CN102096894A - Image fragile watermarking algorithm capable of realizing accurate positioning of tampered region - Google Patents

Abstract

An image is easily tampered in certain application occasions, so that the real content of the image is hard to identify. Therefore, the content of the image is identified by means of tamper detection necessarily. The invention provides an image fragile watermarking algorithm capable of realizing accurate positioning of a tampered region based on discrete cosine transform. The method comprises the following steps: performing the discrete cosine transform on an original image in a partitioning way; adjusting the numerical value of the high-frequency coefficient of the image subblock discrete cosine transform to build the linear relationship of the numerical values between the two different discrete cosine transform coefficients of the same subblock; and obtaining a coefficient-adjusted image by means of the discrete cosine transform. When the content is identified, the tamper detection is realized by judging whether the set numerical value linear relation is existed between the two corresponding discrete cosine transform coefficients of each subblock of the tampered image by utilizing the reversibility of the discrete cosine transform, thus tamper detection is realized. The algorithm provided by the invention can be used for accurately positioning the tampered region of the image.

Description

The pinpoint image Fragile Watermark Algorithm in a kind of realization tampered region

Technical field

The present invention relates to Flame Image Process and information security field.The present invention designs and a kind ofly can realize the pinpoint image Fragile Watermark Algorithm in tampered region, realizes the purpose that picture material is accurately authenticated.

Background technology

In some application scenarios, image is tampered easily, thereby is difficult to distinguish the true content of image.Therefore, be necessary picture material to be authenticated by distorting detection.Digital watermark technology not only can be realized multimedia is carried out copyright protection, can realize again multimedia is carried out content authentication.Whether whether the content of multimedia authentication is meant to confirm whether data are complete, have or not and distort, true reliable with the source.Fragile watermark technology and semi-fragile watermarking technology can be used for realizing multimedia is carried out content authentication, thereby but because performance is slightly different application scenario are separately arranged.

The fragile image digital watermark algorithm of self-embedding [1] is the special fragile image digital watermark algorithm of a class, its special character is that watermark is to produce and be embedded in original image by certain characteristic quantity that coding side extracts original image, rather than extra watermark information is embedded in original image.The way of the fragile image digital watermark algorithm of a more common class self-embedding be with watermark be embedded in original image the lowest bit position (LeastSignificant Bit, LSB).That is, at built-in end, at first the LSB with original image puts 0, puts image after 0 from LSB and extracts certain characteristic quantity and produce watermark, and the LSB that then watermark is embedded in original image obtains containing watermarking images.So just between the high 7 bit contents that contain watermarking images and its LSB, set up mapping relations.In the test side, at first extract the LSB that suffers tampered image, to suffer the LSB of tampered image to put 0 then, and suffer tampered image to extract identical characteristic quantity producing watermark after 0 from putting again, whether the watermark of LSB that extracts by contrast and generation at last consistent distorts detection.If both unanimities then not it is generally acknowledged to be tampered, otherwise it is generally acknowledged and be tampered.This shows that the foundation that the fragile image digital watermark algorithm of self-embedding is distorted detection is to judge whether the mapping relations that exist built-in end to set in the test side.

Researchers have proposed some relatively more classical image authentication watermarking algorithms.Barreto[2] the block fragile watermark algorithm based on PKI is proposed, public key encryption is applied in the Fragile Watermark Algorithm, have good security; Lin[3] utilize before and after the JPEG compression piece to the discrete cosine transform of a relevant position (Discreet Cosine Transformation, DCT) coefficient magnitude concern that unchangeability proposes a kind ofly can distinguish the image authentication algorithm that JPEG compresses and malice is distorted; Li Chun [4] in JPEG compression back great majority this fact that do not change, proposes the semi-fragile watermarking algorithm that a kind of new anti-JPEG compresses according to the magnitude relationship between the adjacent small echo high frequency coefficient of image; Fourth section [5] proposes a kind of fragile watermark technology based on address code, utilize the watermark calculated address sign indicating number behind the scramble, on the piece of the DCT territory of carrier image, find the corresponding reference position according to address code, and change the coefficient value in field, reference position according to chaos sequence, reduced the number of revising carrier image DCT coefficient effectively; Zhang Xianhai [6] proposes a kind of image authentication algorithm based on fragile watermark, use the SHA512 algorithm and produce watermark based on the one-way function of knapsack problem, use sliding window technique and hierarchical structure to come embed watermark, can be with the block of pixels of tampering location to 2 * 2 sizes.

