Frequency domain color digital image blind watermarking method based on Hadamard transform and voting system

Abstract

The present invention makes full use of the high correlation of the image Hadamard domain coefficients and discloses a frequency domain color digital image blind watermarking method based on Hadamard transform and voting system. The present invention quantifies the high correlation coefficient pairs of the Hadamard domain and uses their size relationship to achieve the embedding and blind extraction of color image digital watermarks, and uses the minority obeys majority voting system to improve robustness, while combining symmetric key and asymmetric key encryption methods to improve security. The present invention embeds the color image digital watermark into the color host image, while achieving good invisibility and strong robustness, it also has high real-time and security, effectively solving the problem of slow operation of large-capacity color image digital watermarks, and is suitable for fast, robust and secure digital media copyright protection.

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. Background Art

[0002] With the rapid development of the Internet and multimedia technology, illegal copying, tampering, and malicious attacks on digital works have become increasingly serious problems. Therefore, it is urgent to develop a technology to solve these problems. As a potential solution, digital watermarking has attracted worldwide attention and has become the focus of international academic attention. Digital watermarking refers to embedding an imperceptible watermark information in the original host image. People can claim ownership of digital works by extracting watermarks from data, which means that watermarks should be able to resist most common attacks, such as JPEG compression, noise, filtering, and geometric attacks.

[0003] In recent years, color image watermarking has become one of the hot topics of research. Compared with binary image and grayscale image watermarking, color image digital watermarking has two advantages: hiding more data and obtaining higher fidelity. Faced with the huge amount of information and the ever-increasing quality of digital images, more groups or individuals tend to choose color images as copyright symbols. Therefore, on the basis of ensuring the invisibility of watermarks and the robustness of algorithms, designing a color image digital watermarking method with high real-time performance and high security has become one of the key points and difficulties in the current research of digital watermarking technology. Summary of the invention

[0004] The purpose of the present invention is to provide a frequency domain color digital image blind watermarking method based on Hadamard transform and voting system. The method makes full use of the high correlation of the Hadamard domain coefficients of the image and adopts a voting system where the minority obeys the majority to optimize the performance of the scheme. The characteristics of the method are realized through specific watermark embedding preprocessing, watermark embedding, watermark embedding postprocessing, watermark extraction preprocessing, watermark extraction and watermark extraction postprocessing processes. The watermark embedding preprocessing process is described as follows:

[0005] Step 1: Set a pixel size to N × N 24-bit color image digital watermark W Divide into 3 layered watermark images in the order of red, green and blue primary colors W i ,in, i =1, 2, 3 represent the red, green, and blue layers respectively;

[0006] Step 2: Add each layered watermark image W i Key-based Ka iAn affine transformation of , where i =1, 2, 3 represent the red, green, and blue layers respectively;

[0007] Step 3: Encrypted layered watermark image W i ’ Each pixel represented by a decimal number in is represented by an 8-bit binary number and connected in sequence to form a length of 8 N 2 The layered watermark bit sequence SW i ’ ,in, i =1, 2, 3 represent the red, green, and blue layers respectively;

[0008] The watermark embedding process is described as follows:

[0009] Step 1: Set a pixel size to M × M Original color host image C Divided into 3 layered host images in the order of red, green and blue primary colors C i ; At the same time, each layered host image C i Divided into pixels m × m Image block; according to the length of the layered watermark bit sequence 8 N 2 , using key-based Kb i The MD5 hash pseudo-random scrambling algorithm generates a non-repeating block sequence, and the block sequence is placed in the hierarchical host image according to the position provided by the block sequence. C i Select image blocks, where 8 N 2 <=( M × M ) / ( m × m ), i =1, 2, 3 represent the red, green, and blue layers respectively;

[0010] Step 2: Select an image block A , and perform Hadamard transform on it according to formula (1) to obtain the frequency domain coefficient matrix H A ;

[0011] H A = H m × A (1)

[0012] in, H m express m × m The Hadamard matrix of m is an image block A The number of pixels of the side length, H A Represents an image block A The frequency domain coefficient matrix obtained after Hadamard transform, where m Hadamard square H m It can be generated iteratively by formula (2), and the initial matrix ;

