Self-adaptive image watermark embedding and recovering system

Abstract

The invention relates to the technical field of digital watermarking, and discloses a self-adaptive image watermark embedding and recovering system, which comprises an image preprocessing module used for carrying out discrete wavelet transform on a carrier image to be embedded with a watermark, decomposing the image into a plurality of frequency sub-bands, and selecting a low-frequency sub-band as a watermark embedding region; the watermark preprocessing module is used for carrying out orthogonal transformation on the watermark image, so that orthogonality is kept among a plurality of watermarks, and mutual interference of the watermarks is reduced; and the watermark embedding module is used for embedding the watermark into the low-frequency sub-band of the carrier image based on the output of the watermark preprocessing module. According to the method, the concealment and robustness of the watermarks are improved through the technologies of self-adaptive optimization of the watermark strength, redundancy embedding and orthogonal transformation, and compared with the prior art, the interference between the watermarks is effectively reduced, and the resistance of the watermarks to attacks such as compression, noise, cutting and rotation is enhanced.

Description

Technical Field

[0001] The present invention relates to the technical field of digital watermarking, and specifically to an adaptive image watermark embedding and recovery system. Background Art

[0002] In digital watermarking technology, watermark embedding and extraction are important means to protect the copyright and integrity of digital content. Existing watermarking technologies are widely used in the protection of multimedia content such as images, videos, and audios, mainly by embedding watermark information to achieve functions such as copyright protection and information verification. However, despite certain progress in the existing technologies, many watermarking technologies still have some obvious deficiencies, especially in aspects such as the adaptability of watermark embedding strength, interference between multiple watermarks, robustness, and extraction accuracy.

[0003] In the existing technology, the watermark embedding method adopts a fixed watermark strength and position, without considering the differences in image content. This fixed embedding method is likely to cause visual interference in complex regions and low-contrast regions of the image, resulting in increased watermark visibility. In addition, when dealing with multiple watermarks, many technologies fail to effectively solve the problems of mutual interference and overlap between watermarks, thus affecting the extraction accuracy. Although some technologies attempt to rely too much on low-frequency subbands during the embedding process to ensure invisibility, when encountering attacks such as compression, noise interference, and cropping, the robustness of the watermark is poor and it cannot be stably recovered. More importantly, the existing watermarking technologies do not adequately control the extraction accuracy, and there are often large errors during the recovery process. Especially after the image has been processed such as compressed or cropped, the accuracy and stability of watermark extraction cannot be effectively guaranteed. Summary of the Invention

[0004] Aiming at the deficiencies of the existing technology, the present invention provides an adaptive image watermark embedding and recovery system, which solves the problems of poor watermark invisibility, weak anti-attack ability, and insufficient watermark recovery accuracy in the existing technology.

[0005] To achieve the above objectives, the present invention is realized through the following technical solutions: An adaptive image watermark embedding and recovery system, comprising:

[0006] An image preprocessing module, used for performing discrete wavelet transform on the carrier image to be embedded with a watermark, decomposing the image into multiple frequency subbands, and selecting the low-frequency subband as the watermark embedding area;

[0007] A watermark preprocessing module, used for performing orthogonal transform on the watermark image to keep multiple watermarks orthogonal to each other and reduce watermark mutual interference;

[0008] A watermark embedding module, which is used to embed a watermark into the low-frequency subband of a carrier image based on the output of the watermark preprocessing module, and adjust the embedding strength through singular value decomposition to match the watermark information with the carrier image features, thereby improving the robustness and invisibility;

[0009] A watermark optimization and enhancement module, which is used to optimize the watermark embedding parameters, and by dynamically adjusting the embedding strength, while maintaining the invisibility after the watermark is embedded, improve the information capacity and robustness, and reduce the similarity between watermarks;

[0010] A watermark extraction and recovery module, which is used to recover the watermark information from the image with the embedded watermark based on the inverse discrete wavelet transform and the inverse singular value decomposition, and optimize the extraction parameters based on the optimal control theory, so that it can adaptively optimize the extraction accuracy under different attack conditions and improve the stability of watermark recovery;

[0011] A robustness enhancement module, which is used to improve the extraction success rate of the watermark under JPEG compression, cropping, noise, and rotation attacks, enhance the anti-cropping ability of the watermark through a spatial distributed embedding strategy, enhance the anti-rotation ability of the watermark through the Fourier-Mellin transform, and adopt an adaptive filtering strategy to enhance the anti-noise ability of the watermark.

[0012] Preferably, the image preprocessing module performs a discrete wavelet transform on the carrier image to generate a low-frequency subband and a high-frequency subband, and selects the low-frequency subband as the watermark embedding area to improve the robustness of the watermark.

[0013] Preferably, the watermark preprocessing module performs a transform on the watermark image using an orthogonal transform to keep multiple watermarks orthogonal to each other and reduce the mutual interference between watermarks, where the orthogonal transform includes a discrete cosine transform or a discrete Fourier transform.

[0014] Preferably, the watermark embedding module decomposes the low-frequency subband of the carrier image through singular value decomposition, embeds the watermark information into the obtained singular value matrix, and the embedded singular value matrix is used to reconstruct the low-frequency subband and combine with the inverse discrete wavelet transform to reconstruct the image with the watermark.

