CN101437163A

CN101437163A - Multi-watermark technology based on network information theory

Info

Publication number: CN101437163A
Application number: CNA2009100007071A
Authority: CN
Inventors: 肖俊; 王颖
Original assignee: University of Chinese Academy of Sciences
Current assignee: University of Chinese Academy of Sciences
Priority date: 2009-01-08
Filing date: 2009-01-08
Publication date: 2009-05-20
Anticipated expiration: 2029-01-08
Also published as: CN101437163B

Abstract

本发明提供一种多水印方法，其基本原理是将多水印的实现过程视为一种含边信息的多用户通信过程，利用相关信源编码定理和多用户通信理论等理论分析并获得通信过程中可利用的边信息，包括水印嵌入和提取两个过程，包括步骤：利用载体的对应分块与各水印编码信号之间的最大线性相关值作为边信息选择嵌入位置、检测阈值和嵌入强度；利用感知模型对水印信号进行感知成形；通过固定载体与水印编码信号之间的线性相关值来调节水印嵌入强度获得含水印载体；通过线性相关检测器进行检测。本方案可有效避免多个水印之间的相互干扰，并具有较好的鲁棒性，可用于网络出版、操作跟踪等场合的版权管理中。The present invention provides a multi-watermarking method, the basic principle of which is to regard the realization process of multi-watermarking as a multi-user communication process with side information, and use the relevant information source coding theorem and multi-user communication theory to analyze and obtain the communication process The side information available in , including two processes of watermark embedding and extraction, includes the steps of: using the maximum linear correlation value between the corresponding block of the carrier and each watermark coded signal as the side information to select the embedding position, detection threshold and embedding strength; The watermark signal is perceptually formed by using the perceptual model; the watermark embedding strength is adjusted by fixing the linear correlation value between the carrier and the watermark coded signal to obtain the watermark carrier; the linear correlation detector is used for detection. This scheme can effectively avoid mutual interference between multiple watermarks, and has better robustness, and can be used in copyright management in network publishing, operation tracking and other occasions.

Description

The multi-watermark technology of information theory Network Based

Technical field

The invention belongs to the Communication and Information Systems field, relate to technology such as information theory and coding and image processing, particularly the digital figure watermark technology.

Background technology

Digital watermarking is considered " last line of defense " of multi-media safety as a kind of technology that rises because of copyright protection, has played important function in the transmission of copyright and preservation process.If in a Digital Media, only embedded an information, then be referred to as single watermark; If embedded two or more information, then be referred to as many watermarks.Along with concrete use increasingly extensive of works transaction, usage track, Web publishing etc. under the network environment, demand to multi-watermark technology is urgent day by day, what the present invention relates to mainly is many watermarks field, is characterized in many watermaking systems are considered as the multi-user communication process of an information theory Network Based.

Multi-watermark technology has been obtained certain achievement in research, and wherein the most representative method can be divided three classes: repeat embedding, piecemeal embedding and merge to embed.

Repeating to embed (Re-Watermarking) is direct, the most tangible method, it is directly transplanted by single watermarking algorithm, thereby in fact be exactly repeatedly to adopt single watermarking algorithm that a plurality of watermark informations are embedded one by one, the watermark quantity that this method embeds directly influences the quality of carrier works, and after adopting this method to embed a plurality of watermarks, when detecting i watermark, must have the i time watermarked after, the i+1 time data before watermarked, this will limit its availability in a lot of practical applications.

Piecemeal embed (Segmented Watermarking) with repeat to embed different, it is cut apart being used for watermarked space, each subspace embeds a different watermark information, thereby realizes the embedding of a plurality of watermark informations, and can detect simultaneously.The restricted number of cutting apart can be with this theoretical watermark quantity that embeds, when cutting apart quantity and increase, each subspace will reduce simultaneously, this moment is the watermarked difficulty more that just becomes in each subspace.And the great advantage of this method is to embed a plurality of watermark informations simultaneously, needn't carry out in order, and can detect simultaneously, therefore uses more in practice.