[0013] (2)

[0014] in, k is a positive integer, m =2 k It means that the Hadamard matrix can only work on matrices that are integer powers of 2. represents the Kronecker product;

[0015] Step 3: From the layered watermark bit sequence SW i ' Take out the watermark information to be embedded in order w , using the frequency domain matrix H A The magnitude relationship of the coefficients in the first row depends on the watermark information to be embedded. w And formula (3), (4) and the inter-layer correlation of RGB images, different quantization step sizes are selected between layers T i , embed the watermark information into the frequency domain matrix;

[0016] (3)

[0017] (4)

[0018] in, H A1,1 , H A1,2 , H A1,3 They are the frequency domain coefficient matrices H A The coefficients of the first row, first, second, and third columns, sign(.) is the sign function, abs(.) is the absolute value function, d is the error parameter, T i Indicates iThe quantization step size of the layer, i =1,2, 3 represent the red, green and blue layers respectively, T 1 =0.87× T 3 , T 2 =0.94× T 3 ;

[0019] Step 4: According to the inverse transformation of formula (5), the image block containing the watermark is obtained A * ;

[0020] A * =( H m × H A * ) / m (5)

[0021] in, m is the frequency domain coefficient matrix H A The number of pixels of the side length, H m express m × m The Hadamard matrix of H A Represents an image block A The frequency domain matrix obtained after Hadamard transform;

[0022] Step 5: Block the watermarked image A * Update to its host image in layer C i The corresponding position in i =1,2, 3 represent the red, green and blue layers respectively;

[0023] Step 6: Repeat steps 2 to 5 of this process until all watermark information is embedded, thus obtaining a layered host image with watermark. C i * ,in i =1, 2, 3 represent the red, green, and blue layers respectively;

[0024] Step 7: Layer the red, green and blue watermarked host images C i * Reassemble and get the pixel size M × MWatermarked image C * ,in i =1, 2, 3 represent the red, green, and blue layers respectively;

[0025] The post-processing process of watermark embedding is described as follows:

[0026] Step 1: Use the integer pairing function to pair and encrypt the important parameters in the above steps to generate a large integer key; the important parameters include the quantization step size of the blue channel T 3 , image block size m , affine transformation key Ka i ,in i =1, 2, 3 represent the red, green, and blue layers respectively;

[0027] Step 2: Encrypt the large integer using the elliptic curve encryption algorithm;

[0028] The watermark extraction preprocessing process is described as follows:

[0029] Step 1: Use the elliptic curve decryption algorithm to obtain a large integer key;

[0030] Step 2: Use the inverse integer pairing function to further decrypt the decrypted large integer to obtain important parameters T 3 '、 m '、 Ka i ',in i =1, 2, 3 represent the red, green, and blue layers respectively;

[0031] The watermark extraction process is described as follows:

[0032] Step 1: Set the pixel size to M × M Watermarked image C * Divided into 3 layers with watermark images C i * , and each layered watermarked image C i * It is further divided into pixel sizes m '× m ' non-overlapping image blocks, where i =1, 2, 3 represent the red, green, and blue layers respectively;

[0033] Step 2: Layer the watermarked image C i* In the watermark embedding process, the key-based Kb i The MD5 hash pseudo-random scrambling algorithm selects the image block containing the watermark, where i =1, 2, 3 represent the red, green, and blue layers respectively;

[0034] Step 3: Select an image block with a watermark A * , and perform Hadamard transform on it according to formula (6) to obtain the frequency domain coefficient matrix H A * ;

[0035] H A * = H m × A * (6)

[0036] in, H A * Represents a watermarked image block A * The frequency domain matrix obtained after Hadamard transform, H m Indicates the pixel size is m × m The Hadamard matrix of m Is the watermarked image block A * The number of pixels of the side length;

[0037] Step 4: According to formulas (7), (8), and (9), use the frequency domain coefficient matrix H A * Extraction conditions for the magnitude relationship of the coefficients in the first row, first, second, and third columns;

[0038] (7)

[0039] (8)

[0040] (9)

