CN102103738B - Method for generating and authenticating digital image tampered content recoverable variable capacity watermarks - Google Patents

Abstract

The invention discloses a method for generating and authenticating digital image tampered content recoverable variable capacity watermarks, which comprises the following steps of: dividing 2*2 image blocks into smooth image blocks and non-smooth image blocks, extracting 6bit characteristics from the smooth image blocks, extracting 12bit characteristics from the non-smooth image blocks, encrypting the characteristics of the image blocks to generate image block recovery watermarks, randomly embedding the image block recovery watermarks into other image blocks, and judging the authenticity of the image blocks by comparing the consistency of the characteristics of the image blocks and combining block neighborhood characteristics. For the judged tampered image blocks, different tampering recovery operations are executed in two steps according to whether the corresponding recovery watermarks of the image blocks are tampered to improve the tampering recovery quality. The variable capacity recovery watermarks are generated according to the personal characteristics of the image blocks, and the recovery watermarks are simultaneously used for tampering positioning and tampering recovery, so the watermark embedding capacity is reduced when enough image characteristics are ensured, and watermark-containing image quality and tampering recovery quality are considered; and meanwhile, the security is improved, and known forging attacks such as joint attack, mean attack and the like can be resisted.

Description

Recoverable varying capacity watermark generation of digital image tampering content and authentication method

Technical field

The present invention relates to a kind of word image and distort the generation and the authentication method of the recoverable authenticating water-mark of content, be used to detect the authenticity and integrity of digital picture, and to being recovered by tampered image.

Background technology

Along with computing machine, digital image-forming and network communications technology development, digital picture becomes people and obtains main source and one of information propagation important carrier with exchange message.The digitizing storage of image and the appearance of various image processing softwares make editor, modification and synthetic the becoming very simply of digital picture.Digital image processing techniques have improved the display quality of image; But distort and forge image as being used to fields such as news media, court evidence, scientific discovery, can bring serious negative effect the sincerity of society, the public credibility of government and the authenticity of science etc.Therefore, the authenticity and integrity that how to detect with the discriminating digit image has become both at home and abroad frontier nature research topic in recent years, not only has important academic values, has more great social significance and application prospects.

Can recover watermark and can not only detect the authenticity and integrity with the discriminating digit image, and can be similar to the original contents of recovering the tampered region." distort recovery " and have higher requirement for the authentication digital watermarking, how improving recovery picture quality is to recover the important goal of watermark research.Digital picture can be recovered watermarking algorithm and at first image is divided into the identical image block of size to realize tampering location, and the recovery watermark that generates according to the image block characteristic is not to be embedded in image block self, is embedded in other image block and be based on key.This processing has reduced the possibility that image block and its recovery watermark are distorted simultaneously, helps distorting the recovery of piece, detects the difficulty relatively that becomes but make to distort.The tampering location performance that can recover watermarking algorithm directly influence is distorted the recovery quality, this be because: be not detected if distort, perhaps true picture is mistaken for and distorts, and will cause recovering existing in the image distorting can not get recovering or true picture is recovered by mistake.Can recover the tampering location problem of watermark for solving; (P.-L.Lin such as Lin; C.-K.Hsieh, P.-W.Huang.A hierarchical digital watermarking method for image tamper detection and recovery, Pattern Recognition; 2005,38 (12): 2519-2529) propose to be the authenticity of the additional 2 bit authenticating water-marks of each image block with the detected image piece.Yet since authenticating water-mark be piece independently, cause authenticating water-mark not resist and piece attack together.That is to say that suppose to utilize same key to generate the two identical width of cloth of size and contain watermarking images Y1 and Y2, with the image block of relevant position among the image block replacement Y2 among the Y1, authenticating water-mark can not detect this and distort.Even the authenticating water-mark bit of each image block is increased to 32 bits; The self-embedding watermarking algorithm with high quality resume ability (Z.Qian, G.Feng, X.Zhang like people such as Qian proposition; S.Wang.Image self-embedding with high-quaility restoration capability; Digital Signal Processing, 2011 (21): 278-286), the Additional Verification watermark still can not be resisted and pieced attack together.And the Additional Verification watermark has increased watermark and has embedded capacity, also can make to contain the reduction of watermarking images quality.

