A dynamic redundancy robust digital blind watermarking method and system

Abstract

The application discloses a dynamic redundancy robust digital blind watermarking method and system, and relates to the technical field of image data processing. The method embeds watermark information into pixel data (robust channel) and metadata (fast channel) of a carrier image simultaneously through a double embedding strategy. When the watermark is extracted from the carrier image subjected to geometric attacks, the system firstly attempts to extract from the metadata; if the fast extraction fails, the system automatically switches to the pixel domain for robust extraction. The method comprises the following steps: optimizing and packaging the data to be embedded; embedding the packaged data stream into the pixel and writing the compressed watermark data into the metadata; and adopting the aforementioned double-channel extraction strategy for extraction. The design effectively avoids the defects of simply relying on spatial synchronization, and significantly improves the extraction performance and reliability of the watermark system under geometric attacks.

Description

Technical Field

[0001] This invention relates to the field of image data processing technology, and in particular to a dynamic redundancy robust digital blind watermarking method and system. Background Technology

[0002] Digital watermarking technology is essentially an information hiding technology. It embeds specific identifying information, such as copyright owner and serial number, into digital media like images, audio, and video in an invisible or imperceptible manner, thus providing a crucial security barrier for these digital assets. The core application value of this technology lies in three main aspects: first, copyright protection, which uses watermark verification to clearly declare ownership and prevent unauthorized copying and distribution of digital products; second, content authentication, which checks the integrity of the watermark to determine whether media content has been tampered with or forged; and third, source tracing, which identifies the embedded source information to trace the distribution path and leakage source of content. Among various watermarking technologies, blind watermarking is highly favored due to its unique advantages—it does not require the original, unwatermarked carrier image as a reference during the watermark extraction and detection stages. This makes it highly practical and feasible in real-world applications such as online copyright monitoring and large-scale content auditing, significantly reducing deployment complexity and cost.

[0003] To achieve the aforementioned copyright protection and content authentication functions in practical applications, watermarking algorithms must strike a balance between invisibility and robustness. This means ensuring that embedding the watermark does not impair the visual quality of the image while simultaneously protecting it from common signal processing attacks (such as compression, filtering, noise, and cropping). Transform domain-based algorithms have emerged as a mainstream approach to address this challenge. Their core idea is to transform the image from the spatial domain to the frequency domain (e.g., wavelet domain, cosine domain), embedding the watermark by modifying the frequency coefficients. This is because the human eye is far less sensitive to changes in the frequency domain than to changes in pixel values ​​in the spatial domain. Among numerous transform domain schemes, a hybrid transform domain method integrating Discrete Wavelet Transform (DWT), Discrete Cosine Transform (DCT), and Singular Value Decomposition (SVD) is widely considered by academia and industry to be a high-performance existing technology. This scheme typically begins by using DWT to decompose the image into multiple resolutions to capture its features across different frequency bands and directions. Then, DCT transform is applied to selected subbands, such as low-frequency approximation components, to further concentrate energy. Finally, SVD is performed on the DCT coefficient blocks, embedding the watermark by modifying their stable singular values. This cascaded strategy fully leverages the advantages of each of the three mathematical tools: the multi-resolution characteristics of DWT, the energy compression properties of DCT, and the inherent stability of SVD, resulting in the scheme's recognized superior performance in resisting various attacks and maintaining visual transparency.

[0004] For example, the invention patent announcement CN108416727B discloses a fully blind robust digital watermarking method for color images based on self-embedding technology, which includes: performing a discrete Fourier transform on each sub-block of the approximation sub-image of the luminance component of the color image to obtain the amplitude spectrum matrix of the discrete Fourier transform coefficient matrix of each sub-block; creating a feature watermark and a self-embedded feature watermark based on the amplitude spectrum matrix of all sub-blocks; obtaining the amplitude spectrum matrix of the discrete Fourier transform coefficient matrix of each sub-block of the approximation sub-image of the luminance component of the watermarked color image in the same manner; blindly extracting the feature watermark and the authentication watermark based on the amplitude spectrum matrix of all sub-blocks; and protecting copyright based on the normalized correlation coefficient between the blindly extracted feature watermark and the authentication watermark.