The image authentication watermarking algorithm often requires to keep responsive to distorting, and requires to orient the zone that is tampered.Therefore, whether can accurately orient the zone that is tampered is an important indicator weighing the authenticating water-mark algorithm performance.The present invention sets up the linear relationship of numerical values recited between two different DCT coefficients of same sub-piece by the numerical values recited of adjusting image subblock DCT high frequency coefficient, thereby proposes a kind ofly can realize the pinpoint image Fragile Watermark Algorithm in tampered region.Experimental result of the present invention shows can accurately orient the zone of being distorted.

List of references

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

[2]Barreto?P?S?L?M，Kim?H?Y，Rijmen?V.Toward?Secure?Public-key?BlockwiseFragile?Authentication?Watermarking[A].In：Proceedings?of?2001International?Conference?on?Image?Proceedings[C].Piscataway(NJ)：IEEESignal?Processing?Sac，2002.2：494-497.

[3]Lin?C?Y，Chang?S?F.A?Robust?Image?Authentication?Method?DistinguishingJPEG?Compression?From?Malicious?Manipulation[J].IEEE?Trans.on?Circuitsand?Systems?of?Video?Technology，2001，11(2)：153-168.

[4] Li Chun, Huang is followed the footsteps of. a kind of half fragile image watermark algorithm [J] of anti-JPEG compression. and software journal, 2006,17 (2): 315-324.

[5] fourth section, He Chen, Jiang Lingge, Wang Hongxia. based on the Fragile Watermark Technique [J] of address code. Shanghai Communications University's journal, 2004,38 (4): 620-623.

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

Summary of the invention

The objective of the invention is to design and a kind ofly can realize the pinpoint image Fragile Watermark Algorithm in tampered region, realize picture material is accurately authenticated.

The pinpoint image Fragile Watermark Algorithm in a kind of realization tampered region comprises following two processes:

A, built-in end DCT high frequency coefficient are adjusted;

Detection is distorted in B, test side.

Steps A further comprises following content:

A1: with size is that the size that the original image of M * M is divided into non-overlapping copies is the sub-piece of m * m;

A2: each sub-piece carries out DCT;

A3: adjust the DCT high frequency coefficient according to following rule, thereby between two different DCT coefficients, set up the linear relationship of numerical values recited: order successively

D′(k ₁，l ₁)＝τ*D(x ₁，y ₁)+μ

D′(k ₂，l ₂)＝τ*D(x ₂，y ₂)+μ

.

.

.

D′(k _n，l _n)＝τ*D(x _n，y _n)+μ

Wherein, τ is a weight coefficient, and μ is an increment, and n is the controlled DCT high frequency coefficient of each a sub-piece number, D (x _j, y _j) (1≤x _j≤ m, 1≤y _j≤ m, 1≤j≤n) represent that each sub-piece is in (x _j, y _j) the original DCT coefficient of position, D ' (k _j, l _j) (1≤k _j≤ m, 1≤l _j≤ m, 1≤j≤n) represent that each sub-piece is in (k _j, l _j) DCT high frequency coefficient after being adjusted of position.D ' (k _j, l _j) residing status requirement is different in twos, i.e. k _p=k _qWith l _p=l _q(1≤p≤n, 1≤q≤n, p ≠ q) can not set up simultaneously.And, D ' (k _p, l _p) and D (x _q, y _q) residing position also require in twos different, i.e. k _p=x _qWith l _p=y _q(1≤p≤n, 1≤q≤n) can not set up simultaneously.Like this can be at D (x _j, y _j) and D ' (k _j, l _j) between set up the linear relationship of numerical values recited.The controlled DCT high frequency coefficient of each sub-piece number n can be adjusted according to the characteristics self-adaptation of each sub-piece.

A4: obtain image after the DCT coefficient adjustment by IDCT.

Step B further comprises following content:

B1: with size is that the size that suffers tampered image to be divided into non-overlapping copies of M * M is the sub-piece of m * m;

B2: each sub-piece carries out DCT;

B3: by judging whether whether exist the numerical values recited linear relationship that sets to detect between two corresponding DCT coefficients of each sub-piece is tampered.D " (k _j, l _j) and D " (x _j, y _j) representative suffers two corresponding DCT coefficients of each sub-piece of tampered image.If

D″(k ₁，l ₁)＝τ*D″(x ₁，y ₁)+μ

D″(k ₂，l ₂)＝τ*D″(x ₂，y ₂)+μ

.

.

