Robust color image blind watermarking method based on water circulation optimization algorithm

Abstract

The invention discloses a robust color image blind watermarking method based on a water circulation optimization algorithm. According to the method, an image block coefficient is converted into a frequency domain coefficient matrix through discrete cosine transform, low-frequency energy is gathered through Laplace transform, and watermark embedding and blind extraction are completed by quantizing a maximum energy coefficient. And simulating a water circulation process by utilizing a water circulation optimization algorithm, and dynamically updating the optimal embedding step length in the watermarking method so as to balance the invisibility and robustness of the watermarking. According to the method, the color image watermark is embedded into the color host image, and the method has good watermark invisibility and high robustness.

Description

Technical Field

[0001] The invention belongs to the technical field of information security, and relates to copyright protection of large-capacity color digital images with strong robustness and high invisibility. Background Art

[0002] With the widespread dissemination and utilization of digital images in various fields, their value as information carriers has become increasingly prominent, bringing new challenges to copyright protection. Unauthorized copying, modification and dissemination not only infringe upon the legitimate rights and interests of the creators, but also damage the authenticity and reliability of the image content. In this context, digital watermarks play a vital role in preventing copyright infringement, combating piracy and ensuring the authenticity of information sources. Good digital watermarking technology needs to consider robustness, invisibility, security, embedding capacity and other performance. In addition, the current copyright protection logo tends to be more beautiful and high-information color image digital watermarks, which makes people pay more attention to color images. Therefore, designing a high-capacity digital watermarking algorithm with strong robustness and high invisibility has become a research hotspot in current digital watermarking technology. Summary of the invention