[0005] For example, the invention patent with publication number CN116993570A discloses a robust and reversible blind watermarking method for color digital images, which includes: embedding the watermark by quantizing the DC coefficient of the discrete Fourier transform to make the watermark robust, and then significantly improving the robustness by repeated embedding and voting mechanism. In addition, the distortion caused during the robust watermark embedding process is re-embedded into the image using a pixel value sorting method.

[0006] However, in the process of implementing the inventive technical solution in the embodiments of this application, it was found that the above-mentioned technology has at least the following technical problems:

[0007] In existing technologies, geometric attacks severely impact the robustness of watermark extraction by disrupting the spatial synchronization between watermark embedding and extraction. Unlike noise attacks that directly interfere with frequency domain information, geometric attacks do not directly delete watermark data. Instead, they disrupt the spatial structure of the image (e.g., rotation, scaling, translation, and cropping), causing the watermark extractor to fail to accurately map back to the transform domain units used during embedding (e.g., DWT subbands or DCT blocks). This results in information being obtained from incorrect image regions, leading to watermark decoding failure and poor watermark extraction performance in the carrier image after watermark embedding. Summary of the Invention

[0008] This application provides a dynamic redundant robust digital blind watermarking method and system, solving the problem of performance degradation caused by geometric attacks in existing watermarking technologies. Existing methods fail when faced with geometric attacks such as rotation, scaling, translation, and cropping, as spatial synchronization is disrupted. This invention, through a dual embedding and dual-channel extraction mechanism, retains the advantages of robust pixel-domain embedding while introducing a fast metadata channel, effectively avoiding the shortcomings of relying solely on spatial synchronization. This significantly improves the extraction performance and reliability of the watermarking system under geometric attacks.

[0009] This application provides a dynamic redundancy robust digital blind watermarking method, comprising the following steps: dynamically optimizing the watermark data to be embedded, and encapsulating the dynamically optimized watermark data to be embedded to obtain a complete encapsulated data stream. The dynamic optimization process involves data compression and error correction coding of the watermark data to be embedded to improve the embedding efficiency and resistance to geometric attacks. The encapsulated complete data stream is embedded into the carrier pixel data, and the compressed watermark data to be embedded is simultaneously embedded into the carrier image to obtain a carrier image after double embedding. The carrier image subjected to geometric attacks after double embedding is subjected to watermark extraction, which includes fast channel extraction and robust channel extraction.

[0010] This application provides a dynamic redundancy robust digital blind watermarking system, comprising: an encapsulation processing module, a watermark embedding module, and a watermark extraction module. The encapsulation processing module dynamically optimizes the watermark data to be embedded and encapsulates the dynamically optimized watermark data to obtain a complete encapsulated data stream. The watermark embedding module embeds the complete encapsulated data stream into the carrier pixel data and simultaneously embeds the compressed watermark data into the carrier image, obtaining a carrier image after double embedding. The watermark extraction module extracts the watermark from the carrier image subjected to geometric attacks after double embedding, including fast channel extraction and robust channel extraction.

[0011] One or more technical solutions provided in the embodiments of this application have at least the following technical effects or advantages:

[0012] 1. This invention utilizes dynamic encapsulation processing based on carrier size and embedding strength to achieve adaptive matching between watermark data and carrier characteristics, thus optimizing embedding efficiency. It employs a dual embedding strategy of pixel domain (robust channel) and metadata domain (fast channel), preserving robust watermarking while adding an independent fast extraction channel. During extraction, the system uses a fixed dual-channel extraction logic: it prioritizes the fast channel, automatically switching to the robust channel upon failure. This design balances high extraction speed and excellent anti-attack capabilities while ensuring invisibility.

