A dark watermark generation system and its generation method

Abstract

The present invention relates to the field of digital copyright protection and information security technology, and specifically to a dark watermark generation system and a generation method thereof, comprising an input data module, a robust feature extraction module, an adaptive dark watermark generation module, a deep learning embedding strategy module, and a dark watermark embedding module; wherein: the input data module is used to receive digital content to be protected; the robust feature extraction module is used to generate feature parameters for embedding dark watermarks and determine robust feature positions; the adaptive dark watermark generation module is used to map dark watermark information into feature parameter space to generate adaptive dark watermark data; the embedding strategy module is used to generate dark watermark embedding strategies; the dark watermark embedding module is used to embed dark watermark data into robust feature positions of preprocessed digital content. The present invention achieves high precision and anti-interference of watermark embedding through the combination of robust feature extraction, adaptive watermark generation, and deep learning embedding strategies.

Description

Technical Field

[0001] The present invention relates to the technical field of digital copyright protection and information security, and in particular to a dark watermark generation system and a generation method thereof. Background Art

[0002] In the field of digital copyright protection, with the widespread dissemination of digital content on the Internet and multimedia platforms, ensuring the security of content and the effective protection of copyright has become increasingly important. Digital watermarking technology has been widely used in data in various formats such as images, audio and video. By embedding invisible or imperceptible identifiers into the content itself, content tracking and copyright identification can be achieved. However, traditional watermarking technology often finds it difficult to maintain the robustness and invisibility of embedded watermarks when faced with various signal processing (such as compression, noise addition, conversion, etc.) and malicious tampering. Especially in high compression or complex signal processing operations, the embedded watermark information is easily damaged, resulting in failure of watermark detection and recognition.

[0003] Existing technologies are difficult to effectively cope with the anti-interference requirements and the accuracy of embedded watermarks in complex environments. Specifically, when facing frequent signal processing (such as image compression and format conversion), the embedded watermarks of existing watermark technologies are often easily weakened or even lost. In addition, how to strike a balance between embedding strength and concealment is also a major problem. Traditional methods lack adaptive watermark generation mechanisms and intelligent embedding strategies, and cannot flexibly adjust embedding parameters according to different digital content characteristics, resulting in insufficient robustness of watermarks. Therefore, the present invention proposes a dark watermark generation system and a generation method thereof to solve the above problems. Summary of the invention

[0004] Based on the above objectives, the present invention provides a dark watermark generation system and a generation method thereof.

[0005] A dark watermark generation system includes an input data module, a robust feature extraction module, an adaptive dark watermark generation module, a deep learning embedding strategy module, and a dark watermark embedding module; wherein:

[0006] Input data module: used for receiving digital content to be protected and preprocessing the digital content to remove noise and redundant information to obtain preprocessed digital content;

[0007] Robust feature extraction module: connected to the input data module, it uses multi-scale analysis and frequency domain transformation to process the pre-processed digital content, generate feature parameters for dark watermark embedding, and determine the robust feature position;

[0008] Adaptive dark watermark generation module: connected to the robust feature extraction module, based on feature parameters and preset dark watermark information; using the hybrid domain dark watermark generation algorithm, the dark watermark information is mapped into the feature parameter space to generate adaptive dark watermark data;

[0009] Embedding strategy module: connected to the adaptive dark watermark generation module, used to calculate the embedding position and embedding parameters of the dark watermark according to the adaptive dark watermark data and robust features, and generate the dark watermark embedding strategy;

[0010] Dark watermark embedding module: connected to the embedding strategy module, used to embed the dark watermark data into the robust feature position of the pre-processed digital content according to the dark watermark embedding strategy, complete the watermark embedding operation, and obtain the digital content embedded with the dark watermark.

[0011] Optionally, the input data module includes a data receiving unit, a format recognition unit, a noise removal unit and a redundant information removal unit; wherein:

[0012] Data receiving unit: used to receive digital content data to be protected, the digital content data includes image, audio and video formats, and the data is transmitted to the format recognition unit after receiving;

[0013] Format identification unit: used to identify the format of the received digital content data and confirm whether the content format is image, audio or video;

[0014] Noise removal unit: connected to the format recognition unit, used to execute the corresponding noise removal algorithm according to the content format to reduce irrelevant noise components in the data; the image format data adopts the noise removal algorithm based on frequency domain filtering; the audio format data adopts time domain filtering processing; the video format data applies the joint denoising processing of space and time domains;

[0015] Redundant information removal unit: connected to the noise removal unit, used to identify and remove redundant parts in the data; image format data removes redundant pixel information, audio format data removes silent segments, and video format data removes redundant frames, and finally obtains pre-processed digital content.

[0016] Optionally, the noise removal unit specifically includes:

[0017] Image denoising subunit: used to remove noise from image format data based on frequency domain filtering;

[0018] Audio denoising subunit: used to perform time domain filtering on audio format data;

[0019] Video denoising subunit: used to apply joint denoising processing in spatial and temporal domains to video format data.

[0020] Optionally, the robust feature extraction module includes a multi-scale analysis unit, a frequency domain transformation unit, a feature parameter generation unit and a robust feature position determination unit; wherein:

[0021] Multi-scale analysis unit: used for performing multi-scale decomposition processing on the pre-processed digital content to generate representations of different resolutions and scales; specifically, the pre-processed digital content is decomposed into components of multiple scales through wavelet transform;

[0022] Frequency domain transformation unit: connected to the multi-scale analysis unit, used to perform discrete Fourier transform on each scale component, convert the scale component in the spatial domain into the frequency domain representation, and obtain the corresponding frequency component;

[0023] A characteristic parameter generating unit is connected to the frequency domain transform unit and is used to extract characteristic parameters from the frequency domain data of each scale; specifically, in the frequency domain representation of each scale, the frequency components whose frequency amplitude is greater than a given threshold are selected, and the average amplitude and relative position of the corresponding frequency components are calculated to form characteristic parameters;

[0024] Robust feature position determination unit: connected to the feature parameter generation unit, used to determine the robust feature position suitable for dark watermark embedding according to the generated feature parameters; the robust feature position is selected as the position of several frequency components with the highest frequency amplitude, and the optimal coordinate area for embedding the dark watermark is obtained by calculating the spatial coordinate offset of the position.

