Audio anti-noise steganography method and information blind extraction method for low signal-to-noise ratio environment

Abstract

The application provides an audio anti-noise steganography method and information blind extraction method for a low signal-to-noise ratio environment, which can be applied to the fields of audio data processing and general security technology. The audio anti-noise steganography method comprises the following steps: mapping watermark data into a fault-tolerant coding matrix and performing a two-dimensional linear endomorphism scrambling transformation to generate a two-dimensional data carrier; performing multi-scale decomposition and discrete cosine transformation on an original audio signal to obtain a two-dimensional frequency domain coefficient matrix; determining a preset high-frequency region of each row of coefficients in the two-dimensional frequency domain coefficient matrix as an embedding region, and calculating the amplitude average value of the coefficients in each embedding region as a statistical steady-state energy feature; adjusting the quantization step based on the statistical steady-state energy features of each row in the two-dimensional frequency domain coefficient matrix, and embedding the corresponding coding elements in the two-dimensional data carrier into the embedding regions of each row based on the statistical steady-state energy features and the quantization step, to reconstruct the audio signal containing the watermark data based on the feature coefficient matrix of the watermark data.

Description

Technical Field

[0001] This application relates to the fields of audio data processing and general security technology, and in particular to an audio noise-resistant steganography method and a blind information extraction method for low signal-to-noise ratio environments. Background Technology

[0002] Audio covert transmission technology (i.e., audio watermarking technology) can establish a covert data transmission link by utilizing the redundancy space of the audio signal without compromising the original audio listening quality. Therefore, this technology has significant application value in scenarios such as command issuance in complex electromagnetic environments, data return in analog channels, and integrity authentication of multimedia content.

[0003] However, most audio covert transmission technologies are designed based on the assumption of "lossless or high signal-to-noise ratio channels". When applied to harsh channels such as open networks, underwater acoustic communication or analog broadcasting, the reliability of communication is often difficult to maintain due to severe noise interference and physical damage in the channel transmission link. Summary of the Invention

[0004] In view of the above problems, this application provides an audio noise-resistant steganography method and a blind information extraction method for low signal-to-noise ratio environments.

[0005] According to the first aspect of this application, an audio noise-resistant steganography method for low signal-to-noise ratio environments is provided, comprising:

[0006] The watermark data to be embedded is mapped to a fault-tolerant coding matrix, and based on the dynamic key, a two-dimensional linear self-isomorphic scrambling transformation is performed on the above fault-tolerant coding matrix to generate a two-dimensional data carrier.

[0007] The original audio signal is decomposed into multiple scales and subjected to discrete cosine transform to obtain a two-dimensional frequency domain coefficient matrix.

[0008] The preset high-frequency region of the coefficients in each row of the above two-dimensional frequency domain coefficient matrix is ​​determined as the embedding region of each row of the above two-dimensional frequency domain coefficient matrix, and the average amplitude of the coefficients in each embedding region is calculated as a statistical steady-state energy characteristic.

[0009] Based on the statistical steady-state energy characteristics of each row in the two-dimensional frequency domain coefficient matrix, the quantization step size is adjusted. Based on the statistical steady-state energy characteristics and quantization step size of each row in the two-dimensional frequency domain coefficient matrix, the coding elements in the two-dimensional data carrier corresponding to each row in the two-dimensional frequency domain coefficient matrix are embedded into the embedding regions of each row in the two-dimensional frequency domain coefficient matrix to obtain the feature coefficient matrix of the watermarked data.

[0010] Based on the feature coefficient matrix of the watermarked data, the audio signal of the watermarked data is reconstructed.

[0011] The second aspect of this application provides a method for blind information extraction, comprising:

[0012] Blind extraction is performed on the audio signal of the watermarked data to obtain a composite verification substrate, wherein the composite verification substrate is used to perform watermark verification on the audio signal of the watermarked data.

[0013] The audio signal containing the watermarked data was obtained using the audio noise-resistant steganography method for low signal-to-noise ratio environments.

[0014] According to the audio noise-resistant steganography method and blind information extraction method for low signal-to-noise ratio environments provided in this application, the watermark data to be embedded is mapped to a fault-tolerant coding matrix to utilize its unique fault-tolerant mechanism to combat low signal-to-noise ratio scenarios and reduce interference to the actual watermark data. A two-dimensional linear self-similar scrambling transformation is performed on the fault-tolerant code to construct a fault-tolerant two-dimensional data carrier, enabling the conversion of burst shearing or packet loss in the channel into controllable random errors. Since the original audio signal is decomposed into multiple scales and subjected to discrete cosine transform, the resulting two-dimensional frequency domain coefficient matrix is ​​generated based on high-frequency components, effectively isolating the influence of low-frequency baseband interference. Furthermore, since the effective signal energy in the two-dimensional frequency domain coefficient matrix is ​​highly concentrated in the low-frequency region, the preset high-frequency region of each row of coefficients in the two-dimensional frequency domain coefficient matrix is ​​determined as the embedding region to achieve embedding without significantly affecting sound quality and improve the concealment of information embedding. By calculating the average amplitude of the coefficients within the embedded region instead of using single-point coefficients for modulation, the resulting statistical steady-state energy characteristics cancel out the zero-mean additive white Gaussian noise. Furthermore, due to the ability to work with an adaptive quantization step size mechanism dynamically bound to the statistical steady-state energy characteristics, it possesses automatic gain control capabilities, enabling immunity to amplitude fading during channel transmission. Thus, while ensuring auditory perception transparency, it achieves highly robust and covert transmission of watermarked data embedded in the audio signal under harsh electromagnetic environments or high-compression links.

[0015] Based on this, blind extraction of the audio signal of the watermarked data can be performed to obtain a composite verification substrate, thus enabling dual identity verification with both machine readability and human verification. This achieves self-healing recovery and dual identity authentication of the data, verifying whether noise and other interference in the channel affect the covertly transmitted watermarked data during transmission. Attached Figure Description

[0016] The above-mentioned contents, other objects, features and advantages of this application will become clearer from the following description of embodiments with reference to the accompanying drawings, which will be explained in conjunction with the drawings.

[0017] Figure 1The illustration shows an application scenario of an audio noise-resistant steganography method or a blind information extraction method for low signal-to-noise ratio environments according to embodiments of this application.

[0018] Figure 2 A flowchart of an audio noise-resistant steganography method for low signal-to-noise ratio environments according to an embodiment of this application is shown.

[0019] Figure 3 A flowchart illustrating the process of obtaining the feature coefficient matrix of watermarked data according to an embodiment of this application is shown.

[0020] Figure 4 A flowchart illustrating an audio noise-resistant steganography method for low signal-to-noise ratio environments according to an embodiment of this application is shown.

[0021] Figure 5 A flowchart illustrating the process of obtaining a composite verification substrate according to an embodiment of this application is shown.

[0022] Figure 6 A flowchart of an information blind extraction method according to an embodiment of this application is shown.

[0023] Figure 7 A structural block diagram of an audio noise-resistant steganography device for low signal-to-noise ratio environments according to an embodiment of this application is shown.

[0024] Figure 8 A structural block diagram of an information blind extraction apparatus according to an embodiment of this application is shown.

[0025] Figure 9 A block diagram of an electronic device suitable for implementing an audio noise-resistant steganography method or a blind information extraction method for low signal-to-noise ratio environments, according to an embodiment of this application, is shown. Detailed Implementation

[0026] The embodiments of this application will now be described with reference to the accompanying drawings. However, it should be understood that these descriptions are exemplary only and are not intended to limit the scope of this application. In the following detailed description, numerous specific details are set forth to provide a thorough understanding of the embodiments of this application for ease of explanation. However, it will be apparent that one or more embodiments may be implemented without these specific details. Furthermore, descriptions of well-known structures and technologies are omitted in the following description to avoid unnecessarily obscuring the concepts of this application.

[0027] The terminology used herein is for the purpose of describing particular embodiments only and is not intended to limit the scope of this application. The terms “comprising,” “including,” etc., as used herein indicate the presence of the stated features, steps, operations, and / or components, but do not exclude the presence or addition of one or more other features, steps, operations, or components.

[0028] All terms used herein (including technical and scientific terms) have the meanings commonly understood by those skilled in the art, unless otherwise defined. It should be noted that the terms used herein are to be interpreted in a manner consistent with the context of this specification, and not in an idealized or overly rigid way.

[0029] When using expressions such as "at least one of A, B and C", they should generally be interpreted in accordance with the meaning that is commonly understood by those skilled in the art (e.g., "a system having at least one of A, B and C" should include, but is not limited to, a system having A alone, a system having B alone, a system having C alone, a system having A and B, a system having A and C, a system having B and C, and / or a system having A, B and C, etc.).

[0030] In the technical solution of this application, the user information (including but not limited to user personal information, user image information, user device information, such as location information) and data (including but not limited to data used for analysis, stored data, and displayed data) involved are all information and data authorized by the user or fully authorized by all parties. Furthermore, the collection, storage, use, processing, transmission, provision, disclosure, and application of related data all comply with relevant laws, regulations, and standards, take necessary confidentiality measures, do not violate public order and good morals, and provide corresponding operation entry points for users to choose to authorize or refuse.

[0031] In scenarios involving automated decision-making using personal information, the methods, devices, and systems provided in this application all offer users corresponding entry points for choosing to agree to or reject the automated decision-making results. If the user chooses to reject, the process proceeds to the expert decision-making stage. Here, "automated decision-making" refers to the activity of automatically analyzing and evaluating an individual's behavioral habits, interests, or economic, health, and credit status through computer programs, and then making a decision. Here, "expert decision-making" refers to the activity of making decisions by personnel who specialize in a particular field, possess specialized experience, knowledge, and skills, and have reached a certain level of professional expertise.

[0032] In the process of developing this application, it was discovered that the mainstream technical solutions for covert audio transmission mainly include time-domain steganography and transform-domain steganography. However, in scenarios with low signal-to-noise ratio and impaired channels, covert audio transmission technology faces the following challenges.

[0033] In low signal-to-noise ratio environments, the intensity of noise superimposed in the channel often approaches or even exceeds the embedding intensity of the concealed signal. However, traditional algorithms often rely on small amplitude modifications of single-point transform coefficients to carry information. Such microscopic features are easily completely covered or have their polarity altered by high-intensity environmental noise, causing the receiver to be unable to extract effective feature signals from the noise background, thus leading to a surge in the bit error rate.

[0034] When audio signals are transmitted in analog, multipath, or through a power amplifier, they often experience overall amplitude fading or nonlinear gain fluctuations. However, the algorithms in related technologies typically rely on preset fixed quantization step sizes or static decision thresholds. This means that once the signal energy envelope shifts overall, the decision interval at the receiver will systematically mismatch with the signal characteristics, leading to continuous misjudgments of originally correct information at the extraction end.

[0035] In real-world physical channels, transient electromagnetic pulse interference, signal packet loss, or malicious time-domain shearing attacks are common, leading to sudden data stream loss. However, traditional solutions often lack fault-tolerance mechanisms based on channel coding theory. Once a critical synchronization frame or data segment is lost, the entire data link is broken, and it becomes impossible to recover the original information using the remaining incomplete data.

[0036] Based on the above, traditional audio steganography and watermarking algorithms are mostly limited to ideal or high signal-to-noise ratio channel environments, and it is difficult to maintain the integrity of the communication link under harsh conditions such as strong noise interference, amplitude fading and signal shearing.

[0037] Furthermore, random noise superimposed on poor signals can easily mask weak single-point embedding features (i.e., feature submersion effect), and amplitude fading during transmission can cause overall fluctuations in the received signal energy, resulting in severe gain mismatch in quantization decision mechanisms based on fixed step sizes. In addition, sudden time-domain shearing in the channel often causes the loss of continuous data segments, directly blocking the decoding link.

[0038] Therefore, related technologies cannot establish robust transmission channels with fault-tolerant mechanisms, noise printing, and adaptive gain control capabilities while ensuring auditory perception transparency. This results in extremely high bit error rates in complex electromagnetic environments or underwater acoustic channels, making it difficult to meet the requirements for highly reliable covert communication in low signal-to-noise ratio environments. Consequently, there is an urgent need for a method in related technologies that can overcome feature submersion, gain mismatch, and link breakage problems to achieve highly robust covert transmission and blind extraction of information in harsh communication environments.

[0039] Therefore, embodiments of this application provide an audio noise-resistant steganography method and a blind information extraction method for low signal-to-noise ratio environments, so as to achieve highly robust covert transmission in harsh communication environments while ensuring auditory perception transparency, thereby maintaining communication reliability.

[0040] Figure 1 The illustration shows an application scenario of an audio noise-resistant steganography method or a blind information extraction method for low signal-to-noise ratio environments according to embodiments of this application.