[0041] in, flag 1 , flag 2 , flag 3 Indicates the possible flag states, H A1,1* , H A1,2 * , H A1,3 * They are the frequency domain coefficient matrices H A * The coefficients of the first row, first, second, and third columns;

[0042] Step 5: According to the principle of majority rule shown in formula (10), extract the frequency domain coefficient matrix H A * The watermark contained in w * ;

[0043] (10)

[0044] Among them, && is the logical AND operation, || is the logical OR operation;

[0045] Step 6: Repeat steps 3 to 5 of this process to extract the binary watermark bit sequence of each layer SW i * ,in i =1, 2, 3 represent the red, green, and blue layers respectively;

[0046] Step 7: Sequence the watermark bit SW i * Each 8 bits of binary information is converted into a decimal pixel value, where i =1, 2, 3 represent the red, green, and blue layers respectively;

[0047] The post-processing process of watermark extraction is described as follows:

[0048] Step 1: Perform key-based Ka i 'Inverse affine transformation and obtain the extracted layered watermark image W i * ,in, i =1, 2, 3 represent the red, green, and blue layers respectively;

[0049] Step 2: Combine the extracted layered watermark images W i * Form the final extracted watermark image W * ,in, i=1, 2,3 represent the red, green and blue layers respectively.

[0050] This method uses the relationship between the frequency domain coefficients of the image block after Hadamard transform and the distribution law of the high similarity coefficient in the matrix, and uses a variable quantization step in the frequency domain to complete the embedding and blind extraction of color digital watermarks; while using the voting system to improve robustness, this method also has good invisibility, high real-time performance and security. BRIEF DESCRIPTION OF THE DRAWINGS

[0051] Figure 1 (a) Figure 1 (b) are two original color host images.

[0052] Figure 2 (a) Figure 2 (b) are two original color watermarked images.

[0053] Figure 3 (a) Figure 3 (b) is to Figure 2 The watermark shown in (a) is embedded into the host image in sequence. Figure 1 (a) Figure 1 The structural similarity SSIM values ​​of the watermarked images obtained after (b) are 0.9580 and 0.9646, respectively, and the peak signal-to-noise ratio PSNR values ​​are 36.6825dB and 35.8728dB, respectively.

[0054] Figure 4 (a) Figure 4 (b) is from Figure 3 (a) Figure 3 The normalized cross-correlation coefficient NC values ​​of the watermarks extracted in (b) are 1.00000 and 1.00000, respectively, and the bit error rate BER values ​​are 0.0000 and 0.0000, respectively.

[0055] Figure 5 (a) Figure 5 (b) Figure 5 (c) Figure 5 (d) Figure 5 (e) Figure 5 (f) is to Figure 3 The watermark extracted from the watermarked image shown in (a) after being subjected to JPEG2000 compression (4:1), salt and pepper noise (1%), median filtering (3×3), rotation (30°), scaling (400%), and translation (-10, -10) have normalized cross-correlation coefficients (NC) of 0.9990, 0.9807, 0.9870, 0.9839, 0.9931, and 0.9577, respectively.

[0056] Figure 6 (a) Figure 6 (b) is to Figure 2 The watermarks shown in (b) are embedded into the host image in sequence. Figure 1 (a) Figure 1 The structural similarity SSIM values ​​of the watermarked images obtained after (b) are 0.9575 and 0.9647, respectively, and the peak signal-to-noise ratio PSNR values ​​are 36.7391dB and 35.9018dB, respectively.

[0057] Figure 7 (a) Figure 7 (b) is from Figure 6 (a) Figure 6 The normalized cross-correlation coefficient NC values ​​of the watermarks extracted in (b) are 1.00000 and 1.00000, respectively, and the bit error rate BER values ​​are 0.0000 and 0.0000, respectively.