[0025] Optionally, the robust feature position determination unit:

[0026] The frequency component sorting subunit is used to sort the frequency domain components provided by the characteristic parameter generating unit in descending order according to the amplitude;

[0027] Position selection subunit: used to select the first position from the sorted frequency components. The components with the highest amplitude constitute the initial robust feature position set ;

[0028] Coordinate offset calculation subunit: connected to the position selection subunit, used to calculate the initial robust feature position set The spatial coordinate offset of each position in the is used to determine the optimal coordinate area for embedding the dark watermark.

[0029] Optionally, the adaptive dark watermark generation module includes a watermark information preprocessing unit, a feature parameter mapping unit and an adaptive watermark data generation unit; wherein:

[0030] The watermark information preprocessing unit is used to preprocess the preset dark watermark information; specifically, the dark watermark information is first represented as a binary sequence, where each bit represents an element of the watermark information; then, the binary sequence is encrypted using an encryption algorithm to generate an encrypted watermark sequence;

[0031] Feature parameter mapping unit: connected to the watermark information preprocessing unit, used to map the encrypted watermark sequence into the feature parameter space;

[0032] The specific steps of mapping include:

[0033] Step 1, representing the feature parameters as a vector, where each element represents a feature parameter;

[0034] Step 2, determining the mapping scale coefficient according to the length of the encrypted watermark sequence and the length of the characteristic parameter vector;

[0035] Step 3, for each bit of encrypted watermark information, calculate its mapping position index in the feature parameter space;

[0036] Step 4, mapping each bit of the encrypted watermark information to the corresponding characteristic parameter to form a mapping pair;

[0037] Adaptive watermark data generating unit: connected to the characteristic parameter mapping unit, used to generate adaptive dark watermark data according to the mapping pair; specifically for each mapping pair, adjust the amplitude of the corresponding characteristic parameter according to the value of the watermark bit; if the watermark bit is 1, increase the amplitude of the characteristic parameter; if the watermark bit is 0, reduce the amplitude of the characteristic parameter; combine all the adjusted characteristic parameters to form adaptive dark watermark data.

[0038] Optionally, the embedding strategy module includes a feature input unit, a model training unit, an embedding position calculation unit and an embedding parameter optimization unit; wherein:

[0039] Feature input unit: used to receive adaptive dark watermark data and robust features, and combine the two to form a feature vector for deep learning processing;

[0040] Model training unit: Based on the preprocessed digital content, the deep neural network model is used to train the feature vector and learn the mapping relationship between the digital content features and the watermark embedding strategy;

[0041] Embedding position calculation unit: uses the trained deep learning model to calculate the optimal embedding position of the dark watermark according to the input feature vector, and determines the coordinate area where the dark watermark is embedded;

[0042] Embedding parameter optimization unit: optimizes watermark embedding parameters, including embedding strength and embedding mode, according to the calculated embedding position and adaptive dark watermark data, and generates a dark watermark embedding strategy.

[0043] Optionally, the embedding parameter optimization unit includes:

[0044] Embedding strength calculation subunit: used to determine the optimal strength of dark watermark embedding according to the robust eigenvalue of the embedding position; the specific steps include:

[0045] Eigenvalue analysis: First calculate the average amplitude of the eigenvalues ​​at the embedded position and variance , the formula is: ; ,in, Represents the total number of feature points at the embedded position, Indicates The amplitude of the feature points;

[0046] Embedding strength determination: Embedding strength is calculated based on the average amplitude and variance of feature points , to enhance robustness and invisibility; the embedding strength calculation formula is: ;in, is the empirical coefficient; represents the influence coefficient of amplitude variance on embedding strength;

[0047] Embedding mode optimization subunit: used to determine the optimal embedding mode according to the embedding strength and dark watermark data; the specific steps include:

[0048] Embedding method selection: Select the embedding method according to the frequency attribute of the embedding position feature; if the embedding position belongs to the low-frequency feature area, select amplitude modulation embedding; if it belongs to the medium-frequency feature area, select phase modulation embedding; the embedding method judgment expression is:

[0049] ;

[0050] in, represents the frequency component of the current feature position, and Represents the frequency range of low frequency and medium frequency respectively;

[0051] Embedding parameter adjustment: According to the adaptive dark watermark data, the parameters under different embedding modes are optimized; for amplitude modulation embedding, the embedding strength is adjusted. Applied to the adjustment of characteristic amplitude; for phase modulation embedding, the embedded phase offset is adjusted according to the value of the watermark data bit , and its calculation formula is: ,in, is the phase modulation coefficient, is the watermark data bit. If the watermark bit is 1, then ; If the watermark bit is 0, then .

