Steganography information generation method and device, steganography information extraction method and device, equipment and storage medium

Abstract

The invention discloses steganography information generation and extraction methods and devices, equipment and a storage medium, and belongs to the technical field of information hiding. The steganography information generation method is applied to a sending end, and comprises the following steps: acquiring a secret information bit stream; encrypting the secret information bit stream through a predetermined encryption algorithm by using a secret key shared by two communication parties to obtain an encrypted bit stream; based on the encrypted bit stream, driving a pre-trained sequence generation model to generate a human body posture key point sequence; drawing the generated human body posture key point sequence into a key point connected graph; inputting the key point connected graph and any figure image into a pre-trained posture guide figure image generation model to generate a figure image with a target posture; and uploading the figure image with the target posture through a public channel to transmit secret information. According to the method disclosed by the invention, the security, secrecy, flexibility and high efficiency of the steganography technology are greatly improved.

Description

Technical Field

[0001] The present invention relates to the technical field of information hiding, and more specifically, to a method and device for generating and extracting stego information, equipment and storage medium. Background Art

[0002] With the digitization of information transmission and the widespread use of network media, the public has put forward higher requirements for information security and privacy protection. Traditional basic cryptographic encryption and decryption methods can no longer meet most information security needs. Therefore, steganography came into being. Steganography is different from cryptography. It embeds secret information directly into carrier data that is not abnormal in perception and sends it to the other party, so it is not easy to arouse the suspicion of attackers, and to a certain extent, it guarantees the security of communication.

[0003] Traditional information hiding methods mainly focus on embedded steganography, in which images are the most commonly used carriers. Image steganography methods modify the spatial and frequency domains of images, or send images containing secret information based on a pre-designed mapping relationship library. However, as long as the carrier is modified, detectable traces of modification will be left, so that the steganalysis algorithm may successfully detect the existence of hidden behavior. In order to fundamentally resist steganalysis, researchers in the field of information hiding have proposed the concept of Coverless Information Hiding (CIH). Compared with traditional methods, the CIH method does not need to modify the carrier, but directly generates or retrieves the stego carrier based on the secret information.

[0004] At present, image carrier-free information hiding algorithms are mainly divided into two categories: mapping carrier-free information hiding and generative carrier-free information hiding. Mapping carrier-free information hiding algorithm refers to the process of establishing a mapping relationship between an image and secret information based on the color, pixel and other features of the image itself, and then sending the image containing the secret information to achieve information hiding. Although the mapping image selection method can directly use natural images to transmit secret information, fundamentally resisting the detection of various steganographic algorithms, its steganographic capacity is very limited, and it is necessary to establish a large-scale natural image database in advance to retrieve images that can express secret information. Generative carrier-free information hiding algorithms achieve information hiding in the process of carrier image generation. In recent years, the rapid development of deep learning and neural network technology has created conditions for the advancement of generative steganography. The emergence of models such as generative adversarial networks (GANs) has made the generation and modification of images more efficient and realistic, providing new possibilities for information embedding.

[0005] Among the existing image generation steganography methods, the steganography technology for character image generation is highly practical, because in daily life, many people often use character images as the content of social communication. Therefore, this type of method will hardly arouse the suspicion of attackers.

[0006] In the prior art, some steganographic methods for character image generation are as follows:

[0007] A steganographic method based on generative images generated by anime characters. First, it converts secret information into a set of attribute labels of anime characters, such as hairstyle, hair color, eye color, etc. Then, with the attribute label set as a constraint, a GANs-based anime character generation network is used to generate an image set that meets the constraint. Then, the quality of the generated images is evaluated, and anime characters with better quality are selected as steganographic images. Finally, the receiver extracts the attribute labels representing the secret information according to the corresponding method to obtain the secret information. Each cryptic image generated by this method can hide about 14 bits of secret information, and the steganographic capacity is low; the image itself has a low resolution, is relatively blurry to the human eye, and is easily seen as AI-generated. If it is frequently transmitted on a public channel, it is easy to arouse the suspicion of attackers; the cryptic images generated by this method have low robustness, and after some attacks such as noise, filtering, and image compression, the accuracy of secret information extraction will be greatly reduced.

