watermarking for Multimedia " (I.J.Cox, J.Kilian, T.Leighton and T.Shamoon, IEEE Trans.on Image Processing, 6,12,1673-1687,1997) the spread-spectrum watermark algorithm that proposes in based on dct transform, this algorithm is embedded into watermark sequence (this sequence is the random series that meets normal distribution) in several maximum conversion coefficients except that DC component of integral image dct transform and goes according to additivity principle or the property taken advantage of principle, utilize original image during watermark detection and in the image that obtains, extract a watermark sequence, calculate the degree of correlation of the watermark sequence extract and real watermark information then and utilize threshold value to judge whether contain watermark in the image according to watermark embedding method.This algorithm is realized simple and is had stronger robustness, can resist general image processing operations, but this algorithm when watermark detection, need original image participate in from rather than a kind of blind watermarking algorithm.From then on Zhi Hou digital watermarking algorithm is mostly based on the spread spectrum of transform domain.

By to the discovering of above existing method, for the image after print scanned, the pixel value of its single-point has produced very big variation, and algorithm itself can not correctly extract watermark to the image after print scanned.So head it off is exactly to seek a kind of feature at all, this feature can also keep after having experienced print scanned process well, by to it modification and cause and a kind ofly can recognized patterns hide watermark information; For the needs of practical application, watermarking algorithm should be accomplished blind extraction simultaneously.Above-mentioned digital watermark method all can not satisfy these 2 requirements fully.

Summary of the invention

The objective of the invention is the defective at prior art, propose a kind of new method that embeds and extract based on the digital watermarking of image transformation domain coefficient statistical model, this new watermarking algorithm has high robust, can resist attacks of print_scan.Image still can adopt method of the present invention therefrom to extract embedded watermark information after through D-A and twice conversion of A-D.

For reaching above purpose, a kind of in digital picture the method for embed watermark, may further comprise the steps:

(1) does two-dimensional dct transform with the digital picture piecemeal and to each image block;

(2) generate watermark signal according to the watermark information that will embed;

(3), the medium and low frequency section coefficient of image block dct transform domain is divided into groups according to the length of the watermark signal that obtains by step (2);

(4) analyze the information complexity and the bright-dark degree of each image block, calculate the watermark embed strength of each image block dct transform domain medium and low frequency channel;

(5) choose whole watermark strength, and, come the embed watermark signal by the expectation value that changes the DCT coefficient in each channel packet that obtains by step (3) according to watermark strength by described each channel of step (4);

(6) will carry out the image that the two-dimensional dct inverse transformation just obtains embed watermark respectively through all images piece of above-mentioned steps.

Preferably, in the step (2), described watermark signal is to carry out the watermark bit stream string to be embedded that obtains after the Error Correction of Coding.

Further, described watermark bit stream string is that (Hocquenghem) mode of Error Correction of Coding generates for Bose, Ray-Chaudhuri by watermark signal being carried out BCH.

Preferably, the length of the watermark bit stream string described in the step (2) is identical with the number of the medium and low frequency of image block dct transform domain described in (3) section coefficient grouping.

Further, the number of channel is identical or identical as much as possible in described each medium and low frequency section coefficient grouping.

Preferably, in the step (4), the information complexity of described each image block is the information entropy e of image block ^k, information entropy e ^kBe to calculate by the following formula that quantizes based on gray-scale value:

$e^k={\mathrm{\Σ}}_{i=0}^{15}{P}_i^k\log P_i^k,---\left(\mathrm{EQ}1\right)$

$P_i^k=\frac{1}{N^2}\#\{(n,m):0\&le;n,m\&le;N-1;16i\&le;x_{n,m}^k<16(i+1)\}$

Wherein, the size of each image block is N * N, x _{N, m} ^kThe gray-scale value of presentation video, k are the labels of image block, and (n m) is the label of position in the image block;

Preferably, in the step (4), the f of bright-dark degree of described each image block ^kBe to calculate by following formula:

$f^k=\left\{\begin{array}{ccc}\frac{{\stackrel{\&OverBar;}{x}}^k-(255-T^{+})}{T^{+}}& \mathrm{if}& {\stackrel{\&OverBar;}{x}}^k>255-T^{+}\\ \frac{T^{-}-{\stackrel{\&OverBar;}{x}}^k}{T^{-}}& \mathrm{if}& {\stackrel{\&OverBar;}{x}}^k<T^{-}\\ 0& \mathrm{if}& T^{-}\&le;{\stackrel{\&OverBar;}{x}}^k\&le;255-T^{+}\end{array}\right.---\left(\mathrm{EQ}2\right)$

Wherein, x ^kThe expression label is the average gray value of the image block of k, T ⁺And T ^-Be two pre-set threshold values, practical experience shows that we can get T ⁺=T ^-=50;

Preferably, the watermark embed strength of each channel of each the image block dct transform described in the step (4) is the information entropy e by each image block ^kWith the f of bright-dark degree ^kCalculate according to following formula:

$a_{u,v}^k=(1-e^k)(1+f^k)\frac{\sqrt{u+1}\sqrt{v+1}}{(u+1)+(v+1)}---\left(\mathrm{EQ}3\right)$

Preferably, the watermark embed process described in the step (5), the watermark strength of each channel that obtains according to the whole watermark strength of choosing with by (4) use additivity watermark embedded mode to revise the expectation value of the coefficient in each channel packet.

Preferably, the watermark embed process described in the step (5) when the embed watermark signal is 1 in this grouping, then increases coefficient expectation value that each coefficient in this grouping makes this grouping for just; Otherwise when the embed watermark signal was 0 in this grouping, each coefficient that then reduces in this grouping made the coefficient expectation value of this grouping for negative.

A kind of method of extracting watermark information from digital picture may further comprise the steps:

(1) obtains digital picture and its piecemeal carried out two-dimensional dct transform to each image block then;

(2) expectation value of calculating coefficient in each channel packet obtain in other words all coefficients in this channel packet and, if expectation value is a canonical think with this corresponding watermark signal that divides into groups be 1, otherwise think with this corresponding watermark signal that divides into groups be 0;

(3) the above-mentioned string of binary characters decoding that obtains is obtained watermark information.

Effect of the present invention is: adopt method of the present invention, can successfully realize embedding and the extraction of digital watermarking in print scanned front and back; This method quantity of information that can embed is bigger simultaneously; Can also resist general image processing operations and high-intensity noise in addition attacks; And not needing original image to participate in also not needing to carry out watermark parameters during watermark extracting estimates.Therefore, this method is a kind of blind watermarking algorithm of high robust, has the broad commercial applications prospect.

We know that the coefficient of natural image piecemeal dct transform fixed frequency distributes about former point symmetry, thereby its expectation value is 0.The print scanned process of image is equivalent to the image independent identically distributed noise that superposeed in the spatial domain, so the ac coefficient of image block dct transform is still followed certain distribution about former point symmetry after having experienced print scanned process, thereby the expectation value of its ac coefficient is still 0.In the present invention, the distribution that has changed the ac coefficient of image block dct transform artificially makes its expectation value for just or for negative, image can also remain through this character after print scanned, thereby can successfully realize embedding and the extraction of digital watermarking in print scanned front and back.

Description of drawings

Figure 1A is the process flow diagram of embed watermark in digital picture;

Figure 1B is the process flow diagram that extracts watermark in digital picture;

Fig. 2 is of a size of ' Lena ' gray level image synoptic diagram of 512 * 512, and the image size is 257KB;

Fig. 3 is the medium and low frequency section synoptic diagram in two-dimensional dct transform territory, and dash area is the channel that is used for embed watermark;

It is ' Lena ' image synoptic diagram of 127bits watermark information that Fig. 4 is to use this method to embed length, PSNR=39.17dB wherein, and whole watermark strength a=4, the BCH Error Correction of Coding of use is (127,78,7);

Fig. 5 is the synoptic diagram that the image behind the embed watermark has been added gaussian noise and the resulting image of salt green pepper noise;

