CN103034970B - Multiple information hiding method based on combination of image normalization and principal component analysis (PCA) - Google Patents

Abstract

The invention discloses an information hiding method based on the combination of image normalization and principal component analysis (PCA). The information hiding method belongs to the technical area of a computer image processing and information security. Based on the theory of image normalization and invariant centroid, the information hiding method has a good resistance to the geometric attack. Before implanting the hidden information, the chaotic technology is utilized to carry out the encryption processing on the hidden images so that the confidentiality and security of the hidden information are effectively improved. According to the features of the visual system---illuminance masking and texture masking, a perceptual masking template is designed to determine the intensity factor embedded by each sub-block hidden information and then the hidden information is embedded into multiple subspaces of the images so that the information embedding capacity of the cover images can be effectively improved.

Description

A kind of multi information hidden method combined based on image normalization and PCA

Technical field

The present invention relates to a kind of multi information hidden method combined based on image normalization and PCA, belong to Computer Image Processing and field of information security technology.

Background technology

The ultimate principle of image information concealing technology utilizes ubiquitous redundancy in image information to embed secret information wherein, thus reach the object of hidden important information.It is mainly studied and how to be embedded in view data by secret data and how to select the problem of embedded location, key issue is the design of information insertion algorithm, and how to process the relation between robustness, not sentience and this three of embedded data volume hidden Info.

Information Hiding Algorithms comparatively is early all in spatial domain, this kind of algorithm is by directly revising carrier numerical information, as the pixel of image, thus secret information is directly carried in data, its advantage is quick, and possesses certain resistivity for the operation such as geometric transformation, compression of carrier image.Although the Measures compare being carried out Information hiding by spatial-domain algorithm is easily realized, to the minimum amendment of concealed carrier, all there is very large fragility, if assailant wants to destroy secret information, only need apply signal processing technology simply and just can accomplish.In many cases, even lossy compression method also can cause the loss of information, and embeds information than having more robustness in time domain embedding information at signal frequency domain, and it is all operate on certain frequency domain that existing robustness hides system reality preferably.Transform-domain algorithm has the ability of better resisting compression, cutting and some other image processing method and attacking than spatial-domain algorithm, also maintain the disguise to human vision simultaneously.

Proper subspace mapping algorithm is a kind of information concealing method that development in recent years is got up, the method is by decomposing the feature space of digital picture, being embedded into hiding Info in each sub spaces of image, effectively can realizing hiding of multi information, and there is good robustness.

Proper subspace can be divided into signal subspace and noise subspace, because human eye is not too responsive to little noise, Information hiding can be strengthened its not sentience in noise subspace.Meanwhile, due to the mutual independence between different subspace, in different subspace, carry out the data volume that Information hiding can increase embedding, and hiding of multi information can be realized based on different subspace.In recent years, the information concealing method of feature based subspace is also subject to people and more and more payes attention to, and becomes a kind of important information concealing method gradually.

Principal component analysis is a kind of very important proper subspace extracting method, its objective is unit orthogonal vector base (i.e. pivot) being found one group of optimum by linear transformation, and carry out reconstruction sample with the linear combination of wherein part vector, make this error under lowest mean square meaning of the sample after reconstruction and former state minimum.Due to the orthogonality between subspace, information in different subspace is by independent existence, do not interfere with each other, therefore, under the prerequisite not damaging picture quality, can embed information to improve the information insertion capacity of carrier image by the number increasing subspace, the method is attacked the signal transacting of some routines has good robustness.But, the same with most image digitization information concealing method, Information Hiding in Digital Image based on principal component analysis also lacks enough resistibilitys to geometric attack, when there is the geometric transformation of the routines such as image translation, rotation or convergent-divergent, be just difficult to hide Info to extract from carrier image.

Image normalization is exactly the normalized parameter by computed image, then through a series of conversion, converts pending original image to canonical form image, and this canonical form image has invariant feature to translation, rotation, upset, the conversion of convergent-divergent equiaffine.At present, most image normalization parameter is all calculate based on the geometric moment of image, the conversion parameter determined by geometric invariant moment can be normalized to canonical form image the original image through certain uncertain affined transformation, and the method has a wide range of applications in the field such as computer vision, pattern-recognition.And in digital watermarking field, image normalization technology is mainly applied to watermark synchronization, namely first image was normalized before watermark embedment and watermark extracting, to eliminate the impact of geometric transformation, and then effectively resists the geometric attack to digital watermarking.