[0008] Another generative steganographic method based on the StarGAN model to generate real face images first uses a specific face dataset with style labels to train the StarGAN model to achieve the effect of completing multi-style image generation. During communication, the sender encodes the secret information in binary and uses it as the target label for style conversion. Then, the target style label is input into the model to generate an image of the corresponding style and send it to the receiver. After receiving the information, the receiver uses the discriminator network to obtain the style label code, de-maps and decodes it to obtain the corresponding secret information. The hidden capacity of the secret image generated by this method is subject to the style label of the training dataset. That is, when the style label contains 40 different categories, the theoretical steganographic capacity of the secret image of this method is 40 bits. However, in practical applications, it is also necessary to consider whether there will be semantic contradictions between the labels, such as the conflict between the two labels "female" and "beard". This method also needs to consider that too many style labels will lead to a significant reduction in the secret information extraction rate. Therefore, its actual hidden capacity will be reduced; secondly, its secret image itself has a lower resolution.

[0009] Another scheme combines the mapping and generative carrier-free information hiding of images, and proposes a steganographic method based on natural image selection and StarGAN image generation network. This method divides the secret information into two parts, and then retrieves a suitable image from the database to represent the first part of the secret information during the image selection process; for the second part, an image mapping is established between it and the attribute features, and then the attribute features are modified according to the mapping relationship and the selected image, and the steganographic face image is generated using StarGAN. This method makes up for the disadvantage that the steganographic capacity of the above-mentioned generative steganographic method based on the StarGAN model to generate real face images is much smaller than the number of labels in the training data set in actual applications, and selects labels without semantic contradictions from the same training data set as training labels. The actual steganographic capacity of each steganographic image exceeds 33 bits. However, its resolution is still low and it is relatively blurry when observed by the human eye. Summary of the invention

[0010] In order to overcome the technical problems existing in the above-mentioned prior art, the present invention provides a method and device for generating and extracting stego information, a device and a storage medium.

[0011] In order to solve the above technical problems, the technical solution of the present invention is as follows:

[0012] A method for generating stego information, applied to a sending end, the method comprising:

[0013] Obtain secret information bit stream;

[0014] Using a key shared by both communicating parties, encrypting the secret information bit stream by a predetermined encryption algorithm to obtain an encrypted bit stream;

[0015] Based on the encrypted bit stream, driving a pre-trained sequence generation model to generate a human body posture key point sequence;

[0016] Drawing the generated human body posture key point sequence into a key point connected graph;

[0017] Inputting the key point connectivity graph and any character image into a pre-trained posture-guided character image generation model to generate a character image with a target posture;

[0018] The person image with the target posture is uploaded through a public channel to transmit secret information.

[0019] Further, the step of encrypting the secret information bit stream includes:

[0020] Obtain the secret information bit stream M, obtain the key Key held by both parties in communication, and generate the encrypted bit stream M′ by calculating M′=M⊕Key;

[0021] The training method of the sequence generation model includes:

[0022] The MLE strategy is used to train the LSTM network, and the trained sequence generation model is obtained by optimizing the model parameters;

[0023] In the LSTM network, the output of each time step t is associated with the information of all previous time steps; at time step t, the simplified output formula of the LSTM network is expressed as:

[0024] h t =LSTM(x t ,h t-1 θ LSTM )

[0025] Output = softmax(h t ×W+b o )

[0026] In the above formula, h t is the hidden state at time step t; x t is the input at time step t; h t-1 is the hidden state at time step t-1; θ LSTM represents the weight parameter of the LSTM network; W represents the weight matrix of the output layer; b o Represents the bias in the output layer; Output represents the output result of the LSTM network at time step t, and the Softmax function is a function that converts the input vector into a probability distribution;

[0027] The MLE strategy optimizes the model parameters by defining a loss function, which is expressed as:

[0028]

[0029] In the above formula, I is the length of the sequence; s i is the observed value at time step i.