[0015] Preferably, the watermark optimization and enhancement module optimizes the watermark embedding strength based on the variational method and dynamically adjusts the embedding parameters through an optimization algorithm, so that while ensuring the invisibility after the watermark is embedded, it improves the information capacity and robustness.

[0016] Preferably, the watermark extraction and recovery module performs a discrete wavelet transform on the image with the watermark, extracts the low-frequency subband, calculates the embedded watermark information based on the singular value decomposition, and then uses the inverse orthogonal transform to recover the original watermark.

[0017] Preferably, the watermark extraction and restoration module uses the optimal control theory to dynamically adjust the watermark restoration parameters, enabling it to adaptively optimize the extraction accuracy under different attack conditions and improve the stability of watermark restoration.

[0018] Preferably, for JPEG compression attacks, the robustness enhancement module adopts a quantization adaptive strategy to adjust the quantization step of watermark embedding, enabling the watermark information to be restored even after JPEG compression.

[0019] Preferably, the robustness enhancement module uses a spatial distributed embedding strategy to store the watermark information in multiple regions, enhancing the resistance of the watermark to cropping attacks.

[0020] Preferably, the robustness enhancement module uses the Fourier-Mellin transform to transform the embedded watermark information to enhance the robustness of the watermark against rotation attacks. For noise attacks, the robustness enhancement module adopts an adaptive filtering strategy, including Wiener filtering and median filtering, to reduce the impact of noise on watermark extraction, and combines redundant coding to improve the stability of watermark extraction.

[0021] The present invention provides an adaptive image watermark embedding and restoration system, which has the following beneficial effects:

[0022] 1. The present invention adopts an adaptive watermark strength optimization technique, combines the variational method and the optimal control theory, and dynamically adjusts the strength and position of watermark embedding to ensure that while maximizing the watermark capacity, the impact on image quality is minimized. Compared with the common fixed watermark strength settings in the prior art, the optimization scheme of the present invention can automatically adjust the watermark parameters according to the image content, improving the concealment of the watermark and reducing visual interference.

[0023] 2. The present invention preprocesses the watermark through orthogonal transformation to keep multiple watermark information orthogonal, avoiding mutual interference between watermarks. Compared with traditional watermark technologies, the present invention can effectively improve the independence in a multi-watermark system, significantly enhancing the extraction accuracy of each watermark information. Especially in the complex environment of multiple watermark embeddings, it demonstrates good stability and reliability.

[0024] 3. The present invention introduces a redundant embedding technology to distribute the watermark information to multiple regions of the image. Even if the image undergoes operations such as cropping and rotation, the watermark can still be restored from the remaining part. Compared with the prior art where the watermark is embedded in a single region of the image and is vulnerable to cropping attacks, the present invention significantly improves the anti-cropping ability and robustness of the watermark, ensuring the stability of the watermark in complex scenarios.

[0025] 4. The present invention adopts adaptive filtering and anti-compression enhancement technologies to effectively reduce the impact of noise and compression on watermark recovery. In the prior art, when facing image compression or noise, the accuracy of watermark extraction is often low and is easily interfered. By optimizing the embedding process and combining denoising processing, the present invention can significantly improve the recovery accuracy of watermarks in a compressed and noise-interfered environment, ensuring the efficient extraction of watermark information. BRIEF DESCRIPTION OF THE DRAWINGS

[0026] Figure 1 It is a system architecture diagram of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0027] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the accompanying drawings of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the protection scope of the present invention.

[0028] Please refer to the appended Figure 1 , the embodiments of the present invention provide an adaptive image watermark embedding and recovery system, including:

[0029] An image preprocessing module, configured to perform discrete wavelet transform on the carrier image to be embedded with a watermark, decompose the image into multiple frequency subbands, and select the low-frequency subband as the watermark embedding area;

[0030] In this embodiment, the image preprocessing module first decomposes the input carrier image using discrete wavelet transform (DWT). DWT is a common signal processing method that can divide an image into different frequency components for better information processing. In the present invention, the DWT transform decomposes the input carrier image I(x, y) into four subbands, namely the low-frequency subband LL, the horizontal high-frequency subband LH, the vertical high-frequency subband HL, and the diagonal high-frequency subband HH:

[0031]

[0032] Wherein:

[0033] The LL subband mainly contains the overall structural information of the image, with relatively low visual contrast but concentrated energy;

[0034] The LH, HL, and HH subbands contain the edge, texture, and high-frequency detail information of the image, and are usually more susceptible to noise and compression.

[0035] In a possible implementation, DWT uses the Haar wavelet as the basis function, which is relatively simple to calculate and suitable for the field of image processing. The Haar wavelet transform can calculate each subband through the following formula:

[0036]

[0037] Among them, i and j respectively represent the row and column indices of the sub-band matrix, and I(x, y) is the pixel value of the input image.

[0038] Generally, the embedding region of the watermark is selected in the low-frequency sub-band LL because the data in this part has the least impact on the image and the energy is relatively concentrated, enabling the watermark to better resist compression and noise interference after embedding. At the same time, the high-frequency sub-bands LH, HL, and HH are easily affected by filtering, quantization, and compression, and are not suitable as watermark embedding regions.

