Information steganography method, device and equipment

Abstract

The embodiment of the invention discloses an information steganography method, device and equipment. The method comprises the steps of obtaining target data to be steganography and target watermark information; and converting the input target data or time domain data into frequency domain target data through one or more sequentially connected watermark embedding modules in a pre-trained steganography coding model, and obtaining the target data or time domain target data based on a global association relationship and / or a local association relationship between the target watermark information and the frequency domain target data. Embedding the target watermark information into the frequency domain target data to obtain frequency domain watermark-carrying data, and converting the frequency domain watermark-carrying data into time domain watermark-carrying data; and through a data adjustment module in the steganography coding model, adjusting time-domain watermark-carrying data finally output by one or more sequentially connected watermark embedding modules to obtain target watermark-carrying data corresponding to the target data.

Description

Technical Field

[0001] This document relates to the field of computer technology, and in particular to an information steganography method, device and equipment. Background Art

[0002] In the current digital media environment, which is full of tampered content such as DeepFakes, as people pay more and more attention to their privacy data, it is crucial to embed reliable and persistent information for ownership verification and source tracing in the objects to be steganographic (such as images, videos, etc.). Through complex embedding methods, the embedded watermark is imperceptible to the naked eye, but can be extracted by a specialized decoder. Although deep learning technology has made progress, it still faces challenges in optimizing input utilization and solving global and local instabilities. To this end, it is necessary to provide a better frequency-domain adaptive watermark steganography mechanism to achieve a better balance between invisibility and robustness, so as to provide watermark information invisibility and robustness performance with fewer parameters. Summary of the invention

[0003] The purpose of the embodiments of this specification is to provide a better frequency-domain adaptive watermark steganography mechanism to achieve a better balance between invisibility and robustness, thereby providing watermark information invisibility and robustness performance with fewer parameters.

[0004] In order to implement the above technical solution, the embodiments of this specification are implemented as follows: An information steganography method provided in an embodiment of the present specification comprises: obtaining target data and target watermark information to be steganographic. Through one or more sequentially connected watermark embedding modules in a pre-trained steganographic coding model, the input target data or time-domain data is converted into frequency-domain target data, and based on the global correlation relationship and / or local correlation relationship between the target watermark information and the frequency-domain target data, the target watermark information is embedded into the frequency-domain target data to obtain frequency-domain watermark data, and the frequency-domain watermark data is converted into time-domain watermark data. Through the data adjustment module in the steganographic coding model, the time-domain watermark data finally output by one or more sequentially connected watermark embedding modules is adjusted to obtain the target watermark data corresponding to the target data.

[0005] An information steganography device provided in an embodiment of the present specification comprises: a data acquisition module for acquiring target data and target watermark information to be steganographic. A watermark embedding module for converting the input target data or time-domain data into frequency-domain target data through one or more sequentially connected watermark embedding modules in a pre-trained steganographic coding model, embedding the target watermark information into the frequency-domain target data based on the global correlation relationship and / or local correlation relationship between the target watermark information and the frequency-domain target data, obtaining frequency-domain watermark data, and converting the frequency-domain watermark data into time-domain watermark data. An adjustment module for adjusting the time-domain watermark data finally output by one or more sequentially connected watermark embedding modules through a data adjustment module in the steganographic coding model to obtain target watermark data corresponding to the target data.

[0006] An information steganography device provided in an embodiment of the present specification includes: a processor; and a memory arranged to store computer executable instructions, wherein when the executable instructions are executed, the processor: obtains the target data to be steganographic and the target watermark information. Through one or more sequentially connected watermark embedding modules in a pre-trained steganographic coding model, the input target data or time domain data is converted into frequency domain target data, and based on the global correlation relationship and / or local correlation relationship between the target watermark information and the frequency domain target data, the target watermark information is embedded into the frequency domain target data to obtain frequency domain watermark data, and the frequency domain watermark data is converted into time domain watermark data. Through the data adjustment module in the steganographic coding model, the time domain watermark data finally output by one or more sequentially connected watermark embedding modules is adjusted to obtain the target watermark data corresponding to the target data.

[0007] The embodiments of this specification also provide a storage medium, which is used to store computer executable instructions. When the executable instructions are executed by the processor, the following process is implemented: obtaining the target data and target watermark information to be steganographic. Through one or more sequentially connected watermark embedding modules in a pre-trained steganographic coding model, the input target data or time domain data is converted into frequency domain target data, and based on the global correlation relationship and / or local correlation relationship between the target watermark information and the frequency domain target data, the target watermark information is embedded in the frequency domain target data to obtain frequency domain watermark data, and the frequency domain watermark data is converted into time domain watermark data. Through the data adjustment module in the steganographic coding model, the time domain watermark data finally output by one or more sequentially connected watermark embedding modules is adjusted to obtain the target watermark data corresponding to the target data.

[0008] The embodiments of this specification also provide a computer program product, including a computer program, which implements the following process when executed by a processor: obtaining target data and target watermark information to be steganographic. Converting the input target data or time-domain data into frequency-domain target data through one or more sequentially connected watermark embedding modules in a pre-trained steganographic coding model, embedding the target watermark information into the frequency-domain target data based on the global correlation relationship and / or local correlation relationship between the target watermark information and the frequency-domain target data, obtaining frequency-domain watermark data, and converting the frequency-domain watermark data into time-domain watermark data. Adjusting the time-domain watermark data finally output by one or more sequentially connected watermark embedding modules through a data adjustment module in the steganographic coding model to obtain target watermark data corresponding to the target data. BRIEF DESCRIPTION OF THE DRAWINGS

[0009] In order to more clearly illustrate the technical solutions in the embodiments of this specification or the prior art, the drawings required for use in the embodiments or the prior art description will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in this specification. For those of ordinary skill in the art, other drawings can be obtained based on these drawings without creative labor. Figure 1 This is an embodiment of an information steganography method of this specification; Figure 2 This is another embodiment of the information steganography method of this specification; Figure 3 This is a schematic diagram of an information steganography process in this manual; Figure 4 This is another embodiment of the information steganography method of this specification; Figure 5 This is another embodiment of the information steganography method of this specification; Figure 6 This is a schematic diagram of a model training process in this manual; Figure 7 This is an embodiment of an information steganography device of this specification; Figure 8 This is an embodiment of an information steganography device in this specification. DETAILED DESCRIPTION

[0010] The embodiments of this specification provide an information steganography method, device and equipment.

[0011] In order to enable those skilled in the art to better understand the technical solutions in this specification, the technical solutions in the embodiments of this specification will be clearly and completely described below in conjunction with the drawings in the embodiments of this specification. Obviously, the described embodiments are only part of the embodiments of this specification, not all of the embodiments. Based on the embodiments in this specification, all other embodiments obtained by ordinary technicians in this field without creative work should fall within the scope of protection of this specification.

[0012] The embodiments of this specification provide a frequency-domain adaptive watermark steganography mechanism. In the current digital media environment, which is full of tampered content such as DeepFakes, it is crucial to embed reliable and persistent information for ownership verification and source tracking in the objects to be steganographic (such as images, videos, etc.). Through complex embedding methods, the embedded watermark is imperceptible to the naked eye, but can be extracted by a specialized decoder. Although deep learning technology has made progress, it still faces challenges in optimizing input utilization and solving global and local instabilities.

[0013] Early watermark steganography mechanisms focused on ensuring that protected data is only accessible to authorized parties, while the current development trend focuses on improving robustness, enhancing the ability to resist various malicious attacks, and ensuring that watermarks can be consistently retrieved. This evolution highlights the important role of watermark steganography mechanisms in protecting and tracking digital media. Generally, watermark steganography mechanisms rely mainly on corresponding rules and are set up using the limits of human visual perception. Specifically, taking the steganographic object as an image as an example, a watermark steganography mechanism that relies on manual design can be set up. Among them, the least significant bit (LSB) method in the spatial domain is a common method. This method can modify the lowest bit of each pixel according to the bit information of the message, thereby creating a new image that is visually indistinguishable from the original image. In the frequency domain, the fact that the human eye is insensitive to changes in the frequency domain region is used to embed watermark information by converting the image into a frequency domain image and adjusting the coefficients of high-frequency elements. Although the above-mentioned manually designed methods can embed a large amount of data in an image imperceptibly, they exhibit poor robustness when the image is modified, especially during encryption operations. In addition, since this method uniformly changes the statistical characteristics of the image, it is also easier to detect and the watermark information may be extracted.

[0014] In recent years, watermark steganography based on neural networks has emerged. Among them, HiDDeN introduces a fully trainable end-to-end framework. Taking the steganographic object as an example, an encoder can be used to fuse the cryptanalyst image with the watermark information to generate an encoded image. The encoded image is degraded through a noise layer to simulate potential attacks, and then the decoder attempts to recover the watermark information. Subsequent studies have further improved the model, some of which use the U-Net architecture for encoding and use skip connections to maintain visual integrity. Watermark information customized for the two-dimensional structure of the image is directly merged into the pixel layer or the intermediate feature layer. Channel attention is applied to enhance the fusion of watermark information and image through several methods such as SepMark, although its effectiveness may be limited by relying only on channel differences. Attention-based techniques such as RivaGan overcome the above limitations by using attention weights to distribute watermark information and create an attention mask that can adjust the strength of watermark information according to spatial position and channel, thereby improving the fusion effect. RoSteALS uses pre-trained autoencoders for latent space fusion, bypassing the pitfalls of pixel domain stitching and enhancing robustness, although the lossy nature of the autoencoder itself may lead to potential reduction in image fidelity. However, the above convolutional neural network-based approach has limitations and may fall into local watermarking information that is visible to the naked eye.

