CN1614634A - Generating and authenticating method for positioned authenticating water-mark for recoverying distorted image contents - Google Patents

Abstract

A method for generating identification watermark of positioning type includes carrying out non-mean value scalar quantization based on cipher key for wavelet low frequency coefficient after original image lowest and secondary lower bit set on zero, using random sequence for ciphering to increase safety of watermark, embedding the ciphered water mark on zero set bit, using D-value of image and watermark to position image information and position falsified of watermark at time of certification being executed.

Description

The generation of positioned authenticating water-mark for recoverying distorted image contents and authentication method

Affiliated technical field

The present invention relates to a kind of generation and authentication method of positioned authenticating water-mark for recoverying distorted image contents, be used for the authentication of digital picture, promptly integrality, the authenticity to digital picture proves, and the image information of being distorted is similar to recovery.

Background technology

Universal day by day along with the Internet, particularly along with the development of computer network communication technology, the interchange of multimedia messages has reached the unprecedented degree of depth and range, the exchange of data has become a relative simple process with transmission, yet the full transparency and the ease for operation of the network information make malicious attacker to distort and to forge it easily.In some application (biological example feature identification, legal argument image, news image, medical image, online transaction), people need know for sure the digital product that maybe will use that receives whether true, complete, whether also have use value.Therefore the information security issue under the network environment has become one of current difficult problem that presses for solution.In current authentification of message field, digital watermarking is paid close attention to by academia and business circles day by day as one of technology that addresses this problem.

The digital watermarking that is used to authenticate should satisfy usually following some: 1. imperceptible, 2. distort and detect and station-keeping ability, 3. detecting does not need original image, the 4. security of algorithm; Not only need detection and location to distorting in some application scenario, also require watermarking algorithm can be similar to the image information that recovery is distorted, this just is the higher requirement of authenticating water-mark algorithm proposition.The existing recovery in the authenticating water-mark, common have following two kinds of methods:

(1) the recovered authenticating water-mark algorithm that compresses based on JPEG

JPEG is one of general International Standard of image compression.List of references " Protection of digitalimages using self-embedding " (J.Fridrich, M.Goljan Processings ofNJIT Symposium on Content Security and Data Hiding in DigitalMedia, (Newark, NJ) [C] .New Jersty Institute of Technology May, 1999.) 7 of original graph image heights are done piecemeal (piece of 8*8) DCT (discrete cosine) conversion, to being embedded into the lowest order of this another piece of image behind its coefficient quantization coding, thus the picture material that realizes can also utilizing when distorted image locate the approximate recovery of quantization encoding of the DCT coefficient of preserving in the watermark to be distorted.List of references " self-embedding of image and detection of altering and recovery algorithms " (Zhang Hongbin, raise into electronic letters, vol, 2004.2 pp:196-199) above-mentioned algorithm is improved from two aspects, one side is according to the statistical property of DCT coefficient, bits of coded progress row is improved, the recovery quality of image is significantly improved; On the other hand, the off-set value that embeds by the cipher controlled watermark block security that improves algorithm.But this algorithm can not judge accurately when image is distorted and distort piece that the algorithm secret key space is little simultaneously, has hidden danger.

(2) based on the recovered authenticating water-mark algorithm of error correction controlled coding (ECC)

Error correction controlled coding (ECC:error control coding) is one of three unitary Item in the information theory, be mainly used in the Error Control of digital display circuit, the quality that guarantees digital transmission processes such as communication, storage, information transfer and media play is had significance.List of references " A watermarking sequenceusing parities of error control coding for image authentication andcorrection " (J.Lee, C.S.Won IEEE Trans on ConsumerElectronics, 2000,46 (2) pp:313:317) utilize error-correcting code technique (for example RS sign indicating number) by picture coding is generated watermark, the watermark that generates is embedded into the lowest order of image, utilizes the error correction decode technology to realize location and recovery tampered image.List of references " recoverable fragile digital figure watermark " (Zheng Jiangbin, the big Gan of Feng etc., Chinese journal of computers, 2004.3 pp:371-376) adopt a kind of watermark generating algorithm based on piece intersection interleave RS coding, further improve the station-keeping ability and the larger area of algorithm and distort recovery capability.But this algorithm may carry out wrong recovery to image information when only having watermark information to be distorted.

The common feature of two kinds of above-mentioned watermarking algorithms to image authentication and recovery the time is: the information high seven to eight tested imagery exploitation obtains watermark by specific algorithm, the watermark that obtains is embedded into the lowest order of image, locatees and recover distorting image by the watermark of lowest order.Its weak point is that it only can point out tampered position to watermarking images (image of embed watermark), and can not distinguish distorting of high seven bit image information or distorting of lowest order watermark; Recovery to tampered image is feasible under the situation that watermark information is not distorted, and can cause that when watermark information is distorted the difficulty of tampering location maybe can not adopt corresponding effective method to recover the image information of being distorted.