Summary of the invention

The present invention is directed to the proposition of above problem, and develop a kind of multi information hidden method combined based on image normalization and PCA, it has effectively merged the resist geometric attacks ability of image normalization and PCA and can increase arbitrarily subspace number to increase to hide Info and do not affect the data compression capability of robustness, based on image normalization and Invariant centroid theory, propose a kind of multi information hidden method combined based on image normalization and PCA.

The present invention takes following steps:

Step one: be 256 grades of gray level image F={f (i, j) for initial carrier, 1≤i≤m, 1≤j≤m}, image size is m × m, and hide Info as bianry image W={w (i, j), 1≤i≤s, 1≤j≤s}, image size is s × s; Normalized based on geometric invariant moment is carried out to carrier image, tries to achieve normalized image I and image centroid thereof, and corresponding conversion parameter R;

Step 2, get and embed region by its piecemeal;

Step 3, determine the intensity factor that hides Info;

Step 4, to generate the initial parameter of Logistic chaos sequence for encryption key, carry out Chaotic Scrambling to hiding Info;

Step 5, calculating S _tkL conversion coefficient; Step 6, use amended sub-space feature vectors replace original u _i, carry out KL inverse transformation try to achieve reconstruct after sample vector, add the average μ deducted in step 2 ₀, and be transformed to the image subblock of s × s, be then rearranged into embedding and hide Info area image;

Step 7, embedding to be hidden Info area image and original image remainder carry out being merged into containing the carrier image that hides Info, inverse normalization is carried out by conversion parameter R to containing the carrier image that hides Info, obtain and embed the carrier image F' after hiding Info, complete the embedding hidden Info;

The extraction of step 8, image information;

Step 9, key K (μ, the x adopted in hidden image scrambling process ₀), regenerate chaos sequence, then with the inverse process of scrambling process, the scramble extracted is hidden Info and carries out anti-scramble transformation, recover the EW that hides Info.

On the basis of traditional PCA Information Hiding Techniques, for resisting the geometric attack of image, the image normalization technology based on geometric invariant moment be have employed to initial carrier image, realize the geometry correction of image.

Determination described in step 3 hides Info the intensity factor embedded, and it is characterized in that: ask between Image Subspace at employing PCA algorithm, adopts the illumination of view-based access control model system to shelter and texture masking, design perceptual mask template: α=c ₁(1-NVF)+c ₂nVF determine each sub-block hide Info embed intensity factor.

The principle of the invention: the present invention proposes a kind of multi information hidden method combined based on image normalization and PCA, the method is before carrying out information insertion, first chaos encryption process is carried out to hiding Info, then adopt the image normalization based on geometric invariant moment to eliminate the impact of translation, rotation, upset, the conversion of convergent-divergent equiaffine, make it have affine-invariant features.Then, carry out piecemeal to around normalized image barycenter, and adopt PCA method to ask for its sample vector and total population scatter matrix.Afterwards, in order to obtain good image perception quality and higher robustness, the illumination of this method view-based access control model system is sheltered and texture masking, and the local characteristics according to different sub-block determines that each sub-block embeds the intensity factor hidden Info.The invisibility also ensureing to hide Info to regulate the intensity that hides Info adaptively, utilizes the characteristic of human visual system (Humanvisual system, HVS) to design perceptual mask template.Finally, embed the intensity factor hidden Info based on each sub-block, the hidden image after scramble is embedded in the subspace of block image, realizes Information hiding.Equally, before information extraction, to the inverse transformation carrying out based on geometric invariant moment containing the image that hides Info being subject to geometric attack, then extract watermark by the method for principal component analysis

The present invention compared with prior art has the following advantages:

1, because PCA utilizes proper subspace mapping algorithm to be decomposed by the feature space of digital picture, simultaneously due to the mutual independence between different subspace, in different subspace, carry out the data volume that Information hiding adds embedding, thus it is hiding to achieve multi information.Simultaneously owing to being the image normalization carried out before information insertion based on geometric invariant moment, effectively eliminate geometric transformation impact, well the geometric attack of opposing to digital watermarking.Hiding Info in leaching process, the transformation matrix of use is identical with original image, and this not only can extract and hide Info in identical subspace, too increases the security hidden Info simultaneously.

