CN1584929A - Synchronous self-adaptable watermark method based on image continuity - Google Patents

Abstract

An adaptive synchronous watermark method based on image connectivity includes converting original grey scale image to be two-valued image, obtaining connecting through image, mapping it into grey scale image, selecting relatively big gravity as reference point to embed watermark and ensuring adaptive synchronization of watermark by utilizing unchanged topology and stable gravity of image.

Description

Self-adaptation synchronization watermarking method based on image connectivity

Technical field

The present invention relates to a kind of self-adaptation synchronization watermarking method based on image connectivity, is a kind of copyright protection that is applicable to the digital picture product, and the robustness image watermark method of infringement proof is provided.

Background technology

At present, in the digital watermarking research field, the robust watermarking system is that people study the most a kind of watermark, this be with practical application in the needs of copyright protection inseparable.At present, in the research of robust watermarking, from the relevant document of publishing, when many robustness water mark methods are handled attack at noise-like signals such as opposing Gaussian noise, JPEG compression and filtering, shown the robustness that it is superior, but such as powerless aspect the geometric attacks such as rotation, convergent-divergent, translation, shearing, bird caging, its reason is that watermark need rely on extra nonsynchronous information when hiding anti-.If under geometric attack, designed digital watermarking system can not detect or extract watermark, and then the usable range of this watermaking system will be restricted, and can not be referred to as the watermaking system of a perfect performance.Therefore, research can resist noise like to attack, and well behaved practicality watermaking system that again can resist geometric attacks is the needs in the present practical application, is again the important development direction and the difficult point of watermark research.Along with the development of digital watermark technology, people's expectation finds a kind of algorithm can resist noise like to attack, again can resist geometric attacks.In the digital watermarking research work of recent years, also occurred much can resist geometric attacks robust watermarking, as document " Geometrically Invariant WatermarkingBased on Gravity Center; " (K.Ding, C.He, L.G Jiang, H.X.Wang, IEEE Trans.Fundamental, vol.E87-A, no.2 pp.513-515, Feb.2004.) a kind of anti-geometry attack robust watermark based on center of gravity has been proposed, though contain under watermarking images what anglec of rotation in office and the zoom factor and still shown very strong robustness, but the reference point that is used for the watermark embedding is based on the center of gravity of entire image, therefore to operations such as local geometric warping and shearings, this watermarking project lost efficacy, and the robust watermarking of the resist geometric attacks that other document proposed, as document " A feature-based robust digital image watermarking scheme; " (C.W.Tang and H.M.Hang, IEEE Trans.Signal Processing, vol.51, no.4, pp.950-959, April2003.) watermarking project based on unique point of Ti Chuing, often because it is not very stable being used for the reference point (unique point) of watermark embedding, it is not very desirable causing watermark robustness.In the research process of digital watermarking, people attempt to seek some all the time than more comprehensive watermarking project, both robustness ideals aspect resist geometric attacks, again in anti-noise like attack robustness ideal, because the attack in the reality should be diversified, that have even out of people's expectation, therefore seeking such method has its realistic meaning.

Summary of the invention

The objective of the invention is at the deficiencies in the prior art, a kind of self-adaptation synchronization watermarking method based on image connectivity is provided, be used for the reference point stability height that watermark embeds, make not only under common image is handled, the robustness of watermark is stronger, and the overall situation and local geometric attack are all presented stronger robustness.

For realizing such purpose, the present invention has designed a kind of new spatial domain robustness New Scheme of Image Watermarking, original-gray image at first changes into bianry image, connectedness according to image obtains being communicated with object, and with these coordinate Mapping that are communicated with objects in gray level image, selects the reference point of the center of gravity of bigger connection object as the watermark embedding, utilize the connectedness of image to have topological invariance, and the center of gravity of image is more stable, guarantees that the self-adaptation of watermark is synchronous.