Fig. 6 is the image synoptic diagram that the image cutting 1/4th behind the embed watermark is obtained;

Fig. 7 is the image synoptic diagram that the image behind the embed watermark is obtained as low-pass filtering;

Fig. 8 makes the image that the JPEG compression obtains to the image behind the embed watermark, and wherein the JPEG compression quality is 20%, and the image size after the compression is 15.4KB;

Fig. 9 makes the image that the JPEG compression obtains to the image behind the embed watermark, and wherein the JPEG compression quality is 10%, and the image size after the compression is 7.93KB;

Figure 10 A scans resulting image synoptic diagram to the image print behind the embed watermark;

Figure 10 B intercepts and zooms to the synoptic diagram that is of a size of 512 * 512 resulting images to the shown image of Figure 10 A;

Figure 11 A scans resulting image synoptic diagram to the image print behind the embed watermark;

Figure 11 B intercepts and zooms to the synoptic diagram that is of a size of 512 * 512 resulting images to the shown image of Figure 11 A.

Embodiment

Below in conjunction with accompanying drawing the specific embodiment of the present invention is further described.

1: watermark embed process

As shown in Figure 1, a kind of in digital picture the method for embed watermark, may further comprise the steps:

(1) a width of cloth gray level image (perhaps coloured image passage) that will be as shown in Figure 2 is divided into size and obtains coefficient in transform domain for the piece of N * N carries out two-dimensional dct transform to each image block then.In the present embodiment, the size of getting image block is N=8.Specifically, suppose that gray level image (perhaps coloured image passage) is

$X=\{x_{n,m}^k:0\&le;n\&le;\mathrm{7,0}\&le;m\&le;\mathrm{7,1}\&le;k\&le;K\},$

X wherein _{N, m} ^kThe gray-scale value of represent pixel point, k are the labels of image block, and (n m) is the label of position in the image block.Each image block is carried out two-dimensional dct transform obtain coefficient in transform domain

$\{{\hat{x}}_{u,v}^k:0\&le;u\&le;\mathrm{7,0}\&le;v\&le;\mathrm{7,1}\&le;k\&le;K\}.$

Wherein the formula of two-dimensional dct transform is

${\hat{x}}_{u,v}^k=\frac{1}{4}{C}_u{C}_v\underset{n=0}{\overset{7}{\mathrm{\&Sigma;}}}\underset{m=0}{\overset{7}{\mathrm{\&Sigma;}}}{x}_{n,m}^k\cos \frac{(2n+1)\mathrm{u\&pi;}}{16}\cos \frac{(2m+1)\mathrm{v\&pi;}}{16}$

Wherein, if u=0, then $C_u=1/\sqrt{2};$ If u ≠ 0, then C _u=1.

(2) generate string of binary characters according to the watermark information that will embed, string of binary characters is carried out Error Correction of Coding obtain watermark signal to be embedded: w ₁, w ₂..., w _LW wherein _lValue 1 or 0.Medium and low frequency channel with image block

Λ＝{(u，v，k)：N ₁≤u+v≤N ₂，1≤k≤K}

Be divided into the L group

$\mathrm{\&Lambda;}=Y_{l=1}^L{\mathrm{\&Lambda;}}_l,$

And make the number λ of contained channel in each grouping _l=# Λ _lIdentical or identical as much as possible.In the present embodiment, the medium and low frequency channel of image block is divided into 127 groups, makes that the number of contained channel is 709 or 710 in each grouping.As shown in Figure 3, our boundary of the medium and low frequency section chosen is N here ₁=3 and N ₂=6.