[0058] Figure 8 (a) Figure 8 (b) Figure 8 (c) Figure 8 (d) Figure 8 (e) Figure 8 (f) is to Figure 6 The watermark extracted from the watermarked image shown in (b) after being subjected to JPEG2000 compression (4:1), salt and pepper noise (1%), median filtering (3×3), rotation (30°), scaling (400%), and translation (-10, -10) have the normalized cross-correlation coefficient NC values ​​of 0.9981, 0.9669, 0.9548, 0.9762, 0.9734, and 0.9587, respectively. DETAILED DESCRIPTION

[0059] The purpose of the present invention is to provide a frequency domain color digital image blind watermarking method based on Hadamard transform and voting system. The method makes full use of the high correlation of the Hadamard domain coefficients of the image and adopts a voting system where the minority obeys the majority to optimize the performance of the scheme. The characteristics of the method are realized through specific watermark embedding preprocessing, watermark embedding, watermark embedding postprocessing, watermark extraction preprocessing, watermark extraction and watermark extraction postprocessing processes. The watermark embedding preprocessing process is described as follows:

[0060] Step 1: Preprocessing of color image digital watermark: First, a 24-bit color image with a pixel size of 32×32 is digitally watermarked. W Divide into 3 layered watermark images in the order of red, green and blue primary colors W i ,in, i =1, 2, 3 represent the red, green, and blue layers respectively;

[0061] Step 2: Each layered watermark image is key-based Ka i An affine transformation of , where i =1, 2, 3 represent the red, green, and blue layers respectively;

[0062] Step 3: Encrypted layered watermark image W i ’ Each pixel represented by a decimal number in is represented by an 8-bit binary number (for example, 255 can be converted into the binary number 11111111), and they are connected in sequence to form a length of 8×32 2 = 8192 layered watermark bit sequence SW i ',in i =1, 2, 3 represent the red, green, and blue layers respectively;

[0063] The watermark embedding process is described as follows:

[0064] Step 1: Get the embedding block of the host image: Take an original color host image with a pixel size of 512×512 C Divided into 3 layered host images in the order of red, green and blue primary colors C i ; At the same time, each layered host image C i Divide into image blocks of 4×4 pixel size; according to the length of the layered watermark bit sequence 8 N 2 , using symmetric key based Kb i The MD5 hash pseudo-random scrambling algorithm generates a non-repeating block sequence, and then, according to the position provided by the block sequence, the layered host image C i The image blocks are selected in order to randomize the embedding position, thereby improving the robustness of the watermark against shearing attacks, where 8192 <= (512 × 512) / (4 × 4), where i =1, 2, 3 represent the red, green, and blue layers respectively;

[0065] Step 2: Select an image block A , and perform Hadamard transform on it according to formula (1) to obtain the frequency domain coefficient matrix H A ;

[0066] H A = H m × A (1)

[0067] in, H m express m × m The Hadamard matrix of m is an image block A The number of pixels of the side length, H A Represents an image block A The frequency domain coefficient matrix obtained after Hadamard transform, where m Hadamard square H m It can be generated iteratively by formula (2), and the initial matrix ;

[0068] (2)

[0069] in, k is a positive integer, m =2 k It means that the Hadamard matrix can only work on matrices that are integer powers of 2. represents the Kronecker product. Here, let the image block A for , k =2, then m =4, then according to formula (2) we can calculate the 4th-order Hadamard matrix , then the frequency domain coefficient matrix is ​​calculated according to formula (1): H m for ;

[0070] Step 3: From the layered watermark bit sequence SW i ' Take out the watermark information to be embedded in order w , using the frequency domain matrix H A The magnitude relationship of the coefficients in the first row depends on the watermark information to be embedded. w And formula (3), (4) and the inter-layer correlation of RGB images, different quantization step sizes are selected between layers T i , embed the watermark information into the frequency domain matrix;

[0071] (3)

[0072] (4)

[0073] in, H A1,1 , H A1,2 , HA1,3 They are the frequency domain coefficient matrices H A The coefficients of the first row, first, second, and third columns, sign(.) is the sign function, abs(.) is the absolute value function, d is the error parameter, T i Indicates i The quantization step size of the layer, i =1,2, 3 represent the red, green and blue layers respectively, T 1 =0.87× T 3 , T 2 =0.94× T 3 ; Here, the watermark information to be embedded is w is '1', i =1, T 1 =10.92, d =10.92, then according to formula (3) we can calculate H A1,1 * =242.92, according to formula (4) we can calculate H A1,3 * =221.08,