Method of the present invention mainly comprises watermark embedding and watermark extracting, and concrete steps are as follows:

1, watermark embed process:

At first original-gray image is done analytic trnasformation, to change into bianry image through the image after the analytic trnasformation, be communicated with according to 4-then or 8-connection pattern obtains the connection object of image, and they are sorted from big to small, in order to obtain the center of gravity of stable connection object, select maximum P to be communicated with its center of gravity of calculation and object, and as the reference point of watermark embedding.The center of entire image and the line of each reference point are as start ray, angle of the every rotation of this start ray will form one fan-shaped, this fan-shaped Rotate 180 degree forms former fan-shaped anti-fan-shaped, the watermark information of same version just be embedded in each fan-shaped with it anti-fan-shaped in.Through central point, with M-1 bar line equably each fan-shaped M equal portions that are divided into, in order to reach the purpose of watermark synchronization, from start ray, on every line, embed N watermark bit equably, embed point, hide each watermark bit with the method for odd-even quantization in each watermark.

Because embedded location is by the center of gravity decision that is communicated with object, start ray is along with synchronously rotation of entire image, convergent-divergent and translation, so image is through behind these geometric attacks, the embedding point of watermark can also correctly be detected, and hides watermark bit by quantization method and can tolerate that to a certain degree noise pollution and JPEG compress.

2, watermark extraction process:

At first, carry out analytic trnasformation to containing watermarking images, then with gray level image being changed into bianry image with the same method of watermark embed process, and obtain being communicated with object according to the connection pattern identical with watermark embed process, according to being communicated with the number of pixels that object comprised they are sorted from big to small, get P maximum its center of gravity of connection calculation and object, obtain being used for the reference point that watermark embeds, the same with watermark embed process, utilize these reference point to form and be embedded with the fan-shaped of watermark information, this fan-shaped Rotate 180 ° forms that it is anti-fan-shaped, fan-shaped and the anti-embedding point that finds watermark in fan-shaped embeds point according to the gray scale quantized value in each watermark and extracts a watermark bit at each.

The present invention is based on the topological property of image connectivity, the self-adaptation that has realized watermark is synchronous, be used for the reference point stability height that watermark embeds, make not only under common image is handled, the robustness of watermark is stronger, and all can present stronger robustness to the overall situation and local geometric attack, and the invisibility of watermark is good.Be used to change into the analytical function of bianry image parameter, form fan-shaped parameters such as the number of degrees and can be used as the key of watermaking system, thereby make the center of gravity that is communicated with object, and watermark to embed point be secret, therefore guaranteed the security of watermark.

Description of drawings

Fig. 1 is original-gray image in the embodiment of the invention and the bianry image through transforming after the analytic trnasformation.

Among Fig. 1, a is that original ' Pepper ' figure, b are that original ' Lena ' figure, c are original ' Baboon ' figure, and e is ' Pepper ' binary map, and f is ' Lena ' binary map, and g is ' Baboon ' binary map.

Fig. 2 is the fan-shaped and anti-fan-shaped synoptic diagram that is divided into the M equal portions.

Fig. 3 is that the watermark of ' Pepper ' figure embeds point.

Fig. 4 is the stable synoptic diagram of center of gravity under rotation, convergent-divergent and common image are handled.

Among Fig. 4, a is the stability of 30 ° of tailheaviness of image rotation, b is the stability that image amplifies 2 times of tailheaviness, and to be image equaled the stability that 46 white Gaussian noise (AWGN) pollutes tailheaviness by variance to c, and d is that image is by the stability of JPEG compression (quality factor QF=50%) tailheaviness.

Fig. 5 is a watermark detection accuracy comparison diagram.

Among Fig. 5, a is a geometric attack, and b is that common image is handled.

Fig. 6 is under different geometric attacks, extracts the similarity coefficient of watermark.

Among Fig. 6, a is the average similarity coefficient under the different zoom factor, and b is the average similarity coefficient under the different rotary angle

Fig. 7 is the influence of Δ to watermark invisibility and robustness.

