A strong robust frequency domain symmetric image watermarking method for copyright protection

Abstract

The application discloses a strong robust frequency domain symmetrical image watermarking method for copyright confirmation, utilizes the base image features of different frequency domain coefficients, modulates specific frequency domain coefficients to realize the embedding of symmetrical watermarking, enhances the robustness of the watermarking against large-area cutting attacks by constructing symmetrical peaks with consistent number of different resolution images, and designs a watermark preprocessing module meeting the Kerckhoffs principle to guarantee the security of the watermarking algorithm under the open condition. The application can effectively resist various signal processing attacks and geometric attacks, still has high-precision watermark resynchronization ability when suffering various attacks, enhances the robustness of the watermarking against various attacks while guaranteeing the high perceptual sensitivity of the image, is stronger in the robustness against large-area cutting, and is stable in performance and good in generalization when applied to different resolution images.

Description

Technical Field

[0001] This invention relates to robust watermarking technology for image information hiding, and is used in the fields of image copyright protection and traceability. Background Technology

[0002] Image watermarking technology originated in the spatial domain. It embeds watermarks by directly modifying image pixels. Previous studies have shown that spatial domain watermarking is not robust to signal processing attacks. Later, researchers focused on exploring frequency domain watermarking algorithms. Discrete cosine transform (DCT), discrete sine transform (DST), discrete wavelet transform (DWT), and dual-tree complex wavelet transform (DT CWT) are commonly used transform techniques. Compared with spatial domain watermarking, frequency domain watermarking has stronger robustness and better imperceptibility to signal processing attacks. However, because frequency domain watermarking schemes lack a resynchronization mechanism, even if the watermark still exists in the image after being subjected to a desynchronization attack, it cannot be extracted. For example, the paper "Enhancing image watermarking with adaptive embedding parameter and PSNR guarantee" (reference [2]) published in IEEE Transactions on Multimedia, Vol. 21, No. 10, pp. 2447-2460 in 2019. To resist desynchronization attacks, researchers embed watermarks into geometrically invariant moments (such as Zernike moments) or geometrically invariant domains (such as Fourier-Mellin domains). However, these methods embed the watermark into the complete image, thus making them less robust to lossy cropping attacks. For example, the paper "Adaptive and robust Fourier-Mellin-based imagewatermarking for social networking platforms" published at the IEEE ICME conference in 2023 (Reference [3]) Convolutional neural networks (CNNs) have been introduced into image watermarking design, forming a new image watermarking technology, namely, deep learning-based watermarking technology. This type of watermarking technology has achieved remarkable results. However, it has three inherent limitations that restrict its practical application. First, these schemes exhibit strong resistance to signal processing attacks, but their robustness decreases significantly when subjected to desynchronization attacks. The second limitation comes from the inherent black-box nature of deep learning technology, which reduces the credibility of watermark extraction. Finally, due to computational limitations, current deep learning-based image watermarking techniques are limited in terms of applicable image resolutions, making them difficult to apply to practical image sizes such as 1080p, 2K, and 4K. Symmetric watermarking techniques, on the other hand, possess excellent resynchronization capabilities and are virtually resistant to all desynchronization attacks. They expand the watermark information into a spatial signal and superimpose it onto the pixels of the carrier image through symmetrical tiling. During watermark extraction, symmetric watermarks are detected by identifying symmetrical peaks using a self-convolution function. The peak map serves as a reference for resynchronization.For example, the paper "Local geometric distortions resilient watermarking scheme based on symmetry" (reference [1]) published in IEEE Transactions on Circuits and Systems for Video Technology, Vol. 31, No. 12, pp. 4826-4839. However, although symmetric watermarking has a strong resynchronization capability, it is essentially a spatial watermarking technique and inherits the inherent limitations of spatial watermarking, namely, poor robustness to signal processing attacks. Therefore, this invention designs a robust frequency domain symmetric image watermarking method for copyright confirmation, which improves the performance of existing watermarking algorithms. Summary of the Invention

[0003] In view of the above-mentioned shortcomings of the prior art, the purpose of this invention is to overcome the above limitations and provide a robust frequency domain symmetric image watermarking method for copyright confirmation, further improving the imperceptibility and robustness of existing robust image watermarking schemes.