[0074] Step 4: According to the inverse transformation of formula (5), the image block containing the watermark is obtained A * ;

[0075] A * =( H m × H A * ) / m (5)

[0076] in, m is the frequency domain coefficient matrix H A The number of pixels of the side length, H m express m × m The Hadamard matrix of H A Represents an image block A The frequency domain matrix obtained after Hadamard transform; at this time, let m =4, the frequency domain coefficient matrix after embedding watermark H A* for , then according to formula (5), the watermarked image fast A * for ;

[0077] Step 5: Block the watermarked image A * Update to its host image in layer C i The corresponding position in i =1,2, 3 represent the red, green and blue layers respectively;

[0078] Step 6: Repeat steps 2 to 5 of this process until all watermark information is embedded, thus obtaining a layered host image with watermark. C i * ,in i =1, 2, 3 represent the red, green, and blue layers respectively;

[0079] Step 7: Layer the red, green and blue watermarked host images C i * Reassemble and obtain a watermarked image of size 512×512 C * ,in i =1, 2, 3 represent the red, green, and blue layers respectively;

[0080] The post-processing process of watermark embedding is described as follows:

[0081] Step 1: Use the integer pairing function to pair and encrypt the important parameters in the above steps to generate a large integer key Ψ =42792; among them, the important parameters include the quantization step size of the blue channel in the above steps T 3 =14, number of pixels of the image block side length m =4, affine transformation key Ka i =[6, 1];

[0082] Step 2: Encrypt the large integer using the elliptic curve encryption algorithm;

[0083] The watermark extraction preprocessing process is described as follows:

[0084] Step 1: Use the elliptic curve decryption algorithm based on asymmetric keys to obtain a large integer key Ψ =64341;

[0085] Step 2: Use the inverse integer pairing function to further decrypt the decrypted large integer and obtain three important parameters T 3 '、 m '、 Ka i ',in T 3 ' is the decrypted blue channel quantization step, its value is 14, m ' is the number of pixels on the side of the decrypted image block, and its value is 4. Ka i ' is the decrypted affine transformation key, whose value is [6, 1];

[0086] The watermark extraction process is described as follows:

[0087] Step 1: Convert the 512×512 watermarked image C * Divided into 3 layers with watermark images C i * , and each layered watermarked image C i * Further divided into m '× m ' non-overlapping image blocks, where i =1, 2, 3 represent the red, green, and blue layers respectively; here, the length of the decrypted watermarked image block is m '=4;

[0088] Step 2: Layer the watermarked image C i * In the watermark embedding process, the symmetric key based Kb i The MD5 hash pseudo-random scrambling algorithm selects the image block containing the watermark;

[0089] Step 3: Select an image block with a watermark A * , and perform Hadamard transform on it according to formula (6) to obtain the frequency domain coefficient matrix H A * ;

[0090] H A * = H m × A * (6)

[0091] in,H A * Represents a watermarked image block A * The frequency domain coefficient matrix obtained after Hadamard transform, H m Indicates the pixel size is m × m The Hadamard matrix of m Is the watermarked image block A * The number of pixels of the side length; here, m =4, assuming that the watermarked image block A * for , then according to formula (6) the frequency domain coefficient matrix can be calculated H A * for ;

[0092] Step 4: According to formulas (7), (8), and (9), use the frequency domain coefficient matrix H A * Extraction conditions for the magnitude relationship of the coefficients in the first row, first, second, and third columns;

[0093] (7)

[0094] (8)

[0095] (9)

[0096] in, flag 1 , flag 2 , flag 3 Indicates the possible flag states, H A1,1 * , H A1,2 * , H A1,3 * They are the frequency domain coefficient matrices H A * The coefficients of the first row, first column, second column, and third column; here, let H A1,1 * =242, H A1,2 * =232, HA1,3 * =220, then according to formulas (7), (8), (9) we can get flag 1 ='1', flag 2 ='1', flag 3 ='1';

[0097] Step 5: According to the principle of majority rule shown in formula (10), extract the frequency domain coefficient matrix H A * The watermark contained in w * ;

