CN102208097A - Network image copyright real-time distinguishing method - Google Patents

Abstract

The invention provides a network image copyright real-time distinguishing method. Copyright protection is necessary for digital images during network communication, but the existing robust digital watermarking technology can not achieve the real-time copyright distinguishing. The method comprises the following steps: partitioning an original image to sub-blocks which are not overlapped with each other at an embedded end; processing each sub-block by discrete cosine transformation; generating a feature watermark by comparing the DC coefficient of each sub-block with the average DC coefficient of all sub-blocks; encrypting the Logistic chaotic sequence of the feature watermark; adjusting the magnitudes of two low-frequency coefficients of the discrete cosine transformation of each sub-block and self-embedding the encrypted feature watermark; and processing each sub-block by inverse discrete cosine transformation to obtain images with watermarks. The method has high robustness to attack. Substantially, the method provided by the invention can achieve complete blind detection by combining the self-embedding encrypted feature watermark with watermark blind extraction and authentication, thereby achieving real-time copyright distinguishing of network images.

Description

The real-time discrimination method of a kind of network image copyright

Technical field

The present invention relates to information security field.The present invention designs the real-time discrimination method of a kind of network image copyright, and the copyright of the digital picture of transmission over networks is differentiated in real time.

Background technology

Digital picture is propagated in network and is caused dispute over copyright easily, need carry out copyright protection.The robust digital watermark technology can be used for the digital picture of transmission over networks is carried out copyright protection.

Content according to watermark is classified, and watermark can be divided into meaningful watermark and meaningless watermark.The content that meaningful watermark refers generally to watermark has concrete implication, such as bianry image, gray level image, trade mark, author's personal information etc.; The general corresponding pseudo-random sequence that does not have concrete implication of meaningless watermark.In addition, hot spring etc. ^[1]Also propose " zero watermark ", certain feature of initial carrier is represented in zero watermark in fact.Therefore, zero watermark can be regarded as a kind of meaningful watermark of special shape.

Whether needs are by initial carrier according to the test side, and digital watermark technology can be divided into blind digital watermark technology and non-blind digital watermark technology.Often need be when non-blind digital watermark technology detects watermark in the test side by the information relevant with initial carrier, blind digital watermark technology need be by any information relevant with initial carrier in the test side.In some application scenarios, be subjected to restrictions such as transmission, storage, security, the test side often can't obtain the information relevant with initial carrier, and therefore blind digital watermark technology has more practicality than non-blind digital watermark technology.

The robust digital watermark algorithm is judged copyright by the degree of correlation between the watermark of calculating original watermark and extraction usually in the test side.According to the test side whether needs by with initial carrier, information that original watermark is relevant, the detection behavior of present robust digital watermark algorithm can be divided into 4 classes: (1) the 1st class: the test side both need be by the relevant information of initial carrier, again need be by the relevant information of original watermark.(2) the 2nd classes: the test side need be by the relevant information of initial carrier, but does not need the relevant information of original watermark.(3) the 3rd classes: the test side need be by the relevant information of initial carrier, but need be by the relevant information of original watermark.(4) the 4th classes: the test side neither need be by the relevant information of initial carrier, also need be by the relevant information of original watermark.Traditional non-blind robust watermarking algorithm has the 1st class or the 2nd class detects behavior, and practicality is poor.Blind meaningless robust watermarking algorithm with the 3rd class detection behavior ^[2]The relevant information by any initial carrier does not extract watermark from attacking carrier in the test side, but will produce original pseudorandom watermark sequence by key, calculates both degrees of correlation then and judges copyright.Blind meaningful robust watermarking algorithm with the 3rd class detection behavior ^[3-10]Do not extract watermark from attacking carrier in the test side, calculate the degree of correlation judgement copyright between the watermark of the original watermark that transmits from built-in end and extraction then by the relevant information of any initial carrier.Zero watermarking algorithm ^[1,11]Also be to have the 3rd class to detect behavior.This be because: in the test side, zero watermarking algorithm need take out being stored in notarize original zero watermark at center of third party, calculate then with the zero watermark of extracting between degree of correlation judgement copyright.Robust watermarking algorithm with the 4th class detection behavior does not almost occur, and the present invention claims that this class robust algorithm is complete blind Detecting robust watermarking algorithm.

Present blind robust watermarking algorithm has the 3rd class and detects behavior, still can't reach copyright and differentiate that in real time its practical sexual needs further improve.This be because: need be though have blind robust watermarking algorithm that the 3rd class detects behavior, but need weigh the degree of correlation between the watermark of original watermark and extraction with the judgement copyright by the partial information of original watermark or original watermark by the relevant information of any initial carrier in the test side.So, built-in end transmission original watermark or its partial information to the test side (or third party notarize center) store and just need certain transmission cost and carrying cost, and the process of transmission is very difficult prevents ubiquitous passive attack on the internet fully.For example, assailant's success " eavesdropping " original watermark that is transmitted or the partial information of original watermark, obtain embedded watermark and further forge watermark passing to the test side by analysis, can make real original watermark can't be used for differentiating copyright, reach and disturb the copyright authentication purposes, explain attack thereby make watermarking algorithm to resist ^[12]

Self-embedding fragile watermark technology ^[13-17]Distinguishing feature be that feature that built-in end extracts initial carrier produces watermark and self-embedding authenticates to reach content integrity to initial carrier.At present, " self-embedding " thought only is applied to the fragile watermark technical field basically.

