CN110728614B - Grey wolf optimization algorithm and full three-tree structure wavelet domain color multi-watermarking method - Google Patents

Abstract

The invention belongs to the field of digital image watermarking technology and copyright protection, and provides a wavelet domain color multi-watermarking technology which can adjust the invisibility, robustness and capacity balance of watermarks and can also effectively resist various known attacks such as noise attack, shearing attack, rotation attack and the like. Therefore, in the invention, three color watermark images are firstly encrypted into a gray image by a three-tree encryption technology in the watermark embedding process by using a gray wolf optimization algorithm and a full three-tree structure wavelet domain color multi-watermark method; then, wavelet transformation is carried out on the color host image, the selected frequency part is transformed into a YCbCr color space, singular value decomposition is carried out, and a plurality of encrypted color watermarks are embedded into the singular value part of the carrier image; and finally, combining the singular value part containing the secret image with the feature vector, and then performing inverse wavelet transformation to obtain the carrier image containing the secret image. The invention is mainly applied to the occasion of embedding the image watermark.

Description

Gray wolf optimization algorithm and complete ternary tree structure wavelet domain color multi-watermark method

Technical field

The invention belongs to the field of digital image watermark technology and copyright protection, and relates to a wavelet domain color multi-watermark technology based on gray wolf optimization algorithm and complete ternary tree structure.

Background technique

With the continuous development of computer technology, while digital products (such as images, audio, videos, etc.) bring convenience to people, these digital products also face problems such as illegal copying, reproduction, and dissemination. How to protect these digital products from illegal acquisition or copying is an urgent research topic in the field of network security today. In order to solve the above problems, digital watermark technology and encryption technology have been proposed one after another.

The main principle of digital watermark technology is to embed watermarks with copyright information into digital multimedia without affecting the original value of digital multimedia, and can effectively avoid detection by the human audio-visual system and will not be perceived by the outside world. . In order to improve the performance of watermarks, many solutions have been proposed to adjust the contradiction between the robustness and invisibility of watermarks. The earliest proposed digital watermarks based on DCT (Discrete Cosine Transform), DWT (Discrete Wavelet Transform), FT (Fourier Transform) and other transformations can make the watermark image have good invisibility by adjusting the embedding strength, but such operations are very difficult. It is difficult to balance the invisibility of the embedded watermark image and the robustness of the hiding method to various geometric attacks or image processing attacks. Based on this, digital image encryption was later introduced on the basis of watermarks to improve the robustness of watermarks through encryption.

The main problem with digital watermarking at this stage is how to balance the constraints among the invisibility, robustness, and capacity of the watermark. In order to improve the invisibility of the watermark, the watermark is usually embedded into the host invisibly based on the visual characteristics of the human eye. Improve the robustness of watermarks by encrypting or scrambling them. In terms of capacity, there are binary watermarks, grayscale watermarks, color watermarks and 3D watermarks. In the watermark embedding process, the embedding factor determines the invisibility, robustness and capacity of the watermark to a large extent. However, in common watermarking technology, the embedding factors are determined through artificial experiments, which results in contingency and uncertainty. In addition, different image information determines that the corresponding embedding strength is dynamic. Therefore, how to find the best embedding factor to balance the main performance of watermarking and achieve good watermarking effect is an important research direction in current digital watermarking technology.

Contents of the invention

In order to overcome the shortcomings of the existing technology, the present invention aims to propose a wavelet domain color multi-watermark technology based on the gray wolf optimization algorithm and a complete ternary tree structure. This digital watermark technology is mainly aimed at how to determine the accuracy of different watermarks in the watermark embedding process. Based on the dynamic embedding factor corresponding to the image, a method is proposed to optimize the watermark embedding process based on the Gray Wolf Optimization Algorithm (GWO), and by introducing ternary tree encryption technology, multiple color watermark images are encrypted, thereby improving the performance of the watermark. robustness. The proposed Trinomial Tree-Gray Wolf Optimization (Tree-GWO) watermarking scheme can adjust the balance between the invisibility, robustness and capacity of the watermark. In addition, it can also effectively resist known noise attacks, shear attacks, rotation attacks and other attacks. To this end, the technical solution adopted by the present invention is the gray wolf optimization algorithm and the complete ternary tree structure wavelet domain color multi-watermark method. During the watermark embedding process, three color watermark images are first encrypted into a grayscale image through ternary tree encryption technology. ; Then, perform wavelet transformation on the color host image, transform the selected frequency part into YCbCr color space and then perform singular value decomposition, and embed the encrypted multiple color watermarks into the singular value part of the carrier image; Finally, the singular value containing the secret image is The value part is combined with the feature vector and then subjected to inverse wavelet transform to obtain the carrier image containing the secret image.