In addition, recovering watermark generation method is the key factor that influences bearing accuracy, distorts recovery quality and security.The existing generation recovered the most frequently used method of watermark and had two kinds: the 1. quantization encoding of the important DCT of 8 * 8 image blocks (discrete cosine transform) coefficient, its shortcoming is that bearing accuracy is lower.2. the mean value of 2 * 2 image blocks, bearing accuracy is high, but is subject to that permanent average is attacked and not good enough to the recovery quality of non-flat image block.To utilizing 2 * 2 averages to generate in the algorithm that recovers watermark, suppose two image block B1=(30,30,30,30) and B2=(1,1,59,59), their mean value is identical, is 30.Therefore their recovery watermark is also identical.If with image block B1 replace block B2; Can not detect this through the recovery watermarking algorithm that relatively recovers the watermark detection block authenticity and distort (promptly can not resist permanent average attacks), and the recovery quality of utilizing the algorithm of Additional Verification watermark detection block authenticity also to can not get.This be because corresponding recovery watermark only can reconstruct B2 characteristics of mean, and textural characteristics that can't reconstruct B2.

If image block recovers watermark except the characteristics of mean of preserving image block, also preserve the textural characteristics of image block, then the recovery quality of algorithm all can be improved with the ability of the permanent average attack of opposing.But the unity and coherence in writing characteristic of preserving image block can increase watermark capacity, and to increase textural characteristics be unnecessary to being similar to the such smooth block of B1.

Summary of the invention

The present invention solves its technical matters, and the technical scheme that is adopted is: a kind of digital image tampering content recoverable varying capacity watermark generation and authentication method comprise the steps:

A, watermark generate and embed

A1, image block and classification: with size is 2 * 2 image block X that the original image X of 2m * 2n is divided into m * n non-overlapping copies _i={ x _I1, x _I2, x _I3, x _I4, wherein, i is the image block numbering, i=1, and 2 ..., N, N=m * n are the image block number; And image block is divided into two types of smoothed image piece and non-flat image blocks according to the image block content;

A2, pseudo-random sequence and piece chain generate: generating length based on user key Key is the real-valued pseudo-random sequence R={r of N _i| i=1,2 ..., N} is by index ordered sequence generation piece the chain { (X of R _i, X _{I '}) | i, i ' ∈ [1, N] }, i ' is the value of i element in the R index ordered sequence, X _{I '}Be X _iMapping block;

A3, varying capacity feature extraction: to each image block X _iCalculate and generate v bit block characteristic F _i={ f _I1, f _I2..., f _Iv.Wherein, f _I2～f _I6Be image block X _iIf 5 binary codings of high 5 mean values are image block X _iBe the smoothed image piece, then v=6 and f _I1=0, otherwise v=12, f _I1=1, f _I7～f _I12Details coding for the non-flat image block;

A4, varying capacity watermark generate: i the random number r that utilizes pseudo-random sequence R _iGenerate binary pseudo-random B _i={ b _Ij| j=1,2 ... 12}, encrypted image block feature F _i={ f _I1, f _I2..., f _IvGenerate and recover watermark W _i={ w _I1, w _I2..., w _Iv,

$w_{\mathrm{ij}}=f_{\mathrm{ij}}\⊕b_{\mathrm{ij}}$

Wherein,

Be xor operation; To smoothed image piece b _IjSubscript j span be 1～6, to non-flat image block b _IjSubscript j span be 1～12;

A5, varying capacity watermark embed: image block X _iRecovery watermark W _iEmbed its mapping block X _{I '}Low order generate and to contain watermarking images piece Y _{I '},

If X _iBe the non-flat image block,

If X _iBe the smoothed image piece,

B, watermark extracting and comparison

B1, piecemeal, pseudo-random sequence and piece chain generate: Y ^*Be contain receive after the watermarking images Y transmission by altimetric image, press A1 and go on foot operation handlebar Y ^*Be divided into 2 * 2

Press A2 step identical operations and key K ey again and generate random series R and piece chain

B2, block feature calculation and Reconstruction: For each test image block?

A3 step calculated by the image block feature?

while an image according to its mapping block?

Significant Bit extraction recovery watermark?

reconstruct the block feature?

$F_i^L=W_i^{*}\⊕B_i$

Wherein, B _iFor based on r _iThe length that generates is 12 binary pseudo-random, from its map image piece

Low level extracts watermark information

Obtain by following formula,

B3, characteristic of division compare: more tested image block

The block feature that a high position calculates

With from its mapping block

The block feature that recovers watermark reconstruct is extracted in low order Generate relatively square D={d by following formula _i| i=1,2 ..., N},

Wherein $c=6(1+f_{i1}^{*});$

C, distort detection

C1, calculating generate neighborhood characteristics matrix Δ={ δ of comparator matrix D _i| i=1,2 ..., N},

