CN1246779C - Embedding and extracting method for wavelet domain chaotic semi-frageile digit water print - Google Patents

Abstract

The present invention relates to an embedding and extracting method for wavelet domain chaotic semi-fragile digit water prints. The embedding and extracting method for wavelet domain chaotic semi-fragile digit water prints comprises the following steps: LL2, HL2, LH2, HH2, HL1, LH1 and HH1 are obtained by the two dimensional two-stage wavelet decomposition of the original image, wherein the gray level of an LL2 sub-layer is mapped into a chaotic initial value, and the water print image is generated by a plurality of times of iteration; the tamper of the added water print image is accurately positioned by the extreme sensitivity of chaos for the initial value and the good time-frequency characteristic of a wavelet according to generated water print image quantization middle-frequency coefficients HL2, LH2 and HH2, and the robustness of non-malicious attack is obtained by flexibly regulating quantization parameters delta1 and delta2. The embedding and extracting method for wavelet domain chaotic semi-fragile digit water prints has the advantages of stable performance, high achieving speed, no need of the original image in the extraction of the water prints, strong robustness for the non-malicious attack such as a certain degree of JPEG compression and channel additive Gaussian noise and good tamper positioning and certification capability for common malicious attack, such as the increase of partial image contents, the tamper of the partial image contents, etc.

Description

Wavelet field chaos semi-fragile watermarking embeds and extracting method

Technical field:

The present invention relates to a kind of wavelet field chaos semi-fragile watermarking and embed and extracting method, is that a kind of semi-fragile watermarking that is specially adapted to authentication, content tampering location and the integrity certification of digital product embeds and extracting method.

Background technology:

In recent years, digital watermark technology more and more demonstrates its superior performance on the copyright protection of digital product and content authentication.The research of robust digital watermark that is used for copyright protection is more deep, and the robust watermarking of function admirable should be designed to resist most attacks, is suffering under these copyright informations of attacking the back digital products still can preserve.But consider the content integrity authentication of digital product, and the confidence level of the multi-medium data of transmission, this digital watermark can't adapt to this requirement fully, and the fragile watermark technology can realize this point.Fragile watermark is designed to that generally the overwhelming majority's attack is had fragility.Fragile watermark should satisfy following characteristics usually: (1) invisibility, and the susceptibility that distort the adding watermarking images (2), (3) tampering location ability, (4) extract watermark does not need original image.Two kinds of watermarks of semi-fragile watermarking and front are different, generally are designed to some non-malicious attack is had than strong robustness, simultaneously most malicious attacks had fragile characteristic.This watermark is mainly used in aspects such as copyright protection, tampering location and image integrity proof, is particularly useful for the Copyright Protection in the transmission of wireless channel conditions hypograph.Semi-fragile watermarking is drawn by fragile watermark, is the recent development of fragile watermark technology.In the middle of the fragility/semi-fragile watermarking of research at present, following two kinds of typical methods are arranged:

(1) based on the Hash function method of safety

The Hash function is widely used on conventional cipher is learned, and utilizes the Hash function that the extreme sensitivity of initial value is realized that fragile watermark is to distorting the focus that responsive requirement also is the research of watermark in recent years.List of references " Hierarchical Watermarking for Secure Image Authentication with Localization " (M U Celik, G Sharma, E Saber, et al.IEEE Trans on Image Processing.2002,11 (6): 585-595) at list of references " Public Key Watermark for Image Verification andAuthentication " (P W Wong.IEEE Int Con on Image Processing.1998, on basis 1:455-459), a kind of classification fragile watermark with tampering location ability is proposed, the original image classification, and by Hash function generation summary, pass through afterwards encrypted private key with the watermark XOR, least significant bit (the LSB:Least Significant Bit) plane that is embedded into original image at last, this algorithm reaches by continuous classification bearing accuracy is progressively improved, and its major advantage is to have overcome the ineffectivity of vector being attacked (VQ attack).

But there are a lot of problems equally in these Fragile Watermarking Algorithm based on the Hash function of safety.

Problem one, above-mentioned similar algorithm can't stand JPEG compression or additive white Gaussian noise, and what do not add differentiation also thinks malicious attack to them.

Problem two, above-mentioned algorithm causes embedding content and comprises mutually because different fractions is enjoyed same LSB plane, so the collision problem when needing partitioning algorithm to solve to embed.