[0004] The technical solution to achieve the objective of this invention is as follows:

[0005] A robust frequency-domain symmetric image watermarking method for copyright confirmation includes the following embedding and extraction processes;

[0006] Its embedding process includes the following steps:

[0007] Step 1: Generate watermark units from the watermarked message through watermark preprocessing;

[0008] Step 2: Flip the watermark unit to construct a symmetrical watermark;

[0009] Step 3: Read the Y channel of the image, i.e., the YCbCr color space; and divide the Y channel into non-overlapping two-level blocks, that is, first divide the Y channel into Z×Z primary sub-blocks in a non-overlapping and uniform manner, and then divide each primary sub-block into Q×Q secondary sub-blocks in a non-overlapping and uniform manner.

[0010] Step 4: Perform a frequency domain transformation on each secondary sub-block, embedding the symmetrical watermark into the specific frequency domain coefficients of the image's secondary sub-block, i.e., coefficients with equal and odd DCT domain row and column indices, such as AC. 3，3 AC 5，5 AC 7，7 Indexing starts from 1.

[0011] Step 5: Perform inverse frequency domain transformation on each secondary sub-block to obtain the watermarked Y channel, and finally obtain the watermarked image.

[0012] The extraction process includes the following steps:

[0013] Step A: Read the Y channel of the watermarked image and perform Wiener filtering to estimate the watermark residual signal;

[0014] Step B: Perform a self-convolution operation on the estimated watermark residual signal obtained in the previous step to obtain a symmetrical peak;

[0015] Step C: By using the position of the symmetrical peak, the distortion of the first-level sub-block is recovered. The first-level sub-block with recovered distortion is uniformly divided into Q×Q second-level sub-blocks without overlap. That is, frequency domain transformation is performed on each second-level sub-block, and watermark units are extracted from the frequency domain coefficients of the second-level sub-blocks.

[0016] Step D: Perform a state check on the extracted watermark unit and restore it to its original state, that is, the watermark unit that has not been flipped or rotated.

[0017] Step E: Decode the watermark through post-processing to obtain the embedded watermark message.

[0018] Furthermore, in step 1, the specific process of watermark preprocessing to generate watermark units is as follows:

[0019] (1) Encrypting messages with key Key1, for a message matrix of size M×M, m={m i，j |m i，j Given i ∈ {0, 1}, j ∈ [0, M-1], use key Key1 as the random number seed to generate a random matrix of the same size as the message matrix m, and XOR-encrypt m to obtain...

[0020] (2) Add a state judgment matrix. Similarly, use the key Key1 as a random number seed matrix of size Q×Q, and use it as the state judgment matrix τ. Covering the upper left corner region of τ, we obtain the initial watermark unit w. 0 ;

[0021] (3) Weak message detection and possible key2 encryption, for w 0 Perform weak message detection; if w 0 If it is a weak message, then the information portion needs to be encrypted again using key Key2 (excluding the state judgment matrix area), and the result w should be output. 1 The generated watermark unit w; otherwise, the initial watermark unit w is directly output. 0 As the generated watermark unit w; w 0 To be classified as a non-weak message, two conditions must be met: first, there must be no internal symmetry IS; and second, the availability of the state decision matrix τ must be guaranteed GA. τInternal symmetry (IS) refers to the similarity between the left and right halves of a watermark unit when flipped, and the similarity between the top and bottom halves when flipped. The internal symmetry IS is mathematically defined as follows: when the IS value is less than a threshold, w is considered... 0 This is a weak message:

[0022]

[0023] Due to the effects of flipping and potential 90-degree attacks, the watermark unit will have 8 possible states: initial state (state 1), horizontal flip (state 2), vertical flip (state 3), 180-degree rotation (state 4), vertical flip + 90-degree clockwise rotation (state 5), 90-degree clockwise rotation (state 6), 180-degree rotation + 90-degree clockwise rotation (state 7), and horizontal flip + 90-degree clockwise rotation (state 8); ensuring the availability of the state judgment matrix τ (GA). τ This refers to the situation where, apart from the initial state (state 1), if there exists a state judgment matrix for any other state that is identical to the original state judgment matrix, then the judgment w is determined. 0 This is a weak message:

[0024]

[0025] here This indicates that the watermark unit is set to state z; the definition of U[.] is as follows:

[0026]

[0027] In summary, detecting w 0 Whether a message is weak can be defined as follows:

[0028]

[0029] Here Det(w) 0 ) indicates w 0 The weak message detection result is considered to be w when its internal symmetry IS is less than a certain threshold ε or the state judgment matrix is ​​unavailable. 0 It is a weak message, i.e., Det(w 0 ) = true. When w 0 When it is a weak message, w 0 This cannot be used as the final watermark unit; a key is required. 2 As a random number seed, a random matrix of the same size as the message matrix m is generated. Perform XOR encryption again to obtain By combining the watermark judgment matrix τ, the key is obtained. 2 Encrypted watermark unit w 1Use this as the generated watermark unit w; otherwise, do not use key Key2 for encryption, and directly use the initial watermark unit w. 0 This is the generated watermark unit w.

