A Color Digital Image Watermarking Method

Abstract

The present invention discloses a method for color digital image watermarking. First, the carrier image is transformed to obtain a singular value matrix; secondly, the watermark image is transformed to obtain a singular value matrix; then, the singular value block embedding method is applied to the two singular value matrices respectively, and an improved differential evolution algorithm is used to optimize the embedding strength of the watermark to obtain the optimal embedding strength. In addition, a digital signature is generated using the left and right singular vectors obtained by the singular value decomposition of the watermark, and the digital signature is embedded into the region with a larger standard deviation in the carrier image, and finally the carrier image containing the digital signature and the watermark is obtained. The present invention combines the improved differential evolution algorithm (DE) with the singular value block embedding method (SVBE), and both the invisibility and robustness of the watermark are better, and it can well balance the robustness and invisibility of the watermark method.

Description

Technical Field

[0001] The present invention belongs to the fields of swarm intelligence optimization algorithms and digital watermarks, and particularly relates to a color digital image watermarking method based on an improved differential evolution algorithm (DE) and a singular value block embedding method (SVBE). Background Art

[0002] In recent years, the combination of the related theories of swarm intelligence optimization algorithms and digital image watermarking algorithms has become a relatively popular research direction. Swarm intelligence optimization algorithms are a series of algorithms inspired by natural laws, biological behaviors, and social experiences, and can effectively solve various optimization problems. For a digital watermarking method, a fixed watermark strength will cause the invisibility and robustness of the watermark to be too high or too low, that is, the invisibility and robustness of the watermark will become unbalanced. In digital watermarking methods based on singular value decomposition, most methods use the singular value matrix obtained by singular value decomposition of the carrier image and the watermark image as the embedding position and watermark information of the watermark, and then only use a single embedding strength to add the watermark information to the carrier image. However, in the singular value matrix, the tolerance of larger singular values and smaller singular values to various watermark attacks is different, and only using a single embedding strength cannot well balance the invisibility and robustness of the watermark. Summary of the Invention

[0003] To solve the problems of poor invisibility and robustness of existing watermarking methods and the imbalance between them, the present invention provides a color digital image watermarking method that combines an improved differential evolution algorithm (DE) with a singular value block embedding method (SVBE). The invisibility and robustness of the watermark are both good, and it can well balance the robustness and invisibility of the watermarking method.

[0004] The technical solution adopted by the present invention to solve its technical problems is: a color digital image watermarking method, including the following steps:

[0005] Transform the carrier image to obtain a singular value matrix;

[0006] Transform the watermark image to obtain a singular value matrix;

[0007] Apply the singular value block embedding method to embed the singular value matrix obtained by transforming the watermark image into the singular value matrix obtained by transforming the carrier image with different embedding strengths;

[0008] Improve the differential evolution algorithm and optimize the embedding strength using the improved differential evolution algorithm: First, set a large range of independent variable values, divide the entire range of independent variable values into several intervals, and then, take each interval point as the embedding strength, calculate the fitness function value, find the embedding strength corresponding to the largest three function values, and use this embedding strength range as the range of independent variable values in the differential evolution algorithm; after mutation, crossover, and boundary value processing, synchronously update the independent variable and the fitness function value; use the improved differential evolution algorithm for iterative calculation, retain multiple independent variables with large fitness function values in each generation, and record the optimal independent variable in each generation. When the number of iterations is greater than the preset maximum number of iterations or the accuracy reaches the preset accuracy, jump out of the iteration, record the optimal independent variable when jumping out of the iteration, which is the optimal embedding strength, and embed the watermark image into the carrier image with the optimal embedding strength.

[0009] Further, the transformation of the carrier image is specifically as follows: Divide the carrier image into three color channels of R, G, and B, apply the undecimated lifting wavelet transform to each color channel respectively to obtain the high-frequency subbands, perform the discrete Haar wavelet transform on the high-frequency subbands to obtain the low-frequency subbands, and apply singular value decomposition to the low-frequency subbands to obtain the left and right singular matrices and the singular value matrix.

[0010] Further, the transformation of the watermark image is specifically as follows: Divide the watermark image into three color channels of R, G, and B, perform the integer wavelet transform on each color channel respectively to obtain the high-frequency subbands, and apply the discrete cosine transform and singular value decomposition to the high-frequency subbands to obtain the left and right singular matrices and the singular value matrix.