Comprehensive above the analysis, the present invention attempts " self-embedding " thought of self-embedding Fragile Watermark Algorithm is incorporated into the robust digital watermark field, design has the complete blind Detecting robust watermarking algorithm that the 4th class detects behavior, thereby provide a kind of network image copyright real-time discrimination method, reach the copyright of the digital picture of transmission over networks is differentiated in real time.

List of references

[1] hot spring, grandson's lance cutting edge of a knife or a sword, Wang Shuxun. the notion and the application [J] of zero watermark. electronic letters, vol, 2003,31 (2): 214-216.

[2]Wang?Xiang-yang，Hou?Li-min，and?Wu?Jun.A?feature-based?robust?digital?image?watermarking?against?geometric?attacks[J].Image?and?Vision?Computing，2008，26：980-989.

[3] Niu Shaozhang, button heart Xin, Yang Yixian, Hu Wenqing. Data Hiding Algorithm for Halftone Images [J]. electronic letters, vol, 2004,32 (7): 1180-1183.

[4]Wang?Xiang-yang?and?Cui?Chang-ying.A?novel?image?watermarking?scheme?against?desynchronization?attacks?by?SVR?revision[J].Journal?of?Visual?Communication?and?Image?Representation，2008，19(5)：334-342.

[5] Li Xudong, Zhang Zhen jumps. the digital watermarking algorithm [J] of double-deck division of image and svd. and journal of Zhejiang university (engineering version), 2006,40 (12): 2088-2092.

[6] Yuan Dayang, Xiao Jun, Wang Ying. digital figure watermark algorithm anti-geometry attack robust Journal of Sex Research [J]. electronics and information journal, 2008,30 (5): 1251-1256.

[7] Li Xudong. the spatial domain image digital watermark algorithm [J] of resist geometric attacks. robotization journal, 2008,34 (7): 832-837.

[8] Xu Wenli, Li Lei, Wang Yumin. the robust digital watermark scheme [J] of antinoise, geometric distortion and JPEG compression attack. electronics and information journal, 2008,30 (4): 933-936.

[9] Li Leida, Guo Baolong, table Jin Feng. based on the spatial domain resist geometric attacks image watermark algorithm [J] of odd-even quantization. electronics and information journal, 2009,31 (1): 134-138.

[10]Leida?Li，Jiansheng?Qian，Jeng-Shyang?Pan.Characteristic?region?based?watermark?embedding?with?RST?invariance?and?high?capacity.International?Journal?of?Electronics?and?communications，2011，65：435-442.

[11] Ye Tianyu. the anti-secondary printing-scanning of discrete cosine transform domain robust zero watermarking algorithm [J]. photon journal, 2011,40 (1): 142-148.

[12] Li Qingcheng, Dou Yi. the explanation attack of digital watermarking and relevance feature [J]. computer utility, 2005 (5): 115-117.

[13] Zhang Hongbin, poplar becomes. the self-embedding of image and the detection of altering and recovery algorithms [J]. and electronic letters, vol, 2004,32 (2): 196-199.

[14] and red outstanding person, Zhang Jiashu. based on the piecemeal self-embedding watermarking algorithm [J] of chaos scramble. communication journal, 2006,27 (7): 80-86,93.

[15] Zhang Xianhai, Yang Yongtian. based on the image authentication algorithm research [J] of fragile watermark. electronic letters, vol, 2007,35 (1): 34-39.

[16] Wang Guodong, Liu Fenlin, Liu Yuan, Yao Gang. a kind of Fragile Watermark Algorithm [J] that can distinguish watermark or content tampering. electronic letters, vol, 2008,36 (7): 1349-1354.

[17] and red outstanding person, Zhang Jiashu. watermark information is distorted the self-embedding watermarking algorithm [J] of robust. software journal, 2009,20 (2): 437-450.

Summary of the invention

The objective of the invention is to design the real-time discrimination method of a kind of network image copyright, the copyright of the digital picture of transmission over networks is differentiated in real time.

The real-time discrimination method of a kind of network image copyright comprises following five processes:

A, feature watermark produce;

B, feature watermark are encrypted;

C, encrypted feature watermark self-embedding;

D, feature watermark extract;

E, the blind extraction of authenticating water-mark and deciphering.