[0013] 2. This invention employs an inverse transformation and multi-channel redundancy extraction strategy. By statistically averaging and intelligent clustering decisions on multiple copies cyclically embedded in the YUV three channels, it effectively offsets synchronization errors and local distortions caused by geometric attacks, significantly improving the accuracy of the initial bit sequence. Simultaneously, it innovatively adopts a self-contained data format of "reading length from the tail and extracting data from the head," enabling the decoder to accurately segment valid data and padding bits without any prior information, achieving true blind extraction. Through cascaded processing of error correction decoding and data decompression, a complete recovery from error correction to data reconstruction is constructed. This ensures that even after suffering a strong geometric attack, the watermark system can reliably recover the original watermark information through a robust channel, guaranteeing the system's availability and reliability under extreme conditions.

[0014] 3. This invention employs a fixed dual-channel extraction strategy. First, it attempts to extract data quickly from metadata, achieving millisecond-level extraction if the metadata remains intact. If quick extraction fails, the system seamlessly switches to a robust channel, recovering the watermark by processing pixel data. This strategy combines the efficiency of the fast channel with the reliability of the robust channel, ensuring a high extraction success rate under various attack scenarios (whether metadata is stripped or pixel content is tampered with), thereby improving the overall usability of the system. Attached Figure Description

[0015] Figure 1 Flowchart of the dynamic redundancy robust digital blind watermarking method provided in the embodiments of this application;

[0016] Figure 2 A flowchart of watermark embedding provided for embodiments of this application;

[0017] Figure 3 A flowchart for watermark extraction provided in this application embodiment;

[0018] Figure 4 This is a structural diagram of a dynamic redundancy robust digital blind watermarking system provided in an embodiment of this application. Detailed Implementation

[0019] This application provides a dynamic redundant robust digital blind watermarking method and system, solving the problem of performance degradation caused by geometric attacks in existing watermarking technologies. Existing methods fail when faced with geometric attacks such as rotation, scaling, translation, and cropping, as spatial synchronization is disrupted. While these attacks do not directly delete watermark data, they disrupt the spatial structure of the image, preventing the extractor from accurately mapping back to the transform domain unit used during embedding, leading to information extraction from incorrect image regions and ultimately watermark decoding failure. This invention, through a dual embedding and dual-channel extraction mechanism, retains the advantages of pixel-domain robust embedding while introducing a fast metadata channel, effectively avoiding the shortcomings of relying solely on spatial synchronization. This significantly improves the extraction performance and reliability of the watermarking system under geometric attacks. The overall approach is as follows:

[0020] First, the data to be embedded with the watermark is dynamically optimized and encapsulated based on the carrier image size and watermark embedding strength parameters to generate a complete encapsulated data stream. Then, this data stream is embedded into the pixel data of the carrier image, while the compressed watermark data is written to the carrier metadata area, achieving dual embedding. At the extraction end, the system employs a fixed dual-channel extraction strategy: it prioritizes rapid extraction from metadata; if this fails (e.g., due to metadata loss), it automatically switches to robust extraction from the pixel domain. This method effectively solves the problem of decreased extraction performance caused by synchronization mechanism failure under geometric attacks in existing watermarking technologies, significantly improving the practicality and reliability of the watermarking system.

[0021] To better understand the above technical solutions, the following will provide a detailed explanation of the technical solutions in conjunction with the accompanying drawings and specific implementation methods.

[0022] like Figure 1The diagram shows a flowchart of a dynamic redundancy robust digital blind watermarking method provided in this application embodiment. The method includes the following steps: performing dynamic optimization processing on the watermark data to be embedded, and encapsulating the dynamically optimized watermark data to be embedded to obtain a complete encapsulated data stream. Dynamic optimization processing means performing data compression and error correction coding on the watermark data to be embedded to improve the embedding efficiency and resistance to geometric attacks. The encapsulated complete data stream is embedded into the carrier pixel data, and the compressed watermark data to be embedded is embedded into the carrier image to obtain a carrier image after double embedding processing. The compressed watermark data to be embedded does not contain RS code (error correction coding) and length encapsulation, which can maximize space saving. Watermark extraction is performed on the carrier image subjected to geometric attacks after double embedding processing. Watermark extraction includes fast channel extraction and robust channel extraction.