Among Fig. 7, a is the Y-PSNR of " Pepper " under the different Δs, " Lena ", " Batoon ", and b is 10 ° of image rotations, amplifies 2 times and be under 90 JPEG compresses in quality factor, the average similarity coefficient during different Δ

Fig. 8 contains watermarking images to be subjected to different local geometric attacks.

Among Fig. 8, a is that the connection object that contains watermark is hit in local assault, and b is that the connection object that contains watermark is not hit in local assault.

Embodiment

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

1, watermark embeds:

In theory, the connection object has geometric invariance to the topological transformation of image, in order to obtain the connection object of image, at first uses threshold value T ₁(x y) changes into bianry image f with gray level image f _b(x, y):

Here threshold value T ₁∈ [0,1], T ₁Selection will influence the number of image connectivity object.If directly from original-gray image, generate and be communicated with object, the connection object that then obtains may comprise very big or very little object, for very large object, then can cause the shortcoming of in very big zone, having to a reference point, and for very little object, the reference point that then obtains is not very stable.Therefore, before original-gray image was changed into bianry image, elder generation was to each gray-scale value L of image _iDo following analytic trnasformation:

$L_i^{\′}={\mathrm{aL}}_i^2+{\mathrm{bL}}_i+c,(i=\mathrm{1,2},...,L)$

Here a, b, c ∈ R are the adjustable parameters of control gray-scale value conversion, like this, generate the connection object of size appropriateness by these parameter values of suitable adjusting.Through experiment, find establishing a=0.003, b=0.002, c=0, T ₁=0.5 o'clock, then the connection of the bianry image gained of Sheng Chenging was more moderate to liking.Original-gray image ' Pepper ', ' Lena ', ' Baboon ' are respectively shown in Fig. 1 (a) and (b), (c), the bianry image that transforms after the process analytic trnasformation is respectively shown in Fig. 1 (e), (f), (g), as seen the size that is communicated with object is more even, and this will help the embedding of watermark.After bianry image generates, can obtain the connection object of image according to certain connection pattern.If Ω _i(i=1,2 ..., u) u of expression is communicated with object, N (Ω _i) expression Ω _iThe number of the pixel that is comprised in the individual object, T ₂The threshold value of representing another definition, and $\underset{i}{\min }\left\{N\left({\mathrm{\&Omega;}}_i\right)\right\}\&le;{\mathrm{<figure-callout\; id="2"\; label="T"\; filenames="A20041002503400122.png,A20041002503400132.png"\; state="\{\{state\}\}">T</figure-callout>}}_2\&le;\underset{i}{\max }\left\{N\left({\mathrm{\&Omega;}}_i\right)\right\},$ In order to obtain the center of gravity of stable connection object, select those to satisfy N (Ω _i) 〉=T ₂Bigger its center of gravity of connection calculation and object, and the reference point that embeds as watermark.Suppose to have selected v to be communicated with object, their center of gravity is designated as G ₁, G ₂..., G _v, the centre coordinate of establishing entire image is O (x ₀, y ₀), upper left corner coordinate is (x ₁, y ₁), lower right corner coordinate is (x ₂, y ₂), then

The mode that watermark embeds such as Fig. 2 (a) description, establish $\stackrel{\&RightArrow;}{{\mathrm{OG}}_i}(i=\mathrm{1,2},...,v)$ Be start ray, α angle of every rotation will form one fan-shaped, it is fan-shaped that this fan-shaped Rotate 180 degree forms, be referred to as former fan-shaped anti-fan-shaped, same watermark information just be embedded in each fan-shaped with it anti-fan-shaped in.Through the center O point, equably each fan-shaped M equal portions that are divided into, the length of every straight line is defined as reference length L with M-1 bar straight line _RefIn order to reach the purpose of watermark synchronization, from

Initial, on every line, embed N watermark bit equably, promptly per two distances that embed point are L on every line _Ref/ N, like this, each fan-shaped and its anti-fan-shaped (M+1) N watermark bit that can embed altogether, Fig. 2 (b) is to be the embedding point of example watermark with ' Pepper ' image.Because being embedded in each fan-shaped is the same with its anti-fan-shaped watermark version, so to less fan-shaped of area, the distribution of watermark information is denser, and to bigger fan-shaped of area, the distribution of watermark information is just more sparse.In order to strengthen the robustness of watermark, with the picture centre be the center of circle radius less than r ( $r<\frac{1}{6}R,$ R is the length of side of image) circular area will be not used in embed watermark, it is denser that its reason is that the watermark of this circular area embeds point, causes the accuracy of watermark detection to descend easily.