Steps A further comprises following content:

A1: with size is that the original image of M * M is divided into size and is the not overlapping sub-piece of m * m;

A2: each sub-piece carry out discrete cosine transform (Discrete Cosine Transformation, DCT), B _i(0,0) represents direct current (Direct Current, DC) coefficient, the B of i sub-piece _Average(0,0) represents the average of all sub-piece DC coefficients,

A3: feature watermark W is by each sub-piece DC coefficient B of contrast _i(0,0) and all sub-piece DC coefficient average B _AverageThe magnitude relationship of (0,0) produces.That is:

if?B _i(0，0)＞B _average(0，0)

w _i＝1；

else

w _i＝0

end

Wherein, w _iRepresent the i bit of W.Know that easily the length of feature watermark W is

Step B further comprises following content:

B1: select a real number as initial value x in (1,1) ₀, select a real number as γ in [0,2], shine upon by Logistic then Carry out iteration and produce chaos random number sequence { x ₁, x ₂, x ₃....With initial value x ₀With preceding two keys of parameter γ as the Logistic mapping.

B2: cast out preceding κ random number of chaos sequence, because initial random number instability.With κ+1 to

Individual random number

By $l_i=\frac{\mathrm{sgn}\left(x_{\mathrm{\κ}+i}\right)+1}{2}$ Two-value turns to GF (2) territory, and { 0,1} sequence L, sgn () are sign function, l _iBe the i bit of L,

Logistic mapping average is 0, so select 0 to carry out binaryzation as threshold value.With three key of parameter κ as the Logistic mapping.

B3: utilize sequence L that feature watermark W is encrypted.Encryption method is:

Wherein

Be XOR,

Be encrypted feature watermark W ^eI bit,

Step C further comprises following content:

C1: original image is divided into size and is the not overlapping sub-piece of m * m;

C2: each sub-piece carries out DCT;

C3: by the following method with the medium and low frequency coefficient of each sub-piece DCT matrix of encrypted feature watermark self-adaptation self-embedding original image:

if $w_i^e=0$

if?B _i(r ₁，s ₁)＜B _i(r ₂，s ₂)

$B_i(r_1,s_1)\↔B_i(r_2,s_2);$

else

if?B _i(r ₁，s ₁)≥B _i(r ₂，s ₂)

$B_i(r_1,s_1)\↔B_i(r_2,s_2);$

end

η _i＝μ×B _i(0，0)；

if?B _i(r ₁，s ₁)＞B _i(r ₂，s ₂)

if(B _i(r ₁，s ₁)-B _i(r ₂，s ₂))＜η _i

$B_i(r_1,s_1)=B_i(r_1,s_1)+\frac{{\mathrm{\η}}_i}{2}$

$B_i(r_2,s_2)=B_i(r_2,s_2)-\frac{{\mathrm{\η}}_i}{2};$

else

if(B _i(r ₂，s ₂)-B _i(r ₁，s ₁))＜η _i

$B_i(r_2,s_2)=B_i(r_2,s_2)+\frac{{\mathrm{\η}}_i}{2}$

$B_i(r_1,s_1)=B_i(r_1,s_1)-\frac{{\mathrm{\η}}_i}{2};$

end

Wherein, symbol The size of two numbers about the expression exchange; μ is

The scale factor that self-adaptation embeds, according to practical application to the requirement of invisibility and the robustness selection of compromising; B _i(r _j, s _j) represent i sub-piece DCT matrix at (r _j, s _j) coefficient located, j=1,2, require to satisfy r ₁=r ₂And s ₁=s ₂Can not set up simultaneously, and r ₁=0 and s ₁=0 can not set up r simultaneously ₂=0 and s ₂=0 can not set up simultaneously.

C4: each sub-piece carries out inverse discrete cosine transform, and (Inverse Discrete Cosine Transformation IDCT), obtains containing watermarking images after the reorganization.

Step D further comprises following content:

D1: attack graph looks like to be divided into size and is the not overlapping sub-piece of m * m;

D2: each sub-piece carries out DCT,

Represent the DC coefficient of i sub-piece,

Represent the average of all sub-piece DC coefficients,

D3: feature watermark W ^aBy contrasting each sub-piece DC coefficient

With all sub-piece DC coefficient averages

Magnitude relationship extract.That is:

if $B_i^a\left(\mathrm{0,0}\right)>B_{\mathrm{average}}^a\left(\mathrm{0,0}\right)$

$w_i^a=1;$

else

$w_i^a=0$

end

Wherein, Represent W ^aIb it.

Step e further comprises following content:

E1: attack graph looks like to be divided into size and is the not overlapping sub-piece of m * m;

E2: each sub-piece carries out DCT;

E3: extract authenticating water-mark W from each sub-piece DCT matrix by the following method ^a':

if?B _i(r ₁，s ₁)＞B _i(r ₂，s ₂)

$w_i^{a^{\′}}=0$

else

$w_i^{a^{\′}}=1$

end

Wherein,

Represent W ^a' ibit.

E4: utilize x ₀Produce Logistic Map Chaotic { x with two keys of γ ₁, x ₂, x ₃..., κ random number before giving up, with κ+1 to the

Individual random number

By Two-value turn to GF (2) territory 0,1} sequence L.

E5: by

To the authenticating water-mark W that extracts ^a' be decrypted, wherein

Representative deciphering back authenticating water-mark W ^a" ibit.