With watermarked different a kind of scheme one by one is earlier each watermark to be merged, and then the single watermark embedded mode of synthetic watermark according to routine is embedded in the carrier, promptly merges to embed.This method can be broken away from the carrier capacity preferably and embed the restriction of algorithm to watermarked number, but these class methods require good application image integration technology, could obtain effect preferably, and require a plurality of watermarks to embed simultaneously, be not suitable for the situation that to predict watermark number and all watermark informations, comparatively speaking, this type of algorithm is less.But it should be noted that a transform method of these class methods---quadrature watermark method, use more.So-called quadrature watermark method is meant that the watermark of embedding is the pseudo random sequence of quadrature, when coherent detection, can not influence each other like this; This just is equivalent to as a whole embedding is done in the advanced line linearity stack of the pseudo random sequence of a plurality of quadratures then, and the process of this process and fusion is very similar.And when detecting watermark, because mutually orthogonal between each watermark, can the phase mutual interference, but the existence that this method can only detect watermark information usually is whether, and must quadrature or nearly orthogonal between each watermark, this is restricted in actual applications.

To sum up, existing many water mark methods mostly are the simple deformation of single watermarking algorithm, remain to be further developed at aspect of performances such as capacity, validity, robustnesss.

Containing the side information watermark is to produce on the basis of side information theory, network source encoding theory and dirty paper code theory.Cox equals first watermark to be regarded as in 1999 a kind of communication that contains side information.At present, contain the advantage of side information watermaking system, obtain people's approval gradually at aspects such as capacity, robustness, validity.

The present invention is applied to many watermarks field with side information thought, and a kind of new many water mark methods are provided.

Summary of the invention

Prior theory research and experimental study show that all the use of side information can improve the performance of watermaking system really.Yet the application of side information thought in the watermark field mainly concentrates in single watermark, fully used as yet in many watermarks.

Main purpose of the present invention provides a kind of new many water mark methods, its basic principle is that the implementation procedure with many watermarks is considered as a kind of multi-user communication process that contains side information, make full use of the side information that can from entities such as carrier and watermark, obtain in the embodiment side information ideologize watermark implementation procedures such as network source encoding theory and multi-user communication theory, and utilize these side informations to realize the embedding and the extraction of many watermarks.

The multi-watermark technology that the present invention proposes comprises that a kind of information theory Network Based realizes the many water mark methods of piecemeal of the method and an information theory Network Based of many watermarks.

1. information theory Network Based realizes the method for many watermarks

Prior theory research and experimental study show that all the use of side information can improve the performance of watermaking system really, but what adopt at present only is the simplest side information form, promptly for information source X ₁With information source X ₂, at transmission information source X ₁The time, detector only is concerned about whether can effectively detect X ₁, and be indifferent to and can detect X ₂, as shown in Figure 1; Equally, at transmission information source X ₂The time, detector only is concerned about whether can effectively detect X ₂, can the pass is indifferent to detect X ₁Contain in the side information list watermaking system existing, just as supplementary, more effectively transmit watermark with auxiliary watermaking system from the side information of carrier, this is quite reasonable in single watermaking system.And in many watermaking systems, more than one of the watermark information that need to embed, if still only consider side information between carrier and the different watermarks, the side information thought in so single watermaking system can be applied directly in many watermaking systems; Yet, in fact, may also have correlation between a plurality of watermark informations, and this correlation can be utilized mutually also, and this moment be in identical status between a plurality of watermark informations, it is more reasonable to utilize correlated source coding theorem in the network information theory that it is carried out analysis meeting.Because in the correlated source coding theory, a plurality of information sources are of equal importance, and detector requires to detect effectively all information, and what Fig. 2 showed is a kind of typical correlated source traffic model.