Embed point in each watermark, with odd-even quantization hide watermark bit w (i) ∈ 0,1}, (i=1,2 ..., N _w):

When w (i)=0, the gray-scale value after the quantification is

When w (i)=1, the gray-scale value after the quantification is

Here Integer, f are taken off in expression _d=sgn (f) (| f|mod1), sgn () represents sign function, Δ＞0 expression quantization step, control watermark strength.Because embedded location is by the center of gravity decision that is communicated with object, Along with synchronously rotation of entire image, convergent-divergent and translation, so image is through behind the geometric attack, the embedding point of watermark can also correctly be detected, and hide watermark bit by quantization method and can tolerate that to a certain degree noise pollution and JPEG compress, visible watermarking project of the present invention all has stronger robustness to geometric attack and noise like attack.

2, watermark extracting:

At first, carry out analytic trnasformation, then with following the same threshold value T of watermark embed process to containing watermarking images ₁Change into bianry image, and obtain being communicated with object, based on threshold value T according to connection pattern (8-is communicated with or 4-is communicated with) ₂Obtain being embedded with the center of gravity of the bigger connection object of watermark information.The same with watermark embed process, utilize these centers of gravity to form to be embedded with the fan-shaped of watermark information anti-fan-shaped with it, at each fan-shaped and its anti-embedding point that can find watermark in fan-shaped, like this, no matter how image to rotate, convergent-divergent and translation, watermark always can synchronously be extracted, and embeds point in each watermark, and watermark bit is extracted as follows:

Here f _wGray-scale value behind the expression embed watermark.By calculating as can be known, when the variation range of gray-scale value [Δ/2, Δ/2) when interior, watermark bit can correctly be extracted.Δ is big more, and the visual quality that then contains watermark will descend, but robustness strengthens, thus Δ should be between the robustness of watermark and invisibility compromise value.

In an embodiment of the present invention: be example with gray scale ' Pepper ' image at first, checking is used for the stability of the reference point of watermark synchronization.Bianry image shown in Fig. 1 (e), threshold value T ₂Get 200, obtain 10 bigger connection objects.Under geometry and noise like attack, the stability of the center of gravity of these 10 bigger connection objects as shown in Figure 4, the actual centre of gravity place of ' * ' expression wherein, and ' zero ' be illustrated in the theoretic centre of gravity place under the watermark synchronization, in case the two coincidence, then the watermark success synchronously.Compare with the watermarking project based on unique point in the prior art, the reference point that is used for watermark synchronization in the watermarking project of the present invention is not only stable, and handles easily.Based on the watermarking project of High Order Moment because its unique point is stable inadequately, variation to image texture is relatively more responsive, and the connectedness that is based on the image inherent feature synchronously of watermarking project of the present invention realizes, center of gravity has the insensitivity to geometric transformation and noise pollution in addition, and therefore watermarking project of the present invention is realizing that the self-adaptation synchronization aspects has more advantage.Line to each reference point and picture centre, every 5 the degree obtain one fan-shaped and anti-fan-shaped, each is fan-shaped and anti-fan-shapedly be divided into three parts, the line that then can be used for embed watermark in each is fan-shaped has four, on every line, embed 34 watermark bit equably, fan-shaped and the anti-watermark information that embeds 136 bits in fan-shaped altogether then at each, as shown in Figure 3.When watermark bit embedded, the quantization step Δ was set to 16, and the PSNR that contains watermarking images is 50.18dB, and the invisibility of visible watermark is fine.