Watermarking images is in network transmission process, can cause the destruction of a little watermark signal or image information unavoidably, thereby the false-alarm probability that causes above-mentioned watermark authentication system improves, also give the assailant to have an opportunity to take advantage of on the other hand, the assailant can forge distorting image information by distorting watermark, original real image can not be distorted recovery and can not carry out effectively by authentication, thereby reach certain purpose of assailant.Therefore adopt above-mentioned authenticating water-mark algorithm can cause real image effectively not utilized, reduce the exchange efficiency of digital picture, hinder the popularization and the application of digitized authentication image technology.

Summary of the invention

The generation and the authentication method that the purpose of this invention is to provide a kind of positioned authenticating water-mark for recoverying distorted image contents, the intensity of distorting to image information can not only be distorted and position, be discerned to this method to the difference of watermarking images, can also be under the prerequisite whether the differentiation watermark is distorted, the image information that adopts the approximate recovery of diverse ways to be distorted; Its authentication result is directly perceived, good visual effect, and key space is big, watermarking algorithm safety.

The present invention solves its technical matters, and the technical scheme that is adopted is: a kind of generation of positioned authenticating water-mark for recoverying distorted image contents and authentication method comprise the steps:

(1), watermark generates: with the lowest order zero setting of each pixel of original image I, and based on key K ₀The inferior low level zero setting of picked at random partial pixel obtains the zero setting image

This zero setting image I is made two-dimentional one-level wavelet decomposition: LL, and LH, HL, HH} does wherein low frequency wavelet coefficient LL based on key K ₃Generate the non-average scalar quantization of the 5-7 bit of random series, generate low frequency compressed image I _LLPicked at random 4 bits wherein series arrangement in the corresponding space of its pixel generates and 4 onesize bit bianry image I of original image _Lb ¹, remaining 1-3 bit is arranged in any order and is generated remaining bits bianry image I _Lb ²Utilize key K again ₁And key K ₂Generate different random seriess respectively to 4 bit bianry image I _Lb ¹With remaining bits bianry image I _Lb ²The piecemeal scrambling encryption generates 4 bit watermark W to be embedded respectively ₁With remaining bits watermark W ₂

(2), watermark embeds: with 4 bit watermark W of (1) step generation ₁The lowest order that is embedded into the zero setting image I, remaining bits watermark W ₂Be embedded into the zero setting image I inferior low level of zero setting, generate the watermarking images I that is embedded with watermark ^w

(3), watermark extracting:

Low bit recovery: to watermarking images I ^wThe tested image I of receiving after the transmission ^*, get its lowest order and obtain tested image I ^*Low level take out 4 bit watermark W ₁', according to key K ₀Get tested image I ^*The inferior low level of respective pixel obtains tested image I ^*Remaining bits watermark W ₂', again according to key K ₁And K ₂The backwards calculation course recovery of same procedure goes out to be stored in low frequency compressed image I in the watermark by (1) step _LL';

High-order calculating: with tested image I ^*The lowest order zero setting of each pixel, and based on key K ₀The inferior low level zero setting of picked at random partial pixel is made two-dimentional one-level wavelet decomposition to it: LL, and LH, HL, HH} adopts and identical method and key K of (1) step ₃And K ₁, low frequency wavelet coefficient LL is wherein calculated the tested image I of generation ^*Low frequency compressed image I _LL ^*Calculate 4 bit watermark W with a high position ₁ ^*

(4) authentication:

The low frequency compressed image I that (3) step is gone out from watermark recovery _LL' with tested image I ^*Low frequency compressed image I _LL ^*Subtract each other, obtain image difference ${\mathrm{\&Delta;I}}_{\mathrm{LL}}=I_{\mathrm{LL}}^{\&prime;}-I_{\mathrm{LL}}^{*};$ If image difference Δ I _LLBe zero entirely, assert tested image I ^*Image information and watermark all do not distorted; If image difference Δ I _LLIn have some non-zero points to concentrate on certain zone, and this region area Δ S assert tested image I greater than threshold value template T ^*The image information of middle corresponding region is distorted, otherwise, assert tested image I ^*Image information do not distorted; If compressed image difference DELTA I _LLIn be not the non-zero points of stochastic distribution, assert tested image I ^*Watermark do not distorted, otherwise assert tested image I ^*Watermark distorted; The tampered position of watermark is then by the watermark difference $\mathrm{\&Delta;W}=|{\mathrm{<figure-callout\; id="1"\; label="W"\; filenames="A20041008140000172.png,A20041008140000191.png"\; state="\{\{state\}\}">W</figure-callout>}}_1^{\&prime;}-{\mathrm{<figure-callout\; id="1"\; label="W"\; filenames="A20041008140000172.png,A20041008140000191.png"\; state="\{\{state\}\}">W</figure-callout>}}_1^{*}|$ Determine that this watermark difference DELTA W takes out 4 bit watermark W for the low level that obtains in (3) step ₁' calculate 4 bit watermark W with a high position ₁ ^*The absolute value that subtracts each other, non-zero points concentrated area among the watermark difference DELTA W is exactly tested image I ^*The position that middle watermark is distorted;