.

D″(k _n，l _n)＝τ*D″(x _n，y _n)+μ

All set up, think that so this sub-piece is not tampered; Otherwise equation wherein has one or more to be false, and then thinks to be tampered.Distorted sub-piece with deceiving mark entirely with detected.Test side τ, μ, n, x _j, y _j, k _j, l _jValue consistent with built-in end.

The present invention provides a new way for realizing that picture material accurately authenticates.The present invention sets up the linear relationship of numerical values recited between two different DCT coefficients of same sub-piece by the numerical values recited of adjusting image subblock DCT high frequency coefficient, thereby proposes a kind ofly can realize the pinpoint image Fragile Watermark Algorithm in tampered region.Experimental result of the present invention shows can accurately locate the zone of being distorted.

Description of drawings

Fig. 1 is the process flow diagram that built-in end DCT high frequency coefficient is adjusted, and Fig. 2 is the process flow diagram that detection is distorted in the test side.

Fig. 3 is original Lena image, and Fig. 4 is that coefficient adjustment is after the Lena bmp image that IDCT obtains.

Fig. 5 is the Kids image, and Fig. 6 is a tampered image, and Fig. 7 distorts detected image.

Fig. 8 is a tampered image, and Fig. 9 distorts detected image.

Figure 10 is a tampered image, and Figure 11 distorts detected image.

Embodiment

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

1, the definition of DCT and characteristic

DCT is a kind of real number field conversion, and transformation kernel is the cosine function of real number.DCT is widely used in JPEG, MPEG, H.261 waits international coding standard.The two-dimensional dct of a M * N original signal A is defined as follows:

$C_{\mathrm{pq}}=a_p{a}_q\underset{m=0}{\overset{M-1}{\mathrm{\Σ}}}\underset{n=0}{\overset{N-1}{\mathrm{\Σ}}}{A}_{\mathrm{mn}}\cos \frac{\mathrm{\π}(2m+1)p}{2M}\cos \frac{\mathrm{\π}(2n+1)q}{2N}---\left(1\right)$

Wherein

0≤p≤M-1,0≤q≤N-1, A _MnBe original signal component, C _PqBe the DCT coefficient.The two-dimensional inverse discrete cosine conversion (Inverse Discrete Cosine Transformation IDCT) is defined as follows:

$A_{\mathrm{mn}}=\underset{p=0}{\overset{M-1}{\mathrm{\Σ}}}\underset{q=0}{\overset{N-1}{\mathrm{\Σ}}}{a}_p{a}_q{c}_{\mathrm{pq}}\cos \frac{\mathrm{\π}(2m+1)p}{2M}\cos \frac{\mathrm{\π}(2n+1)q}{2N}---\left(2\right)$

0≤m≤M-1 wherein, 0≤n≤N-1.

Original signal is transformed into DC coefficient, low frequency coefficient, intermediate frequency coefficient and high frequency coefficient through DCT.DCT has " concentration of energy " characteristic, and promptly the energy of original signal mainly concentrates on DC coefficient and low frequency coefficient, and the energy of high frequency coefficient is minimum.DCT has decorrelation, and promptly when the statistical property of original signal during near Markov process (Markov processes), its decorrelation approaches K-L (Karhunen-Loeve) conversion.In video compress, DCT is considered to the accurate optimal mapping of performance near Karhunen-Loeve transformation.In addition, from two-dimensional dct definition (1) and two-dimentional IDCT definition (2) reversibility of DCT, i.e. original signal component A as can be seen _MnBe transformed into DCT coefficient C through formula (1) _Pq, pass through formula (2) again and can undistortedly revert to the original signal component A _MnIn like manner, original DCT coefficient C _PqBe transformed into component of signal A through formula (2) _Mn, pass through formula (1) again and can undistortedly revert to original DCT coefficient C _Pq

The present invention utilizes the reversibility of DCT to realize content authentication.The numerical values recited of at first adjusting the sub-piece DCT of original image high frequency coefficient is set up the linear relationship of numerical values recited between two different DCT coefficients of same sub-piece, obtains image after the coefficient adjustment by IDCT then.If this sub-piece is not tampered, pass through DCT so after, still keep the numerical values recited linear relationship set between two corresponding DCT coefficients of this sub-piece.If this sub-piece is tampered, pass through DCT so after, will be difficult to keep the numerical values recited linear relationship that sets between two corresponding DCT coefficients of this sub-piece, distort thereby can detect.