(3) according to formula (EQ1), (EQ2) with the watermark strength a that (EQ3) calculates each channel _{U, v} ^kChoose whole watermark strength a＞0 and according to additivity watermark mode to watermark signal w _lBe embedded into channel packet Λ _lEach channel in:

${\hat{y}}_{u,v}^k={\hat{x}}_{u,v}^k+a{a}_{u,v}^k({2w}_l-1)$ $\&ForAll;(u,v,k)\&Element;{\mathrm{\&Lambda;}}_l.$

(4) image block of each embed watermark is made image after the two-dimensional dct inverse transformation obtains embed watermark.Wherein the formula of two-dimensional dct inverse transformation is

$x_{n,m}^k=\frac{1}{4}\underset{u=0}{\overset{7}{\mathrm{\&Sigma;}}}\underset{v=0}{\overset{7}{\mathrm{\&Sigma;}}}{C}_u{C}_v{\hat{x}}_{u,v}^k\cos \frac{(2n+1)\mathrm{u\&pi;}}{16}\cos \frac{(2m+1)\mathrm{v\&pi;}}{16}$

As shown in Figure 4, we are the BCH Error Correction of Coding of the watermark information operation parameter of 78 bits for (127,78,7) to length, and choose whole watermark strength a=4.

2: watermark extraction process

A kind of method of extracting watermark from digital picture may further comprise the steps:

(1) gray level image (perhaps coloured image passage) that obtains is divided into size and obtains coefficient in transform domain for the piece of N * N (according to the method identical with watermark embed process, N=8 in the present embodiment) carries out two-dimensional dct transform to each image block then.Suppose that the coefficient in transform domain that obtains is

$\{{\hat{y}}_{u,v}^k:0\&le;u\&le;\mathrm{7,0}\&le;v\&le;\mathrm{7,1}\&le;k\&le;K\}.$

(2) expectation value of calculating the coefficient of each grouped channels according to the packet mode identical with watermark embed process is calculated watermark extracting in other words

$T_l={\mathrm{\&Sigma;}}_{(u,v,k)\&Element;{\mathrm{\&Lambda;}}_l}{\hat{y}}_{u,v}^k$

Wherein l ∈ 1,2 ..., L}.If the sub-T of watermark extracting _lThink watermark signal for＞0 ${\tilde{w}}_l=1,$ Otherwise think watermark signal ${\tilde{w}}_l=0.$

(3) the watermark bit stream string to obtaining by (2)

Decode, obtain the actual watermark information of hiding.

Below the concise and to the point theoretical foundation of explaining algorithm of the present invention.We will be illustrated as the attack what this watermark embedding extraction method can be resisted noise.The image of supposing to comprise watermark has been subjected to independent identically distributed noise and has attacked (this also can be similar to think the influence that print scanned process produces image) in the spatial domain, in other words our gray level image (perhaps coloured image passage) of obtaining

$Z=\{z_{n,m}^k:0\&le;n\&le;\mathrm{7,0}\&le;m\&le;\mathrm{7,1}\&le;k\&le;K\}$

Be gray level image (perhaps coloured image passage)

$Y=\{y_{n,m}^k:0\&le;n\&le;\mathrm{7,0}\&le;m\&le;\mathrm{7,1}\&le;k\&le;K\}$

The independent identically distributed noise field that superposeed in the spatial domain, in other words

$z_{n,m}^k=y_{n,m}^k+f_{n,m}^k,$

Wherein $\{f_{n,m}^k:0\&le;n\&le;\mathrm{7,0}\&le;m\&le;\mathrm{7,1}\&le;k\&le;K\}$ It is independent identically distributed stochastic variable.So we can obtain at frequency domain

${\hat{z}}_{u,v}^k={\hat{y}}_{u,v}^k+g_{u,v}^k,$

Notice that two-dimensional dct transform is an orthogonal transformation, thereby

(1) $\{g_{n,v}^k:0\&le;u\&le;\mathrm{7,0}\&le;v\&le;\mathrm{7,1}\&le;k\&le;K\}$ It is independent random variables;

(2) for fixing channel (u, v), stochastic variable $\{g_{u,v}^k:1\&le;k\&le;K\}$ Has identical distribution;

(3) if (u, v) ≠ 0, stochastic variable g then _{U, v} ^kExpectation be 0.