We are on the basis of analyzing network information theories such as correlated source coding theorem, multi-user communication, the implementation procedure of many watermarks is analogized to multi-user communication process based on side information, and the information theory Network Based that provides Fig. 3 to show realizes the embedding grammar of many watermarks.In the method,, from carrier and a plurality of watermark information, extract side information, each watermark information is contained the coding or the preliminary treatment of side information in the watermark encoder stage; Embed the stage in watermark, still fully excavate the side information in carrier and each watermark, to determine embedded location, embedment strength effectively.And, need utilize some side informations in watermark encoder and the watermark embed process to determine detection threshold equally in the watermark detection stage.

2. many water mark methods of the piecemeal of information theory Network Based

Traditional many water mark methods of typical piecemeal are directed to single watermarking algorithm, carrier is carried out piecemeal after, not enough to the correlation utilization between a plurality of of carrier and the many watermarks, be also referred to as blind many watermarking algorithms, performances such as its validity are not high.

We utilize the relevant information between carrier works and watermark information and a plurality of watermark to optimize watermark embedded location, watermark embed strength and watermark detection threshold value, proposed a kind of many water mark methods of piecemeal that make full use of the information theory Network Based of various side informations, what Fig. 4 showed is the block diagram that watermark embeds and extracts.This method replaces a watermark information with big reference templates such as carrier works piecemeal and generation and piecemeals with a reference template, carries out the watermark information preliminary treatment and obtains the watermark information template; Utilize the Watson sensor model to calculate the perception gap, and utilize the perception gap that the watermark information template is carried out the perception shaping and handle,, reduce embedding distortion to improve the fidelity of carrier to greatest extent; Utilize carrier works side information to seek the optimum position that watermark embeds in the mode of maximum correlation, make that the interference between carrier and a plurality of watermark information reaches minimum, the validity when guaranteeing watermark detection; Utilize the carrier side information to determine embedment strength in the mode that fixed linear is relevant, making has fixing linear correlation values between carrier partition and the corresponding watermark information, thereby further guarantees the validity of watermark detection.

Description of drawings

Fig. 1 utilizes form for the simplest side information, and this also is that side information the most frequently used in single watermaking system utilizes mode.

Fig. 2 is a kind of typical correlated source traffic model.

Fig. 3 realizes the block diagram that many watermarks embed for information theory Network Based provided by the invention.

Fig. 4 is many water mark methods of piecemeal schematic diagram of information theory Network Based provided by the invention.

The validity test result of method when selecting carrier image at random that Fig. 5 proposes for the present invention.

The validity test result of method when selecting watermark information at random that Fig. 6 proposes for the present invention.

Embodiment

Describe each detailed problem that relates in the method provided by the present invention in detail below in conjunction with accompanying drawing.

1. information theory Network Based realizes the method for many watermarks

As shown in Figure 3, the information theory Network Based provided by the invention specific embodiments that realizes many watermarks is as follows:

(1) analyze a plurality of watermark m[i to be embedded] (1≤i≤N) and support C _oBetween correlation and the correlation between a plurality of watermark information, be that side information is to watermark information m[i with these passes] contain the coding of side information, obtain watermark encoder signal W _i, wherein N represents the watermark number that needs embed.

(2) in conjunction with the embedding strategy of fixing robustness, fixedly the embedding strategy etc. of fidelity contains the embedding strategy of side information, according to support C _oWith all watermark encoder signal W _i(relation between 1≤i≤N) realizes W ₁Embedding, in telescopiny, comprise and utilize side information to select embedded location, embedment strength, finally obtain moisture seal signal C _W1

(3) repeat above-mentioned steps, all be encoded and embed, finally obtain moisture seal signal C up to all watermark signals _W

According to foregoing description, among the present invention watermark encoder with embed device and routine and contain identical in the side information list watermaking system, can adopt various codings and embedded mode, and watermark extraction process also with routine to contain the watermark of side information list identical, its special character is, relates to a plurality of watermarks in the telescopiny, thereby from the angle analysis problem of multi-user communication, and, set up the method that realizes that many watermarks embed based on the correlated source coding theory with a plurality of watermark information fair plays.Compare with existing many water mark methods, its difference is: in the process of a plurality of watermark informations being encoded and embedding, encoder and embedding device have made full use of the correlation between carrier and watermark and a plurality of watermark, and this is the embodiment of network source encoding theory.