[0023] In this embodiment, the watermark data to be embedded is a segment of UTF-8 encoded text or binary data. To construct a highly robust watermarking system, this invention first inputs the original watermark data to be embedded into the zlib (Zip Library) compression module. This open-source library implements the DEFLATE algorithm, which can significantly reduce the number of bits in the watermark data through efficient lossless compression. On the one hand, this allows the limited embedding capacity to carry more original information; on the other hand, the space saved by compression provides key resources for subsequent enhancement of error correction redundancy. Then, the binary data stream output by the compressed zlib is fed into the Reed-Solomon (RS) encoding module to generate the RS-encoded data stream. As a powerful non-binary block forward error correction code, RS code enables the decoder to actively detect and correct symbol errors by adding a check symbol. This feature makes it particularly effective against burst errors caused by JPEG compression or noise.

[0024] like Figure 2The diagram shows a watermark embedding flowchart provided in an embodiment of this application. Specifically, dynamic capacity calculation and data stream encapsulation are achieved by dynamically optimizing the watermark data to be embedded and encapsulating the dynamically optimized watermark data to obtain a complete encapsulated data stream: the product of the carrier image width and the carrier image height is recorded as the total number of pixels in the carrier image; a security factor is obtained based on the watermark embedding strength; and the maximum number of bits that the image can safely carry is obtained based on the security factor and the total number of pixels in the carrier image. To satisfy the invisibility and stability of the watermark, the amount of watermark data finally embedded in the image must be exactly the maximum number of bits that the image can safely carry. The formula for calculating the maximum number of bits that the image can safely carry is: safe_bit_num = (pixel_num × 0.0). The formula is: (025+190)×C; where pixel_num is the total number of pixels in the carrier image, and C is the security factor. This calculation formula dynamically correlates the embedding capacity with the image size, completely solving the problem of resource waste or insufficient capacity caused by fixed capacity in existing technologies. The bit length of the dynamically optimized watermark data to be embedded is calculated and converted into a binary sequence. The padding length is obtained based on the maximum number of bits that the image can safely carry, the binary sequence length, and the bit length of the dynamically optimized watermark data to be embedded. The padding bits and the binary sequence are concatenated to the dynamically optimized watermark data to be embedded, resulting in a complete encapsulated data stream. The padding bits consist of a sequence of meaningless bits, such as all zeros, equal to the padding length. For example, if the binary sequence is a 32-bit unsigned integer, to achieve complete blind extraction of the watermark, the system adopts a self-contained encapsulation scheme that uses length information as tail metadata. This scheme encapsulates the dynamically optimized watermark data, i.e., the bit stream after RS ​​encoding, along with the padding bits equal to the padding length and a fixed 32-bit unsigned integer length sequence. This 32-bit sequence precisely records the bit length of the preceding RS-encoded data and is placed at the end of the entire data stream. This tail structure, which embeds key metadata into the data stream itself, ensures that the decoding end can locate and parse the length value from within the carrier without any prior information, thereby accurately defining the boundary of valid data. This is the core foundation for realizing the fully blind watermark extraction function.

[0025] In this embodiment, the security factor C is a fixed empirical value (such as 0.9) used to balance the theoretical maximum embedding capacity and robustness. The theoretical maximum embedding capacity is reduced to ensure the invisibility and robustness of the watermark.

[0026] Specifically, the double-redundant watermark embedding and the generation of the final watermarked image are achieved by embedding the encapsulated complete data stream into the carrier pixel data, and then embedding the compressed watermark data into the carrier image. The step of embedding the encapsulated complete data stream into the carrier pixel data includes: extracting the low-frequency sub-band of the carrier image and dividing it into blocks; performing matrix eigenvalue decomposition on each block to obtain the block coefficient matrix, the matrix eigenvalue decomposition including orthogonal transformation and singular value decomposition; modifying the two primary singular values ​​of each block coefficient matrix according to the watermark bit sequence carried by the encapsulated complete data stream, the primary singular values ​​being the two first values ​​in the singular value sequence sorted from largest to smallest; and performing an inverse transformation reconstruction on the modified block coefficient matrices.