[0011] Further, the application of the singular value block embedding method is specifically as follows: Divide the singular value matrix obtained by transforming the carrier image and the singular value matrix obtained by transforming the watermark image into S block1 , S block2 and S block3 three parts, and apply different embedding strengths to each part:

[0012]

[0013] where floor(·) is the floor function;

[0014] Assume that the singular value matrix obtained by transforming the carrier image is marked as SC, and the singular value matrix obtained by transforming the watermark image is marked as SW. Divide SC and SW in the above manner respectively. All three sub-blocks of SC are recorded as SC i ; all three sub-blocks of SW are recorded as SW i ; the embedding strengths of the three sub-blocks are respectively recorded as α i , where i = 1, 2, 3, and SW iWith different embedding strengths α i Embed into SC i and reconstruct it into a new singular value matrix S i :

[0015] S i = SC i + α i × SW i

[0016] Perform inverse SVD decomposition, inverse Haar wavelet decomposition, and inverse redistributed invariant lifting wavelet transform on the S i matrix, and recombine the R, G, and B color channels to obtain the watermarked carrier image.

[0017] Furthermore, the optimal independent variable in each generation of the improved differential evolution algorithm is selected according to the magnitude of the fitness function fitness

[0018]

[0019] where C represents the carrier image, C' represents the watermarked carrier image, PSNR(C, C') represents the peak signal-to-noise ratio between the carrier image and the watermarked carrier image, W represents the original watermark image, W' represents the watermark image recovered from the carrier image, NC(W, W') represents the normalized cross-correlation coefficient between the original watermark image and the watermark image recovered from the carrier, and n represents the number of simulation attacks.

[0020] Furthermore, the steps also include:

[0021] Generate a digital signature, block the watermarked carrier image, and embed the digital signature into the regions with large standard deviations in the watermarked carrier image, extract the digital signature from the carrier image, and determine whether the watermark extraction process is allowed based on the number of correctly extracted digital signatures.

[0022] Furthermore, the generation of the digital signature is specifically: encrypt the left and right singular value matrices obtained by singular value decomposition of the watermark image through the SHA384 hash function, then convert the encrypted sequences into binary sequences respectively, extract the first six bits of the sequences respectively for exclusive OR operation, and perform exclusive OR encryption on the obtained result with a six-bit private key to obtain the final digital signature sig i .

[0023] Furthermore, the blocking of the watermarked carrier image and the embedding of the digital signature into the regions with large standard deviations in the watermarked carrier image are specifically:

[0024] The respective color channels of the watermarked carrier image are divided into 8×8 blocks, the standard deviation of each block is calculated, and the two regions with the largest standard deviation in each color channel are selected for singular value decomposition. The watermark is embedded into the left singular matrix U obtained by modulating the first column and second row U 2,1 and the first column and third row U 3,1 of the two elements. The 6-bit digital signature is embedded into the left singular matrix U according to the following embedding formula:

[0025]

[0026]

[0027]

[0028] where sign(·) represents the sign function; |·| represents taking the absolute value; U ave is the average value of the sum of the absolute values of the two coefficients: U ave =(|U 2,1 | + |U 3,1 |) / 2, and Ts is the threshold value.

[0029] Furthermore, the extraction of the digital signature from the carrier image is specifically as follows:

[0030] The attacked carrier image containing the digital signature and watermark is decomposed into three color channels: R, G, and B. Each color channel matrix is divided into 8×8 non-overlapping sub-blocks. The sub-blocks where the digital signature is embedded are found according to the addresses saved during the embedding process. Singular value decomposition is performed on the sub-blocks where the digital signature is embedded. Six digital signatures are extracted according to the size relationship between U 2,1 and U 3,1 of the two elements:

[0031]

[0032] When the number of correctly extracted digital signatures ≥ 3, the watermark extraction process is allowed.