[0052] Optionally, the dark watermark embedding module includes an embedding decision unit and a watermark data embedding unit; wherein:

[0053] Embedding decision unit: used to receive the embedding strategy from the embedding parameter optimization unit, including embedding strength and embedding mode;

[0054] Watermark data embedding unit: used to perform a specific dark watermark data embedding operation, and embed the dark watermark data into the robust feature position of the pre-processed digital content according to the embedding strategy;

[0055] The specific steps include:

[0056] Embedding position localization: Based on the robust feature position information, locate the specific embedding coordinates in the pre-processed digital content;

[0057] Watermark data processing: Encode the watermark information according to the embedding strategy and adjust the embedding depth and method of each data bit; according to the embedding method, adjust the amplitude or phase of the data to ensure the correct embedding of each data bit;

[0058] Data embedding execution: embedding according to the encoded watermark data at the determined robust feature location; using frequency domain or time domain methods to accurately embed the watermark data into the digital content;

[0059] Embedding effect verification: After embedding is completed, the robustness of the watermark is verified by simulating attacks, including compression or noise, to ensure that the watermark can still be correctly detected and recovered under different attack conditions.

[0060] A dark watermark generation method is implemented by the above-mentioned dark watermark generation system, comprising the following steps:

[0061] S1: receiving digital content to be protected, removing noise and redundant parts in the digital content through denoising and redundant information removal steps, and obtaining pre-processed digital content;

[0062] S2: Analyze the pre-processed digital content through multi-scale analysis and frequency domain transformation, generate feature parameters for dark watermark embedding, and determine the robust feature position;

[0063] S3: Using the dark watermark information and feature parameters, the dark watermark information is mapped to the robust feature parameter space through the hybrid domain dark watermark generation algorithm to generate adaptive dark watermark data;

[0064] S4: Based on the adaptive dark watermark data and robust features, the embedding position and embedding parameters of the dark watermark are calculated, and a dark watermark embedding strategy is generated;

[0065] S5: according to the dark watermark embedding strategy, the dark watermark data is embedded into the robust feature position of the preprocessed digital content to complete the watermark embedding operation;

[0066] S6: After embedding is completed, the watermark embedding effect is verified by simulating attack signals to ensure that the embedded dark watermark is robust under various attack conditions.

[0067] Beneficial effects of the present invention:

[0068] The present invention realizes precise control of watermark embedding position and parameters by combining robust feature extraction, adaptive dark watermark generation and deep learning embedding strategy. The method extracts robust features suitable for dark watermark embedding through multi-scale analysis and frequency domain transformation, and can effectively cope with various signal processing operations, so that the embedded watermark still maintains strong anti-interference performance under complex environments such as compression, noise addition and format conversion. At the same time, the adaptive dark watermark generation module dynamically adjusts the watermark data according to the characteristic parameters of the digital content, thereby improving the concealment and invisibility of the watermark, and minimizing the impact on the original content while ensuring that the watermark is not destroyed.

[0069] The present invention generates the best embedding position and embedding strength by combining and analyzing adaptive watermark data with robust features, so that the watermark embedding operation has higher flexibility and intelligence level. Finally, the embedded dark watermark can maintain a stable detection effect under a variety of complex attack conditions, thereby greatly improving the copyright protection capability of digital content during the dissemination process. BRIEF DESCRIPTION OF THE DRAWINGS

[0070] In order to more clearly illustrate the technical solutions in the present invention or the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings in the following description are only for the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0071] Figure 1 Schematic diagram of a dark watermark generation system according to an embodiment of the present invention;

[0072] Figure 2 Schematic diagram of a dark watermark generation method according to an embodiment of the present invention. DETAILED DESCRIPTION

[0073] The present invention is described in detail below in conjunction with the accompanying drawings and specific embodiments. At the same time, it is explained here that in order to make the embodiments more detailed, the following embodiments are the best and preferred embodiments, and those skilled in the art may also adopt other alternatives to implement some known technologies; and the accompanying drawings are only for more specific description of the embodiments, and are not intended to specifically limit the present invention.

[0074] It should be noted that the references to "one embodiment", "embodiment", "exemplary embodiments", "some embodiments" and the like in the specification indicate that the embodiments described may include specific features, structures or characteristics, but not every embodiment may include the specific features, structures or characteristics. In addition, when a specific feature, structure or characteristic is described in conjunction with an embodiment, it should be within the knowledge of a person skilled in the art to implement such feature, structure or characteristic in conjunction with other embodiments (whether or not explicitly described).

[0075] In general, a term can be understood, at least in part, from its use in context. For example, depending, at least in part, on the context, the term "one or more" as used herein can be used to describe any feature, structure, or characteristic in the singular sense, or can be used to describe a combination of features, structures, or characteristics in the plural sense. Additionally, the term "based on" can be understood as not necessarily intended to convey an exclusive set of factors, but can instead, depending, at least in part, on the context, allow for the presence of other factors that are not necessarily explicitly described.

[0076] like Figure 1 As shown, a dark watermark generation system includes an input data module, a robust feature extraction module, an adaptive dark watermark generation module, a deep learning embedding strategy module, and a dark watermark embedding module; wherein:

[0077] Input data module: used to receive the digital content to be protected and pre-process the digital content to remove noise and redundant information to obtain pre-processed digital content suitable for dark watermark embedding;

[0078] Robust feature extraction module: connected to the input data module, it uses multi-scale analysis and frequency domain transformation to process the pre-processed digital content, generate feature parameters for dark watermark embedding, and determine the robust feature position;

[0079] Adaptive dark watermark generation module: connected to the robust feature extraction module, based on feature parameters and preset dark watermark information; using the hybrid domain dark watermark generation algorithm, the dark watermark information is mapped into the feature parameter space to generate adaptive dark watermark data;

[0080] Embedding strategy module: connected to the adaptive dark watermark generation module, used to calculate the embedding position and embedding parameters of the dark watermark according to the adaptive dark watermark data and robust features, and generate the dark watermark embedding strategy;

[0081] Dark watermark embedding module: connected to the embedding strategy module, used to embed the dark watermark data into the robust feature position of the pre-processed digital content according to the dark watermark embedding strategy, complete the watermark embedding operation, and obtain the digital content embedded with the dark watermark.

