CN1664849A - Image identification method based on parameter integer wavelet transformation - Google Patents

Abstract

This invention relates to an image identifying technology based on parameter integer wavelet change belonging to multimedia information safety region, which comprises the following steps: first, pre-processing original watermark in double-value format to equalize the ratio of 0 and 1; then, adopting lift lattice parameter integer wavelet to decomposing image data and embedding watermark in changed low-frequency factor; last acquiring image data with watermark through wavelet change. The invention can by identifying validity and integrity of image spreading by network, assure the confidence level of image's transmission and issuance.

Description

Image authentication method based on parameter integer wavelet transformation

Technical field

The present invention is a kind of image authentication technology based on parameter integer wavelet transformation, belongs to the multi-media signal security fields.

Background technology

Digital media (digital picture, digital video, DAB general designation) bringing ultimate attainment audiovisual to enjoy love to people when, also exists weakness: by existing multimedia treatment technology, its content can be modified easily or be synthetic.Because these are by still only wonderful only Xiao of Digital Media after distorting, human eye can not discovered the vestige behind a bit manual amendment, and this just gives heavy strike of author of medium inevitably: the works that work laboriously to create just become other people achievement in seconds.Confirm that whether these Digital Medias are distorted, and need authenticate them.

Traditional authentication techniques based on digital signature (Digital Signature) are carrying out having following deficiency when Digital Media authenticates: 1) efficient is not high; 2) digital signature information and image information binding, vulnerable; 3) can't positioning tampering.This makes becomes new research focus based on fragile watermark Digital Media authentication techniques.Fragile watermark is to any variation sensitivity that adds all in image.Whether, truly whether and whether complete existence by detecting embed watermark guarantee the confidence level of original image.In general, fragile watermark need satisfy following three basic demands: the 1) high susceptibility to distorting; 2) invisibility; 3) security.Fragile watermark can be divided into complete fragile watermark and semi-fragile watermarking again.Fragile watermark does not allow its carrier that a bit variation is arranged fully, and semi-fragile watermarking then allows its carrier can carry out the operation that keeps such as contents such as JPEG lossy compression method, the malice of destroying its content is distorted then very responsive.In actual applications, semi-fragile watermarking has more wide application prospect than complete fragile watermark.

The research of fragile watermark mainly concentrates on the image field at present.Its technology can be divided into two classes: spatial domain technology and transform domain technology.The former comes embed watermark by direct modification image pixel value.The conversion coefficient that the latter then chooses after the image transformation carries out the watermark embedding.Because wavelet transform DWT (Discrete Wavelet Transform) has when good-frequency localization property and be complementary with human visual system HVS (Human Visual System), make it in Joint Photographic Experts Group JPEG2000 of new generation, occupy core position, and replace discrete cosine transform DCT (Discrete CosineTransform) to become the main tool of transform domain digital watermarking algorithm gradually.

In the existing fragile watermark scheme based on traditional DWT, conversion all is to adopt same wavelet basis, if the algorithm victim knows that then original watermark information might illegally be recovered, thereby can carry out multiple attack.In addition, not high based on the watermarking algorithm efficient of traditional DWT, can not satisfy the needs of real system.

Summary of the invention

The objective of the invention is to propose a kind of semi-fragile watermarking technology of the parameter integer wavelet transformation based on Via Lifting Scheme, be used to carry out the authenticity and integrity authentication of digital media data, guarantee the confidence level of media data.

To achieve these goals, the inventive method adopts following four processes:

1) binary watermarking pre-service is with two-value original watermark and two-value pseudo random matrix XOR mutually, the watermark that produces 0 and 1 balanced proportion;

2) structure is based on the parameter integer small echo of Via Lifting Scheme;

3) embed watermark adopts the parameter integer small echo that step 2 constructed that view data is carried out wavelet decomposition, and watermark is embedded in the low frequency sub-band coefficient behind the wavelet transformation;

4) distort detection and authentication, the classifying rules of distorting of its foundation is: (1) is if the nothing modification is then thought in λ=0; (2) as if λ＞0, and η＜τ, then think accidental modification; (3) if η 〉=τ then thinks malicious attack.λ wherein, η be respectively distort detect erroneous point and distort detect be in the erroneous point densely distributed have a few in shared ratio, τ is the real number between predefined 0.5 to 1, generally is taken as 0.55.