The specific steps are detailed as follows:

Encryption of watermark images: First, divide the three color watermark images f _i into their respective R, G, and B channels, i = 1, 2, 3, and then encrypt them based on ternary tree encryption technology, and the encryption result is Cen; Then perform wavelet transform on the encryption result Cen, perform singular value decomposition on the obtained intermediate frequency part LHCen, and perform variational image decomposition on the obtained singular value part SCen to obtain the texture part Cen_u and detail part Cen_v of the encrypted image;

The watermark embedding process: First, perform wavelet transformation on the color host image IH, and convert the mid-frequency parts LHr, LHg, and LHb of the RGB color space obtained through RGB2YCbCr into the YCbCr color space: LHY, LHCb, and LHCr; for the blue chromaticity of the vertical component The component LHCb and the red chroma component LHCr of the vertical component are subjected to singular value decomposition; then Cen_u and Cen_v in step 1 are embedded into the singular value component SCb of LHCb and the singular value component SCr of LHCr respectively; then through inverse singular value transformation and The inverse wavelet transform is used to obtain the host image IW after the watermark is embedded;

The watermark extraction process: first perform wavelet decomposition on the image IW embedded with the watermark, and convert the mid-frequency parts LHEr, LHEg, and LHEb of the RGB color space obtained through RGB2YCbCr into the YCbCr color space: LHEY, LHECb, LHECr; for the blue color of the vertical component The metric component LHECb and the red chromaticity component LHECr of the vertical component are subjected to singular value decomposition; the obtained singular values are combined with the previous feature vectors to obtain the watermarked singular values SssCb and SssCr; CEN_u and CEN_v are extracted from SssCb and SssCr respectively, and finally Get the extracted encrypted image CEN;

Step 4: Decryption of the watermark image: Use ternary tree technology to decrypt the extracted encrypted image CEN to obtain the decrypted watermark image _Fi ;

Step 5: Robustness test: Conduct various attack tests on the host image with embedded watermark. By calculating the mean square value MSE, peak signal-to-noise ratio PSNR, and correlation coefficient CC value, evaluate the extraction from the attacked host image after embedding the watermark. Invisibility and robustness of watermarked images.

The specific steps of the watermark embedding process are as follows:

Step 1: Perform wavelet decomposition on the color host image IH:

[LL,LH,HL,HH]=DWT(IH) (6)

Among them, LL, LH, HL, and HH are the low-frequency, horizontal, vertical, and high-frequency components of the host image after wavelet change respectively;

Step 2: Convert the intermediate frequency parts LHr, LHg, and LHb of the obtained RGB color space into YCbCr color space: LHY, LHCb, and LHCr through RGB2YCbCr:

[LHY,LHCb,LHCr]=RGB2YCbCr(LHr,LHg,LHb) (7)

Step 3: Perform wavelet transformation and singular value decomposition on the blue chroma component LHCb of the vertical component and the red chroma component LHCr of the vertical component:

[UCb,SCb,VCb]=SVD(DWT(LHCb)) (8)

[UCr,SCr,VCr]=SVD(DWT(LHCr)) (9)

Among them, UCb, SCb, and VCb are respectively the left singular value component, characteristic singular value component, and right singular value component of LHCb; UCr, SCr, and VCr are respectively the left singular value component, characteristic singular value component, and right singular value component of LHCr.

Step 4: Embed Cen_u and Cen_v in step 1 into the singular value component SCb of LHCb and the singular value component SCr of LHCr respectively:

SsCb＝SCb+af·Cen_u (10)

SsCr＝SCr+af·Cen_v (11)

where af is the watermark embedding strength;

Step 5: Perform singular value decomposition on the singular values SsCb and SsCr embedded in the watermark:

[U1Cb,S1Cb,V1Cb]=SVD(SsCb) (12)

[U1Cr,S1Cr,V1Cr]=SVD(SsCr) (13)

Step 6: Perform inverse transformation on the obtained singular values to obtain the intermediate frequency part containing the watermark:

LH11Cb＝UCb·S1Cb·VCb ^-1 (14)

LH11Cr＝UCr·S1Cr·VCr ^-1 (15)

Step 7: Perform inverse wavelet transform on the obtained intermediate frequency part:

LHhCb＝IDWT(LL1Cb,LH11Cb,HL1Cb,HH1Cb) (16)

LHhCr＝IDWT(LL1Cr,LH11Cr,HL1Cr,HH1Cr) (17)

Step 8: Convert the intermediate frequency parts LHhCb and LHhCr of the obtained YCbCr color space into RGB color space through YCbCr2RGB:

[LH1r,LH1g,LH1b]=YCbCr2RGB(LH1Y,LHhCb,LHhCr) (18)

Step 9: Perform inverse wavelet transform on the reorganized intermediate frequency part LH1 to obtain the host image with embedded watermark:

IW＝IDWT(LL,LH1,HL,HH) (19)

The optimization process of the gray wolf optimization algorithm for the watermark embedding factor af is as follows:

1) Initialize the gray wolf population, as well as a, A and C; %a, A and C are parameters

2) Calculate the fitness function value of each gray wolf individual, and save the top three wolves with the best fitness values, the best gray wolf individual X _a , the second best gray wolf individual X _b , and the third best gray wolf Individual X _d ;

3) Record t as the current number of iterations, and max iteration as the maximum number of iterations. When the number of iterations t<maxiteration, for each gray wolf individual, update the position of the current gray wolf individual, update a, A and C, and calculate all gray wolf individuals. The fitness function value of the wolf; repeat the above steps until t≥max iteration, then X _a is the optimal solution found, that is, the optimal embedding strength af to be found.