In case it is under attack to contain watermarking images, possible some watermark can not detect, establish from each fan-shaped and anti-fan-shaped the watermark number of extraction be N _Total, the number that correctly detects watermark is N _Correct, then the accuracy of watermark detection (RC) is defined as

RC＝N _correct/N _total

Table 1 and table 2 have been listed several frequently seen geometric attack type and Flame Image Process type respectively, Fig. 5 has compared under the listed different attack type of table 1 and table 2, watermarking project of the present invention and difference based on the watermark detection accuracy of the watermarking project of unique point, as seen its difference major part is greater than zero, this explanation the solution of the present invention watermark detection accuracy is higher, thereby the robustness of watermarking project of the present invention is stronger as can be known, and this benefits from the topological property of connection object and the high stability of centre of gravity place thereof.Extract watermark w ' (i) ∈ 0,1}, (i=1,2 ..., N _w) performance estimate that by similarity coefficient ρ it is defined as follows:

$\mathrm{\&rho;}=\frac{1}{N_w}\underset{i=1}{\overset{N_w}{\mathrm{\&Sigma;}}}|w\left(i\right)-w^{\&prime;}\left(i\right)|$

ρ is big more, and then the robustness of watermark is strong more.If ρ ₁, ρ ₂..., ρ _NtotalExpression N _TotalThe similarity coefficient of the watermark of individual extraction in order to estimate the performance of watermarking project of the present invention, calculates the mean value of these similarity coefficients

$E\left(\mathrm{\&rho;}\right)=\frac{1}{N_{\mathrm{total}}}\underset{j=1}{\overset{N_{\mathrm{total}}}{\mathrm{\&Sigma;}}}{\mathrm{\&rho;}}_j$

When containing watermarking images at different zoom factors, the different swing geometry of numbers is attacked down, from 7 fan-shaped and anti-fan-shaped extraction 7 watermarks average similarity coefficient as shown in Figure 6, as seen watermarking project of the present invention has stronger robustness aspect resist geometric attacks, from Fig. 6, also can find out, when image dwindles or shears, the robustness of watermark is more relatively poor, this is because these operations may make the connection object that is embedded with watermark disappear or watermark embeds the some disappearance, cause the stability decreases of reference point, therefore, in Fig. 6 (a), zoom factor is that 0.5 o'clock average similarity coefficient is less.When containing the watermarking images rotation, be communicated with also synchronously rotation of object, so the also synchronously rotation of its center of gravity, synchronous adaptively thereby the detection of watermark can realize.In watermarking project based on unique point, only rotate 5 degree, ' Pepper ' contains watermarking images just can not detect watermark, and watermarking project of the present invention all is effective to the rotation of any angle, therefore, watermarking project proposed by the invention has improved the robustness of watermark resist geometric attacks.Embed point in each watermark, watermark strength is by quantization step Δ decision, Δ to the influence of watermark invisibility and robustness as shown in Figure 7, as seen, Δ is big more, the general trend of watermark robustness is to strengthen, and invisibility is poor more, can find out from Fig. 7 (a), when Δ increases to 50, also all more than 39dB, this illustrates watermarking project of the present invention when guaranteeing than strong robustness to the PSNR that contains watermarking images, and invisibility can not descend very big yet.Can find out from Fig. 7 (b), the variation of Δ value is smaller to the influence of the attack of geometry and common signal processing, this is because rotation and convergent-divergent influence very little cause to the center of gravity of bigger connection object, therefore, watermarking project of the present invention especially aspect resist geometric attacks robustness more outstanding.