[0082] The input data module includes a data receiving unit, a format recognition unit, a noise removal unit and a redundant information removal unit; wherein:

[0083] Data receiving unit: used to receive the digital content data to be protected, the digital content data includes image, audio and video formats, and the data is transmitted to the format recognition unit after receiving;

[0084] Format recognition unit: used to recognize the format of the received digital content data, confirm whether the content format is image, audio or video, and transmit the data to the corresponding noise removal unit and redundant information removal unit according to the format characteristics;

[0085] Noise removal unit: connected to the format recognition unit, used to execute the corresponding noise removal algorithm according to the content format to reduce irrelevant noise components in the data; the image format data adopts the noise removal algorithm based on frequency domain filtering; the audio format data adopts time domain filtering processing; the video format data applies the joint denoising processing of space and time domains;

[0086] Redundant information removal unit: connected to the noise removal unit, used to identify and remove redundant parts in the data; remove redundant pixel information from image format data, remove silent segments from audio format data, and remove redundant frames from video format data, and finally obtain pre-processed digital content suitable for dark watermark embedding; through the above-mentioned processing steps of each unit of the input data module, the received digital content has the basic conditions for embedding dark watermarks after the noise and redundant information are removed, ensuring the embedding accuracy of the dark watermark generation system and the effectiveness of data quality.

[0087] The noise removal unit specifically includes:

[0088] Image denoising subunit: used to remove noise from image format data based on frequency domain filtering;

[0089] The frequency domain filtering steps are as follows:

[0090] First, the image data is subjected to discrete Fourier transform (DFT) to convert the spatial domain image into frequency domain representation. The calculation formula is: ,in, Represents the pixel value of the spatial domain image, represents the frequency domain image value, is the image width, is the image height, and is the frequency coordinate, represents an imaginary unit;

[0091] Then, a bandpass filter is applied to remove high-frequency and low-frequency components in the frequency domain, leaving only the intermediate-frequency components. The transfer function of the bandpass filter is:

[0092] ,

[0093] in, is the bandpass filter function, represents the frequency distance, and Respectively represent the low-frequency and high-frequency cutoff frequencies of the filter; when the frequency distance At the set low frequency cutoff High frequency cutoff frequency When the frequency is between , the output value of the bandpass filter is 1, and the intermediate frequency component in the signal will be retained; Outside this range (i.e. less than or greater than ), the output value is 0, and the corresponding high-frequency and low-frequency components will be removed;

[0094] Finally, the filtered image data is restored to the spatial domain by inverse discrete Fourier transform to obtain the denoised image data.

[0095] Audio denoising subunit: used to perform time domain filtering on audio format data;

[0096] The steps of time domain filtering are as follows:

[0097] First, the audio signal is sampled to obtain a discrete signal ,in Indicates The audio value of the sampling point, is the discrete time index;

[0098] Then, a sliding average filter is used to smooth the audio signal. The filtering formula is: ,in, Indicates The audio value after denoising of sampling points, represents half the size of the sliding window, Indicates the relative position within the sliding window;

[0099] Finally, the audio signal is smoothed by a sliding average filter to obtain denoised audio data.

[0100] Video denoising subunit: used to apply joint denoising processing in spatial and temporal domains to video format data;

[0101] The steps of joint denoising in spatial and temporal domains are as follows:

[0102] First, a Gaussian filter is applied to each frame in the video for spatial denoising, and the calculation formula is: ,in, represents the Gaussian filter kernel value, and represents the spatial coordinates, and are the coordinates of the filter center, is the standard deviation of the filter;

[0103] Then, the pixels between adjacent frames are denoised in the time domain, the pixel differences between frames are calculated and pixels with small differences are retained to remove rapidly changing noise. The time domain denoising formula is:

[0104] ,

[0105] in, Indicates the current frame Frame at coordinates The pixel value at Represents the pixel value at the same position in the previous frame, is the pixel difference threshold; its working principle is to calculate the current frame ( frame) and the previous frame ( frame) at the same pixel position The pixel value difference between the current frame and the previous frame is used to determine whether to retain the pixel of the current frame; the pixel value difference between the current frame and the previous frame is less than the threshold When the pixel value of the current frame is retained; when the pixel value difference is greater than When using the pixel value of the previous frame Replace the pixel value of the current frame; this method can remove rapidly changing noise and retain inter-frame information with high continuity, thereby effectively reducing the temporal noise in the video data;

[0106] Finally, the denoised video data is obtained through joint denoising in spatial and temporal domains.

[0107] Through the processing steps of the noise removal unit, the noise of data in different formats is effectively removed in the preprocessing stage, ensuring the high efficiency of dark watermark embedding and the clarity of data.

[0108] The robust feature extraction module includes a multi-scale analysis unit, a frequency domain transformation unit, a feature parameter generation unit and a robust feature position determination unit; wherein:

[0109] Multi-scale analysis unit: used to perform multi-scale decomposition processing on the pre-processed digital content to generate representations of different resolutions and scales; specifically, the pre-processed digital content is decomposed into components of multiple scales through wavelet transform, where each component corresponds to detail information or approximate information at different resolutions, and is used to capture the characteristics of the signal at different levels;

[0110] Frequency domain transformation unit: connected to the multi-scale analysis unit, used to perform discrete Fourier transform (DFT) on each scale component, convert the scale component in the spatial domain into the frequency domain representation, and obtain the corresponding frequency component;