The objective function corresponding to the embedding factor af is:

Among them, N is the total number of attack types, m is the number of watermarks, CC is the correlation coefficient value, IH is the original color host image, IW is the host image after embedding the watermark, f is the three original color watermark images, and F is the extracted and recovered For three watermark images, the corresponding objective function value should be as close to 4 as possible when finding the best embedding factor af through the gray wolf optimization algorithm.

The characteristics and beneficial effects of the present invention are:

Compared with the digital watermark algorithms that have been proposed, the present invention mainly focuses on how to determine the dynamic embedding factors corresponding to different watermark images during the watermark embedding process, and proposes a method to optimize the watermark embedding process based on the Gray Wolf Optimization Algorithm (GWO) operation, and by introducing ternary tree encryption technology, multiple color watermark images are encrypted before embedding, thereby improving the robustness of the watermark. The advantages of the proposed ternary tree-grey wolf optimization (Tree-GWO) watermarking scheme are: (1) Based on the previous single binary or grayscale or single color watermark technology, by introducing ternary tree encryption technology, three The watermark image encryption becomes a grayscale encryption result, which not only greatly improves the watermark embedding capacity, but also the ternary tree encryption technology can effectively resist various attacks; (2) For the existing watermark embedding factors determined by artificial experiments, Randomness and uncertainty, the present invention introduces the gray wolf optimization algorithm, designs an objective function based on watermark embedding extraction, and uses the GWO optimization algorithm to optimize the watermark embedding factor, thereby solving the problem of embedding corresponding to different image features. The dynamic characteristics of the factor, and effectively balance the mutual constraints between robustness, invisibility and capacity in the watermark embedding process; (3) In order to test the wide applicability of the present invention, not only various natural color images were Performance tests were also performed on medical images, which showed that the present invention is not only suitable for copyright protection of natural images, but also for copyright protection of medical images; (4) the present invention tested color hosts and color watermarks, and for gray The same applies to image and audio watermarks.

Picture description:

Figure 1 is a flow chart of the embedding and extraction of the proposed Tree-GWO watermark technology. In the figure:

(a) Schematic diagram of the color multi-watermark embedding and encryption principle provided by the present invention;

(b) Schematic diagram of the principle of color multi-watermark extraction and decryption provided by the present invention;

Figure 2 is a schematic diagram of the convergence history of the Gray Wolf optimization algorithm to find the optimal embedding factor;

Figure 3 shows four original color watermark images to be embedded and two original color host images:

(a) is Baboon;

(b) for Fruits;

(c) for Peppers;

(d) is Cell;

(e) for Lena;

Figure 4 shows the encrypted image;

Figure 5 shows the extracted and decrypted watermark image:

(a) Decryption result Baboon;

(b) Decryption result Fruits;

(c) Decryption result Peppers;

Figure 6 shows the embedded watermark image attacked by Gaussian noise with a strength of 0.2 and the corresponding extracted and decrypted watermark:

(a) Host Cell with embedded watermark attacked by 0.2 Gaussian noise;

(b) Extract the decrypted Baboon from Figure 6(a);

(c) Extract the decrypted Fruits from Figure 6(a);

(d) Extract the decrypted Peppers from Figure 6(a);

(e) Host Lena with embedded watermark attacked by 0.2 Gaussian noise;

(f) Extract the decrypted Baboon from Figure 6(e);

(g) Extract the decrypted Fruits from Figure 6(e);

(h) Extract the decrypted Peppers from Figure 6(e);

Figure 7 shows the embedded watermark image subjected to a shear attack with a strength of 50% and the corresponding extracted and decrypted watermark:

(a) Host Cell with watermark embedded under 50% clipping attack;

(b) Extract the decrypted Baboon from Figure 7(a);

(c) Extract the decrypted Fruits from Figure 7(a);

(d) Extract the decrypted Peppers from Figure 7(a);

(e) Host Lena with embedded watermark subjected to 50% clipping attack;

(f) Extract the decrypted Baboon from Figure 7(e);

(g) Extract the decrypted Fruits from Figure 7(e);

(h) Extract the decrypted Peppers from Figure 7(e);

Figure 8 shows the embedded watermark image after a rotation attack of 15° and the corresponding extracted and decrypted watermark:

(a) Host Cell with watermark embedded under rotation attack of 15°;

(b) Extract the decrypted Baboon from Figure 8(a);

(c) Extract the decrypted Fruits from Figure 8(a);

(d) Extract the decrypted Peppers from Figure 8(a);

(e) Host Lena with embedded watermark subjected to a rotation attack of 15°;

(f) Extract the decrypted Baboon from Figure 8(e);

(g) Extract the decrypted Fruits from Figure 8(e);

(h) Extract the decrypted Peppers from Figure 8(e).