[0041] like Figure 1As shown, application scenario 100 according to this embodiment may include a first terminal device 101, a second terminal device 102, a third terminal device 103, a network 104, and a server 105. The network 104 serves as a medium for providing a communication link between the first terminal device 101, the second terminal device 102, the third terminal device 103, and the server 105. The network 104 may include various connection types, such as wired or wireless communication links, or fiber optic cables, etc.

[0042] Users can use the first terminal device 101, the second terminal device 102, and the third terminal device 103 to interact with the server 105 via the network 104 to receive or send messages, etc. Various communication client applications can be installed on the first terminal device 101, the second terminal device 102, and the third terminal device 103, such as shopping applications, web browser applications, search applications, instant messaging tools, email clients, social media platform software, etc. (for example only).

[0043] The first terminal device 101, the second terminal device 102, and the third terminal device 103 can be various electronic devices with displays and support web browsing, including but not limited to smartphones, tablets, laptops, and desktop computers.

[0044] Server 105 can be a server that provides various services, such as a backend management server that supports websites browsed by users using the first terminal device 101, the second terminal device 102, and the third terminal device 103 (this is just an example). The backend management server can analyze and process data such as received user requests, and feed back the processing results (such as web pages, information, or data obtained or generated according to user requests) to the terminal devices.

[0045] For example, server 105 can map the watermark data to be embedded into a fault-tolerant coding matrix, and based on a dynamic key, perform a two-dimensional linear automorphic scrambling transformation on the fault-tolerant coding matrix to generate a two-dimensional data carrier; perform multi-scale decomposition and discrete cosine transform on the original audio signal to obtain a two-dimensional frequency domain coefficient matrix; determine the preset high-frequency region of the coefficients in each row of the two-dimensional frequency domain coefficient matrix as the embedding region of each row of the two-dimensional frequency domain coefficient matrix, and calculate the average amplitude of the coefficients in each embedding region as a statistical steady-state energy feature; thereby, adjust the quantization step size based on the statistical steady-state energy feature of each row of the two-dimensional frequency domain coefficient matrix, and based on the statistical steady-state energy feature and quantization step size of each row of the two-dimensional frequency domain coefficient matrix, embed the coding elements in the two-dimensional data carrier corresponding to each row of the two-dimensional frequency domain coefficient matrix into the embedding region of each row of the two-dimensional frequency domain coefficient matrix respectively, to obtain the feature coefficient matrix of the watermarked data, and finally reconstruct the audio signal of the watermarked data based on the feature coefficient matrix of the watermarked data.

[0046] Server 105 can also perform blind extraction of the audio signal of the watermarked data to obtain a composite verification substrate, wherein the composite verification substrate is used to perform watermark verification on the audio signal of the watermarked data.

[0047] It should be noted that the audio noise-resistant steganography method or blind information extraction method for low signal-to-noise ratio environments provided in this application embodiment can generally be executed by server 105. Correspondingly, the audio noise-resistant steganography device or blind information extraction device for low signal-to-noise ratio environments provided in this application embodiment can generally be located in server 105. The audio noise-resistant steganography method or blind information extraction method for low signal-to-noise ratio environments provided in this application embodiment can also be executed by a server or server cluster that is different from server 105 and capable of communicating with the first terminal device 101, the second terminal device 102, the third terminal device 103, and / or server 105. Correspondingly, the audio noise-resistant steganography device or blind information extraction device for low signal-to-noise ratio environments provided in this application embodiment can also be located in a server or server cluster that is different from server 105 and capable of communicating with the first terminal device 101, the second terminal device 102, the third terminal device 103, and / or server 105.

[0048] It should be understood that Figure 1 The number of the first terminal device 101, second terminal device 102, third terminal device 103, network 104, and server 105 in the diagram is merely illustrative. Depending on the implementation requirements, any number of the first terminal device 101, second terminal device 102, third terminal device 103, network 104, and server 105 may be included.

[0049] The following will be based on Figure 1 The described scene, through Figures 2-6 This application provides a detailed description of the audio noise-resistant steganography method or information blind extraction method for low signal-to-noise ratio environments.

[0050] Figure 2 A flowchart of an audio noise-resistant steganography method for low signal-to-noise ratio environments according to an embodiment of this application is shown.

[0051] like Figure 2 As shown, the method 200 includes operations S210 to S250.

[0052] In operation S210, the watermark data to be embedded is mapped to a fault-tolerant coding matrix, and based on the dynamic key, a two-dimensional linear automorphic scrambling transformation is performed on the fault-tolerant coding matrix to generate a two-dimensional data carrier.

[0053] In one embodiment, the watermark data to be embedded represents the data to be embedded into the audio signal to be transmitted. It uses the audio signal as a carrier to realize functions such as covert communication and identity authentication, while not affecting the normal auditory effect of the audio signal.

[0054] For example, the watermark data to be embedded can be digital information, and its form can include numerical values, text, images, etc.

[0055] Because of the risk of link interruption during transmission in poor channels, the watermark data to be embedded can be constructed into data with fault tolerance capabilities. By utilizing its unique fault tolerance mechanism, a defense against channel time-domain shearing and packet loss can be built, providing a basis for combating low signal-to-noise ratio scenarios. This allows the watermark data to still be extracted from the audio signal even after malicious shearing.

[0056] Specifically, the watermark data to be embedded can be mapped to a fault-tolerant coding matrix.

[0057] In one embodiment, the watermark data to be embedded is mapped to a fault-tolerant coding matrix that follows a maximum fault-tolerant redundancy protocol.

[0058] For example, the fault-tolerant coding matrix can be a two-dimensional 0-1 matrix based on QR codes to carry the watermark data to be embedded. Because this fault-tolerant coding matrix has built-in error-correcting redundancy coding, it has fault-tolerant redundancy characteristics. Even if some areas are damaged or interfered with, it can still ensure the complete and reliable recovery of data, and has strong anti-interference and error correction capabilities.

[0059] In one embodiment, in order to further combat interference and damage to the watermark data embedded in the audio signal in a poor channel, a two-dimensional linear automorphic scrambling transformation is performed on the fault-tolerant coding matrix based on a dynamic key to generate a two-dimensional data carrier.

[0060] Specifically, based on a dynamic key, a two-dimensional linear automorphic scrambling transformation is performed on the fault-tolerant coding matrix to establish a linear mapping relationship between pixel positions, generating an encrypted data matrix exhibiting a noise-like form, which serves as a two-dimensional data carrier and a noise-like anti-interference basis. The row and column coordinates of the elements in the fault-tolerant coding matrix directly correspond to the spatial positions of image pixels; that is, the positions of elements in the fault-tolerant coding matrix can be considered as pixel positions.

[0061] According to an embodiment of this application, a two-dimensional linear automorphic scrambling transformation is performed on the fault-tolerant coding matrix, and the mapping equation of its iterative transformation can be shown in the following formula (1).

[0062] (1);

[0063] in, This represents the coordinates of an element in the fault-tolerant coding matrix. Indicates the mapped target coordinates. The parameter represents the order of the fault-tolerant coding matrix. and number of iterations The key is a dynamic key; where the number of iterations is... The configuration causes the spatial autocorrelation of the image to decay to a noise-like state.

[0064] Based on the above formula (1), a two-dimensional linear automorphic scrambling transformation is performed on the fault-tolerant coding matrix, that is, the fault-tolerant coding matrix is ​​subjected to the above formula (1). The next iteration generates a two-dimensional data carrier, which is also a two-dimensional matrix.

[0065] Therefore, based on the dynamic key, a two-dimensional linear automorphic scrambling transformation is performed on the fault-tolerant coding matrix, that is, a full-dimensional interleaving scrambling iteration is performed on the fault-tolerant coding matrix. By destroying the spatial domain correlation of the data, the burst errors that may occur in the channel transmission can be dispersed into random errors during the extraction process.

[0066] Specifically, based on the mapping relationship shown in formula (1), a two-dimensional linear self-similar scrambling transformation is performed on the fault-tolerant coding matrix. The spatial distribution of each coding element in the fault-tolerant coding matrix is ​​changed through iteration to reduce the correlation between adjacent elements. Thus, even if burst errors occur during channel transmission, after the inverse scrambling transformation during the extraction process, the burst errors can be dispersed into discrete errors distributed throughout the overall data range, thereby improving the error correction capability of the fault-tolerant coding matrix and ensuring the reliability of data transmission.

[0067] In operation S220, the original audio signal is decomposed into multiple scales and subjected to discrete cosine transform to obtain a two-dimensional frequency domain coefficient matrix.

[0068] In one embodiment, the original audio signal is the audio signal to be transmitted. The watermark data to be embedded can be embedded into the original audio signal so that the embedded audio signal carries the watermark data for communication transmission.

[0069] Since the low-frequency baseband in the original audio signal is used to carry speech semantics, in order to avoid interference from the low-frequency baseband that carries speech semantics, the original audio signal can be decomposed into multiple scales to separate the low and high frequencies of the original audio signal. Then, the high-frequency components can be used to covertly transmit the watermark data to be embedded.

[0070] Based on this, a discrete cosine transform is performed on the high-frequency components to obtain a two-dimensional frequency domain coefficient matrix, which can then be used to embed watermark data into the two-dimensional frequency domain coefficient matrix.

[0071] In operation S230, the preset high-frequency region of the coefficients in each row of the two-dimensional frequency domain coefficient matrix is ​​determined as the embedding region of each row of the two-dimensional frequency domain coefficient matrix, and the average amplitude of the coefficients in each embedding region is calculated as a statistical steady-state energy characteristic.

[0072] In one embodiment, since the effective signal energy (i.e. the energy representing the audio information itself) in the two-dimensional frequency domain coefficient matrix obtained after discrete cosine transform is highly concentrated in the low-frequency region, the preset high-frequency region of each row of coefficients in the two-dimensional frequency domain coefficient matrix can be determined as the embedding region of each row of the two-dimensional frequency domain coefficient matrix. This allows the watermark data to be embedded to be embedded into the embedding region of the two-dimensional frequency domain coefficient matrix, thereby achieving the embedding of watermark data into the original audio signal without significantly affecting the sound quality, thus improving the concealment of information embedding.

[0073] Since noise can overwhelm embedded information in low signal-to-noise ratio environments, statistical steady-state energy features are extracted from the embedded region to reduce the masking effect of noise on embedded information and improve the anti-interference capability against random noise interference.

[0074] Specifically, in low signal-to-noise ratio environments, the additive white Gaussian noise superimposed in the channel usually follows a zero-mean distribution, and the watermark data to be embedded is embedded in the embedding region of the two-dimensional frequency domain coefficient matrix. Thus, by taking the arithmetic mean of the amplitudes of the coefficients in the embedding region, the noise components can cancel each other out during the summation process, thereby obtaining the robust features after statistical denoising, i.e., the statistical steady-state energy features, to effectively prevent weak feature signals from being submerged by environmental noise, specifically, to prevent the feature signals of the watermark data embedded in the embedding region from being submerged by noise.

[0075] In operation S240, the quantization step size is adjusted based on the statistical steady-state energy characteristics of each row in the two-dimensional frequency domain coefficient matrix. Based on the statistical steady-state energy characteristics and quantization step size of each row in the two-dimensional frequency domain coefficient matrix, the encoded elements corresponding to each row in the two-dimensional data carrier are embedded into the embedding regions of each row in the two-dimensional frequency domain coefficient matrix to obtain the feature coefficient matrix of the watermarked data.

[0076] According to embodiments of this application, in order to solve the gain mismatch problem caused by channel amplitude fading, the quantization step size is dynamically adjusted based on the statistical steady-state energy characteristics of each row in the two-dimensional frequency domain coefficient matrix. This establishes an adaptive gain control mechanism where the quantization step size changes with the statistical steady-state energy characteristics to resist amplitude fading. Amplitude fading refers to the sudden decrease, attenuation, or weakening of the overall or local amplitude (i.e., volume or energy) of an audio signal during transmission or processing, which can easily lead to unreliable extraction of embedded watermark data.

[0077] Specifically, the gain characterizes the factor by which the modified amount is amplified or reduced when embedding watermarked data, and is used to control the strength of the watermark embedding. If the gain is larger, the watermark is stronger and has better anti-interference; if the gain is smaller, it is more concealed, but is easily drowned out by noise. Therefore, through an adaptive gain control mechanism, the watermark embedding strength can be automatically and dynamically adjusted according to the real-time strength of the audio signal, the noise level, channel fading, etc., so that the embedding strength can be appropriately reduced when the signal is strong, and the watermark gain can be automatically increased when the signal is weak or fading occurs, thereby effectively resisting the signal amplitude fading.