(5), recover: watermark is not distorted if the authentication identification in (4) step is distorted by the image information of altimetric image, then directly utilizes image difference Δ I _LLThe positioning tampering position also utilizes the low frequency compressed image I that recovers from watermark _LLThe information of ' relevant position, the tested image I of approximate recovery ^*Middle image information of being distorted; If assert that image information is distorted, watermark is also distorted, then earlier to image difference Δ I _LLDo " corrosion-expansion " and handle,, utilize the low frequency compressed image I that from watermark, recovers again with the position that accurate positioning image information is distorted _LLThe information of ' relevant position, the tested image I of approximate recovery ^*Middle image information of being distorted.

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

1, the present invention can detect the tampered region greater than the setting threshold template and will not guarantee the reliability and the authenticity of digital picture by authentication; Simultaneously can also distinguish distorting to watermark, to this watermark that does not influence image's authenticity distort give by the authentication, on the basis that guarantees image's authenticity, improved the efficient of the application and the exchange of digital picture, help digitized authentication image The Application of Technology and popularization.

Low frequency compressed image I by from watermark, recovering when 2, authenticating _LL' with tested image I ^*Low frequency compressed image I _LL ^*Subtract each other the image difference that obtains, intuitively orient the position that image information is distorted, the intensity that difference size reflection correspondence image information is distorted; By tested image I ^*Low level take out 4 bit watermark W ₁' calculate 4 bit watermark W with a high position ₁ ^*Subtract each other the watermark difference that (taking absolute value) obtains, intuitively orient the position that watermark is distorted.

3, when image information is distorted, the present invention can adopt diverse ways to recover the image information of being distorted under the prerequisite whether the differentiation watermark is distorted.When making the watermark of certain limit be distorted, the present invention still can be similar to effectively and recover the image information of being distorted, and guarantees the availability of image to greatest extent.

Can be when 4, generating watermark according to the purposes of image and different requirements to recovery effects, the small echo low frequency coefficient is done the quantification of different bits, generate the watermark of a constant volume; Take into account watermark and embed capacity and invisibility requirement, the watermark that generates is embedded into the lowest order and the part time low level of original image.

5, when generating watermark, utilize the embedded location of cipher controlled non-uniform quantizing step-length and time low level watermark, enlarged key space, make watermarking algorithm safer.

Above-mentioned based on key K ₀During the inferior low level zero setting of picked at random partial pixel, the object of its picked at random both can be single pixel, also can be the zone that plurality of pixels is formed.

Embodiment

The invention will be further described below in conjunction with embodiment.

Fig. 1 generates the block diagram of step for embodiment of the invention watermark

Fig. 2 handles synoptic diagram for the non-average scalar quantization that embodiment of the invention watermark generates in the step

Fig. 3 is the watermark extracting of the embodiment of the invention and the block diagram of authenticating step

Fig. 4 generates and the emulation legend that authenticates for carry out watermark with the inventive method, and wherein: I is an original image, I ^wBe watermarking images, Δ I _LLBe image difference, Δ W is the watermark difference.

Fig. 5 is for to distort and to recover by the emulation legend of tampered image information with the inventive method differentiate between images and watermark.Wherein: I ₁ ^*For image information and watermark all distorted by altimetric image, I ₂ ^*For image information distorted by altimetric image, I ₃ ^*For watermark distorted by altimetric image; Δ I _LL(1), Δ I _LL(2) and Δ I _LL(3) be respectively and I ₁ ^*, I ₂ ^*And I ₃ ^*Corresponding image difference, Δ W (1), Δ W (2) and Δ W (3) are respectively and I ₁ ^*, I ₂ ^*And I ₃ ^*Corresponding watermark difference.I _C(1) and I _C(2) be respectively and I ₁ ^*And I ₂ ^*Corresponding approximate recovery image, I _C' (1) is to utilize the photoshop software for editing to I _C(1) the recovery image behind the removal noise spot.

Embodiment

A kind of image is distorted generation and the authentication method of content positioned authenticating water-mark for recoverying, comprises the steps:

One, watermark generates:

Fig. 1 illustrates the watermark generative process: with the lowest order zero setting of each pixel of original image I, and based on key K ₀The inferior low level zero setting of picked at random partial pixel obtains the zero setting image

This zero setting image I is made two-dimentional one-level wavelet decomposition: LL, and LH, HL, HH} does wherein low frequency wavelet coefficient LL based on key K ₃Generate the non-average scalar quantization of the 5-7 bit of random series, generate low frequency compressed image I _LLPicked at random 4 bits wherein series arrangement in the corresponding space of its pixel generates and 4 onesize bit bianry image I of original image _Lb ¹, remaining 1-3 bit is arranged in any order and is generated remaining bits bianry image I _Lb ²Utilize key K again ₁And key K ₂Generate different random seriess respectively to 4 bit bianry image I _Lb ¹With remaining bits bianry image I _Lb ²The piecemeal scrambling encryption generates 4 bit watermark W to be embedded respectively ₁With remaining bits watermark W ₂