2, DCT high frequency coefficient regulation rule

According to the reversibility of DCT, the present invention designs to distort by the DCT high frequency coefficient of adjusting each sub-piece of original image and detects rule.The reason of why selecting the DCT high frequency coefficient of each sub-piece to adjust is: after (1) original image carried out piecemeal DCT, the DCT high frequency coefficient contained the minimum energy of original image.Have close visual effect for the assurance coefficient adjustment after the image of IDCT is compared with original image, select the DCT high frequency coefficient is adjusted.(2) than other coefficients, the DCT high frequency coefficient is the most responsive to external interference, helps distorting detection.

Fig. 1 is the process flow diagram that built-in end DCT high frequency coefficient is adjusted, and comprises following process:

Step1: with size is that the size that the original image of M * M is divided into non-overlapping copies is the sub-piece of m * m;

Step2: each sub-piece carries out DCT;

Step3: adjust the DCT high frequency coefficient according to following rule, thereby between two different DCT coefficients, set up the linear relationship of numerical values recited: order successively

D′(k ₁，l ₁)＝τ*D(x ₁，y ₁)+μ

D′(k ₂，l ₂)＝τ*D(x ₂，y ₂)+μ

.

.

.

D′(k _n，l _n)＝τ*D(x _n，y _n)+μ (3)

Wherein, τ is a weight coefficient, and μ is an increment, and n is the controlled DCT high frequency coefficient of each a sub-piece number, D (x _j, y _j) (1≤x _j≤ m, 1≤y _j≤ m, 1≤j≤n) represent that each sub-piece is in (x _j, y _j) the original DCT coefficient of position, D ' (k _j, l _j) (1≤k _j≤ m, 1≤l _j≤ m, 1≤j≤n) represent that each sub-piece is in (k _j, l _j) DCT high frequency coefficient after being adjusted of position.The implication of formula (3) is (x that is in τ times _j, y _j) the original DCT coefficient D (x of position _j, y _j) with increment μ sum as the new (k that is in _j, l _j) the DCT high frequency coefficient D ' (k of position _j, l _j).D ' (k _j, l _j) residing status requirement is different in twos, i.e. k _p=k _qWith l _p=l _q(1≤p≤n, 1≤q≤n, p ≠ q) can not set up simultaneously.And, D ' (k _p, l _p) and D (x _q, y _q) residing position also require in twos different, i.e. k _p=x _qWith l _p=y _q(1≤p≤n, 1≤q≤n) can not set up simultaneously.Through type (3) can be at D (x _j, y _j) and D ' (k _j, l _j) between set up the linear relationship of numerical values recited.The adjustment of DCT high frequency coefficient is after the visual effect of the image that IDCT obtains depends on τ, μ, n, x _j, y _j, k _j, l _jValue.The controlled DCT high frequency coefficient of each sub-piece number n can be adjusted according to the characteristics self-adaptation of each sub-piece, but for purposes of simplicity of explanation, the present invention is with the n value unanimity of each height piece.

Step4: obtain image after the DCT coefficient adjustment by IDCT.

3, distort detection algorithm

Fig. 2 is the process flow diagram that detection is distorted in the test side, comprises following process:

Step1: with size is that the size that suffers tampered image to be divided into non-overlapping copies of M * M is the sub-piece of m * m;

Step2: each sub-piece carries out DCT;

Step3: by judging whether whether exist the numerical values recited linear relationship that sets to detect between two corresponding DCT coefficients of each sub-piece is tampered.D " (k _j, l _j) and D " (x _j, y _j) representative suffers two corresponding DCT coefficients of each sub-piece of tampered image.If

D″(k ₁，l ₁)＝τ*D″(x ₁，y ₁)+μ

D″(k ₂，l ₂)＝τ*D″(x ₂，y ₂)+μ

.

.

.

D″(k _n，l _n)＝τ*D″(x _n，y _n)+μ (4)

All set up, think that so this sub-piece is not tampered; Otherwise,, then think to be tampered if the equation in the formula (4) has one or more to be false.Distorted sub-piece with deceiving mark entirely with detected.τ, μ, n, x in the formula (4) _j, y _j, k _j, l _jValue consistent with formula (3).

4, distort the theoretical analysis that detects false-alarm probability and false dismissal probability

The detection algorithm of distorting according to the 3rd part carries out theoretical analysis to distorting detection false-alarm probability and false dismissal probability.