[0015] In addition to the spatial domain fusion mechanism, there is also research on the frequency domain method that uses neural networks to mine frequency domain image characteristics to realize watermark information, especially embedding watermark information into a specific frequency band of the red channel, recognizing that high-frequency components provide better invisibility but weaker robustness, while low-frequency components provide stronger robustness but reduced invisibility. However, this method's single choice design for low or high frequencies will result in an inability to balance robustness and invisibility.

[0016] To this end, the embodiments of this specification provide a better frequency domain adaptive watermark steganography mechanism, which directly integrates the watermark information into the generation process and adopts a unique fusion method that combines the advantages of the spatial domain and the frequency domain. This method enhances details through data-based learning and builds a comprehensive adaptive framework using spectral characteristics. Within the Transformer architecture, the data-based embedding is converted to the frequency domain for conditional processing. Subsequently, an adaptive filter adjusts the degree of integration of the watermark information according to different frequency bands, achieving a better balance between invisibility and robustness, thereby being able to provide watermark information invisibility and robustness performance with fewer parameters. For specific processing, please refer to the specific content in the following embodiments.

[0017] like Figure 1As shown, the embodiment of this specification provides an information steganography method, and the execution subject of the method can be a terminal device or a server, etc., wherein the terminal device can be a mobile terminal device such as a mobile phone, a tablet computer, or a computer device such as a laptop or a desktop computer, or an IoT device (specifically such as a smart watch, a car device, etc.), etc., wherein the server can be an independent server, or a server cluster composed of multiple servers, etc. The server can be a background server in the financial field or the online shopping field, or a background server of an application, etc. In this embodiment, the execution subject is a server as an example for detailed description. For the case where the execution subject is a terminal device, please refer to the following server situation processing, which will not be repeated here. The method can specifically include the following steps: In step S102, the target data to be steganographic and the target watermark information are obtained.

[0018] Among them, the target data can be any data to be steganographic, specifically, the target data can be image data, video data, etc., which can be set according to actual conditions, and the embodiments of this specification do not limit this. The target watermark information can be any watermark information that needs to be embedded in the target data. In practical applications, the target watermark information can be generated in a variety of ways, for example, using a pseudo-random sequence generator or a chaotic system, etc. The target watermark information can also be a meaningful binary image, grayscale image or color image, etc. Usually, in order to carry more copyright information, a binary image or a grayscale image can be used to represent the watermark information. Specifically, a product serial number or logo (logo or trademark) can be used as watermark information. In addition, for meaningful watermark information, in order to enhance the security of the watermark information and improve its ability to resist malicious attacks, a scrambling mechanism can be used to pre-process the watermark information to eliminate the correlation between the watermark information.

[0019] In implementation, when it is necessary to perform copyright protection, tamper prevention, data traceability, etc. on certain data (i.e., target data to be steganographic), the corresponding watermark information (i.e., target watermark information) can be obtained. The target watermark information can be obtained in a variety of different ways. For example, the target watermark information can be generated by using a pseudo-random sequence generator or a chaotic system, or the specified data can be obtained from a specified database as the target watermark information, etc. The specific setting can be based on actual conditions. The target data to be steganographic and the target watermark information can be provided to the server, and the server can obtain the target data to be steganographic and the target watermark information.

[0020] In step S104, the input target data or time domain data is converted into frequency domain target data through one or more sequentially connected watermark embedding modules in a pre-trained steganographic coding model, and based on the global correlation relationship and / or local correlation relationship between the target watermark information and the frequency domain target data, the target watermark information is embedded in the frequency domain target data to obtain the frequency domain watermark data, and the frequency domain watermark data is converted into the time domain watermark data.

[0021] Among them, the stego coding model can be a model for embedding watermark information into the frequency domain data corresponding to the target data, and the stego coding model can be a model constructed by a specified algorithm and / or network model. Specifically, the stego coding model can be constructed by a time-frequency conversion algorithm, a neural network, an adaptive filtering network, etc., wherein the time-frequency conversion algorithm can be, for example, a Fourier transform algorithm, a wavelet transform algorithm, etc., which can be set according to actual conditions, and the embodiments of this specification do not limit this. In the embodiments of this specification, the stego coding model can include a watermark embedding module and a data adjustment module. The watermark embedding module can be used to embed watermark information into the frequency domain data corresponding to the target data. The watermark embedding module can be constructed by a specified algorithm and / or network model, such as the watermark embedding module can be constructed by a time-frequency conversion algorithm, a neural network, an adaptive filtering network, etc. The data adjustment module can be used to adjust the watermarked data finally output by the watermark embedding module to obtain watermarked data with better effect. The data adjustment module can be constructed by a specified algorithm and / or network model, for example, the data adjustment module can be constructed by a neural network, etc., which can be set according to actual conditions.

[0022] In implementation, a frequency domain adaptive watermark steganography mechanism can be pre-set, and a suitable algorithm and / or network model can be selected through the watermark steganography mechanism. The selected algorithm and / or network model can be used to construct a model architecture of a steganographic coding model, in which the watermark embedding module plays a key role. In practical applications, since the model architecture of the Transformer module is an important model architecture in the field of deep learning, the Transformer module abandons the usual recurrent neural network (RNN) and convolutional neural network (CNN) and instead adopts a self-attention mechanism (Self-Attention). The Transformer module can process all positions of the input sequence in parallel and capture long-distance dependencies through the self-attention mechanism. Based on this, the watermark embedding module in the steganographic coding model can adopt a model architecture similar to the Transformer module. Based on this, the internal structure of the watermark embedding module can be similarly designed according to the internal structure of the Transformer module. Specifically, the watermark embedding module can be provided with a time-frequency conversion algorithm network layer, an embedding network layer, an adaptive filtering network layer, etc., or a more complex or simplified network structure can be provided. For example, the watermark embedding module can be provided with an embedding network layer and an adaptive filtering network layer, etc., and frequency domain conversion and watermark embedding can be performed through the embedding network layer. The specific setting can be based on actual conditions, and the embodiments of this specification do not limit this. After constructing the watermark embedding module in the above manner, the steganographic coding model can be further constructed, a certain number of training samples (such as video samples, etc.) can be obtained, and the training samples can be used to train the steganographic coding model until the preset loss function converges to obtain the trained steganographic coding model.

[0023] The target data can be input into the above-mentioned trained steganographic coding model. The target data first enters the watermark embedding module in the steganographic coding model. The target data is subjected to time-frequency transformation through the network layer preset in the watermark embedding module to convert the target data into frequency domain target data. Then, the target watermark information can be embedded into the frequency data corresponding to the target data. In addition, considering that the target watermark information is usually embedded into the high-frequency part of the frequency data corresponding to the target data, which will make them exhibit poor robustness when the target data is modified (especially when the target data is encrypted), and also uniformly change the statistical characteristics of the target data, so that it is easier to detect and the target watermark information may be extracted. For this purpose, an adaptive filtering mechanism can be set, through which the global correlation relationship between the target watermark information and the frequency domain target data and / or the local correlation relationship between the target watermark information and the frequency domain target data can be determined. In this way, through the adaptive filtering mechanism, the global correlation relationship between the target watermark information and the frequency domain target data and / or the local correlation relationship between the target watermark information and the frequency domain target data can be determined. The filtering mechanism can select a more suitable embedding position for the target watermark information in the frequency domain target data (it may be the high-frequency part that is more conducive to the embedding of the watermark, it may be the low-frequency part that is more conducive to the embedding of the watermark, it may also include both the high-frequency part that is more conducive to the embedding of the watermark and the low-frequency part that is more conducive to the embedding of the watermark, etc.), and is not limited to the designated embedding position such as the high-frequency part or the specific frequency band of the designated channel. Based on this, in the process of embedding the target watermark information into the frequency data corresponding to the target data, the global correlation relationship between the target watermark information and the frequency domain target data and / or the local correlation relationship between the target watermark information and the frequency domain target data can be determined through the adaptive filtering mechanism, and based on the global correlation relationship and / or the local correlation relationship between the target watermark information and the frequency domain target data, the target watermark information can be embedded in the frequency domain target data to obtain the frequency domain watermarked data, and then the frequency domain watermarked data can be converted into the time domain watermarked data.The above is a processing process of a watermark embedding module. For multiple watermark embedding modules connected in sequence, the first watermark embedding module can output a time domain data (data in the time domain carrying the target watermark information), and the output time domain data can be input into the second watermark embedding module. The time domain data is subjected to time-frequency transformation through the network layer pre-set in the second watermark embedding module to convert the time domain data into frequency domain target data. Then, the global correlation relationship between the target watermark information and the frequency domain target data corresponding to the time domain data and / or the local correlation relationship between the target watermark information and the frequency domain target data corresponding to the time domain data can be determined through the adaptive filtering mechanism. Based on the global correlation relationship and / or the local correlation relationship between the target watermark information and the frequency domain target data corresponding to the time domain data, the target watermark information can be embedded into the frequency domain target data corresponding to the time domain data to obtain the corresponding frequency domain watermarked data, which can then be converted into the corresponding time domain watermarked data. The subsequent watermark embedding modules repeat the above processing process until the last watermark embedding module, thereby obtaining the corresponding output data (i.e., the time domain watermarked data).