[0098] (10)

[0099] Among them, && is the logical AND operation, || is the logical OR operation; here, let flag 1 ='1', flag 2 ='1', flag 3 ='1', then according to formula (10) we can calculate w * ='1';

[0100] Step 6: Repeat steps 3 to 5 of this process to extract the binary watermark bit sequence of each layer SW i * ,in i =1, 2, 3 represent the red, green, and blue layers respectively;

[0101] Step 7: Sequence the watermark bit SW i * Each 8 bits of binary information is converted into a decimal pixel value, where i =1, 2, 3 represent the red, green, and blue layers respectively;

[0102] The post-processing process of watermark extraction is described as follows:

[0103] Step 1: Perform key-based Ka i 'Inverse affine transformation and obtain the extracted layered watermark image W i * ,in, i =1, 2, 3 represent the red, green, and blue layers respectively;

[0104] Step 2: Combine the extracted layered watermark images W i * Form the final extracted watermark image W * ,in, i =1, 2,3 represent the red, green and blue layers respectively.

[0105] This method uses the relationship between the frequency domain coefficients of the image block after Hadamard transform and the distribution law of the high similarity coefficient in the matrix, and uses a variable quantization step in the frequency domain to complete the embedding and blind extraction of the color digital watermark; while using the voting system to improve the robustness, this method also has good invisibility, high real-time performance and security.

[0106] Verification of the effectiveness of the present invention

[0107] In order to prove the effectiveness of the present invention, Figure 1 (a) Figure 1 The two 24-bit standard images with a pixel size of 512×512 shown in (b) are used as host images and are respectively Figure 2 (a) Figure 2 The two 24-bit color images with a pixel size of 32×32 shown in (b) are used as digital watermarks for verification.

[0108] Figure 3 (a) Figure 3 (b) is to Figure 2 The watermark shown in (a) is embedded into the host image in sequence. Figure 1 (a) Figure 1 (b) The structural similarity SSIM values ​​of the watermarked images obtained after are 0.9580 and 0.9646, respectively, and the peak signal-to-noise ratio PSNR values ​​are 36.6825dB and 35.8728dB, respectively; Figure 4 (a) Figure 4 (b) is from Figure 3 (a) Figure 3 The normalized correlation coefficient NC values ​​of the watermarks extracted in (b) are 1.00000 and 1.00000, respectively, and the bit error rate BER values ​​are 0.0000 and 0.0000, respectively; Figure 5 (a) Figure 5 (b) Figure 5 (c) Figure 5 (d) Figure 5 (e) Figure 5 (f) is to Figure 3The watermark extracted from the watermarked image shown in (a) after being subjected to JPEG2000 compression (4:1), salt and pepper noise (1%), median filtering (3×3), rotation (30°), scaling (400%), and translation (-10, -10) have normalized cross-correlation coefficients (NC) of 0.9990, 0.9807, 0.9870, 0.9839, 0.9931, and 0.9577, respectively.

[0109] Figure 6 (a) Figure 6 (b) is to Figure 2 The watermarks shown in (b) are embedded into the host image in sequence. Figure 1 (a) Figure 1 (b) The structural similarity SSIM values ​​of the watermarked images obtained after are 0.9575 and 0.9647, respectively, and the peak signal-to-noise ratio PSNR values ​​are 36.7391dB and 35.9018dB, respectively; Figure 7 (a) Figure 7 (b) is from Figure 6 (a) Figure 6 The normalized correlation coefficient NC values ​​of the watermarks extracted in (b) are 1.00000 and 1.00000, respectively, and the bit error rate BER values ​​are 0.0000 and 0.0000, respectively; Figure 8 (a) Figure 8 (b) Figure 8 (c) Figure 8 (d) Figure 8 (e) Figure 8 (f) is to Figure 6 The watermark extracted from the watermarked image shown in (b) after being subjected to JPEG2000 compression (4:1), salt and pepper noise (1%), median filtering (3×3), rotation (30°), scaling (400%), and translation (-10, -10) have the normalized cross-correlation coefficient NC values ​​of 0.9981, 0.9669, 0.9548, 0.9762, 0.9734, and 0.9587, respectively.