[0030] Furthermore, in step 2, the watermark unit w is flipped continuously Z×Z times in the horizontal and vertical directions to construct a symmetrical watermark W.

[0031] Further, in step 3, the color space of the original image I is converted from RGB to YCbCr and its grayscale channel Y is extracted. First, Y is uniformly and non-overlappingly divided into Z×Z first-level sub-blocks, and then each first-level sub-block is uniformly and non-overlappingly divided into Q×Q second-level sub-blocks.

[0032] Furthermore, in step 4, a frequency domain transformation is performed on each secondary sub-block, and the symmetrical watermark W is embedded into the specific frequency domain coefficients F of the image secondary sub-block, that is, the coefficients with equal and odd row and column indices in the DCT domain, such as AC. 3，3 AC 5，5 AC 7，7 Indexing starts from 1:

[0033]

[0034] Here, △≥0 represents the watermark strength, F w This represents the frequency domain coefficients of the watermark.

[0035] Further, in step 5, an inverse frequency domain transform is performed on each secondary sub-block to obtain the watermarked Y channel. The color space is then converted from YCbCr to RGB, ultimately yielding the watermarked image I. w .

[0036] Further, in step A, the attacked image I... * The color space was converted from RGB to YCbCr and its grayscale channel Y was extracted. Wiener filtering was then applied to obtain the watermark residual signal.

[0037]

[0038] Here, μ Y and Let Y represent the local mean and local variance, respectively. mean represents calculating the mean, and max represents calculating the maximum value.

[0039] Further, in step B, a self-convolution operation is performed on the watermark residual signal to calculate its symmetry S:

[0040]

[0041] here, It is Zero-padding is applied to twice the original size. D is the downsampling function. FFT and IFFT represent the Fast Fourier Transform and its inverse, respectively. Peak points in S are selected to determine the location of the first-level sub-blocks.

[0042] Further, in step C, the distortion of the first-level sub-block is restored, and then the first-level sub-block is uniformly and non-overlappingly divided into Q×Q second-level sub-blocks. A frequency domain transformation is performed on each second-level sub-block, and then specific frequency domain coefficients F are applied. * (This embodiment of the invention uses AC in DCT transformation) 3，3 Extract watermark units from coefficients

[0043]

[0044] Furthermore, in step D, it is necessary to process the watermark unit. Perform a state determination. Construct the null hypothesis as follows:

[0045]

[0046] When H0 is rejected Watermark status judgment module The correlation with the original τ is the strongest. Therefore, we calculate the correlation ρ as follows:

[0047]

[0048] Here, ⊙ represents the XOR operation, and sum represents matrix summation. We construct the following null hypothesis set Λ:

[0049]

[0050] By selecting the slice 0 with the highest correlation ρ in Λ for rejection, the determination is made. The watermark state is restored to state 1.

[0051] Further, in step E, Key2 is used for decryption. Next, weak message detection is performed. If the decryption result is determined to be a weak message, it proves that key Key2 was used when embedding the watermark, and the final message matrix is ​​obtained by directly decrypting using key1. Otherwise, it proves that Key2 encryption was not used when embedding the watermark. Key2 encryption needs to be performed again, followed by Key1 decryption to obtain the final extracted watermark message.

[0052] This invention proposes a robust frequency-domain symmetric image watermarking method for copyright verification. Utilizing the features of base images with different frequency-domain coefficients, it embeds symmetric watermarks by modulating specific frequency-domain coefficients, solving the problem of traditional frequency-domain watermarking schemes lacking a resynchronization mechanism when subjected to desynchronization attacks, thus achieving strong robustness at high visual quality. By constructing symmetric peaks with a consistent number of peaks across images of different resolutions, the robustness of the watermark against large-area cropping attacks is enhanced. Furthermore, a watermark preprocessing module satisfying the Kerckhoffs principle is designed to ensure the security of the watermarking algorithm even when it is publicly available.