[0024] In step S106, the time-domain watermarked data finally output by one or more sequentially connected watermark embedding modules are adjusted through the data adjustment module in the stego-coding model to obtain target watermarked data corresponding to the target data.

[0025] There is almost no or no visual difference between the target watermarked data and the target data, that is, the similarity between the target watermarked data and the target data is greater than a preset threshold, and the threshold can be set according to the fact that there is almost no or no visual difference between the target watermarked data and the target data. The data adjustment module can be implemented in a variety of ways, for example, the data adjustment module can be constructed by a linearization algorithm and / or a data reshaping algorithm, etc., which can be set according to actual conditions.

[0026] In implementation, through the above processing, one or more sequentially connected watermark embedding modules can eventually output time domain watermark data, and the output time domain watermark data can be input into the data adjustment module. The data adjustment module can perform tuning processing on the above final output watermark data to obtain watermark data with better effect. Finally, the adjusted watermark data can be used as the target watermark data corresponding to the target data.

[0027] The embodiment of the present specification provides an information steganography method, which obtains target data and target watermark information to be steganographic, and then converts the input target data or time-domain data into frequency-domain target data through one or more sequentially connected watermark embedding modules in a pre-trained steganographic coding model, embeds the target watermark information into the frequency-domain target data based on the global correlation relationship and / or local correlation relationship between the target watermark information and the frequency-domain target data, obtains the frequency-domain watermark data, and converts the frequency-domain watermark data into the time-domain watermark data; adjusts the time-domain watermark data finally output by one or more sequentially connected watermark embedding modules through a data adjustment module in the steganographic coding model. In this way, the inherent properties of the target data in the frequency domain are fully considered, and the global correlation and / or local correlation between the target watermark information and the frequency domain target data are introduced, and the watermark is adaptively embedded, thereby enhancing the ability to resist watermark removal while maintaining resistance to detection, and the watermark is more invisible. In addition, this scheme does not select a fixed frequency band for watermark embedding, but considers the global correlation and / or local correlation between the target watermark information and the frequency domain target data, and adaptively selects the frequency band for watermark fusion (or watermark embedding), thereby further enhancing the ability to resist watermark removal and the watermark is more invisible.

[0028] In practical applications, the watermark embedding module may include a frequency domain adaptive fusion submodule and a feedforward network. Based on this, the specific processing methods of the step S104 may be varied. An optional processing method is provided below. Figure 2 As shown, the processing may specifically include the following steps S1042 and S1044.

[0029] In step S1042, the input target data or time domain data is converted into frequency domain target data through the frequency domain adaptive fusion submodule in the watermark embedding module, and based on the global correlation relationship and / or local correlation relationship between the target watermark information and the frequency domain target data, the target watermark information is embedded in the frequency domain target data to obtain the frequency domain watermarked data, and the frequency domain watermarked data is converted into the time domain watermarked data.

[0030] Among them, the frequency domain adaptive fusion submodule can be constructed by a specified algorithm and / or network model, such as the frequency domain adaptive fusion submodule can be constructed by a time-frequency conversion algorithm, a neural network, an adaptive filtering network, etc., which can be specifically set according to actual conditions. The frequency domain adaptive fusion submodule can be used to convert the input target data or time domain data into frequency domain target data, embed the target watermark information into the frequency domain target data based on the global correlation relationship and / or local correlation relationship between the target watermark information and the frequency domain target data, obtain the frequency domain watermarked data, and convert the frequency domain watermarked data into the time domain watermarked data.

[0031] The specific processing process of the above step S1042 can refer to the relevant content in the above step S104, which will not be repeated here.

[0032] In step S1044, the watermarked data in the time domain converted by the frequency domain adaptive fusion submodule and the fused data of the target data are mapped through the feedforward network in the watermark embedding module.

[0033] Among them, the feedforward network can be constructed by a specified algorithm and / or network model. For example, the feedforward network can be constructed by a feedforward neural network FNN, etc., which can be set according to actual conditions. The feedforward network can include an input layer, a hidden layer, and an output layer. The information flow can only be transmitted in one direction from the input layer-hidden layer-output layer, and feedback usually does not occur. Through the feedforward network, the input data or sequence can be nonlinearly transformed to introduce more nonlinear capabilities to the stego coding model, thereby improving the expressiveness and generalization capabilities of the stego coding model. In practical applications, the feedforward network can be composed of two linear layers and a nonlinear activation function. This structure allows the input data (specifically, it can be a vector) to be mapped to a higher dimensional space and different feature information is fused through nonlinear transformation. The feedforward network can process the input data in the following way: the input data is mapped to a high-dimensional space through a linear transformation, then nonlinearly transformed through a ReLU activation function, and finally mapped back to a low-dimensional space. The above processing not only extracts deeper features, but also strengthens the expressiveness of the input data, so that the expression effect of the relevant part is better. In addition, the feedforward network also pays attention to the information of the location of the data. It can enhance the processing ability of the steganographic coding model for information at different locations without changing the information of the original location, so that the steganographic coding model can better understand and generate data with specific location information.

[0034] In implementation, Figure 3 As shown, the time-domain watermarked data converted by the frequency-domain adaptive fusion submodule can be fused with the target data (for example, the time-domain watermarked data and the target data can be superimposed or concatenated) to obtain fused data, which can be input into the feedforward network in the watermark embedding module, mapped to a high-dimensional space through the linear transformation mechanism of the feedforward network, then nonlinearly transformed through the ReLU activation function, and finally mapped back to a low-dimensional space, so that the effect of the watermarked data is better (that is, there is almost no or no visual difference between the watermarked data and the target data).

[0035] In practical applications, such as Figure 3As shown, the frequency domain adaptive fusion submodule includes a frequency domain conversion unit, a fusion unit, an adaptive filtering unit, a time domain conversion unit and a watermark encoding unit. Based on this, the specific processing method of the above step S1042 can be various. The following is an optional processing method, such as Figure 4 As shown, the processing may specifically include the following steps S10422 to S10428.

[0036] In step S10422, the input target data or time domain data is converted into frequency domain target data through a frequency domain conversion unit.

[0037] In step S10424, the target watermark information is encoded by the watermark encoding unit to obtain watermark encoding information.

[0038] In step S10426, the watermark coding information is embedded into the frequency domain target data through the fusion unit to obtain the frequency domain watermarked data, and in the process of embedding the watermark coding information into the frequency domain target data, the global correlation relationship and / or local correlation relationship between the watermark coding information and the frequency domain target data is determined through the adaptive filtering unit.

[0039] Among them, a long-range adaptive filter (AdaptiveFilter) can be set in the adaptive filtering unit. The long-range adaptive filter is a digital filter that can automatically adjust the performance according to the input data to process digital signals. For some applications, since the parameters to be operated (such as the characteristics of some noise data, etc.) are not known in advance, it is required to use adaptive coefficients for processing. In this case, an adaptive filter is usually used. The adaptive filter uses feedback to adjust the filter coefficients and frequency response. The adaptive process involves using a cost function to determine how to change the filter coefficients to reduce the cost of the next iteration process. Based on the above content, the long-range adaptive filter is provided with the adaptive filtering mechanism as described above. In implementation, in the embodiments of this specification, the new and efficient watermark steganography mechanism of FAST (Frequency-Adaptive STeganography) can be used to embed the target watermark information into the frequency domain of the target data. In the process of watermark steganography, the Transformer architecture is used to perform adaptive long-range fusion in the spatial domain and the frequency domain. The FAST watermark steganography mechanism explores the potential of watermark embedding learning in the spatial domain, and fully utilizes the spectral characteristics in the frequency domain through adaptability and long-range fusion. Considering that the frequency domain usually exhibits superior performance, since it can effectively separate and represent data structure and texture, thereby enhancing robustness and invisibility without compromising visual quality, watermark fusion in the frequency domain is usually better than that in the spatial domain. Therefore, the FAST watermark steganography mechanism places special emphasis on long-term adaptive conditioning on the frequency spectrum. Specifically, the FAST watermark steganography mechanism performs watermark processing under frequency domain conditions and uses a long-range adaptive filter. This long-range adaptive filter cleverly navigates the global and local relationships in the fusion process of the target watermark information and the target data, and adjusts the intensity distribution of the target watermark information in different frequency bands. In addition, given the transformation relationship between the frequency domain and the spatial domain, the long-range adaptive filter can be used as an effective alternative to spatial self-attention, thereby reducing computational complexity.