2. many water mark methods of the piecemeal of information theory Network Based

As shown in Figure 4, the many water mark methods of piecemeal provided by the invention at first utilize the corresponding piecemeal of carrier works and the maximum linear correlation between the watermark signal to select embedded location as side information, the detection threshold and the embedment strength factor, secondly the carrier works gap array correspondence that will utilize the Watson model to obtain is assigned to its perception is shaped, obtain the watermark encoder signal, regulate watermark embed strength by linear correlation values between immobilization carrier works and the watermark encoder signal then, finish telescopiny, utilize the linear correlation detector to detect at last, the output watermark information.Relate generally to following 4 kinds of modes of utilizing side information in the implementation process:

(1) utilize the corresponding piecemeal of carrier and the maximum linear correlation between the watermark information to select embedded location, detection threshold and embedment strength as side information;

(2) utilize the Watson model to extract the carrier side information, use the gap array that the information model of representing watermark is carried out perception and be shaped, obtain the watermark encoder signal;

(3) utilize carrier to make linear correlation values between the watermark encoder signal, regulate watermark embed strength in the mode of linear correlation values between immobilization carrier and the watermark encoder signal as side information;

(4) detection threshold of determining according to the maximum linear correlation between corresponding piecemeal of carrier and the watermark information detects watermark information in the mode of linear correlation.

Select the implementation step of embedded location, detection threshold and embedment strength 2.1 utilize side information

(1) establish carrier image size and be M1 * N1, watermark quantity to be embedded is N, then according to formula (1) to k watermark m[k] carry out preliminary treatment and obtain watermark information template w _m[k],

w_{m} [k] = \{\begin{matrix} w_{r} [k], & m [k] = 1 \\ {- w}_{r} [k], & m [k] = 0 \end{matrix} - - - (1)

W wherein _r[k] (1≤k≤N) be size for M * M and between [0,1] pseudo random sequence of Normal Distribution, w _r[k, i, j] (1≤i, k value that sequence is located at [i, j] of the expression of j≤M).

(2), carrier is divided into the piece of size for M * M and non-overlapping copies according to watermark quantity

, and use c ₀[t, x, y] (1≤i, pixel value in every of the expression of j≤M).

(3) calculate the information model w that represents k watermark _mAll piecemeals of [k] and carrier

Between linear correlation values, if watermark m[k]=1, then select and w _m[k] has the piece of maximum linear correlation as embedded location, and with maximum related value and the yardstick zoom factor β [k] of time difference of big correlation as watermark, simultaneously will time big correlation threshold tau during as watermark detection _Lc[k]; If m[k]=0, then select and w _m[k] has the piece of minimal linear correlation as embedded location, and with the difference of minimum correlation value and time little correlation yardstick zoom factor β [k] as watermark, simultaneously will time little correlation threshold tau during as watermark detection _Lc[k].Wherein zoom factor β [k] is used for follow-up definite embedment strength.

2.2 utilize the Watson model to extract the implementation step that side information carries out watermark encoder

(1), utilize DC coefficient after each piecemeal discrete cosine transform of carrier to the predefined frequency sensitive kilsyth basalt of Watson model t[i, j according to formula (2)] adjust, obtain brightness and shelter function threshold t _L[k, i, j],

t_{L} [k, i, j] = t [i, j] {(C_{o} [k, 0,0] / C_{0,0})}^{α_{T}} - - - (2)

α wherein _TBe a constant, value is 0.649 usually, C _o[k, 0,0] is the DC coefficient of k piecemeal after discrete cosine transform in the carrier image, C _0,0Be the mean value of all piecemeal DC coefficient of carrier image, also can be set at the constant of representative image expection intensity.

(2) calculate contrast masking threshold s[k, i, j according to formula (3)],

s[k，i，j]＝max{t _L[k，i，j]，|C _o[k，i，j]| ^w[i，j]t _L[k，i，j] ^1-w[i，j]} (3)

W[i in the formula, j] be a constant between 0 and 1, and can be different because of the difference of coefficient of frequency, all w[i in the Watson model, j] all be taken as 0.7.Final thresholding s[k, i, j] estimated be every variation size that in a just perceptible scope, can stand of piece discrete cosine transform, also be called the gap.