[0110] The algorithm has been run nearly 10,000 times on the platform 2.30GHZ CPU, 16.00GB RAM, Win10, MATLAB 7.10.0 (R2017a). The average embedding time of the digital watermark is 0.6515 seconds, the average extraction time is 0.3375 seconds, and the total time is 0.9890 seconds.

[0111] The encryption system of the present method combines symmetric key and asymmetric key encryption algorithms, wherein the key space of the affine transformation in a single color channel is 2 84 , the total key space of the three channels of the color image is 2 252; The key space of the MD5 hash pseudo-random scrambling algorithm in a single color channel is 2 21 , the total key space of the three channels of the color image is 2 63 ; Therefore, the total key space of the symmetric key encryption algorithm of this method is 2 315 In addition, the elliptic curve encryption algorithm based on asymmetric keys can calculate the public key from the private key, and the process is irreversible. Therefore, if the private key is unknown, the elliptic curve encryption algorithm is almost impossible to crack.

[0112] In summary, the embedded color image digital watermark has good invisibility, which meets the invisibility requirement of the watermark algorithm; moreover, the color image digital watermark extracted from various attacked images has good identifiability and high NC value, which shows that the method has strong robustness; at the same time, according to the running time analysis and key space analysis, this method has higher real-time performance and security.

Claims

1. A frequency domain color digital image blind watermarking method based on Hadamard transform and voting system, It is characterized in that This is achieved through specific watermark embedding preprocessing, watermark embedding, watermark embedding postprocessing, watermark extraction preprocessing, watermark extraction and watermark extraction postprocessing. The watermark embedding preprocessing process is described as follows: Step 1: Divide a 24-bit color image digital watermark W with a pixel size of N×N into three layered watermark images W according to the order of the three primary colors of red, green and blue. i , where i=1,2,3 represent the red, green, and blue layers respectively; Step 2: Each layered watermark image W i Based on the key Ka i Affine transformation of , where i=1,2,3 represent the red, green, and blue layers respectively; Step 3: The encrypted layered watermark image W i Each pixel represented by the decimal number in ' is represented by an 8-bit binary number and connected in sequence to form a length of 8N 2 The layered watermark bit sequence SW i ', where i = 1, 2, 3 represent the red, green, and blue layers respectively; The watermark embedding process is described as follows: Step 1: Divide an original color host image C with a pixel size of M×M into three layered host images C according to the order of the three primary colors of red, green and blue i ; At the same time, each layered host image C i Divide into image blocks of pixel size m×m; according to the length of the layered watermark bit sequence 8N 2 , using the key Kb i The MD5 hash pseudo-random scrambling algorithm generates a non-repeating block sequence, and the block sequence is located in the layered host image C according to the position provided by the block sequence. i Select image blocks from where 8N 2 <=(M×M) / (m×m), i=1,2,3 represent the red, green and blue layers respectively; Step 2: Select an image block A and perform Hadamard transform on it according to formula (1) to obtain the frequency domain coefficient matrix H A ; H A =H m ×A (1) Among them, H m represents the m×m Hadamard matrix, where m is the number of pixels on the side of the image block A, and H A represents the frequency domain coefficient matrix obtained after the Hadamard transform of the image block A, where the m-order Hadamard matrix H m It can be generated iteratively by formula (2), and the initial matrix Where k is a positive integer, m = 2 k It means that the Hadamard matrix can only work on matrices that are integer powers of 2. represents the Kronecker product; Step 3: From the layered watermark bit sequence SW i 'In order, take out the watermark information w to be embedded, and use the frequency domain matrix H A The magnitude relationship of the coefficients in the first row is based on the watermark information w to be embedded, formulas (3), (4) and the inter-layer correlation of the RGB image. Different quantization step sizes T are selected between each layer. i , embed the watermark information into the frequency domain matrix; Among them, H A1,1 , H A1,2 , H A1,3 They are the frequency domain coefficient matrices H A The coefficients of the first row, first, second, and third columns, sign(.) is the sign