[0111] A characteristic parameter generating unit is connected to the frequency domain transform unit and is used to extract characteristic parameters from the frequency domain data of each scale; specifically, in the frequency domain representation of each scale, the frequency components whose frequency amplitude is greater than a given threshold are selected, and the average amplitude and relative position of the corresponding frequency components are calculated to form characteristic parameters;

[0112] The specific calculation steps for extracting feature parameters are as follows:

[0113] In the frequency domain representation of each scale, traverse all frequency components ,in and represents the frequency coordinate;

[0114] For each frequency component The amplitude To judge, when Greater than the preset threshold When , the component is included in the feature set In Chinese, that is: ,in, It is the amplitude threshold set after noise removal, used to select significant frequency components;

[0115] For feature set Calculate the average amplitude of the frequency components in and relative position , where the average amplitude The calculation formula is: ,in, is a feature set Total number of servings;

[0116] Compute feature set The relative positions of all frequency components in the average frequency coordinates are obtained , and its calculation formula is: and ,in, and Represents the average position coordinates of the frequency components in the feature set;

[0117] The average amplitude and the average position As characteristic parameters, to ensure that the characteristic parameters are stable and robust under compression and signal processing conditions.

[0118] Robust feature position determination unit: connected to the feature parameter generation unit, used to determine the robust feature position suitable for dark watermark embedding according to the generated feature parameters; the robust feature position is selected as the position of several frequency components with the highest frequency amplitude, and the optimal coordinate area for embedding the dark watermark is obtained by calculating the spatial coordinate offset of the position to ensure that the dark watermark is still significant and stable after the content is compressed or signal processed; through the above-mentioned unit processing steps of the robust feature extraction module, the feature parameters and robust feature positions suitable for dark watermark embedding are obtained to enhance the robustness and anti-interference of the dark watermark after compression and signal processing operations, and ensure the effective preservation of the watermark information.

[0119] Robust feature location determination unit:

[0120] Frequency component sorting subunit: used to sort the frequency domain components provided by the feature parameter generation unit in descending order according to the amplitude; let the feature set ,in and Respectively represent The frequency coordinates of the frequency components, Indicates The amplitude of the frequency components, the sorting result satisfies ,in is the total number of frequency components in the feature set;

[0121] Position selection subunit: used to select the first position from the sorted frequency components. The components with the highest amplitude constitute the initial robust feature position set ;Right now: ,in, Indicates the number of selected frequency components, which is used to ensure that the robust feature positions are concentrated on the components with larger amplitudes, thereby ensuring the stability of the feature positions after compression and signal processing;

[0122] Coordinate offset calculation subunit: connected to the position selection subunit, used to calculate the initial robust feature position set The spatial coordinate offset of each position in the is used to determine the optimal coordinate area for embedding the dark watermark.

[0123] The specific calculation steps are as follows:

[0124] First, for the set The position coordinates of all frequency components in Perform spatial transformation to obtain the corresponding spatial coordinate offset value ; The transformation formula is: ; ,in, and Respectively represent The horizontal and vertical coordinates of the frequency components in the spatial domain, is the width of the digital content, is the height of the digital content, is the total number of frequency sampling points;

[0125] Then, calculate the set The average coordinate of all spatial coordinate offset values ​​in , as the center position of the dark watermark embedding: ; ;in, and are the average abscissa and average ordinate of the spatial coordinates of all selected frequency components, respectively;

[0126] Finally, Centered on the distribution radius of the selected frequency component Determine the optimal coordinate area, radius The calculation formula is: ,in, From the center The average distance to each feature position is used to limit the range of the optimal coordinate area.

[0127] The adaptive dark watermark generation module includes a watermark information preprocessing unit, a feature parameter mapping unit and an adaptive watermark data generation unit; wherein:

[0128] Watermark information preprocessing unit: used to preprocess the preset dark watermark information; specifically, the dark watermark information is first represented as a binary sequence, where each bit represents an element of the watermark information; then, the encryption algorithm is applied to encrypt the binary sequence to generate an encrypted watermark sequence to enhance the security of the watermark;

[0129] Feature parameter mapping unit: connected to the watermark information preprocessing unit, used to map the encrypted watermark sequence into the feature parameter space;

[0130] The specific steps of mapping include:

[0131] Step 1, representing the feature parameters as a vector, where each element represents a feature parameter;

[0132] Step 2, determining the mapping scale coefficient according to the length of the encrypted watermark sequence and the length of the characteristic parameter vector;

[0133] Step 3, for each bit of encrypted watermark information, calculate its mapping position index in the feature parameter space;

[0134] Step 4, mapping each bit of the encrypted watermark information to the corresponding characteristic parameter to form a mapping pair;

[0135] Adaptive watermark data generation unit: connected to the feature parameter mapping unit, used to generate adaptive dark watermark data according to the mapping pair; specifically for each mapping pair, adjust the amplitude of the corresponding feature parameter according to the value of the watermark bit; if the watermark bit is 1, increase the amplitude of the feature parameter; if the watermark bit is 0, reduce the amplitude of the feature parameter; combine all the adjusted feature parameters to form adaptive dark watermark data; through the above steps, the adaptive dark watermark generation module maps the preset dark watermark information into the feature parameter space, generates adaptive dark watermark data, and provides a basis for the subsequent watermark embedding process.

[0136] The embedding strategy module includes a feature input unit, a model training unit, an embedding position calculation unit and an embedding parameter optimization unit; wherein:

[0137] Feature input unit: used to receive adaptive dark watermark data and robust features, and combine the two to form a feature vector for deep learning processing;

[0138] Model training unit: Based on the preprocessed digital content, the deep neural network model is used to train the feature vector and learn the mapping relationship between the digital content features and the watermark embedding strategy;

[0139] Embedding position calculation unit: uses the trained deep learning model to calculate the optimal embedding position of the dark watermark according to the input feature vector, and determines the coordinate area where the dark watermark is embedded;

[0140] Embedding parameter optimization unit: According to the calculated embedding position and adaptive dark watermark data, optimize the watermark embedding parameters, including embedding strength and embedding method, generate dark watermark embedding strategy, and ensure the invisibility and robustness of the watermark.

