CN101159059B - Self-adapting attack method for discrete cosine transform domain digital watermarking system - Google Patents

Abstract

The invention discloses a self-adaptable attack method for a discrete cosine transform domain digital watermarking system and belongs to the technical field of digital watermarking. The method includes: acquiring a digital watermarking image; dividing the digital watermarking image into a plurality of sub-blocks each having n multiplied by n pixels, and acquiring human viral system classificationresults on the sub-blocks to obtain superposition strength of watermarking components; performing one attack on each sub-block with a periodical transformation matrix to obtain the attacked sub-blocks matrix; calculating signal-to-noise ratio of the peak value, and performing a second attack or outputting the image. The periodical transformation matrix should satisfy the two following formulas: (1) Phi<m>(Xn)=A<m>Xn(modN), wherein A is a periodical transformation matrix, A=(aij)n multiplied by n, aij is an integer, Phi<m>(Xn) and Xn are original divided sub-blocks matrixes, Xn=(x1 to xn)<T>, Xn'=(x1' to xn')<T>, x1 to xn belong to {0, 1, ellipsis, N-1}, and N is the transformation order; and (2) A<mN>(modN)=En, wherein mN is the period for transformation of Xn'=Phi(Xn) and En is an n-ordered identity matrix. The inventive method can detect robustness of the present watermarking system, and as the same method is adopted for attack and recovery, calculation is simple and effective. The invention can promote development of the digital watermarking system.

Description

A kind of self-adapting attack method to the discrete cosine transform domain digital watermarking system

Technical field

The present invention relates to digital watermarking (Digital Watermarking) technical field, particularly a kind of self-adapting attack method to discrete cosine transform (DCT) territory digital watermarking system.

Background technology

Digital watermarking system can be embedded into watermark in the image, and image is carried out copyright protection; Can also extract watermark information to the image that contains watermark after attacking simultaneously, thereby be used for detection attack method.The direct purpose of attacking watermark is to remove watermark, can stop the correct extraction of watermark, perhaps can't correctly read the watermark information that is embedded in the image.By attacking watermark, every performance index of digital watermarking system can also be tested, thereby the development of digital watermarking system technology can be promoted, design the optional network specific digit watermaking system that can effectively resist some attack.

In the watermark attacking system, how to guarantee the mass conservation of hidden carrier (being the image of embed watermark), this is that people are seeking the technical matters that solves always.Because, if after attacking, bigger change has taken place in the quality of hidden carrier, can influence the use of hidden carrier, and can make people recognize that hidden carrier has been subjected to attack.Whether change the quality of hidden carrier according to attacking digital watermarking system, existing digital watermarking system attack method is divided into two big classes: first kind attack method, by the quality of the hidden carrier of remarkable change, digital watermarking system is attacked.The representative of this class is geometric attack method and image processing method; The second class attack method, carrying out watermark by the quality of the hidden carrier of trickle change attacks, to such an extent as to this change is not by under the prerequisite of human visual system's perception, the representative of this class is that mosaic is attacked, and these class methods are not had very high accuracy by the digital watermarking system of people's visually-perceptible for the quality change of the hidden carrier of test.

But the mosaic attack method still has some shortcomings.For example, the principle that mosaic is attacked is that image is cut into many enough little pieces, if the fritter that is cut into will make detection method malfunctioning less than the piece of the minimum of detection algorithm needs; And through re-assemblying, just can recover original image fully after attacking; Though attacking, mosaic do not change hidden carrier quality, but the carrier after attacking comprises the complete watermark that digital watermarking system can not be discerned, digital watermarking system can be verified the existence of watermark, therefore, mosaic is attacked can not prove the validity that complete hidden carrier is attacked, and promptly attacks to be detected easily or to be found.In addition, although attacking, mosaic can stop attacking the detection of the hidden carrier in back, but if will from the hidden carrier after attacking, recover watermark, also need to carry out inverse transformation, need the participation of a lot of additional informations just can finish, and be not the process of finishing automatically, recover the poor visual effect of back image.

Therefore, the attack method of existing digital watermarking system is detected easily and finds, the picture quality poor visual effect, and attack the back and recover the water mark method complexity, can not finish automatically.