Problem three, similar these traditional tampering location methods generally are based on the susceptibility of Hash function to distorting.Hash function calculation complexity height is not suitable for real-time processing, and approved safe Hash function kind is few.

(2) based on the restrained vector quantization semi-fragile watermarking

Vector quantization is widely used in image code domain, in general is a kind of compaction coding method that diminishes.List of references " Semi-fragile image watermarking method based on index constrainedvector quantization " (Lu Z M, Liu C H, Xu D G, et al:Electronics Letters, 2003,39 (1): 35-36) propose a kind of semi-fragile watermarking algorithm (hereinafter to be referred as ICVQ) based on restrained vector quantization.This algorithm is classified Codebook of Vector Quantization according to the embed watermark bit, seeks optimum quantification code book again in the codebook subset under satisfying condition, and reaches the purpose of embed watermark bit with this.This scheme can be tolerated JPEG compression to a certain degree, has tentatively reached the designing requirement of semi-fragile watermarking.But obviously there are many weak points in this scheme:

Problem one can only be resisted not half JPEG compression.

Problem two, the Y-PSNR of the image of embed watermark is too low.

Problem three must design Codebook of Vector Quantization by off-line, and the extraction of watermark must be transmitted this code book.

Finding to sum up, for a well behaved semi-fragile watermarking, must meet following several basic demand: (1) is invisible, and promptly the digital multimedia data behind the embed watermark must have very high Y-PSNR value: (2) have the tamper-proofing ability; (3) can have very strong robustness to non-malicious attack, and malicious attack is had fragility preferably: (4) do not need original image when detecting: (5) have the tampering location ability.And prior art still fails well to address these problems.

Summary of the invention:

The objective of the invention is at the deficiencies in the prior art, provide a kind of location type chaos semi-fragile watermarking of wavelet transformed domain to embed and extracting method, common non-malicious attack had very strong robustness as to a certain degree JPEG compression and channel additive Gaussian noise, have good tampering location and proof ability and common malicious attack such as topography's content increased, distorts etc., speed advantage is obvious, and visual effect is good.

For realizing such purpose, the present invention through two-dimentional secondary wavelet decomposition, obtains original image: { LL ₂, HL ₂, LH ₂, HH ₂, HL ₁, LH ₁, HH ₁, LL wherein ₂The pixel gray-scale value of sublayer is mapped as the chaos initial value, generates watermarking images through the several times chaos iteration, quantizes intermediate frequency coefficient { HL according to the watermarking images that generates then ₂, LH ₂, HH ₂, utilize extreme sensitivity and the small echo good time-frequency characteristic of chaos to initial value, accurately locate adding distorting of watermarking images; By flexible quantization parameter Δ ₁And Δ ₂, obtain robustness to non-malicious attack.

The present invention includes following concrete steps:

(1) initialization.Selective system parameter: Key1, Key2, Δ ₁, Δ ₂Wherein Key1 and Key2 are key, Δ ₁And Δ ₂Be quantization parameter.Original image is carried out two-dimentional secondary wavelet decomposition, obtain: { LL ₂, HL ₂, LH ₂, HH ₂, HL ₁, LH _l, HH ₁, and establish LL ₂The size of sublayer is r * c.With key K ey1 is the initial value of chaotic maps, obtain chaos random series location (i, j) ∈ 0,1,2}, 1≤i≤r 1≤j≤c, this sequence is as the intermediate frequency sublayer of selecting the quantization parameter place; With key K ey2 is the initial value of chaotic maps, and (this sequence will be as the number of times of chaos iteration for i, j) 1≤i≤r 1≤j≤c to obtain chaos random series iternum.

(2) generation of watermark:

Be mapped as the initial value of chaos, here

Integer is taken off in expression, through iteration iternum (i, j) after, obtain W (i, j) ∈ 0,1}.To LL ₂All coefficients of sublayer are carried out aforesaid operations, obtain size and are the watermarking images W of r * c.

(3) watermark embeds: (i, j) { 0,1,2}, 1≤i≤r 1≤j≤c select the intermediate frequency sublayer { HL at corresponding quantization parameter place to ∈ according to location ₂, LH ₂, HH ₂, with Δ ₂Be quantization parameter, the intermediate frequency coefficient of odd-even quantization correspondence.Two dimension secondary inverse wavelet transform obtains the image of embed watermark.