[0078] Based on this, according to the statistical steady-state energy characteristics and quantization step size of each row in the two-dimensional frequency domain coefficient matrix, the coding elements corresponding to each row in the two-dimensional data carrier are embedded into the embedding regions of each row in the two-dimensional frequency domain coefficient matrix to obtain the feature coefficient matrix of the watermarked data. Specifically, based on the coding elements in the two-dimensional data carrier and the quantization step size, the statistical steady-state energy characteristics of each row in the two-dimensional frequency domain coefficient matrix are modulated, and the coefficients in the embedding region are compensated based on the modulated energy characteristics. This realizes the mapping of the hidden information carried by the two-dimensional data carrier (i.e., the watermarked data to be embedded) to a small offset in the spectral energy of the original audio signal (i.e., the feature coefficient matrix of the obtained watermarked data has a small offset from the coefficients of each row in the two-dimensional frequency domain coefficient matrix), thereby realizing the embedding of the two-dimensional data carrier into the two-dimensional frequency domain coefficient matrix.

[0079] In operation S250, the audio signal containing the watermarked data is reconstructed based on the feature coefficient matrix of the watermarked data.

[0080] According to an embodiment of this application, the feature coefficient matrix of the watermarked data carries the watermarked data. However, since the final transmitted signal should be an audio signal, the audio signal of the watermarked data can be reconstructed based on the feature coefficient matrix of the watermarked data, and then transmitted based on the audio signal of the watermarked data.

[0081] According to embodiments of this application, the watermark data to be embedded is mapped to a fault-tolerant coding matrix to utilize its unique fault-tolerant mechanism to combat low signal-to-noise ratio scenarios and reduce interference to the actual watermark data. A two-dimensional linear self-similar scrambling transformation is performed on the fault-tolerant code, realizing the construction of a fault-tolerant two-dimensional data carrier, enabling the conversion of burst shearing or packet loss in the channel into controllable random errors. Since the original audio signal undergoes multi-scale decomposition and discrete cosine transform, the resulting two-dimensional frequency domain coefficient matrix is ​​generated based on high-frequency components, effectively isolating the influence of low-frequency baseband interference. Furthermore, since the effective signal energy in the two-dimensional frequency domain coefficient matrix is ​​highly concentrated in the low-frequency region, the preset high-frequency region of each row of coefficients in the two-dimensional frequency domain coefficient matrix is ​​determined as the embedding region, achieving embedding without significantly affecting sound quality and improving the concealment of information embedding. By calculating the average amplitude of the coefficients within the embedded region instead of using single-point coefficients for modulation, the resulting statistical steady-state energy characteristics cancel out the zero-mean additive white Gaussian noise. Furthermore, due to the ability to work with an adaptive quantization step size mechanism dynamically bound to the statistical steady-state energy characteristics, it possesses automatic gain control capabilities, enabling immunity to amplitude fading during channel transmission. Thus, while ensuring auditory perception transparency, it achieves highly robust and covert transmission of watermarked data embedded in the audio signal under harsh electromagnetic environments or high-compression links.

[0082] In one embodiment, the watermark data to be embedded can be any of the following: numerical values, text, images, etc. The watermark data to be embedded can also be fused information, such as a fusion of text and images. The following example, using a fusion of text and images as the watermark data to be embedded, illustrates how to map the watermark data to be embedded into a fault-tolerant coding matrix.

[0083] According to an embodiment of this application, the watermark data to be embedded includes digital identification information and visual authentication patterns; mapping the watermark data to be embedded into a fault-tolerant coding matrix includes: performing fault-tolerant redundancy coding on the digital identification information to obtain a two-dimensional matrix data structure; performing feature normalization on the visual authentication patterns, and mapping and fusing the feature-normalized visual authentication patterns into the non-critical data areas of the two-dimensional matrix data structure to obtain a fault-tolerant coding matrix.

[0084] Among them, the non-critical data area represents the area in the two-dimensional matrix data structure that does not affect the integrity of the digital identification information.

[0085] In one embodiment, the digital identification information can be represented by numbers or text, and the visual authentication pattern can be represented by an image.

[0086] According to an embodiment of this application, when the watermark data to be embedded includes digital identification information and visual authentication patterns, the digital identification information can first be encoded with fault-tolerant redundancy to obtain a two-dimensional matrix data structure. This two-dimensional matrix data structure can be a two-dimensional matrix based on QR codes to carry the digital identification information.

[0087] Specifically, the digital identification information can be encoded with fault tolerance and redundancy, for example, by processing the digital identification information to compress it into a two-dimensional matrix.

[0088] Based on this, since the two-dimensional matrix based on QR codes has certain fault tolerance and redundancy characteristics, the visual authentication pattern can be mapped and integrated into the non-critical data area of ​​the two-dimensional matrix data structure. This way, while integrating the visual authentication pattern with the digital identification information, the integrity of the visual authentication pattern is avoided.

[0089] However, visual authentication images have inconsistencies in pixel size and grayscale distribution. Directly mapping and fusing visual authentication images into a two-dimensional matrix data structure can easily interfere with subsequent scrambling processes. Therefore, before mapping and fusing visual authentication images into a two-dimensional matrix data structure, feature normalization is performed on the visual authentication images, and the feature-normalized visual authentication images are then mapped and fused into the two-dimensional matrix data structure to generate a fault-tolerant coding matrix.

[0090] According to an embodiment of this application, when the watermark data to be embedded includes digital identification information and visual authentication patterns, the visual authentication patterns and digital identification information are fused together and then embedded into the audio signal so that the watermark data embedded in the audio signal can be verified by dual verification of the visual authentication patterns and digital identification information to see if it is interfered with.

[0091] Based on the above, when the watermark data to be embedded is any one of numerical values, text, or images, the watermark data to be embedded can be directly subjected to fault-tolerant redundancy encoding to obtain a fault-tolerant encoding matrix, which is a two-dimensional matrix based on QR codes.

[0092] According to an embodiment of this application, a two-dimensional frequency domain coefficient matrix is ​​obtained by performing multi-scale decomposition and discrete cosine transform on the original audio signal, including: performing multi-scale decomposition on the original audio signal through discrete wavelet transform to obtain high-frequency detail components; segmenting the high-frequency detail components based on a preset segment length to obtain a wavelet domain coefficient matrix; and performing discrete cosine transform on each row of the wavelet domain coefficient matrix in sequence to obtain a two-dimensional frequency domain coefficient matrix.

[0093] The number of coefficients in each row of the wavelet domain coefficient matrix is ​​determined according to the preset segment length.

[0094] In one embodiment, the original audio signal is decomposed into multiple scales, for example, by using discrete wavelet transform.

[0095] Specifically, an orthogonal wavelet basis is selected as the decomposition kernel to perform single-level discrete wavelet decomposition on the original audio signal, in order to extract low-frequency approximate components and high-frequency detail components from the mixed spectrum of the original audio signal. Among them, the orthogonal wavelet basis (Haar wavelet basis) has compact support properties.

[0096] In one embodiment, the multi-scale decomposition of the original audio signal by discrete wavelet transform can be achieved by the following formula (2).

[0097] (2);

[0098] in, Represents the original audio signal. Represents the low-frequency approximate component. Represents high-frequency detail components. This indicates the original audio signal. Perform discrete wavelet transform.

[0099] Since the low-frequency approximation component carries speech semantics, the low-frequency approximation component and high-frequency detail component in the mixed spectrum of the original audio signal are separated, and the high-frequency detail component is selected as the physical channel for covert transmission, so that the influence of the low-frequency approximation component carrying speech semantics on the embedded watermark data can be avoided.

[0100] In one embodiment, the high-frequency detail component is a set of numerical coefficients, each representing the intensity of the audio signal at a corresponding time-frequency. Specifically, the high-frequency detail component is a one-dimensional vector.

[0101] In one embodiment, the audio signal also contains non-stationary noise, which changes over time but can be approximated as constant over a very short time. Therefore, the high-frequency detail components can be segmented to divide a "non-stationary" segment into many small "approximately stationary" segments, and each segment can be processed to effectively cancel the non-stationary noise.

[0102] Based on this, since high-frequency detail components are selected as the transmission channel for the watermark data to be embedded, and there is non-stationary noise in the high-frequency detail components, the high-frequency detail components can be segmented based on a preset segment length to obtain a wavelet domain coefficient matrix. This segmentation of the high-frequency detail components can isolate the influence of non-stationary noise, so as to avoid interference from non-stationary noise to the watermark data that needs to be transmitted covertly.

[0103] In one embodiment, the extracted high-frequency detail components are separated according to a preset segment length L. The data is segmented into sets of m independent units, that is, the high-frequency detail components are segmented according to a preset segment length L. Divide the matrix into m segments, each segment having a length of L. These m segments form a wavelet domain coefficient matrix, where the number of coefficients in each row of the wavelet domain coefficient matrix is ​​L, and the number of coefficients in each column of the wavelet domain coefficient matrix is ​​m. In other words, the wavelet domain coefficient matrix is ​​a matrix with L columns and m rows.

[0104] For example, if the length of the high-frequency detail component is 512 and the preset segment length L is 16, then the high-frequency detail component can be segmented according to the length of 16. Thus, the resulting two-dimensional frequency domain coefficient matrix is ​​a matrix with 16 columns and 64 rows.

[0105] The preset segment length can be determined based on the amount of information in the watermark data to be embedded. Specifically, when the original audio signal duration is fixed and the amount of information in the watermark data to be embedded is the same, a larger preset segment length allows a single segment to carry more information, thus requiring a shorter audio signal duration to embed the same amount of watermark data; conversely, a smaller preset segment length allows a single segment to carry less information, thus requiring a longer audio signal duration to embed the same amount of watermark data.

[0106] In one embodiment, after the original audio signal undergoes discrete wavelet transform, the timing information of the high-frequency detail components can be automatically preserved; performing discrete cosine transform on each row of the wavelet domain coefficient matrix in sequence, that is, performing discrete cosine transform on the segments of the high-frequency detail components in sequence, can also automatically preserve the timing information.

[0107] For example, the segments of the high-frequency detail components are: [0]、 [1] [2]..., the subscript numbers [0], [1], [2], etc. can be used to represent time. Since the discrete cosine transform is performed on each row of the wavelet domain coefficient matrix in sequence, the matrix row direction still retains the temporal order.

[0108] Among them, the high-frequency detail components contain time information, and the arrangement of coefficients in the high-frequency detail components corresponds to the time sequence of the audio signal.

[0109] In one embodiment, discrete cosine transform is performed sequentially on each row of the wavelet domain coefficient matrix to remove redundant correlations between coefficients. This concentrates the effective energy of each row in the wavelet domain coefficient matrix into the low-frequency region on the left side of the matrix, achieving a compact energy distribution that facilitates subsequent information embedding. Thus, in the two-dimensional frequency domain coefficient matrix obtained after the wavelet domain coefficient matrix undergoes discrete cosine transform, the high-frequency region is located on the right side of the matrix. Furthermore, the coefficients in the high-frequency region are generally small and have high redundancy, making the high-frequency region suitable for embedding watermark data.

[0110] Since a two-dimensional frequency domain coefficient matrix is ​​obtained by sequentially performing discrete cosine transform on each row of the wavelet domain coefficient matrix, the elements in the two-dimensional frequency domain coefficient matrix represent the transform coefficients of the corresponding cosine basis functions, and the elements have positive and negative signs.

[0111] According to embodiments of this application, since the low-frequency approximate components of the original audio signal carry speech semantics, the high-frequency detail components of the original audio signal have an auditory masking effect. The original audio signal can be decomposed at multiple scales to extract the high-frequency detail components from the mixed spectrum of the original audio signal. These high-frequency detail components can then be used as a covert transmission channel, enabling the construction of a covert barrier using the auditory masking effect of the high-frequency detail components to hide the embedding traces of watermark data. Furthermore, based on a preset segment length, the high-frequency detail components are segmented to perform granular slicing, thereby physically isolating strong impulse noise (i.e., non-stationary noise) at a single time point within a local micro-segment (i.e., a row in the wavelet domain coefficient matrix), preventing its energy diffusion and contamination of the overall link's signal-to-noise ratio. Thus, by performing multi-scale decomposition and discrete cosine transform on the original audio signal, the resulting two-dimensional frequency domain coefficient matrix can effectively isolate low-frequency baseband interference from non-stationary noise.

[0112] According to an embodiment of this application, determining a preset high-frequency region of the coefficients in each row of a two-dimensional frequency domain coefficient matrix as the embedding region of each row of the two-dimensional frequency domain coefficient matrix includes: determining a preset high-frequency region located at the end of each row of coefficients in the two-dimensional frequency domain coefficient matrix based on a preset embedding bit; and determining the preset high-frequency region of each row of coefficients as the embedding region of each row of the two-dimensional frequency domain coefficient matrix.

[0113] The number of coefficients in the preset high-frequency region is determined based on the preset embedding bit depth.