[0040] In step S10428, the frequency domain watermark data is converted into time domain watermark data through a time domain conversion unit.

[0041] In practical applications, the frequency domain conversion unit is a unit constructed based on a Fourier transform algorithm, a discrete cosine transform algorithm or a discrete wavelet transform algorithm, and the watermark encoding unit is a unit constructed by a multi-layer perceptron.

[0042] In the related processing of the above steps S10422 to S10428, by utilizing the global context perception capability of the Transformer architecture for watermark embedding, the FAST watermark steganography mechanism exhibits good robustness in the face of various attacks while avoiding significant data redundancy. Moreover, the frequency domain conditional variant of the FAST watermark steganography mechanism can not only flexibly integrate watermark information over a wide frequency range, but also achieve this high efficiency with fewer model parameters. In addition, the FAST watermark steganography mechanism introduces a Transformer conditioning method in the spectrum field, so that it can seamlessly adapt to a variety of other data editing tasks.

[0043] In practical applications, if the size of the target data is large, the target data can also be divided into blocks, and then each data block is processed, such as Figure 5 As shown, please refer to the processing of the following steps S110 and S112 for details.

[0044] In step S108, the target data is divided into blocks to obtain a plurality of different data blocks.

[0045] In implementation, for example, Figure 3 As shown, the target data can be divided into 8x8 or 4x4 blocks to obtain multiple different data blocks.

[0046] In step S110, each data block is encoded by the data block encoding unit in the steganographic coding model to obtain data block encoding information.

[0047] Among them, Figure 3 As shown, the data block encoding information may be a vector, specifically an embedded feature vector, etc., which may be specifically set according to actual conditions.

[0048] Based on the processing of the above steps S108 and S110, the specific processing methods of the above step S104 can be various. An optional processing method is provided below, which may specifically include the following: converting the input data block encoding information into a frequency domain target data block through one or more sequentially connected watermark embedding modules in a pre-trained steganographic coding model, embedding the target watermark information into the frequency domain target data block based on the global correlation relationship and / or local correlation relationship between the target watermark information and the frequency domain target data block to obtain a frequency domain watermarked data block, and converting the frequency domain watermarked data block into a time domain watermarked data block, and splicing multiple time domain watermarked data blocks into time domain watermarked data.

[0049] The above steganalysis coding model can be obtained by model training in the following ways: Figure 6 As shown, please refer to the processing of the following steps A02 to A14 for details.

[0050] In step A02, sample data and watermark sample information are obtained.

[0051] The sample data may be any sample data to be steganographic, for example, the sample data may be image data, text data, video data, etc., which may be set according to actual conditions, and the embodiments of this specification do not limit this. The watermark sample information may be any watermark information that needs to be embedded in the sample data. In practical applications, the watermark sample information may be generated in a variety of ways, for example, using a pseudo-random sequence generator or a chaotic system, etc. The watermark sample information may also be a meaningful binary image, grayscale image, or color image, etc.

[0052] In step A04, the input sample data or time domain data is converted into frequency domain sample data through one or more sequentially connected watermark embedding modules in the steganographic coding model, and the watermark sample information is embedded into the frequency domain sample data based on the global correlation relationship and / or local correlation relationship between the watermark sample information and the frequency domain sample data to obtain the frequency domain watermark sample, and the frequency domain watermark sample is converted into the time domain watermark sample. The specific processing process of the above step A04 can be found in the above related content, which will not be repeated here.

[0053] In step A08, the data adjustment module in the stego-coding model is used to adjust the time-domain watermarked samples finally output by one or more sequentially connected watermark embedding modules to obtain target watermarked samples corresponding to the sample data.

[0054] The specific processing process of the above step A06 can refer to the above related content and will not be repeated here.

[0055] In step A10, interference processing is performed on the target watermarked sample by using the interference model to obtain an interfered target watermarked sample.

[0056] In practice, the target watermarked sample can be attacked, specifically, by Figure 6 The interference model set in the middle noise layer performs interference processing (such as cropping, scribing, encoding, scaling, rotating, etc.) on the target watermarked sample to attack the target watermarked sample and obtain the interfered target watermarked sample.

[0057] In step A12, a watermark decoding process is performed based on the input interfered target watermark sample and the target watermark sample through a steganographic decoding model to obtain target watermark sample information.

[0058] In step A14, based on the watermark sample information, the target watermark sample information, the sample data and the time domain watermarked samples finally output by one or more sequentially connected watermark embedding modules, the steganalysis encoding model, the interference model and the steganalysis decoding model are jointly trained through a preset loss function to obtain a trained steganalysis encoding model.

[0059] Among them, the loss function may include multiple types, for example, mean square error loss function, cross entropy loss function, similarity loss function, etc., which can be set according to actual conditions.

[0060] In implementation, the loss information between the watermark sample information and the target watermark sample information can be calculated through the mean square error loss function. The similarity between the sample data and the watermarked sample in the time domain finally output by one or more sequentially connected watermark embedding modules can be calculated, and the maximization of the similarity is used as the loss information between the two. The model parameters of the steganalysis encoding model, the interference model and the steganalysis decoding model are adjusted through the above loss information. Then, the above processing process can be repeated to iteratively train the steganalysis encoding model, the interference model and the steganalysis decoding model to obtain the trained steganalysis encoding model, the trained interference model and the trained steganalysis decoding model.

[0061] In practical applications, the above-mentioned loss function includes a first sub-loss function and a second sub-loss function. The first sub-loss function is a loss function between watermark sample information and target watermark sample information, and the second sub-loss function is a loss function between sample data and the time domain watermark-bearing sample finally output by one or more sequentially connected watermark embedding modules. The first sub-loss function includes a mean square error loss function, and the second sub-loss function includes one or more of a mean square error loss function, a mean absolute error loss function, and a perceptual similarity loss function.

[0062] In practical applications, the second sub-loss function can be obtained by the mean square error loss function + mean absolute error loss function + perceptual similarity loss function.

[0063] In practical applications, the target data includes one or more of image data, text data, audio data and video data, and the target watermark information includes one or more of image data, character string, audio data and video data.

[0064] The embodiment of the present specification provides an information steganography method, which obtains target data and target watermark information to be steganographic, and then converts the input target data or time-domain data into frequency-domain target data through one or more sequentially connected watermark embedding modules in a pre-trained steganographic coding model, embeds the target watermark information into the frequency-domain target data based on the global correlation relationship and / or local correlation relationship between the target watermark information and the frequency-domain target data, obtains the frequency-domain watermark data, and converts the frequency-domain watermark data into the time-domain watermark data; adjusts the time-domain watermark data finally output by one or more sequentially connected watermark embedding modules through a data adjustment module in the steganographic coding model. In this way, the inherent properties of the target data in the frequency domain are fully considered, and the global correlation and / or local correlation between the target watermark information and the frequency domain target data are introduced, and the watermark is adaptively embedded, thereby enhancing the ability to resist watermark removal while maintaining resistance to detection, and the watermark is more invisible. In addition, this scheme does not select a fixed frequency band for watermark embedding, but considers the global correlation and / or local correlation between the target watermark information and the frequency domain target data, and adaptively selects the frequency band for watermark fusion (or watermark embedding), thereby further enhancing the ability to resist watermark removal and the watermark is more invisible.

[0065] In addition, by leveraging the global context-awareness of the Transformer architecture for watermark embedding, the FAST watermark steganography mechanism exhibits good robustness in the face of various attacks while avoiding significant data redundancy. Moreover, the frequency domain conditional variant of the FAST watermark steganography mechanism can not only flexibly integrate watermark information over a wide frequency range, but also achieve this high efficiency with fewer model parameters. In addition, the FAST watermark steganography mechanism introduces a Transformer conditioning method in the spectral domain, which can seamlessly adapt to a variety of other data editing tasks.

[0066] The following is a detailed description of an information steganography method provided by an embodiment of this specification in combination with a specific application scenario, wherein the target data is image data, the sample data may be an image sample, the target watermark information may be a string watermark, and the watermark sample information may be a string watermark sample.

[0067] The embodiments of this specification provide an information steganography method, the execution subject of the method may be a terminal device or a server, etc., wherein the terminal device may be a mobile terminal device such as a mobile phone, a tablet computer, or a computer device such as a laptop or a desktop computer, or an IoT device (specifically a smart watch, a vehicle-mounted device, etc.), etc., wherein the server may be an independent server, or a server cluster composed of multiple servers, etc., and the server may be a background server in the financial field or the online shopping field, etc., or a background server of an application, etc. In this embodiment, the execution subject is taken as an example for detailed description, and for the case where the execution subject is a terminal device, please refer to the following server case processing, which will not be repeated here. The method may specifically include the following steps: In step B02, image samples and character string watermark samples are obtained.

[0068] In step B04, the image sample is divided into blocks to obtain a plurality of different sample image blocks.

[0069] In step B06, each sample image block is encoded by the data block encoding unit in the steganographic coding model to obtain sample image block encoding information.

[0070] In step B08, the input sample image block encoding information is converted into a frequency domain sample image block by the frequency domain conversion unit in the frequency domain adaptive fusion submodule in the watermark embedding module.