δ _i＝∑d _j，j＝i±1，i±n，i+n±1，i-n±1

C2, according to comparator matrix D and and neighborhood characteristics matrix Δ, generate initial state and distort the detection matrix T ⁰ $T^0=\left(t_i^0\right|i=\mathrm{1,2},...,N),$

The detection matrix T is distorted in C3, the initial state that generates according to C2 step ⁰The detection matrix T is distorted in generation,

j=i ± 1 wherein; I ± n, i ± n ± 1.t _iThe tested image block of=1 expression Distorted t _iThe tested image block of=0 expression

Be real;

D, distort recovery

If element is not 0 entirely among the T, show to be had the tampered image piece in the altimetric image, then obtain distorting recovery image Y to carried out following two steps successively by altimetric image ^R

D1, characteristic recovery: to being judged to be the image block of distorting

If its mapping block Y _{I '} ^*Be real, with B2 step reconstructed blocks characteristic Right

Recover,

Wherein, Γ ^-1() is the inverse function of image block feature coding function; Generate to indicate matrix L simultaneously, distort piece in order to what mark was not resumed,

D2, neighborhood recover: if element is not 0 entirely among the L, to l _i=1 correspondence image piece

Utilization is adjacent 12 valid pixel average correction image pieces in the pixel

In the value of each pixel.

Compared with prior art, the invention has the beneficial effects as follows:

1, characteristic volume-variable: the present invention is based on the not isometric block feature of image block contents extraction---smoothed image piece 6 bits; Non-flat image block 12 bits; Recover watermark to generate varying capacity adaptively; When guaranteeing high quality resume tampered image piece and resisting permanent average attack, make the characteristic capacity of extraction reach minimum.

2, the recovery watermark capacity is variable: recover not increase when watermark generates any redundant information; The block feature of two kinds of different lengths (6 and 12 bit) is encrypted the generation recovery watermark isometric with it; Recovering does not increase any redundant information when watermark generates, and adopts distinct methods that varying capacity is recovered in other image block of watermark embedding.When keeping the image block characteristic fully, make embedding recover watermark capacity and reach minimum.Simultaneously, algorithm need not to increase additional authenticating water-mark, has further reduced watermark and has embedded capacity, thereby improved the quality of moisture printing word image.

3, recover watermark and be used to distort detection: the consistency detection of image block has been utilized all recovery watermarks of image block---and the smoothed image piece is with 6 bits recovery watermark, and the non-flat image block is with 12 bits recovery watermark.All recover watermark participation consistency detection, can improve algorithm and resist the ability that attack was attacked and pieced together to permanent average, also lay a good foundation for the detection performance of distorting that further improves algorithm.

4, high security: at first the characteristic of image block is encrypted to generate and recover watermark information based on key; And then be embedded at random in other image block based on key; And the consistance of the recovery watermark information detected image piece of utilization embedding, set up the correlativity at random between the image block.Simultaneously, image block characteristic information as much as possible has been preserved in varying capacity recovery watermark.These characteristics make algorithm can resist known Forgery Attack, as piece attack, permanent average attack etc. together, thus the security that has improved algorithm.

5, the high quality of recovering: at first, adopt 12 bits to preserve its characteristic, keep the characteristic of image block as much as possible, distort the recovery quality for raising and lay a good foundation to the abundant non-flat image block of details.Secondly, high bearing accuracy (2 * 2 pixel) and good distorting detect performance and recover quality possibility is provided for improving to distort.At last, to different tampered image pieces, carry out the different recovery operations of distorting in two steps and further improved the quality of distorting the recovery image in the algorithm.

Below in conjunction with embodiment the present invention is described further.

Description of drawings

Fig. 1 is that the varying capacity of the embodiment of the invention recovers the watermark generation and embeds the step block diagram.

Fig. 2 is classification of embodiment of the invention non-flat image block and subclass coding synoptic diagram.

Fig. 3 is for generally distorting down the present invention and existing two kinds of performance comparison diagrams that can recover watermarking algorithm.

Fig. 4 attacks the present invention and existing two kinds of performance comparison diagrams that can recover watermarking algorithm down for permanent average.

Fig. 5 pieces together to attack the present invention and existing two kinds of performance comparison diagrams that can recover watermarking algorithm down.

Embodiment

Embodiment

A kind of digital image tampering content recoverable varying capacity watermark generation and authentication method comprise watermark generation and embedding, watermark extracting and comparison, distort and detect and distort recovery four parts.