2, information concealing method is in the past all in spatial domain, just information extraction from the original position embedded mostly, suffer the information extraction after geometric attack often can not be satisfactory, and all can produce very large fragility to the minimum amendment of concealed carrier, in a lot of situation, even lossy compression method also can cause the loss of information.Given this, we propose to carry out chaos encryption process before information insertion, the image normalization of geometric invariant moment, eliminate the impact of geometric transformation, its key accurately will calculate before being watermark extracting and be subject to geometric attack parameter, carries out inverse transformation, then extract watermark by principal component analysis method to the watermarking images that contains being subject to geometric attack, the method not only has good robustness to the signal transacting of some routines, also has good resistibility to geometric attack.In addition, because PCA converts the good data compression had, effectively can realize Data Dimensionality Reduction, reduce the complexity calculated in image procossing.

Accompanying drawing explanation

Fig. 1 is based on the embedding grammar process flow diagram that hides Info of principal component analysis.

Fig. 2 (a) original image hides Info.

Fig. 2 (b) initial carrier image.

Fig. 3 image normalization and embedding regional choice.

Fig. 4 (a) initial carrier image.

Fig. 4 (b) is containing the image that hides Info.

The enciphering hiding information that Fig. 4 (c) extracts.

What Fig. 4 (d) extracted hides Info.

Fig. 5 (a) initial carrier image

Fig. 5 (b) hides Info 1.

Fig. 5 (c) hides Info 2.

Fig. 5 (d) enciphering hiding information 1.

Fig. 5 (e) enciphering hiding information 2.

Fig. 5 (g) is containing the image that hides Info.

Fig. 5 (i) extracts enciphering hiding information 1.

Fig. 5 (j) extracts enciphering hiding information 2.

Fig. 5 (k) extracts and hides Info 1.

Fig. 5 (l) extracts and hides Info 2.

Embodiment

Doing the present invention below in conjunction with accompanying drawing and further illustrate, one embodiment of the present of invention are:

As shown in Figure 1: the process flow diagram of performing step of the present invention;

Step one: be 256 grades of gray level image F={f (i for initial carrier, j), 1≤i≤m, 1≤j≤m}, image size is m × m, as shown in Figure 2 (a) shows, hide Info as bianry image W={w (i, j), 1≤i≤s, 1≤j≤s}, image size is s × s, as shown in Fig. 2 (b); Normalized based on geometric invariant moment is carried out to carrier image, tries to achieve normalized image I and image centroid thereof, as shown in Figure 3, and corresponding conversion parameter R;

The described image normalization processes based on geometric invariant moment is as four steps below:

(1) coordinate centralization

Original image f (x, y) is carried out coordinate centralization according to following formula;

$\left(\begin{array}{c}{x}_a\\ y_a\end{array}\right)=\left(\begin{array}{cc}1& 0\\ 0& 1\end{array}\right)\left(\begin{array}{c}x\\ y\end{array}\right)-\left(\begin{array}{c}{d}_1\\ d_2\end{array}\right)---\left(1\right)$

Wherein, m ₁₀, m ₀₁, m ₀₀be the geometric moment of original image f (x, y), its computing formula is: $m_{\mathrm{pq}}=\underset{x=0}{\overset{N_1-1}{\mathrm{\&Sigma;}}}\underset{y=0}{\overset{N_2-1}{\mathrm{\&Sigma;}}}{x}^p{y}^{q}f(x,y),0\&le;x<m,0\&le;y<m,$ F (x _a, y _a) be the image after coordinate centralization.