[0141] The embedded parameter optimization unit includes:

[0142] Embedding strength calculation subunit: used to determine the optimal strength of dark watermark embedding according to the robust eigenvalue of the embedding position; the specific steps include:

[0143] Eigenvalue analysis: First calculate the average amplitude of the eigenvalues ​​at the embedded position and variance , the formula is: ; ,in, Represents the total number of feature points at the embedded position, Indicates The amplitude of each feature point; through eigenvalue analysis, the characteristic intensity distribution characteristics of the embedded position are obtained;

[0144] Embedding strength determination: Embedding strength is calculated based on the average amplitude and variance of feature points , to enhance robustness and invisibility; the embedding strength calculation formula is: ;in, is the empirical coefficient used to adjust the benchmark of embedding strength; The coefficient that represents the influence of amplitude variance on embedding strength, and adjusts the variation of embedding strength under different feature distributions;

[0145] Embedding mode optimization subunit: used to determine the optimal embedding mode according to the embedding strength and dark watermark data; the specific steps include:

[0146] Embedding method selection: Select the embedding method according to the frequency attribute of the embedding position feature; if the embedding position belongs to the low-frequency feature area, select amplitude modulation embedding; if it belongs to the medium-frequency feature area, select phase modulation embedding; the embedding method judgment expression is:

[0147] ;

[0148] in, represents the frequency component of the current feature position, and Represents the frequency range of low frequency and medium frequency respectively;

[0149] Embedding parameter adjustment: According to the adaptive dark watermark data, the parameters under different embedding modes are optimized; for amplitude modulation embedding, the embedding strength is adjusted. Applied to the adjustment of feature amplitude; for phase modulation embedding, the embedded phase offset is adjusted according to the value of the watermark data bit (0 or 1) , and its calculation formula is: ,in, is the phase modulation coefficient, is the watermark data bit. If the watermark bit is 1, then ; If the watermark bit is 0, then ; Adjust the embedding effect of different data bits in the frequency domain through phase modulation; through the synergy of the embedding strength calculation subunit and the embedding mode optimization subunit, the embedding parameter optimization unit can generate the optimal dark watermark embedding strategy according to the embedding position and adaptive dark watermark data, ensuring the embedding accuracy, invisibility and robustness of the dark watermark.

[0150] The dark watermark embedding module includes an embedding decision unit and a watermark data embedding unit; wherein:

[0151] Embedding decision unit: used to receive the embedding strategy from the embedding parameter optimization unit, including embedding strength and embedding mode;

[0152] Watermark data embedding unit: used to perform a specific dark watermark data embedding operation, and embed the dark watermark data into the robust feature position of the pre-processed digital content according to the embedding strategy;

[0153] The specific steps include:

[0154] Embedding position localization: Based on the robust feature position information, locate the specific embedding coordinates in the pre-processed digital content;

[0155] Watermark data processing: Encode the watermark information according to the embedding strategy and adjust the embedding depth and method of each data bit; according to the embedding method, adjust the amplitude or phase of the data to ensure the correct embedding of each data bit;

[0156] Data embedding execution: embed the encoded watermark data at the determined robust feature location; use frequency domain or time domain methods to accurately embed the watermark data into the digital content. The specific method depends on the embedding method previously decided (amplitude modulation or phase modulation). For amplitude modulation, the amplitude of the frequency component is adjusted; for phase modulation, the phase of the frequency component is adjusted;

[0157] Embedding effect verification: After embedding is completed, the robustness of the watermark is verified by simulating attacks, including compression or noise, to ensure that the watermark can still be correctly detected and recovered under different attack conditions; through the above units and steps, the dark watermark embedding module can effectively embed the dark watermark data into the robust feature position of the preprocessed digital content, complete the watermark embedding operation, and ensure the invisibility, robustness and security of the dark watermark.

[0158] like Figure 2 As shown, a dark watermark generation method is implemented by the above-mentioned dark watermark generation system, comprising the following steps:

[0159] S1: receiving the digital content to be protected, removing the noise and redundancy in the digital content through denoising and redundant information removal steps, and obtaining the pre-processed digital content, so as to ensure that the processed content has the basic conditions for embedding the dark watermark;

[0160] S2: Analyze the pre-processed digital content through multi-scale analysis and frequency domain transformation, generate feature parameters for dark watermark embedding, and determine the robust feature position;

[0161] S3: Using the dark watermark information and feature parameters, the dark watermark information is mapped to the robust feature parameter space through the hybrid domain dark watermark generation algorithm to generate adaptive dark watermark data;

[0162] S4: Based on the adaptive dark watermark data and robust features, the embedding position and embedding parameters of the dark watermark are calculated, and a dark watermark embedding strategy is generated;

[0163] S5: according to the dark watermark embedding strategy, the dark watermark data is embedded into the robust feature position of the preprocessed digital content to complete the watermark embedding operation;

[0164] S6: After embedding is completed, the watermark embedding effect is verified by simulating attack signals (such as compression, noise addition, etc.) to ensure that the embedded dark watermark is robust under various attack conditions.

[0165] The present invention covers any substitution, modification, equivalent method and scheme made on the essence and scope of the present invention. In order to make the public have a thorough understanding of the present invention, specific details are described in detail in the following preferred embodiments of the present invention, but those skilled in the art can fully understand the present invention without the description of these details. In addition, in order to avoid unnecessary confusion about the essence of the present invention, well-known methods, processes, procedures, components and circuits are not described in detail.