[0053] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0054] It retains high-precision watermark resynchronization capability even under various attacks, enhancing the robustness of the watermark against various attacks while ensuring high image quality, especially against large-area cropping. It also exhibits stable performance and good generalization when applied to images of different resolutions. Attached Figure Description

[0055] Figure 1 This is a flowchart illustrating the watermark embedding process according to an embodiment of the present invention.

[0056] Figure 2 This is a flowchart illustrating the watermark extraction process according to an embodiment of the present invention.

[0057] Figure 3 This is a flowchart of the watermark preprocessing process according to an embodiment of the present invention.

[0058] Figure 4 This is a schematic diagram of the internal symmetry (IS) of an embodiment of the present invention.

[0059] Figure 5 This is a schematic diagram of the eight watermark unit states in an embodiment of the present invention.

[0060] Figure 6 This is a flowchart of the watermark post-processing according to an embodiment of the present invention.

[0061] Figure 7 This is a comparison table showing the image accuracy against cropping in embodiments of the present invention on a public dataset, compared with existing methods.

[0062] Figure 8 This is a comparison table showing the image accuracy of the embodiments of the present invention against different types of attacks on a public dataset and against existing methods.

[0063] Figure 9 This is a comparison table of PSNR between embodiments of the present invention and existing methods on a public dataset.

[0064] Figure 10This is a comparison chart of the subjective visual quality of the embodiments of the present invention with existing methods on a public dataset. Detailed Implementation

[0065] The present invention will be further described below with reference to the accompanying drawings and specific embodiments.

[0066] like Figures 1-2 As shown, this embodiment of the invention implements a robust frequency domain symmetric image watermarking method for copyright confirmation, which includes watermark preprocessing, flipping to construct a symmetric watermark, image two-level segmentation, symmetric watermark frequency domain embedding, symmetric peak detection, distortion recovery, watermark frequency domain extraction, state judgment, and watermark decoding, constituting the entire robust image watermarking framework. Figure 1 The embedding process of this invention is given; Figure 2 The extraction process of this invention is given.

[0067] This embodiment includes the following embedding steps:

[0068] S1: Generate watermark units from the watermark message through watermark preprocessing. Figure 3 The flowchart of watermark preprocessing is given in the document.

[0069] The specific steps for watermark preprocessing are as follows:

[0070] S1.1: Encrypt the message with key Key1. For a message matrix of size M×M, m={m i，j |m i，j ∈{0, 1}, i, j∈[0, M-1]}, we use the key Key1 as the random number seed to generate a random matrix of the same size as the message matrix m, and then XOR-encrypt m to obtain In this embodiment, M = 8.

[0071] S1.2: Add a state judgment matrix. Similarly, use the key Key1 as a random number seed matrix of size Q×Q, and use it as the state judgment matrix τ. Covering the upper left corner region of τ, we obtain the initial watermark unit w. 0 In this embodiment, Q = 10.

[0072] S1.3: Weak message detection and possible key2 encryption. (For w) 0 Perform weak message detection; if w 0 If it is a weak message, then the information portion needs to be encrypted again using key Key2 (excluding the state judgment matrix area), and the result w should be output. 1 This is the generated watermark unit w. Otherwise, directly output the initial watermark unit w. 0 This is the generated watermark unit w. 0To be classified as a non-weak message, two conditions must be met: first, there must be no internal symmetry (IS); second, the availability of the state decision matrix τ must be guaranteed (GA). τ Internal symmetry (IS) refers to the similarity between the left and right halves of the watermark unit when flipped, and the similarity between the top and bottom halves when flipped. Figure 4 A schematic diagram of internal symmetry (IS) is given. Internal symmetry (IS) can be mathematically defined as follows: when the IS value is less than a certain threshold ε, w is considered... 0 This is a weak message:

[0073]

[0074] Due to the effects of flipping and possible 90-degree attacks, the watermark unit will have 8 possible states. Figure 5 Eight watermark unit state diagrams are given. The availability of the state judgment matrix τ is guaranteed (GA). τ This means that if, except for state 1, there exists a state judgment matrix that is consistent with the original state judgment matrix, then the judgment w is determined. 0 This is a weak message:

[0075]

[0076] here This indicates that the watermark unit is set to state z. The definition of U[.] is as follows:

[0077]

[0078] In summary, detecting w 0 Whether a message is weak can be defined as follows:

[0079]

[0080] Here Det(w) 0 ) indicates w 0 Weak message detection results are considered invalid when the internal symmetry (IS) is less than a certain threshold ε or the state decision matrix is ​​unavailable. 0 It is a weak message, i.e., Det(w 0 ) = true. When w 0 When it is a weak message, w 0 This cannot be used as the final watermark unit; we need to use the key. 2 As a random number seed, a random matrix of the same size as the message matrix m is generated. Perform XOR encryption again to obtain By combining the watermark judgment matrix τ, the key is obtained. 2 Encrypted watermark unit w 1This is used as the generated watermark unit w. Otherwise, the initial watermark unit w is directly used instead of being encrypted with key Key2. 0 The generated watermark unit w. In this embodiment, ε = 14.

[0081] S2: Construct a symmetrical watermark by flipping the watermark unit. The watermark unit w is flipped Z×Z times continuously in both the horizontal and vertical directions to construct a symmetrical watermark W. In this embodiment, Z = 8.

[0082] S3: Convert the color space of the original image I from RGB to YCbCr and extract its grayscale channel Y. First, divide Y into Z×Z primary sub-blocks uniformly and non-overlappingly, and then divide each primary sub-block into Q×Q secondary sub-blocks uniformly and non-overlappingly.

[0083] S4: Perform frequency domain transformation on each secondary sub-block, and embed the symmetrical watermark W into the special frequency domain coefficients F of the image secondary sub-block (the embodiment of this invention uses AC in DCT transformation). 3，3 In the coefficients:

[0084]

[0085] Here, △≥0 represents the watermark strength, F w This represents the frequency domain coefficients of the watermark.

[0086] S5: Perform inverse frequency domain transform on each secondary sub-block to obtain the watermarked Y channel. Convert from YCbCr color space to RGB color space to obtain the final watermarked image I. w .

[0087] This embodiment includes the following extraction steps:

[0088] SA: The attacked image I * The color space was converted from RGB to YCbCr and its grayscale channel Y was extracted. Wiener filtering was then applied to obtain the watermark residual signal.

[0089]

[0090] Here, μ Y and Let Y represent the local mean and local variance, respectively. `mean` calculates the mean, and `max` calculates the maximum value.

[0091] SB: Perform a self-convolution operation on the watermark residual signal and calculate its symmetry S:

[0092]

[0093] here, It is Zero-padding is applied to twice the original size. D is the downsampling function. FFT and IFFT represent the Fast Fourier Transform and its inverse, respectively. Peak points in S are selected to determine the location of the first-level sub-blocks.

[0094] SC: Restore the distortion of the first-level sub-block, and then divide the first-level sub-block into Q×Q second-level sub-blocks uniformly and non-overlappingly. Perform frequency domain transformation on each second-level sub-block, and then apply specific frequency domain coefficients F. * (This embodiment of the invention uses AC in DCT transformation) 3，3 Extract watermark units from coefficients

[0095]

[0096] SD: Perform state checks on the extracted watermark units and restore them to their original state. The null hypothesis is constructed as follows:

[0097]

[0098] When H0 is rejected Watermark status judgment module The correlation with the original τ is the strongest. Therefore, we calculate the correlation p as follows:

[0099]

[0100] Here, ⊙ represents the XOR operation, and sum represents matrix summation. We construct the following null hypothesis set Λ:

[0101]

[0102] By selecting the slice 0 with the highest correlation ρ in Λ for rejection, the determination is made. The watermark state is restored to state 1.

[0103] SE: Watermark decoding is performed through watermark post-processing to obtain the embedded watermark message. Figure 6 A diagram illustrating the watermark decoding process is provided. Key2 decryption is used. Next, weak message detection is performed. If the decryption result is determined to be a weak message, it proves that key Key2 was used when embedding the watermark, and the final message matrix is ​​obtained by directly decrypting using key1. Otherwise, it proves that Key2 encryption was not used when embedding the watermark. Key2 encryption needs to be performed again, followed by Key1 decryption to obtain the final extracted watermark message.