(2) x-shearing normalization

To the image f (x after coordinate centralization _a, y _a) carry out conversion process according to following formula;

$\left(\begin{array}{c}{x}_b\\ y_b\end{array}\right)=\left(\begin{array}{cc}1& \mathrm{\&beta;}\\ 0& 1\end{array}\right)\left(\begin{array}{c}{x}_a\\ y_a\end{array}\right)---\left(2\right)$

Wherein, parameter μ ₁₁, μ ₀₂centered by change image f (x _a, y _a) center square, its computing formula is:

$u_{\mathrm{pq}}=\underset{x=0}{\overset{N_1-1}{\mathrm{\&Sigma;}}}\underset{y=0}{\overset{N_2-1}{\mathrm{\&Sigma;}}}{(x-\stackrel{\&OverBar;}{x})}^p{(y-\stackrel{\&OverBar;}{y})}^{q}f(x,y),0\&le;x<m,0\&le;y<m$

$\stackrel{\&OverBar;}{x}=\frac{m_{10}}{m_{00}},\stackrel{\&OverBar;}{y}=\frac{m_{01}}{m_{00}}---\left(3\right)$

F (x _b, y _b) represent the image after x-shearing normalization.

(3) convergent-divergent normalization

To the image f (x after x-shearing normalization _by _b) carry out conversion process according to following formula;

$\left(\begin{array}{c}{x}_c\\ y_c\end{array}\right)=\left(\begin{array}{cc}\mathrm{\&alpha;}& 0\\ 0& \mathrm{\&delta;}\end{array}\right)\left(\begin{array}{c}{x}_b\\ y_b\end{array}\right)---\left(4\right)$

Described μ ₂₀, μ ₀₂for image f (x _b, y _b) center square, f (x _c, y _c) represent the image after dimension normalization;

(4) rotational normalization

To the image f (x after convergent-divergent normalization _c, y _c) carry out conversion process according to following formula;

$\left(\begin{array}{c}{x}_d\\ y_d\end{array}\right)=\left(\begin{array}{cc}\cos \mathrm{\&phi;}& \sin \mathrm{\&phi;}\\ -\sin \mathrm{\&phi;}& \cos \mathrm{\&phi;}\end{array}\right)\left(\begin{array}{c}{x}_c\\ y_c\end{array}\right)---\left(5\right)$

Described i=f (x _d, y _d) represent the image after rotational normalization.

Step 2, get and embed region by its piecemeal;

Point centered by the barycenter of normalized image I, is first divided into piece image F the sub-block that N number of size is s × s, each sub-block is rearranged to the vector x obtaining n=s × s dimension _i, i=1,2 ..., N, in order to make its average be zero, needs to deduct its mean vector namely:

A＝(x ₁-μ ₀,x ₂-μ _0,...,x _N-μ ₀)（6）

In formula, A is called training sample matrix, be the population mean of sample, at this moment, the autocorrelation matrix R in KLT becomes covariance matrix, its maximal possibility estimation and total population scatter matrix:

$S_t=\frac{1}{N}\underset{i=1}{\overset{N}{\mathrm{\&Sigma;}}}(x_i-{\mathrm{\&mu;}}_0){(x_i-{\mathrm{\&mu;}}_0)}^T---\left(7\right)$

Due to AA ^tand A ^ta has identical eigenwert, and AA ^tcorresponding to eigenvalue λ _iproper vector u _iwith A ^tthe corresponding proper vector v of A _ithere is following relationship:

$u_i=\frac{1}{\sqrt{{\mathrm{\&lambda;}}_i}}A{v}_i---\left(8\right)$

Then to the sample x in any X _ican be expressed as:

$x_i=\underset{j=1}{\overset{n}{\mathrm{\&Sigma;}}}{y}_j{u}_j---\left(9\right)$

Wherein y _j=x _i ^tu _j(j=1,2 ..., n), y=[y ₁, y ₂..., y _n] ^t.Choose front d the component of y as feature, then can demonstrate,prove for d component of variance maximum (namely energy is maximum), and all with S _td proper vector reconstruct in there is minimum square error, thus be commonly called main composition, with corresponding subspace is called as signal subspace, and then obtains sample vector A and total population scatter matrix S _t.