[0039] In some embodiments, in order to further reduce the impact of the watermark on the image quality, the image preprocessing module also performs singular value decomposition (SVD) on the LL sub-band to extract key singular value information for watermark embedding. Specifically, the SVD transformation of the LL sub-band can be expressed as follows:

[0040] LL = USV T

[0041] Where:

[0042] U is the left singular matrix, containing orthogonal basis information;

[0043] S is the singular value matrix, containing the main energy information of the image;

[0044] V is the right singular matrix, having the same orthogonal property as U.

[0045] As an option, in the image preprocessing stage, only the singular value matrix S is extracted for watermark embedding, while the U and V matrices are kept unchanged, which can ensure that the impact of the embedding process on the image structure is minimized.

[0046] In another possible implementation, in order to further improve the adaptability of the watermark embedding region, the image preprocessing module can also combine edge detection algorithms, such as the Canny operator or the Sobel operator, to pre-analyze the carrier image and determine the main structure region of the image. Specifically, the Canny operator can calculate the gradient information of the image:

[0047]

[0048] Where:

[0049] G(x, y) represents the gradient intensity of the image at the pixel point (x, y);

[0050] G x and G yThe gradient calculation values in the horizontal and vertical directions respectively can be calculated through the Sobel convolution kernel.

[0051] Before watermark embedding, regions with lower gradients can be selected for watermark embedding to reduce the impact of embedding on the edge structure of the image. This method can further optimize the watermark embedding position and improve the imperceptibility of the watermark.

[0052] In some embodiments, to improve the calculation efficiency, the image can be processed in blocks. Specifically, the carrier image can be divided into multiple small blocks of M×N, and each small block is respectively subjected to DWT and SVD transformations. This method helps to enhance the local adaptability of the watermark, enabling the embedding strength of the watermark in different regions to be dynamically adjusted, thereby improving the overall performance of the system.

[0053] Generally speaking, the core objective of the image preprocessing module is to extract the optimal watermark embedding region to ensure the stability and concealment after watermark embedding. Through discrete wavelet transform, the image can be decomposed into different frequency sub-bands, providing appropriate storage space for the watermark. Combining singular value decomposition can further optimize the embedding process, enabling the watermark information to be embedded with the least distortion. For some special applications, methods such as edge detection and adaptive block division can also be combined to further optimize the watermark embedding position, making the entire system have stronger adaptability and stability.

[0054] The watermark preprocessing module is used to perform orthogonal transformation on the watermark image to keep multiple watermarks orthogonal and reduce the interference between watermarks;

[0055] Generally, the watermark preprocessing module will first perform orthogonal transformation on the watermark image. The main objective of orthogonal transformation is to minimize the correlation between multiple watermarks, ensuring that each watermark information is as independent as possible during the embedding process and avoiding interference and information overlap between watermarks. The advantage of this is that when multiple watermarks are embedded in the same carrier image, they can effectively avoid influencing each other, thereby improving the extraction accuracy of the watermark. After orthogonal transformation, the watermark information will have higher anti-interference ability and better concealment.

[0056] In this embodiment, the watermark preprocessing module adopts discrete cosine transform (DCT) or discrete Fourier transform (DFT) as the orthogonal transformation method. Specifically, the watermark image W is first subjected to DCT transformation to obtain the transformation result W′ in the frequency domain, which can be expressed as:

[0057] W′ = TW

[0058] Among them, W is the original watermark image, and T is an orthogonal transformation matrix. Common transformations include DCT or DFT, specifically depending on the implementation method. Through the orthogonal transformation, the watermark information is decomposed into different frequency components. Generally, the low-frequency components contain the basic information of the image, while the high-frequency components contain more details. The watermark information in the high-frequency components is relatively stable. Therefore, the watermark preprocessing module will give priority to processing the low-frequency components to ensure the robustness of the watermark after embedding.

[0059] As an option, after the orthogonal transformation, the watermark information can be further optimized. Generally, by selecting an appropriate frequency range and the strength of the watermark after transformation, the mutual interference between the watermark and the image can be further reduced, and the extraction accuracy of the watermark can be improved. In this process, the high-frequency part of the watermark is suppressed, avoiding high-frequency noise from interfering with the image quality and watermark extraction.

[0060] Specifically, the orthogonal transformation adopted in this embodiment includes the discrete cosine transform (DCT), which is commonly used in the fields of image compression and image processing. DCT can transform the image data from the spatial domain to the frequency domain and embed the watermark information into the low-frequency part of the image, thereby enhancing its anti-interference ability. The watermark information W' after the DCT transformation is as follows:

[0061]

[0062] where W p,q is the pixel value of the watermark image, M and N are the height and width of the watermark image respectively, and i and j are the frequency coordinates after transformation. Through this transformation, the frequency components of the watermark image are decomposed into a set of coefficients. The low-frequency components retain the main information of the watermark, while the high-frequency components retain more details.

[0063] In another possible implementation, in order to further improve the independence of the watermark, the watermark preprocessing module can also combine the singular value decomposition (SVD) to process the watermark information. SVD is a matrix decomposition method that can decompose the watermark information matrix into three matrices: U, S, and V, where S is the matrix containing singular values, representing the energy and information of the watermark image. By performing SVD processing on the watermark, we can obtain the main information of the watermark image, thereby optimizing the subsequent embedding process.