[0104] In this example, the effective watermark payload is 40 bits, which is encoded using CRC-16 and BCH(63, 56, 1) to form a 63-bit codeword. When the number of error bits is ≤1, the watermark can be completely extracted and verified. The test images consist of 100 images at different resolutions, including 30 images each of 512×512, 1080p, and 2K resolutions, and 10 images of random resolutions. Furthermore, tests were conducted on images in different formats such as JPG, BMP, and PNG to verify that the watermarking scheme can be used with various image formats.

[0105] This example uses Peak Signal-to-Noise Ratio (PSNR) and Image Accuracy as evaluation metrics: the higher the PSNR value, the better the imperceptibility, the higher the Image Accuracy, and the better the watermark robustness.

[0106]

[0107] Figure 7 This paper compares the robustness of our invention against cropping attacks with existing methods on publicly available image datasets. The results show that, with the same watermark capacity, our solution achieves the best robustness against cropping attacks. Figure 8 This paper compares the robustness of the present invention against different types of attacks with existing methods on publicly available image datasets (bold data represents the best results, and underlined data represents the second-best results). The results show that, under the same watermark capacity, the present invention achieves optimal or second-best robustness against other different attack types. Figure 9 The PSNR of this invention was compared with that of existing methods on a publicly available video dataset. The results show that this invention has better imperceptibility under the same watermark capacity. Figure 10 This image compares the subjective visual quality of the present invention with the latest watermarking schemes. It is evident that the watermarking effect of the present invention is not noticeable in either complex or smooth areas (magnified areas), exhibiting good subjective imperceptibility. Both objective test results and subjective visual observation demonstrate that the present invention offers high visual appeal and effectively guarantees good visual quality for watermarked images.

Claims

1. A robust frequency-domain symmetric image watermarking method for copyright confirmation, characterized in that, This includes the following embedding and extraction processes; Its embedding process includes the following steps: Step 1: Generate watermark units from the watermarked message through watermark preprocessing; Step 2: Flip the watermark unit to construct a symmetrical watermark; Step 3: Read the Y channel of the image, i.e., the YCbCr color space; and perform non-overlapping two-level segmentation of the Y channel, that is, first divide the Y channel into non-overlapping and uniform blocks. Each first-level sub-block is then divided into non-overlapping, uniformly sized sub-blocks. Two-level sub-blocks; Step 4: Perform frequency domain transformation on each secondary sub-block and embed the symmetrical watermark into the special frequency domain coefficients of the image secondary sub-block, that is, the coefficients whose DCT domain row and column indices are equal and all are odd numbers; Step 5: Perform inverse frequency domain transformation on each secondary sub-block to obtain the watermarked Y channel, and finally obtain the watermarked image; The extraction process includes the following steps: Step A: Read the Y channel of the watermarked image and perform Wiener filtering to estimate the watermark residual signal; Step B: Perform a self-convolution operation on the estimated watermark residual signal obtained in Step A to obtain a symmetrical peak; Step C: By analyzing the position of the symmetrical peaks, recover the distortion of the first-level sub-blocks, and then uniformly divide the recovered distortion-corrected first-level sub-blocks into non-overlapping sections. Each secondary sub-block is processed by performing a frequency domain transformation on each secondary sub-block and extracting watermark units from the frequency domain coefficients of the secondary sub-block. Step D: Perform a state check on the extracted watermark unit and restore it to its original state, that is, the watermark unit that has not been flipped or rotated. Step E: Decode the watermark through post-processing to obtain the embedded watermark message; In step 1, the specific process of watermark preprocessing to generate watermark units is as follows: (1) Key Encrypted messages, for a size of message matrix Using a key As a random number seed generation and message matrix Random matrices of the same size, for XOR encryption is performed to obtain ; (2) Add a state judgment matrix, also using the key. The seed size for generating random numbers is [size]. A random matrix, used as the state judgment matrix. ,Will Coverage The initial watermark unit is obtained from the upper left corner area. ; (3) Weak message detection and possible Encryption, for Perform weak message detection, if If it is a weak message, then a key is needed. The information portion is encrypted again, excluding the state judgment matrix area, and the result is output. For the generated watermark unit Otherwise, directly output the initial watermark unit. As the generated watermark unit ; To be classified as a non-weak message, two conditions must be met: first, there must be no internal symmetry. Secondly, ensure the state judgment matrix Availability Internal symmetry This refers to the similarity between the left and right halves of the watermark unit when flipped, and the similarity between the top and bottom halves when flipped; regarding internal symmetry. We make the following mathematical definition: when When the value is less than the threshold, a determination is made. This is a weak message: Due to the influence of flipping and possible 90-degree attacks, the watermark unit will have 8 possible states, ensuring the state judgment matrix Availability This refers to the situation where, apart from the initial state, if the state judgment matrix of any other state is consistent with the original state judgment matrix, then a judgment is made. This is a weak message: here This indicates that the watermark unit is set to a certain state. The definition is as follows: In summary, the detection Whether a message is weak can be defined as follows: here express The weak message detection results, when its internal symmetry Less than a certain threshold If the state judgment matrix is ​​unavailable, it is considered that... It is a weak message, that is ;when When it is a weak message, This cannot be used as the final watermark unit; a key is required. As a random number seed generation and message matrix Random matrices of the same size, for Perform XOR encryption again to obtain Combined with watermark judgment matrix ,get Encrypted watermark unit Use it as the generated watermark unit Otherwise, do not use the key. Encryption, directly using the initial watermark unit As the generated watermark unit .