[0114] According to embodiments of this application, since the coefficients located at the right end of the two-dimensional frequency domain coefficient matrix correspond to high-frequency information, their impact on the listening experience is relatively small compared to low frequencies, and their impact on the perception effect is not significant. Therefore, the regions located at the ends of each row in the two-dimensional frequency domain coefficient matrix are more suitable for embedding watermark data. Based on this, a preset high-frequency region is determined at the ends of each row of the coefficients in the two-dimensional frequency domain coefficient matrix. Specifically, the regions located at the ends of each row in the two-dimensional frequency domain coefficient matrix that correspond to a preset number of embedding bits are selected as the preset high-frequency region.

[0115] For example, in a preset embedding bit depth of 1 In the case of selecting the coefficients at the end of a certain row in the two-dimensional frequency domain coefficient matrix... The bit coefficient, and the The region of the bit coefficients is determined as the preset high-frequency region of this row, and the number of coefficients in the preset high-frequency region of this row is . .

[0116] According to embodiments of this application, a preset high-frequency region is selected from a two-dimensional frequency domain coefficient matrix as the embedding region based on a preset embedding bit depth, so that the watermark data to be embedded is subsequently embedded into the corresponding embedding region, thereby realizing the embedding of the watermark data into the audio signal. Furthermore, since a high-frequency region is selected as the embedding region, the embedded watermark data has little impact on the perception effect.

[0117] Based on the above, the statistical steady-state energy characteristics of each row in the two-dimensional frequency domain coefficient matrix can be obtained by the following formula (3).

[0118] (3);

[0119] in, Represents the first in the two-dimensional frequency domain coefficient matrix The statistical steady-state energy characteristics of the line, It is a positive integer. Indicates the preset embedding bit length. Indicates the preset segment length. Represents the first in the two-dimensional frequency domain coefficient matrix The first line The magnitude of each coefficient.

[0120] Based on this The range of values ​​for is ( ), so that by using formula (3), based on the second in the two-dimensional frequency domain coefficient matrix, The amplitude of each coefficient within the embedded region of the row is used to calculate the statistical steady-state energy characteristics. .

[0121] For example, in The value is 16. When the value is 4, the second value in the two-dimensional frequency domain coefficient matrix is... The embedded region of the row includes the first The 13th to 16th coefficients in the row, among which, The range of values ​​for is ( ).

[0122] According to an embodiment of this application, adjusting the quantization step size based on the statistical steady-state energy characteristics of each row in the two-dimensional frequency domain coefficient matrix includes: adjusting the quantization step size based on the statistical steady-state energy characteristics of each row in the two-dimensional frequency domain coefficient matrix according to a preset intensity coefficient, so as to obtain the quantization step size corresponding to the statistical steady-state energy characteristics of each row in the two-dimensional frequency domain coefficient matrix.

[0123] According to an embodiment of this application, the second frequency domain coefficient matrix in the two-dimensional frequency domain is... The quantization step size of a row can be obtained by the following formula (4).

[0124] (4);

[0125] in, This represents the preset intensity coefficient, i.e., the intensity coefficient of adaptive quantization; Represents the first in the two-dimensional frequency domain coefficient matrix Statistical steady-state energy characteristics of the line; Indicates the quantization step size.

[0126] Based on the above, according to the first... Statistical steady-state energy characteristics of lines The corresponding quantization step size can be dynamically determined using formula (4). This makes the quantization step size proportional to the signal energy (i.e., the statistical steady-state energy characteristic).

[0127] According to embodiments of this application, by making the quantization step size With signal energy By proportionally varying the quantization step size, an adaptive decision threshold can be constructed that varies with signal energy. When amplitude fading occurs during channel transmission, causing an overall decrease in signal energy, the quantization step size... The corresponding reduction keeps the relative position of the quantization decision interval stable, thereby improving the robustness to linear gain distortion under blind detection conditions.

[0128] The adaptive decision threshold refers to a dynamically adjusted decision threshold based on factors such as the location of the embedding region in the two-dimensional frequency domain coefficient matrix, signal energy, and the sign of the coefficients within the embedding region. This threshold is used to stably distinguish between watermarks and noise under different locations and noise intensities. The quantization decision interval refers to dividing a continuous numerical range into several fixed-width intervals, each interval corresponding to a decision result.

[0129] Figure 3 A flowchart illustrating the process of obtaining the feature coefficient matrix of watermarked data according to an embodiment of this application is shown.

[0130] like Figure 3 As shown, the method 300 includes operations S310 to S340.

[0131] In operation S310, at least one coding element in the two-dimensional data carrier is determined that corresponds to each row of the two-dimensional frequency domain coefficient matrix.

[0132] The number of at least one encoded element is determined according to a preset embedding bit depth.

[0133] In one embodiment, for each row of the embedding region in the two-dimensional frequency domain coefficient matrix, at least one coded element at the corresponding position in the two-dimensional data carrier can be embedded into the corresponding embedding region. For example, within a preset embedding bit depth... When the value is 4, the 4-bit encoded element at the corresponding position in the two-dimensional data carrier can be embedded into the embedding region of the corresponding row, and the number of coefficients included in the embedding region is also 4.

[0134] Specifically, the encoded elements in the two-dimensional data carrier (i.e., a two-dimensional matrix) are embedded according to a preset number of bits. This partitioning allows the two-dimensional data carrier to be divided into m independent units. Therefore, for the m-th element in the two-dimensional frequency domain coefficient matrix... The embedded region of the row, which corresponds to the first row of the two-dimensional data carrier. The first independent unit, then for the first... embedded region of the row The bit coefficient, the first bit of the two-dimensional data carrier In each independent unit Bit-coded elements need to be embedded into it sequentially.

[0135] In operation S320, based on the quantization step size and at least one coding element corresponding to each row of the two-dimensional frequency domain coefficient matrix, the statistical steady-state energy features corresponding to each row of the two-dimensional frequency domain coefficient matrix are modulated into target energy features corresponding to at least one coding element corresponding to each row of the two-dimensional frequency domain coefficient matrix.

[0136] According to an embodiment of this application, the second coefficient in the two-dimensional frequency domain coefficient matrix can be obtained by the following formula (5). Statistical steady-state energy characteristics of lines The modulation is based on the target energy features corresponding to at least one coded element at a corresponding position in the two-dimensional data carrier, that is, the target energy features corresponding to the first coded element in the two-dimensional frequency domain coefficient matrix. The target energy features corresponding to each coefficient within the embedded region of the row.

[0137] (5);

[0138] Among them, for the first in the two-dimensional frequency domain coefficient matrix OK, Represents the relationship between the two-dimensional data carrier and the first The encoded element at the corresponding position of the embedded region of the row. Represents the second frequency domain coefficient matrix. The quantization step size of the row, Represents the second frequency domain coefficient matrix. The first row located in the embedded region The target energy characteristics corresponding to each coefficient.

[0139] For example, in the preset embedding bit length In the case of 4, based on the second-dimensional data carrier... The first coding element among the four elements in each independent unit can be used to obtain the target energy feature corresponding to the coding element based on the value of the coding element through the above formula (5).

[0140] Among them, the first in the two-dimensional frequency domain coefficient matrix The number of target energy features corresponding to the embedding region of the row is the same as the preset embedding bit number.

[0141] In operation S330, the difference between the target energy feature and the statistical steady-state energy feature corresponding to at least one coding element in each row of the two-dimensional frequency domain coefficient matrix is ​​calculated as the difference in mean energy feature.

[0142] In one embodiment, the first coefficient of the two-dimensional frequency domain coefficient matrix can be calculated using the following formula (6). The first row located in the embedded region Differences in mean energy characteristics before and after modulation of each coefficient .

[0143] (6);

[0144] Among them, for the two-dimensional frequency domain coefficient matrix, the first Line has a unique statistical steady-state energy characteristic. The second-dimensional frequency domain coefficient matrix The embedded region of the row includes The digit coefficient, then the digit coefficient The embedded region of a row is used to embed the corresponding position in the two-dimensional data carrier. For the bit-coded element, then for the two-dimensional frequency domain coefficient matrix, the first bit... Yes, it has The modulated target energy characteristics The range of values ​​for is ( ).

[0145] In operation S340, based on the positive and negative signs of the coefficients in the embedding region of each row of the two-dimensional frequency domain coefficient matrix, the coefficients in the embedding region of each row of the two-dimensional frequency domain coefficient matrix are compensated by the difference in mean energy characteristics to obtain the feature coefficient matrix of the watermarked data.

[0146] Since the coefficients in the embedded regions of each row of the two-dimensional frequency domain coefficient matrix have positive and negative signs, phase-preserving energy compensation is performed to prevent auditory abnormalities caused by phase reversal due to energy injection.

[0147] Specifically, a phase-locked constraint mechanism is used to compensate for the coefficients in the corresponding embedding region along the original polarity direction by using the difference in mean energy characteristics. The phase-locked constraint mechanism is used to ensure that the coefficients do not undergo zero-crossing flips before and after modulation, thereby eliminating high-frequency pops caused by phase abrupt changes.

[0148] In one embodiment, the positive and negative polarity signs of the coefficients in the embedded regions of each row of the two-dimensional frequency domain coefficient matrix are obtained. Based on the polarity guiding operator (i.e., the positive and negative polarity signs of the coefficients), the coefficients in the embedded regions of each row of the two-dimensional frequency domain coefficient matrix are uniformly compensated by the difference value of the mean energy characteristics, so as to keep the phase polarity of the spectral coefficients unchanged while changing the amplitude of the statistical energy (i.e., the coefficients in the embedded regions of the two-dimensional frequency domain coefficient matrix).

[0149] According to an embodiment of this application, the compensation formula based on the difference in mean energy characteristics is shown in the following formula (7).

[0150] (7);

[0151] in, Represents the first in the two-dimensional frequency domain coefficient matrix The first row located in the embedded region One coefficient; Represents the first in the two-dimensional frequency domain coefficient matrix The row located in the embedded region after compensation of the first One coefficient; This refers to a signal phase-locked constraint mechanism, which guides energy to always be injected along the polarity direction of the original signal, thereby strictly maintaining the phase polarity of the spectral coefficients while adjusting the energy amplitude, eliminating high-frequency pops caused by phase abrupt changes, and ensuring the auditory perception transparency of audio containing hidden information.

[0152] Therefore, by using the above formulas (5) to (7), it is possible to embed the two-dimensional data carrier into the two-dimensional frequency domain coefficient matrix to obtain the feature coefficient matrix of the watermarked data.

[0153] According to an embodiment of this application, a phase-locked constraint mechanism is used to perform phase-preserving energy compensation on the coefficients in the embedded regions of each row of the two-dimensional frequency domain coefficient matrix to prevent spectral phase flipping, thereby ensuring that the spectral coefficients do not undergo zero-crossing flipping before and after modulation, and thus eliminating high-frequency popping caused by phase abrupt changes.

[0154] According to an embodiment of this application, an audio signal of watermarked data is reconstructed based on the feature coefficient matrix of the watermarked data, including: performing inverse discrete cosine transform on each row of the feature coefficient matrix of the watermarked data in sequence to obtain the coefficient matrix after inverse transform; concatenating and splicing each row of the coefficient matrix after inverse transform according to a preset segmentation order to obtain the high-frequency detail components of the watermarked data; and performing inverse discrete wavelet transform on the low-frequency approximation components and the high-frequency detail components of the watermarked data to reconstruct the audio signal of the watermarked data.

[0155] The low-frequency approximation component is obtained by performing a discrete wavelet transform on the original audio signal.

[0156] According to an embodiment of this application, since the watermark data to be embedded is embedded into the high-frequency detail components stripped from the original audio signal during the embedding process, and the audio signal is ultimately required for covert transmission, the feature coefficient matrix of the watermark data is reconstructed to obtain the audio signal containing the watermark data, thus realizing the embedding of the watermark data into the audio signal.

[0157] In one embodiment, during the embedding process, the original audio signal is subjected to discrete wavelet transform to obtain high-frequency detail components. These high-frequency detail components are then segmented to obtain a wavelet domain coefficient matrix. Each row of the wavelet domain coefficient matrix is ​​then sequentially subjected to discrete cosine transform to obtain a two-dimensional frequency domain coefficient matrix for embedding the watermark data. Based on this, to reconstruct the audio signal, each row of the feature coefficient matrix containing the watermark data is sequentially subjected to inverse discrete cosine transform to obtain an inverse transform coefficient matrix. Following a preset segmentation order, each row of the inverse transform coefficient matrix is ​​concatenated and spliced ​​to obtain the high-frequency detail components of the watermark data. Finally, an inverse discrete wavelet transform is performed on the low-frequency approximation components and the high-frequency detail components of the watermark data to reconstruct the audio signal containing the watermark data.

[0158] Since the timing information is automatically preserved during the discrete wavelet transform, segmentation, and discrete cosine transform of the original audio signal, the rows of the coefficient matrix after the inverse transform can be concatenated and spliced ​​based on the timing information of this stage (i.e., the preset segmentation order) to obtain the high-frequency detail components of the watermarked data, thus constructing a complete high-frequency detail component that carries the embedded watermark data (i.e., hidden information).