In this example: non-average scalar quantization is adopted based on key K ₃Generate the non-average scalar quantization of 5 bits of random series, original image I zero setting is that inferior low level in its lowest order and picked at random 1/4th carries out, and the zero setting position is used for embed watermark, and is identical with the corresponding zero setting image of watermarking images with the assurance original image.The employing of choosing of four/low levels is chosen a pixel, key K from the image-region (piece) of 2*2 ₀Control one of picked at random from four pixels, K ₀Span be 1,2,3 and 4 four integer.Formula to watermark embedded location zero is described as:

Wherein, i=0,1 ..., (m/2)-1, j=0,1 ..., (n/2)-1,

It is that the image of m*n is made two-dimentional one-level wavelet decomposition to size that present embodiment is selected the DB1 wavelet basis, and wherein low frequency wavelet coefficient LL is done based on key K ₃The non-average scalar quantization of 5 bits, generate low frequency compressed image I _LLFig. 2 shows the process and method of non-mean quantization, is described as with formula:

I _LL＝Q(LL) (2)

Rule of correspondence Q is:

Wherein, i=1,2 ..., m/2, j=1,2 ..., n/2,

Be called the mean quantization step-length, max and min are respectively the maximal value and the minimum value of element among the LL.{ δ _k, k=0,1,2 ... 31} is a random series.

Present embodiment adopts the logistic chaotic maps to generate random series { δ _k, k=0,1,2 ... 31}, the Logistic mapping definition is as follows:

δ _k+1＝μδ _k(1-δ _k) (3)

Wherein, 0≤δ _k≤ 1,0≤μ≤4 are called branch parameter.When 3.58＜μ≤4, the logistic mappings work is in chaos state, μ in the present embodiment=4.Key K ₃Initial value as this chaotic maps.

Because the chaotic maps One's name is legion, and chaos system is to the extreme sensitivity of initial value, good randomness and the characteristics of regeneration easily, thereby present embodiment adopts the random series of chaos sequence as the fine setting quantization step, has better security.

This routine medium and low frequency compression figure I _LLEach element be converted to 5 scale-of-two, get 4 at random and generate 4 bit bianry image I _Lb ¹, remaining bit generates remaining bits bianry image I _Lb ²Formula specifically describes as follows:

${\left(I_{\mathrm{LL}}\right)}_{(m/2)*(n/2)}\&RightArrow;{\left(I_{\mathrm{Lb}}^k\right)}_{(m/2)*(n/2)},k=\mathrm{1,2}---\left(4\right)$

Wherein, ${\left(I_{{\mathrm{LL}}_{\mathrm{ij}}}\right)}_{10}={\left(b_4{b}_3{b}_2{b}_1{b}_0\right)}_2,$ $I_{{\mathrm{<figure-callout\; id="1"\; label="Lb\; ij"\; filenames="A20041008140000172.png,A20041008140000191.png"\; state="\{\{state\}\}">Lb</figure-callout>}}_{\mathrm{ij}}}^1\&RightArrow;\left[\begin{array}{cc}{b}_3& b_2\\ b_1& {\mathrm{<figure-callout\; id="0"\; label="b"\; filenames="A20041008140000172.png"\; state="\{\{state\}\}">b</figure-callout>}}_0\end{array}\right],$ ${\mathrm{<figure-callout\; id="2"\; label="I\; Lb\; ij"\; filenames="A20041008140000172.png,A20041008140000191.png"\; state="\{\{state\}\}">I</figure-callout>}}_{{\mathrm{Lb}}_{\mathrm{ij}}}^{{}_{}}\&RightArrow;\left[b_4\right]$

According to the key K of setting ₁And K ₂Generate two random seriess, respectively to 4 bit bianry image I _Lb ¹With remaining bits bianry image I _Lb ²Carry out the piecemeal scrambling encryption, the bianry image after the encryption is exactly 4 bit watermark W to be embedded ₁With remaining bits watermark W ₂

Two, watermark embed process:

Watermark embeds: with 4 bit watermark W of step generation ₁The lowest order that is embedded into the zero setting image I, remaining bits watermark W ₂Be embedded into the zero setting image I inferior low level of zero setting, generate the watermarking images I that is embedded with watermark ^w

Embedding grammar is described as with formula:

I ^w(k _i+2*i，k _j+2*j)＝ I(k _i+2*i，k _j+2*j)+W ₂ (5)

I ^w＝I ^w*2+W ₁

Wherein, i=0,1 ..., (m/2)-1, j=0,1 ..., (n/2)-1,

k _j=mod (K ₀, 2)

Three, watermark extracting:

Fig. 3 illustrates watermark extraction process of the present invention:

Low bit recovery: to watermarking images I ^wThe tested image I of receiving after the transmission ^*, get its lowest order and obtain tested image I ^*Low level take out 4 bit watermark W ₁', according to key K ₀Get by the inferior low level of altimetric image respective pixel and obtain by altimetric image remaining bits watermark W ₂', again according to key K ₁And K ₂Backwards calculation course recovery by a step same procedure goes out to be stored in low frequency compressed image I in the watermark _LL';

High-order calculating: with tested image I ^*The lowest order zero setting of each pixel, and based on key K ₀The inferior low level zero setting of picked at random partial pixel is made two-dimentional one-level wavelet decomposition to it: LL, and LH, HL, HH} adopts and identical method and key K of a step ₃And K ₁, low frequency wavelet coefficient LL is wherein calculated the tested image I of generation ^*Low frequency compressed image I _LL ^*Calculate 4 bit watermark W with a high position ₁ ^*

At the zero setting position of original image embed watermark, make that the zero setting image of watermarking images and original image is identical.If do not distorted, according to tested image I by altimetric image ^*Low frequency compressed image I by same procedure and same key generation _LL ^*With 4 bit watermark W ₁ ^*Low frequency compressed image I with original image _LLWith 4 bit watermark W ₁Identical.

Four, authentication:

The verification process that Fig. 3 illustrates present embodiment is: in three step watermark extraction process, and the low frequency compressed image I that from watermark, recovers _LL' with by the low frequency compressed image I of altimetric image _LL ^*Subtract each other, obtain image difference

${\mathrm{\&Delta;I}}_{\mathrm{LL}}=I_{\mathrm{LL}}^{\&prime;}-I_{\mathrm{LL}}^{*}.$

If image difference Δ I _LLBe zero entirely, assert tested image I ^*Image information and watermark all do not distorted; As the Δ I among Fig. 4 _LLShown in.

If image difference Δ I _LLIn have some non-zero points to concentrate on certain zone, and this region area Δ S assert tested image I greater than threshold value template T ^*The image information of corresponding region is distorted, otherwise, assert tested image I ^*Image information do not distorted; As Fig. 5 Δ I _LL(1) and Δ I _LLThe non-zero points concentration zones that goblet shape is arranged (2) can be assert corresponding tested image I ₁ ^*And I ₂ ^*Distorted in this regional image information, and reflected the shape (goblet shape) of being distorted part, Δ I exactly _LL(1) and Δ I _LL(2) gray-scale value that should the zone in can intuitively reflect the intensity of distorting to this area image information.

If compressed image difference DELTA I _LLIn be not the non-zero points of stochastic distribution, assert tested image I ^*Watermark do not distorted, otherwise assert tested image I ^*Watermark distorted; The tampered position of watermark is then by the watermark difference $\mathrm{\&Delta;W}=|{\mathrm{<figure-callout\; id="1"\; label="W"\; filenames="A20041008140000172.png,A20041008140000191.png"\; state="\{\{state\}\}">W</figure-callout>}}_1^{\&prime;}-{\mathrm{<figure-callout\; id="1"\; label="W"\; filenames="A20041008140000172.png,A20041008140000191.png"\; state="\{\{state\}\}">W</figure-callout>}}_1^{*}|$ Determine that this watermark difference DELTA W is that the low level that obtains in three steps takes out 4 bit watermark W ₁' calculate 4 bit watermark W with a high position ₁ ^*The absolute value that subtracts each other, non-zero points concentrated area among the watermark difference DELTA W is exactly tested image I ^*The position that middle watermark is distorted.As Fig. 5 Δ I _LL(1) and Δ I _LL(3) there is the non-zero points (noise) that is stochastic distribution in, assert corresponding tested image I ₁ ^*And I ₃ ^*In have watermark to be distorted, and the non-zero points concentration zones of goblet shape among figure Δ W (1) and the Δ W (3) reflects position and shape that tested image watermark is distorted exactly.

Should consider choosing of threshold value template T, the scramble piece the size and by in the altimetric image by the factor of size two aspects of tampered region.In the present embodiment, the size of scramble piece is 2*2 (being the image block that the 2*2 pixel constitutes), and the size of threshold value template T is 3*3.

Five, recover:

The process that present embodiment is distorted recovery is: the authentication identifications in four steps are distorted by the image information of altimetric image and watermark is not distorted, and then directly utilize image difference Δ I _LLThe positioning tampering position also utilizes the low frequency compressed image I that recovers from watermark _LLThe information of ' relevant position, the tested image I of approximate recovery ^*Middle image information of being distorted; Distorted simultaneously if assert image information and watermark, then earlier to image difference Δ I _LLDo " corrosion-expansion " and handle,, utilize the low frequency compressed image I that from watermark, recovers again with the position that accurate positioning image information is distorted _LLThe information of ' relevant position, the tested image I of approximate recovery ^*Middle image information of being distorted.