Summary of the invention

For the validity that can prove that hidden carrier is attacked, the invention provides a kind of self-adapting attack method to DCT territory digital watermarking system.Described technical scheme is as follows:

A kind of self-adapting attack method to the discrete cosine transform domain digital watermarking system, described method comprises:

Step a: obtain digital watermarking image;

Step b: described digital watermarking image is divided into the sub-piece of several n * n pixel, calculates entropy, mean square deviation and the illumination of each described sub-piece respectively by following formula,

Entropy:

Wherein L represents gray level, p _iRepresenting gray-scale value is the pixel and the ratio of all putting of i;

Mean square deviation:

X wherein _IjDenotation coordination is (i, the pixel value of sub-piece j);

Illumination:

According to entropy, mean square deviation and the illumination of each described sub-piece,, obtain human visual system's classification results of each described sub-piece again in conjunction with threshold value, the threshold value of mean square deviation and the threshold value of illumination of the entropy of each described sub-piece; Obtain the intensity subregion result of the watermark component of each described sub-piece superposition according to human visual system's classification results of each described sub-piece, wherein, described n is a natural number;

Step c: according to the intensity subregion result of the watermark component of each described sub-piece superposition, adopt periodically transformation matrix that each described sub-piece is once attacked, obtain sub-block matrix through attacking;

Described periodicity transformation matrix satisfies following two formulas:

(1).

Wherein, A is the periodicity transformation matrix, A=(a _Ij) _{N * n}, a _IjBe integer,

X _nBe the sub-block matrix of original piecemeal, X _n=(x ₁..., x _n) ^T, X _n'=(x ' ₁..., x ' _n) ^T, x ₁..., x _n∈ 0,1 ..., N-1}, N are the exponent numbers of conversion;

(2).

Wherein, m _NBe conversion Cycle, E _nBe n rank unit matrix;

Steps d: calculate the Y-PSNR of each described sub-block matrix, select to attack once more or output image.

According to the intensity subregion result of the watermark component of each described sub-piece superposition, adopt periodically transformation matrix that each described sub-piece is once attacked among the described step c, specifically comprise:

Steps A: each described sub-piece is carried out discrete cosine transform;

Step B: each is carried out discrete cosine transform coefficient through the sub-piece after the discrete cosine transform attack;

Step C: the sub-piece after each process discrete cosine transform coefficient attack is carried out inverse discrete cosine transform, obtain new sub-piece;

Step D: according to the intensity subregion result of the watermark component of each described sub-piece superposition, selected point is once attacked in each described new sub-piece.

According to described Y-PSNR, select specifically to comprise in attack once more or the output image in the described steps d:

If described Y-PSNR 〉=28 and do not reach predetermined number of times of attack are changeed step c and are once attacked, if described Y-PSNR＜28 or reach predetermined number of times of attack halt attacks output image.

The present invention is also by testing new periodicity digital watermarking attack method on digital picture spatial domain and dct transform domain thereof.Experimental result shows that the new attack method effectively stops the detection and the extraction of digital watermarking, has attack effect preferably; Attack method has periodically simultaneously, and digital watermarking image has good visual effect quality after recovering.

Description of drawings

Fig. 1 is the self-adapting attack method overview flow chart of the digital digital watermarking system in the embodiment of the invention;

Fig. 2 is the process flow diagram of once attacking that the digital watermarking in the embodiment of the invention is attacked;

Fig. 3 is testing process figure after the attack of attacking of the digital watermarking in the embodiment of the invention;

Fig. 4 is testing process figure after the recovery of attacking of the digital watermarking in the embodiment of the invention.

Embodiment

For making the purpose, technical solutions and advantages of the present invention clearer, embodiment of the present invention is described further in detail below in conjunction with accompanying drawing.

The embodiment of the invention changes under the not perceived situation in the quality of carrier, utilizes the cyclical theory of matrixing, carries out self-adapting attack, through after repeatedly attacking fully reconstruct attack preceding hidden carrier, thereby prove the validity of attack method automatically.

As shown in Figure 1, the concrete steps of the embodiment of the invention are as follows:

Step 101: obtain digital watermarking image, this image is gray scale or coloured image, and form can be bmp, png, jpg etc.