[0003] The purpose is to provide a robust color image blind watermarking method based on a water cycle optimization algorithm, which is characterized by being implemented through a specific watermark embedding process, an extraction process, and an optimization process of a water cycle algorithm. The watermark embedding process is described as follows: Step 1: First, take a picture of size Color digital watermark image Perform dimensionality reduction processing and decompose it into three single-channel watermark images of R, G, and B ; Then, use the two-dimensional Logistic adjusted sine mapping based on the key Ka to watermark each single channel image Perform encryption scrambling to obtain the scrambled watermark single-channel image; then, the single-channel watermark image Each pixel in the image is converted from a decimal number to an 8-bit binary number and connected in sequence to form a binary number of length 8N. 2 Single channel watermark sequence , where N is the color digital watermark image The number of edge pixels, Respectively represent the three channels of R, G, and B; Step 2: Make a picture of size Color host image Perform dimensionality reduction processing and decompose it into three single-channel watermark images of R, G, and B , and then each Divide into non-overlapping pixel blocks of 4×4 pixels, where M is the color host image The number of edge pixels, Respectively represent the three channels of R, G, and B; Step 3: Use the pseudo-random sequence generated by the random function to screen the divided pixel blocks, and store the selected pixel block positions in the key Kb; Step 4: Select a 4×4 pixel block ,right Perform DCT transform to obtain its frequency domain coefficient block , then select 2×2 coefficient block of mid-neighboring low-frequency coefficients , perform Laplace transform on it to obtain the second-order frequency domain coefficient block ; Step 5: From a single-channel watermark sequence Take out one bit of watermark information to be embedded in the sequence , select the second-order low-frequency coefficients in the first row and first column , the watermark is embedded by formula (1) to get ; (1) in, is the rounding function towards zero, is the embedding step size; Step 6: Update to the second-order frequency domain coefficient block Get , and then perform an inverse Laplace transform to obtain and replace ,get , and finally perform inverse DCT transform to obtain the pixel block containing the watermark ; Step 7: Repeat steps 4 to 6 until all watermark information is embedded in the pixel blocks to obtain a single-channel watermarked image. Finally, the three single-channel watermarked images are reassembled to obtain the watermarked image ; The watermark extraction process is described as follows: Step 1: The host image containing the watermark information Perform dimensionality reduction processing and decompose it into three single-channel watermarked images of R, G, and B , and then each Divide into non-overlapping pixel blocks of size 4×4, where Respectively represent the three channels of R, G, and B; Step 2: Select the pixel block containing watermark information through a pseudo-random sequence based on the key Kb; Step 3: Select a 4×4 pixel block containing a watermark ,right Perform DCT transform to obtain its frequency domain coefficient block , then select 2×2 coefficient block of mid-neighboring low-frequency coefficients , perform Laplace transform on it to obtain the second-order frequency domain coefficient block ; Step 4: Select The second-order low-frequency coefficients of the first row and first column , extracted by formula (2) The watermark contained in ; (2) in, is the remainder function, is the embedding step size; Step 5: Repeat steps 3 to 4 until all watermark information is extracted to form a single-channel binary watermark sequence. , the watermark bit sequence Each 8-bit binary information in is grouped together and converted into a decimal number to obtain a decimal sequence, where Respectively represent the three channels of R, G, and B; Step 6: Convert the decimal watermark bit sequence Perform inverse two-dimensional Logistic adjusted sine mapping decryption based on the key Ka to obtain the layered watermark , and finally reassemble to get the final watermark image ,in Respectively represent the three channels of R, G, and B; The process of its water cycle optimization algorithm is: Step 1: Set the initial parameters required by the water cycle optimization algorithm, including the upper bound of the solution , the lower bound of the solution , the number of populations , the number of rivers , evaporation condition constant , maximum number of iterations ; Step 2: Set the objective function of the water cycle optimization algorithm according to formula (3) to obtain The maximum value of The corresponding embedding step size T; (3) in, is the averaging function, The embedding step size is The NC values ​​measured under different attack types, is the average NC value measured under 10 attack types, Represents different attack types. It is the peak signal-to-noise ratio function of the image after embedding the watermark. It is the structural similarity function of the image after embedding the watermark. is the host image, is a watermarked image; Step 3: According to the above watermark embedding process and watermark extraction process, watermark embedding and extraction operations are continuously performed, and the water cycle optimization algorithm and the measured objective function value are used to measure whether the embedding step length is locally optimal, and the embedding step length is continuously optimized, and finally the optimal embedding step length in this iteration is determined; Step 4: Repeat step 3 until the maximum number of iterations is reached. The global optimal embedding step corresponding to the last iteration is the embedding step of the algorithm. .

[0004] The present invention converts the image block coefficients into frequency domain coefficient matrices by performing discrete cosine transform, and combines Laplace transform to further concentrate low-frequency energy, and finally completes watermark embedding and blind extraction by quantizing the maximum energy coefficient. This method embeds the color image watermark into the color host image, has good watermark invisibility and strong robustness, and achieves a balance between invisibility and robustness through the water cycle optimization algorithm. BRIEF DESCRIPTION OF THE DRAWINGS

[0005] Figure 1 It is a size of The original color host image.

[0006] Figure 2 It is a size of The original color watermark image.

[0007] Figure 3 Yes Figure 2 The watermark image shown is embedded into the host image Figure 1 The obtained watermarked image has a structural similarity SSIM value of 0.977 and a peak signal-to-noise ratio PSNR value of 42.061dB.

[0008] Figure 4 is from Figure 3 The normalized cross-correlation coefficient NC value of the watermark extracted from is 1.000.