[0033] The beneficial effects of the present invention include: 1. By performing a redistributed invariant lifting wavelet transform, a discrete Haar wavelet transform, and a singular value decomposition on the carrier image, a watermark embedding region with strong robustness can be obtained;

[0034] 2. By performing an integer wavelet decomposition, a discrete cosine transform, and a singular value decomposition on the watermark image, the amount of embedded information is reduced to a certain extent, enhancing the invisibility of the watermark method;

[0035] 3. Improve the DE algorithm. On the one hand, narrow the value range of independent variables to make the DE algorithm more efficient. On the other hand, reduce the recalculation of fitness function values, which reduces the time complexity of the algorithm to a certain extent;

[0036] 4. Propose a new SVBE method. Embed the singular value matrix obtained by transforming the watermark image into the singular value matrix obtained by transforming the carrier image with different embedding strengths, and use the improved DE algorithm to optimize the embedding strength of the watermark, which can well balance the robustness and invisibility of the watermark method;

[0037] 5. Use the digital signature protection method. Encrypt the left and right singular matrices obtained by transforming the watermark image using the SHA384 hash function, and further encrypt them with a private key to obtain a digital signature and embed it into the watermarked carrier image. On the one hand, it can prevent the false positive problem, and on the other hand, it improves the security of the watermark method. Description of the Drawings

[0038] Figure 1 is the watermark embedding flow chart;

[0039] Figure 2 is the watermark extraction flow chart;

[0040] Figure 3 is the flow chart for optimizing the watermark embedding strength using the improved DE algorithm;

[0041] Figure 4 is the schematic diagram of singular value matrix partitioning;

[0042] Figure 5 is the parameter table of the improved differential evolution algorithm;

[0043] Figure 6 is the PSNR value of the carrier image after embedding the watermark;

[0044] Figure 7 is the NC value between the extracted watermark image and the original watermark;

[0045] Figure 8 is the number of bits for recovering the digital signature;

[0046] Figure 9 is the average NC value, PSNR value and fitness function value embedded using the non - SVBE method and the SVBE method. Detailed Implementation Manner

[0047] The technical solution of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are part of the embodiments of the present invention, rather than all of them. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts belong to the scope of protection of the present invention.

[0048] In the description of the present invention, it should be noted that the orientation or positional relationship indicated by the terms "vertical", "horizontal", "inner", "outer", etc. is based on the orientation or positional relationship shown in the accompanying drawings. It is only for the convenience of describing the present invention and simplifying the description, rather than indicating or implying that the device or component referred to must have a specific orientation, be constructed and operated in a specific orientation, and therefore should not be construed as a limitation to the present invention. In addition, the terms "first", "second", and "third" are only used to distinguish components and cannot be understood as indicating or implying relative importance.

[0049] In addition, the technical features involved in different embodiments of the present invention described below can be combined with each other as long as they do not conflict with each other.

[0050] A color digital image watermarking method based on an improved differential evolution algorithm and a singular value block embedding method. First, the carrier image is transformed to obtain a singular value matrix; secondly, the watermark image is transformed to obtain a singular value matrix; then, the SVBE method is applied to the two singular value matrices respectively, and the improved DE algorithm is used to optimize the embedding strength of the watermark to obtain the optimal embedding strength. In addition, digital signatures are generated using the left and right singular vectors obtained by singular value decomposition of the watermark, and the digital signatures are embedded into the regions with larger standard deviations in the carrier image, and finally the carrier image containing the digital signature and the watermark is obtained. The watermark extraction process is the inverse process of watermark embedding. The watermark embedding flowchart is as Figure 1 shown, and the watermark extraction flowchart is as Figure 2 shown.

[0051] The specific steps of the above implementation scheme include:

[0052] The color carrier image is divided into three color channels of R, G, and B. The redundant distribution invariant lifting wavelet transform is applied to each color channel respectively to obtain the high-frequency subbands. The discrete Haar wavelet transform is performed on the high-frequency subbands to obtain the low-frequency subbands. The singular value decomposition is applied to the low-frequency subbands to obtain the left and right singular matrices and the singular value matrix, and the singular value matrix is used in the SVBE method;

[0053] The color watermark image is divided into three color channels of R, G, and B. The integer wavelet transform is performed on each color channel respectively to obtain the high-frequency subbands. The discrete cosine transform and the singular value decomposition are applied to the high-frequency subbands to obtain the left and right singular matrices and the singular value matrix, and the singular value matrix is used in the SVBE method;

[0054] Apply the SVBE method to the singular value matrix generated from the color carrier image and the singular value matrix generated from the color watermark image. Divide the singular matrix obtained by transforming the carrier and the singular value matrix obtained by transforming the watermark into three regions. The larger singular values form one region, the medium-sized singular values form one region, and the smaller singular values form one region. Embed the three singular value blocks of the watermark into the corresponding three blocks of the carrier image with different embedding strengths, as Figure 4 is a schematic diagram of the singular value block division. Specifically, divide the singular value matrix into three parts S block1 , S block2 and S block3 , and apply different embedding strengths to each part.