In the drawings, the parts represented by each number are listed as follows:

In Figure 1(a): IH: original color host image; DWT: discrete wavelet transform; LL: low-frequency component; HL: horizontal component; LH: vertical component; HH: high-frequency component; LHr: red channel of vertical component; LHg : Green channel of the vertical component; LHb: Blue channel of the vertical component; RGB2YCbCr: RGB color space converted to YCbCr color space; LHY: Luminance component of the vertical component; LHCb: Blue chroma component of the vertical component; LHCr: Vertical component Red chromaticity component; SVD: singular value decomposition; U: left singular value vector of singular values; S: eigenvalue vector of singular values; U: right singular value vector of singular values; GreyWolf Optimizer: gray wolf optimization algorithm; af: Watermark embedding strength; Cen: encryption result of watermark; VID: variational image decomposition; Cen_u: texture part of the encryption result of watermark; Cen_v: detail part of the encryption result of watermark; ISVD: inverse singular value decomposition; IDWT: inverse discrete wavelet Transformation; YCbCr2RGB: Convert YCbCr color space to RGB color space; IW: Host image after embedding watermark.

In Figure 1(b): LHE: the vertical component of the host image embedded with watermark; CEN_u: the extracted texture part of the encrypted image; CEN_v: the extracted detail part of the encrypted image; CEN: the extracted encrypted image.

Detailed ways

The present invention aims to propose a wavelet domain color multi-watermark technology based on the gray wolf optimization algorithm and a complete ternary tree structure. This digital watermark technology is mainly aimed at how to determine the dynamic embedding factors corresponding to different watermark images during the watermark embedding process. A method is based on the Gray Wolf Optimization Algorithm (GWO) to optimize the watermark embedding process, and by introducing ternary tree encryption technology, multiple color watermark images are encrypted to improve the robustness of the watermark. The proposed Trinomial Tree-Gray Wolf Optimization (Tree-GWO) watermarking scheme can adjust the balance between the invisibility, robustness and capacity of the watermark. In addition, it can also effectively resist known noise attacks, shear attacks, rotation attacks and other attacks. To this end, the technical solution adopted by the present invention is a wavelet domain color multi-watermark technology based on the gray wolf optimization algorithm and a complete ternary tree structure. During the watermark embedding process, the three color watermark images are first encrypted into a gray image through ternary tree encryption technology. degree image; then, perform wavelet transformation on the color host image, transform the selected frequency part into YCbCr color space and then perform singular value decomposition, and embed the encrypted multiple color watermarks into the singular value part of the carrier image; finally, the secret image will be included The singular value part is combined with the eigenvector and then subjected to inverse wavelet transformation to obtain the carrier image containing the secret image. The optimal watermark embedding factor is found through the gray wolf optimization algorithm. Furthermore, in order to test the wide applicability of the present invention, we test the feasibility of the proposed algorithm in the case of color natural host images and color medical images respectively.

The specific steps are detailed as follows:

Step 1: Encryption of watermark images: First, divide the three color watermark images fi ( _i =1, 2, 3) into their respective R, G, and B channels, and then encrypt them based on ternary tree encryption technology. The result is Cen, thereby enhancing the robustness and invisibility of the watermark image; then wavelet transform is performed on the encryption result Cen, singular value decomposition is performed on the obtained intermediate frequency part LHCen, and variational image decomposition is performed on the obtained singular value part SCen (VID) Obtain the texture part Cen_u and detail part Cen_v of the encrypted image;

Step 2: Watermark embedding process: First, perform wavelet transformation on the color host image IH, and convert the obtained intermediate frequency parts LHr, LHg, and LHb of the RGB color space into YCbCr color space through RGB2YCbCr: LHY, LHCb, LHCr; for LHCb (vertical The blue chroma component of the component) and LHCr (the red chroma component of the vertical component) are subjected to singular value decomposition (SVD); then Cen_u and Cen_v in step 1 are embedded into the singular value component SCb of LHCb and the singular value component of LHCr respectively. in SCr; then through inverse singular value transform and inverse wavelet transform, the host image IW with embedded watermark can be obtained;

Step 3: Watermark extraction process: First, perform wavelet decomposition on the image IW embedded with the watermark, and convert the obtained intermediate frequency parts LHEr, LHEg, and LHEb of the RGB color space into YCbCr color space through RGB2YCbCr: LHEY, LHECb, LHECr; for LHECb ( The blue chroma component of the vertical component) and LHECr (the red chroma component of the vertical component) are subjected to singular value decomposition (SVD); the obtained singular values are combined with the previous feature vectors to obtain the watermarked singular values SssCb and SssCr; from SssCb Extract CEN_u and CEN_v respectively from SssCr and SssCr, and finally obtain the extracted encrypted image CEN;

Step 4: Decryption of the watermark image: Use ternary tree technology to decrypt the extracted encrypted image CEN to obtain the decrypted watermark image _Fi ;

Step 5: Robustness test: Conduct various attack tests on the host image with embedded watermark. By calculating MSE (mean square value), PSNR (peak signal-to-noise ratio), and CC (correlation coefficient) values, evaluate the impact of the attack on the host image after embedding the watermark. Invisibility and robustness of watermarked images extracted from the host image of the attack.