[0030] Further, based on the encrypted bit stream, a pre-trained sequence generation model is driven to generate a sequence of key points of human posture including candidate set generation and secret drive coordinate selection;

[0031] The candidate set generation includes:

[0032] The first i-1 elements in the key point coordinate sequence S are input into the sequence generation model in sequence. At time step i, the output of the sequence generation model is s i The conditional probability distribution of i |s1,s2,…,s i-1 );

[0033] Will s i The probabilities of all legal coordinates are arranged in descending order, and those with a probability lower than the preset threshold τ are screened out. i The remaining coordinates constitute the preliminary candidate set;

[0034] To be able to work with binary encoding Bin i ={b1,b2,…,b K}Build a mapping relationship and select the top K=2 with the highest probability Li coordinates as the final candidate set Set i ={e1,e2,…,e K}; where L i Represents the constructed candidate set Set i The final embeddable hidden capacity is calculated by the following formula:

[0035] L i =log2(size(Set i ))

[0036] In the above formula, size(Set i ) represents the candidate set Set i The number of coordinates in the middle, log2 is the logarithmic function with base 2;

[0037] Wherein, the secret drive coordinate selection includes:

[0038] In generating the candidate set Set i Afterwards, based on the mapping relationship, a segment with a length of L is extracted from the encrypted bit stream. i If the length of the encrypted bit stream is greater than 0 and less than L i Then fill the header with zeros, select the coordinates as the element si through the mapping relationship and add it to the key point coordinate sequence S;

[0039] The steps of candidate set generation and secret-driven coordinate selection are repeated until the human pose keypoint sequence is completely constructed.

[0040] A stego information generating device, the stego information generating device comprising:

[0041] A secret information acquisition module, wherein the secret information acquisition module is used to acquire a secret information bit stream;

[0042] An encryption module, wherein the encryption module uses a key shared by both communicating parties to encrypt the secret information bit stream through a predetermined encryption algorithm to obtain an encrypted bit stream;

[0043] A sequence generation module, which drives a pre-trained sequence generation model to generate a human posture key point sequence based on the encrypted bit stream;

[0044] A key point connected graph drawing module, wherein the key point connected graph drawing module draws the generated human posture key point sequence into a key point connected graph;

[0045] A posture-guided image generation module, wherein the posture-guided image generation module inputs the key point connectivity graph and any character image into a pre-trained posture-guided character image generation model to generate a character image with a target posture;

[0046] An image transmission module uploads the image of the person with the target posture through a public channel to transmit secret information.

[0047] A method for extracting steganalytic information, applied to a receiving end, the method comprising:

[0048] Acquire a person image with a target pose from a public channel;

[0049] Extracting a human body posture key point sequence from the human image with the target posture using a posture estimation model;

[0050] Recovering an encrypted bit stream from the extracted human posture key point sequence using a sequence generation model;

[0051] The original secret information bit stream is restored based on the restored encrypted bit stream and the key shared by the communicating parties.

[0052] Furthermore, YOLOv8n-Pose is adopted as the pose estimation model to realize key point extraction.

[0053] A stego information extraction device, the stego information extraction device comprising:

[0054] An image acquisition module, the image acquisition module is used to acquire a person image with a target posture from a public channel;

[0055] A posture estimation module, wherein the posture estimation module uses a posture estimation model to extract a human posture key point sequence from the human image having the target posture;

[0056] A bitstream recovery module, wherein the bitstream recovery module uses a sequence generation model to recover an encrypted bitstream from the extracted human posture key point sequence;

[0057] A data decryption module is used to restore the original secret information bit stream based on the restored encrypted bit stream and a key shared by both communicating parties.

[0058] A device for generating stego information comprises: a processor, a memory and a program stored in the memory and executable on the processor, wherein the program implements the aforementioned method for generating stego information when executed by the processor.

[0059] A device for extracting stego information comprises: a processor, a memory and a program stored in the memory and executable on the processor, wherein the program implements the aforementioned method for extracting stego information when executed by the processor.