When with the inventive method the picture material of being distorted part being recovered, if only image information is distorted, the picture quality that then recovers is relatively good; If image information and watermark are distorted simultaneously, the picture quality that then recovers is poor slightly.This is that therefore the image at the quilt place of distorting of recovering also noise spot can occur because expansion-corrosion operation can not be removed the noise spot in the image information tampered region.But,, can utilize image editing software (as photoshop) to be easy to such an extent that the noise spot that recovers in the image is removed, obtain picture quality and recover image preferably because these noise spots clearly.I among Fig. 5 _C(1) and I _C(2) be respectively and I ₁ ^*And I ₂ ^*Corresponding approximate recovery image, I _C' (1) is to utilize the photoshop software for editing to I _C(1) the approximate recovery image behind the removal noise spot.As seen can effectively recover the picture material of being distorted with method of the present invention.

Become to give birth in the process in watermark among the present invention, low frequency wavelet coefficient LL makes the method for non-average scalar quantization except the non-average scalar quantization of 5 bits that can adopt this example, can also adopt the non-average mark quantization methods of 6,7 bits; Opposed null graph is made two-dimentional one-level wavelet decomposition except that can adopting this routine DB1 wavelet basis as I, also can adopt other any existing wavelet basis.

Effect of the present invention can be verified by following performance evaluation:

One, invisibility analysis:

The watermark that fragile watermark requires to add can not be perceiveed, and in order to weigh the difference between watermarking images and the original image, definition Y-PSNR PSNR (Peak Signal-to-Noise Ratio) is:

$\mathrm{PSNR}=10\log 10\left[\frac{255*255}{\frac{1}{m*n}\underset{i=1}{\overset{m}{\mathrm{\&Sigma;}}}\underset{j=1}{\overset{n}{\mathrm{\&Sigma;}}}{[I(i,j)-I^w(i,j)]}^2}\right]---\left(6\right)$

In the present embodiment watermark information is embedded into the LSB position of image and 1/4th inferior low level, because the lowest order of original image and part time each low bit are independently, so [I (i, j)-I ^w(i, j)] ²Mathematical expectation be: $E\left({[I(i,j)-I^w(i,j)]}^2\right)=\frac{3}{4}*(1+0)*\frac{1}{2}+\frac{1}{4}((0+1+4+9)*\frac{1}{4})=\frac{5}{4}$ So the mathematical expectation of original image and watermarking images Y-PSNR is:

$E\left(\mathrm{PSNR}\right)=10\log 10\frac{255*255}{\frac{1}{m*n}\underset{i=1}{\overset{m}{\mathrm{\&Sigma;}}}\underset{j=1}{\overset{n}{\mathrm{\&Sigma;}}}E\left({[I(i,j)-I^w(i,j)]}^2\right)}=47.1617$

As seen, can satisfy the requirement of fragile watermark invisibility at the algorithm of lowest order and four/low level embed watermarks.In like manner, at lowest order and four/secondary low level embed watermark, the mathematical expectation of original image and watermarking images Y-PSNR is: E (PSNR)=45.7004; At lowest order and 3/4ths low level embed watermarks, the mathematical expectation of original image and watermarking images Y-PSNR is: E (PSNR)=43.9395, all can satisfy the requirement of fragile watermark invisibility.The purposes of consideration image and the different requirements to recovery effects can be selected different quantization bits.

Two, tampering location capability analysis:

The present invention passes through image difference ${\mathrm{\&Delta;I}}_{\mathrm{LL}}=I_{\mathrm{LL}}^{\&prime;}-I_{\mathrm{LL}}^{*}$ The position of coming positioning image information to be distorted.The low frequency compressed image quantizes to obtain to the small echo low frequency coefficient, and the ranks number of low frequency compressed image is half of original image.With the DB1 wavelet basis that adopts in the present embodiment is the detectability of distorting that example illustrates this algorithm.

The two-dimensional wavelet transformation of DB1 wavelet basis can be similar to following formula from the mathematics angle and describe:

I _LL(i，j)＝sum(I(2*i-1∶2*i，2*j-1∶2*j))/2 (7)

Wherein, i=1,2 ..., m/2, j=1,2 ..., n/2

The span of eight each pixels of gray level image is [0,255], so I _LLSpan be [0,510], the mean quantization step-length when doing 5 bit average scalar quantizations is q＜(510/32)＜16.To a pixel value or a 2*2 pixel value and change surpass 32, error image is just unequal, thereby can orient by the position of tampered image information, the reflection of difference size is to the intensity of distorting of image information.

Three, analysis is distorted in differentiation:

Identifying algorithm of the present invention passes through image difference ${\mathrm{\&Delta;I}}_{\mathrm{LL}}=I_{\mathrm{LL}}^{\&prime;}-I_{\mathrm{LL}}^{*}$ Position that positioning image is distorted and differentiation are to image information or distorting watermark.