[0027] Embedding the compressed watermark data into the carrier image is achieved by writing the compressed watermark data into the metadata area of ​​the carrier image. Specifically, this includes: using Base64 encoding to text-encode the compressed watermark data; if the carrier image is in PNG format, writing the text-encoded watermark data into a pre-created tEXt type standard data block in the carrier image; if the carrier image is in JPEG format, writing the compressed watermark data into the UserComment field of its EXIF ​​information. EXIF ​​is a type of metadata embedded in JPEG image files.

[0028] In this embodiment, the present invention first performs one-level or multi-level DWT decomposition on the YUV or grayscale color space of the carrier pixel data to obtain the low-frequency subband LL. The LL subband is then divided into blocks, and DCT transformation is performed on each block. SVD is performed on the coefficient matrix of each DCT block. Each bit in the data stream is embedded by quantizing and modifying one or more of its largest singular values. Inverse SVD, inverse DCT, and inverse DWT are performed sequentially to reconstruct the image and generate an image version containing pixel-level watermarks. The present invention constructs a redundant watermark storage mechanism independent of the image pixel data. After compression and encoding, the complete watermark information is directly written into the metadata area of ​​the image file, such as the EXIF ​​user comment field or custom APPn segment in JPEG format, or the tEXt block in PNG format, creating an independent "metadata backup channel" parallel to the main watermark channel. The storage location and parsing logic of the backup data are independent of the image's pixel content. Therefore, any attack targeting the image's visual content, such as filtering, noise addition, color adjustment, or even severe geometric cropping and scaling, will remain intact and readable as long as the file container structure is not damaged. During watermark extraction, the watermark extractor does not need to execute complex image processing algorithms; it can directly locate and read the metadata segment in the file header, achieving millisecond-level fast watermark extraction and providing users with a near-instantaneous verification experience. This mechanism, together with the watermark body embedded in the pixels, constitutes a dual protection strategy. The metadata backup channel ensures a reliable recovery path even if the watermark body cannot be extracted due to image content corruption; conversely, if the file metadata is maliciously stripped, the watermark body deeply embedded in the pixels still serves as a final safeguard. The two act as backups for each other, greatly enhancing the robustness and reliability of the entire watermarking system.

[0029] like Figure 3 The diagram shows a watermark extraction flowchart provided in an embodiment of this application. Specifically, watermark extraction follows a fixed dual-channel strategy: First, the watermark is extracted via a fast channel, which attempts to directly read and decode the watermark information from the image's metadata. If fast channel extraction fails (e.g., the metadata is erased or corrupted), the system extracts the watermark via a robust channel based on the number of embedded information bits. Furthermore, before enabling the robust channel, if the image suffers a geometric attack, a geometric attack recovery attempt will be performed first.

[0030] The number of embedded information bits is obtained based on the maximum number of bits that the image can securely carry, the number of bits in the carrier image subjected to geometric attacks after double embedding processing, and a preset capacity utilization factor. The number of embedded information bits is the length of the payload. Specifically, the number of embedded information bits Nfinal is obtained as follows: Nfinal = Min(SizeEncoded, num * R); where SizeEncoded represents the number of bits in the carrier image subjected to geometric attacks after double embedding processing, num represents the maximum number of bits that the image can securely carry, and R represents the capacity utilization factor, which is a scalar coefficient between 0 and 1, and its value is based on the priority of capacity and robustness.

[0031] The steps for enabling fast channel watermark extraction are as follows: perform text decoding (Base64 decoding) on ​​the watermark data, and directly send the decoding result to the zlib decompression module for data decompression to obtain the original watermark information.