[0159] The preset segmentation order is determined when segmenting the high-frequency detail components. Therefore, the order in which the rows of the coefficient matrix after the inverse transformation are concatenated and spliced ​​is consistent with the segmentation order.

[0160] When performing multi-scale decomposition on the original audio signal, not only can high-frequency detail components be extracted, but also low-frequency approximation components can be extracted. These low-frequency approximation components are used to carry speech semantics. Then, the low-frequency approximation components and the high-frequency detail components of the watermarked data can be subjected to inverse discrete wavelet transform to fuse the low-frequency approximation components with the high-frequency detail components of the watermarked data, thereby obtaining the audio signal of the watermarked data.

[0161] In one embodiment, the audio signal containing the watermark data can be reconstructed using the following formula (8).

[0162] (8);

[0163] in, This represents the high-frequency detail components of the watermarked data. The audio signal representing the watermarked data. This represents the inverse discrete wavelet transform.

[0164] In one embodiment, during the inverse cascade reconstruction of the high-frequency detail components carrying covert information into an audio signal, a spectral envelope smoothing method can also be used to perform phase-preserving energy diffusion to maintain the continuity of the spectral envelope, thereby generating an audio signal containing covert information that can resist channel filtering and amplitude fading, i.e., generating an audio signal containing covert information that is transparent to auditory perception.

[0165] According to embodiments of this application, during the reconstruction of the audio signal containing watermarked data, the impact on the perceived quality of the original audio signal is reduced and robustness to channel filtering and amplitude fading is improved by maintaining the amplitude relationship and structural continuity between the components. Furthermore, due to the reconstructed audio signal... Maintaining the smoothness of the spectral envelope in the time-frequency domain can better resist the truncation effect of channel filters and ensure the reliable transmission of covert information.

[0166] Figure 4 A flowchart illustrating an audio noise-resistant steganography method for low signal-to-noise ratio environments according to an embodiment of this application is shown.

[0167] like Figure 4 As shown, audio noise-resistant steganography refers to embedding the watermark data to be embedded into the original audio signal to achieve covert transmission.

[0168] Specifically, the construction of a fault-tolerant carrier based on full-dimensional interleaving refers to encoding the watermark data to be embedded to obtain a fault-tolerant coding matrix, which is a highly fault-tolerant and error-correcting two-dimensional coding matrix; and then performing full-dimensional interleaving scrambling iteration on the fault-tolerant coding matrix to obtain a two-dimensional data carrier, which is the encryption scrambling base. For example, the construction of a fault-tolerant carrier based on full-dimensional interleaving can be achieved by executing the above operation S210.

[0169] High-frequency subband extraction based on multi-scale decomposition refers to performing multi-scale decomposition on the original audio signal to extract high-frequency detail components, and then performing piecewise segmentation and discrete cosine transform on these components to obtain a two-dimensional frequency domain coefficient matrix, which carries the watermark data. For example, high-frequency subband extraction based on multi-scale decomposition can be achieved by performing the above operation S220. The coefficients of the high-frequency subbands constitute a wavelet domain coefficient matrix.

[0170] Steady-state energy quantization modulation based on adaptive gain control refers to a method that, to combat random noise interference in low signal-to-noise ratio environments, selects an embedding region within a two-dimensional frequency domain coefficient matrix to embed concealed information, and calculates the arithmetic mean of the amplitudes of the coefficients within the embedding region as a statistical steady-state energy characteristic, thereby canceling out noise components during the summation process. To address the gain mismatch problem caused by channel amplitude fading, the quantization step size is dynamically adjusted based on the statistical steady-state energy characteristics of the embedding region, establishing an adaptive gain control mechanism where the quantization step size changes with the statistical steady-state energy characteristics to combat amplitude fading.

[0171] Based on this, according to the encoded elements in the two-dimensional data carrier and the quantization step size, the statistical steady-state energy characteristics of the two-dimensional frequency domain coefficient matrix are modulated to obtain the modulated target energy characteristics. Then, based on the phase-locked constraint mechanism, according to the positive and negative signs of the coefficients within the embedding region, the target energy characteristics are used to compensate the coefficients in the corresponding embedding region, thus realizing the embedding of the two-dimensional data carrier into the two-dimensional frequency domain coefficient matrix. Here, the positive and negative signs of the coefficients represent the phase-locked modulation coefficients.

[0172] For example, steady-state energy quantization modulation based on adaptive gain control can be achieved by executing the above operations S230 and S240.

[0173] Feature-to-audio signal reconstruction based on inverse cascaded transform refers to performing inverse discrete cosine transform on each row of the feature coefficient matrix of the watermarked data, and then cascading and concatenating the rows of the transformed coefficient matrix according to a preset segmentation order to obtain the high-frequency detail components of the watermarked data. Inverse discrete wavelet transform is then performed on the low-frequency approximation components and the high-frequency detail components of the watermarked data to reconstruct the audio signal of the watermarked data. The high-frequency detail components of the watermarked data are used to carry hidden information, and the audio signal of the watermarked data is auditorily transparent protective audio.

[0174] For example, by performing the above operation S250, the reconstruction of features into audio signals based on inverse cascade transformation can be achieved.

[0175] According to an embodiment of this application, a blind information extraction method is also provided, comprising: blindly extracting an audio signal containing watermarked data to obtain a composite verification substrate. The composite verification substrate is used for watermark verification of the audio signal containing the watermarked data.

[0176] The audio signal containing the watermark data is obtained through the above-mentioned audio noise-resistant steganography method for low signal-to-noise ratio environments. Specifically, the audio signal containing the watermark data can be obtained through the above-mentioned operations S210 to S250.

[0177] According to embodiments of this application, when the embedded watermark data includes digital identification information and visual authentication patterns, the composite verification substrate characterizes the fusion of digital identification information and visual authentication patterns.

[0178] In one embodiment, blind extraction of the audio signal of the watermarked data refers to performing an inverse transformation on the audio signal of the watermarked data to obtain a composite verification substrate.

[0179] According to an embodiment of this application, since the visual authentication pattern has been fused with the QR code (Quick Response Code, i.e., a two-dimensional matrix data structure) before embedding, the extracted result is a composite verification base. Furthermore, non-separate hybrid verification can be performed based on the composite verification base.

[0180] Specifically, on the one hand, the composite verification substrate can be directly output for manual verification, allowing for intuitive identity verification by checking the clarity of the visual authentication pattern; on the other hand, the composite verification substrate can also be input into a QR code decoder for decoding. Utilizing the error-tolerant characteristics of QR codes, the decoder can identify and automatically ignore random errors scattered after inverse transformation, thus accurately parsing the watermark data even in the presence of errors, achieving dual traceability authentication through "machine error-tolerant decoding" and "human visual verification".

[0181] According to the embodiments of this application, when the embedded watermark data includes digital identification information and visual authentication patterns, blind extraction of the audio signal containing the watermark data can extract a composite verification substrate, thereby achieving dual identity verification with machine readability and manual verification. This realizes data fault tolerance recovery and dual identity authentication, in order to verify whether noise and other interference in the channel affect the covertly transmitted watermark data during transmission.

[0182] Under the premise of blind extraction of the audio signal containing watermarked data, a feature extraction channel consistent with the embedding end (i.e., the audio noise-resistant steganography process) is constructed. To address the broadband noise or baseband interference that may be mixed into the audio signal to be identified (i.e., the audio signal containing watermarked data) during transmission, this embodiment does not directly demodulate the full-band signal, but performs precise spectrum stripping, that is, directly performs blind extraction based on the audio signal containing watermarked data, without involving the original audio signal.

[0183] Figure 5 A flowchart illustrating the process of obtaining a composite verification substrate according to an embodiment of this application is shown.

[0184] like Figure 5 As shown, the method 500 includes operations S510 to S540.

[0185] The following operations S510 to S540 are equivalent to the inverse transformations of the above operations S210 to S250.

[0186] In the S510, the audio signal of the watermarked data is decomposed into multiple scales and subjected to discrete cosine transform to obtain the two-dimensional frequency domain coefficient matrix of the watermarked data.

[0187] In one embodiment, the high-frequency component of the audio signal containing watermark data carries the watermark data, while the low-frequency component carries the speech semantics. Then, through multi-scale decomposition, the high-frequency detail component carrying the watermark data and the low-frequency approximate component carrying the speech semantics are extracted from the audio signal containing watermark data.

[0188] Based on this, the high-frequency detail components carrying the watermark data are sliced ​​according to the preset segment length to obtain the wavelet domain coefficient matrix of the watermark data, thereby accurately extracting the high-frequency detail components that may contain hidden information.

[0189] Specifically, the extraction process must be consistent with the embedding process. Using the isomorphic mapping mechanism, the audio signal to be identified is subjected to multi-scale decomposition and discrete cosine transform with parameters that are strictly consistent with those in the embedding process, so as to accurately lock the two-dimensional frequency domain coefficient matrix carrying the hidden information.

[0190] In one embodiment, a synchronous demodulation time slot is established based on a preset segment length, thereby constructing a time-frequency feature extraction channel that is fully aligned with the transmitting end at the physical level, ensuring the stability of the time-domain reference for subsequent statistical feature calculation.

[0191] Specifically, for the audio signal of watermarked data Perform a spectrum stripping operation identical to the embedding process. If the embedding process uses discrete wavelet transform, then the extraction process uses the db1 orthogonal wavelet basis (Daubechies 1 wavelet, i.e., Haar wavelet) to perform single-level discrete wavelet decomposition on the audio signal containing the watermark data, retaining only the high-frequency detail components carrying the watermark data. By utilizing the multi-resolution characteristics of wavelet transform, the mixed signal containing noise is decoupled, and the frequency domain location of the concealed channel is accurately located, thus constructing a physically isolated secure frequency band for subsequent feature extraction.

[0192] After locking the frequency band, the high-frequency detail components carrying the watermark data are... Segment the data. The segment length will be based on the preset segment length used during embedding. The high-frequency detail components extracted Discretization by A sequence set composed of individual units ensures the extraction of high-frequency detail components during the extraction process. The partitioning and embedding processes are strictly aligned to construct synchronous demodulation time slots, thereby providing a stable temporal benchmark for subsequent statistical feature extraction.

[0193] Since the same preset segment length is used for segmentation in both the embedding and extraction processes, a time-division multiplexing channel capable of isolating local strong impulse noise (i.e., non-stationary noise) is constructed.

[0194] In operation S520, based on the preset number of embedding bits, the embedding region of each row in the two-dimensional frequency domain coefficient matrix of the watermarked data is determined, and the average amplitude of the coefficients in each embedding region is calculated as the residual statistical steady-state energy feature.

[0195] In order to resist random noise interference in low signal-to-noise ratio environments during the embedding process, statistical steady-state energy features are extracted. In order to accurately restore the data in low signal-to-noise ratio environments, feature extraction based on statistical averaging is also performed during the extraction process.

[0196] In one embodiment, a two-dimensional frequency domain coefficient matrix of the watermarked data is obtained by performing a discrete cosine transform on the wavelet domain coefficient matrix of the watermarked data, wherein the coefficients in the two-dimensional frequency domain coefficient matrix of the watermarked data are slightly offset from those in the two-dimensional frequency domain coefficient matrix.

[0197] In one embodiment, based on the same preset embedding bit depth as during the embedding process. The embedding region is determined from the two-dimensional frequency domain coefficient matrix of the watermarked data. Specifically, the number of bits of the coefficients in the embedding region of each row of the two-dimensional frequency domain coefficient matrix of the watermarked data is a preset embedding bit number. Furthermore, the embedding region of each row in the two-dimensional frequency domain coefficient matrix of the watermarked data starts from the end of each row in the two-dimensional frequency domain coefficient matrix of the watermarked data, and a preset embedding bit depth is selected. Obtained.

[0198] Based on this, the average amplitude of the coefficients within the embedding region of each row in the two-dimensional frequency domain coefficient matrix of the watermarked data is calculated as a residual statistical steady-state energy characteristic. For example, the value of the first coefficient in the two-dimensional frequency domain coefficient matrix of the watermarked data can be calculated using the above formula (3). The residual statistical steady-state energy characteristics of the line.

[0199] Therefore, by calculating the residual statistical steady-state energy characteristics after channel attenuation... By utilizing the zero-mean statistical property of additive white Gaussian noise, the amplitudes of the coefficients within the embedding regions of each row in the two-dimensional frequency domain coefficient matrix of the watermarked data are averaged, causing the random noise components superimposed on the audio signal of the watermarked data to cancel each other out during the summation. Therefore, even if the audio signal is weak, the residual statistical steady-state energy characteristics... It can still accurately reflect the effective energy state within the embedded region, rather than instantaneous noise interference.