Distort when detecting and authenticating,, the type whether view data to be certified is distorted and distorted is made Comprehensive Assessment in conjunction with subjective evaluation and objective evaluation.Subjective evaluation is by eye-observation watermark differential chart, realizes distorting the preliminary judge of detection.Objective evaluation then is according to certain rule, according to distorting classifying rules, makes assessment to distorting type.

Compare with existing fragile watermark technology, the present invention has following advantage:

Wavelet basis when 1) selecting medium to carry out wavelet transformation via parameter has solved medium when carrying out traditional wavelet, the changeless shortcoming of wavelet basis, and security is than the watermarking project height based on traditional wavelet.

2) adopt Via Lifting Scheme to realize integer wavelet transformation, make the watermark authentication scheme more at a high speed with efficiently.

3) in conjunction with subjectivity and objective evaluation, the accuracy of distinguishing accidental modification and malicious attack is improved.

4) can resist quality factor and be low to moderate 40% JPEG (Joint Photograph Coding Experts Group) lossy compression method.

Below we illustrate effect of the present invention from theoretical and experimental data.

1) binary watermarking pre-service

Binary watermarking that the present invention adopts embeds for convenience because wherein 0 and 1 ratio gap is bigger, must carry out pre-service to original watermark.By with the pseudo random matrix of binary watermarking and two-value XOR mutually, can reach the purpose of balanced 0 and 1 ratio.

2) based on the structure of the parameter integer small echo of Via Lifting Scheme

Adopt Via Lifting Scheme to decompose original signal and can divide three steps: cut apart, predict, upgrade.We can construct parameterized integer wavelet by Via Lifting Scheme.For one-dimensional signal { x _l} _{L ∈ Z}, be example with 9/7 biorthogonal wavelet, its parameterized integer wavelet transformation based on Via Lifting Scheme is:

$\left\{\begin{array}{c}{s}_l^{\left(0\right)}=x_{2l}\\ d_l^{\left(0\right)}=x_{2l+1}\end{array}\right.,$ $\left\{\begin{array}{c}{d}_l^{\left(1\right)}=d_l^{\left(0\right)}+\mathrm{Int}\left(\mathrm{\α}(s_l^{\left(0\right)}+s_{l+1}^{\left(0\right)})\right)\\ s_l^{\left(1\right)}=s_l^{\left(0\right)}+\mathrm{Int}\left(\mathrm{\β}(d_l^{\left(1\right)}+d_{l-1}^{\left(0\right)})\right)\end{array}\right.,$

$\left\{\begin{array}{c}{d}_l^{\left(2\right)}=d_l^{\left(1\right)}+\mathrm{Int}\left(\mathrm{\γ}(s_l^{\left(1\right)}+s_{l+1}^{\left(1\right)})\right)\\ s_l^{\left(2\right)}=s_l^{\left(1\right)}+\mathrm{Int}\left(\mathrm{\δ}(d_l^{\left(2\right)}+d_{l-1}^{\left(2\right)})\right)\end{array}\right.,$