(3) utilize the gap that the watermark information template is carried out perception and be shaped, promptly at first the watermark information template is carried out discrete cosine transform and obtain W _m[k] utilizes formula (4) to W then _m[k] carries out perception and is shaped and obtains perception shaping information W under the discrete cosine transform domain _s[k],

W _s[k]＝W _m[k]s[k] (4)

At last to W _s[k] carries out inverse discrete cosine transform and obtains spatial domain shaping template w _s[k], i.e. watermark encoder signal.

Determine the implementation step of embedment strength 2.3 utilize side information

(1) embedment strength of supposing k watermark is α [k], then watermark encoder signal w _s[k] will be embedded in the corresponding piecemeal by formula (5),

c _w[k]＝c _o[k]+α[k]w _s[k] (5)

At this moment, can calculate moisture seal signal c according to formula (6) _w[k] and watermark information template w _mLinear correlation values between [k],

z _lc(w _m[k]，c _w[k])＝(C ₀[k]·w _m[k]+w _a[k]·w _m[k])/(M×M) (6)

W wherein _α[k] determines according to formula (7).

w _a[k]＝α[k]w _s[k] (7)

In formula (6), use τ _Lc[k]+β [k] replaces z _Lc(w _m[k], c _w[k]), can obtain watermark embed strength and be

α [k] = \frac{L (τ_{lc} [k] + β [k]) - c_{0} [k] \cdot w_{m} [k]}{w_{s} [k] \cdot w_{m} [k]} - - - (8)

(2) embedment strength that adopts (8) formula to obtain obtains containing watermarking images according to formula (5) embed watermark information.

2.4 extract the implementation step of watermark in conjunction with side information

(1) adopt with telescopiny in identical mode to carrier c to be detected _nCarry out piecemeal, and obtain each piecemeal c _n[t]

(2) with the embedded location in the telescopiny as key, may contain the piecemeal c of watermark information according to formula (9) _nDetect watermark information in [k],

Z wherein _Lc(w _r[k], c _n[k]) expression carrier block c _n[k] and watermark information w _rLinear correlation values between [k], τ _Lc[k] is the detection threshold that utilizes the linear correlation values between carrier partition and the watermark information template to determine in the telescopiny.

3. test result

We are applied in the method that is proposed under the multiple situations such as different images, different watermark informations, have carried out sufficient experiment, have verified the validity of many water mark methods proposed by the invention.Provide below with the method for the present invention's proposition and the experiment comparative result of the many water mark methods of conventional blind piecemeal.In the experiment, the carrier of employing is 256 * 256 gray level image, and the watermark quantity of embedding is 128.The many water mark methods of conventional blind piecemeal are divided into the piece of 256 non-overlapping copies with gray level image, and every block size is 16 * 16, selects 128 piecemeals to be used to embed 128 watermark informations then from 256 piecemeals at random, and identical embedment strength is adopted in all watermarks.

(1) validity of carrier image experiment at random

Consider that different images may have certain influence at random to experiment, in order to prove absolutely the broad applicability of this method, we test carrier at random, and it is constant promptly to fix 128 watermark informations, 1000 experiments are carried out in the continuous conversion of carrier image in 1000 width of cloth images.

Fig. 5 is the distribution situation of the average error rate of the watermark information that extracts of two kinds of methods, experimental result shows: under the situation that carrier image is selected at random, the average error rate of this method mainly is distributed near 0, can obtain good validity, and the average error rate of the many water mark methods of conventional blind piecemeal mainly is distributed on 0.6, and its validity is relatively poor.

(2) random watermark validity experiment

Consider that different watermarks also may affect to experiment, in order to prove absolutely the broad applicability of this method, we test random watermark, and promptly the immobilization carrier image is constant, 128 watermark informations constantly change, and 1000 groups of watermark informations are carried out 1000 experiments.