Step 3, determine the intensity factor that hides Info;

Design perceptual mask template is as follows:

α＝c ₁·(1-NVF)+c ₂·NVF（10）

Described c ₁and c ₂for regulating the embedment strength that hides Info, c ₁can be used to regulate the visual quality containing the image that hides Info.Noise visible function (Noise visibility function, NVF) adopts following form:

$\mathrm{NVF}(x,y)=\frac{1}{1+k\&CenterDot;{\mathrm{\&sigma;}}_x^2(x,y)},k=\frac{D}{{\mathrm{\&sigma;}}_{x\max }^2}---\left(11\right)$

Wherein, k is adjustment parameter, and every width image has different adjustment parameters, represent the local variance of original image in window; represent the maximal value of original image local variance.

Step 4, to generate the initial parameter of Logistic chaos sequence for encryption key, carry out Chaotic Scrambling to hiding Info, it is as follows that it comprises step:

(1) key K (μ, x is inputted ₀), produce chaos sequence X, by X by ascending order arrangement, namely [X ', l]=sort (X), X' is the chaos sequence of ascending order arrangement, and l is index sequence;

(2) W that will hide Info is converted to one-dimensional vector, is then rearranged to scramble with index sequence l and hides Info W'.

Step 5, by formula (9) calculate S _tkL conversion coefficient, get its eigenwert y _i, i ∈ [1,2., N] corresponding to sub-space feature vectors u _i, wherein, N is the sub-block number of cutting, then, at u _iembed scramble to hide Info W', concrete formula is:

$u_{\mathrm{ij}}^w=u_{\mathrm{ij}}+{\mathrm{\&alpha;}}_j\&CenterDot;w_j^{\&prime;},j=\mathrm{1,2},...,s\&times;s---\left(12\right)$

Its intensity factor α _jlocal characteristics according to different sub-block calculates by formula (10); Step 6, use amended sub-space feature vectors replace original u _i, carry out KL inverse transformation try to achieve reconstruct after sample vector, add the average μ deducted in step (2) ₀, and be transformed to the image subblock of s × s, be then rearranged into embedding and hide Info area image;

Step 7, embedding to be hidden Info area image and original image remainder carry out being merged into containing the carrier image that hides Info, inverse normalization is carried out by conversion parameter R to containing the carrier image that hides Info, obtain and embed the carrier image F' after hiding Info, complete the embedding hidden Info;

The extraction of step 8, image information

Leaching process is as follows:

(1) utilize the image normalization technology based on square, treat detected image F' and be normalized, to obtain corresponding normalized image I';

(2) with hide Info embed time identical, equally extraction piecemeal are carried out to the embedding of the normalized image I' region that hides Info, are then transformed to sample vector and try to achieve its total population scatter matrix S by formula (7) _t';

(3) utilize the KL transformation matrix hidden Info when embedding to S _t' carry out feature extraction, obtain corresponding to y _i, i ∈ [1,2 ..., N] sub-space feature vectors u _i';

(4) intensity factor when embedding according to each image subblock, setting threshold determination function carries out the extraction embedding the EW that hides Info, and formula is as follows:

$e{w}_i=\left\{\begin{array}{cc}1& u_i^{\&prime;}-u_i\&GreaterEqual;\mathrm{\&alpha;}\\ 0& u_i^{\&prime;}-u_i<\mathrm{\&alpha;}\end{array}\right.i=\mathrm{1,2},...,s\&times;s---\left(13\right)$

Step 9, key K (μ, the x adopted in hidden image scrambling process ₀), regenerate chaos sequence, then with the inverse process of scrambling process, the scramble extracted is hidden Info and carries out anti-scramble transformation, recover the EW that hides Info.On the basis of traditional PCA Information Hiding Techniques, for resisting the geometric attack of image, the image normalization technology based on geometric invariant moment be have employed to initial carrier image, realize the geometry correction of image.

Determination described in the step 3 intensity factor embedded that hides Info is asked between Image Subspace at employing PCA algorithm, adopts the illumination of view-based access control model system to shelter and texture masking, design perceptual mask template:

α＝c ₁·(1-NVF)+c ₂·NVF（10）

Determine each sub-block hide Info embed intensity factor.