(4) extraction of watermark and distort authentication: to above-mentioned 1～2 step of fulfillment of the embed watermark received, first version W ' of the watermark that obtains embedding.In addition, according to location (i, j) ∈ 0,1,2}, 1≤i≤r 1≤j≤c finds that coefficient value that is quantized when embedding, for example, be u (i, j), so the watermark of second version can by

Calculate.The relatively watermark of two versions, calculate distort matrix T=| W '-W " |, be that 1 element has been located tampered position so in the matrix.

Wavelet transformed domain chaos semi-fragile watermarking of the present invention embeds and extracting method, for the authentication and the semi-fragile watermarking of location type have been opened up a new path.Intermediate frequency coefficient embed watermark by behind the modification original image wavelet transformation has obtained higher Y-PSNR.The quantization parameter Δ ₁And Δ ₂The flexible non-malice that can resist to a certain degree on the one hand of adjusting distort, can carry out tampering location to malicious attack on the other hand.The application of chaos random series has increased the security of algorithm.Algorithm of the present invention is simple, realize easily, when extracting watermark, do not need original image, common non-malicious attack had very strong robustness as to a certain degree JPEG compression and channel additive Gaussian noise, have good tampering location and proof ability and common malicious attack such as topography's content increased, distorts etc., speed advantage is obvious, and visual effect is good simultaneously.These good performances have increased its range of application greatly.

Description of drawings:

Fig. 1 is a watermarking algorithm block diagram of the present invention.

The detection embodiment that Fig. 2 increases for picture material.

Wherein, Fig. 2 (a) is the image that is embedded with watermark before distorting, the image of Fig. 2 (b) for distorting, and Fig. 2 (c) is for distorting testing result.

Fig. 3 is the Y-PSNR value of adding watermarking images and the relation of Δ 2.

Embodiment:

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

1) initialization

Initialization step corresponding diagram 1 first frame of broken lines.The selective system parameter comprises Key1, Key2, Δ ₁, Δ ₂Original image is carried out the secondary 2-d wavelet decompose, it is as follows to obtain seven sublayers: { LL ₂, HL ₂, LH ₂, HH ₂, HL ₁, LH ₁, HH ₁, and establish LL ₂The size of sublayer is r * c.With Key1 is the initial value of chaotic maps, obtain chaos random series location (i, j) ∈ 0,1,2}, 1≤i≤r 1≤j≤c, this sequence is as the intermediate frequency sublayer of selecting the quantization parameter place.With Key2 is the initial value of chaotic maps, and (this sequence will be as the number of times of chaos iteration for i, j) 1≤i≤r 1≤j≤c to obtain chaos random series iternum.

In the scheme that the present invention proposes, repeatedly use chaos system to produce pseudo-random sequence.Chaos is nonlinear equation (group) separating under given conditions, exporting at random by an equation (group) of determining of it determines, the initial parameter of chaos system can be used as the key of whole watermaking system, and the characteristics of performance at random that chaos system is good and regeneration have easily increased the security performance of system.Below be example with the one-dimensional discrete chaotic maps, f:U → U, U R

z(n+1)＝f(λ，z(n))，λ∈R，z(n)∈U (1)

N=0 wherein, 1, Λ represents iterations, λ is the parameter of control system chaotic behavior.By a quantization function, can real-valued z (n) be quantified as z (n) ∈ 0,1}.

2) generation of watermark

1 second frame of broken lines of the generation corresponding diagram of watermark.

Be mapped as the initial value of chaos, here

Integer is taken off in expression, through iteration iternum (i, j) after, obtain W (i, j) ∈ 0,1}.To LL ₂All coefficients of sublayer are carried out aforesaid operations, obtain size and are the watermarking images W of r * c.

Here Δ ₁Value reflected the robust performance of watermark, Δ ₁Big more, the robustness that non-malicious attack is resisted in watermark is just big more.But big Δ ₁Value is meaned the minimizing of number of this chaos initial value, and the security of system just reduces like this, so must do compromise on robustness and security of system.