[0064] Specifically, the watermark image W can be decomposed by SVD into:

[0065] W = USV T

[0066] where:

[0067] U and V are orthogonal matrices, representing the eigenvectors of the watermark image;

[0068] S is a diagonal matrix containing the singular values of the watermark image.

[0069] During the processing, the watermark preprocessing module adjusts and optimizes the singular value matrix S of the watermark to further enhance the anti-interference ability and extraction accuracy of the watermark image. The watermark image after this processing is more suitable for embedding because of its strong independence and is not easily affected by other watermarks or image noise.

[0070] In some embodiments, the watermark preprocessing module also combines frequency-domain compression techniques, such as quantization strategies, to further reduce the redundancy of the watermark information. The frequency-domain compression technique can compress the volume of the watermark information, reduce the impact on the image quality during embedding, and improve the transmission efficiency of the watermark.

[0071] In summary, the core function of the watermark preprocessing module is to perform orthogonal transformation and processing on the watermark image to ensure that the watermark information has high independence and robustness, and can effectively resist image compression, noise, and other attacks. By selecting appropriate transformation methods, such as DCT, DFT, or SVD, the watermark preprocessing module in this embodiment can maximize the stability of the watermark during the embedding process and provide high-quality watermark information for the watermark embedding module, thus ensuring the invisibility and extraction accuracy of the final image watermark.

[0072] The watermark embedding module is used to embed the watermark into the low-frequency sub-band of the carrier image based on the output of the watermark preprocessing module, and adjust the embedding strength through singular value decomposition to make the watermark information match the characteristics of the carrier image, improving robustness and invisibility;

[0073] In this embodiment, the main task of the watermark embedding module is to embed the watermark into the carrier image through appropriate technical means based on the low-frequency sub-band (such as the LL sub-band) provided by the image preprocessing module and the processed watermark information provided by the watermark preprocessing module. In a specific implementation, the watermark embedding module first performs singular value decomposition (SVD) on the low-frequency sub-band, then embeds the watermark information into the decomposed singular value matrix, and finally reconstructs the image through inverse SVD to generate the carrier image with the embedded watermark.

[0074] Watermark Embedding Process

[0075] In this embodiment, the specific steps of watermark embedding are as follows:

[0076] First, obtain the low-frequency sub-band of the image (i.e., the LL sub-band obtained after DWT decomposition) from the image preprocessing module and perform singular value decomposition on it. Specifically, let the watermark image be W i ′ after watermark preprocessing, then perform SVD processing on the low-frequency sub-band LL of the carrier image to obtain the following decomposition result:

[0077] LL = USV T

[0078] Among them, U and V are the left and right singular matrices respectively, which contain the main feature information of the image; S is the singular value matrix, which contains the energy information of the image.

[0079] Next, the watermark embedding module embeds the orthogonally transformed watermark information W i ′ into the singular value matrix S. The embedding process uses the following formula:

[0080]

[0081] Where:

[0082] S′ is the singular value matrix after embedding the watermark;

[0083] W i ′ is the i-th watermark information after preprocessing;

[0084] α i is a factor that controls the watermark strength. Usually, it is optimized by the variational method to balance the amount of watermark information and the embedding strength, ensuring the balance between the maximum information capacity and the minimum visual impact.

[0085] In a possible implementation, the embedding strength α i can be dynamically adjusted through an optimization algorithm (such as a genetic algorithm or a gradient descent method) to ensure the adaptive adjustment of the watermark strength in different regions. Specifically, in the low-contrast region of the image, a stronger watermark is embedded; while in the high-contrast region of the image, a weaker watermark is embedded.

[0086] As an option, if it is necessary to improve the stability and robustness of the system, the watermark embedding module can also combine the singular value compression technology to compress the embedded watermark information to reduce the redundancy of the watermark data and improve the embedding efficiency. Specifically, during the embedding process, some smaller singular values in the singular value matrix S can be suppressed, thereby improving the concealment of the watermark and avoiding affecting the image quality.

[0087] The embedded singular value matrix S′ will then be reconstructed together with U and V through the Inverse Singular Value Decomposition (InverseSVD) to recover the new low-frequency subband LL′, and its expression is:

[0088] LL′ = US′V T

[0089] In some embodiments, the watermark embedding module can also perform local watermark embedding on other high-frequency subbands LH, HL, and HH to further increase the watermark capacity and anti-attack ability. Although these high-frequency subbands are vulnerable to compression and noise interference, if properly embedded, they can also enhance the redundancy and recovery ability of the watermark in specific scenarios. At this time, through different weightings of each subband, the watermark information can be embedded with different intensities to ensure the image quality and the stability of the watermark.

[0090] Generate the watermarked image

[0091] After the watermark embedding is completed, the embedded watermark low-frequency subband LL′ will be recombined with other high-frequency subbands LH, HL, and HH. The combined image is then reconstructed through the inverse discrete wavelet transform (IDWT) to generate the final watermarked carrier image. The specific combination and reconstruction process are as follows:

[0092] I′ = IDWT(LL′, LH, HL, HH)

[0093] where I′ is the final watermarked image. In this process, the watermark information embedded in the low-frequency subband LL′ is not easily perceptible, and at the same time, the structural features of the image are well preserved.