[0071] In step B10, the watermark encoding unit in the frequency domain adaptive fusion submodule in the watermark embedding module encodes the character string watermark sample to obtain watermark sample encoding information.

[0072] In step B12, the watermark sample coding information is embedded into the frequency domain sample image block through the fusion unit in the frequency domain adaptive fusion submodule in the watermark embedding module to obtain the frequency domain watermark sample image block, and in the process of embedding the watermark sample coding information into the frequency domain sample image block, the global correlation relationship and / or local correlation relationship between the watermark sample coding information and the frequency domain sample image block is determined through the adaptive filtering unit.

[0073] In step B14, the watermark sample image block in the frequency domain is converted into a watermark sample image block in the time domain by the time domain conversion unit in the frequency domain adaptive fusion submodule in the watermark embedding module.

[0074] In step B16, the time domain watermarked sample image blocks converted by the frequency domain adaptive fusion submodule and the fused data of the sample image blocks are mapped through the feedforward network in the watermark embedding module, and multiple time domain watermarked sample image blocks are spliced ​​into time domain watermarked image samples.

[0075] In step B18, the time domain watermarked image samples finally output by one or more sequentially connected watermark embedding modules are adjusted through the data adjustment module in the stego-coding model to obtain target watermarked image samples corresponding to the image samples.

[0076] In step B20, interference processing is performed on the target watermarked image sample by using the interference model to obtain an interfered target watermarked image sample.

[0077] In step B22, a watermark decoding process is performed based on the input disturbed target watermark image sample and the target watermark image sample through the steganalysis decoding model to obtain a target character string watermark sample.

[0078] In step B24, based on the string watermark samples, the target string watermark samples, the image samples and the time domain watermarked image samples finally output by one or more sequentially connected watermark embedding modules, the steganalysis encoding model, the interference model and the steganalysis decoding model are jointly trained through a preset loss function to obtain a trained steganalysis encoding model.

[0079] Among them, the loss function includes a first sub-loss function and a second sub-loss function, the first sub-loss function is a loss function between a string watermark sample and a target string watermark sample, the second sub-loss function is a loss function between an image sample and a time domain watermarked image sample finally output by one or more sequentially connected watermark embedding modules, the first sub-loss function includes a mean square error loss function, and the second sub-loss function includes one or more of a mean square error loss function, a mean absolute error loss function and a perceptual similarity loss function.

[0080] During the training of the above model, the new and efficient watermark steganography mechanism of FAST can be used to embed the string watermark into the frequency domain of the image data. In the process of watermark steganography, the Transformer architecture is used to perform adaptive long-range fusion in the spatial and frequency domains. The FAST watermark steganography mechanism explores the potential of image embedding learning in the spatial domain and fully utilizes the spectral characteristics in the frequency domain through adaptive and long-range fusion. In order to learn image embedding in the spatial domain, a block method based on the Transformer architecture is selected because its inherent global conditional mechanism can integrate conditions in the entire sequence generation process. Considering that the frequency domain usually exhibits superior performance, since it can effectively separate and represent image structure and texture, thereby enhancing robustness and invisibility without compromising visual quality, watermark fusion in the frequency domain is usually better than the spatial domain method. Therefore, the FAST watermark steganography mechanism places special emphasis on long-term adaptive conditioning on the frequency spectrum. Specifically, the FAST watermark steganography mechanism performs watermark processing under frequency domain conditions and uses a long-range adaptive filter. This long-range adaptive filter cleverly navigates the global and local relationships in the process of fusion of string watermark and target image data, and adjusts the intensity distribution of string watermark in different frequency bands. In addition, given the transformation relationship between the frequency domain and the spatial domain, the long-range adaptive filter can be used as an effective alternative to spatial self-attention, thereby reducing computational complexity.

[0081] In step B26, the image data to be steganographic and the character string watermark are obtained.

[0082] In step B28, the image data is divided into blocks to obtain a plurality of different image blocks.

[0083] In step B30, each image block is encoded by the data block encoding unit in the steganographic coding model to obtain image block encoding information.

[0084] In step B32, the input image block coding information or the image block data in the time domain is converted into frequency domain image block data through the frequency domain conversion unit.

[0085] In step B34, the watermark encoding unit encodes the character string watermark to obtain watermark encoding information.

[0086] In step B36, the watermark coding information is embedded into the frequency domain image block data through the fusion unit to obtain the frequency domain watermarked image block, and in the process of embedding the watermark coding information into the frequency domain image block data, the global correlation relationship and / or local correlation relationship between the watermark coding information and the frequency domain image block data is determined through the adaptive filtering unit.

[0087] In step B38, the frequency domain watermarked image block is converted into a time domain watermarked image block by the time domain conversion unit.

[0088] In step B40, the time domain watermarked image block converted by the frequency domain adaptive fusion submodule is mapped to the fusion data of the image block through the feedforward network in the watermark embedding module, and multiple time domain watermarked image blocks are spliced ​​into a time domain watermarked image.

[0089] In step B42, the time-domain watermarked image finally output by one or more sequentially connected watermark embedding modules is adjusted through the data adjustment module in the steganographic coding model to obtain the target watermarked image corresponding to the image data.

[0090] The embodiment of the present specification provides an information steganography method, which obtains target data and target watermark information to be steganographic, and then converts the input target data or time-domain data into frequency-domain target data through one or more sequentially connected watermark embedding modules in a pre-trained steganographic coding model, embeds the target watermark information into the frequency-domain target data based on the global correlation relationship and / or local correlation relationship between the target watermark information and the frequency-domain target data, obtains the frequency-domain watermark data, and converts the frequency-domain watermark data into the time-domain watermark data; adjusts the time-domain watermark data finally output by one or more sequentially connected watermark embedding modules through a data adjustment module in the steganographic coding model. In this way, the inherent properties of the target data in the frequency domain are fully considered, and the global correlation and / or local correlation between the target watermark information and the frequency domain target data are introduced, and the watermark is adaptively embedded, thereby enhancing the ability to resist watermark removal while maintaining resistance to detection, and the watermark is more invisible. In addition, this scheme does not select a fixed frequency band for watermark embedding, but considers the global correlation and / or local correlation between the target watermark information and the frequency domain target data, and adaptively selects the frequency band for watermark fusion (or watermark embedding), thereby further enhancing the ability to resist watermark removal and the watermark is more invisible.

[0091] In addition, by leveraging the global context-awareness of the Transformer architecture for watermark embedding, the FAST watermark steganography mechanism exhibits good robustness in the face of various attacks while avoiding significant data redundancy. Moreover, the frequency domain conditional variant of the FAST watermark steganography mechanism can not only flexibly integrate watermark information over a wide frequency range, but also achieve this high efficiency with fewer model parameters. In addition, the FAST watermark steganography mechanism introduces a Transformer conditioning method in the spectral domain, which can seamlessly adapt to a variety of other image editing tasks.

[0092] The above is the information steganography method provided in the embodiment of this specification. Based on the same idea, the embodiment of this specification also provides an information steganography device, such as Figure 7 shown.

[0093] The information steganography device comprises: a data acquisition module 701, a watermark embedding module 702 and an adjustment module 703, wherein: Data acquisition module 701, acquires target data to be steganographic and target watermark information; The watermark embedding module 702 converts the input target data or time domain data into frequency domain target data through one or more sequentially connected watermark embedding modules in the pre-trained stego coding model, embeds the target watermark information into the frequency domain target data based on the global correlation relationship and / or local correlation relationship between the target watermark information and the frequency domain target data to obtain frequency domain watermark data, and converts the frequency domain watermark data into time domain watermark data; The adjustment module 703 adjusts the time-domain watermarked data finally output by one or more sequentially connected watermark embedding modules through the data adjustment module in the stego-coding model to obtain the target watermarked data corresponding to the target data.

[0094] In the embodiment of this specification, the watermark embedding module 702 includes: A watermark embedding unit, which converts the input target data or time domain data into frequency domain target data through the frequency domain adaptive fusion submodule in the watermark embedding module, embeds the target watermark information into the frequency domain target data based on the global correlation relationship and / or local correlation relationship between the target watermark information and the frequency domain target data to obtain frequency domain watermarked data, and converts the frequency domain watermarked data into time domain watermarked data; The feedforward unit maps the watermarked data in the time domain converted by the frequency domain adaptive fusion submodule to the fused data of the target data through the feedforward network in the watermark embedding module.

[0095] In the embodiment of the present specification, the frequency domain adaptive fusion submodule includes a frequency domain conversion unit, a fusion unit, an adaptive filtering unit, a time domain conversion unit and a watermark encoding unit. The watermark embedding unit converts the input target data or time domain data into frequency domain target data through the frequency domain conversion unit; encodes the target watermark information through the watermark encoding unit to obtain watermark encoding information; embeds the watermark encoding information into the frequency domain target data through the fusion unit to obtain frequency domain watermarked data, and in the process of embedding the watermark encoding information into the frequency domain target data, determines the global correlation relationship and / or local correlation relationship between the watermark encoding information and the frequency domain target data through the adaptive filtering unit; and converts the frequency domain watermarked data into time domain watermarked data through the time domain conversion unit.