E6: calculate the feature watermark W that extracts after attacking ^aWith the authenticating water-mark W after the deciphering ^a" between the normalization degree of correlation (Normalized Correlation NC) estimates anti-attack robust to judge copyright.The NC value defined is:

$\mathrm{\θ}=\left(\underset{i=1}{\overset{{\left(\frac{M}{m}\right)}^2}{\mathrm{\Σ}}}(w_i^a\×w_i^{a^{\′\′}})\right)/(\sqrt{\underset{i=1}{\overset{{\left(\frac{M}{m}\right)}^2}{\mathrm{\Σ}}}{\left(w_i^a\right)}^2}\×\sqrt{\underset{i=1}{\overset{{\left(\frac{M}{m}\right)}^2}{\mathrm{\Σ}}}{\left(w_i^{a^{\′\′}}\right)}^2}).$

The present invention provides a copyright to differentiate new method in real time for the digital picture of transmission over networks.The present invention at first is divided into original image the piece of non-overlapping copies at built-in end, each sub-piece carries out DCT, magnitude relationship generation feature watermark between DC coefficient by each sub-piece relatively and the average of all sub-piece DC coefficients, feature watermark is carried out the Logistic chaos sequence to be encrypted, adjust the big or small self-embedding encrypted feature watermark of two DCT medium and low frequency coefficients of each sub-piece then, each sub-piece carries out IDCT and obtains containing watermarking images at last.In the test side, extract feature watermark and authenticating water-mark from the attack graph picture respectively.Process and built-in end feature watermark production process that feature watermark extracts are similar, and the blind leaching process of authenticating water-mark is the inverse process of built-in end encrypted feature watermark self-embedding process.Experimental result shows: feature watermark produces algorithm and encrypted feature watermark self-embedding algorithm all has very strong robustness to various attack.Therefore, algorithm of the present invention has the robustness of very strong opposing various attack.The present invention realizes complete blind Detecting in conjunction with watermark of self-embedding encrypted feature and blind extraction authenticating water-mark in fact, realizes the copyright of network image is differentiated in real time.

Description of drawings

Fig. 1 is the algorithm flow chart that feature watermark produces.

Fig. 2 is the algorithm flow chart that feature watermark is encrypted.

Fig. 3 is the algorithm flow chart of encrypted feature watermark self-embedding.

Fig. 4 is the algorithm flow chart that feature watermark extracts.

Fig. 5 is the algorithm flow chart of blind extraction of authenticating water-mark and deciphering.

Fig. 6 is original Lena image, and Fig. 7 is original Barbara image, and Fig. 8 is original Elain image, and Fig. 9 contains watermark Lena image, and Figure 10 contains watermark Barbara image, and Figure 11 contains watermark Elain image.

Embodiment

Below in conjunction with drawings and Examples technical scheme of the present invention is described further.

1 feature watermark produces

Fig. 1 is the algorithm flow chart that feature watermark produces, and comprises following process ^[11]:

A1: with size is that the original image of M * M is divided into size and is the not overlapping sub-piece of m * m;

A2: each sub-piece carries out DCT, B _i(0,0) represents the DC coefficient of i sub-piece, B _Average(0,0) represents the average of all sub-piece DC coefficients,

A3: feature watermark W is by each sub-piece DC coefficient B of contrast _i(0,0) and all sub-piece DC coefficient average B _AverageThe magnitude relationship of (0,0) produces.That is:

if?B _i(0，0)＞B _average(0，0)

w _i＝1；

else

w _i＝0

end (1)

Wherein, w _iRepresent the i bit of W.Know that easily the length of feature watermark W is

Can be known that by above process W produces by the DC coefficient characteristics of extracting original image, this process is not made any modification to original image in fact.Therefore, the present invention calls feature watermark to W.

2 feature watermarks are encrypted

In nonlinear dynamic system, chaos phenomenon is deterministic, similar process at random, does not restrain and not periodically, and is very responsive to initial value.The Logistic chaotic maps is the nonlinear dynamic system of a widespread use, is defined as:

$x_{n+1}=1-\mathrm{\γ}{x}_n^2---\left(2\right)$

x _N+1Span be (1,1), n=0,1,2,3 ...If γ is in [0,2] interior value, then this mapping is in chaos state.The present invention utilizes the initial value susceptibility of Logistic mapping that the feature watermark that produces is encrypted.The process of Logistic mapping encrypting feature watermark is seen Fig. 2, comprises following process:

B1: select a real number as initial value x in (1,1) ₀, select a real number as γ in [0,2], carry out iteration by (2) formula then and produce chaos random number sequence { x ₁, x ₂, x ₃....With initial value x ₀With preceding two keys of parameter γ as the Logistic mapping.

B2: cast out preceding κ random number of chaos sequence, because initial random number instability.With κ+1 to

Individual random number

Through type (3) two-value turn to GF (2) territory 0,1} sequence L, promptly

$l_i=\frac{\mathrm{sgn}\left(x_{\mathrm{\κ}+i}\right)+1}{2}---\left(3\right)$

Wherein sgn () is a sign function, l _iBe the i bit of L, Logistic mapping average is 0, so select 0 to carry out binaryzation as threshold value.With three key of parameter κ as the Logistic mapping.