The optimization process of the gray wolf optimization algorithm for the watermark embedding factor af in step 2 is as follows:

Gray Wolf Optimizer (GWO) is a recently developed meta-heuristic search algorithm that simulates the social class and hunting mechanism of gray wolves and is used to solve non-convex engineering optimization problems. The main steps are:

1) Initialize the gray wolf population, as well as a, A and C; %a, A and C are parameters

2) Calculate the fitness function value of each gray wolf individual, and save the top three wolves with the best fitness values X _a (the best gray wolf individual), X _β (the second best gray wolf individual), X _δ (The third best gray wolf individual);

4) Record t as the current number of iterations, and max iteration as the maximum number of iterations. When the number of iterations t<maxiteration, for each gray wolf individual, update the position of the current gray wolf individual, update a, A and C, and calculate all gray The fitness function value of the wolf. Repeat the above steps until t≥max iteration, then X _a is the optimal solution found, that is, the optimal embedding strength af sought in this solution.

The objective function corresponding to the embedding factor af in the present invention is:

Where N is the total number of attack types, m is the number of watermarks, CC is the correlation coefficient value, IH is the original color host image,

IW is the host image after embedding the watermark, f is the three original color watermark images, and F is the three extracted and restored watermark images. In order to achieve good robustness, invisibility and capacity at the same time, the corresponding objective function value when finding the best embedding factor af through the gray wolf optimization algorithm should be as close as possible to 4.

In order to overcome the shortcomings of the existing technology, the present invention mainly focuses on how to determine the dynamic embedding factors corresponding to different watermark images in the watermark embedding process, and proposes a method to optimize the watermark embedding process based on the Gray Wolf Optimization Algorithm (GWO). , and by introducing ternary tree encryption technology, multiple color watermark images are encrypted before embedding, thereby improving the robustness of the watermark. The advantages of the proposed ternary tree-grey wolf optimization (Tree-GWO) watermarking scheme are: (1) Based on the previous single binary or grayscale or single color watermark technology, by introducing ternary tree encryption technology, three The watermark image encryption becomes a grayscale encryption result, which not only greatly improves the watermark embedding capacity, but also the ternary tree encryption technology can effectively resist various attacks; (2) For the existing watermark embedding factors determined by artificial experiments, Randomness and uncertainty, the present invention introduces the gray wolf optimization algorithm, designs an objective function based on watermark embedding extraction, and uses the GWO optimization algorithm to optimize the watermark embedding factor, thereby solving the problem of embedding corresponding to different image features. The dynamic characteristics of the factor, and effectively balance the mutual constraints between robustness, invisibility and capacity in the watermark embedding process; (3) In order to test the wide applicability of the present invention, not only various natural color images were Performance tests were also performed on medical images, which showed that the present invention is not only suitable for copyright protection of natural images, but also for copyright protection of medical images; (4) the present invention tested color hosts and color watermarks, and for gray The same applies to image and audio watermarks.

In order to clearly explain the purpose, technical solutions and advantages of the present invention, the watermark encryption algorithm and watermark embedding in the present invention will be described in further detail below.

(1) Triple tree encryption process of color watermark images:

Step 1: Divide the three color watermark images f _i (i=1,2,3) into their respective R, G, and B channels, and then encrypt them based on ternary tree encryption technology. The encryption result is Cen:

Cen＝Encrypt(f ₁ ,f ₂ ,f ₃ ) (22)

Among them, Encrypt(·) is the ternary tree encryption process.

Step 2: Perform wavelet transform on the encrypted result Cen:

[LLCen,LHCen,HLCen,HHCen]=DWT(Cen) (23)

Among them, LLCen, LHCen, HLCen, and HHCen are the low-frequency, horizontal, vertical, and high-frequency components of the encrypted image after wavelet change, respectively.

Step 3: Perform singular value decomposition on the intermediate frequency part LHCen obtained by wavelet transform:

[UCen,SCen,VCen]=SVD(LHCen) (24)

Step 4: Perform VID (variational image decomposition) on the singular value part SCen obtained by singular value decomposition:

[SCen_u,SCen_v]=VID(SCen) (25)

(2) Watermark embedding process:

Step 1: Perform wavelet decomposition on the color host image IH:

[LL,LH,HL,HH]=DWT(IH) (26)

Step 2: Convert the intermediate frequency parts LHr, LHg, and LHb of the obtained RGB color space into YCbCr color space: LHY, LHCb, and LHCr through RGB2YCbCr:

[LHY,LHCb,LHCr]=RGB2YCbCr(LHr,LHg,LHb) (27)

Step 3: Perform wavelet transform and singular value decomposition on LHCb (blue chroma component of vertical component) and LHCr (red chroma component of vertical component):

[UCb,SCb,VCb]=SVD(DWT(LHCb)) (28)

[UCr,SCr,VCr]=SVD(DWT(LHCr)) (29)

Among them, UCb, SCb, and VCb are respectively the left singular value component, characteristic singular value component, and right singular value component of LHCb; UCr, SCr, and VCr are respectively the left singular value component, characteristic singular value component, and right singular value component of LHCr.

Step 4: Embed Cen_u and Cen_v in step 1 into the singular value component SCb of LHCb and the singular value component SCr of LHCr respectively:

SsCb＝SCb+af·Cen_u (30)

SsCr＝SCr+af·Cen_v (31)

where af is the watermark embedding strength, which is obtained by the gray wolf optimization algorithm.