2. The robust frequency-domain symmetric image watermarking method for copyright confirmation as described in claim 1, characterized in that, In step 2, the process of constructing a symmetrical watermark by flipping the watermark unit is as follows: The watermark unit... Continuous in both horizontal and vertical directions Second flip to construct a symmetrical watermark .

3. The robust frequency-domain symmetric image watermarking method for copyright confirmation as described in claim 1, characterized in that, In step 3, the specific process of secondary segmentation is as follows: The original image is divided into blocks... The color space was converted from RGB to YCbCr and its grayscale channels were extracted. First of all Divide into uniform and non-overlapping groups Each first-level sub-block is further divided into equal and non-overlapping sub-blocks. Two secondary sub-blocks.

4. The robust frequency-domain symmetric image watermarking method for copyright confirmation as described in claim 1, characterized in that, In step 4, a frequency domain transformation is performed on each secondary sub-block, and the specific frequency domain coefficients corresponding to the symmetrical watermark W are embedded into the secondary sub-block of the image. : here Indicates watermark strength. This represents the frequency domain coefficients of the watermark.

5. The robust frequency-domain symmetric image watermarking method for copyright confirmation as described in claim 1, characterized in that, In step C, the specific process of extracting the watermark unit is as follows: restore the distortion of the first-level sub-block, and then divide the first-level sub-block into uniform and non-overlapping sections. Each secondary sub-block is divided into two sub-blocks; a frequency domain transform is performed on each secondary sub-block, and specific frequency domain coefficients are obtained. Extract watermark unit : 。 6. The robust frequency-domain symmetric image watermarking method for copyright confirmation as described in claim 1, characterized in that, In step D, the specific process of determining the state of the watermark unit is as follows: The null hypothesis is constructed as follows: When Rejected hour, Watermark status judgment module With the original The correlation is the highest; calculate the correlation. Then it is as follows: here `x` represents the XOR operation, and `sum` represents matrix summation; here, we construct the following null hypothesis set. : choose Highest correlation of By refusing, it is confirmed. The watermark state is restored to state 1.

7. The robust frequency-domain symmetric image watermarking method for copyright confirmation as described in claim 1, characterized in that, In step E, the specific method for watermark decoding is as follows: using Decryption Next, weak message detection is performed. If the decryption result is determined to be a weak message, it proves that a key was used when embedding the watermark. Use directly Decryption yields the final message matrix. ; Otherwise, it proves that no watermark was used during watermark embedding. Encryption is required. Encrypt, then proceed Decryption yields the final watermark extraction message. .

8. The robust frequency-domain symmetric image watermarking method for copyright confirmation as described in claim 1, characterized in that, The message matrix of .

9. The robust frequency-domain symmetric image watermarking method for copyright confirmation as described in claim 1, 2, 3, or 5, characterized in that, The size of each watermark unit and the number of secondary sub-blocks are consistent. .

10. The robust frequency-domain symmetric image watermarking method for copyright confirmation as described in claim 1, 2, or 3, characterized in that, The number of flips is consistent with the number of first-level sub-blocks. .

11. The robust frequency-domain symmetric image watermarking method for copyright confirmation as described in claim 4 or 5, characterized in that, The frequency domain coefficients are the AC coefficients of the DCT transform. .

Images