[0055]

[0056] Among them, floor(·) is the floor function. Assume that the singular value matrix obtained by transforming the carrier image is marked as SC, and the singular value matrix obtained by transforming the watermark image is marked as SW. Divide SC and SW in the above way respectively. All three sub-blocks of SC (from top to bottom) are marked as SC i ; The three sub-blocks of SW (from top to bottom) are recorded as SW i . The embedding strengths of the three singular value blocks are respectively recorded as α i , where i = 1, 2, 3. Embed SW i into SC i with different embedding strengths α i .

[0057] S i = SC i + α i × SW i

[0058] Perform inverse SVD decomposition, inverse Haar wavelet decomposition, and inverse redistributed invariant lifting wavelet transform on the updated S i matrix, and recombine the R, G, and B color channels to obtain the watermarked carrier image.

[0059] Improve the DE algorithm. First, set a relatively large value range (0, 1) for the independent variable. Then, take a step size of 0.1 as the interval and divide the entire range into 10 regions. Next, use each interval point as the embedding strength to calculate the function values of the fitness function, find the embedding strengths corresponding to the three largest function values, and use this embedding strength range as the value range of the individual (independent variable) in the DE algorithm. After mutation, crossover, and boundary value processing, synchronously update the individual (independent variable) and the fitness function value to reduce the recalculation of the fitness function value caused by only updating the individual (independent variable), and use the improved DE algorithm to optimize the embedding strength of the watermark, such as Figure 3 The flowchart for optimizing the watermark embedding strength by the improved DE algorithm: Use the improved DE algorithm for iterative calculation, retain multiple individuals with larger fitness function values in each generation, and record the optimal individual in each generation. When the number of iterations is greater than the preset maximum number of iterations or the accuracy reaches the preset accuracy, jump out of the iteration. At this time, the individual (independent variable) vector is the optimal embedding strength.

[0060] Adopt the method of digital signature to enhance the security of the digital watermark method. Generate a 6-bit digital signature according to the left and right singular matrices obtained by the color watermark transformation. Divide each color channel of the color carrier image into 8*8 image blocks, calculate the standard deviation of each block, and obtain the 2 image blocks with the largest standard deviation. Then perform SVD on them. By modulating the size relationship between the two elements in the second row and the third row of the first column of the left singular matrix U, embed the 6-bit digital signature into 6 left singular matrices U. Perform inverse SVD and recombine the three color channels to obtain the carrier image containing the watermark and the digital signature. Specifically, encrypt the left and right singular value matrices obtained by the singular value decomposition of the watermark through the SHA384 hash function, then convert the encrypted sequences into binary sequences respectively, extract the first six bits of each sequence for exclusive OR operation, and perform exclusive OR encryption on the obtained result with the six-bit private key to obtain the final digital signature sigi. Then divide each color channel of the watermark-containing carrier image into 8*8 blocks, calculate the standard deviation of each block, select the 2 regions with the largest standard deviation in each color channel for singular value decomposition, and by modulating the size relationship between the two elements in the second row and the third row of the first column of the left singular matrix U 2,1 and the two elements in the second row and the third row of the first column of the left singular matrix U 3,1 embed the 6-bit digital signature into the left singular matrix U;

[0061]

[0062]

[0063]

[0064] sign(·) represents the sign function; |·| represents taking the absolute value; Uave is the average of the sum of the absolute values of two coefficients: U ave =(|U 2,1 | + |U 3,1 |) / 2. Ts is the threshold;

[0065] The carrier image with digital signature and watermark under attack is decomposed into three color channels: R, G, and B. Each color channel matrix is divided into 8×8 non-overlapping sub-blocks. The sub-blocks where the digital signature is embedded are found according to the addresses saved during the embedding process. Singular value decomposition is performed on the sub-blocks where the digital signature is embedded. Six digital signatures are extracted according to the size relationship between the two elements U 2,1 and U 3,1 :

[0066]

[0067] When the number of correctly extracted digital signatures ≧ 3, the watermark extraction process is allowed.