[0166] The above is only a preferred embodiment of the present invention. It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the principle of the present invention. These improvements and modifications should also be regarded as the scope of protection of the present invention.

Claims

1. A dark watermark generation system, characterized in that: It includes an input data module, a robust feature extraction module, an adaptive dark watermark generation module, a deep learning embedding strategy module, and a dark watermark embedding module; among which: Input data module: used for receiving digital content to be protected and preprocessing the digital content to remove noise and redundant information to obtain preprocessed digital content; Robust feature extraction module: connected to the input data module, it uses multi-scale analysis and frequency domain transformation to process the pre-processed digital content, generate feature parameters for dark watermark embedding, and determine the robust feature position; The robust feature extraction module includes a multi-scale analysis unit, a frequency domain transformation unit, a feature parameter generation unit and a robust feature position determination unit; wherein: Multi-scale analysis unit: used for performing multi-scale decomposition processing on the pre-processed digital content to generate representations of different resolutions and scales; specifically, the pre-processed digital content is decomposed into components of multiple scales through wavelet transform; Frequency domain transformation unit: connected to the multi-scale analysis unit, used to perform discrete Fourier transform on each scale component, convert the scale component in the spatial domain into the frequency domain representation, and obtain the corresponding frequency component; A characteristic parameter generating unit is connected to the frequency domain transform unit and is used to extract characteristic parameters from the frequency domain data of each scale; specifically, in the frequency domain representation of each scale, the frequency components whose frequency amplitude is greater than a given threshold are selected, and the average amplitude and relative position of the corresponding frequency components are calculated to form characteristic parameters; A robust feature position determination unit is connected to the feature parameter generation unit and is used to determine a robust feature position suitable for embedding a dark watermark according to the generated feature parameters; the robust feature position is selected as the position of several frequency components with the highest frequency amplitude, and the optimal coordinate area for embedding the dark watermark is obtained by calculating the spatial coordinate offset of the position; The robust feature position determination unit comprises: The frequency component sorting subunit is used to sort the frequency domain components provided by the characteristic parameter generating unit in descending order according to the amplitude; Position selection subunit: used to select the first position from the sorted frequency components. The components with the highest amplitude constitute the initial robust feature position set ; Coordinate offset calculation subunit: connected to the position selection subunit, used to calculate the initial robust feature position set The spatial coordinate offset of each position in the image is used to determine the best coordinate area for embedding the dark watermark; The specific calculation steps include: First, for the set The position coordinates of all frequency components in Perform spatial transformation to obtain the corresponding spatial coordinate offset value ; The transformation formula is: ; ,in, and Respectively represent The horizontal and vertical coordinates of the frequency components in the spatial domain, is the width of the digital content, is the height of the digital content, is the total number of frequency sampling points; Then, calculate the set The average coordinate of all spatial coordinate offset values ​​in , as the center position of the dark watermark embedding: ; ;in, and are the average abscissa and average ordinate of the spatial coordinates of all selected frequency components, respectively; Finally, Centered on the distribution radius of the selected frequency component Determine the optimal coordinate area, radius The calculation formula is: ,in, From the center The average distance to each feature position is used to limit the range of the optimal coordinate area; Adaptive dark watermark generation module: connected to the robust feature extraction module, based on feature parameters and preset dark watermark information; using the hybrid domain dark watermark generation algorithm, the dark watermark information is mapped into the feature parameter space to generate adaptive dark watermark data; The adaptive dark watermark generation module includes a watermark information preprocessing unit, a feature parameter mapping unit and an adaptive watermark data generation unit; wherein: The watermark information preprocessing unit is used to preprocess the preset dark watermark information; specifically, the dark watermark information is first represented as a binary sequence, where each bit represents an element of the watermark information; then, the binary sequence is encrypted using an encryption algorithm to generate an encrypted watermark sequence; Feature parameter mapping unit: connected to the watermark information preprocessing unit, used to map the encrypted watermark sequence into the feature parameter space; The specific steps of mapping include: Step 1, representing the feature parameters as a vector, where each element represents a feature parameter; Step 2, determining the mapping scale coefficient according to the length of the encrypted watermark sequence and the length of the characteristic parameter vector; Step 3, for each bit of encrypted watermark information, calculate its mapping position index in the feature parameter space; Step 4, mapping each bit of the encrypted watermark information to the corresponding characteristic parameter to form a mapping pair; Adaptive watermark data generating unit: connected to the characteristic parameter mapping unit, used to generate adaptive dark watermark data according to the mapping pair; specifically for each mapping pair, according to the value of the watermark bit, adjust the amplitude of the corresponding characteristic parameter; if the watermark bit is 1, increase the amplitude of the characteristic parameter; if the watermark bit is 0, reduce the amplitude of the characteristic parameter; combine all the adjusted characteristic parameters to form adaptive dark watermark data; Embedding strategy module: connected to the adaptive dark watermark generation module, used to calculate the embedding position and embedding parameters of the dark watermark according to the adaptive dark watermark data and robust features, and generate the dark watermark embedding strategy; Dark watermark embedding module: connected to the embedding strategy module, used to embed the dark watermark data into the robust feature position of the pre-processed digital content according to the dark watermark embedding strategy, complete the watermark embedding operation, and obtain the digital content embedded with the dark watermark.