B3: utilize that binaryzation obtains { 0,1} sequence L encrypts feature watermark W.Encryption method is:

$w_i^e=w_i\⊕l_i---\left(4\right)$

Wherein Be XOR,

Be encrypted feature watermark W ^eIbit,

3 encrypted feature watermark self-embeddings

Fig. 3 is the algorithm flow chart of encrypted feature watermark self-embedding, comprises following process:

C1: original image is divided into size and is the not overlapping sub-piece of m * m;

C2: each sub-piece carries out DCT;

C3: by the following method with each sub-piece DCT matrix of encrypted feature watermark self-adaptation self-embedding original image:

if $w_i^e=0$

if?B _i(r ₁，s ₁)＜B _i(r ₂，s ₂)

$B_i(r_1,s_1)\↔B_i(r_2,s_2);$

else

if?B _i(r ₁，s ₁)≥B _i(r ₂，s ₂)

$B_i(r_1,s_1)\↔B_i(r_2,s_2);$

end

η _i＝μ×B _i(0，0)；

if?B _i(r ₁，s ₁)＞B _i(r ₂，s ₂)

if(B _i(r ₁，s ₁)-B _i(r ₂，s ₂))＜η _i

$B_i(r_1,s_1)=B_i(r_1,s_1)+\frac{{\mathrm{\η}}_i}{2}$

$B_i(r_2,s_2)=B_i(r_2,s_2)-\frac{{\mathrm{\η}}_i}{2};$

else

if(B _i(r ₂，s ₂)-B _i(r ₁，s ₁))＜η _i

$B_i(r_2,s_2)=B_i(r_2,s_2)+\frac{{\mathrm{\η}}_i}{2}$

$B_i(r_1,s_1)=B_i(r_1,s_1)-\frac{{\mathrm{\η}}_i}{2};$

end (5)

Wherein, symbol

The size of two numbers about the expression exchange; μ is

The scale factor that self-adaptation embeds, according to practical application to the requirement of invisibility and the robustness selection of compromising; B _i(r _j, s _j) represent i sub-piece DCT matrix at (r _j, s _j) coefficient located, j=1,2, require to satisfy r ₁=r ₂And s ₁=s ₂Can not set up simultaneously, and r ₁=0 and s ₁=0 can not set up r simultaneously ₂=0 and s ₂=0 can not set up simultaneously.

C4: each sub-piece carries out IDCT, obtains containing watermarking images after the reorganization.

The present invention selects DCT medium and low frequency coefficient to adjust the watermark of self-embedding encrypted feature in (5) formula.This is often not have very strong robustness to attacking because revise the DCT high frequency coefficient.

4 feature watermarks extract

Test side feature watermark leaching process and built-in end feature watermark production process are similar.Fig. 4 is the algorithm flow chart that feature watermark extracts, and comprises following process ^[11]:

D1: attack graph looks like to be divided into size and is the not overlapping sub-piece of m * m;

D2: each sub-piece carries out DCT,

Represent the DC coefficient of i sub-piece, Represent the average of all sub-piece DC coefficients,

D3: feature watermark W ^aBy contrasting each sub-piece DC coefficient

With all sub-piece DC coefficient averages

Magnitude relationship extract.That is:

if $B_i^a\left(\mathrm{0,0}\right)>B_{\mathrm{average}}^a\left(\mathrm{0,0}\right)$

$w_i^a=1;$

else

$w_i^a=0$

end (6)

Wherein,

Represent W ^aIbit.

Blind extraction of 5 authenticating water-marks and deciphering

The leaching process of test side authenticating water-mark is the inverse process of built-in end encrypted feature watermark self-embedding process.Fig. 5 is the algorithm flow chart of blind extraction of authenticating water-mark and deciphering, comprises following process:

E1: attack graph looks like to be divided into size and is the not overlapping sub-piece of m * m;

E2: each sub-piece carries out DCT;

E3: extract authenticating water-mark W from each sub-piece DCT matrix by the following method ^a':

if?B _i(r ₁，s ₁)＞B _i(r ₂，s ₂)

$w_i^{a^{\′}}=0$

else

$w_i^{a^{\′}}=1$

end (7)

Wherein,

Represent W ^a' i bit.

E4: utilize x ₀Produce Logistic Map Chaotic { x with two keys of γ ₁, x ₂, x ₃..., κ random number before giving up, with κ+1 to the

Individual random number

Through type (3) two-value turn to GF (2) territory 0,1} sequence L.

E5: the authenticating water-mark W of the corresponding decryption method of through type (4) to extracting ^a' be decrypted, promptly

$w_i^{a^{\′\′}}=w_i^{a^{\′}}\⊕l_i---\left(8\right)$

Wherein, Representative deciphering back authenticating water-mark W ^a" ibit.