[0060] A readable storage medium stores a program, and when the program is executed by a processor, the program implements the aforementioned stego information generation method or the aforementioned stego information extraction method.

[0061] Compared with the prior art, the technical solution of the present invention has the following beneficial effects:

[0062] The present invention provides a method and device for generating and extracting steganographic information, a device and a storage medium. By guiding a character image generation model through posture, the generated secret image conforms to the posture of a real human body and is no different from an ordinary image, so it is not easy to attract the attention of attackers, which greatly improves the security, confidentiality and efficiency of the steganographic technology. BRIEF DESCRIPTION OF THE DRAWINGS

[0063] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0064] Figure 1 is a flow chart of a method for generating stego information provided in one embodiment of the present application;

[0065] Figure 2 It is a schematic diagram of the sequence generation process in the stego information generation method of the present application;

[0066] Figure 3 It is a schematic diagram of the DPTN-TA model structure in the steganalysis information generation method of the present application;

[0067] Figure 4 is a schematic diagram of a stego information generating device provided in one embodiment of the present application;

[0068] Figure 5 is a flow chart of a method for extracting steganalytic information provided in one embodiment of the present application;

[0069] Figure 6It is a network structure diagram of YOLOv8n-Pose in the steganalysis information extraction method of this application;

[0070] Figure 7 is a schematic diagram of a stego information extraction device provided in one embodiment of the present application;

[0071] Figure 8 It is a general flow chart of the stego information generation method and the stego information extraction method of the present application;

[0072] Description of the markings in the figure:

[0073] 1. Secret information acquisition module; 2. Encryption module; 3. Sequence generation module; 4. Key point connectivity graph drawing module; 5. Posture guidance image generation module; 6. Image transmission module;

[0074] 7. Image acquisition module; 8. Posture estimation module; 9. Bit stream recovery module; 10. Data decryption module. DETAILED DESCRIPTION

[0075] Exemplary embodiments of the present invention will be described in more detail below with reference to the accompanying drawings. Although exemplary embodiments of the present invention are shown in the drawings, it should be understood that the present invention can be implemented in various forms and should not be limited to the embodiments set forth herein.

[0076] Embodiment 1:

[0077] like Figure 1 As shown, the present invention provides a technical solution:

[0078] A method for generating stego information, applied to a sending end, the method comprising:

[0079] Obtain secret information bit stream;

[0080] Using a key shared by both communicating parties, encrypting the secret information bit stream by a predetermined encryption algorithm to obtain an encrypted bit stream;

[0081] Based on the encrypted bit stream, driving a pre-trained sequence generation model to generate a human body posture key point sequence;

[0082] Drawing the generated human body posture key point sequence into a key point connected graph;

[0083] Inputting the key point connectivity graph and any character image into a pre-trained posture-guided character image generation model to generate a character image with a target posture;

[0084] The person image with the target posture is uploaded through a public channel to transmit secret information.

[0085] This embodiment encrypts the secret information to ensure that the secret information is not easily stolen by a third party even if it is transmitted on a public channel; a human posture key point sequence is used to generate a posture that conforms to the real human body, which is no different from an ordinary image and is not likely to attract the attention of an attacker; the posture guides the character image generation model, and the generated image has a high sense of reality, and even experienced observers find it difficult to detect abnormalities; there is no need to modify the original image, which fundamentally avoids the detection of steganalysis technology; by guiding image generation, the final output character image not only conforms to the target posture, but also has no significant visual difference from the natural image; this method improves the quality and realism of the stego image while increasing the stego capacity.

[0086] Embodiment 2:

[0087] Based on Example 1, the process of generating a key point sequence is as follows:

[0088] In order to generate a sequence that is consistent with the statistical characteristics of the real human body posture key point coordinate data, this paper uses the maximum likelihood estimation (MLE) strategy to train the long short-term memory network (LSTM). In this model, the LSTM network aims to capture the dependencies between time steps in the sequence data and generate new data sequences by learning the statistical characteristics of these data.