False-alarm probability is meant the probability that the detecting device report is distorted under situation about not being tampered.According to the reversibility of DCT, be not tampered as the fruit piece, so D " (k _j, l _j) and D " (x _j, y _j) between still keeping the numerical values recited linear relationship, so the false-alarm probability P of each sub-piece _FAEqual 0.

False dismissal probability is meant that the report of under situation about being tampered detecting device does not have the probability of distorting.After being tampered, establish D " (k _j, l _j) and D " (x _j, y _j) between still to keep the probability of numerical values recited linear relationship be δ, the DCT coefficient is all kept the probability of numerical values recited linear relationship is δ to n so ⁿTherefore, the false dismissal probability P of each sub-piece _MEqual δ ⁿThe DCT high frequency coefficient is represented the HFS of original image, even in the spatial domain original image is done very little change, high frequency coefficient also can change on the DCT territory but be reflected at, thus the DCT high frequency coefficient relatively sensitivity be changed easily.When so the group piece is tampered, D " (k _j, l _j) and D " (x _j, y _j) between still keep the probability δ of numerical values recited linear relationship to level off to 0, so false dismissal probability P of each sub-piece _MTo level off to 0 with n index.

Embodiment:

1, built-in end DCT high frequency coefficient is adjusted

Original Lena image is that size is 512 * 512 256 gray level bmp images, sees Fig. 3.Original Lena image segmentation is become 8 * 8 sub-pieces of non-overlapping copies.Make τ=1, μ=0.05 and n=6, adjust each sub-piece corresponding D CT high frequency coefficient according to DCT high frequency coefficient regulation rule at built-in end.The adjustment scheme is specific as follows:

D′(5，6)＝D(3，5)+0.05

D′(8，4)＝D(5，8)+0.05

D′(7，8)＝D(8，7)+0.05

D′(8，6)＝D(7，6)+0.05

D′(8，5)＝D(7，7)+0.05

D′(8，3)＝D(7，5)+0.05 (5)

Adjusted DCT high frequency coefficient is IDCT with the original DCT coefficient that other do not adjust, obtains size and be 512 * 512 numerical matrix, then this numerical matrix is saved as the bmp image that data type is the unit8 type, see Fig. 4.The matrix of differences of preserving between this numerical matrix and Fig. 4 is used to distort detection.PSNR between original image and Fig. 4 is 40.7178, so algorithm has good imperception.

2, detection is distorted in the test side

Because there is round-off error in the matlab numerical evaluation, actual distorting when detecting need be revised formula (4).Formula (4) is modified to:

|D″(k _i，l _i)-τ*D″(x _i，y _i)-μ|≤λ (6)

Wherein λ is very little positive number.During experiment, the λ value is 0.000000001.Correspondingly, according to formula (5) and formula (6), distort the detection rule and be modified to:

|D″(5，6)-D″(3，5)-0.05|≤λ

|D″(8，4)-D″(5，8)-0.05|≤λ

|D″(7，8)-D″(8，7)-0.05|≤λ

|D″(8，6)-D″(7，6)-0.05|≤λ

|D″(8，5)-D″(7，7)-0.05|≤λ

|D″(8，3)-D″(7，5)-0.05|≤λ (7)

When all inequality of formula (7) are all set up, think that then this sub-piece is not tampered; Otherwise,, think that then this sub-piece is tampered if the inequality in the formula (7) has one or more to be false.With original Lena image as a setting, distorted sub-piece and carried out mark detected with black.

A1: with the gray level image Kids (see figure 5) of 64 * 64 sizes replace Fig. 4 (image after distorting is seen Fig. 6 for 78:141,187:250) Qu Yu content.Image after distorting is added the matrix of differences of being preserved, distort detection then.Distort detected image and see Fig. 7.As seen, can accurately detect the tampered region.

A2: produce one 128 * 128 size 0, the 1} stochastic matrix, be superimposed upon Fig. 4 (image after distorting is seen Fig. 8 for 370:497,125:252) Qu Yu content.As can be seen, { 0, the 1} stochastic matrix is very little for the visual effect influence of image in stack.Image after distorting is added the matrix of differences of being preserved, distort detection then.Distort detected image and see Fig. 9.As seen, can accurately detect the tampered region.

A3: with Fig. 4 (69:164,45:140) area contents replaces that (image after distorting is seen Figure 10 for 121:216,121:216) Qu Yu content.Image after distorting is added the matrix of differences of being preserved, distort detection then.Distort detected image and see Figure 11.As seen, can accurately detect the tampered region.