[0096] In the embodiments of this specification, the frequency domain conversion unit is a unit constructed based on Fourier transform algorithm, discrete cosine transform algorithm or discrete wavelet transform algorithm, and the watermark encoding unit is a unit constructed by a multi-layer perceptron.

[0097] In the embodiment of this specification, the device further includes: A block division module divides the target data into blocks to obtain a plurality of different data blocks; A data encoding module, which encodes each data block through a data block encoding unit in the stego-coding model to obtain data block encoding information; The watermark embedding module 702 converts the input data block encoding information into a frequency domain target data block through one or more sequentially connected watermark embedding modules in a pre-trained steganographic coding model, embeds the target watermark information into the frequency domain target data block based on the global correlation relationship and / or local correlation relationship between the target watermark information and the frequency domain target data block, obtains a frequency domain watermarked data block, converts the frequency domain watermarked data block into a time domain watermarked data block, and splices multiple time domain watermarked data blocks into time domain watermarked data.

[0098] In the embodiment of this specification, the device further includes: A sample data acquisition module, which acquires sample data and watermark sample information; A sample watermark embedding module, which converts the input sample data or time domain data into frequency domain sample data through one or more sequentially connected watermark embedding modules in the stego-coding model, embeds the watermark sample information into the frequency domain sample data based on the global correlation relationship and / or local correlation relationship between the watermark sample information and the frequency domain sample data to obtain a frequency domain watermark sample, and converts the frequency domain watermark sample into a time domain watermark sample; A sample adjustment module, which adjusts the time-domain watermarked samples finally output by one or more sequentially connected watermark embedding modules through the data adjustment module in the stego-coding model, to obtain a target watermarked sample corresponding to the sample data; An interference module performs interference processing on the target watermarked sample through an interference model to obtain an interfered target watermarked sample; A decoding module, which performs watermark decoding processing based on the input interfered target watermark sample and the target watermark sample through a stego decoding model to obtain target watermark sample information; The training module jointly trains the steganalysis encoding model, the interference model and the steganalysis decoding model through a preset loss function based on the watermark sample information, the target watermark sample information, the sample data and the time-domain watermarked samples finally output by the one or more sequentially connected watermark embedding modules to obtain a trained steganalysis encoding model.

[0099] In an embodiment of the present specification, the loss function includes a first sub-loss function and a second sub-loss function, the first sub-loss function is a loss function between the watermark sample information and the target watermark sample information, the second sub-loss function is a loss function between the sample data and the time domain watermarked sample finally output by one or more sequentially connected watermark embedding modules, the first sub-loss function includes a mean square error loss function, and the second sub-loss function includes one or more of a mean square error loss function, a mean absolute error loss function and a perceptual similarity loss function.

[0100] In the embodiment of this specification, the target data includes one or more of image data, text data, audio data and video data, and the target watermark information includes one or more of image data, character string, audio data and video data.

[0101] The embodiment of the present specification provides an information steganography device, which obtains target data and target watermark information to be steganographic, and then converts the input target data or time-domain data into frequency-domain target data through one or more sequentially connected watermark embedding modules in a pre-trained steganographic coding model, embeds the target watermark information into the frequency-domain target data based on the global correlation relationship and / or local correlation relationship between the target watermark information and the frequency-domain target data, obtains the frequency-domain watermark data, and converts the frequency-domain watermark data into the time-domain watermark data; adjusts the time-domain watermark data finally output by one or more sequentially connected watermark embedding modules through a data adjustment module in the steganographic coding model. In this way, the inherent properties of the target data in the frequency domain are fully considered, and the global correlation and / or local correlation between the target watermark information and the frequency domain target data are introduced, and the watermark is adaptively embedded, thereby enhancing the ability to resist watermark removal while maintaining resistance to detection, and the watermark is more invisible. In addition, this scheme does not select a fixed frequency band for watermark embedding, but considers the global correlation and / or local correlation between the target watermark information and the frequency domain target data, and adaptively selects the frequency band for watermark fusion (or watermark embedding), thereby further enhancing the ability to resist watermark removal and the watermark is more invisible.

[0102] In addition, by leveraging the global context-awareness of the Transformer architecture for watermark embedding, the FAST watermark steganography mechanism exhibits good robustness in the face of various attacks while avoiding significant data redundancy. Moreover, the frequency domain conditional variant of the FAST watermark steganography mechanism can not only flexibly integrate watermark information over a wide frequency range, but also achieve this high efficiency with fewer model parameters. In addition, the FAST watermark steganography mechanism introduces a Transformer conditioning method in the spectral domain, which can seamlessly adapt to a variety of other data editing tasks.

[0103] The above is an information steganography device provided in the embodiment of this specification. Based on the same idea, the embodiment of this specification also provides an information steganography device, such as Figure 8 shown.

[0104] The information steganography device may provide a terminal device or a server, etc. for the above-mentioned embodiment.

[0105] The information steganography device may have relatively large differences due to different configurations or performances, and may include one or more processors 801 and memory 802, and the memory 802 may store one or more storage applications or data. Among them, the memory 802 may be a short-term storage or a persistent storage. The application stored in the memory 802 may include one or more modules (not shown in the figure), and each module may include a series of computer executable instructions in the information steganography device. Furthermore, the processor 801 may be configured to communicate with the memory 802 to execute a series of computer executable instructions in the memory 802 on the information steganography device. The information steganography device may also include one or more power supplies 803, one or more wired or wireless network interfaces 804, one or more input and output interfaces 805, and one or more keyboards 806.

[0106] Specifically, in this embodiment, the information steganography device includes a memory and one or more programs, wherein the one or more programs are stored in the memory, and the one or more programs may include one or more modules, and each module may include a series of computer executable instructions in the information steganography device, and the one or more programs are configured to be executed by one or more processors, including the following computer executable instructions: Obtaining target data and target watermark information to be steganographic; The input target data or time-domain data is converted into frequency-domain target data through one or more sequentially connected watermark embedding modules in a pre-trained stego-coding model, and the target watermark information is embedded into the frequency-domain target data based on the global correlation relationship and / or local correlation relationship between the target watermark information and the frequency-domain target data to obtain frequency-domain watermarked data, and the frequency-domain watermarked data is converted into time-domain watermarked data; The data adjustment module in the stego-coding model adjusts the time-domain watermarked data finally output by one or more sequentially connected watermark embedding modules to obtain target watermarked data corresponding to the target data.

[0107] Each embodiment in this specification is described in a progressive manner, and the same or similar parts between the embodiments can be referred to each other, and each embodiment focuses on the differences from other embodiments. In particular, for the information steganography device embodiment, since it is basically similar to the method embodiment, the description is relatively simple, and the relevant parts can be referred to the partial description of the method embodiment.

[0108] The embodiment of the present specification provides an information steganography device, which obtains target data and target watermark information to be steganographic, and then converts the input target data or time-domain data into frequency-domain target data through one or more sequentially connected watermark embedding modules in a pre-trained steganographic coding model, embeds the target watermark information into the frequency-domain target data based on the global correlation relationship and / or local correlation relationship between the target watermark information and the frequency-domain target data, obtains the frequency-domain watermark data, and converts the frequency-domain watermark data into the time-domain watermark data; adjusts the time-domain watermark data finally output by one or more sequentially connected watermark embedding modules through a data adjustment module in the steganographic coding model. In this way, the inherent properties of the target data in the frequency domain are fully considered, and the global correlation and / or local correlation between the target watermark information and the frequency domain target data are introduced, and the watermark is adaptively embedded, thereby enhancing the ability to resist watermark removal while maintaining resistance to detection, and the watermark is more invisible. In addition, this scheme does not select a fixed frequency band for watermark embedding, but considers the global correlation and / or local correlation between the target watermark information and the frequency domain target data, and adaptively selects the frequency band for watermark fusion (or watermark embedding), thereby further enhancing the ability to resist watermark removal and the watermark is more invisible.

[0109] Furthermore, based on the above Figures 1 to 6 In one embodiment, the present specification further provides a storage medium for storing computer executable instruction information. In a specific embodiment, the storage medium may be a USB flash drive, an optical disk, a hard disk, etc. When the computer executable instruction information stored in the storage medium is executed by the processor, the following process can be implemented: Obtaining target data and target watermark information to be steganographic; The input target data or time-domain data is converted into frequency-domain target data through one or more sequentially connected watermark embedding modules in a pre-trained stego-coding model, and the target watermark information is embedded into the frequency-domain target data based on the global correlation relationship and / or local correlation relationship between the target watermark information and the frequency-domain target data to obtain frequency-domain watermarked data, and the frequency-domain watermarked data is converted into time-domain watermarked data; The data adjustment module in the stego-coding model adjusts the time-domain watermarked data finally output by one or more sequentially connected watermark embedding modules to obtain target watermarked data corresponding to the target data.