Step 102: according to the digital watermarking image in the step 101, this digital watermarking image is divided into the experimental process piece, obtain human visual system (HVS) classification results of each height piece of digital watermarking image, obtaining each height piece needs the intensity subregion result of the watermark component of superposition, and concrete steps comprise:

A: digital watermarking image is divided into the sub-piece of some n * n pixel, is designated as X _n, n is a natural number; Be divided into 8 * 8 pixels with digital watermarking image in the present embodiment, promptly n=8 is an example;

B: obtain the HVS classification of each height piece, the method for obtaining the HVS classification has a lot, adopts entropy, mean square deviation and the illumination of calculating each 8 * 8 sub-block of pixels in the present embodiment, is determined by following formula respectively:

Entropy:

Wherein L represents gray level, equals 256 in gray-scale map,

p _iRepresenting gray-scale value is the pixel and whole somes ground ratio of i;

Variance:

N=8 wherein;

Illumination:

Digital watermarking image entropy h is the assembly average of its each gray level bit number of expression, and the less piece of entropy should be a smooth region, and the bigger zone of entropy may be texture or edge, and the texture corresponding variance is less, and the edge corresponding variance is bigger.Illumination susceptibility has been described human eye can be discovered the noise figure that in a certain gray scale background size, and brightness has different noise susceptibilities to human eye to different background, and is the most responsive when emphasis brightness (gray-scale value 128) usually.Therefore, according to entropy, mean square deviation and the illumination of each height piece,, thereby obtained the HVS sorting result again in conjunction with the threshold value of entropy, mean square deviation and illumination parameter.

According to above-mentioned HVS sorting result, sub-pieces all in the digital watermarking image is divided into four class zones, as table 1, the value of the threshold value of parameter for measuring and determine in the experiment;

Table 1

The 1st regional illumination is lower, and texture is simple, and HVS is comparatively responsive to the change of pixel value wherein, and the intensity of the watermark component of superposition should be minimum; The 2nd regional illumination is higher, the texture complexity, and be the edge, the intensity of the watermark component of superposition should be less; The 4th regional illumination is higher, texture complexity, and be not the edge, and HVS is the most weak to the change susceptibility of pixel value wherein, and the intensity of the watermark component of superposition should be maximum; All the other are the 3rd zone, and the intensity of the watermark component of superposition should be less.The threshold value of classification is determined by experiment.

C: the intensity subregion result who obtains each regional watermark component of digital picture: it is minimum that the 1st zone needs the intensity of the watermark component of superposition to answer, and it is maximum that the 4th zone needs the intensity of the watermark component of superposition to answer.The 2nd the zone and the 3rd zone need superposition the watermark component intensity between the 1st the zone and the 4th zone between.

Step 103:, adopt the periodicity transformation matrix that each height piece is once attacked according to the four class zones that obtain in the step 102.

In above-mentioned four class zones, can choose any amount point array mode above-mentioned four class zones are attacked respectively.As preferred embodiment, because the intensity of the watermark component of the 4th regional superposition should be maximum, so choosing more point in the sub-piece corresponding to the 4th zone attacks, because the intensity of the watermark component of the 1st regional superposition should be minimum, chooses less point and attack in the sub-piece corresponding to the 1st zone; For the 2nd zone and the 3rd zone, can choose quantity and attack between the point in the 1st zone and the 4th zone.

Once the process flow diagram of Gong Jiing as shown in Figure 2, concrete steps comprise:

1. the sub-piece to 8 * 8 pixels carries out dct transform;

2. attack carrying out the DCT coefficient again through each the height piece in the digital watermarking image behind the dct transform;

3. carry out the DCT inverse transformation then, obtain the sub-piece of 8 * 8 new pixels;

4. once attack according to intensity selected point in each sub-piece of the watermark component of superposition, the quantity of selected point is according to mentioned above principle.

Periodically matrixing is defined as follows:

Make X _n=(x ₁..., x _n) ^T, X _n'=(x ' ₁..., x ' _n) ^T, x ₁..., x _n∈ 0,1 ..., N-1},

Note conversion X ' _n=AX _n(modN) (A=(a wherein _Ij) _{N * n}, a _IjBe integer) be

And note

(wherein m ∈ 2 ..., N-1}); The matrix norm computing is meant carries out modular arithmetic to each unit in the matrix, promptly to any matrix A=(a _Ij) _{N * n}, A (modN)=(a is arranged _Ij(modN)) _{N * n}

If there is m _N∈ 2 ..., N-1} makes

And to arbitrarily

M is arranged _N≤ m then claims conversion

Have periodically, and the cycle is m _N

Can obtain conversion from above-mentioned

Have following character:

Character 1:

Character 2: if conversion

Cycle be m _N, then

And to any A ^m(modN)=E _nM is arranged _N≤ m, E here _nBe n rank unit matrix;

Conversion

Having periodic sufficient and necessary condition is | A| and N are coprime, here | A| is the determinant of matrix A;

Can obtain the transformation period of A with following method: matrix A is carried out the multiplying of some apotype N, know to obtain unit matrix, have several A to multiply each other, how many attack cycles is exactly.At the situation of 2 peacekeeping N=2, provide the concrete computation process of transformation period below.