A, watermark generate and embed

Fig. 1 is that the watermark in the embodiment of the invention method generates and the concrete steps block diagram that embeds, and comprises following five steps:

A1, image block and classification: with size is 2 * 2 image block X that the original image X of 2m * 2n is divided into m * n non-overlapping copies _i={ x _I1, x _I2, x _I3, x _I4, wherein, i is the image block numbering, i=1, and 2 ..., N, N=m * n are the image block number; And image block is divided into two types of smoothed image piece and non-flat image blocks according to the image block content;

Image block classification in the present embodiment: to 2 * 2 image block X _i={ x _I1, x _I2, x _I3, x _I4, if the difference of two maximum pixel and two big minimum pixels is no more than h (h=3 in this example) in the image block, then this image block is the smoothed image piece, otherwise is the texture image piece.

During enforcement; The span of two maximum pixel that image block classification is adopted and the poor h of two big minimum pixels is generally 2 to 16; The h value is more little, is considered to the many more of non-flat image block, and the recovery quality of image is high more; But the watermark capacity of corresponding embedding is big more, can reduce the quality that contains watermarking images.Otherwise the smoothed image piece is many more, and the watermark capacity of corresponding embedding is more little, and the quality that contains watermarking images is high more, but the quality of recovery image is low more.

A2, pseudo-random sequence and piece chain generate: generating length based on user key Key is the real-valued pseudo-random sequence R={r of N _i| i=1,2 ..., N} is by index ordered sequence generation piece the chain { (X of R _i, X _{I '}) | i, i ' ∈ [1, N] }, i ' is the value of i element in the R index ordered sequence, X _{I '}Be X _iMapping block;

A3, varying capacity feature extraction: to each image block X _iCalculate and generate v bit block characteristic F _i={ f _I1, f _I2..., f _Iv.Wherein, f _I2～f _I6Be image block X _iIf 5 binary codings of high 5 mean values are image block X _iBe the smoothed image piece, then v=6 and f _I1=0, otherwise v=12, f _I1=1, f _I7～f _I12Details coding for the non-flat image block;

In the present embodiment, image block X _iThe binary coding of high 5 mean values is according to formula

Calculate wherein (.) _BThe binary coding of expression integer.

The details coding f of non-flat image block _I7～f _I12Obtain through the image block classification.The position appears in two pixels maximum according to 2 * 2 image block intermediate values, and the non-flat image block is divided into 6 types, and maximum two pixels are positioned at image block: the f of top _I7～f _I9Be encoded to 001, the f of below _I7～f _I9Be encoded to 010, the f in left side _I7～f _I9Be encoded to 011, the f on right side _I7～f _I9Be encoded to 100, a left side is the f of (upper left, bottom right) tiltedly _I7～f _I9Be encoded to 101, the right tiltedly f of (upper right, down left) _I7～f _I9Be encoded to 110, as shown in Figure 2.f _I10～f _I12Be two maximal value pixels and with two minimum value pixels with the uniform quantization binary coding of difference, promptly

Wherein j ' (j=1,2,3,4) is the index ordered sequence of four pixels of image block.

A4, varying capacity watermark generate: i the random number r that utilizes pseudo-random sequence R _iGenerate binary pseudo-random B _i={ b _Ij| j=1,2 ... 12}, encrypted image block feature F _i={ f _I1, f _I2..., f _IvGenerate and recover watermark W _i={ w _I1, w _I2..., w _Iv,

$w_{\mathrm{ij}}=f_{\mathrm{ij}}\⊕b_{\mathrm{ij}}$

Wherein,

is xor operation.To smoothed image piece, pseudo-random sequence B _i={ b _Ij| j=1,2 ... 12} only uses preceding 6 bits; To non-flat image block, pseudo-random sequence B _i={ b _Ij| j=1,2 ... 12 bits of 12} are all used.

A5, varying capacity watermark embed: image block X _iRecovery watermark W _iEmbed its mapping block X _{I '}Low order, generate and to contain watermarking images piece Y _{I '},

If X _iBe the non-flat image block,

If X _iBe the smoothed image piece,

B, watermark extracting and comparison

B1, piecemeal, pseudo-random sequence and piece chain generate: Y ^*Be contain receive after the watermarking images Y transmission by altimetric image, press A1 and go on foot operation handlebar Y ^*Be divided into 2 * 2 Press A2 step identical operations and key K ey again and generate random series R and piece chain

B2, block feature calculation and Reconstruction: For each test image block?

A3 step calculated by the image block feature?

while an image according to its mapping block?

Significant Bit extraction recovery watermark?

reconstruct the block feature?