2. A dark watermark generation system according to claim 1, characterized in that: The input data module includes a data receiving unit, a format recognition unit, a noise removal unit and a redundant information removal unit; wherein: Data receiving unit: used to receive digital content data to be protected, the digital content data includes image, audio and video formats, and the data is transmitted to the format recognition unit after receiving; Format identification unit: used to identify the format of the received digital content data and confirm whether the content format is image, audio or video; Noise removal unit: connected to the format recognition unit, used to execute the corresponding noise removal algorithm according to the content format to reduce irrelevant noise components in the data; the image format data adopts the noise removal algorithm based on frequency domain filtering; the audio format data adopts time domain filtering processing; the video format data applies the joint denoising processing of space and time domains; Redundant information removal unit: connected to the noise removal unit, used to identify and remove redundant parts in the data; image format data removes redundant pixel information, audio format data removes silent segments, and video format data removes redundant frames, and finally obtains pre-processed digital content.

3. A dark watermark generation system according to claim 2, characterized in that: The noise removal unit specifically includes: Image denoising subunit: used to remove noise from image format data based on frequency domain filtering; Audio denoising subunit: used to perform time domain filtering on audio format data; Video denoising subunit: used to apply joint denoising processing in spatial and temporal domains to video format data.

4. A dark watermark generation system according to claim 1, characterized in that: The embedding strategy module includes a feature input unit, a model training unit, an embedding position calculation unit and an embedding parameter optimization unit; wherein: Feature input unit: used to receive adaptive dark watermark data and robust features, and combine the two to form a feature vector for deep learning processing; Model training unit: Based on the preprocessed digital content, the deep neural network model is used to train the feature vector and learn the mapping relationship between the digital content features and the watermark embedding strategy; Embedding position calculation unit: uses the trained deep learning model to calculate the optimal embedding position of the dark watermark according to the input feature vector, and determines the coordinate area where the dark watermark is embedded; Embedding parameter optimization unit: optimizes watermark embedding parameters, including embedding strength and embedding mode, according to the calculated embedding position and adaptive dark watermark data, and generates a dark watermark embedding strategy.

5. A dark watermark generation system according to claim 4, characterized in that: The embedding parameter optimization unit comprises: Embedding strength calculation subunit: used to determine the optimal strength of dark watermark embedding according to the robust eigenvalue of the embedding position; the specific steps include: Eigenvalue analysis: First calculate the average amplitude of the eigenvalues ​​at the embedded position and variance , the formula is: ; ,in, Represents the total number of feature points at the embedded position, Indicates The amplitude of the feature points; Embedding strength determination: Embedding strength is calculated based on the average amplitude and variance of feature points , to enhance robustness and invisibility; the embedding strength calculation formula is: ;in, is the empirical coefficient; represents the influence coefficient of amplitude variance on embedding strength; Embedding mode optimization subunit: used to determine the optimal embedding mode according to the embedding strength and dark watermark data; the specific steps include: Embedding method selection: Select the embedding method according to the frequency attribute of the embedding position feature; if the embedding position belongs to the low-frequency feature area, select amplitude modulation embedding; if it belongs to the medium-frequency feature area, select phase modulation embedding; the embedding method judgment expression is: ； in, represents the frequency component of the current feature position, and Represents the frequency range of low frequency and medium frequency respectively; Embedding parameter adjustment: According to the adaptive dark watermark data, the parameters under different embedding modes are optimized; for amplitude modulation embedding, the embedding strength is adjusted. Applied to the adjustment of characteristic amplitude; for phase modulation embedding, the embedded phase offset is adjusted according to the value of the watermark data bit , and its calculation formula is: ,in, is the phase modulation coefficient, is the watermark data bit. If the watermark bit is 1, then ; If the watermark bit is 0, then .

6. A dark watermark generation system according to claim 1, characterized in that: The dark watermark embedding module includes an embedding decision unit and a watermark data embedding unit; wherein: Embedding decision unit: used to receive the embedding strategy from the embedding parameter optimization unit, including embedding strength and embedding mode; Watermark data embedding unit: used to perform a specific dark watermark data embedding operation, and embed the dark watermark data into the robust feature position of the pre-processed digital content according to the embedding strategy; The specific steps include: Embedding position localization: Based on the robust feature position information, locate the specific embedding coordinates in the pre-processed digital content; Watermark data processing: Encode the watermark information according to the embedding strategy and adjust the embedding depth and method of each data bit; according to the embedding method, adjust the amplitude or phase of the data to ensure the correct embedding of each data bit; Data embedding execution: embedding according to the encoded watermark data at the determined robust feature location; using frequency domain or time domain methods to accurately embed the watermark data into the digital content; Embedding effect verification: After embedding is completed, the robustness of the watermark is verified by simulating attacks, including compression or noise, to ensure that the watermark can still be correctly detected and recovered under different attack conditions.

7. A dark watermark generation method, implemented by a dark watermark generation system according to any one of claims 1 to 6, characterized in that: The following steps are involved: S1: receiving digital content to be protected, removing noise and redundant parts in the digital content through denoising and redundant information removal steps, and obtaining pre-processed digital content; S2: Analyze the pre-processed digital content through multi-scale analysis and frequency domain transformation, generate feature parameters for dark watermark embedding, and determine the robust feature position; S3: Using the dark watermark information and feature parameters, the dark watermark information is mapped to the robust feature parameter space through the hybrid domain dark watermark generation algorithm to generate adaptive dark watermark data; S4: Based on the adaptive dark watermark data and robust features, the embedding position and embedding parameters of the dark watermark are calculated, and a dark watermark embedding strategy is generated; S5: according to the dark watermark embedding strategy, the dark watermark data is embedded into the robust feature position of the preprocessed digital content to complete the watermark embedding operation; S6: After embedding is completed, the watermark embedding effect is verified by simulating attack signals to ensure that the embedded dark watermark is robust under various attack conditions.

Images