[0009] Figure 5 (a) Figure 5 (b) Figure 5 (c) Figure 5 (d) Figure 5(e) Figure 5 (f) Figure 5 (g) Figure 5 (h) Figure 5 (i) Figure 5 (j) is Figure 3 The watermarks extracted from the watermarked images shown in the figure are attacked with JPEG70, JPEG2000 (7:1), salt and pepper noise (1%), Gaussian low-pass filtering (5×5), shearing (25%), scaling (50%), rotation (15°), translation (35, -25), affine (0.01, 0.01), and oblique transformation (50, 1; 512, 60; 450, 512; 1,400), and their normalized cross-correlation coefficient NC values ​​are 0.994, 0.995, 0.927, 0.976, 0.747, 0.960, 0.971, 0.985, 0.838, and 0.835, respectively. DETAILED DESCRIPTION

[0010] The purpose of the present invention is to provide a robust color image blind watermarking method based on a water cycle optimization algorithm, which is characterized by being implemented through a specific watermark embedding process, an extraction process, and an optimization process of a water cycle algorithm. The watermark embedding process is described as follows: Step 1: First, take a picture of size Color digital watermark image Perform dimensionality reduction processing and decompose it into three single-channel watermark images of R, G, and B ; Then, use the two-dimensional Logistic adjusted sine mapping based on the key Ka to watermark each single channel image Perform encryption scrambling to obtain the scrambled watermark single-channel image; then, the single-channel watermark image Each pixel in is converted from a decimal number to an 8-bit binary number (for example, a pixel value is 168, which is converted into a binary watermark sequence of 10101000), and then concatenated to form a binary watermark sequence of length Single channel watermark sequence ,in, Color digital watermark image The number of edge pixels, Respectively represent the three channels of R, G, and B; at this time Ka = [ 0 . 56823603 , 0 . 00427592 , 0 . 71750067 ] ; Step 2: Make a picture of size Color host image Perform dimensionality reduction processing and decompose it into three single-channel watermark images of R, G, and B , and then each Divide into non-overlapping pixel blocks of 4×4 pixels, where For color host images The number of edge pixels, Respectively represent the three channels of R, G, and B; Step 3: Use the pseudo-random sequence generated by the random function to screen the divided pixel blocks, and store the selected pixel block positions in the key Kb, where: Kb = [ 9472 , 524 , … , x n ] , ; Step 4: Select a 4×4 pixel block ,right Perform DCT transform to obtain its frequency domain coefficient block , then select 2×2 coefficient block of mid-neighboring low-frequency coefficients , perform Laplace transform on it to obtain the second-order frequency domain coefficient block Here, let Block = [ 7 1 7 1 7 0 7 0 7 1 7 1 7 0 7 0 7 2 7 1 7 0 7 0 7 2 7 2 7 1 7 1 ] ,but Block D ( 4 × 4 ) = [ 2 8 3 . 2 5 0 0 2 . 1 7 4 4 0 . 2 5 0 0 − 0 . 6 3 0 1 − 1 . 4 4 1 9 − 0 . 1 7 6 8 − 0 . 1 3 5 3 − 0 . 0 7 3 2 0 . 7 5 0 0 − 0 . 3 2 6 6 − 0 . 2 5 0 0 − 0 . 1 3 5 3 − 0 . 2 1 4 6 0 . 4 2 6 8 0 . 3 2 6 6 0 . 1 7 6 8 ] , Block D ( 2 × 2 ) = [ 283 . 2500 − 1 . 4419 2 . 1744 − 0 . 1768 ] , Block L ( 2 × 2 ) = [ 2265 . 4442 − 296 . 7825 − 264 . 2362 − 285 . 