$F_i^L=W_i^{*}\⊕B_i$

Wherein, B _iFor based on r _iThe length that generates is 12 binary pseudo-random, from its map image piece

Low level extracts watermark information

Obtain by following formula,

B3, characteristic of division compare: more tested image block The block feature that a high position calculates

With from its mapping block

The block feature that recovers watermark reconstruct is extracted in low order

Generate relatively square D={d by following formula _i| i=1,2 ..., N},

Wherein

that is to say; To the smoothed image piece, relatively whether be identical with

preceding 6 bits; To the non-flat image block, relatively whether be identical with

all 12 bits.

C, distort detection

C1, calculating generate neighborhood characteristics matrix Δ={ δ of comparator matrix D _i| i=1,2 ..., N},

δ _i＝∑d _j，j＝i±1，i±n，i+n±1，i-n±1

Wherein, neighborhood characteristics matrix Δ={ δ of D _i| i=1,2 ...; The generation method of N} is existing method; Specific practice sees document He Hong-jie for details, Zhang Jia-shu, Chen Fen.A self-recovery fragile watermarking scheme for image authentication with superior localization; Science in China Series F-Information Sciences, 2008.51 (10): 1487-1507.

C2, according to comparator matrix D and and neighborhood characteristics matrix Δ, generate initial state and distort the detection matrix T ⁰ $T^0=\left(t_i^0\right|i=\mathrm{1,2},...,N),$

The detection matrix T is distorted in C3, the initial state that generated according to the last step ⁰The detection matrix T is distorted in generation,

j=i ± 1 wherein; I ± n, i ± n ± 1.t _iThe tested image block of=1 expression

Distorted t _iThe tested image block of=0 expression

Be real;

D, distort recovery

If element is not 0 entirely among the T, show to be had the tampered image piece in the altimetric image, then obtain distorting recovery image Y to carried out following two steps successively by altimetric image ^R

D1, characteristic recovery: to being judged to be the image block of distorting

If its mapping block Y _{I '} ^*Be real, with B2 step reconstructed blocks characteristic

Right

Recover,

Wherein, Γ ^-1() is the inverse function of image block feature coding function; Generate to indicate matrix L simultaneously, distort piece in order to what mark was not resumed,

D2, neighborhood recover: if element is not 0 entirely among the L, to l _i=1 correspondence image piece

Utilization is adjacent 12 valid pixel average correction image pieces in the pixel

In the value of each pixel.

Effect of the present invention can be through following performance evaluation and experimental verification

During analysis verification; The bit number (bpp:bit per pixel) that characteristic capacity and watermark capacity generate or embed with unit picture element is respectively weighed; The quality that contains watermarking images is weighed (unit is dB) with the Y-PSNR of it and original image, and chooses document [1], [2] two kinds of recoverable watermark generations of representative digital image tampering content and identifying algorithm and do comparison:

Document [1] Z.Qian, G.Feng, X.Zhang, S.Wang.Image self-embedding with high-quaility restoration capability, 2011 (21): 278-286

Document [2] He Hong-jie; Zhang Jia-shu; Chen Fen.A self-recovery fragile watermarking scheme for image authentication with superior localization; Science in China Series F-Information Sciences, 2008.51 (10): 1487-1507.

1, characteristic capacity and reconstruction quality

Feature extraction is to generate the important step of recovering watermark.Desirable situation is to preserve characteristics of image as much as possible with the least possible capacity.Adopt similar feature extracting method, the image block characteristic of extraction is many more, and is good more according to the picture quality of feature reconstruction, but characteristic information also increases.Table 1 has provided the present invention and existing document [1] and document and has recovered the characteristic capacity of watermarking algorithm and the statistics of reconstruction quality for [2] two kinds.Can be found out that by table 1 characteristic information of the present invention is maximum, it is also best that correspondingly it utilizes the picture quality of feature reconstruction.The characteristic information of document [1] is minimum; Its reconstruction quality is better; Especially to its smoother digital picture, this is because document [1] is to utilize the important DCT coefficient of 8 * 8 image blocks to generate the recovery watermark, and dct transform can effectively have been removed redundancy between pixel; But cause the tampering location precision of algorithm not high more greatly, and possibly reduce the quality of recovering image owing to image block.It can also be seen that by table 1 document [1] all is 1bpp to the characteristic information that the different digital image extracts, document [2] all is 1.5bpp to the characteristic information that the different digital image extracts, and the characteristic capacity that this explanation document [1] and [2] extracts is all fixed.On the contrary, the present invention is variable to the characteristic capacity that the different digital image extracts, and variation range is 1.80～2.62bpp.Image is level and smooth more, and the characteristic capacity of extraction is more little.