[0094] Optimization and adjustment

[0095] To further improve the robustness and invisibility of the watermark, the watermark embedding module can also combine with the adaptive optimization of the image content. Specifically, the embedding module can pre-analyze the image through edge detection algorithms (such as the Canny operator or the Sobel operator), select the flat areas of the image for watermark embedding, and avoid affecting the important structural parts of the image. In addition, by adopting local embedding intensity adjustment, the watermark intensity can be dynamically adjusted according to the complexity and importance of the image content, making the distribution of the watermark in the image more uniform and further improving the invisibility.

[0096] In this embodiment, the watermark embedding module performs singular value decomposition on the low-frequency subband of the image and embeds the watermark information. By combining the optimized embedding intensity and the flexible application of multiple subbands, the efficient embedding of the watermark information in the image is achieved. Through the inverse SVD reconstruction and inverse DWT reconstruction processes, the watermarked image is generated, ensuring the image quality and the recoverability of the watermark. At the same time, the watermark embedding module also further improves the stability and robustness of the watermark through adaptive optimization techniques and local adjustments, ensuring that the watermark information can be reliably extracted under various attacks such as compression, noise, and cropping.

[0097] A watermark optimization and enhancement module, which is used to optimize the watermark embedding parameters. By dynamically adjusting the embedding intensity, the watermark can improve the information capacity and robustness while maintaining invisibility after embedding, and reduce the similarity between watermarks;

[0098] In this embodiment, the watermark optimization and enhancement module uses techniques such as the variational method and optimal control theory to adaptively adjust the parameters in the watermark embedding process. Specifically, this module optimizes the embedding strength and position, so as to maximize the amount of watermark information while minimizing the impact of the watermark on the image quality. Through the optimized embedding scheme, the watermark can stably exist in the image, and even when facing common attacks such as compression, noise, and cropping, the watermark information can be reliably extracted.

[0099] Watermark optimization process

[0100] In this embodiment, the core goal of watermark optimization is to balance the invisibility and robustness of the watermark. Generally, the embedding strength and position of the watermark are affected by the image content. When the watermark strength is too large, it will affect the image quality; while when the strength is too small, the watermark may not be effectively extracted. Therefore, the watermark optimization module optimizes the embedding strength of the watermark through the variational method, so as to ensure that while maximizing the stored information in the image, the watermark maintains good invisibility.

[0101] In some embodiments, after the watermark embedding module performs the embedding operation on the singular value matrix S, the watermark optimization and enhancement module will adjust the strength of the watermark. The optimized embedding process is as follows:

[0102]

[0103] Where:

[0104] S′ is the optimized singular value matrix, which contains the watermark information;

[0105] W i ′ is the i-th watermark information after preprocessing;

[0106] α i is the strength factor of watermark embedding, which needs to be dynamically adjusted through an optimization algorithm.

[0107] The purpose of the variational method is to find the optimal α i value, so as to maximize the capacity of the watermark information while ensuring the invisibility of the watermark. Specifically, the goal of variational method optimization is to minimize the similarity between the watermark and the image during the watermark embedding process. This process reduces the impact of the watermark information on the image quality by optimizing the adaptability of the watermark strength to the image content.

[0108] In some embodiments, the watermark optimization module can also adjust the recovery parameters during the embedding process through optimal control theory to further improve the extraction accuracy of the watermark. For example, in the watermark extraction stage, optimal control theory can help adjust the recovery parameters to ensure the minimization of errors during the extraction process. By dynamically adjusting the control parameters, the system can adapt to different image contents and attack environments, thereby enhancing the robustness of the watermark.

[0109] Enhancing the Robustness of the Watermark

[0110] In this embodiment, the watermark optimization and enhancement module not only optimizes the embedding strength of the watermark, but also further improves the robustness of the watermark under various attack conditions. Specifically, the watermark optimization and enhancement module enhances the image after watermark embedding to ensure that the watermark can still be stably recovered when the image undergoes common attacks such as compression, noise, cropping, and rotation.

[0111] In certain embodiments, the enhancement strategies include the following methods:

[0112] (1) Adaptive adjustment of the embedding strength

[0113] During the watermark embedding process, the watermark optimization and enhancement module adaptively adjusts the embedding strength to ensure that the watermark has an appropriate strength in different regions of the image. In the flat regions of the image, the watermark strength is relatively large to increase the watermark capacity; in the detailed regions of the image, the watermark strength is relatively small to reduce the impact on the image structure.

[0114] Specifically, the embedding strength α i can be adaptively adjusted according to the gradient and texture information of the image in different regions of the image. For low-gradient regions (regions with small changes in the horizontal or vertical directions), the watermark can be embedded more strongly; while in high-gradient regions, the watermark strength is correspondingly reduced to avoid affecting the details of the image.

[0115] (2) Anti-noise and compression enhancement

[0116] To improve the robustness of the watermark against noise, the watermark optimization and enhancement module adopts denoising processing, such as Wiener filtering or median filtering. These filtering algorithms can effectively remove noise and improve the extraction stability of the watermark information. In a compression environment, the watermark optimization module also adjusts the embedding method of the watermark to make it more adaptable to the changes of the compression algorithm, ensuring that the watermark can still be effectively recovered after JPEG compression.

[0117] (3) Embedding redundant watermark information

[0118] To improve the robustness of the system, the watermark optimization and enhancement module can also adopt redundant watermark information and embed the same watermark information in multiple regions of the image. By storing the watermark information redundantly, the system can still recover the watermark information through other regions when some regions of the image are attacked or lost.