$\left\{\begin{array}{c}{d}_l^{\left(3\right)}=d_l^{\left(2\right)}+\mathrm{Int}\left((\mathrm{\ζ}-{\mathrm{\ζ}}^2)s_l^{\left(2\right)}\right)\\ s_l^{\left(3\right)}=s_l^{\left(2\right)}+\mathrm{Int}\left((-1/\mathrm{\ζ})d_l^{\left(3\right)}\right)\end{array}\right.,$ $\left\{\begin{array}{c}{d}_l^{\left(4\right)}=d_l^{\left(3\right)}+\mathrm{Int}\left((\mathrm{\ζ}-1)s_l^{\left(3\right)}\right)\\ s_l^{\left(4\right)}=s_l^{\left(3\right)}+d_l^{\left(4\right)}\end{array}\right.,$ $\left\{\begin{array}{c}{s}_l={s_l}^{\left(4\right)}\\ d_l={d_l}^{\left(4\right)}\end{array}\right.----\left(1\right)$

S wherein _lAnd d _lBe respectively low frequency and high fdrequency component, S ^(l) _l, d ⁽ⁱ⁾ _l(i=0,1,2) is intermediate result, and Int (x) is a bracket function.The relation of parameter beta, γ, δ ζ and α is as follows:

$\left\{\begin{array}{c}\mathrm{\β}=-\frac{1}{4{(1+2\mathrm{\α})}^2}\\ \mathrm{\γ}=\frac{-1-4\mathrm{\α}-4{\mathrm{\α}}^2}{1+4\mathrm{\α}}\\ \mathrm{\δ}=\frac{1}{16}(4-\frac{2+4\mathrm{\α}}{{(1+2\mathrm{\α})}^4}+\frac{1-8\mathrm{\α}}{{(1+2\mathrm{\α})}^2})\\ \mathrm{\ζ}=\frac{2\sqrt{2}(1+2\mathrm{\α})}{1+4\mathrm{\α}}\end{array}\right.----\left(2\right)$

Restructuring procedure then is the inverse process of said process.For 2D signal, then decompose earlier to be listed as again to decompose and obtain wavelet decomposition figure, during reconstruct then earlier in column direction reconstruct, then in line direction reconstruct.

3) embed watermark

Original image is carried out wavelet decomposition, and the coefficient of keeping the score after separating is f, to the subband wavelet coefficient of embed watermark, finishes the embedding of watermark by following formula.

When watermark is " 0 ", embed as follows:

$f^{*}(i,j)=\left\{\begin{array}{cc}R\_\mathrm{LFB}(f(i,j)-01000b,11000b)& C\_\mathrm{LFB}\left(f(i,j)\right)\≤01000b\\ R\_\mathrm{LFB}(f(i,j),11000b)& \mathrm{otherwise}\end{array}\right.$

When watermark is " 1 ", embed as follows:

$f^{*}(i,j)=\left\{\begin{array}{cc}R\_\mathrm{LFB}(f(i,j)+10000b,01000b)& C\_\mathrm{LFB}\left(f(i,j)\right)\≥11000b\\ R\_\mathrm{LFB}(f(i,j),01000b)& \mathrm{otherwise}\end{array}\right.$

Wherein, minimum five of C_LFB (a) expression number a, (a, b) expression replaces among the number a minimum five with number b to R_LFB.

4) watermark extracting

The same with the watermark embedding, at first testing image is carried out wavelet decomposition, the coefficient of keeping the score after separating is that (i j), equally, for the subband wavelet coefficient of embed watermark, calculates f '

$w^{*\′}(i,j)=\left\{\begin{array}{cc}1& \mathrm{LFB}\left(f^{\′}(i,j)\right)=0\\ 0& \mathrm{LFB}\left(f^{\′}(i,j)\right)=1\end{array}\right.$

Here minimum the 5th of LFB (a) expression coefficient a, and the final watermark bit information of extracting is

w′(i，j)＝w ^*′(i，j)p _n(i，j)

P wherein _n(i, an element in the two-value pseudo random matrix that is produced when j) being watermark pretreatment.

5) distort detection and verification process

Make w ' (i, j) watermark for extracting.The watermark differential chart is

D(i，j)＝|w(i，j)-w′(i，j)|

In conjunction with subjective evaluation and objective evaluation, we can judge distorting type.