The characteristic capacity of table 1 the present invention and existing literature method and the statistics of reconstruction quality

2, watermark capacity and contain the watermarking images quality

Because the watermark capacity that can recover watermarking algorithm is bigger, embed the anamorphose of introducing for reducing watermark as far as possible, can recover watermark mostly at the low order embed watermark information in spatial domain.Therefore, the watermark embedding capacity that can recover watermarking algorithm is big more, brings anamorphose also serious more, and the quality that contains watermarking images that promptly generates is poor more.Table 2 has provided the present invention and existing document [1] and [2] two kinds of watermark capacity and the statistics that contains the watermarking images quality of recovering watermarking algorithm of document.Find out that by table 2 watermark capacity of document [1] and [2] is all fixed, and be respectively 3bpp and 1.5bpp, and watermark capacity of the present invention is variable.Document [1] is similar to the quality of the moisture printing word image that different images generates, and its Y-PSNR is about 38dB respectively, and document [2] is also similar to the quality of the moisture printing word image that different images generates, and its Y-PSNR is about 48dB respectively.The present invention differs bigger to the quality of the moisture printing word image that different images generates, and its Y-PSNR is between 38---and between the 43dB, image is level and smooth more, and the quality that contains watermarking images of generation is good more.

Contrast table 1 can be found out with table 2; The present invention and document [2] do not increase redundant information when watermark embeds; And document [1] is distorted the recovery quality for positioning tampering and raising; The watermark capacity that embeds is 3 times of characteristic information, has reduced the quality that contains watermarking images, and the Y-PSNR that contains watermarking images that the present invention generates all is higher than document [1].

The watermark capacity of table 2 the present invention and existing document and the statistics that contains the watermarking images quality

3, generally distort

Fig. 3 provides and generally distorts under the condition, and the present invention and existing document [1] and document recover distorting of watermarking algorithm for [2] two kinds and detect and distort restoration result.(a) component is that size is the Beach original-gray image of 328 * 328 pixels among Fig. 3; (b) component is the moisture printing word image that utilizes algorithm of the present invention to generate among Fig. 3; Its Y-PSNR is 42.93dB; (c) component is a tampered image among Fig. 3, wherein comprises following distorting: add literal " ITP Southwest Jiaotong University "; Below add a fawn, distort pixel ratio and be about 7.78%.(d) component among Fig. 3; (e) component and (f) component are respectively the testing results of distorting of the present invention, [2] two kinds of recoveries of existing document [1] and document watermarking algorithm; From figure, find out; The bearing accuracy of document [1] is lower than the present invention and document [2], and this is because the image block size of the present invention and document [2] is 2 * 2 pixels, and the image block size of document [1] is 8 * 8 pixels.For the quantitative comparison algorithm distort the detection performance, be that the loss of the unit the present invention, document [1] and the document [2] that calculate is respectively 0.3%, 0% and 3.45% with 2 * 2 pixels, empty inspection rate is respectively 0.39%, 5.13% and 0.02%.(g) component among Fig. 3, what (h) component and (i) component were respectively the present invention, document [1] and document [2] distorts the recovery image, and the Y-PSNR of they and original image is respectively 43.63dB, 43.06dB and 29.94dB.The present invention is similar with the recovery effects of document [1], has all obtained distorting preferably recovery effects.The recovery effects of document [2] is relatively poor, in the tampered region of (i) component noise spot is arranged among Fig. 3, this is because these recovery watermark informations of distorting piece also are destroyed, and the efficient recovery watermark information is reformed distorts that piece causes and document [2] can't have.

Can be found out by above-mentioned comparison, generally distort under the condition that the present invention has similar bearing accuracy with document [2], is superior to document [1], simultaneously, the present invention and document [1] have the similar recovery quality of distorting, and are superior to document [2].Be that the present invention has good tampering location precision simultaneously and distorts the recovery quality.