Step 5: Perform singular value decomposition on the singular values SsCb and SsCr embedded in the watermark:

[U1Cb,S1Cb,V1Cb]=SVD(SsCb) (32)

[U1Cr,S1Cr,V1Cr]=SVD(SsCr) (33)

Step 6: Perform inverse transformation on the obtained singular values to obtain the intermediate frequency part containing the watermark:

LH11Cb＝UCb·S1Cb·VCb ^-1 (34)

LH11Cr＝UCr·S1Cr·VCr ^-1 (35)

Step 7: Perform inverse wavelet transform on the obtained intermediate frequency part:

LHhCb＝IDWT(LL1Cb,LH11Cb,HL1Cb,HH1Cb) (36)

LHhCr＝IDWT(LL1Cr,LH11Cr,HL1Cr,HH1Cr) (37)

Step 8: Convert the intermediate frequency parts LHhCb and LHhCr of the obtained YCbCr color space into RGB color space through YCbCr2RGB:

[LH1r,LH1g,LH1b]=YCbCr2RGB(LH1Y,LHhCb,LHhCr) (38)

Step 9: Perform inverse wavelet transform on the reorganized intermediate frequency part LH1 to obtain the host image with embedded watermark:

IW＝IDWT(LL,LH1,HL,HH) (39)

(3) Watermark extraction process:

Step 1: Perform wavelet decomposition on the image IW with embedded watermark:

[LLE,LHE,HLE,HHE]=DWT(IW) (40)

Step 2: Convert the mid-frequency parts LHEr, LHEg, and LHEb of the obtained RGB color space into YCbCr color space through RGB2YCbCr: LHEY, LHECb, LHECr:

[LHEY,LHECb,LHECr]=RGB2YCbCr(LHEr,LHEg,LHEb) (41)

Step 3: Perform wavelet transform and singular value decomposition on LHECb (blue chroma component of vertical component) and LHECr (red chroma component of vertical component):

[U2Cb,S2Cb,V2Cb]=SVD(DWT(LHECb)) (42)

[U2Cr,S2Cr,V2Cr]=SVD(DWT(LHECr)) (43)

Step 4: Combine the obtained singular values with the previous feature vectors to obtain the watermarked singular values SssCb and SssCr:

SssCb＝U1Cb·S2Cb·V1Cb ^-1 (44)

SssCr＝U1Cr·S2Cr·V1Cr ^-1 (45)

Step 5: Extract CEN_u and CEN_v from SssCb and SssCr respectively, and finally obtain the extracted encrypted image, and then perform inverse singular value transform and inverse wavelet transform to obtain CEN:

CEN_u＝(SssCb-SCb)/af (46)

CEN_v＝(SssCr-SCr)/af (47)

CEN＝IDWT(ISVD(CEN_u+CEN_v)) (48)

Step 6: Decryption of the watermark image: Use ternary tree technology to decrypt the extracted encrypted image CEN to obtain the decrypted watermark image _Fi (i=1,2,3):

F ₁ ,F ₂ ,F ₃ =Decrypt(CEN) (49)

Step 7: Robustness test: Conduct various attack tests on the host image with embedded watermark. By calculating MSE (mean square value), PSNR (peak signal-to-noise ratio), and CC (correlation coefficient) values, evaluate the impact of the attack on the host image after embedding the watermark. Invisibility and robustness of watermarked images extracted from the host image of the attack. And calculate the fitness function value of each iteration process.

In order to verify the effectiveness of the method, experimental results are given for the encryption and decryption process of inputting four color images.

Figure 1(a) is a schematic diagram of the principle of color multi-watermark embedding and encryption provided by the present invention. First, the color host image IH is subjected to wavelet transformation, and the intermediate frequency parts LHr, LHg, and LHb of the RGB color space are converted into the YCbCr color space through RGB2YCbCr: LHY, LHCb, LHCr; for LHCb (the blue chroma component of the vertical component) and LHCr (the red chromaticity component of the vertical component) is subjected to singular value decomposition (SVD); then Cen_u and Cen_v in step 1 are embedded into the singular value component SCb of LHCb and the singular value component SCr of LHCr respectively; then through the inverse singular value Transform and inverse wavelet transform can obtain the host image IW with embedded watermark. Figure 1(b) shows the watermark extraction process: first, the image IW embedded with the watermark is decomposed by wavelet, and the obtained intermediate frequency parts LHEr, LHEg, and LHEb of the RGB color space are converted into YCbCr color space: LHEY, LHECb, LHECr through RGB2YCbCr; Perform singular value decomposition (SVD) on LHECb (the blue chroma component of the vertical component) and LHECr (the red chroma component of the vertical component); combine the obtained singular values with the previous eigenvectors to obtain the watermarked singular values SssCb and SssCr ; Extract CEN_u and CEN_v from SssCb and SssCr respectively, and finally obtain the extracted encrypted image CEN; use ternary tree technology to decrypt the extracted encrypted image CEN to obtain the decrypted watermark image _Fi .