[0200] In operation S530, the recovery quantization step size is estimated based on the residual statistical steady-state energy characteristics of each row in the two-dimensional frequency domain coefficient matrix of the watermarked data. Based on the residual statistical steady-state energy characteristics and recovery quantization step size of each row in the two-dimensional frequency domain coefficient matrix of the watermarked data, the two-dimensional frequency domain coefficient matrix of the watermarked data is restored to a two-dimensional encryption matrix.

[0201] According to the embodiments of this application, in the embedding process, in view of the problem of overall fluctuation of signal energy caused by amplitude fading in the transmission channel, the decision method with fixed quantization step size is abandoned. In the extraction process, an adaptive recovery strategy based on the signal energy (i.e., residual statistical steady-state energy characteristics) of the audio signal with watermark data is adopted.

[0202] Specifically, based on the implementation method of transforming a two-dimensional data carrier into a two-dimensional frequency domain coefficient matrix during the embedding process, the recovery quantization step size can be estimated based on the residual statistical steady-state energy characteristics of each row of the embedding region in the two-dimensional frequency domain coefficient matrix of the watermarked data during the extraction process. Then, by using the residual statistical steady-state energy characteristics and the recovery quantization step size of each row of the embedding region in the two-dimensional frequency domain coefficient matrix of the watermarked data, the two-dimensional frequency domain coefficient matrix of the watermarked data is restored to the two-dimensional data carrier, resulting in a two-dimensional encryption matrix.

[0203] If the hidden watermark data is not affected during transmission, the resulting two-dimensional encryption matrix is ​​the two-dimensional data carrier; if the hidden watermark data is affected during transmission, the resulting two-dimensional encryption matrix differs from the two-dimensional data carrier and cannot be restored to the two-dimensional data carrier.

[0204] In operation S540, based on the dynamic key of the embedding process, the two-dimensional encryption matrix is ​​subjected to an inverse transformation of two-dimensional linear automorphic scrambling to obtain a composite verification basis.

[0205] According to an embodiment of this application, during the embedding process, a two-dimensional linear self-similar scrambling transformation is performed on the fault-tolerant coding matrix based on a dynamic key to obtain a two-dimensional data carrier. During the extraction process, based on the dynamic key from the embedding process, an inverse two-dimensional linear self-similar scrambling transformation is performed on the two-dimensional encryption matrix to convert burst errors in the channel into discrete random errors. Its unique fault-tolerant mechanism isolates the impact of errors on the composite verification substrate (ignoring the impact of random errors on the watermark information), thereby outputting the composite verification substrate. This allows for dual authentication based on the visual authentication pattern and digital identification watermark information in the composite verification substrate.

[0206] Perform dual authentication on the two-dimensional encryption matrix, including performing reverse deinterleaving, then decoding the deinterleaved matrix, and outputting the text identifier and visual integrity pattern after fault-tolerant decoding for dual authentication.

[0207] Specifically, by using a private key (i.e., the dynamic key of the embedding process) to perform an inverse transformation of two-dimensional linear automorphic scrambling on a two-dimensional encryption matrix, local burst errors caused by signal shearing during channel transmission can be spatially dispersed and transformed into discrete random errors distributed across the entire graph, thereby recovering the composite verification basis and achieving fault-tolerant data recovery.

[0208] Based on the above, for a two-dimensional encryption matrix, spatial scrambling can be removed by inverse transformation of dimensional linear automorphism scrambling, thereby achieving reverse parsing and fault-tolerant recovery of the two-dimensional encryption matrix.

[0209] According to an embodiment of this application, a two-dimensional linear automorphic scrambling inverse transformation is performed on a two-dimensional encryption matrix, and the iterative mapping equation of the inverse transformation can be shown in the following formula (9).

[0210] (9);

[0211] in, The coordinates of the elements in the two-dimensional encryption matrix. These are the corresponding coordinates in the restored composite verification basis.

[0212] According to embodiments of this application, the audio signal of the watermarked data undergoes multi-scale decomposition and discrete cosine transform to accurately extract high-frequency detail components potentially containing hidden information. Based on this, to accurately reconstruct the data in a low signal-to-noise ratio environment, feature extraction based on statistical averaging is performed. This involves calculating the average amplitude of coefficients within each row of the embedded region in the two-dimensional frequency domain coefficient matrix of the watermarked data, using this as a residual statistical steady-state energy feature to cancel additive white Gaussian noise during summation. The recovery quantization step size is estimated based on the residual statistical steady-state energy features of each row of the embedded region in the two-dimensional frequency domain coefficient matrix of the watermarked data. This allows for the reconstruction of the two-dimensional data carrier based on the residual statistical steady-state energy features and the recovery quantization step size, resulting in a two-dimensional encryption matrix. The two-dimensional encryption matrix undergoes an inverse two-dimensional linear automorphic scrambling transformation. This inverse transformation disperses local burst errors occurring during channel transmission through geometric inverse mapping, distributing them throughout the matrix space and transforming them into discrete random errors. This breaks the continuity of errors and creates the necessary conditions for subsequent decoding. Therefore, by using fault tolerance mechanisms to ignore random errors, data fault recovery can be achieved.

[0213] According to an embodiment of this application, the recovery quantization step size is estimated based on the residual statistical steady-state energy characteristics of each row in the two-dimensional frequency domain coefficient matrix of the watermarked data, and the two-dimensional frequency domain coefficient matrix of the watermarked data is restored to a two-dimensional encryption matrix based on the residual statistical steady-state energy characteristics of each row in the two-dimensional frequency domain coefficient matrix of the watermarked data and the recovery quantization step size. This includes: estimating the recovery quantization step size based on the residual statistical steady-state energy characteristics of each row in the two-dimensional frequency domain coefficient matrix of the watermarked data according to a preset strength coefficient, obtaining the recovery quantization step size corresponding to the residual statistical steady-state energy characteristics of each row in the two-dimensional frequency domain coefficient matrix of the watermarked data respectively; determining at least one coding element corresponding to the embedding region of each row in the two-dimensional frequency domain coefficient matrix of the watermarked data according to the quantization interval range to which the residual statistical steady-state energy characteristics of each row in the two-dimensional frequency domain coefficient matrix of the watermarked data belong; and obtaining a two-dimensional encryption matrix based on the positional mapping relationship between the coding elements in the two-dimensional data carrier and the embedding regions of each row in the two-dimensional frequency domain coefficient matrix determined by the embedding process, according to at least one coding element corresponding to multiple embedding regions in the two-dimensional frequency domain coefficient matrix of the watermarked data respectively.

[0214] The quantization interval range corresponding to each row in the two-dimensional frequency domain coefficient matrix of the watermarked data is determined based on the recovery quantization step size of each row.

[0215] According to an embodiment of this application, the recovery quantization step size can be obtained by the following formula (10) based on the same preset strength coefficient as the embedding process.

[0216] (10); among which, This refers to the intensity coefficient based on the adaptive quantization preset at the embedded end, i.e., the preset intensity coefficient. Represents the residual statistical steady-state energy eigenvalue. This indicates the restoration of the quantization step size.

[0217] Therefore, using the above formula (10), the residual statistical steady-state energy eigenvalues ​​of each row in the two-dimensional frequency domain coefficient matrix of the watermarked data are utilized. The recovery quantization step size can be estimated. This enables adaptive following of changes in channel gain.

[0218] In one embodiment, based on residual statistical steady-state energy characteristics With recovery quantization step size The ratio relationship allows for blind-state decision-making and data bit reconstruction to obtain a two-dimensional encryption matrix.

[0219] Specifically, the blind decision logic can be represented by the following formula (11). The blind decision logic is used to achieve blind extraction without requiring the original audio reference.

[0220] (11);

[0221] in, The integer index corresponds to the nth row in the two-dimensional frequency domain coefficient matrix of the watermarked data. One coefficient; and This indicates the range of the quantization interval, which is the quantization decision interval.

[0222] Based on formula (11), the residual statistical steady-state energy characteristics are judged. The relative position within the quantization interval determines the encoded element at that position. .

[0223] Based on the principle of ratio invariance, the adaptive blind-state decision logic is established as follows: using the residual statistical steady-state energy characteristics, the recovery quantization step size is estimated in reverse, so that the decision threshold of the recovery can be scaled proportionally with the fluctuation of signal energy; by judging the relative position of the residual statistical steady-state energy characteristics in the quantization interval, adaptive gain control of channel amplitude fading is achieved without the need for the original audio reference, and the binarized data, i.e., the two-dimensional encryption matrix, is accurately restored.

[0224] In one embodiment, based on the positional mapping relationship between the encoded elements in the two-dimensional data carrier and the embedding regions of each row of the two-dimensional frequency domain coefficient matrix determined by the embedding process, at least one encoded element corresponding to each of the multiple embedding regions in the two-dimensional frequency domain coefficient matrix of the watermarked data is sequentially mapped to the corresponding position in the matrix to obtain a two-dimensional encryption matrix.

[0225] Among them, the position mapping relationship can characterize the mapping relationship between the row of the embedded region in the two-dimensional frequency domain coefficient matrix of the watermarked data and the position of at least one encoded element in the two-dimensional encryption matrix.

[0226] According to embodiments of this application, the residual statistical steady-state energy characteristics are used to inversely estimate the quantization step size, and the relative position of the residual statistical steady-state energy characteristics within the quantization interval is determined, enabling accurate reconstruction of the two-dimensional encryption matrix without the need for the original audio reference. Based on this, even if the audio signal experiences channel impairment during transmission, the residual statistical steady-state energy characteristics... and estimated recovery quantization step size When scaling is performed proportionally, the relative positions of the two ratios remain unchanged, demonstrating dynamic automatic gain control capabilities, which can accurately reconstruct the two-dimensional encryption matrix without requiring the original audio reference.

[0227] Figure 6 A flowchart of an information blind extraction method according to an embodiment of this application is shown.

[0228] like Figure 6 As shown, blind information extraction refers to the ability to extract watermark data from audio information containing watermark data without the need for the original audio signal reference.

[0229] Hidden feature channel alignment based on isomorphic mapping refers to performing the same multi-scale decomposition and discrete cosine transform as the embedding process on the audio signal of the watermarked data to obtain the two-dimensional frequency domain coefficient matrix of the watermarked data, so as to ensure that the number of rows and columns of the two-dimensional frequency domain coefficient matrix and the two-dimensional frequency domain coefficient matrix of the watermarked data are the same.

[0230] Blind-state coding recovery based on adaptive gain control refers to the process of accurately restoring data in low signal-to-noise ratio environments. This involves performing feature extraction based on statistical averaging. Specifically, embedding regions carrying hidden information are selected from each row of the two-dimensional frequency domain coefficient matrix of the watermarked data, and the arithmetic mean of the amplitudes of the coefficients within these regions is calculated as the residual statistical steady-state energy feature. The recovery quantization step size is then estimated based on this residual statistical steady-state energy feature. Using the residual statistical steady-state energy feature of each row of the two-dimensional frequency domain coefficient matrix of the watermarked data and the recovery quantization step size, the two-dimensional frequency domain coefficient matrix of the watermarked data is restored to a two-dimensional data carrier, resulting in a two-dimensional encryption matrix. For example, blind-state coding recovery based on adaptive gain control can be achieved by executing operations S520 and S530.

[0231] Dual authentication based on a composite verification basis using inverse deinterleaving refers to the process of performing an inverse transformation of a two-dimensional encryption matrix using a dynamic key embedded in the process, reconstructing the encryption scrambling basis, and obtaining a composite verification basis. The composite verification basis is then decoded, outputting a decoded visual authentication pattern and digital identification information. The digital identification information is fault-tolerant and repaired, while the visual authentication pattern serves as an auxiliary tool to achieve dual authentication.

[0232] Therefore, based on the decoded visual authentication pattern and digital identification information, the clarity of the visual authentication pattern is used to verify whether noise and other interference in the channel affect the covertly transmitted watermark data, thus obtaining the verification result.

[0233] To comprehensively evaluate the effectiveness of the proposed audio noise immunity steganography method for low signal-to-noise ratio channels in harsh communication environments, this application constructs a communication simulation environment that includes various channel impairment simulations.

[0234] In this embodiment, the experiment selected 7,000 mixed audio data points covering pop music, rock music, classical music, human voices (simulated through voice commands), and ambient sounds (simulated through background noise) as the communication carrier. The pop music, rock music, and classical music samples came from the GTZAN dataset (a music genre dataset), the human voice samples came from the sample_voice_data dataset, and the ambient sound samples came from the ESC-50 dataset (Environmental Sound Classification 50). The mixed audio data of pop music, rock music, classical music, human voices, and ambient sounds in the experiment were all selected with authorization.

[0235] In one embodiment, the experimental parameters are configured as follows: audio sampling rate of 22.05kHz; channel coding using H-level fault-tolerant QR code with a size of 128×128 pixels; scrambling key parameters (i.e., dynamic key). The channel is constructed using the db1 wavelet basis; the granular segment length is the preset segment length. ; Statistical feature locking region, i.e., preset embedding bit depth Adaptive quantization intensity coefficient, i.e., preset intensity coefficient .