3) watermark embeds

Watermark embeds the 3rd frame of broken lines of corresponding diagram 1.Definition odd-even quantization function: y=g (x, b, d) x ∈ R b ∈ 0,1} d ∈ Z ⁺, it quantizes to corresponding odd even zone to coefficient x according to two-value data b, and d is a quantization parameter.The present invention proposes a kind of simple and practical odd-even quantization method and revise frequency coefficient x, thereby reach the purpose of hiding watermark information.Corresponding relation is as follows:

When out of order watermark bit b=0, guarantee that amended coefficient x is in even zone, at this moment, in even zone, then x revises towards this regional intermediate value as if x, if x is between the left half-court in strange zone, then survey a most contiguous idol zone, and revise, if x is between the right half-court in strange zone towards this regional intermediate value to negative direction, then survey a most contiguous idol zone, and revise towards this regional intermediate value to positive dirction; When b=1, guarantee that amended coefficient is in strange zone, at this moment, in strange zone, then x revises towards this regional intermediate value as if x, if x is between the left half-court in idol zone, then survey a most contiguous strange zone, and revise, if x is between the right half-court in idol zone towards this regional intermediate value to negative direction, then survey a most contiguous strange zone, and revise towards this regional intermediate value to positive dirction.Provide concrete function calculation formula below:

Here,

x _r＝sgn(x)(|x|mod1) (5)

The wavelet field intermediate frequency coefficient is implemented odd-even quantization, and specific implementation process is as follows:

FOR i＝1 to r

FOR j＝1 to c

SWITCH location(i，j)

CASE0：HL ₂(i，j)＝g(HL ₂(i，j)，W(i，j)，Δ ₂)

CASE1：LH ₂(i，j)＝g(LH ₂(i，j)，W(i，j)，Δ ₂)

CASE2：HH ₂(i，j)＝g(HH ₂(i，j)，W(i，j)，Δ ₂)

Finish the telescopiny of watermark.Last two-dimentional two-stage inverse wavelet transform obtains the image of embed watermark.

The Δ equally, here ₂Value also reflect the robust performance of watermark, Δ ₂Big more, the robustness that non-malicious attack is resisted in watermark is just big more, but picture quality also descends thereupon simultaneously.So must robustness and add watermark image do compromise qualitatively.

4) extraction of watermark and the authentication of distorting

Watermark extracting and to distort verification process as follows: to the above-mentioned l of fulfillment～2 steps of the embed watermark received, first version W ' of the watermark that obtains embedding.In addition, according to location (i, j) ∈ 0,1,2}, 1≤i≤r 1≤j≤c finds that coefficient value that is quantized when embedding, for example, be u (i, j), the watermark of second version can be calculated by following formula so:

(6)

The relatively watermark of two versions is calculated and is distorted the detection matrix

T＝|W′-W″| (7)

If W '=W ", T=0 so, this shows that the image that adds watermark was not distorted, otherwise, be that 1 element has been located tampered position in the matrix.Notice that distorting the size that detects matrix is r * c, because what implement is the classification of secondary 2-d wavelet, add 1/16 of watermarking images size so distort the size that detects matrix the chances are, this means to distort to detect and a bit represented in the adding watermarking images one 4 * 4 fritter on the matrix.

5) experimental result

With Matlab 6 these algorithms of simulation.Increase in the experiment in a picture material of the present invention,, make Δ for 512 * 512 * 8 ' Peppers ' gray-scale map ₁=30, Δ ₂=16, the image that adds watermark is shown in Fig. 2 (a), and Y-PSNR is 41.6dB.The Y-PSNR of aforementioned ICVQ method is 31.85dB, and as seen the invisibility of this algorithm is better, and is very little to the modification of original image after the adding watermark.If on the image that adds watermark, add two capsicums again shown in Fig. 2 (b), the algorithm that utilizes the present invention to propose detects this and distorts, the tampered region testing result is shown in Fig. 2 (c), and visible the present invention can point-devicely locate distorting of picture material, and the tampered region profile is obvious.

The normalized cross correlation value of general using on the document (Normalized Cross-correlation NC) represents to extract difference between watermark and the original watermark, and the NC value is defined as follows:

$\mathrm{NC}=\frac{\underset{i=1}{\overset{r}{\mathrm{\Σ}}}\underset{j=1}{\overset{c}{\mathrm{\Σ}}}W(i,j)W^{\′}(i,j)}{\sqrt{\underset{i=1}{\overset{r}{\mathrm{\Σ}}}\underset{j=1}{\overset{c}{\mathrm{\Σ}}}|W(i,j){|}^2}\sqrt{\underset{i=1}{\overset{r}{\mathrm{\Σ}}}\underset{j=1}{\overset{c}{\mathrm{\Σ}}}|W^{\′}(i,j){|}^2}}---\left(8\right)$

Table 1 is being subjected to the different quality factor (Quality Factor for the image that adds watermark, QF) under the additive white Gaussian noise situation under JPEG under compression and the different variances sigma 2, utilize watermark that the used algorithm of the present invention extracts and the NC value between the original watermark.Table 1 has also been listed the NC value that suggest plans obtains in the aforementioned ICVQ method under the JPEG of same quality factor compression simultaneously, and algorithm robustness proposed by the invention obviously obviously is better than the document algorithm.