If the low frequency compressed image I that then recovers is not distorted in watermark from watermark _LL' equal the low frequency compressed image of watermarking images I; If watermark is distorted, to distorting after scramble recovers, of localized watermark from the low frequency compressed image I of watermark recovery _LL' on the noise spot of stochastic distribution can appear, but still can reflect the substance of watermarking images.I _LL ^*It is tested image I ^*The low frequency compressed image, when being distorted by the image information of altimetric image, error image Δ I _LLMiddle corresponding position is non-vanishing, and non-vanishing zone is similar to the shape in the zone of distorting; When watermark is distorted, Δ I _LLMiddle non-vanishing the distorting a little of fritter that similar stochastic distribution occurs.

Four, algorithm security analysis

It is disclosed requiring employed watermarking algorithm under the certain situation, and the security of algorithm just relies on key to control so, and therefore rational key space just becomes the important assurance of algorithm security.

In the watermark generating algorithm of the present invention, key K ₀What be used for time low level watermark embedded location chooses K ₃Be used for the adjustment of non-mean quantization step-length, K ₁And K ₂Be used for the piecemeal scrambling encryption, used a plurality of keys, make the security of algorithm of the present invention greatly improve.

The simulation analysis of computer of the inventive method:

To " vase " gray level image of 208*328*8, with Matlab emulation algorithm of the present invention, this emulation is to W ₁The scramble piece be 2*2, (scramble piece choose also as key one one) is to W ₂The scramble piece be 1*1, threshold value template T is 3*3, the invisibility of watermarking images adopts Y-PSNR (dB) to weigh.

Fig. 4 is for carrying out the emulation legend that watermark embeds and authenticates with algorithm of the present invention, I is an original image among Fig. 4, I ^wBe watermarking images, Δ I _LLBe image difference, Δ W is the watermark difference.Watermarking images I after original image I and the adding watermark ^wY-PSNR be 48.1424, show that this algorithm has good imperception, utilize algorithm of the present invention to watermarking images I ^wAuthenticate the visual difference DELTA I that obtains _LLWith watermark difference DELTA W be zero entirely.

Fig. 5 is for to distort and to recover by the emulation legend of tampered image information with the inventive method differentiate between images and watermark.Use the photoshop image editing software at watermarking images I ^wWineglass of last interpolation is designated as I by the image after distorting ₁ ^*By using watermarking images I ^wLowest order and be embedded with the inferior low level substitute I of watermark ₁ ^*Corresponding positions obtain the only tested image I of tampered image information ₂ ^*By using tampered image I ₁ ^*Lowest order and be embedded with inferior low level replacement watermarking images I of watermark ^wCorresponding positions only distorted the tested image I of watermark ₃ ^*Δ I _LL(1), Δ I _LL(2) and Δ I _LL(3) be respectively and I ₁ ^*, I ₂ ^*And I ₃ ^*Corresponding image difference, Δ W (1), Δ W (2) and Δ W (3) are respectively and I ₁ ^*, I ₂ ^*And I ₃ ^*Corresponding watermark difference; I _C(1) and I _C(2) be respectively and I ₁ ^*And I ₂ ^*Corresponding approximate recovery image to distorted image contents, I _C' (1) is to utilize the photoshop software for editing to I _C(1) recovery effects behind the removal noise spot.

Image difference Δ I _LL(1) and Δ I _LL(2) tested image I is judged in the zone that exists non-zero points to concentrate in ₁ ^*And I ₂ ^*This regional image information is distorted, according to the shape decidable of non-zero points concentrated area by the shape of tampered image information (this example is " goblet shape "), and as can be seen to the intensity of distorting of image; Image difference Δ I _LL(1) and Δ I _LL(3) there be " noise " of similar stochastic distribution in, tested image I is described ₁ ^*And I ₃ ^*In have watermark to be distorted, position that watermark is distorted and shape are judged (this example for " goblet shape ") by the non-zero points concentrated area among watermark difference DELTA W (1) and the Δ W (3); Δ I wherein _LL" noise " only arranged (3), and do not have the zone that non-zero points is concentrated, judge this tested image I ₃ ^*Only there is watermark to be distorted.

I _C(1) and I _C(2) be respectively and I ₁ ^*And I ₂ ^*Corresponding approximate recovery image, I _C' (1) is to utilize the photoshop software for editing to I _C(1) recovery effects behind the removal noise spot.

Tampered image I ₁ ^*With watermarking images I ^wY-PSNR be 23.9679, recover image I accordingly _C(1) with watermarking images I ^wY-PSNR be 33.8166, utilize the photoshop software for editing to I _C(1) image I behind the removal noise spot _C' (1) and watermarking images I ^wY-PSNR be 37.5325; Tampered image I ₂ ^*With watermarking images I ^wY-PSNR be 23.9796, recover image I accordingly _C(2) with watermarking images I ^wY-PSNR be 38.3160; Tampered image I ₃ ^*With original watermark image I ^wY-PSNR be 60.8615 because this image only distorted watermark and do not had tampered image information, so adopt recovery algorithms of the present invention recovery image that obtains and the image I of only distorting watermark ₃ ^*Identical (not showing among the figure).