Subjective evaluation is according to the watermark differential chart, relies on human visual observation, and then the type of whether being distorted and distorting is belonged to accidental attacks or judgement is made in malicious attack.

Objective evaluation be a kind of quantized distort evaluating method.At first need the low frequency sub-band that watermark embeds is done as giving a definition:

Dense point: (i is dense to be meant that its eight close on picture element and have at least one to be to detect erroneous point j) to current watermark detection erroneous point x.

Sparse point: (i is sparse be meant eight to close on the picture element neither one be to detect erroneous point j) to current watermark detection erroneous point x.

Area={ low frequency sub-band picture element sum }; Area _{, dense}={ dense some number }; Area _Sparse={ sparse some number };

area _total＝area _dense+area _sparse； $\mathrm{\λ}=\frac{{\mathrm{area}}_{\mathrm{total}}}{\mathrm{area}};$ $\mathrm{\η}$ $=\frac{{\mathrm{area}}_{\mathrm{dense}}}{{\mathrm{area}}_{\mathrm{total}}};$

Be defined as follows judgment criterion again:

(1) λ=0 does not then have modification.

(2) as if λ＞0, and η＜τ, think accidental modification so.α is a real number of getting in advance between fixed 0.5 to 1.

(3) η 〉=τ then thinks malicious attack.

Make authentication result at last, do not revise or accidental the modification if having, image then to be checked is true; If malicious attack, image then to be checked is false.

Description of drawings

The invisibility image of Fig. 1 watermark.Wherein (a) is original image; (b) be the watermarking images (PSNR=42.26dB) of α=-1.5 for embedding parameter.

Fig. 2 detects and the station-keeping ability test pattern for distorting.Wherein (a) is for distorting figure (hair highlight); (b) be watermark extracting figure; (c) be the watermark error image.

Fig. 3 is the watermark extracting figure after the watermarking images lease making different quality factor JPEG compression.Wherein the quality factor of (a) is 60%; (b) quality factor is 50%; (c) quality factor is 40%.

Fig. 4 embeds parameter and the extracting parameter watermark extracting figure when identical or different.Wherein (a) is for embedding the watermarking images that contains of parameter alpha=-1.5; (b) be the watermark extracting figure of extracting parameter α=-1.5; (c) be the watermark extracting figure of extracting parameter α=-1.5000001.

Visible watermark image and original image visually do not have difference among Fig. 1.

Fig. 2 can find out that present technique is the positioning tampering zone accurately.

Fig. 3 as seen, technology described herein can be resisted the quality factor and is low to moderate 40% JPEG compression.

Can see among Fig. 4 that extracting parameter is with respect to the minor variations of embedding parameter, meeting is so that the watermark that extracts is assorted Disorderly no chapter distributes, and this illustrates with other parameter, also can not obtain even the minimum parameter of difference is carried out watermark extracting Get useful information.

Claims (2)

1, a kind of image authentication method based on parameter integer wavelet transformation is characterized in that this method is made up of four steps

1) binary watermarking pre-service is with two-value original watermark and two-value pseudo random matrix XOR mutually, the watermark that produces 0 and 1 balanced proportion;

2) structure is based on the parameter integer small echo of Via Lifting Scheme;

3) embed watermark adopts the parameter integer small echo that step 2 constructed that view data is carried out wavelet decomposition, and watermark is embedded in the low frequency sub-band coefficient behind the wavelet transformation;

4) distort detection and authentication, the classifying rules of distorting of its foundation is: (1) is if the nothing modification is then thought in λ=0; (2) as if λ＞0, and δ＜α, then think accidental modification; (3) if δ 〉=α then thinks malicious attack.λ wherein, δ be respectively distort detect erroneous point and distort detect be in the erroneous point densely distributed have a few in shared ratio, α is the real number between predefined 0.5 to 1.

2, a kind of image authentication method based on parameter integer wavelet transformation as claimed in claim 1 is characterized in that the α of step 4) gets real number 0.55.

Images