E6: calculate the feature watermark W that extracts after attacking ^aWith the authenticating water-mark W after the deciphering ^a" between NC estimate anti-attack robust to judge copyright.The NC value defined is:

$\mathrm{\θ}=\left(\underset{i=1}{\overset{{\left(\frac{M}{m}\right)}^2}{\mathrm{\Σ}}}(w_i^a\×w_i^{a^{\′\′}})\right)/(\sqrt{\underset{i=1}{\overset{{\left(\frac{M}{m}\right)}^2}{\mathrm{\Σ}}}{\left(w_i^a\right)}^2}\×\sqrt{\underset{i=1}{\overset{{\left(\frac{M}{m}\right)}^2}{\mathrm{\Σ}}}{\left(w_i^{a^{\′\′}}\right)}^2}).---\left(9\right)$

Reach blind extraction when obviously, authenticating water-mark is extracted in the test side.The present invention is W ^a' the reason that is called authenticating water-mark is that the test side is to W ^a' be used to carry out copyright authentication after being decrypted.By the 4th part and the 5th part as can be known, the test side only need utilize the attack graph picture just can extract feature watermark respectively and authenticating water-mark calculates NC, need be by the relevant information of any original image and original watermark.Therefore, the present invention can realize complete blind Detecting, realizes the copyright of network image is differentiated in real time.

Embodiment:

1 experiment parameter explanation

Selecting 256 grades of sizes is that 512 * 512 Lena, Barbara and Elain three width of cloth gray level images are as experimental image, respectively shown in Fig. 6,7 and 8.The size of sub-piece all is 16 * 16, so the length of feature watermark W is 1024bit.Logistic chaotic maps initial value x ₀Value is 0.28, and parameter γ value is 1.5, casts out preceding κ=200 random number during Logistic chaos sequence encrypted feature watermark W.Encrypted feature watermark W ^eEach sub-piece of the original Lena of self-adaptation self-embedding, Barbara and Elain image is in the DCT coefficient of (3,5) and (4,4) position. The scale factor μ value that self-adaptation embeds is 0.028.Obtain contain watermark Lena, Barbara and the Elain image is seen Fig. 9,10 and 11 respectively, and the PSNR between original Lena, Barbara and the Elain image is 35.9595,36.1211 and 35.7637.Therefore, three width of cloth test patterns all had good imperception.

2 anti-attack robust experiments

Weigh the robustness of opposing various attack with the NC value.The middle row of deletion at random refers to move up line by line from bottom's first row beginning of deleted row, and vacant capable completion is black.Downward bias is divided a word with a hyphen at the end of a line and is referred to that entire image moves down several row, above several capable completions black, last several row shift out to be lost.Offset column refers to that entire image moves to right to the right, and several row completions in front are black, and last several row shift out to be lost.Resample and adopt the nearest method of interpolation."/" below contains PSNR between watermarking images and the original image for three width of cloth after attacking in each table, and "/" top is two NC values between the watermark sequence.

(1) the feature watermark W that extracts ^aWith the authenticating water-mark W after the deciphering ^a" between the NC value

Do not exist when attacking, contain the feature watermark W that watermark Lena, Barbara and Elain image extract separately ^aWith the authenticating water-mark W after the deciphering ^a" between the NC value all be 1.0000.Three width of cloth after the attack contain the feature watermark W that watermarking images extracts separately ^aWith the authenticating water-mark W after the deciphering ^a" between the NC value see Table 1.As can be seen from Table 1, algorithm of the present invention all shows very strong robustness to various attack.

(2) the feature watermark W that extracts ^aAnd the NC value between the primitive character watermark W

Do not exist when attacking, contain the feature watermark W that watermark Lena, Barbara and Elain image extract separately ^aAnd the NC value between the primitive character watermark W all is 1.0000.Three width of cloth after the attack contain the feature watermark W that watermarking images extracts separately ^aAnd the NC value between the primitive character watermark W sees Table 2.Therefore, feature watermark generation algorithm of the present invention all has very strong robustness to various attack.In addition, the data of contrast table 1 and table 2 can find that the anti-attack robust that feature watermark produces algorithm is better than algorithm of the present invention slightly.

(3) the authenticating water-mark W after the deciphering ^a" and the NC value between the primitive character watermark W

Do not exist when attacking, contain the authenticating water-mark W after watermark Lena, Barbara and Elain image are deciphered separately ^a" and the NC value between the primitive character watermark W all is 1.0000.The authenticating water-mark W that three width of cloth after the attack contain watermarking images after deciphering separately ^a" and the NC value between the primitive character watermark W sees Table 3.Therefore, encrypted feature watermark self-embedding algorithm of the present invention all has very strong robustness to various attack.In addition, the data of contrast table 1 and table 3 can find that the anti-attack robust of encrypted feature watermark self-embedding algorithm is better than algorithm of the present invention slightly.

Three width of cloth after table 1 is attacked contain the feature watermark W that watermarking images extracts separately ^aWith the authenticating water-mark W after the deciphering ^a" between the NC value

3 analyze and discuss

Algorithm of the present invention can be realized complete blind Detecting, its reason is to reach simultaneously following 2 points: (1) produces feature watermark from original image by feature extraction, original image is arrived in self-embedding after encrypting, rather than extra watermark form is embedded into original image; (2) encrypted feature watermark self-embedding algorithm itself just can reach blind extraction authenticating water-mark in the test side.