4, permanent average is attacked

Fig. 4 provides under the average attack condition, and the present invention, existing document [1] and document recover distorting of watermarking algorithm for [2] two kinds and detect and distort restoration result.(a) component is that size is the Lena original-gray image of 256 * 256 pixels among Fig. 4, and (b) component is the moisture printing word image that utilizes algorithm of the present invention to generate among Fig. 4, and its Y-PSNR is 42.54dB.(c) component is a tampered image among Fig. 4, and wherein, the square area of size 140 * 140 pixels is suffered the average attack, and the ratio of distorting is about 30%.(d) component among Fig. 4, (e) component and (f) component are respectively the testing result of distorting of the present invention, document [1] and document [2].(f) component does not have white portion among Fig. 4, explains that document [2] can not detect this and distort, and promptly loss is respectively 100%, and this explanation document [2] can not be resisted permanent average and attack.Owing to detect and do not distort piece, algorithm can not distorted recovery to tampered image, and it is distorted and recovers image (shown in (i) component among Fig. 4) with identical by altimetric image ((c) component among Fig. 4), with the Y-PSNR of original image be 11.20dB.The present invention and document [1] can both detect this and distort, and are that the present invention that calculates of unit and the loss of document [1] are respectively 0.21% and 0% with 2 * 2 pixels, and empty inspection rate is respectively 0.94% and 3.6%.(d) component also can be found out with (e) component in the comparison diagram 4, and the present invention distorts and has a few omission image block in the detection, and this is that to attack be to implement to 2 * 2 image blocks because the unit of positioning tampering of the present invention is 2 * 2 pixels and average.Document [1] is that unit judges and to distort piece with 8 * 8 pixels, thus document [1] distort the image block that does not have omission in the district.Distort at algorithm and not to be judged as the image block of distorting in the detection and not carry out and distort recovery operation, cause distorting of algorithm of the present invention to recover quality a little less than document [1].What (g) component and (h) component were respectively the present invention and document [1] among Fig. 4 distorts the recovery image, and the Y-PSNR of they and original image is respectively 30.81dB and 31.39dB.(g) component can be found out from Fig. 4, and the present invention distorts a few noise spot of existence of recovering image (show " cap " zone), and this is because there is the omission piece in this zone, not carry out to distort that recovery operation causes.

Can find out that by above-mentioned comparison document [2] can not be resisted to the average of 2 * 2 enforcements and attack.The present invention and document [1] can detect such and distort and effectively the tampered region is carried out and recover, and bearing accuracy of the present invention is higher than document [1], and empty inspection rate is lower than document [1], but loss causes distorting and recovers quality a little less than document [1] a little more than document [1].

5, piece attack together

Fig. 5 provides and pieces together under the attack condition, and the present invention, existing document [1] and document recover distorting of watermarking algorithm for [2] two kinds and detect and distort restoration result.Among Fig. 5 (a) component and (b) component be respectively utilize that same key generates contain watermark Monlisa and Napoleon gray level image, their big or small equal 208 * 328 pixels.The head zone that will contain watermark Monelisa image replaces with the same area (piecing attack together) that contains watermark Napoleon image, and the tampered image that obtains is shown in (c) component among Fig. 4, and the ratio of distorting is about 18%.(d) component among Fig. 5; (e) component and (f) component are respectively the testing result of distorting of the present invention, document [1] and document [2]; With 2 * 2 pixels is that the loss of the unit the present invention, document [1] and the document [2] that calculate is respectively 2.03%; 86.67% and 20.22%, empty inspection rate is respectively 0.13%, 2.27% and 0.26%.Can be found out that by (e) component among Fig. 5 document [1] only detects the border of piecing the zone together, the inside of piecing the zone together is through authentication.Why document [1] can detect is positioned at 8 * 8 image blocks of piecing the zone boundary together, is only to have changed the partial pixel in these image blocks because piece attack together.Do not carry out and distort recovery operation owing to be detected the image block of distorting, it is lower to cause distorting of document [1] to recover quality, and shown in (h) component among Fig. 5, its Y-PSNR is 17.72dB.The present invention and document [2] can be resisted and piece attack together, and (d) component and (f) component in the comparison diagram 5 (d) are pieced many than in (f) component of the black color dots in zone together in the component, and this explains that loss of the present invention is lower than document [2].The image block that the present invention can have the efficient recovery watermark information also to be distorted, the recovery quality of distorting therefore of the present invention obviously is superior to document [2].Among Fig. 5 (g) component and (i) component be respectively the present invention and document [2] distort the recovery image, their Y-PSNR is respectively 33.82dB and 24.83dB.

Found out relatively that by above-mentioned document [1] can not be resisted and piece attack together, the present invention and document [2] can be resisted and piece attack together, and of the present invention distorting through detecting performance and distorting the recovery quality all obviously is superior to document [2].

Table 3 has provided three kinds and has distorted the comparative result that distorting of the present invention under the mode, existing document [1] and [2] two kinds of recoveries of document watermarking algorithm detected performance (false drop rate) and distorted recovery quality (distort and recover image and the Y-PSNR that contains watermarking images).Can find out that to sum up the recovery watermark capacity in the algorithm of the present invention is variable.To the recovery watermark of different digital image generation different capabilities, image is level and smooth more, and the recovery watermark amount of generation is few more.Under the prerequisite that satisfies the preservation of image block quantity of information, reduce the recovery watermark as far as possible and embedded capacity, thereby improved the quality that contains watermarking images.Simultaneously, algorithm of the present invention has higher tampering location performance and distorts the recovery quality.Generally distort, average is attacked, piece together attack under, accurately positioning tampering position, high probability tamper detection piece, high quality resume tampered image.