From Fig. 4 (a) to 4(d) (Fig. 4 (a) is initial carrier image, Fig. 4 (b) is containing the image that hides Info, Fig. 4 (c) is the enciphering hiding information extracted, what Fig. 4 (d) extracted hides Info) can find out, in the experiment of digital image hidden information, the success that the information concealing method combined based on image normalization and PCA achieves image is hidden, moreover from Fig. 5 (a) to Fig. 5 (l), ((Fig. 5 (a) is initial carrier image to Fig. 5 (a) to Fig. 5 (l), Fig. 5 (b) hides Info 1, Fig. 5 (c) hides Info 2, Fig. 5 (d) is enciphering hiding information 1, Fig. 5 (e) is enciphering hiding information 2, Fig. 5 (g) is containing the image that hides Info, Fig. 5 extracts enciphering hiding information 1, Fig. 5 (j) extracts enciphering hiding information 2, Fig. 5 (k) extracts to hide Info 1, Fig. 5 (l) extracts to hide Info 2.) it can also be seen that the multi information hidden method combined based on image normalization and PCA successfully achieves the embedding of multiple information, improve the embedding capacity of carrier image, there is good robustness, to geometric attack, also there is good resistibility.

The above; be only the present invention's preferably embodiment; but protection scope of the present invention is not limited thereto; anyly be familiar with those skilled in the art in the technical scope that the present invention discloses; be equal to according to technical scheme of the present invention and inventive concept thereof and replace or change, all should be encompassed within protection scope of the present invention.

Claims (3)

1., based on the multi information hidden method that image normalization and PCA combine, it is characterized in that: comprise the following steps:

Step one: be 256 grades of gray level image F={f (i, j) for initial carrier, 1≤i≤m, 1≤j≤m}, image size is m × m, and hide Info as bianry image W={w (i, j), 1≤i≤s, 1≤j≤s}, image size is s × s; Normalized based on geometric invariant moment is carried out to carrier image, tries to achieve normalized image I and image centroid thereof, and corresponding conversion parameter R;

Step 2, get and embed region by its piecemeal;

Point centered by the barycenter of normalized image I, is first divided into piece image F the sub-block that N number of size is s × s, each sub-block is rearranged to the vector x obtaining n=s × s dimension _i, i=1,2 ..., N, in order to make its average be zero, needs to deduct its mean vector namely:

A＝(x ₁-μ ₀,x ₂-μ ₀,...,x _N-μ ₀)(1)

In formula, A is called training sample matrix, be the population mean of sample, at this moment, the autocorrelation matrix R in KL conversion becomes covariance matrix, its maximal possibility estimation and total population scatter matrix:

$S_t=\frac{1}{N}\underset{i=1}{\overset{N}{\mathrm{\&Sigma;}}}(x_i-{\mathrm{\&mu;}}_0){(x_i-{\mathrm{\&mu;}}_0)}^T---\left(2\right)$

Due to AA ^tand A ^ta has identical eigenwert, and AA ^tcorresponding to eigenvalue λ _iproper vector u _iwith A ^tthe corresponding proper vector v of A _ithere is following relationship:

$u_i=\frac{1}{\sqrt{{\mathrm{\&lambda;}}_i}}A{v}_i---\left(3\right)$

Then to the sample x in any X _ican be expressed as:

$x_i=\underset{j=1}{\overset{n}{\mathrm{\&Sigma;}}}{y}_j{u}_j---\left(4\right)$

Wherein y _j=x _i ^tu _j(j=1,2 ..., n), y=[y ₁, y ₂..., y _n] ^t; Choose front d the component of y as feature, then can demonstrate,prove for d component of variance maximum (namely energy is maximum), with corresponding subspace is called as signal subspace, and then obtains sample vector A and total population scatter matrix S _t;

Step 3, determine the intensity factor that hides Info

Design perceptual mask template is as follows:

α＝c ₁·(1-NVF)+c ₂·NVF(5)

Described c ₁and c ₂for regulating the embedment strength that hides Info, c ₁can be used to regulate the visual quality containing the image that hides Info, noise visible function (Noise visibility function, NVF) adopts following form:

$\mathrm{NVF}(x,y)=\frac{1}{1+k\&CenterDot;{\mathrm{\&sigma;}}_x^2(x,y)},k=\frac{D}{{\mathrm{\&sigma;}}_{x\max }^2}---\left(6\right)$

Wherein, k is adjustment parameter, and every width image has different adjustment parameters, represent the local variance of original image in window; represent the maximal value of original image local variance;

Step 4, to generate the initial parameter of Logistic chaos sequence for encryption key, carry out Chaotic Scrambling to hiding Info, it is as follows that it comprises step:

(1) key K (μ, x is inputted ₀), wherein μ, x ₀for the initial parameter of chaos sequence, produce chaos sequence X, by X by ascending order arrangement, namely [X ', l]=sort (X), the chaos sequence that X ' arranges for ascending order, l is index sequence;

(2) W that will hide Info is converted to one-dimensional vector, is then rearranged to scramble with index sequence l and hides Info W ';

Step 5, by formula (4) calculate S _tkL conversion coefficient, get its eigenwert y _i, i ∈ [1,2 ..., N] corresponding to sub-space feature vectors u _i, wherein, N is the sub-block number of cutting, then, at u _iembed scramble to hide Info W ', concrete formula is:

$u_{\mathrm{ij}}^w=u_{\mathrm{ij}}+{\mathrm{\&alpha;}}_j\&CenterDot;w_j^{\&prime;},j=\mathrm{1,2},...,s\&times;s---\left(7\right)$

Its intensity factor α _jlocal characteristics according to different sub-block calculates by formula (5);

Step 6, with amended sub-space feature vectors u _i ^wreplace original u _i, carry out KL inverse transformation try to achieve reconstruct after sample vector, add the average μ deducted in step 2 ₀, and be transformed to the image subblock of s × s, be then rearranged into embedding and hide Info area image;

Step 7, embedding to be hidden Info area image and original image remainder carry out being merged into containing the carrier image that hides Info, inverse normalization is carried out by conversion parameter R to containing the carrier image that hides Info, obtain and embed the carrier image F ' after hiding Info, complete the embedding hidden Info;

The extraction of step 8, image information

Leaching process is as follows:

(1) utilize the image normalization technology based on square, treat detected image F ' and be normalized, to obtain corresponding normalized image I ';

(2) with hide Info embed time identical, equally extraction piecemeal are carried out to the embedding of the normalized image I ' region that hides Info, are then transformed to sample vector and try to achieve its total population scatter matrix St' by formula (2);

(3) utilize the KL transformation matrix hidden Info when embedding to S _t' carry out feature extraction, obtain corresponding to yi, i ∈ [1,2 ..., N] sub-space feature vectors u _i';

(4) intensity factor when embedding according to each image subblock, setting threshold determination function carries out the extraction embedding the EW that hides Info, and formula is as follows:

${\mathrm{ew}}_i=\left\{\begin{array}{cc}1& u_i^{\&prime;}-u_i\&GreaterEqual;\mathrm{\&alpha;}\\ 0& u_i^{\&prime;}-u_i<\mathrm{\&alpha;}\end{array}\right.i=\mathrm{1,2},...,s\&times;s---\left(8\right);$

Step 9, key K (μ, the x adopted in hidden image scrambling process ₀), regenerate chaos sequence, then with the inverse process of scrambling process, the scramble extracted is hidden Info and carries out anti-scramble transformation, recover the EW that hides Info.

2. a kind of multi information hidden method combined based on image normalization and PCA according to claim 1, it is characterized in that: on the basis of traditional PCA Information Hiding Techniques, for resisting the geometric attack of image, image normalization technology based on geometric invariant moment be have employed to initial carrier image, realize the geometry correction of image.

3. a kind of multi information hidden method combined based on image normalization and PCA according to claim 1, it is characterized in that: the determination described in step 3 hide Info embed intensity factor, it is characterized in that: before employing PCA algorithm asks for Image Subspace, the illumination of view-based access control model system is adopted to shelter and texture masking, design perceptual mask template: α=c ₁(1-NVF)+c ₂nVF determine each sub-block hide Info embed intensity factor.