From a1-a3 as can be seen, the tampered region can accurately be detected, thereby has verified that also preamble is about distorting the correctness of the theoretical analysis that detects false-alarm probability and false dismissal probability.

3, sum up

In some application scenarios, image is tampered easily, thereby is difficult to distinguish the true content of image.Therefore, be necessary picture material to be authenticated by distorting detection.The present invention provides a new way for realizing that picture material accurately authenticates.The present invention carries out piecemeal DCT to original image, and the numerical values recited of adjusting image subblock DCT high frequency coefficient is set up the linear relationship of numerical values recited between two different DCT coefficients of same sub-piece, obtains image after the coefficient adjustment by IDCT.Owing to have only high frequency coefficient to be adjusted, algorithm has good imperception.During content authentication, utilize the reversibility of DCT, whether existed between two corresponding DCT coefficients of each sub-piece of tampered image the numerical values recited linear relationship that sets to realize distorting detection by judging.Theoretical analysis shows that the false-alarm probability that each sub-piece is distorted detection equals 0, and false dismissal probability levels off to 0 with n index.The present invention has carried out some and has distorted test experience, and experimental result shows can accurately locate the zone of being distorted.

Claims (3)

1. realize the pinpoint image Fragile Watermark Algorithm in tampered region for one kind, can be used for picture material is accurately authenticated, comprise following two processes:

A, built-in end DCT high frequency coefficient are adjusted;

Detection is distorted in B, test side.

2. the pinpoint image Fragile Watermark Algorithm in a kind of realization according to claim 1 tampered region, steps A further comprises following content:

A1: with size is that the size that the original image of M * M is divided into non-overlapping copies is the sub-piece of m * m;

A2: each sub-piece carries out DCT;

A3: adjust the DCT high frequency coefficient according to following rule, thereby between two different DCT coefficients, set up the linear relationship of numerical values recited: order successively

D′(k ₁，l ₁)＝τ*D(x ₁，y ₁)+μ

D′(k ₂，l ₂)＝τ*D(x ₂，y ₂)+μ

.

.

.

D′(k _n，l _n)＝τ*D(x _n，y _n)+μ

Wherein, τ is a weight coefficient, and μ is an increment, and n is the controlled DCT high frequency coefficient of each a sub-piece number, D (x _j, y _j) (1≤x _j≤ m, 1≤y _j≤ m, 1≤j≤n) represent that each sub-piece is in (x _j, y _j) the original DCT coefficient of position, D ' (k _j, l _j) (1≤k _j≤ m, 1≤l _j≤ m, 1≤j≤n) represent that each sub-piece is in (k _j, l _j) DCT high frequency coefficient after being adjusted of position.D ' (k _j, l _j) residing status requirement is different in twos, i.e. k _p=k _qWith l _p=l _q(1≤p≤n, 1≤q≤n, p ≠ q) can not set up simultaneously.And, D ' (k _p, l _p) and D (x _q, y _q) residing position also require in twos different, i.e. k _p=x _qWith l _p=y _q(1≤p≤n, 1≤q≤n) can not set up simultaneously.Like this can be at D (x _j, y _j) and D ' (k _j, l _j) between set up the linear relationship of numerical values recited.The controlled DCT high frequency coefficient of each sub-piece number n can be adjusted according to the characteristics self-adaptation of each sub-piece.

A4: obtain image after the DCT coefficient adjustment by IDCT.

3. the pinpoint image Fragile Watermark Algorithm in a kind of realization according to claim 1 tampered region, step B further comprises following content:

B1: with size is that the size that suffers tampered image to be divided into non-overlapping copies of M * M is the sub-piece of m * m;

B2: each sub-piece carries out DCT;

B3: by judging whether whether exist the numerical values recited linear relationship that sets to detect between two corresponding DCT coefficients of each sub-piece is tampered.D " (k _j, l _j) and D " (x _j, y _j) representative suffers two corresponding DCT coefficients of each sub-piece of tampered image.If

D″(k ₁，l ₁)＝τ*D″(x ₁，y ₁)+μ

D″(k ₂，l ₂)＝τ*D″(x ₂，y ₂)+μ

.

.

.

D″(k _n，l _n)＝τ*D″(x _n，y _n)+μ

All set up, think that so this sub-piece is not tampered; Otherwise equation wherein has one or more to be false, and then thinks to be tampered.Distorted sub-piece with deceiving mark entirely with detected.Test side τ, μ, n, x _j, y _j, k _j, l _jValue consistent with built-in end.

Images