[0119] (4) Anti-rotation and cropping enhancement

[0120] The watermark optimization and enhancement module can also adopt technologies such as the Fourier-Mellin transform (FMT) to enhance the robustness of the watermark against rotation and cropping attacks. The Fourier-Mellin transform can make the watermark information invariant to the rotation of the image, ensuring that the watermark information can still be accurately extracted even when the image rotates.

[0121] In some embodiments, the redundant watermark information can be distributed to multiple regions of the image through a spatial embedding strategy. In this way, even when the image is cropped or rotated, the system can still recover the watermark information from other regions, improving the robustness of the watermark.

[0122] The watermark optimization and enhancement module ensures the maximization of the embedding effect of the watermark in the image through a series of optimization algorithms and enhancement strategies. By dynamically adjusting the watermark strength, enhancing resistance to noise and compression, introducing redundant information, and enhancing resistance to rotation and cropping, the watermark optimization and enhancement module can significantly improve the robustness and invisibility of the watermark. During the image processing process, the module optimizes various parameters of watermark embedding to ensure the stable existence of the watermark information in the face of various attacks and the accurate extraction of the watermark, ultimately improving the performance and reliability of the entire watermark system.

[0123] The watermark extraction and recovery module is used to recover the watermark information from the image embedded with the watermark based on the inverse discrete wavelet transform and the inverse singular value decomposition, and optimize the extraction parameters based on the optimal control theory to make it adaptively optimize the extraction accuracy under different attack conditions and improve the stability of watermark recovery;

[0124] In this embodiment, the main operation steps of the watermark extraction and recovery module include the inverse discrete wavelet transform (IDWT), the inverse singular value decomposition (SVD), and the extraction optimization based on the optimal control theory. Through these technical means, the module can extract and recover the watermark information from the watermarked image, ensuring the stability and reliability of the watermark.

[0125] Watermark extraction process

[0126] In this embodiment, the watermark extraction process first extracts the low-frequency sub-band from the watermarked image and performs discrete wavelet transform (DWT) on it to decompose the image into multiple frequency sub-bands. Specifically, by performing DWT on the watermarked image, the low-frequency sub-band LL′ is extracted, and then singular value decomposition (SVD) is performed to obtain the reconstructed singular value matrix S″. The mathematical representation of this process is as follows:

[0127] LL′ = U′S″V′ T

[0128] Where:

[0129] LL′ is the low-frequency sub-band of the watermarked image;

[0130] U′ and V′ are the corresponding singular matrices;

[0131] S″ is the singular value matrix recalculated after embedding the watermark.

[0132] In some embodiments, the singular value matrix S″ may have been affected by the watermark information, so it needs to be compared with the singular value matrix S in the original image to extract the embedded watermark information. This process can be carried out through the following formula:

[0133]

[0134] Where:

[0135] W i ′ is the i-th extracted watermark information;

[0136] S″ and S are the singular value matrices of the watermarked image and the original image respectively;

[0137] α i is the watermark strength coefficient, which needs to be restored according to the parameters adjusted during watermark embedding.

[0138] Watermark recovery

[0139] After watermark extraction, the recovery process will restore the extracted watermark information to the original watermark image through inverse orthogonal transformation (such as inverse DCT or inverse DFT). Specifically, the watermark information W i ′ will be restored through the following formula:

[0140] W i = T -1 W i ′

[0141] Where:

[0142] T -1 is the inverse transformation matrix of the orthogonal transformation;

[0143] W i ′ is the extracted watermark information;

[0144] W i is the restored watermark image.

[0145] In some embodiments, the error in the restoration process may affect the quality of the watermark. Therefore, the present invention also introduces the optimal control theory to dynamically adjust the parameters in the restoration process. Through the optimal control theory, the accuracy of watermark restoration can be optimized by minimizing the extraction error. In this process, the error E of watermark restoration can be expressed as:

[0146] E = ||W i -W i ′|| F

[0147] Where:

[0148] E is the restoration error;

[0149] ||·|| F represents the Frobenius norm, which is used to measure the difference between the restored watermark and the original watermark;

[0150] W i is the original watermark image, and W i ′ is the extracted watermark image.

[0151] Through the optimal control algorithm, the various parameters involved in the restoration process can be dynamically adjusted to minimize the restoration error and ensure the accurate restoration and stability of the watermark.

[0152] Enhancing the robustness of watermark extraction

[0153] To improve the robustness of the watermark extraction process, some optimization strategies are also introduced in this embodiment. By combining noise filtering and image enhancement techniques, the influence of noise interference on watermark extraction can be effectively reduced. Generally, the image may be affected by noise during watermark extraction, especially during JPEG compression or other image compression processes, and the noise will seriously affect the restoration of watermark information. For this reason, the present invention uses an adaptive filter (such as a Wiener filter) to denoise the image, thereby improving the extraction accuracy of watermark information.

[0154] As an option, the watermark extraction module can also adopt a redundant coding strategy to embed watermark information in multiple regions to increase the restoration stability of the watermark. For example, when some regions of the image are cropped or damaged, through redundant coding, even if some watermark information is lost, the complete watermark information can still be restored from other regions.