Fig. 6 is the distribution situation of the average error rate of the watermark information that extracts of two kinds of methods, experimental result shows: under the situation that watermark is selected at random, the average error rate of the method that the present invention proposes is distributed near 0, can obtain good validity, and the average error rate of the many water mark methods of conventional blind piecemeal is distributed between 0.5 and 0.6, and its validity is relatively poor.

(3) robustness experiment

In order further to test the performance of method provided by the invention, we test the robustness of this method, and what table 1 showed is that various common images are handled the attack average error rate of a plurality of watermarks of extraction down.As seen from Table 1, handle under the attack at normal image, method provided by the invention can obtain the lower error rate, promptly has robustness preferably.

To sum up, many water mark methods that the present invention proposes have good validity, and normal image processing attack is had robustness preferably.

The robustness of table 1. this method under the normal image processing is attacked

Attack	The error rate
Attack	The error rate	Amplify 2 times	0
Dwindle 1/2	0.023	Amplify 2 times	0
Dwindle 1/2	0.023	Low-pass filtering (3 * 3)	0.257
Medium filtering (3 * 3)	0.047	Low-pass filtering (3 * 3)	0.257
Medium filtering (3 * 3)	0.047	Shear 1/9	0.148
JPEG compresses (50%)	0.035	Shear 1/9	0.148
JPEG compresses (50%)	0.035	JPEG compresses (70%)	0.092
Gas noise (0.0008)	0.098	JPEG compresses (70%)	0.092

Claims

1. A method for realizing multiple watermarks based on network information theory, characterized in that:

Multi-watermarking is regarded as a multi-user communication process with side information, and the characteristics of carrier, multiple watermarks, channels and detectors are analyzed by using the correlation source coding theorem and multi-user communication theory. Extract the side information of watermark encoding, watermark embedding and extraction stages, and use these side information to realize multi-watermark encoding, embedding and extraction.

2. The method according to claim 1, wherein the step of embedding multiple watermarks is:

(1) Use the relationship between each watermark information m[i] (1≤i≤N) and the carrier C _o to encode the watermark information m[i] with side information, and obtain the watermark encoded signal W _i , where N Indicates the number of watermarks to be embedded;

(2) Using the embedding strategy with side information, realize the embedding of W ₁ according to the relationship between the carrier C _o and all watermarked signals W _i (1≤i≤N), and obtain the watermarked signal C _W1 ;

(3) Repeat the above steps until all the watermark signals are embedded, and finally obtain the watermarked signal C _W .

3. A block multi-watermarking method based on network information theory, characterized in that:

(1) Use the maximum linear correlation value between the corresponding block of the carrier and the watermark information as side information to select the embedding position, detection threshold and embedding strength;

(2) Use the Watson model to extract the side information of the carrier, and use the gap array to perform perceptual shaping on the information template representing the watermark to obtain the watermark coded signal;

(3) Using the linear correlation value between the carrier and the watermark coded signal as side information, the watermark embedding strength is adjusted by fixing the linear correlation value between the carrier and the watermark coded signal.

4. according to claim 3, the concrete steps that realize a plurality of watermark embeddings are:

(1) Suppose the size of the carrier image is M1×N1, and the number of watermarks to be embedded is N, then preprocess the kth watermark information m[k] according to formula (1) to obtain the watermark information template w _m [k],

{w w}_{m m} [[k k]] = = \{\begin{matrix} {w w}_{r r} [[k k]],, & m m [[k k]] = = 11 \\ {- - w w}_{r r} [[k k]],, & m m [[k k]] = = 00 \end{matrix} - - - - - - ((11))

Where w _r [k] (1≤k≤N) is a pseudo-random sequence with a size of M×M and a normal distribution between [0, 1], w _r [k, i, j] (1≤i, j≤M) indicates the value of the kth sequence at [i, j];

(2) According to the number of watermarks to be embedded, the carrier is divided into blocks with a size of M×M and non-overlapping blocks