[0236] First, the system's "channel concealment" (imperceptibility) is assessed. In covert communication missions, the sensory quality of the carrier signal determines how easily the communication behavior can be detected by hostile monitoring parties.

[0237] In one embodiment, the signal-to-noise ratio (SNR) and objective difference grade (ODG) for different carrier types can be shown in Table 1 below.

[0238] The subjective difference rating was calculated by using a double-blind listening test on 23 test subjects (12 boys and 11 girls), playing 10 sets of audio samples before and after processing, and collecting scores.

[0239] Table 1

[0240]

[0241] Based on Table 1, the average signal-to-noise ratio (SNR) of each type of audio exceeds 20 dB, and the subjective difference rating is extremely low (close to no difference). This indicates that the audio noise-resistant steganography method for low SNR environments proposed in this application successfully utilizes the frequency masking effect of the human ear by embedding data into high-frequency detail subbands and strictly controlling the embedding strength. This makes the signal containing hidden information audibly indistinguishable from the original signal, thereby establishing a covert communication link with an extremely low probability of interception.

[0242] In another embodiment, the focus is on evaluating the system's "communication reliability" (robustness). This experiment designs a composite attack scenario for poorly configured channels, using bit error rate as the core evaluation metric. To verify the technical advantages of this application, it is compared with the DWT-DCT (Discrete Wavelet Transform-Discrete Cosine Transform) algorithm model and the DWT-DCT-SVD (Discrete Wavelet Transform-Discrete Cosine Transform-Singular Value Decomposition) algorithm model. The DWT-DCT algorithm model is represented by baseline-1, and the DWT-DCT-SVD algorithm model is represented by baseline-2.

[0243] The channel impairment types simulated in the experiment can include the following: Amplitude scaling (simulating fading): reducing the audio amplitude to 80% of its original value; Sample suppression (simulating interruption): randomly setting 0.1% of sample points to zero; Echo addition (simulating multipath): attenuating the audio volume by 5% and then superimposing it after a 50ms delay; Random noise (simulating low signal-to-noise ratio): adding uniformly distributed noise while maintaining an SNR of 20dB; Sample quantization (simulating lossy processing): 2 9 Order-of-magnitude quantization; median filtering (simulated lossy processing): median filtering with a kernel of 3.

[0244] In one embodiment, the performance of different methods under simulated channel impairments is shown in Table 2 below. The performance can be characterized by the average bit error rate.

[0245] Table 2

[0246]

[0247] Based on Table 2, compared with the two baseline models (baseline-1 and baseline-2), the cross-layer collaborative noise reduction mechanism proposed in this application can significantly reduce the bit error rate under various adverse channel conditions and ensure the connectivity of the communication link.

[0248] Traditional algorithms use a fixed quantization step size. When the overall signal energy decreases, a systematic mismatch occurs between the original decision interval and the received signal characteristics. This application, however, introduces adaptive quantization mapping logic, which adjusts the quantization step size accordingly. Statistical energy of the signal itself Linear binding effectively immunizes against linear gain distortion. When the signal fades, the decision threshold scales synchronously with the signal energy while maintaining its relative position. As shown in Table 2, the experimental results demonstrate that under amplitude scaling attacks, the bit error rate of this application is only 0.30%, significantly better than traditional algorithms.

[0249] Furthermore, since baseline-1 directly modifies single-point coefficients, it is highly susceptible to noise interference, leading to feature submersion. This application, however, utilizes spectral mean energy transfer to calculate the arithmetic mean of subband coefficients. Simultaneously, based on the law of large numbers, zero-mean Gaussian noise cancels each other out during the summation and averaging process, significantly improving the signal-to-noise ratio of feature extraction and ensuring accurate information reconstruction even under noise interference. As shown in Table 2, the experimental results demonstrate that, even in environments with superimposed random noise, the bit error rate of this application is kept at a low level, significantly lower than that of baseline-1.

[0250] This application constructs a collaborative protection system combining full-dimensional interleaving and fault-tolerance mechanisms. Full-dimensional interleaving discretizes burst shearing in the channel into random errors at the receiver. Subsequently, the fault-tolerance mechanism inherent in QR codes effectively ensures that the original payload can be recovered losslessly using redundant check information even if some data units are erroneous or lost. Experimental data in Table 2 demonstrates that the fault-tolerance capability of this mechanism effectively ensures the integrity of the link. Therefore, this application can still maintain an extremely low bit error rate even against attacks such as adversarial sample suppression.

[0251] In another embodiment, the system's "effective transmission rate" (capacity) is finally evaluated.

[0252] In one embodiment, the embedding capacity of different algorithms can be shown in Table 3 below. The effective transmission rate can be used to represent the watermark carrying capacity (i.e., the embedding capacity).

[0253] Table 3

[0254]

[0255] Based on Table 3, due to the granular time-division multiplexing strategy adopted in this application, the transmission space is greatly explored while ensuring the noise immunity of the high-frequency channel, and a high-capacity transmission of 2756.25 bps is achieved, which is sufficient to support the real-time transmission of complex instructions, identity authentication images or encryption keys, and solves the bottleneck of low transmission efficiency caused by excessive sacrifice of bandwidth in traditional noise immunity algorithms.

[0256] In summary, the audio noise-resistant steganography method and blind extraction method for low signal-to-noise ratio environments proposed in this application effectively overcome the feature submersion and fading mismatch problems in low signal-to-noise ratio environments by constructing a three-layer collaborative noise-resistant framework consisting of link-layer fault-tolerant interleaving, physical-layer statistical denoising, and energy modulation-layer adaptive quantization. This framework also endows the system with fault tolerance against sudden shearing, thereby significantly improving the robustness, imperceptibility, and effective transmission capacity of covert communication in complex channels.

[0257] Based on the above-mentioned audio noise-resistant steganography method for low signal-to-noise ratio environments, this application also provides an audio noise-resistant steganography device for low signal-to-noise ratio environments. The following will combine... Figure 7 The device is described in detail.

[0258] Figure 7 A structural block diagram of an audio noise-resistant steganography device for low signal-to-noise ratio environments according to an embodiment of this application is shown.

[0259] like Figure 7 As shown, the audio noise-resistant steganography device 700 for low signal-to-noise ratio environments in this embodiment includes a generation module 710, an acquisition module 720, a determination module 730, an embedding module 740, and a synthesis module 750.

[0260] The generation module 710 is used to map the watermark data to be embedded into a fault-tolerant coding matrix, and based on the dynamic key, to perform a two-dimensional linear automorphic scrambling transformation on the fault-tolerant coding matrix to generate a two-dimensional data carrier. In one embodiment, the generation module 710 can be used to perform the operation S210 described above, which will not be repeated here.

[0261] The acquisition module 720 is used to perform multi-scale decomposition and discrete cosine transform on the original audio signal to obtain a two-dimensional frequency domain coefficient matrix. In one embodiment, the acquisition module 720 can be used to perform the operation S220 described above, which will not be repeated here.

[0262] The determining module 730 is used to determine the preset high-frequency regions of the coefficients in each row of the two-dimensional frequency domain coefficient matrix as the embedding regions of each row of the two-dimensional frequency domain coefficient matrix, and to calculate the average amplitude of the coefficients in each embedding region as a statistical steady-state energy characteristic. In one embodiment, the determining module 730 can be used to perform the operation S230 described above, which will not be repeated here.

[0263] The embedding module 740 is used to adjust the quantization step size based on the statistical steady-state energy characteristics of each row in the two-dimensional frequency domain coefficient matrix, and based on the statistical steady-state energy characteristics and quantization step size of each row in the two-dimensional frequency domain coefficient matrix, embeds the encoded elements corresponding to each row in the two-dimensional data carrier into the embedding regions of each row of the two-dimensional frequency domain coefficient matrix, thereby obtaining the feature coefficient matrix of the watermarked data. In one embodiment, the embedding module 740 can be used to perform the operation S240 described above, which will not be repeated here.

[0264] The synthesis module 750 is used to reconstruct the audio signal containing the watermarked data based on the feature coefficient matrix of the watermarked data. In one embodiment, the synthesis module 750 can be used to perform the operation S250 described above, which will not be repeated here.

[0265] According to embodiments of this application, any plurality of modules among the generation module 710, obtaining module 720, determining module 730, embedding module 740, and synthesis module 750 can be combined into one module, or any one of these modules can be split into multiple modules. Alternatively, at least part of the functionality of one or more of these modules can be combined with at least part of the functionality of other modules and implemented in one module. According to embodiments of this application, at least one of the generation module 710, obtaining module 720, determining module 730, embedding module 740, and synthesis module 750 can be at least partially implemented as hardware circuitry, such as a field-programmable gate array (FPGA), a programmable logic array (PLA), a system-on-a-chip, a system-on-a-substrate, a system-on-package, an application-specific integrated circuit (ASIC), or implemented in hardware or firmware by any other reasonable means of integrating or packaging circuitry, or implemented in software, hardware, or firmware, or in any suitable combination of any of these three implementation methods. Alternatively, at least one of the generation module 710, the acquisition module 720, the determination module 730, the embedding module 740, and the synthesis module 750 may be implemented at least partially as a computer program module that can perform corresponding functions when the computer program module is run.

[0266] Based on the above-described blind information extraction method, this application also provides a blind information extraction device. The following will be combined with... Figure 8 The device is described in detail.

[0267] Figure 8 A structural block diagram of an information blind extraction apparatus according to an embodiment of this application is shown.

[0268] like Figure 8 As shown, the information blind extraction device 800 of this embodiment includes a blind extraction module 810.

[0269] The blind extraction module 810 is used to blindly extract the audio signal of the watermarked data to obtain a composite verification substrate, wherein the composite verification substrate is used to perform watermark verification on the audio signal of the watermarked data.

[0270] According to embodiments of this application, the blind extraction module 810 can be divided into multiple modules. According to embodiments of this application, the blind extraction module 810 can be at least partially implemented as a hardware circuit, such as a Field-Programmable Gate Array (FPGA), a Programmable Logic Array (PLA), a System-on-a-Chip, a System-on-a-Substrate, a System-on-Package, an Application-Specific Integrated Circuit (ASIC), or any other reasonable method of integrating or packaging circuitry, or implemented in software, hardware, or firmware, or in any suitable combination of any of these three methods. Alternatively, the blind extraction module 810 can be at least partially implemented as a computer program module, which, when run, can perform corresponding functions.

[0271] Figure 9 A block diagram of an electronic device suitable for implementing an audio noise-resistant steganography method or a blind information extraction method for low signal-to-noise ratio environments, according to an embodiment of this application, is shown.

[0272] like Figure 9 As shown, an electronic device 900 according to an embodiment of this application includes a processor 901, which can perform various appropriate actions and processes according to a program stored in ROM 902 or a program loaded from storage portion 908 into RAM 903. The processor 901 may include, for example, a general-purpose microprocessor (e.g., a CPU), an instruction set processor and / or an associated chipset and / or a special-purpose microprocessor (e.g., an application-specific integrated circuit (ASIC)), etc. The processor 901 may also include onboard memory for caching purposes. The processor 901 may include a single processing unit or multiple processing units for performing different actions of the method flow according to an embodiment of this application. Wherein, ROM is read-only memory, and RAM is random access memory.

[0273] RAM 903 stores various programs and data required for the operation of electronic device 900. Processor 901, ROM 902, and RAM 903 are interconnected via bus 904. Processor 901 executes various operations of the method flow according to embodiments of this application by executing programs in ROM 902 and / or RAM 903. It should be noted that the programs may also be stored in one or more memories other than ROM 902 and RAM 903. Processor 901 may also execute various operations of the method flow according to embodiments of this application by executing programs stored in said one or more memories.

[0274] According to embodiments of this application, the electronic device 900 may further include an input / output (I / O) interface 905, which is also connected to a bus 904. The electronic device 900 may also include one or more of the following components connected to the input / output (I / O) interface 905: an input section 906 including a keyboard, mouse, etc.; an output section 907 including a cathode ray tube (CRT), liquid crystal display (LCD), etc., and a speaker, etc.; a storage section 908 including a hard disk, etc.; and a communication section 909 including a network interface card such as a LAN card, modem, etc. The communication section 909 performs communication processing via a network such as the Internet. A drive 910 is also connected to the input / output (I / O) interface 905 as needed. A removable medium 911, such as a disk, optical disk, magneto-optical disk, semiconductor memory, etc., is installed on the drive 910 as needed so that computer programs read from it can be installed into the storage section 908 as needed.

[0275] This application also provides a computer-readable storage medium, which may be included in the device / apparatus / system described in the above embodiments; or it may exist independently and not assembled into the device / apparatus / system. The computer-readable storage medium carries one or more programs, which, when executed, implement the method according to the embodiments of this application.