Table 3 is distorted and is detected performance and distort the recovery mass ratio

Annotate: false drop rate=distort ratio * loss+(1-distorts ratio) * empty inspection rate.

Claims (1)

1. digital image tampering content recoverable varying capacity watermark generation and authentication method comprise the steps:

A, watermark generate and embed

A1, image block and classification: with size is 2 * 2 image block X that the original image X of 2m * 2n is divided into m * n non-overlapping copies _i={ x _I1, x _I2, x _I3, x _I4, wherein, i is the image block numbering, i=1, and 2 ..., N, N=m * n are the image block number; And image block is divided into two types of smoothed image piece and non-flat image blocks according to the image block content;

A2, pseudo-random sequence and piece chain generate: generating length based on user key Key is the real-valued pseudo-random sequence R={r of N _i| i=1,2 ..., N} is by index ordered sequence generation piece the chain { (X of R _i, X _{I '}) | i, i ' ∈ [1, N] }, i ' is the value of i element in the R index ordered sequence, X _{I '}Be X _iMapping block;

A3, varying capacity feature extraction: to each image block X _iCalculate and generate v bit block characteristic F _i={ f _I1, f _I2..., f _Iv, wherein, f _I2～f _I6Be image block X _iIf 5 binary codings of high 5 mean values are image block X _iBe the smoothed image piece, then v=6 and f _I1=0, otherwise v=12, f _I1=1, f _I7～f _I12Details coding for the non-flat image block;

A4, varying capacity watermark generate: i the random number r that utilizes pseudo-random sequence R _iGenerate binary pseudo-random B _i={ b _Ij| j=1,2 ... 12}, encrypted image block feature F _i={ f _I1, f _I2..., f _IvGenerate and recover watermark W _i={ w _I1, w _I2..., w _Iv,

Wherein, Be xor operation; To smoothed image piece b _IjSubscript j span be 1～6, to non-flat image block b _IjSubscript j span be 1～12;

A5, varying capacity watermark embed: image block X _iRecovery watermark W _iEmbed its mapping block X _{I '}Low order generate and to contain watermarking images piece Y _{I '},

If X _iBe the non-flat image block,

If X _iBe the smoothed image piece,

B, watermark extracting and comparison

B1, piecemeal, pseudo-random sequence and piece chain generate: Y ^*Be contain receive after the watermarking images Y transmission by altimetric image, press A1 and go on foot operation handlebar Y ^*Be divided into 2 * 2

Press A2 step identical operations and key K ey again and generate random series R and piece chain

B2, block feature calculation and Reconstruction: For each test image block?

A3 step calculated by the image block feature?

while an image according to its mapping block? low watermark bit extraction recovery?

reconstruct the block feature?

Wherein, B _iFor based on r _iThe length that generates is 12 binary pseudo-random, from its map image piece

Low level extracts watermark information

Obtain by following formula,

B3, characteristic of division compare: more tested image block

The block feature that a high position calculates

With from its mapping block The block feature that recovers watermark reconstruct is extracted in low order

Generate comparator matrix D={d by following formula _i| i=1,2 ..., N},

Wherein

C, distort detection

C1, calculating generate neighborhood characteristics matrix Δ={ δ of comparator matrix D _i| i=1,2 ..., N},

δ _i＝∑d _j，j＝i±1，i±n，i+n±1，i-n±1

C2, according to comparator matrix D and neighborhood characteristics matrix Δ, generate initial state and distort the detection matrix T ⁰

The detection matrix T is distorted in C3, the initial state that generates according to C2 step ⁰The detection matrix T is distorted in generation,

Wherein J=i ± 1, i ± n, i ± n ± 1, t _iThe tested image block of=1 expression

Distorted t _iThe tested image block of=0 expression

Be real;

D, distort recovery

If element is not 0 entirely among the T, show to be had the tampered image piece in the altimetric image, then obtain distorting recovery image Y to carried out following two steps successively by altimetric image ^R

D1, characteristic recovery: to being judged to be the image block of distorting If its mapping block Y _{I '} ^*Be real, with B2 step reconstructed blocks characteristic

Right

Recover,

Wherein, Γ ^-1() is the inverse function of image block feature coding function; Generate to indicate matrix L simultaneously, distort piece in order to what mark was not resumed,

D2, neighborhood recover: if element is not 0 entirely among the L, to l _i=1 correspondence image piece

Utilization is adjacent 12 valid pixel average correction image pieces in the pixel In the value of each pixel.

Images