3968 ] ; Step 5: From a single-channel watermark sequence Take out one bit of watermark information to be embedded in the sequence , select the second-order low-frequency coefficients in the first row and first column , the watermark is embedded by formula (1) to get ; (1) in, is the rounding function towards zero, is the embedding step length; assuming that when , When, yes After watermark embedding, we get ; Step 6: Update to the second-order frequency domain coefficient block Get , and then perform an inverse Laplace transform to obtain and replace ,get , and finally perform inverse DCT transform to obtain the pixel block containing the watermark ;at this time, Block L ( 2 × 2 ) * = [ 22 23.6533 − 264 . 2362 − 296 . 7825 − 285 . 3968 ] , Block D ( 2 × 2 ) * = [ 227 . 6779 − 2 . 3706 1 . 2457 − 1 . 1055 ] , Block D ( 4 × 4 ) * = [ 2 7 7 . 6 7 7 9 1 . 2 4 5 7 0 . 2 5 0 0 − 0 . 6 3 0 1 − 2 . 3 7 0 6 − 1 . 1 0 5 5 − 0 . 1 3 5 3 − 0 . 0 7 3 2 0 . 7 5 0 0 − 0 . 3 2 6 6 − 0 . 2 5 0 0 − 0 . 1 3 5 3 − 0 . 2 1 4 6 0 . 4 2 6 8 0 . 3 2 6 6 0 . 1 7 6 8 ] , Block * = [ 6 8 . 6 0 3 9 6 9 . 0 1 3 8 6 8 . 5 9 3 4 6 9 . 0 0 3 3 6 9 . 0 1 3 8 6 9 . 2 8 7 7 6 8 . 6 7 5 0 6 8 . 9 4 8 8 7 0 . 5 9 3 4 6 9 . 6 7 5 0 6 8 . 7 9 0 3 6 8 . 8 7 1 8 7 1 . 0 0 3 3 7 0 . 9 4 8 8 6 9 . 8 7 1 8 6 9 . 8 1 7 3 ] ; Step 7: Repeat steps 4 to 6 until all watermark information is embedded in the pixel blocks to obtain a single-channel watermarked image. Finally, the three single-channel watermarked images are reassembled to obtain the watermarked image ; The watermark extraction process is described as follows: Step 1: The host image containing the watermark information Perform dimensionality reduction processing and decompose it into three single-channel watermarked images of R, G, and B , and then each Divide into non-overlapping pixel blocks of size 4×4, where Respectively represent the three channels of R, G, and B; Step 2: Select the pixel block containing watermark information through a pseudo-random sequence based on the key Kb; here, Kb = [ 9472 , 524 , … , x n ] , ; Step 3: Select a 4×4 pixel block containing a watermark ,right Perform DCT transform to obtain its frequency domain coefficient block , then select 2×2 coefficient block of mid-neighboring low-frequency coefficients , perform Laplace transform on it to obtain the second-order frequency domain coefficient block ;at this time, Block * = [ 6 9 6 9 6 9 6 9 6 9 6 9 6 9 6 9 7 1 7 0 6 9 6 9 7 1 7 1 7 0 7 0 ] , Block D ( 4 × 4 ) * = [ 2 7 8 . 2 5 0 0 1 . 2 5 0 5 0 . 2 5 0 0 − 0 . 2 4 7 4 − 2 . 3 6 5 7 − 1 . 0 3 0 3 − 0 . 1 3 5 3 0 . 2 8 0 3 0 . 7 5 0 0 − 0 . 3 2 6 6 − 0 . 2 5 0 0 − 0 . 1 3 5 3 0 . 1 6 8 1 0 . 7 8 0 3 0 . 3 2 6 6 0 . 0 3 0 3 ] , Block D ( 2 × 2 ) * = [ 2 7 8 . 2 5 0 0 − 2 . 3 6 5 7 1 . 2 5 0 5 − 1 . 0 3 0 3 ] , Block L ( 2 × 2 ) * = [ 2 2 2 8 . 1 4 5 6 − 2 9 7 . 8 1 9 5 − 2 6 4 . 9 6 4 5 − 2 8 5 . 6 6 5 5 ] Step 4: Select The second-order low-frequency coefficients of the first row and first column , extracted by formula (2) The watermark contained in ; (2) in, is the remainder function, is the embedding step size; here, when , , ,so, , which is similar to the watermark information embedded above The assumptions are consistent; Step 5: Repeat steps 3 to 4 until all watermark information is extracted to form a single-channel binary watermark sequence. , the watermark bit sequence Each 8-bit binary information in is grouped together and converted into a decimal number to obtain a decimal sequence, where Respectively represent the three channels of R, G, and B; Step 6: Convert the decimal watermark bit sequence Perform inverse two-dimensional Logistic adjusted sine mapping decryption based on the key Ka to obtain the layered watermark , and finally reassemble to get the final watermark image ,in Respectively represent the three channels of R, G, and B; at this time, Ka = [ 0 . 