Get matrix:

$A_2=\left[\begin{array}{cc}2& 3\\ 3& 4\end{array}\right]$ $\mathrm{mod}(A_2*A_2,2)=\mathrm{mod}(\left[\begin{array}{cc}13& 18\\ 18& 13\end{array}\right],2)=\left[\begin{array}{cc}1& 0\\ 0& 1\end{array}\right]$

According to character 2, the cycle is 2.

After once attacking, obtain new sub-block matrix: X ' _n=AX _n(modN), X ' wherein _nBe 8 * 8 sub-block matrix after once attacking, A is the periodicity transformation matrix, X _nBe 8 * 8 sub-block matrix of original piecemeal, N is the exponent number of conversion;

Step 104: calculate the Y-PSNR (PSNR) of each height piece, select to continue to attack or output image.

Can select to continue to attack or output image according to PSNR value and number of times of attack, as preferred embodiment, if PSNR 〉=28 and do not reach predetermined number of times of attack, forwarding step 103 to further attacks, otherwise, if PSNR＜28 or reach predetermined number of times of attack then halt attacks, the image after output is attacked.

The separation of PSNR=28 for whether continuing to attack is the value of measuring in experiment, also can be provided with PSNR for other values as separation.

Predetermined number of times of attack can realize by the following method: calculate the period m of attacking used matrix _N, number of times of attack is 1 to m _NBetween a random number.

After attacking digital watermarking image in the embodiment of the invention, by digital watermarking system detect and recover after reached good effect.

Be illustrated in figure 3 as testing process figure after the attack that digital watermarking attacks: we attack the image that existing watermaking system generates, and obtain new image, and this image uses original watermaking system to detect, and detect the existence less than watermark.

Be illustrated in figure 4 as testing process figure after the recovery that digital watermarking attacks: we obtain image after carrying out the several times conversion to attack, can revert to the image before attacking fully, thereby this image can be detected the existence of watermark by original watermaking system.

The method that present embodiment provides has restorability when guaranteeing good attack effect.

Present embodiment employing n dimension (n=2,3 ..., 8), the periodicity matrixing of mould N (value of N can be adjusted according to the pixel value of input picture), the sub-piece of preceding n * n that the periodicity matrixing is got A in the step 103 according to dimension n obtains its period m _NCan obtain by table 2.As number of times of attack m＜m _NThe time, can't detect watermark information; When m is m _NIntegral multiple the time, can detect and extract and attack before identical watermark information, and the Y-PSNR of image be the infinity, this shows that image reverts to original state fully.The variation of PSNR was as shown in table 2 when 2 dimensions were attacked different number of times, when 3 dimensions are attacked different number of times the variation of PSNR as shown in table 3, it is as shown in table 4 that 2 dimensions are attacked Y-PSNRs (PSNR), the degree of correlation (CF) and similarity.

Table 2

Number of times of attack	1	2	3	4	5
						PSNR(dB)	43.37	43.17	43.22	43.02	43.34
Number of times of attack	8	10	15	20
						PSNR(dB)	43.28	42.92	43.23	∞

Table 3

Number of times of attack	1	2	3	4	5
						PSNR(dB)	42.36	42.19	42.27	42.11	42.29
Number of times of attack	30	40	50	70
						PSNR(dB)	42.45	42.18	42.25	∞

Table 4

	PSNR	CF	Similarity
				The relative Lena original graph of Cox watermark figure	31.9949	0.9979	29.8207
1 the relative Cox watermark of cycle attack graph figure	43.37	6.3804e-005	0.2384
				Recover the relative Cox watermark figure of figure	INF	1	1

As can be seen, all more than 40, effect can be satisfactory with respect to the Y-PSNR of original graph for attack graph from table 2 and 3; Simultaneously, when number of times of attack equaled to attack the cycle, Y-PSNR was infinitely great, shows that image recovers fully.As can be seen from Table 4, after the attack once, the degree of correlation that we obtain and similarity are very low, can think that image and original image are uncorrelated, dissmilarity; And the Y-PSNR of recovery figure, the degree of correlation shows all that with similar tripe image recovers fully.

The present invention has the matrixing of periodic feature by employing, and the self-adaptive digital watermark system that utilizes human visual system's characteristic is attacked, and makes digital watermarking system can't correctly extract watermark.Image graph picture before periodic transformation can nondestructively be resumed the attack through limited number of time can correctly extract watermark, thereby the proof of success attack is provided to the copyright owner.