[0068] Calculate the peak signal-to-noise ratio (PSNR) between the carrier image and the watermarked carrier image. Apply the improved DE algorithm to the watermarking method. Attack the watermarked carrier image through simulating various attack methods, extract the watermark in the damaged image, and calculate the normalized cross-correlation coefficient (NC) between the damaged watermark and the original watermark;

[0069] According to the evaluation indexes of robustness and invisibility, a fitness function fitness for evaluating the comprehensive performance of the watermark is proposed. Apply this function to the improved DE algorithm as an index for evaluating the superiority of individuals, and select the optimal individual in each generation of the improved DE algorithm according to the size of the fitness function.

[0070]

[0071] C represents the carrier image, C’ represents the watermarked carrier image, PSNR(C, C’) represents the peak signal-to-noise ratio between the carrier image and the watermarked carrier image, W represents the original watermark image, W’ represents the watermark image recovered from the carrier image, NC(W, W’) represents the normalized cross-correlation coefficient between the original watermark image and the watermark image recovered from the carrier, and n represents the number of simulated attacks.

[0072] This method expands the application scope of the transform domain watermarking method in the field of digital watermarking, and designs a new transform domain by means of the theory of redistributed invariant lifting wavelet transform, Haar wavelet transform and singular value decomposition. Optimize the watermarking method by using the improved DE algorithm and the SVBE method, and finally obtain a color digital watermarking method with good invisibility and robustness.

[0073] Example 1

[0074] The improved DE algorithm parameter table in this embodiment is as follows Figure 5 shown. Set to generate a 6-bit digital watermark from the G color channel of the watermark image, and set the threshold T in the digital signature embedding process s = 0.1.

[0075] Step 1: The carrier image X is divided into three color channels: R, G, and B. Then, the redistributed invariant lifting wavelet transform is applied to the three color channel matrices to obtain a low-frequency subband, two mid-frequency subbands, and a high-frequency subband.

[0076]

[0077] where i = R, G, B; LL i represents the low-frequency subband; HL i and LH i represent the mid-frequency subbands, and HH i represents the high-frequency subband.

[0078] Step 2: Extract the high-frequency subband HH i , and apply the discrete Haar wavelet transform to HH i to obtain the low-frequency subband HLL i . Then perform singular value decomposition on the HLL i subband to obtain the diagonal singular value matrix S i , and use the SVBE method to block the singular value matrix to obtain the singular value block matrix

[0079]

[0080] where j represents the jth block; j = 1, 2, 3.

[0081] Step 3: Divide the watermark image into three color channels, which are labeled as W i (i = R, G, B). Then, apply the integer wavelet transform to the three color channel matrices to obtain four subbands LLW i , HLW i , LHW i and HHW i . The process is as follows:

[0082]

[0083] Step 4: First, apply the DCT (Discrete Cosine Transform) to the high-frequency subband HHW i to obtain DHHW i . Then, perform singular value decomposition on the matrix DHHW i to obtain the singular value matrix SW i. Finally, apply the SVBE method to the singular value matrix SW i , to obtain

[0084]

[0085] where j represents the j-th block; j = 1, 2, 3.

[0086] Step 5: Add the matrix generated by the watermark image to the matrix generated by the carrier image , and use different embedding strengths α for each singular value block i . i = 1, 2, 3.

[0087]

[0088] Step 6: and are reconstructed into a new singular value matrix Snew i , then, obtain the updated subband HLL' through inverse singular value decomposition i . The updated subband HLL' i is used to reconstruct the high-frequency subband HLL'.

[0089]

[0090] Step 7: The updated subband HLL' is used to reconstruct the R, G, and B color channels of the watermarked carrier image. Finally, these three color channels are recombined to obtain the watermarked carrier image.

[0091]

[0092] Step 8: First, select the G color channel of the color watermark image, and then decompose this channel into the left singular vector U w and the right singular vector V w through singular value decomposition. Convert the matrix U W of size n×n W and V into one-dimensional arrays A and B respectively. Then, encrypt A and B through the SHA-384 hash function, and convert the encrypted sequence into a binary sequence. Extract the first six bits of the binary sequence and record them as a and b respectively. Finally, apply the XOR operation to a and b and record the result as c. Apply the XOR operation using the private key k and c to obtain the final digital signature s.