The NC value of table 1 under different JPEG compressions and additive white Gaussian noise condition.

AWGN： σ 2	6	12	18	24	30	36
							OurNC	0.97	0.94	0.91	0.87	0.83	0.79
JPEG：QF	100	90	80	70	60	50
							OurNC	1.00	0.99	0.97	0.93	0.88	0.82
ICVQ’s NC	0.99	/	0.81	/	/	/

Add watermark image the Y-PSNR value only and Δ ₂Relevant, Fig. 3 has shown Y-PSNR value and the Δ that adds watermarking images ₂Relation.

Claims (1)

1, a kind of wavelet field chaos semi-fragile watermarking embeds and extracting method, it is characterized in that comprising following concrete steps:

1) initialization: selective system parameter: Key1, Key2, Δ ₁, Δ ₂, wherein Key1 and Key2 are key, Δ ₁And Δ ₂Be quantization parameter, original image is carried out two-dimentional secondary wavelet decomposition: { LL ₂, HL ₂, LH ₂, HH ₂, HL ₁, LH ₁, HH ₁, and establish LL ₂The size of sublayer is r * c, with Key1 is the initial value of chaotic maps, obtain chaos random series location (i, j) ∈ { 0,1,2}, 1≤i≤r 1≤j≤c, this sequence is as the intermediate frequency sublayer of selecting the quantization parameter place, with Key2 is the initial value of chaotic maps, (this sequence will be as the number of times of chaos iteration for i, j) 1≤i≤r 1≤j≤c to obtain chaos random series iternum;

2) watermark generates: Be mapped as the initial value of chaos, here

Integer is taken off in expression, through iteration iternum (i, j) after, (i, j) { 0,1} is to LL for ∈ to obtain W ₂All coefficients of sublayer are carried out aforesaid operations, obtain size and are the watermarking images W of r * c; Adjust the quantization parameter Δ ₁Value, make it on the robustness of watermark and security of system, do compromise;

3) watermark embeds: (i, j) { 0,1,2}, 1≤i≤r 1≤j≤c select the intermediate frequency sublayer { HL at corresponding quantization parameter place to ∈ according to location ₂, LH ₂, HH ₂, with Δ ₂Be quantization parameter, the intermediate frequency coefficient of odd-even quantization correspondence, two-dimentional two-stage inverse wavelet transform obtains the image of embed watermark; Wherein, the corresponding odd even regional relation of odd-even quantization method is as follows: when out of order watermark bit b=0, the corresponding region is the idol zone, guarantee that amended coefficient x is in even zone, if this moment, x was in even zone, then x revises towards this regional intermediate value, if x is between the left half-court in strange zone, then survey a most contiguous idol zone to negative direction, and towards this regional intermediate value modification, if x between the right half-court in strange zone, then surveys a most contiguous idol zone to positive dirction, and revises towards this regional intermediate value; When out of order watermark bit b=1, the corresponding region is strange zone, guarantee that amended coefficient x is in strange zone, if x is in strange zone at this moment, then x revises towards this regional intermediate value, if x is between the left half-court in idol zone, then survey a most contiguous strange zone, and revise, if x is between the right half-court in idol zone towards this regional intermediate value to negative direction, then survey a most contiguous strange zone, and revise towards this regional intermediate value to positive dirction; Adjust the quantization parameter Δ ₂Value, make its robustness of watermark and add watermark image do compromise qualitatively;

4) watermark extracting and distort authentication: to above-mentioned 1～2 step of fulfillment of the embed watermark received, first version W ' of the watermark that obtains embedding, according to location (i, j) ∈ 0,1,2}, 1≤i≤r 1≤j≤c, (i j), obtains the watermark of second version to find the coefficient value u that is quantized when embedding The relatively watermark of two versions, calculate distort matrix T=| W '-W " |, be that 1 element has been located tampered position in the matrix.

Images