[0110] Each embodiment in this specification is described in a progressive manner, and the same or similar parts between the embodiments can be referred to each other, and each embodiment focuses on the differences from other embodiments. In particular, for the above-mentioned storage medium embodiment, since it is basically similar to the method embodiment, the description is relatively simple, and the relevant parts can be referred to the partial description of the method embodiment.

[0111] The embodiment of the present specification provides a storage medium, which obtains target data and target watermark information to be steganographic, and then converts the input target data or time-domain data into frequency-domain target data through one or more sequentially connected watermark embedding modules in a pre-trained steganographic coding model, embeds the target watermark information into the frequency-domain target data based on the global correlation relationship and / or local correlation relationship between the target watermark information and the frequency-domain target data, obtains the frequency-domain watermark data, and converts the frequency-domain watermark data into the time-domain watermark data; adjusts the time-domain watermark data finally output by one or more sequentially connected watermark embedding modules through a data adjustment module in the steganographic coding model. , the target watermark data corresponding to the target data is obtained. In this way, the inherent properties of the target data in the frequency domain are fully considered, and the global correlation and / or local correlation between the target watermark information and the frequency domain target data are introduced, and the watermark is adaptively embedded, thereby enhancing the ability to resist watermark removal while maintaining resistance to detection, and the watermark is more invisible. In addition, this scheme will not select a fixed frequency band for watermark embedding, but considers the global correlation and / or local correlation between the target watermark information and the frequency domain target data, and adaptively selects the frequency band for watermark fusion (or watermark embedding), thereby further enhancing the ability to resist watermark removal, and the watermark is more invisible.

[0112] Furthermore, based on the above Figures 1 to 6 In one or more embodiments of the present specification, a computer program product is provided, including a computer program. When the computer program in the computer program product is executed by a processor, the following process can be implemented: Obtaining target data and target watermark information to be steganographic; The input target data or time-domain data is converted into frequency-domain target data through one or more sequentially connected watermark embedding modules in a pre-trained stego-coding model, and the target watermark information is embedded into the frequency-domain target data based on the global correlation relationship and / or local correlation relationship between the target watermark information and the frequency-domain target data to obtain frequency-domain watermarked data, and the frequency-domain watermarked data is converted into time-domain watermarked data; The data adjustment module in the stego-coding model adjusts the time-domain watermarked data finally output by one or more sequentially connected watermark embedding modules to obtain target watermarked data corresponding to the target data.

[0113] Each embodiment in this specification is described in a progressive manner, and the same or similar parts between the embodiments can be referred to each other, and each embodiment focuses on the differences from other embodiments. In particular, for the above-mentioned computer program product embodiment, since it is basically similar to the method embodiment, the description is relatively simple, and the relevant parts can be referred to the partial description of the method embodiment.

[0114] The embodiment of the present specification provides a computer program product, which obtains target data and target watermark information to be steganographic, and then converts the input target data or time-domain data into frequency-domain target data through one or more sequentially connected watermark embedding modules in a pre-trained steganographic coding model, embeds the target watermark information into the frequency-domain target data based on the global correlation relationship and / or local correlation relationship between the target watermark information and the frequency-domain target data, obtains the frequency-domain watermark data, and converts the frequency-domain watermark data into the time-domain watermark data; adjusts the time-domain watermark data finally output by one or more sequentially connected watermark embedding modules through a data adjustment module in the steganographic coding model. In this way, the inherent properties of the target data in the frequency domain are fully considered, and the global correlation and / or local correlation between the target watermark information and the frequency domain target data are introduced, and the watermark is adaptively embedded, thereby enhancing the ability to resist watermark removal while maintaining resistance to detection, and the watermark is more invisible. In addition, this scheme does not select a fixed frequency band for watermark embedding, but considers the global correlation and / or local correlation between the target watermark information and the frequency domain target data, and adaptively selects the frequency band for watermark fusion (or watermark embedding), thereby further enhancing the ability to resist watermark removal and the watermark is more invisible.

[0115] The above is a description of a specific embodiment of the specification. Other embodiments are within the scope of the appended claims. In some cases, the actions or steps recorded in the claims can be performed in an order different from that in the embodiments and still achieve the desired results. In addition, the processes depicted in the drawings do not necessarily require the specific order or continuous order shown to achieve the desired results. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0116] In the 1990s, it was very clear whether the improvement of a technology was hardware improvement (for example, improvement of the circuit structure of diodes, transistors, switches, etc.) or software improvement (improvement of the method flow). However, with the development of technology, many improvements of the method flow today can be regarded as direct improvements of the hardware circuit structure. Designers almost always obtain the corresponding hardware circuit structure by programming the improved method flow into the hardware circuit. Therefore, it cannot be said that the improvement of a method flow cannot be implemented with a hardware entity module. For example, a programmable logic device (PLD) (such as a field programmable gate array (FPGA)) is such an integrated circuit whose logical function is determined by the user's programming of the device. Designers can "integrate" a digital system on a PLD by programming themselves, without having to ask chip manufacturers to design and make dedicated integrated circuit chips. Moreover, nowadays, instead of manually making integrated circuit chips, this kind of programming is mostly implemented by "logic compiler" software, which is similar to the software compiler used when developing and writing programs, and the original code before compilation must also be written in a specific programming language, which is called hardware description language (HDL). There is not only one kind of HDL, but many kinds, such as ABEL (Advanced Boolean Expression Language), AHDL (Altera Hardware Description Language), Confluence, CUPL (Cornell University Programming Language), HDCal, JHDL (Java Hardware Description Language), Lava, Lola, MyHDL, PALASM, RHDL (Ruby Hardware Description Language), etc. The most commonly used ones are VHDL (Very-High-Speed ​​Integrated Circuit Hardware Description Language) and Verilog. Those skilled in the art should also know that it is only necessary to program the method flow slightly in the above-mentioned hardware description languages ​​and program it into the integrated circuit, and then it is easy to obtain the hardware circuit that implements the logic method flow.

[0117] The controller may be implemented in any suitable manner, for example, the controller may take the form of a microprocessor or processor and a computer-readable medium storing a computer-readable program code (e.g., software or firmware) executable by the (micro)processor, a logic gate, a switch, an application-specific integrated circuit (ASIC), a programmable logic controller, and an embedded microcontroller, examples of which include but are not limited to the following microcontrollers: ARC 625D, Atmel AT91SAM, Microchip PIC18F26K20, and Silicone Labs C8051F320, and the memory controller may also be implemented as part of the control logic of the memory. It is also known to those skilled in the art that, in addition to implementing the controller in a purely computer-readable program code manner, the controller may be implemented in the form of a logic gate, a switch, an application-specific integrated circuit, a programmable logic controller, and an embedded microcontroller by logically programming the method steps. Therefore, such a controller may be considered as a hardware component, and the devices for implementing various functions included therein may also be considered as structures within the hardware component. Or even, the devices for implementing various functions may be considered as both software modules for implementing the method and structures within the hardware component.

[0118] The systems, devices, modules or units described in the above embodiments may be implemented by computer chips or entities, or by products with certain functions. A typical implementation device is a computer. Specifically, the computer may be, for example, a personal computer, a laptop computer, a cellular phone, a camera phone, a smart phone, a personal digital assistant, a media player, a navigation device, an email device, a game console, a tablet computer, a wearable device, or a combination of any of these devices.

[0119] For the convenience of description, the above devices are described in terms of functions and are divided into various units. Of course, when implementing one or more embodiments of this specification, the functions of each unit can be implemented in the same or multiple software and / or hardware.

[0120] Those skilled in the art will appreciate that the embodiments of this specification may be provided as methods, systems, or computer program products. Therefore, one or more embodiments of this specification may take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware. Moreover, one or more embodiments of this specification may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program codes.

[0121] The embodiments of this specification are described with reference to flowcharts and / or block diagrams of methods, devices (systems), and computer program products according to the embodiments of this specification. It should be understood that each process and / or box in the flowchart and / or block diagram, as well as the combination of processes and / or boxes in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable fraud case serial and parallel device to produce a machine, so that the instructions executed by the processor of the computer or other programmable fraud case serial and parallel device generate instructions for implementing the processes in the process. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0122] These computer program instructions may also be stored in a computer readable memory capable of directing a computer or other programmable fraud case serial and parallel device to operate in a specific manner, so that the instructions stored in the computer readable memory produce an article of manufacture including an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0123] These computer program instructions may also be loaded onto a computer or other programmable device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, whereby the instructions executed on the computer or other programmable device provide for implementing the process. Figure 1 A process or multiple processes and / or boxes Figure 1 The steps for the functions specified in one or more boxes.

[0124] In a typical configuration, a computing device includes one or more processors (CPU), input / output interfaces, network interfaces, and memory.

[0125] Memory may include non-permanent storage in a computer-readable medium, random access memory (RAM) and / or non-volatile memory in the form of read-only memory (ROM) or flash RAM. Memory is an example of a computer-readable medium.

[0126] Computer readable media include permanent and non-permanent, removable and non-removable media that can be implemented by any method or technology to store information. Information can be computer readable instructions, data structures, program modules or other data. Examples of computer storage media include, but are not limited to, phase change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technology, compact disk read-only memory (CD-ROM), digital versatile disk (DVD) or other optical storage, magnetic cassettes, magnetic tape magnetic disk storage or other magnetic storage devices or any other non-transmission media that can be used to store information that can be accessed by a computing device. As defined in this article, computer readable media does not include temporary computer readable media (transitory media), such as modulated data signals and carrier waves.