[0093] Step 9: First, divide the carrier image into three color channels: R, G, and B. Then divide each of the three components into 8×8 non-overlapping sub-blocks. Calculate the standard deviation of each sub-block, find the two sub-blocks with the largest standard deviation in each component, and store the positions of each sub-block.

[0094] Step 10: Perform singular value decomposition on the selected blocks, and embed the digital signature sig 2,1 generated from the watermark image according to the size relationship between U 3,1 and U i (i = 1, 2,..., 6). There are two cases here: The first case is that there is no need to correct U 2,1 and U 3,1 of the current sub-block.

[0095]

[0096] In the second case, it is necessary to correct U 2,1 and U 3,1 of the current sub-block.

[0097]

[0098]

[0099] sign(·) represents the sign function; |·| represents taking the absolute value; U ave is the average value of the sum of the absolute values of two coefficients: U ave =(|U 2,1 | + |U 3,1 |) / 2.

[0100] Step 11: Apply inverse singular value decomposition to the updated U matrix to obtain the modified sub-blocks, and use the modified sub-blocks to replace the original sub-blocks to obtain the carrier image with digital signature and watermark.

[0101] Step 12: Combine the three embedding strengths into a vector as an individual in the improved DE algorithm. Set the value range of the individual: First, set a relatively large value range (0, 1) for the independent variable (individual), and then divide the entire range into 10 regions with an interval of 0.1 in length. Then, use each interval point as the embedding strength, calculate the function value of the fitness function, find the embedding strengths corresponding to the largest three function values, and use this embedding strength range as the value range of the individual (independent variable) in the DE algorithm. After mutation, crossover, and boundary value processing, synchronously update the individual (independent variable) and the fitness function value, and record the optimal individual and its fitness function value in each generation. After reaching the accuracy condition or reaching the maximum number of iterations, jump out of the loop, record the optimal individual in the last generation when jumping out of the loop, which is the optimal embedding strength, and then embed the watermark image into the carrier image with the optimal embedding strength.

[0102] Step 13: Respectively use two color cell tissue images Gland1 and Gland2 with a size of 512*512 and two color retina images Retina1 and Retina with a size of 512*512 as carrier images, select common images House (W1), Splash (W2), Jellybeans (W3), and Female (W4) as watermark images, repeat steps 1-13, and calculate the PSNR values of the carrier before and after embedding, as Figure 6 shown.

[0103] Step 14: Use Gland1 and Gland2 as carrier images and W1, W2, W3, and W4 as watermark images, repeat steps 1-13 to complete the embedding of the watermark and digital signature. Use 33 common attack methods (denoted as Attack1, Attack2,..., Attack33 respectively) to attack the carrier images containing the watermark and digital signature. The attacks involved are as follows: no attack, 3x3 mean filtering, 5x5 mean filtering, 3x3 median filtering, 5x5 median filtering, adding Gaussian noise with a variance of 0.1, adding Gaussian noise with a variance of 0.01, adding Gaussian noise with a variance of 0.001, adding salt-and-pepper noise with a density of 0.1, adding salt-and-pepper noise with a density of 0.01, adding salt-and-pepper noise with a density of 0.001, adding multiplicative noise with a variance of 0.1, adding multiplicative noise with a variance of 0.01, adding multiplicative noise with a variance of 0.001, 50% cropping attack, scaling attack with a scaling factor of 2, scaling attack with a scaling factor of 0.5, JEPG compression attack with a quality factor of 10, JEPG compression attack with a quality factor of 30, JEPG compression attack with a quality factor of 50, JEPG compression attack with a quality factor of 100, rotation attack with an angle of 10, rotation attack with an angle of 20, rotation attack with an angle of 50, rotation attack with an angle of 90, histogram equalization, motion blur, Gaussian low-pass filtering, sharpening attack, JEPG2000 compression attack with a compression ratio of 5, JEPG2000 compression attack with a compression ratio of 30, JEPG2000 compression attack with a compression ratio of 90, Poisson noise.