function, abs(.) is the absolute value function, d is the error parameter, T i represents the quantization step size of the i-th layer, i=1,2,3 represent the red, green and blue layers respectively, T 1 =0.87×T 3 , T 2 =0.94×T 3 ; Step 4: According to the inverse transformation of formula (5), the image block A containing the watermark is obtained * ; A * =(H m ×H A * ) / m (5) Where m is the frequency domain coefficient matrix H A The number of pixels of the side length, H m represents the m×m Hadamard matrix, H A represents the frequency domain matrix obtained after the Hadamard transform of the image block A; Step 5: Add the watermarked image block A * Update to its host image C in the hierarchy i The corresponding positions in , where i = 1, 2, 3 represent the red, green, and blue layers respectively; Step 6: Repeat steps 2 to 5 of this process until all watermark information is embedded, thereby obtaining the layered host image C containing the watermark. i * , where i = 1, 2, 3 represent the red, green, and blue layers respectively; Step 7: Layer the red, green and blue layers containing the watermark into the host image C i * Recombine and obtain the watermarked image C with a pixel size of M×M * , where i = 1, 2, 3 represent the red, green, and blue layers respectively; The post-processing process of watermark embedding is described as follows: Step 1: Use the integer pairing function to pair and encrypt the important parameters in the above steps to generate a large integer key; the important parameters include the quantization step size T of the blue channel. 3 , image block size m, affine transformation key Ka i , where i = 1, 2, 3 represent the red, green, and blue layers respectively; Step 2: Encrypt the large integer using the elliptic curve encryption algorithm; The watermark extraction preprocessing process is described as follows: Step 1: Use the elliptic curve decryption algorithm to obtain a large integer key; Step 2: Use the inverse integer pairing function to further decrypt the decrypted large integer and obtain the important parameter T 3 ', m', Ka i ', where i = 1, 2, 3 represent the red, green and blue layers respectively; The watermark extraction process is described as follows: Step 1: Convert the watermarked image C with a pixel size of M×M * Divided into 3 layers of watermarked images C i * , and each layered watermarked image C i * It is further divided into non-overlapping image blocks of pixel size m'×m', where i=1, 2, 3 represent the red, green, and blue layers respectively; Step 2: Layer the watermarked image C i * In the watermark embedding process, the key Kb is used. i The MD5 hash pseudo-random scrambling algorithm selects the image block containing the watermark, where i=1, 2, 3 represents the red, green, and blue layers respectively; Step 3: Select an image block A containing a watermark * , and perform Hadamard transform on it according to formula (6) to obtain the frequency domain coefficient matrix H A * ; H A * =H m ×A * (6) Among them, H A * Represents the watermarked image block A * The frequency domain matrix obtained after Hadamard transform, H m Represents the Hadamard matrix with pixel size m×m, where m is the watermarked image block A * The number of pixels of the side length; Step 4: According to formulas (7), (8), and (9), use the frequency domain coefficient matrix H A * Extraction conditions for the magnitude relationship of coefficients in the first row, first, second, and third columns; Among them, flag 1 、flag 2 、flag 3 Indicates the possible flag status, H A1,1 * , H A1,2 * , H A1,3 * They are the frequency domain coefficient matrices H A * The coefficients of the first row, first, second, and third columns; Step 5: According to the principle of majority rule shown in formula (10), extract the frequency domain coefficient matrix H A * The watermark contained in * ; Among them, && is the logical AND operation, || is the logical OR operation; Step 6: Repeat steps 3 to 5 of this process to extract each layer of binary watermark bit sequence SW i * , where i = 1, 2, 3 represent the red, green, and blue layers respectively; Step 7: Set the watermark bit sequence SW i * Each 8 bits of binary information is converted into a decimal pixel value, where i = 1, 2, 3 represents the red, green, and blue layers respectively; The post-processing process of watermark extraction is described as follows: Step 1: Perform the key Ka on each layer of converted decimal pixels i 'Inverse affine transformation and obtain the extracted layered watermark image W i * , where i=1,2,3 represent the red, green, and blue layers respectively; Step 2: Combine the extracted layered watermark images W i * Form the final extracted watermark image W * , where i=1, 2, 3 represent the red, green, and blue layers respectively.

Images