Then we can obtain for the sub-Tl of watermark detection

$T_l={\mathrm{\&Sigma;}}_{(u,v,k)\&Element;{\mathrm{\&Lambda;}}_l}{\hat{z}}_{u,v}^k={\mathrm{\&Sigma;}}_{(u,v)}{\mathrm{\&Sigma;}}_{k\&Element;{\mathrm{\&Lambda;}}_{l,u,v}}{\hat{z}}_{u,v}^k={\mathrm{\&Sigma;}}_{(u,v)}{\mathrm{\&Sigma;}}_{k\&Element;{\mathrm{\&Lambda;}}_{l,u,v}}({\hat{y}}_{u,v}^k+g_{u,v}^k),$

So

$T_l={\mathrm{\&Sigma;}}_{(u,v)}{\mathrm{\&Sigma;}}_{k\&Element;{\mathrm{\&Lambda;}}_{l,u,v}}{\hat{x}}_{u,v}^k+{\mathrm{\&Sigma;}}_{(u,v)}{\mathrm{\&Sigma;}}_{k\&Element;{\mathrm{\&Lambda;}}_{l,u,v}}{g}_{u,v}^k+a({2w}_l-1){\mathrm{\&Sigma;}}_{(u,v,k)\&Element;{\mathrm{\&Lambda;}}_l}{a}_{u,v}^k,---\left(\mathrm{EQ}4\right)$

Index set Λ wherein _{L, u, v}Definition be

Λ _l，u，v＝{k：(u，v，k)∈Λ _l}。

Suppose that (u, it is p that v) coefficient distributes for the channel of the two-dimensional dct transform of image Y _{U, v}, we know as (u, v) ≠ 0 o'clock p _{U, v}Be function, thereby its expectation value is 0 about former point symmetry.Notice at watermark label l and channel label (u, v) fixing and set Λ _{L, u, v}When comprising abundant element, set $\{{\hat{x}}_{u,v}^k:k\&Element;{\mathrm{\&Lambda;}}_{l,u,v}\}$ Constituted distribution p _{U, v}A sampling, so

${\mathrm{\&Sigma;}}_{k\&Element;{\mathrm{\&Lambda;}}_{l,u,v}}{\hat{x}}_{u,v}^k\&ap;\#{\mathrm{\&Lambda;}}_{l,u,v}\&CenterDot;E\left(p_{u,v}\right)\&ap;0.$

Simultaneously, by the discussion of front as can be known, set $\{g_{u,v}^k:k\&Element;{\mathrm{\&Lambda;}}_{l,u,v}\}$ Constituted certain expectation and be a sampling of 0 distribution, so in like manner as can be known

${\mathrm{\&Sigma;}}_{k\&Element;{\mathrm{\&Lambda;}}_{l,u,v}}{g}_{u,v}^k\&ap;0.$

Thereby equation (EQ4) can be reduced to

$T_l\&ap;a({2w}_l-1){\mathrm{\&Sigma;}}_{(u,v,k)\&Element;{\mathrm{\&Lambda;}}_l}{a}_{u,v}^k,$

Notice each coefficient a _{U, v} ^kAll be positive number, so the sub-T of watermark detection _lSymbol can determine that watermark signal is ' 0 ' or ' 1 '.