[0104] Step 15: Decompose the attacked carrier images containing the digital signature and watermark into three color channels: R, G, and B. Each color channel matrix is divided into 8×8 non-overlapping sub-blocks. Find the sub-blocks where the digital signature is embedded according to the addresses saved during the embedding process. Perform singular value decomposition on the sub-blocks where the digital signature is embedded, and extract 6 digital signatures according to the size relationship of the two elements of U 2,1 and U 3,1 Extract 6 digital signatures according to the size relationship of the two elements. As Figure 8The number of digital signatures extracted from each attacked carrier image when W1 is used as the watermark. In the watermarking method, when the correct number of extracted digital signatures is greater than or equal to 3, the watermark extraction process is allowed.

[0105]

[0106] Step 16: Extract the watermark from the damaged image in Step 14. During the watermark extraction process, the digital signatures must first be authenticated. In the tested image combinations, the authentication passed (i.e., the number of extracted digital signatures was greater than or equal to 3). The watermark extraction step is the opposite of the watermark embedding step. The watermark extraction process is as Figure 2 shown. Calculate the normalized cross-correlation coefficient (NC value) between the extracted watermark and the original watermark, as Figure 7 shown.

[0107] Step 17: Calculate the average NC value, PSNR value, and fitness function value without using the SVBE method and using the SVBE method, respectively, as Figure 9 shown.

[0108] The color digital image watermarking method based on the improved differential evolution algorithm and the SVBE method proposed by the present invention first performs three combined transformations on the carrier image to obtain a singular value matrix, and then performs three combined transformations on the watermark image to obtain a singular value matrix. Then, the SVBE method is applied to the two singular value matrices respectively, and the improved DE algorithm is used to optimize the embedding strength of the watermark to obtain the optimal embedding strength. Watermark embedding is performed with the optimal embedding strength, obtaining a digital watermark image method with good robustness, invisibility, and certain security. The present invention is simulated in MATLAB 2020a, and the operating environment is Win10 Intel(R) CPU 3.1 GHz, ARM 16.0 GB. Through Figures 6 - 9 experiments, it is shown that the robustness, invisibility, and security obtained in this example are better than those of other methods.

[0109] Obviously, the above embodiments are merely examples given for clear illustration and are not limitations on the implementation manners. For those of ordinary skill in the art, other different forms of changes or modifications can be made based on the above description. It is not necessary and impossible to list all the implementation manners here. And the obvious changes or modifications derived therefrom are still within the protection scope of the present invention.

Claims

1. A method for color digital image watermarking, characterized in that, It includes the following steps: Transform the carrier image to obtain a singular value matrix; Transform the watermark image to obtain a singular value matrix; Apply the singular value block embedding method to embed the singular value matrix obtained by transforming the watermark image into the singular value matrix obtained by transforming the carrier image with different embedding strengths; Improve the differential evolution algorithm and use the improved differential evolution algorithm to optimize the embedding strength: First, set a large independent variable value range, divide the entire independent variable value range into several intervals, and then use each interval point as the embedding strength to calculate the fitness function value. Find the embedding strengths corresponding to the largest three function values, and use this embedding strength range as the value range of the independent variable in the differential evolution algorithm; after mutation, crossover, and boundary value processing, synchronously update the independent variable and the fitness function value; use the improved differential evolution algorithm for iterative calculation, retain multiple independent variables with large fitness function values in each generation, and record the optimal independent variable in each generation. When the number of iterations is greater than the preset maximum number of iterations or the accuracy reaches the preset accuracy, jump out of the iteration, record the optimal independent variable when jumping out of the iteration, which is the optimal embedding strength, and embed the watermark image into the carrier image with the optimal embedding strength.

2. The method for color digital image watermarking according to claim 1, characterized in that, The transformation of the carrier image specifically includes: dividing the carrier image into three color channels of R, G, and B, applying the redistributed invariant lifting wavelet transform to each color channel respectively to obtain the high-frequency sub-band, performing the discrete Haar wavelet transform on the high-frequency sub-band to obtain the low-frequency sub-band, and applying the singular value decomposition to the low-frequency sub-band to obtain the left and right singular matrices and the singular value matrix.