Figure 2 is a schematic diagram of the convergence history of the Gray Wolf optimization algorithm to find the optimal embedding factor. It can be seen that after the 6th iteration, the objective function begins to converge to the maximum value. This shows that the optimization efficiency of the gray wolf optimization algorithm is relatively high.

Figures 3(a)-3(c) are three original color watermark images (size 256×256×3), Figure 3(d) and Figure 3(e) are the original color natural host image Lena (size 512 ×512×3) and the original color medical host image Cell (size is 512×512×3).

As can be seen from Figure 4, the information of the color watermark image is encrypted and encrypted into a grayscale image with a size of 256×256.

Figure 5(a) and Figure 5(b) show the host images Lena and Cell respectively after embedding watermarks. It can be found that there is no difference between them and the original host images just through the naked eye. For convenience, the present invention only shows the watermark image extracted from Figure 5(a) by this scheme without attack (Figure 5(c)-(e)).

In order to verify the robustness of the watermark in the present invention, various attack experiments were conducted. For the sake of simplicity, this description only shows three common attacks.

Figure 6(a) shows the host Lena image with embedded watermark attacked by Gaussian noise with a strength of 0.2, and the corresponding extracted and decrypted watermarks Baboon, Fruits, and Peppers (Figure 6(b)-(d)). Figure 6(e) shows the host Cell image after the watermark was embedded and attacked by Gaussian noise with a strength of 0.2, and the corresponding extracted and decrypted watermarks Baboon, Fruits, and Peppers (Figure 6(f)-(h)).

Figure 7(a) shows the host Lena image after embedding watermarks and the corresponding extracted and decrypted watermarks Baboon, Fruits, and Peppers (Figure 7(b)-(d)) that were subjected to a shearing attack with a strength of 50%. Figure 7(e) shows the host Cell image after embedding watermarks and the corresponding extracted and decrypted watermarks Baboon, Fruits, and Peppers that are subject to a 50% clipping attack (Figure 7(f)-(h)).

Figure 8(a) shows the watermark-embedded host Lena image after a rotation attack of 15° and the corresponding extracted and decrypted watermarks Baboon, Fruits, and Peppers (Figure 7(b)-(d)). Figure 8(e) shows the host Cell image embedded with watermark after a rotation attack of 15° and the corresponding extracted and decrypted watermarks Baboon, Fruits, and Peppers (Figure 8(f)-(h)).

According to the extraction and decryption results under different attacks in Figures 6-8, it can be found that even if the image after embedding the watermark is contaminated by a large degree of noise or part of the information is missing, the present invention can still extract and decrypt the original color image that can be identified , verified the feasibility of this system and met various needs in practical applications.

Although the present invention has been described above in conjunction with the illustrations, the present invention is not limited to the above-mentioned specific embodiments. The above-mentioned specific embodiments are only illustrative and not restrictive. Those of ordinary skill in the art will not Under the inspiration of the present invention, many modifications can be made without departing from the spirit of the present invention, and these all fall within the protection of the present invention.

Those skilled in the art can understand that the accompanying drawing is only a schematic diagram of a preferred embodiment, and the above-mentioned serial numbers of the embodiments of the present invention are only for description and do not represent the advantages and disadvantages of the embodiments.

The above are only preferred embodiments of the present invention and are not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention shall be included in the protection of the present invention. within the range.

Claims (5)

1. A gray wolf optimization algorithm and a full three-tree structure wavelet domain color multi-watermark method are characterized in that three color watermark images are firstly encrypted into a gray image through a three-tree encryption technology in the watermark embedding process; then, wavelet transformation is carried out on the color host image, the selected frequency part is transformed into a YCbCr color space, singular value decomposition is carried out, and a plurality of encrypted color watermarks are embedded into the singular value part of the carrier image; and finally, combining the singular value part containing the secret image with the feature vector, and then performing inverse wavelet transformation to obtain the carrier image containing the secret image.

2. The gray wolf optimization algorithm and the full three-tree structure wavelet domain color multi-watermarking method as claimed in claim 1, wherein the specific steps are refined as follows:

encryption of the watermark image: first, three color watermark images f _i Dividing the data into three channels R, G, B, i=1, 2 and 3, and then encrypting based on a trigeminal encryption technology, wherein the encryption result is Cen; then carrying out wavelet transformation on the encryption result Cen, carrying out singular value decomposition on the obtained intermediate frequency part LHCen, and carrying out variational image decomposition on the obtained singular value part SCen to obtain a texture part Cen_u and a detail part Cen_v of the encryption image;

embedding the watermark: the color host image IH is first wavelet transformed and the intermediate frequency portion LHr, LHg, LHb of the resulting RGB color space is converted to YCbCr color space by RGB2YCbCr: LHY, LHCb, LHCr; singular value decomposition is performed on a blue chrominance component LHCb of the vertical component and a red chrominance component LHCr of the vertical component; then embedding the Cen_u and Cen_v into a singular value component SCb of LHCb and a singular value component SCr of LHCr respectively; then obtaining a host image IW embedded with the watermark through inverse singular value transformation and inverse wavelet transformation;

the watermark extraction process comprises the following steps: the watermark embedded image IW is first subjected to wavelet decomposition, and the intermediate frequency part LHEr, LHEg, LHEb of the obtained RGB color space is converted into YCbCr color space by RGB2YCbCr: LHEY, LHECb, LHECr; singular value decomposition is performed on the blue chrominance component LHECb of the vertical component and the red chrominance component LHECr of the vertical component; combining the obtained singular values with the previous eigenvectors to obtain singular values SssCb and SssCr containing watermarks; CEN_u and CEN_v are respectively extracted from SssCb and SssCr, and an extracted encrypted image CEN is finally obtained;

decryption of watermark images: the extracted encrypted image CEN is decrypted by using the trigeminal technique to obtain a decrypted watermark image F _i 。