[0155] In addition, in terms of anti-rotation, the watermark extraction module enhances the robustness of the watermark against rotation attacks through the Fourier-Mellin transform. Specifically, the Fourier-Mellin transform of the watermark image enables the watermark information to remain invariant to the rotation of the image. Even if the image is rotated, the watermark information can still be accurately restored.

[0156] The watermark extraction and recovery module ensures the accurate extraction of the watermark information in the watermarked image through a series of technical means. Through the combination of the inverse discrete wavelet transform (IDWT), the inverse singular value decomposition (SVD), and the optimal control theory, the watermark information can be efficiently recovered. At the same time, the adoption of adaptive filtering and redundant watermark embedding strategies improves the robustness of the system, enabling the watermark to be stably and accurately restored after experiencing attacks such as compression, noise, cropping, and rotation. Finally, the watermark extraction and recovery module can effectively ensure the accurate recovery of the watermark and improve the reliability of the system in various environments.

[0157] The robustness enhancement module is used to improve the extraction success rate of the watermark under attacks such as JPEG compression, cropping, noise, and rotation, enhance the anti-cropping ability of the watermark through a spatially distributed embedding strategy, enhance the anti-rotation ability of the watermark through the Fourier-Mellin transform, and adopt an adaptive filtering strategy to enhance the anti-noise ability of the watermark;

[0158] In this embodiment, the robustness enhancement module enhances the stability of the watermark through various technical means, such as redundant embedding, adaptive filtering, anti-compression enhancement, anti-rotation and anti-cropping enhancement. By combining these technologies, the robustness enhancement module can ensure that the watermark remains effective and recoverable under various attacks on the image.

[0159] Robustness Enhancement Process

[0160] In this embodiment, the robustness enhancement module ensures the stability of the watermark through various technical means. Specifically, the functions of this module include the following aspects:

[0161] (1) Anti-JPEG Compression Enhancement

[0162] Generally, JPEG compression is one of the main factors affecting the stability of the watermark. Especially at high compression ratios, the watermark information often gets lost or becomes difficult to recover. To improve the ability of the watermark to be recovered after compression, the robustness enhancement module first designs an adaptive quantization strategy for JPEG compression. When embedding the watermark, according to the characteristics of the compressed image, the embedding position and strength of the watermark information are optimized to ensure that the watermark information can be effectively retained.

[0163] Specifically, when embedding the watermark information, the quantization step is used to adjust the strength of the watermark. At this time, the low-frequency part of the image is usually less affected by compression, so the watermark is more embedded in these areas. For higher-frequency areas, the watermark strength is appropriately weakened to avoid excessive information loss. The quantized watermark information is adjusted by the following formula:

[0164] W′ i,j = Quantize(W i,j ,α)

[0165] Where:

[0166] W′ i,j is the quantized watermark information;

[0167] W i,j is the watermark information before embedding;

[0168] α is the quantization step size, which controls the watermark strength.

[0169] Through this process, the watermark can still maintain a certain robustness after JPEG compression.

[0170] (2) Anti-noise enhancement

[0171] In practical applications, the watermark extraction process is usually affected by noise interference. Especially after the image is transmitted or stored, the noise often affects the recovery accuracy of the watermark. To effectively reduce the impact of noise on watermark extraction, the robustness enhancement module adopts adaptive filtering techniques such as Wiener filtering and median filtering.

[0172] Wiener filtering is a commonly used denoising technique. It removes the noise in the image by calculating the local mean and variance of the image. In the watermark extraction process, Wiener filtering can effectively remove the high-frequency noise in the image, thus ensuring the stable extraction of the watermark information under noise interference. Its mathematical expression is as follows:

[0173] W ′ = WienerFilter(W,σ)

[0174] Where:

[0175] W ′ is the watermark information after denoising;

[0176] W is the original watermark information;

[0177] σ is the standard deviation of the noise, which determines the strength of the filtering.

[0178] Median filtering can remove the salt-and-pepper noise in the image. In the process of watermark information recovery, combining median filtering can further enhance the robustness.

[0179] (3) Anti-Cropping and Rotation Enhancement

[0180] In some attack scenarios, images may be subject to cropping and rotation operations, which can lead to the loss of watermark information or incorrect recovery. To address this issue, the robustness enhancement module employs redundant embedding technology and the Fourier-Mellin transform (FMT).

[0181] Specifically, during the watermark embedding process, the anti-cropping ability of the watermark is enhanced through redundant embedding, that is, embedding the same watermark information in multiple regions simultaneously. For example, the same watermark information is stored in multiple regions of the image, so even if the image is cropped, the watermark in the remaining part can still be fully recovered. The way of redundant distribution of watermark information can be expressed by the following formula:

[0182]

[0183] Where:

[0184] W′ i,j is the watermark information after redundant embedding;

[0185] α k is the weight of the redundant information;

[0186] W k is the watermark information in the k-th region.

[0187] The Fourier-Mellin transform (FMT) is a rotation-invariant transform that can effectively handle the changes in watermark information caused by image rotation. Through the Fourier-Mellin transform, the watermark information of the rotated image can be mapped to an invariant space, thus ensuring that the watermark information can still be recovered for the rotated image. Specifically, the Fourier-Mellin transform can be calculated by the following formula:

[0188]

[0189] Where:

[0190] is the result after the Fourier-Mellin transform of the watermark information;

[0191] W(x,y) is the pixel value of the watermark image;

[0192] x -s is the rotation-invariant filter.