Three width of cloth after table 2 is attacked contain the feature watermark W that watermarking images extracts separately ^aAnd the NC value between the primitive character watermark W

The authenticating water-mark W that three width of cloth after table 3 is attacked contain watermarking images after deciphering separately ^a" and the NC value between the primitive character watermark W

Algorithm of the present invention has strong robustness to various attack, and its reason is to reach simultaneously following 2 points: (1) feature watermark produces algorithm itself and just has very strong anti-attack robust, can come as seen from Table 2.(2) encrypted feature watermark self-embedding algorithm itself just has very strong anti-attack robust, can come as seen from Table 3.In addition, the anti-attack robust of algorithm of the present invention is worse than the reason that feature watermark produces algorithm and encrypted feature watermark self-embedding algorithm slightly and is that encrypted feature watermark self-embedding meeting has certain influence to the DC coefficient that produces feature watermark.

4 sum up

Present blind robust watermarking algorithm still need be by the partial information of original image or original watermark in the test side, thereby can't really realize complete blind Detecting, thereby can't reach the copyright of the digital picture of transmission over networks is differentiated in real time.At this problem, the present invention is incorporated into the robust watermarking field with " self-embedding " thought of Fragile Watermark Algorithm, proposes the complete blind Detecting robust watermarking of a kind of self-embedding algorithm, and realization is differentiated in real time to the copyright of the digital picture of transmission over networks.Algorithm of the present invention possesses following 4 characteristics simultaneously:

(1) " self-embedding " characteristic: produce feature watermark by feature extracting method from original image, again encrypted feature watermark self-embedding is contained watermarking images to the original image generation.

(2) " complete blind extraction " characteristic: neither need be during detection by any information relevant with initial carrier, again need be by any information relevant with original watermark.In the test side, at first extract feature watermark from attacking the imagery exploitation feature extracting method identical with built-in end; Blindly extract authenticating water-mark from the inverse process of attacking imagery exploitation built-in end encrypted feature watermark self-embedding algorithm then; Calculate the feature watermark of extraction and the NC between the authenticating water-mark at last and judge copyright.Therefore, the test side only need utilize the attack graph picture.

(3) robustness: feature watermark produces algorithm and encrypted feature watermark self-embedding algorithm all has very strong robustness to various attack, and further, algorithm of the present invention has the robustness of very strong opposing various attack.

(4) security: will carry out the Logistic chaos sequence before the feature watermark self-embedding original image earlier and encrypt, not know that the assailant of encryption key can't accurately decrypt feature watermark.

The complete blind Detecting robust watermarking of the self-embedding technology that the present invention proposes has very important Research Significance.This is embodied in:

(1) saves transmission cost and carrying cost.Any information that built-in end need not to transmit initial carrier and original watermark to the test side (or third party notarize center) store, thereby save transmission cost and carrying cost.

(2) blocking-up is explained and is attacked.Any information that built-in end need not to transmit original watermark is to the test side (or third party notarize center), can effectively block to explain and attack.

Therefore, the complete blind Detecting robust watermarking of the self-embedding algorithm that the present invention proposes provides a kind of network image copyright real-time discrimination method, and realization is differentiated in real time to the copyright of the digital picture of transmission over networks.

Claims (6)

1. real-time discrimination method of network image copyright is differentiated in real time to the copyright of the digital picture of transmission over networks, comprises following five processes:

A, feature watermark produce;

B, feature watermark are encrypted;

C, encrypted feature watermark self-embedding;

D, feature watermark extract;

E, the blind extraction of authenticating water-mark and deciphering.

2. the real-time discrimination method of a kind of network image copyright according to claim 1, steps A further comprises following content:

A1: with size is that the original image of M * M is divided into size and is the not overlapping sub-piece of m * m;

A2: each sub-piece carry out discrete cosine transform (Discrete Cosine Transformation, DCT), B _i(0,0) represents direct current (Direct Current, DC) coefficient, the B of i sub-piece _Average(0,0) represents the average of all sub-piece DC coefficients,

A3: feature watermark W is by each sub-piece DC coefficient B of contrast _i(0,0) and all sub-piece DC coefficient average B _AverageThe magnitude relationship of (0,0) produces.That is:

if?B _i(0，0)＞B _average(0，0)

w _i＝1；

else

w _i＝0

end

Wherein, w _iRepresent the i bit of W.Know that easily the length of feature watermark W is

3. the real-time discrimination method of a kind of network image copyright according to claim 1, step B further comprises following content:

B1: select a real number as initial value x in (1,1) ₀, select a real number as γ in [0,2], shine upon by Logistic then Carry out iteration and produce chaos random number sequence { x ₁, x ₂, x ₃....With initial value x ₀With preceding two keys of parameter γ as the Logistic mapping.

B2: cast out preceding κ random number of chaos sequence, because initial random number instability.With κ+1 to

Individual random number

By Two-value turns to GF (2) territory, and { 0,1} sequence L, sgn () are sign function, l _iBe the ibit of L, Logistic mapping average is 0, so select 0 to carry out binaryzation as threshold value.With three key of parameter κ as the Logistic mapping.