56823603 , 0 . 00427592 , 0 . 71750067 ] ; The process of its water cycle optimization algorithm is: Step 1: Set the initial parameters required by the water cycle optimization algorithm, including the upper bound of the solution , the lower bound of the solution , the number of populations , the number of rivers , evaporation condition constant , maximum number of iterations , here is set , , , , , ; Step 2: Set the objective function of the water cycle optimization algorithm according to formula (3) to obtain The maximum value of The corresponding embedding step ; (3) in, is the averaging function, The embedding step size is The NC values ​​measured under different attack types, is the average NC value measured under 10 attack types, Represents different attack types. It is the peak signal-to-noise ratio function of the image after embedding the watermark. It is the structural similarity function of the image after embedding the watermark. is the host image, is the watermarked image; the number of populations , here we assume that the generated initial solution is , , , , , , , , , , , , , , , , , , , , and its corresponding initial solution is , , , , , , , , , , , , , , , , , , , ; Step 3: According to the above watermark embedding process and watermark extraction process, watermark embedding and extraction operations are continuously performed, and the embedding step length is measured by the water cycle optimization algorithm and the measured objective function value to measure whether it is locally optimal, and the embedding step length is continuously optimized, and finally the optimal embedding step length in this iteration is determined; at this time, through the first column iteration of the water cycle algorithm, it is determined that the optimal function value in the first round is 0.91287, and its corresponding embedding step length is 196.4673; Step 4: Repeat step 3 until the maximum number of iterations is reached. The global optimal embedding step corresponding to the last iteration is the embedding step of the algorithm. ; At this time, through the water cycle algorithm, the optimal function value determined in the second round of iteration is 0.91946, and its corresponding embedding step length is 224.2344; the optimal function value determined in the third round is 0.91946, and its corresponding embedding step length is 224.2344; the optimal function value determined in the fifth round of iteration is 0.92008, and its corresponding embedding step length is 225.2195; the optimal function value determined in the sixth round of iteration is 0.922, and its corresponding embedding step length is 227.7404; the optimal function value determined in the seventh round of iteration is 0.92265, and its corresponding embedding step length is 228.0266, until in the fiftieth round of iteration, the optimal function value is determined to be 0.92347, and the corresponding embedding step length is 228.067. The embedding step length determined at this time is the embedding step length of the algorithm. .