[0276] According to embodiments of this application, the computer-readable storage medium can be a non-volatile computer-readable storage medium, such as including but not limited to: portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination thereof. In this application, the computer-readable storage medium can be any tangible medium containing or storing a program that can be used by or in conjunction with an instruction execution system, apparatus, or device. For example, according to embodiments of this application, the computer-readable storage medium may include ROM 902 and / or RAM 903 and / or one or more memories other than ROM 902 and RAM 903 described above.

[0277] Embodiments of this application also include a computer program product comprising a computer program containing program code for performing the methods shown in the flowchart. When the computer program product is run on a computer system, the program code enables the computer system to implement the audio noise-resistant steganography method or information blind extraction method for low signal-to-noise ratio environments provided in the embodiments of this application.

[0278] When the computer program is executed by the processor 901, it performs the functions defined in the system / apparatus of this application embodiment. According to the embodiments of this application, the systems, apparatuses, modules, units, etc., described above can be implemented by computer program modules.

[0279] In one embodiment, the computer program may rely on a tangible storage medium such as an optical storage device or a magnetic storage device. In another embodiment, the computer program may also be transmitted and distributed in the form of signals over a network medium, and downloaded and installed via the communication section 909, and / or installed from a removable medium 911. The program code contained in the computer program can be transmitted using any suitable network medium, including but not limited to: wireless, wired, etc., or any suitable combination thereof.

[0280] In such an embodiment, the computer program can be downloaded and installed from a network via the communication section 909, and / or installed from the removable medium 911. When the computer program is executed by the processor 901, it performs the functions defined in the system of this application embodiment. According to the embodiments of this application, the systems, devices, apparatuses, modules, units, etc., described above can be implemented by computer program modules.

[0281] According to embodiments of this application, program code for executing the computer programs provided in the embodiments of this application can be written in any combination of one or more programming languages. Specifically, these computational programs can be implemented using high-level procedural and / or object-oriented programming languages, and / or assembly / machine languages. Programming languages ​​include, but are not limited to, languages ​​such as Java, C++, Python, "C", or similar programming languages. The program code can be executed entirely on the user's computing device, partially on the user's device, partially on a remote computing device, or entirely on a remote computing device or server. In cases involving remote computing devices, the remote computing device can be connected to the user's computing device via any type of network, including a local area network (LAN) or a wide area network (WAN), or it can be connected to an external computing device (e.g., via the Internet using an Internet service provider).

[0282] The flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of systems, methods, and computer program products according to various embodiments of this application. In this regard, each block in a flowchart or block diagram may represent a module, segment, or portion of code containing one or more executable instructions for implementing a specified logical function. It should also be noted that in some alternative implementations, the functions indicated in the blocks may occur in a different order than those indicated in the drawings. For example, two consecutively indicated blocks may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved. It should also be noted that each block in a block diagram or flowchart, and combinations of blocks in a block diagram or flowchart, may be implemented using a dedicated hardware-based system that performs the specified function or operation, or using a combination of dedicated hardware and computer instructions.

[0283] Those skilled in the art will understand that the features described in the various embodiments of this application can be combined and / or combined in various ways, even if such combinations or combinations are not explicitly described in this application. In particular, the features described in the various embodiments of this application can be combined and / or combined in various ways without departing from the spirit and teachings of this application. All such combinations and / or combinations fall within the scope of this application.

[0284] The embodiments of this application have been described above. However, these embodiments are merely illustrative and not intended to limit the scope of this application. Although various embodiments have been described above, this does not mean that the measures in the various embodiments cannot be used advantageously in combination. Without departing from the scope of this application, those skilled in the art can make various substitutions and modifications, all of which should fall within the scope of this application.

Claims

1. An audio noise-resistant steganography method for low signal-to-noise ratio environments, characterized in that, The method includes: The watermark data to be embedded is mapped to a fault-tolerant coding matrix, and based on the dynamic key, a two-dimensional linear self-isomorphic scrambling transformation is performed on the fault-tolerant coding matrix to generate a two-dimensional data carrier. The original audio signal is decomposed into multiple scales and subjected to discrete cosine transform (DCT) to obtain a two-dimensional frequency domain coefficient matrix. This process includes: performing multi-scale decomposition on the original audio signal using discrete wavelet transform (DCT) to obtain high-frequency detail components; segmenting the high-frequency detail components based on a preset segment length to obtain a wavelet domain coefficient matrix; and sequentially performing DCT on each row of the wavelet domain coefficient matrix to obtain the two-dimensional frequency domain coefficient matrix. The number of coefficients in each row of the wavelet domain coefficient matrix is ​​determined based on the preset segment length. Determining the preset high-frequency region of the coefficients in each row of the two-dimensional frequency domain coefficient matrix as the embedding region of each row of the two-dimensional frequency domain coefficient matrix includes: determining the preset high-frequency region located at the end of the coefficients in each row of the two-dimensional frequency domain coefficient matrix based on a preset embedding bit number, wherein the number of coefficients in the preset high-frequency region is determined according to the preset embedding bit number; determining the preset high-frequency region of the coefficients in each row as the embedding region of each row of the two-dimensional frequency domain coefficient matrix, and calculating the average amplitude of the coefficients in each embedding region as a statistical steady-state energy characteristic; The quantization step size is adjusted based on the statistical steady-state energy characteristics of each row in the two-dimensional frequency domain coefficient matrix. Based on the statistical steady-state energy characteristics and quantization step size of each row in the two-dimensional frequency domain coefficient matrix, the encoded elements in the two-dimensional data carrier corresponding to each row of the two-dimensional frequency domain coefficient matrix are embedded into the embedding regions of each row of the two-dimensional frequency domain coefficient matrix, resulting in a feature coefficient matrix of the watermarked data, including: Determine at least one coding element in the two-dimensional data carrier that corresponds to each row of the two-dimensional frequency domain coefficient matrix, wherein the number of the at least one coding element is determined according to a preset embedding bit number; Based on the quantization step size and at least one coding element corresponding to each row of the two-dimensional frequency domain coefficient matrix, the statistical steady-state energy features corresponding to each row of the two-dimensional frequency domain coefficient matrix are modulated into target energy features corresponding to at least one coding element corresponding to each row of the two-dimensional frequency domain coefficient matrix. The target energy characteristics are obtained using the following formula: ； Among them, for the first in the two-dimensional frequency domain coefficient matrix OK, Represents the relationship between the two-dimensional data carrier and the first The encoded element at the corresponding position of the embedded region of the row. Represents the second two-dimensional frequency domain coefficient matrix. The quantization step size of the row, Represents the second two-dimensional frequency domain coefficient matrix. The first row located in the embedded region The target energy characteristics corresponding to each coefficient Represents the first element in the two-dimensional frequency domain coefficient matrix. Statistical steady-state energy characteristics of the line; Calculate the difference between the target energy feature corresponding to at least one encoded element in each row of the two-dimensional frequency domain coefficient matrix and the statistical steady-state energy feature, and use it as the mean energy feature difference; The difference in mean energy characteristics is obtained using the following formula: ； in, This indicates that for the second two-dimensional frequency domain coefficient matrix... The first row located in the embedded region The difference in mean energy characteristics of each coefficient before and after modulation; Based on the positive and negative signs of the coefficients in the embedding region of each row of the two-dimensional frequency domain coefficient matrix, the coefficients in the embedding region of each row of the two-dimensional frequency domain coefficient matrix are compensated by the difference in mean energy characteristics to obtain the feature coefficient matrix of the watermarked data. The compensation formula based on the difference in mean energy characteristics is as follows: ； in, Represents the first element in the two-dimensional frequency domain coefficient matrix. The row located in the embedded region One coefficient; Represents the first element in the two-dimensional frequency domain coefficient matrix. The row located in the embedded region after compensation of the first One coefficient; This indicates a signal phase-locked constraint mechanism. Preset embedding bit width; Based on the feature coefficient matrix of the watermarked data, the audio signal of the watermarked data is reconstructed.

2. The method according to claim 1, characterized in that, The adjustment of the quantization step size based on the statistical steady-state energy characteristics of each row in the two-dimensional frequency domain coefficient matrix includes: Based on a preset intensity coefficient, the quantization step size is adjusted according to the statistical steady-state energy characteristics of each row in the two-dimensional frequency domain coefficient matrix to obtain the quantization step size corresponding to the statistical steady-state energy characteristics of each row in the two-dimensional frequency domain coefficient matrix.

3. The method according to claim 1, characterized in that, The process of reconstructing the audio signal containing the watermarked data based on the feature coefficient matrix of the watermarked data includes: The inverse discrete cosine transform is performed sequentially on each row of the feature coefficient matrix of the watermarked data to obtain the coefficient matrix after the inverse transform. According to the preset segmentation order, the rows of the coefficient matrix after inverse transformation are concatenated and spliced ​​to obtain the high-frequency detail components of the watermarked data. The inverse discrete wavelet transform is performed on the low-frequency approximation component and the high-frequency detail component of the watermarked data to reconstruct the audio signal of the watermarked data. The low-frequency approximation component is obtained by performing a discrete wavelet transform on the original audio signal.

4. The method according to claim 1, characterized in that, The watermark data to be embedded includes digital identification information and visual authentication patterns; The step of mapping the watermark data to be embedded into a fault-tolerant coding matrix includes: The digital identification information is subjected to fault-tolerant redundancy encoding to obtain a two-dimensional matrix data structure; The visual authentication pattern is feature normalized, and the feature-normalized visual authentication pattern is mapped and fused into the non-critical data area of ​​the two-dimensional matrix data structure to obtain the fault-tolerant coding matrix. The non-critical data region represents the area in the two-dimensional matrix data structure that does not affect the integrity of the digital identification information.

5. A method for blind information extraction, characterized in that, include: Blind extraction is performed on the audio signal of the watermarked data to obtain a composite verification substrate, wherein the composite verification substrate is used to perform watermark verification on the audio signal of the watermarked data; The audio signal containing the watermark data is obtained by the audio noise-resistant steganography method for low signal-to-noise ratio environments according to any one of claims 1 to 4.

6. The method according to claim 5, characterized in that, The process of blindly extracting the audio signal from the watermarked data to obtain a composite verification basis includes: Multi-scale decomposition and discrete cosine transform are performed on the audio signal of the watermarked data to obtain the two-dimensional frequency domain coefficient matrix of the watermarked data. Based on the preset number of embedding bits, the embedding region of each row in the two-dimensional frequency domain coefficient matrix of the watermarked data is determined, and the average amplitude of the coefficients in each embedding region is calculated as the residual statistical steady-state energy feature. The recovery quantization step size is estimated based on the residual statistical steady-state energy characteristics of each row in the two-dimensional frequency domain coefficient matrix of the watermarked data, and the two-dimensional frequency domain coefficient matrix of the watermarked data is restored to a two-dimensional encryption matrix based on the residual statistical steady-state energy characteristics and the recovery quantization step size of each row in the two-dimensional frequency domain coefficient matrix of the watermarked data. Based on the dynamic key of the embedding process, the two-dimensional encryption matrix is ​​subjected to an inverse transformation of two-dimensional linear automorphic scrambling to obtain the composite verification basis.

7. The method according to claim 6, characterized in that, The method of estimating the recovery quantization step size based on the residual statistical steady-state energy characteristics of each row in the two-dimensional frequency domain coefficient matrix of the watermarked data, and recovering the two-dimensional frequency domain coefficient matrix of the watermarked data to a two-dimensional encryption matrix based on the residual statistical steady-state energy characteristics of each row in the two-dimensional frequency domain coefficient matrix of the watermarked data and the recovery quantization step size, includes: Based on the preset intensity coefficient, the recovery quantization step size is estimated according to the residual statistical steady-state energy characteristics of each row in the two-dimensional frequency domain coefficient matrix of the watermarked data, and the recovery quantization step size corresponding to the residual statistical steady-state energy characteristics of each row in the two-dimensional frequency domain coefficient matrix of the watermarked data is obtained respectively. Based on the quantization interval range to which the residual statistical steady-state energy characteristics of each row in the two-dimensional frequency domain coefficient matrix of the watermarked data belong, at least one coding element corresponding to the embedding region of each row in the two-dimensional frequency domain coefficient matrix of the watermarked data is determined, wherein the quantization interval range corresponding to each row in the two-dimensional frequency domain coefficient matrix of the watermarked data is determined according to the recovery quantization step size of each row. Based on the positional mapping relationship between the encoded elements in the two-dimensional data carrier and the embedded regions of each row of the two-dimensional frequency domain coefficient matrix determined by the embedding process, the two-dimensional encryption matrix is ​​obtained according to at least one encoded element corresponding to each of the multiple embedded regions in the two-dimensional frequency domain coefficient matrix of the watermarked data.

Images