With regard to success attack whether, can whether quality change, directly detect whether watermark, attack pattern can prove automatically, hidden carrier aspects the such as whether algorithm of the hidden carrier by the proof of successful attack, before accurately reconstruct is attacked and attack and reconstruct consistent, attack method and existing typical attack method that the present invention proposes compare, and see Table 5.

Table 5

	Success attack	Whether quality changes	Can directly detect watermark	Whether attack pattern can prove automatically	Whether hidden carrier is by the proof of successful attack	Hidden carrier before accurately reconstruct is attacked	Attack whether consistent with the algorithm of reconstruct
								Geometric attack	Be	Change big	Not	Not	Be	Uncertain	Not
Mosaic attacks	Be	Not	Be	Be	Be	Be	Not

	Success attack	Whether quality changes	Can directly detect watermark	Whether attack pattern can prove automatically	Whether hidden carrier is by the proof of successful attack	Hidden carrier before accurately reconstruct is attacked	Attack whether consistent with the algorithm of reconstruct
								Masterplate is attacked and self-adapting attack	Be	Change little	Not	Not	Be	Uncertain	Not
Method of the present invention	Be	Change little	Not	Be	Be	Be	Be

The above only is preferred embodiment of the present invention, and is in order to restriction the present invention, within the spirit and principles in the present invention not all, any modification of being done, is equal to replacement, improvement etc., all should be included within protection scope of the present invention.

Claims (3)

1. the self-adapting attack method to the discrete cosine transform domain digital watermarking system is characterized in that, described method comprises:

Step a: obtain digital watermarking image;

Step b: described digital watermarking image is divided into the sub-piece of several n * n pixel, calculates entropy, mean square deviation and the illumination of each described sub-piece respectively by following formula,

Entropy: Wherein L represents gray level, p _iRepresenting gray-scale value is the pixel and the ratio of all putting of i;

Mean square deviation:

X wherein _IjDenotation coordination is (i, the pixel value of sub-piece j);

Illumination: $\stackrel{\&OverBar;}{x}=\underset{i,j=1..n}{\mathrm{\&Sigma;}}{X}_{\mathrm{ij}}/n^2;$

According to entropy, mean square deviation and the illumination of each described sub-piece,, obtain human visual system's classification results of each described sub-piece again in conjunction with threshold value, the threshold value of mean square deviation and the threshold value of illumination of the entropy of each described sub-piece; Obtain the intensity subregion result of the watermark component of each described sub-piece superposition according to human visual system's classification results of each described sub-piece, wherein, described n is a natural number;

Step c: according to the intensity subregion result of the watermark component of each described sub-piece superposition, adopt periodically transformation matrix that each described sub-piece is once attacked, obtain sub-block matrix through attacking;

Described periodicity transformation matrix satisfies following two formulas:

(1).

Wherein, A is the periodicity transformation matrix, A=(a _Ij) _{N * n}, a _IjBe integer, X _nBe the sub-block matrix of original piecemeal, X _n=(x ₁..., x _n) ^T, X _n'=(x ' ₁..., x ' _n) T, x ₁..., x _n∈ 0,1 ..., N-1}, N are the exponent numbers of conversion;

(2).

Wherein, m _NBe conversion

Cycle, E _nBe n rank unit matrix;

Steps d: calculate the Y-PSNR of each described sub-block matrix, select to attack once more or output image.

2. the self-adapting attack method to the discrete cosine transform domain digital watermarking system as claimed in claim 1, it is characterized in that, among the described step c according to the intensity subregion result of the watermark component of each described sub-piece superposition, adopt the periodicity transformation matrix that each described sub-piece is once attacked, specifically comprise:

Steps A: each described sub-piece is carried out discrete cosine transform;

Step B: each is carried out discrete cosine transform coefficient through the sub-piece after the discrete cosine transform attack;

Step C: the sub-piece after each process discrete cosine transform coefficient attack is carried out inverse discrete cosine transform, obtain new sub-piece;

Step D: according to the intensity subregion result of the watermark component of each described sub-piece superposition, selected point is once attacked in each described new sub-piece.

3. the self-adapting attack method to the discrete cosine transform domain digital watermarking system as claimed in claim 1 is characterized in that, according to described Y-PSNR, selects specifically to comprise in attack once more or the output image in the described steps d:

If described Y-PSNR 〉=28 and do not reach predetermined number of times of attack are changeed step c and are once attacked, if described Y-PSNR＜28 or reach predetermined number of times of attack halt attacks output image.

Images