[0127] It should also be noted that the terms "include", "comprises" or any other variations thereof are intended to cover non-exclusive inclusion, so that a process, method, commodity or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, commodity or device. In the absence of more restrictions, the elements defined by the sentence "comprises a ..." do not exclude the existence of other identical elements in the process, method, commodity or device including the elements.

[0128] Those skilled in the art will appreciate that the embodiments of this specification may be provided as methods, systems or computer program products. Therefore, one or more embodiments of this specification may be in the form of a complete hardware embodiment, a complete software embodiment or an embodiment combining software and hardware. Furthermore, one or more embodiments of this specification may be in the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0129] One or more embodiments of the present specification may be described in the general context of computer-executable instructions executed by a computer, such as program modules. Generally, program modules include routines, programs, objects, components, data structures, etc. that perform specific tasks or implement specific abstract data types. One or more embodiments of the present specification may also be practiced in distributed computing environments where tasks are performed by remote processing devices connected through a communication network. In a distributed computing environment, program modules may be located in local and remote computer storage media, including storage devices.

[0130] Each embodiment in this specification is described in a progressive manner, and the same or similar parts between the embodiments can be referred to each other, and each embodiment focuses on the differences from other embodiments. In particular, for the system embodiment, since it is basically similar to the method embodiment, the description is relatively simple, and the relevant parts can be referred to the partial description of the method embodiment.

[0131] The above description is only an embodiment of this specification and is not intended to limit this document. For those skilled in the art, this specification may have various changes and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of this specification should be included in the scope of the claims of this specification.

Claims

1. A method for information steganography, the method comprising: Obtaining target data and target watermark information to be steganographic; The input target data or time-domain data is converted into frequency-domain target data through one or more sequentially connected watermark embedding modules in a pre-trained stego-coding model, and the target watermark information is embedded into the frequency-domain target data based on the global correlation relationship and / or local correlation relationship between the target watermark information and the frequency-domain target data to obtain frequency-domain watermarked data, and the frequency-domain watermarked data is converted into time-domain watermarked data; The data adjustment module in the stego-coding model adjusts the time-domain watermarked data finally output by one or more sequentially connected watermark embedding modules to obtain target watermarked data corresponding to the target data.

2. According to the method of claim 1, the watermark embedding module comprises a frequency domain adaptive fusion submodule and a feedforward network, the one or more watermark embedding modules sequentially connected through the pre-trained steganographic coding model convert the input target data or time domain data into frequency domain target data, embed the target watermark information into the frequency domain target data based on the global correlation relationship and / or local correlation relationship between the target watermark information and the frequency domain target data to obtain frequency domain watermarked data, and convert the frequency domain watermarked data into time domain watermarked data, comprising: The input target data or time domain data is converted into frequency domain target data through the frequency domain adaptive fusion submodule in the watermark embedding module, and based on the global correlation relationship and / or local correlation relationship between the target watermark information and the frequency domain target data, the target watermark information is embedded into the frequency domain target data to obtain frequency domain watermarked data, and the frequency domain watermarked data is converted into time domain watermarked data; Through the feedforward network in the watermark embedding module, mapping processing is performed on the time domain watermarked data converted by the frequency domain adaptive fusion submodule and the fused data of the target data.

3. The method according to claim 2, wherein the frequency domain adaptive fusion submodule comprises a frequency domain conversion unit, a fusion unit, an adaptive filtering unit, a time domain conversion unit and a watermark encoding unit, wherein the input target data or time domain data is converted into frequency domain target data through the frequency domain adaptive fusion submodule in the watermark embedding module, and the target watermark information is embedded into the frequency domain target data based on the global correlation relationship and / or local correlation relationship between the target watermark information and the frequency domain target data to obtain frequency domain watermarked data, and the frequency domain watermarked data is converted into time domain watermarked data, comprising: The input target data or time domain data is converted into frequency domain target data by the frequency domain conversion unit; The target watermark information is encoded by the watermark encoding unit to obtain watermark encoding information; The watermark coding information is embedded into the frequency domain target data through the fusion unit to obtain the frequency domain watermarked data, and in the process of embedding the watermark coding information into the frequency domain target data, the global correlation relationship and / or local correlation relationship between the watermark coding information and the frequency domain target data is determined through the adaptive filtering unit; The frequency domain watermark data is converted into time domain watermark data by the time domain conversion unit.

4. The method according to claim 3, wherein the frequency domain conversion unit is a unit constructed based on a Fourier transform algorithm, a discrete cosine transform algorithm or a discrete wavelet transform algorithm, and the watermark encoding unit is a unit constructed by a multi-layer perceptron.

5. The method according to claim 2, further comprising: Dividing the target data into blocks to obtain a plurality of different data blocks; By means of the data block encoding unit in the stego-coding model, each data block is encoded to obtain data block encoding information; The one or more sequentially connected watermark embedding modules in the pre-trained stego-coding model convert the input target data or time-domain data into frequency-domain target data, embed the target watermark information into the frequency-domain target data based on the global correlation relationship and / or local correlation relationship between the target watermark information and the frequency-domain target data to obtain frequency-domain watermarked data, and convert the frequency-domain watermarked data into time-domain watermarked data, including: Through one or more sequentially connected watermark embedding modules in a pre-trained steganographic coding model, the input data block coding information is converted into a frequency domain target data block, and based on the global correlation relationship and / or local correlation relationship between the target watermark information and the frequency domain target data block, the target watermark information is embedded in the frequency domain target data block to obtain a frequency domain watermarked data block, and the frequency domain watermarked data block is converted into a time domain watermarked data block, and multiple time domain watermarked data blocks are spliced ​​into time domain watermarked data.

6. The method according to any one of claims 1 to 5, further comprising: Get sample data and watermark sample information; The input sample data or time domain data is converted into frequency domain sample data through one or more sequentially connected watermark embedding modules in the stego-coding model, and the watermark sample information is embedded into the frequency domain sample data based on the global correlation relationship and / or local correlation relationship between the watermark sample information and the frequency domain sample data to obtain the frequency domain watermark sample, and the frequency domain watermark sample is converted into the time domain watermark sample; By means of the data adjustment module in the stego-coding model, the watermarked samples in the time domain finally outputted by one or more sequentially connected watermark embedding modules are adjusted to obtain the target watermarked samples corresponding to the sample data; Performing interference processing on the target watermarked sample by using an interference model to obtain an interfered target watermarked sample; Through the stego-decoding model, watermark decoding processing is performed based on the input interference target watermark sample and the target watermark sample to obtain target watermark sample information; Based on the watermark sample information, the target watermark sample information, the sample data and the time domain watermarked samples finally output by the one or more sequentially connected watermark embedding modules, the steganalysis encoding model, the interference model and the steganalysis decoding model are jointly trained through a preset loss function to obtain a trained steganalysis encoding model.

7. According to the method of claim 6, the loss function includes a first sub-loss function and a second sub-loss function, the first sub-loss function is a loss function between the watermark sample information and the target watermark sample information, the second sub-loss function is a loss function between the sample data and the watermarked samples in the time domain finally output by the one or more sequentially connected watermark embedding modules, the first sub-loss function includes a mean square error loss function, and the second sub-loss function includes one or more of a mean square error loss function, a mean absolute error loss function and a perceptual similarity loss function.

8. According to the method of claim 1, the target data includes one or more of image data, text data, audio data and video data, and the target watermark information includes one or more of image data, character string, audio data and video data.

9. An information steganography device, comprising: A data acquisition module, which acquires the target data to be steganographic and the target watermark information; A watermark embedding module, which converts the input target data or time-domain data into frequency-domain target data through one or more sequentially connected watermark embedding modules in a pre-trained stego-coding model, embeds the target watermark information into the frequency-domain target data based on a global correlation relationship and / or a local correlation relationship between the target watermark information and the frequency-domain target data to obtain frequency-domain watermarked data, and converts the frequency-domain watermarked data into time-domain watermarked data; The adjustment module adjusts the time-domain watermarked data finally output by one or more sequentially connected watermark embedding modules through the data adjustment module in the stego-coding model to obtain the target watermarked data corresponding to the target data.

10. An information steganography device, comprising: processor; as well as a memory arranged to store computer executable instructions which, when executed, cause the processor to: Obtaining target data and target watermark information to be steganographic; The input target data or time-domain data is converted into frequency-domain target data through one or more sequentially connected watermark embedding modules in a pre-trained stego-coding model, and the target watermark information is embedded into the frequency-domain target data based on the global correlation relationship and / or local correlation relationship between the target watermark information and the frequency-domain target data to obtain frequency-domain watermarked data, and the frequency-domain watermarked data is converted into time-domain watermarked data; The data adjustment module in the stego-coding model adjusts the time-domain watermarked data finally output by one or more sequentially connected watermark embedding modules to obtain target watermarked data corresponding to the target data.

Images