3. The gray wolf optimization algorithm and the full trigeminal structure wavelet domain color multi-watermark method of claim 2, further comprising the step of robustness testing: and carrying out various attack tests on the host image embedded with the watermark, and evaluating the invisibility and the robustness of the watermark image extracted from the host image which is attacked after the watermark is embedded by calculating a mean square value MSE, a peak signal to noise ratio PSNR and a correlation coefficient CC value.

4. The gray wolf optimization algorithm and the full three-tree structure wavelet domain color multi-watermarking method according to claim 2, wherein the watermark embedding process comprises the following specific steps:

step 1: wavelet decomposing the color host image IH:

[LL,LH,HL,HH]＝DWT(IH) (6)

wherein LL, LH, HL, HH is the low frequency, horizontal, vertical, high frequency component after wavelet change of the host image respectively;

step 2: intermediate frequency portion LHr, LHg, LHb of the resulting RGB color space is converted to YCbCr color space by RGB2YCbCr: LHY, LHCb, LHCr:

[LHY,LHCb,LHCr]＝RGB2YCbCr(LHr,LHg,LHb) (7)

step 3: wavelet transform and singular value decomposition are performed on a blue chrominance component LHCb of the vertical component and a red chrominance component LHCr of the vertical component:

[UCb,SCb,VCb]＝SVD(DWT(LHCb)) (8)

[UCr,SCr,VCr]＝SVD(DWT(LHCr)) (9)

wherein UCb, SCb, VCb is left singular value component, characteristic singular value component, right singular value component of LHCb respectively; UCr, SCr, VCr are left singular value component, characteristic singular value component, right singular value component of LHCr, respectively;

step 4: embedding cen_u and cen_v in step 1 into the singular value component SCb of LHCb and the singular value component SCr of LHCr, respectively:

SsCb＝SCb+af·Cen_u (10)

SsCr＝SCr+af·Cen_v (11)

wherein af is the optimal embedding factor;

step 5: singular value decomposition is performed on singular values SsCb and SsCr embedded with watermarks:

[U1Cb,S1Cb,V1Cb]＝SVD(SsCb) (12)

[U1Cr,S1Cr,V1Cr]＝SVD(SsCr) (13)

step 6: performing inverse transformation on the obtained singular values to obtain an intermediate frequency part containing the watermark:

LH11Cb＝UCb·S1Cb·VCb ^-1 (14)

LH11Cr＝UCr·S1Cr·VCr ^-1 (15)

step 7: and performing inverse wavelet transformation on the obtained intermediate frequency part:

LHhCb＝IDWT(LL1Cb,LH11Cb,HL1Cb,HH1Cb) (16)

LHhCr＝IDWT(LL1Cr,LH11Cr,HL1Cr,HH1Cr) (17)

step 8: converting intermediate frequency parts LHhCB and LHhCr of the obtained YCbCr color space into RGB color space through YCbCr2RGB:

[LH1r,LH1g,LH1b]＝YCbCr2RGB(LH1Y,LHhCb,LHhCr) (18)

step 9: performing inverse wavelet transformation on the recombined intermediate frequency part LH1 to obtain a host image embedded with the watermark:

IW＝IDWT(LL,LH1,HL,HH) (19)。

5. the gray-wolf optimization algorithm and the full trigeminal structure wavelet domain color multi-watermark method according to claim 1, wherein the optimization process of the gray-wolf optimization algorithm of the optimal embedding factor af is as follows:

1) Initializing a gray wolf population, and a, A and C; a. a and C are parameters

2) Calculating fitness function value of each individual wolf, and storing the best individual wolf X of the first three wolves with the best fitness value _a Second best individual X of wolves _b Third best individual X of wolf _d ；

3) T is the current iteration number, max iteration is the maximum iteration number, and when the iteration number t is<When max is equal, for each wolf individual, updating the position of the current wolf individual, updating a, A and C, and calculating the fitness function value of all the wolves; repeating the steps until t is more than or equal to max time, and then X _a For the best solution found, i.e. the best embedding factor af to be found;

the objective function corresponding to the optimal embedding factor af is:

wherein N is the total number of attacked types, m is the number of watermarks, CC is a correlation coefficient value, IH is an original color host image, IW is a host image after watermark embedding, F is three original color watermark images, F is three watermark images recovered by extraction, and the objective function value corresponding to the best embedding factor af is found by optimizing through a gray wolf optimization algorithm and should be as close as possible to 4.