By several examples the robustness that embeds and extract water mark method in digital picture of the present invention is described again at last.Fig. 5 has added gaussian noise and the resulting image of salt green pepper noise to the image behind the embed watermark, and at this moment picture quality has been suffered serious destruction but still can therefrom successfully have been extracted watermark information.Fig. 6 crops 1/4th of the lower right corner with the image behind the embed watermark, still can therefrom successfully extract watermark information, if the part that image is cropped other also can therefrom be extracted watermark information.Fig. 7 makes the image that low-pass filtering obtains afterwards to the image behind the embed watermark, and at this moment image has produced smoothly, and we can therefrom successfully extract watermark information.Fig. 8 makes the image that the JPEG compression obtains afterwards to the image behind the embed watermark, and wherein the JPEG compression quality is 20%, and the image size after the compression is 15.4KB, and we can therefrom successfully extract watermark information.Fig. 9 makes the image that the JPEG compression obtains afterwards to the image behind the embed watermark, wherein the JPEG compression quality is 10%, image size after the compression is 7.93KB, at this moment picture quality has suffered that tangible destruction has lost use value, the failure of extraction watermark information, the watermark extracting success ratio is 107/127=0.8425.Figure 10 A is the image that obtains after the scanning of the image print behind the embed watermark; Wherein the output DPI of image is 150, prints under 600DPI by the capital printer C850 that speeds, and the size of image print on paper is 8.67cm * 8.67cm, scans under 600DPI through hewlett-packard scanner ScanJet4890 then; Be stored in the computing machine with electronic format through the image after the said process, what obtain is to be of a size of 2416 * 2408 image with white frame, and wherein we add the border of the dark border of a pixel with the explanation scan image in the scan image most external for convenience of description.Figure 10 B is us by Figure 10 A being carried out image boundary is extracted and estimating that to printing the image inclination angle that scanning process produced (image inclination angle in the implementation case is 0.22 degree) makes it the position level to image 0.22 degree that turns clockwise then, at last postrotational image is removed white frame and intercept and zoom to and be of a size of 512 * 512 image, we can therefrom successfully extract watermark information by the watermark detection algorithms among the present invention.Figure 11 A is the image that obtains after the scanning of the image print behind the embed watermark; Wherein the output DPI of image is 300, prints under 600DPI by the capital printer C850 that speeds, and the size of image print on paper is 4.33cm * 4.33cm, scans under 600DPI through hewlett-packard scanner ScanJet4890 then; Be stored in the computing machine with electronic format through the image after the said process, what obtain is to be of a size of 1420 * 1444 image with white frame, and wherein we add the border of the dark border of a pixel with the explanation scan image in the scan image most external for convenience of description.Figure 11 B is us by Figure 11 A being carried out image boundary is extracted and estimating that to printing the image inclination angle that scanning process produced (image inclination angle in the implementation case is 0.12 degree) makes it the position level to image 0.12 degree that turns clockwise then, at last postrotational image is removed white frame and intercept and zoom to and be of a size of 512 * 512 image, we can therefrom successfully extract watermark information by the watermark detection algorithms among the present invention.

Method of the present invention is not limited to the embodiment described in the embodiment, and those skilled in the art's technical scheme according to the present invention draws other embodiment, belongs to technological innovation scope of the present invention equally.

Claims (12)

1. watermark embedding method based on the coefficient in transform domain of digital images statistical model may further comprise the steps:

(1) does two-dimensional dct transform with the digital picture piecemeal and to each image block;

(2) generate watermark signal according to the watermark information that will embed;

(3), the medium and low frequency section coefficient of image block dct transform domain is divided into groups according to the length of the watermark signal that obtains by step (2);

(4) analyze the information complexity and the bright-dark degree of each image block, calculate the watermark embed strength of each image block dct transform domain medium and low frequency channel;

(5) choose whole watermark strength, and, come the embed watermark signal by the expectation value that changes the DCT coefficient in each channel packet that obtains by step (3) according to watermark strength by described each channel of step (4);

(6) will carry out the image that the two-dimensional dct inverse transformation just obtains embed watermark respectively through all images piece of above-mentioned steps.

2. a kind of watermark embedding method based on the coefficient in transform domain of digital images statistical model as claimed in claim 1 is characterized in that: in the step (2), described watermark signal is to carry out the watermark bit stream string to be embedded that obtains after the Error Correction of Coding.

3. a kind of watermark embedding method as claimed in claim 2 based on the coefficient in transform domain of digital images statistical model, it is characterized in that: in the step (2), described watermark bit stream string is by watermark signal is carried out BCH (Bose, Ray-Chaudhuri, Hocquenghem) mode of Error Correction of Coding generates.

4. a kind of watermark embedding method based on the coefficient in transform domain of digital images statistical model as claimed in claim 3 is characterized in that: the length of described watermark bit stream string is identical with the number of the section of image block dct transform domain medium and low frequency described in the step (3) coefficient grouping.