[0011] The present invention converts the image block coefficients into frequency domain coefficient matrices by performing discrete cosine transform, and combines Laplace transform to further concentrate low-frequency energy, and finally completes watermark embedding and blind extraction by quantizing the maximum energy coefficient. This method embeds the color image watermark into the color host image, has good watermark invisibility and strong robustness, and achieves a balance between invisibility and robustness through the water cycle optimization algorithm.

Claims

1. The purpose is to provide a robust color image blind watermarking method based on water cycle optimization algorithm, characterized in that This is achieved through a specific watermark embedding process, extraction process, and optimization process of the water cycle algorithm. The watermark embedding process is described as follows: Step 1: First, take a picture of size Color digital watermark image Perform dimensionality reduction processing and decompose it into three single-channel watermark images of R, G, and B ; Then, use the two-dimensional Logistic adjusted sine mapping based on the key Ka to watermark each single channel image Perform encryption scrambling to obtain the scrambled watermark single-channel image; then, the single-channel watermark image Each pixel in the image is converted from a decimal number to an 8-bit binary number and connected in sequence to form a binary number of length 8N. 2 Single channel watermark sequence , where N is the color digital watermark image The number of edge pixels, Respectively represent the three channels of R, G, and B; Step 2: Make a picture of size Color host image Perform dimensionality reduction processing and decompose it into three single-channel watermark images of R, G, and B , and then each Divide into non-overlapping pixel blocks of 4×4 pixels, where M is the color host image The number of edge pixels, Respectively represent the three channels of R, G, and B; Step 3: Use the pseudo-random sequence generated by the random function to screen the divided pixel blocks, and store the selected pixel block positions in the key Kb; Step 4: Select a 4×4 pixel block ,right Perform DCT transform to obtain its frequency domain coefficient block , then select 2×2 coefficient block of mid-neighboring low-frequency coefficients , perform Laplace transform on it to obtain the second-order frequency domain coefficient block ; Step 5: From a single-channel watermark sequence Take out one bit of watermark information to be embedded in the sequence , select the second-order low-frequency coefficients in the first row and first column , the watermark is embedded by formula (1) to get ; (1) in, is the rounding function towards zero, is the embedding step size; Step 6: Update to the second-order frequency domain coefficient block Get , and then perform an inverse Laplace transform to obtain and replace ,get , and finally perform inverse DCT transform to obtain the pixel block containing the watermark ; Step 7: Repeat steps 4 to 6 until all watermark information is embedded in the pixel blocks to obtain a single-channel watermarked image. Finally, the three single-channel watermarked images are reassembled to obtain the watermarked image ; The watermark extraction process is described as follows: Step 1: The host image containing the watermark information Perform dimensionality reduction processing and decompose it into three single-channel watermarked images of R, G, and B , and then each Divide into non-overlapping pixel blocks of size 4×4, where Respectively represent the three channels of R, G, and B; Step 2: Select the pixel block containing watermark information through a pseudo-random sequence based on the key Kb; Step 3: Select a 4×4 pixel block containing a watermark ,right Perform DCT transform to obtain its frequency domain coefficient block , then select 2×2 coefficient block of mid-neighboring low-frequency coefficients , perform Laplace transform on it to obtain the second-order frequency domain coefficient block ; Step 4: Select The second-order low-frequency coefficients of the first row and first column , extracted by formula (2) The watermark contained in ; (2) in, is the remainder function, is the embedding step size; Step 5: Repeat steps 3 to 4 until all watermark information is extracted to form a single-channel binary watermark sequence. , the watermark bit sequence Each 8-bit binary information in is grouped together and converted into a decimal number to obtain a decimal sequence, where Respectively represent the three channels of R, G, and B; Step 6: Convert the decimal watermark bit sequence Perform inverse two-dimensional Logistic adjusted sine mapping decryption based on the key Ka to obtain the layered watermark , and finally reassemble to get the final watermark image ,in Respectively represent the three channels of R, G, and B; The process of its water cycle optimization algorithm is: Step 1: Set the initial parameters required by the water cycle optimization algorithm, including the upper bound of the solution , the lower bound of the solution , the number of populations , the number of rivers , evaporation condition constant , maximum number of iterations ; Step 2: Set the objective function of the water cycle optimization algorithm according to formula (3) to obtain The maximum value of The corresponding embedding step size T; (3) in, is the averaging function, The embedding step size is The NC values ​​measured under different attack types, is the average NC value measured under 10 attack types, Represents different attack types. It is the peak signal-to-noise ratio function of the image after embedding the watermark. It is the structural similarity function of the image after embedding the watermark. is the host image, is a watermarked image; Step 3: According to the above watermark embedding process and watermark extraction process, watermark embedding and extraction operations are continuously performed, and the water cycle optimization algorithm and the measured objective function value are used to measure whether the embedding step length is locally optimal, and the embedding step length is continuously optimized, and finally the optimal embedding step length in this iteration is determined; Step 4: Repeat step 3 until the maximum number of iterations is reached. The global optimal embedding step corresponding to the last iteration is the embedding step of the algorithm. .

Images