[0193] Through this method, even if the image is rotated, the watermark information can still be accurately recovered.

[0194] In this embodiment, the robustness enhancement module enhances the robustness of the watermark through various means, ensuring that the watermark information can still be stably extracted when facing common attacks such as JPEG compression, noise, cropping, and rotation. Through techniques such as adaptive quantization, adaptive filtering, redundant embedding, and Fourier-Mellin transform, the robustness enhancement module significantly improves the system's resistance to various interferences. In practical applications, this module can significantly enhance the extraction accuracy and stability of the watermark while ensuring the image quality, thereby improving the security and reliability of the entire watermark system.

[0195] Although the embodiments of the present invention have been shown and described, those of ordinary skill in the art can understand that various changes, modifications, substitutions, and variations can be made to these embodiments without departing from the principles and spirit of the present invention. The scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. An adaptive image watermark embedding and restoration system, characterized in that: include: The image preprocessing module is used to perform discrete wavelet transform on the carrier image to be embedded with watermark, decompose the image into multiple frequency sub-bands, and select the low-frequency sub-band as the watermark embedding area; The watermark preprocessing module is used to perform orthogonal transformation on the watermark image to maintain orthogonality between multiple watermarks and reduce mutual interference between watermarks; The watermark embedding module is used to embed the watermark into the low-frequency subband of the carrier image based on the output of the watermark preprocessing module, and adjust the embedding strength through singular value decomposition to match the watermark information with the characteristics of the carrier image, thereby improving robustness and concealment; The watermark optimization and enhancement module is used to optimize the watermark embedding parameters. By dynamically adjusting the embedding strength, the watermark embedding can maintain its concealment while improving the information capacity and robustness and reducing the similarity between watermarks. The watermark extraction and recovery module is used to recover the watermark information from the watermarked image based on inverse discrete wavelet transform and inverse singular value decomposition, and optimize the extraction parameters based on optimal control theory, so that it can adaptively optimize the extraction accuracy under different attack conditions and improve the stability of watermark recovery; The robustness enhancement module is used to improve the success rate of watermark extraction under JPEG compression, cropping, noise, and rotation attacks, and enhance the watermark's anti-cropping ability through a spatially distributed embedding strategy, enhance the watermark's anti-rotation ability through Fourier-Mellin transform, and enhance the watermark's anti-noise ability using an adaptive filtering strategy.

2. The adaptive image watermark embedding and restoration system according to claim 1, characterized in that: The image preprocessing module performs discrete wavelet transform on the carrier image to generate low-frequency sub-bands and high-frequency sub-bands, and selects the low-frequency sub-bands as the watermark embedding area to improve the robustness of the watermark.

3. The adaptive image watermark embedding and restoration system according to claim 1, characterized in that: The watermark preprocessing module transforms the watermark image using orthogonal transformation to maintain orthogonality between multiple watermarks and reduce mutual interference between watermarks, wherein the orthogonal transformation includes discrete cosine transformation or discrete Fourier transformation.

4. The adaptive image watermark embedding and restoration system according to claim 1, characterized in that: The watermark embedding module decomposes the low-frequency subband of the carrier image through singular value decomposition, and embeds the watermark information into the decomposed singular value matrix. The embedded singular value matrix is ​​used to reconstruct the low-frequency subband and reconstruct the watermarked image in combination with inverse discrete wavelet transform.

5. The adaptive image watermark embedding and restoration system according to claim 1, characterized in that: The watermark optimization and enhancement module optimizes the watermark embedding strength based on the variational method and dynamically adjusts the embedding parameters through the optimization algorithm, so that after the watermark is embedded, the information capacity and robustness are improved while ensuring the concealment.

6. The adaptive image watermark embedding and restoration system according to claim 1, characterized in that: The watermark extraction and restoration module performs discrete wavelet transform on the watermarked image, extracts the low-frequency subband, calculates the embedded watermark information based on singular value decomposition, and then restores the original watermark using inverse orthogonal transform.

7. The adaptive image watermark embedding and restoration system according to claim 1, characterized in that: The watermark extraction and recovery module adopts optimal control theory to dynamically adjust watermark recovery parameters, so that it can adaptively optimize extraction accuracy under different attack conditions and improve the stability of watermark recovery.

8. The adaptive image watermark embedding and restoration system according to claim 1, characterized in that: The robustness enhancement module adopts a quantization adaptive strategy to adjust the quantization step size of watermark embedding to target JPEG compression attacks, so that the watermark information can still be restored after JPEG compression.

9. The adaptive image watermark embedding and restoration system according to claim 1, characterized in that: The robustness enhancement module stores watermark information in multiple areas through a spatially distributed embedding strategy to enhance the watermark's resistance to cropping attacks.

10. The adaptive image watermark embedding and restoration system according to claim 1, characterized in that: The robustness enhancement module uses Fourier-Mellin transform to transform the embedded watermark information to enhance the robustness of the watermark against rotation attacks. The robustness enhancement module uses adaptive filtering strategies, including Wiener filtering and median filtering, to reduce the impact of noise on watermark extraction, and combines redundant coding to improve the stability of watermark extraction.

Images