5. a kind of watermark embedding method based on the coefficient in transform domain of digital images statistical model as claimed in claim 4 is characterized in that: in the step (3), the number of channel is identical or identical as much as possible in described each medium and low frequency section coefficient grouping.

6. as the described a kind of watermark embedding method of one of claim 1 to 5, it is characterized in that based on the coefficient in transform domain of digital images statistical model:

In the step (4), the information complexity of described each image block is the information entropy e of image block ^k, information entropy e ^kBe to calculate by the following formula that quantizes based on gray-scale value:

$e^k={\mathrm{\&Sigma;}}_{i=0}^{15}{P}_i^k\log P_i^k,---\left(\mathrm{EQ}1\right)$

$P_i^k=\frac{1}{N^2}\#\{(n,m):0\&le;n,m\&le;N-1;16i\&le;x_{n,m}^k<16(i+1)\}$

Wherein, the size of each image block is N * N, x _{N, m} ^kThe gray-scale value of presentation video, k are the labels of image block, and (n m) is the label of position in the image block.

7. a kind of watermark embedding method based on the coefficient in transform domain of digital images statistical model as claimed in claim 6 is characterized in that:

In the step (4), the f of bright-dark degree of described each image block ^kBe to calculate by following formula: $f^k=\left\{\begin{array}{ccc}\frac{{\stackrel{\&OverBar;}{x}}^k-(255-T^{+})}{T^{+}}& \mathrm{if}& {\stackrel{\&OverBar;}{x}}^k>255-T^{+}\\ \frac{T^{-}-{\stackrel{\&OverBar;}{x}}^k}{T^{-}}& \mathrm{if}& {\stackrel{\&OverBar;}{x}}^k<T^{-}\\ 0& \mathrm{if}& T^{-}\&le;{\stackrel{\&OverBar;}{x}}^k\&le;255-T^{+}\end{array}\right.,---\left(\mathrm{EQ}2\right)$

Wherein, x ^kThe expression label is the average gray value of the image block of k, T ⁺And T ^-Be two pre-set threshold values.

8. a kind of watermark embedding method based on the coefficient in transform domain of digital images statistical model as claimed in claim 7 is characterized in that:

T ⁺＝T ^-＝50。

9. a kind of watermark embedding method based on the coefficient in transform domain of digital images statistical model as claimed in claim 8 is characterized in that:

In the step (4), the watermark embed strength of each channel of described each image block dct transform is the information entropy e by each image block ^kWith the f of bright-dark degree ^kCalculate according to following formula:

$a_{u,v}^k=(1-e^k)(1+f^k)\frac{\sqrt{u+1}\sqrt{v+1}}{(u+1)+(v+1)}---\left(\mathrm{EQ}3\right).$

10. a kind of watermark embedding method as claimed in claim 9 based on the coefficient in transform domain of digital images statistical model, it is characterized in that: in the described watermark embed process of step (5), the watermark strength of each channel that obtains according to the whole watermark strength of choosing with by step (4) uses additivity watermark embedded mode to revise the expectation value of the coefficient in each channel packet.

11. a kind of watermark embedding method as claimed in claim 10 based on the coefficient in transform domain of digital images statistical model, it is characterized in that: in the watermark embed process described in the step (5), when the embed watermark signal is 1 in this grouping, then increase coefficient expectation value that each coefficient in this grouping makes this grouping for just; Otherwise when the embed watermark signal was 0 in this grouping, each coefficient that then reduces in this grouping made the coefficient expectation value of this grouping for negative.

12. the watermark extracting method based on the coefficient in transform domain of digital images statistical model may further comprise the steps:

(1) obtains digital picture and its piecemeal carried out two-dimensional dct transform to each image block then;

(2) expectation value of calculating coefficient in each channel packet obtain in other words all coefficients in this channel packet and, if expectation value is a canonical think with this corresponding watermark signal that divides into groups be 1, otherwise think with this corresponding watermark signal that divides into groups be 0;

(3) the above-mentioned string of binary characters decoding that obtains is obtained watermark information.

Images