3. The method for color digital image watermarking according to claim 1, characterized in that, The transformation of the watermark image specifically includes: dividing the watermark image into three color channels of R, G, and B, performing the integer wavelet transform on each color channel respectively to obtain the high-frequency sub-band, applying the discrete cosine transform and the singular value decomposition to the high-frequency sub-band to obtain the left and right singular matrices and the singular value matrix.

4. The method for color digital image watermarking according to claim 1, characterized in that, The application of the singular value block embedding method is specifically as follows: The singular value matrix obtained by transforming the carrier image and the singular value matrix obtained by transforming the watermark image are divided into S block1 , S block2 and S block3 into three parts, and different embedding strengths are applied to each part: Where, floor(·) is the downward rounding function; Suppose the singular value matrix obtained through carrier image transformation is denoted as SC, and the singular value matrix obtained through watermark image transformation is denoted as SW. SC and SW are partitioned in the above manner respectively, and all three sub-blocks of SC are recorded as SC i ; all three sub-blocks of SW are recorded as SW i ; the embedding strengths of the three sub-blocks are recorded as α i , where i = 1, 2, 3. Embed SW i into SC i with different embedding strengths α i to reconstruct a new singular value matrix S i : S i = SC i + α i × SW i Perform inverse SVD decomposition, inverse Haar wavelet decomposition, and inverse redistribution invariant lifting wavelet transform on the S i matrix, and recombine the R, G, and B color channels to obtain the watermarked carrier image.

5. The method for color digital image watermarking according to claim 1, characterized in that, The optimal independent variable in each generation of the improved differential evolution algorithm is selected according to the size of the fitness function fitness, Where, C represents the carrier image, C’ represents the carrier image with the watermark, PSNR(C, C’) represents the peak signal-to-noise ratio of the carrier image and the carrier image with the watermark, W represents the original watermark image, W’ represents the watermark image recovered from the carrier image, NC(W, W’) represents the normalized cross-correlation coefficient between the original watermark image and the watermark image recovered from the carrier, and n represents the number of simulated attacks.

6. The method for color digital image watermarking according to claim 1, characterized in that, The steps also include: Generate a digital signature, block the carrier image with the watermark, and embed the digital signature into the area with a large standard deviation in the carrier image with the watermark, extract the digital signature in the carrier image, and judge whether the extraction process of the watermark is allowed according to the number of correctly extracted digital signatures.

7. The method for color digital image watermarking according to claim 6, characterized in that, The generation of the digital signature is specifically as follows: the left and right singular value matrices obtained by singular value decomposition of the watermark image are encrypted through the SHA384 hash function, then the encrypted sequences are respectively converted into binary sequences, the first six bits of the sequences are respectively extracted for exclusive OR operation, and the obtained result is encrypted by exclusive OR with a six-bit private key to obtain the final digital signature sig i .

8. The method for color digital image watermarking according to claim 7, characterized in that, The blocking of the carrier image with the watermark and the embedding of the digital signature into the area with a large standard deviation in the carrier image with the watermark specifically include: The respective color channels of the watermarked carrier image are divided into 8×8 blocks, the standard deviation of each block is calculated, and the two regions with the largest standard deviation in each color channel are selected for singular value decomposition. The watermark is embedded into the left singular matrix U obtained by modulating the singular value decomposition by comparing the magnitude relationship between the second element in the second row and the third element in the third row of the first column of the left singular matrix U 2,1 and the second element in the third row of the first column of U 3,1 The 6-bit digital signature is embedded into the left singular matrix U according to the following embedding formula: where sign(·) represents the sign function; |·| represents taking the absolute value; U ave is the average value of the sum of the absolute values of two coefficients: U ave =(|U 2,1 | + |U 3,1 |) / 2, and Ts is the threshold value.

9. The method for color digital image watermarking according to claim 8, characterized in that, The extraction of the digital signature in the carrier image specifically includes: Decompose the carrier image with digital signature and watermark to be attacked into three color channels: R, G, and B. Each color channel matrix is divided into 8×8 non-overlapping sub-blocks. Locate the sub-blocks where the digital signature is embedded according to the addresses saved during the embedding process, perform singular value decomposition on the sub-blocks where the digital signature is embedded, and extract 6 digital signatures according to the size relationship between the two elements of U 2,1 and U 3,1 : When the number of correctly extracted digital signatures ≧ 3, the extraction process of the watermark is allowed.

Images