[0032] The steps for enabling robust channel watermark extraction based on the number of embedded information bits include: performing a reverse transformation (DWT-DCT-SVD) operation on the geometrically corrected image, the exact opposite of the embedding operation, and extracting the encapsulated data stream bit by bit from the singular values ​​of each block of the image after geometric attack recovery; since embedding may have been performed in all three YUV channels and cyclic embedding was used to increase redundancy, the extraction process calculates the average of the singular values ​​from multiple redundant copies obtained by cyclic embedding of the same bit in multiple YUV channels as a simulated value, and algorithms such as K-Means clustering can be used to binarize the extracted simulated value to obtain a preliminary bit sequence; based on the number of embedded information bits, the length of the binary sequence is truncated at the end of the preliminary bit sequence, and this... The integer decoded from the field serves as the exact length of the watermark data after error correction coding. The portion between the data block and the length information is considered useless padding and is discarded. This invention utilizes a "read length at the tail, retrieve data from the head" strategy, enabling the decoder to accurately segment effective data from a long bitstream without any external information, achieving true self-containment and self-decoding. The RS decoder performs error correction decoding on the data block of the exact length of the watermark data after error correction coding, which is truncated from the beginning of the initial bit sequence. The decoder uses redundant check symbols in the data block to automatically detect and correct bit errors generated during transmission or attacks, generating a compressed data stream. The compressed data stream is then input into the zlib inflate module for data decompression, generating the original watermark information.

[0033] In this embodiment, the present invention achieves robustness in the watermark extraction process through a fixed dual-channel extraction strategy. This strategy ensures that even if one channel fails (such as metadata being erased or pixels being severely damaged), the watermark can still be recovered through the other channel. The fast channel ensures efficient extraction under no-attack or minor-attack conditions, while the robust channel serves as a key safeguard, ensuring extraction capability under severe geometric attacks. The self-contained data encapsulation format adopted makes the extraction process completely free of external parameter input, achieving true blind extraction. This ultimately results in a watermarking system that can achieve millisecond-level response in attack-free environments and maintain stable extraction performance under strong geometric attacks.

[0034] like Figure 4 The diagram shown is a structural schematic of the dynamic redundancy robust digital blind watermarking system provided in this application embodiment. The dynamic redundancy robust digital blind watermarking system provided in this application embodiment includes: an encapsulation processing module, a dual embedding module, and a watermark extraction module. The encapsulation processing module is used to dynamically optimize the watermark data to be embedded and encapsulate the dynamically optimized watermark data to be embedded to obtain a complete encapsulated data stream. The dual embedding module is used to embed the complete encapsulated data stream into the carrier pixel data and simultaneously embed the compressed watermark data to be embedded into the carrier image to obtain a carrier image after dual embedding. The watermark extraction module is used to extract the watermark from the carrier image that has been subjected to geometric attacks after dual embedding. The watermark extraction includes fast channel extraction and robust channel extraction.

[0035] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0036] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0037] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0038] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0039] Although preferred embodiments of the invention have been described, those skilled in the art, upon learning the basic inventive concept, can make other changes and modifications to these embodiments. Therefore, the appended claims are intended to be interpreted as including the preferred embodiments as well as all changes and modifications falling within the scope of the invention.

[0040] Obviously, those skilled in the art can make various modifications and variations to this invention without departing from its spirit and scope. Therefore, if these modifications and variations fall within the scope of the claims of this invention and their equivalents, this invention also intends to include these modifications and variations.