B3: utilize sequence L that feature watermark W is encrypted.Encryption method is:

Wherein

Be XOR,

Be encrypted feature watermark W ^eIbit,

4. the real-time discrimination method of a kind of network image copyright according to claim 1, step C further comprises following content:

C1: original image is divided into size and is the not overlapping sub-piece of m * m;

C2: each sub-piece carries out DCT;

C3: by the following method with the medium and low frequency coefficient of each sub-piece DCT matrix of encrypted feature watermark self-adaptation self-embedding original image:

if $w_i^e=0$

if?B _i(r ₁，s ₁)＜B _i(r ₂，s ₂)

$B_i(r_1,s_1)\↔B_i(r_2,s_2);$

else

if?B _i(r ₁，s ₁)≥B _i(r ₂，s ₂)

$B_i(r_1,s_1)\↔B_i(r_2,s_2);$

end

η _i＝μ×B _i(0，0)；

if?B _i(r ₁，s ₁)＞B _i(r ₂，s ₂)

if(B _i(r ₁，s ₁)-B _i(r ₂，s ₂))＜η _i

$B_i(r_1,s_1)=B_i(r_1,s_1)+\frac{{\mathrm{\η}}_i}{2}$

$B_i(r_2,s_2)=B_i(r_2,s_2)-\frac{{\mathrm{\η}}_i}{2};$

else

if(B _i(r ₂，s ₂)-B _i(r ₁，s ₁))＜η _i

$B_i(r_2,s_2)=B_i(r_2,s_2)+\frac{{\mathrm{\η}}_i}{2}$

$B_i(r_1,s_1)=B_i(r_1,s_1)-\frac{{\mathrm{\η}}_i}{2};$

end

Wherein, symbol

The size of two numbers about the expression exchange; μ is

The scale factor that self-adaptation embeds, according to practical application to the requirement of invisibility and the robustness selection of compromising; B _i(r _j, s _j) represent i sub-piece DCT matrix at (r _j, s _j) coefficient located, j=1,2, require to satisfy r ₁=r ₂And s ₁=s ₂Can not set up simultaneously, and r ₁=0 and s ₁=0 can not set up r simultaneously ₂=0 and s ₂=0 can not set up simultaneously.

C4: each sub-piece carries out inverse discrete cosine transform, and (Inverse Discrete Cosine Transformation IDCT), obtains containing watermarking images after the reorganization.

5. the real-time discrimination method of a kind of network image copyright according to claim 1, step D further comprises following content:

D1: attack graph looks like to be divided into size and is the not overlapping sub-piece of m * m;

D2: each sub-piece carries out DCT,

Represent the DC coefficient of i sub-piece, Represent the average of all sub-piece DC coefficients,

D3: feature watermark W ^aBy contrasting each sub-piece DC coefficient

With all sub-piece DC coefficient averages

Magnitude relationship extract.That is:

if $B_i^a\left(\mathrm{0,0}\right)>B_{\mathrm{average}}^a\left(\mathrm{0,0}\right)$

$w_i^a=1;$

else

$w_i^a=0$

end

Wherein,

Represent W ^aIbit.

6. the real-time discrimination method of a kind of network image copyright according to claim 1, step e further comprises following content:

E1: attack graph looks like to be divided into size and is the not overlapping sub-piece of m * m;

E2: each sub-piece carries out DCT;

E3: extract authenticating water-mark W from each sub-piece DCT matrix by the following method ^a':

if?B _i(r ₁，s ₁)＞B _i(r ₂，s ₂)

$w_i^{a^{\′}}=0$

else

$w_i^{a^{\′}}=1$

end

Wherein,

Represent W ^a' ibit.

E4: utilize x ₀Produce Logistic Map Chaotic { x with two keys of γ ₁, x ₂, x ₃..., κ random number before giving up, with κ+1 to the

Individual random number

By

Two-value turn to GF (2) territory 0,1} sequence L.

E5: by

To the authenticating water-mark W that extracts ^a' be decrypted, wherein

Representative deciphering back authenticating water-mark W ^a" ibit.

E6: calculate the feature watermark W that extracts after attacking ^aWith the authenticating water-mark W after the deciphering ^a" between the normalization degree of correlation (Normalized Correlation NC) estimates anti-attack robust to judge copyright.The NC value defined is:

$\mathrm{\θ}=\left(\underset{i=1}{\overset{{\left(\frac{M}{m}\right)}^2}{\mathrm{\Σ}}}(w_i^a\×w_i^{a^{\′\′}})\right)/(\sqrt{\underset{i=1}{\overset{{\left(\frac{M}{m}\right)}^2}{\mathrm{\Σ}}}{\left(w_i^a\right)}^2}\×\sqrt{\underset{i=1}{\overset{{\left(\frac{M}{m}\right)}^2}{\mathrm{\Σ}}}{\left(w_i^{a^{\′\′}}\right)}^2}).$

Images