And use c ₀ [t, x, y] (1≤i, j≤M) to represent the pixel value in each block;

(3) Use the linear correlation value between the carrier block and the watermark information template to determine the watermark embedding position and detection threshold, that is, calculate the information template w _m [k] representing the kth watermark information and all the blocks of the carrier

The linear correlation value between, if the watermark information m[k]=1, select the block with the largest linear correlation value with w _m [k] as the embedding position, and use the difference between the largest correlation value and the second largest correlation value as the watermark The scaling factor β[k] of , and the second largest correlation value is used as the threshold τ _lc [k] for watermark detection; if m[k]=0, select the block with the smallest linear correlation value with w _m [k] As the embedding position, the difference between the minimum correlation value and the second smallest correlation value is used as the scaling factor β[k] of the watermark, and the second smallest correlation value is used as the threshold τ _lc [k] for watermark detection, where the scaling factor β[ k] for subsequent determination of embedding strength;

(4) Extract the side information according to the Watson model, and use the gap to perceptually shape the watermark information template, that is, first perform discrete cosine transform on the watermark information template to obtain W _m [k], and then use formula (2) to perform W _m [k] Perceptual shaping obtains perceptual shaping information W _s [k] in the discrete cosine transform domain,

W _s [k]=W _m [k]s[k] (2)

Where s[k] represents the overall perceptual distance of the carrier block c ₀ [k], and finally inverse discrete cosine transform is performed on W _s [k] to obtain the perceptual shaping template w _s [k] in the spatial domain, that is, the watermark coded signal;

(5) Assuming that the embedding strength of the kth watermark is α[k], then the encoded watermark signal w _s [k] will be embedded into the corresponding block according to formula (3),

c _w [k]=c _o [k]+α[k]w _s [k] (3)

At this time, the linear correlation value between the watermarked signal c _w [k] and the watermark information template w _m [k] can be calculated according to formula (4),

z _lc (w _m [k], c _w [k]) = (c ₀ [k]·w _m [k]+w _a [k]·w _m [k])/(M×M) (4)

Where w _α [k] is determined according to formula (5);

w _a [k]=α[k]w _s [k] (5)

In formula (4), replace z _lc (w _m [k], c _w [k]) with τ _lc [k]+β[k] to obtain the watermark embedding strength α[k];

α α [[k k]] = = \frac{L L (({τ τ}_{lc lc} [[k k]] + + β β [[k k]])) - - {c c}_{00} [[k k]] \cdot \cdot {w w}_{m m} [[k k]]}{{w w}_{s the s} [[k k]] \cdot &Center Dot; {w w}_{m m} [[k k]]} - - - - - - ((66))

(6) Use the embedding strength obtained by formula (6), and embed watermark information according to formula (3).

5. according to claim 3, the concrete steps that realize that a plurality of watermarks are extracted are:

(1) Use the same method as in the embedding process to divide the carrier c _n to be detected into blocks, and record each block as

(2) Use the embedding position in the embedding process as the key, and detect the watermark information in the block c _n [k] that may contain watermark information according to formula (7),

m m' ' [[k k]] = = \{\begin{matrix} 11 & {z z}_{11 c c} (({w w}_{r r} [[k k]],, {c c}_{n no} [[k k]])) > > {τ τ}_{lc lc} [[k k]] \\ no   watermark no watermark & | | {z z}_{lc lc} (({w w}_{r r} [[k k]],, {c c}_{n no} [[k k]])) | | \leq \leq {τ τ}_{lc lc} [[k k]] \\ 00 & {z z}_{lc lc} (({w w}_{r r} [[k k]],, {c c}_{n no} [[k k]])) < < {- - τ τ}_{lc lc} [[k k]] \end{matrix} - - - - - - ((77))

where z _lc (w _r [k], c _n [k]) represents the linear correlation value between the carrier block c _n [k] and the pseudo-random sequence w _r [k] used to preprocess the watermark, τ _lc [k] is the detection threshold determined by the linear correlation value between the carrier block and the watermark information template during the embedding process.