From the telescopiny of watermark as can be known; watermarking project of the present invention can be resisted the local geometric distortion; as long as those connection objects that are used to the watermark embedding are not damaged; their center of gravity just can be accurately positioned so; thereby can find the embedding point of watermark; even some regional area is by geometric warping; but watermark information may can not extracted in the zone that only is twisted; but other zone still can extract watermark; Fig. 8 (a) has shown that local assault hit the situation of the connection object that is embedded with watermark; and having shown local assault, Fig. 8 (b) do not hit the situation of the connection object that is embedded with watermark; table 3 has been listed the similarity coefficient that extracts watermark under these two kinds of attacks; object 5 as can be seen; 6; 7 similarity coefficient is less; this illustrates that these three objects are just in time attacked; but from the similarity coefficient of other object, still can detect watermark, thereby realize the function of copyright protection.

Table 1 geometric attack type

??No.	Attack type	??No.	Attack type
				??1	The center cuts off 5%	??11	The center cuts off 5%+JPEG70
??2	The center cuts off 10%	??12	The center cuts off 10%+JPEG70
				??3	Shearing-x-1%-y-1%	??13	Rotate 1 °+shearing 5%+ and amplify 2 times
??4	Shearing-x-0%-y-5%	??14	Rotate 1 °+shearing 5%
				??5	Shearing-x-5%-y-5%	??15	Rotate 2 °+shearing 5%
??6	Shearing-x-1%-y-1%+JPEG70	??16	Rotate 5 °+shearing 5%
				??7	Shearing-x-0%-y-5%+JPEG70	??17	Rotate 1 °+shearing 5%+ and amplify 2 times+JPEG70
??8	Shearing-x-5%-y-5%+JPEG70	??18	Rotate 1 °+shearing 5%+JPEG70
				??9	Remove 1 row, 5 row+JPEG70	??19	Rotate 2 °+shearing 5%+JPEG70
??10	Remove 5 row, 17 row+JPEG70	??20	Rotate 5 °+shearing 5%+JPEG70

Table 2 common image is handled type

??No.	Attack type	??No.	Attack type
				??1	Contain watermarking images	??7	??JPEG30
??2	The even noise of additivity (scale=0.1)	??8	Medium filtering 2 * 2+JPEG90
				??3	The even noise of additivity (scale=0.15)	??9	Medium filtering 3 * 3+JPEG90
??4	The even noise of additivity (scale=0.2)	??10	Medium filtering 2 * 2
				??5	JPEG80	??11	Medium filtering 3 * 3
??6	JPEG60	??12	Gray balance

Table 3 extracts the similarity coefficient of watermark under local geometric is attacked

Claims (1)

1, a kind of self-adaptation synchronization watermarking method based on image connectivity is characterized in that comprising following concrete steps:

1) watermark embed process: at first original-gray image is done analytic trnasformation, to change into bianry image through the image after the analytic trnasformation, be communicated with according to 4-then or 8-connection pattern obtains the connection object of image, and they are sorted from big to small, in order to obtain the center of gravity of stable connection object, select P maximum its center of gravity of connection calculation and object, and the reference point that embeds as watermark, the center of entire image and the line of each reference point are as start ray, angle of the every rotation of this start ray will form one fan-shaped, this fan-shaped Rotate 180 degree forms former fan-shaped anti-fan-shaped, the watermark information of same version just be embedded in each fan-shaped with it anti-fan-shaped in, through central point, with M-1 bar line equably each fan-shaped M equal portions that are divided into, in order to reach the purpose of watermark synchronization, from start ray, on every line, embed N watermark bit equably, embed point, hide each watermark bit with the method for odd-even quantization in each watermark;

2) watermark extraction process: at first carry out analytic trnasformation to containing watermarking images, then with gray level image being changed into bianry image with the same method of watermark embed process, and obtain being communicated with object according to the connection pattern identical with watermark embed process, according to being communicated with the number of pixels that object comprised they are sorted from big to small, get P maximum its center of gravity of connection calculation and object, obtain being used for the reference point that watermark embeds, the same with watermark embed process, utilize these reference point to form and be embedded with the fan-shaped of watermark information, this fan-shaped Rotate 180 ° forms that it is anti-fan-shaped, fan-shaped and the anti-embedding point that finds watermark in fan-shaped embeds point according to the gray scale quantized value in each watermark and extracts a watermark bit at each.

Images