Claims

1. A dynamic redundancy robust digital blind watermarking method, characterized by, Includes the following steps: The watermark data to be embedded is dynamically optimized, and the optimized watermark data is encapsulated based on the carrier size parameters and watermark embedding strength to obtain a complete encapsulated data stream. The dynamic optimization process refers to data compression and error correction coding of the watermark data to be embedded in order to improve the embedding efficiency and resistance to geometric attacks. The encapsulated complete data stream is embedded into the carrier pixel data, and the compressed watermark data to be embedded is embedded into the carrier image to obtain the carrier image after double embedding. Watermark extraction is performed on carrier images subjected to geometric attacks after dual embedding processing. The watermark extraction includes fast channel extraction and robust channel extraction. The steps of dynamically optimizing the data to be embedded with watermarks and encapsulating the dynamically optimized data to obtain a complete encapsulated data stream include: The total number of pixels in the carrier image is obtained based on the carrier image width and carrier image height. The security factor is obtained based on the watermark embedding strength; The maximum number of bits that an image can safely carry is determined based on the security factor and the total number of pixels in the carrier image. Calculate the bit length of the watermark data to be embedded after dynamic optimization processing, and convert it into a binary sequence; The padding length is obtained based on the maximum number of bits that an image can securely carry, the length of the binary sequence, and the bit length of the watermark data to be embedded after dynamic optimization. The padding bits and binary sequence are concatenated into the dynamically optimized data to be embedded with the watermark, resulting in a complete encapsulated data stream. The padding bits consist of a sequence of meaningless bits, the number of which is equal to the padding length; The step of embedding the encapsulated complete data stream into the carrier pixel data includes: Extract the low-frequency sub-band of the carrier image and divide the sub-band into blocks; Perform matrix eigenvalue decomposition on each block to obtain the coefficient matrix of each block. The matrix eigenvalue decomposition includes orthogonal transformation and singular value decomposition. Based on the watermark bit sequence carried by the encapsulated complete data stream, modify the two primary singular values ​​of each block coefficient matrix; Perform inverse transformation and reconstruction on each modified block coefficient matrix; The process of embedding the compressed watermark data into the carrier image is achieved by writing the compressed watermark data into the metadata area of ​​the carrier image, specifically including: The compressed data to be embedded with the watermark is then text-encoded. If the carrier image is in PNG format, the text-encoded watermark data to be embedded is written into a pre-created standard data block in the carrier image. The data block includes a keyword field and a text value field, wherein the keyword field is a pre-set identifier and the text value field stores the text-encoded watermark data to be embedded. If the carrier image is in JPEG format, the compressed watermark data to be embedded is written into the user comment field of its EXIF ​​information. The step of extracting watermarks from carrier images that have been subjected to geometric attacks after double embedding processing includes: The number of embedded information bits is obtained based on the maximum number of bits that the image can safely carry, the number of bits of the carrier image subjected to geometric attacks after dual embedding processing, and a preset capacity utilization factor. Watermarks are extracted via a fast channel. If the fast channel watermark extraction is successful, no additional processing is required. If fast channel watermark extraction fails, the watermark is extracted using robust channel based on the number of embedded information bits. The steps for extracting the watermark via a fast channel are as follows: text decoding of the watermark data, and data decompression of the decoding result to obtain the original watermark information; The step of extracting the watermark through a robust channel based on the number of embedded information bits includes: extracting the encapsulated data stream bit by bit from the singular values ​​of each block of the image after geometric attack recovery; The average of the singular values ​​in multiple redundant copies obtained by cyclic embedding of the same bit is used as the simulated value. The extracted simulated value is then binarized to obtain the preliminary bit sequence. Based on the number of embedded information bits, the length of the binary sequence at the end of the initial bit sequence is extracted, and the integer obtained by decoding this field is used as the exact length of the watermark data after error correction coding. Error correction decoding is performed on the data block of the exact length of the watermark data after error correction encoding, which is truncated from the front end of the initial bit sequence, to generate a compressed data stream; The compressed data stream is decompressed to generate the original watermark information.

2. A dynamic redundancy robust digital blind watermarking system applying the dynamic redundancy robust digital blind watermarking method as claimed in claim 1, characterized in that, It includes a packaging processing module, a watermark embedding module, and a watermark extraction module: The encapsulation processing module is used to dynamically optimize the watermark data to be embedded, and encapsulate the dynamically optimized watermark data to be embedded to obtain a complete encapsulated data stream. The watermark embedding module is used to embed the encapsulated complete data stream into the carrier pixel data, and at the same time embed the compressed watermark data to be embedded into the carrier image, so as to obtain the carrier image after dual embedding processing. The watermark extraction module is used to extract watermarks from carrier images that have been subjected to geometric attacks after double embedding processing. The watermark extraction includes fast channel extraction and robust channel extraction.

Images