As seen, algorithm of the present invention can be accomplished: accurate the distorting, distinguish and distort intensity of positioning tampering position, differentiate between images information and watermark, and can whether be distorted under the prerequisite by the approximate picture material of being distorted recovered of diverse ways distinguishing watermark.

Claims (2)

1, a kind of generation of positioned authenticating water-mark for recoverying distorted image contents and authentication method comprise the steps:

(1), watermark generates: with the lowest order zero setting of each pixel of original image I, and based on key K ₀The inferior low level zero setting of picked at random partial pixel obtains the zero setting image This zero setting image I is made two-dimentional one-level wavelet decomposition: LL, and LH, HL, HH} does wherein low frequency wavelet coefficient LL based on key K ₃Generate the non-average scalar quantization of the 5-7 bit of random series, generate low frequency compressed image I _LLPicked at random 4 bits wherein series arrangement in the corresponding space of its pixel generates and 4 onesize bit bianry image I of original image _Lb ¹, remaining 1-3 bit is arranged in any order and is generated remaining bits bianry image I _Lb ²Utilize key K again ₁And key K ₂Generate different random seriess respectively to 4 bit bianry image I _Lb ¹With remaining bits bianry image I _Lb ²The piecemeal scrambling encryption generates 4 bit watermark W to be embedded respectively ₁With remaining bits watermark W ₂

(2), watermark embeds: with 4 bit watermark W of (1) step generation ₁The lowest order that is embedded into the zero setting image I, remaining bits watermark W ₂Be embedded into the zero setting image I inferior low level of zero setting, generate the watermarking images I that is embedded with watermark ^w

(3), watermark extracting:

Low bit recovery: to watermarking images I ^wThe tested image I of receiving after the transmission ^*, get its lowest order and obtain tested image I ^*Low level take out 4 bit watermark W ₁', according to key K ₀Get by the inferior low level of altimetric image respective pixel and obtain by altimetric image remaining bits watermark W ₂', again according to key K ₁And K ₂The backwards calculation course recovery of same procedure goes out to be stored in low frequency compressed image I in the watermark by (1) step _LL';

High-order calculating: with tested image I ^*The lowest order zero setting of each pixel, and based on key K ₀The inferior low level zero setting of picked at random partial pixel is made two-dimentional one-level wavelet decomposition to it: LL, and LH, HL, HH} adopts and identical method and key K of (1) step ₃And K ₁, low frequency wavelet coefficient LL is wherein calculated the tested image I of generation ^*Low frequency compressed image I _LL ^*Calculate 4 bit watermark W with a high position ₁ ^*

(4) authentication:

The low frequency compressed image I that (3) step is gone out from watermark recovery _LL' with tested image I ^*Low frequency compressed image I _LL ^*Subtract each other, obtain image difference Δ I _LL=I _LL'-I _LL ^*If image difference Δ I _LLBe zero entirely, assert tested image I ^*Image information and watermark all do not distorted; If image difference Δ I _LLIn have some non-zero points to concentrate on certain zone, and this region area Δ S assert tested image I greater than threshold value template T ^*The image information of middle corresponding region is distorted, otherwise, assert tested image I ^*Image information do not distorted; If compressed image difference DELTA I _LLIn be not the non-zero points of stochastic distribution, assert tested image I ^*Watermark do not distorted, otherwise assert tested image I ^*Watermark distorted; The tampered position of watermark is then by watermark difference DELTA W=|W ₁'-W ₁ ^*| determine that this watermark difference DELTA W takes out 4 bit watermark W for the low level that obtains in (3) step ₁' calculate 4 bit watermark W with a high position ₁ ^*The absolute value that subtracts each other, non-zero points concentrated area among the watermark difference DELTA W is exactly tested image I ^*The position that middle watermark is distorted;

(5), recover: watermark is not distorted if the authentication identification in (4) step is distorted by the image information of altimetric image, then directly utilizes image difference Δ I _LLThe positioning tampering position also utilizes the low frequency compressed image I that recovers from watermark _LLThe information of ' relevant position, the tested image I of approximate recovery ^*Middle image information of being distorted; If assert that image information is distorted, watermark is also distorted, then earlier to image difference Δ I _LLDo " corrosion-expansion " and handle,, utilize the low frequency compressed image I that from watermark, recovers again with the position that accurate positioning image information is distorted _LLThe information of ' relevant position, the tested image I of approximate recovery ^*Middle image information of being distorted.

2, the generation of a kind of positioned authenticating water-mark for recoverying distorted image contents according to claim 1 and authentication method is characterized in that: described based on key K ₀The inferior low level zero setting of picked at random partial pixel, the object of its picked at random both can be single pixels, also can be the zones that plurality of pixels is formed.

Images