[0089] LSTM is a special recurrent neural network (RNN) that solves the gradient vanishing and gradient exploding problems encountered by traditional RNN when processing long sequence data by introducing a gating mechanism. This network structure is suitable for modeling sequence data because it can capture long-term dependencies in time series. In the LSTM network, the output of each time step t is associated with the information of all previous time steps. At time step t, the simplified output formula of the model can be expressed as:

[0090] h t =LSTM(x t ,h t-1 θ LSTM )

[0091] Output = softmax(h t ×W+b o )

[0092] Among them, h t is the hidden state at time step t; x t is the input at time step t; θ LSTMrepresents the weight parameter of the LSTM network; W represents the weight matrix of the output layer; b o represents the bias in the output layer.

[0093] In the field of human posture estimation, a given set of human key point coordinates can be abstracted as a set of discrete points on the image plane, which depict the motion contour of the human body. Specifically, for a full-body image with a width of w and a height of h, its posture key points can be represented as a sequence of n two-dimensional coordinates S = {s i |i=1,2,…,n}. Among them, p i represents the i-th key point in the sequence, whose coordinates (x i ,y i ) is constrained to range from (0,0) to (w-1,h-1).

[0094] The first i-1 elements in the key point coordinate sequence S are input into the model in sequence. At time step i, the output of the model is s i The conditional probability distribution of i |s1,s2,…,s i-1 ), where s i The range of values ​​is any coordinate position in the two-dimensional plane. In order to train LSTM to generate sequences consistent with the distribution of real human posture key point data, this application adopts the MLE strategy. The MLE strategy is implemented by defining a loss function that measures the probability of the model predicting the current observation value given the previous observation value. The loss function can be expressed as:

[0095]

[0096] Where I is the length of the sequence; s i is the observed value at time step i. By minimizing the loss function, we are actually looking for model parameters that can maximize the probability of the observed data sequence, so as to obtain a sequence generation model that is consistent with the distribution of real human key point data.

[0097] After the sequence generation model is trained, the secret information is used to drive the model to generate a sequence of key points of human posture.

[0098] The sequence generation process includes the following steps:

[0099] 1. Generate encrypted bit stream;

[0100] Obtain the secret information bit stream M; obtain the key Key held by both parties in communication; and obtain the encrypted bit stream M'={m1,m2,…,m n};

[0101] 2. Generate candidate sets;

[0102] This application embeds secret information in the selection of coordinates by constructing a candidate set of coordinates for the i-th element; the first i-1 elements in the key point coordinate sequence S are input into the model in sequence. At time step i, the output of the model is s i The conditional probability distribution of i |s1,s2,…,s i-1 ); Then, s i The probabilities of all legal coordinates are arranged in descending order, and the coordinates with probabilities lower than the preset threshold τ are screened out. i The remaining coordinates initially constitute the candidate set Set i ; Then, to be able to use the binary encoding Bin i ={b1,b2,…,b K}Build a mapping relationship and select the top K=2 with the highest probability Li coordinates as the final candidate set Set i ={e1,e2,…,e K}. Note that L i Represents the constructed candidate set Set i The final embedded hidden capacity can be calculated by the following formula:

[0103] L i =log2(size(Set i ))

[0104] In the above formula, size(Set i ) represents the candidate set Set i The number of coordinates in the function, log2 is the logarithmic function with base 2.

[0105] To illustrate this step, assume that the number of coordinates in the preliminary candidate set is 10. According to the above formula, L is 3, and only the 8 coordinates with the highest probability are retained in the final candidate set. A binary code is assigned to each coordinate in sequence, so the binary code corresponding to Set = {e1, e2, ..., e8} is Bin = {000, 001, ..., 111}.

[0106] 3. Secret drive coordinate selection;

[0107] Generate candidate set Set i , and after building the mapping relationship; then, cut off the length L from M' i Binary string (if the length of M' is greater than 0 and less than L i Then fill the head with zeros), select the coordinates as element s according to the mapping relationship i , and s i Add sequence S.

[0108] Repeat steps 2 and 3 until the key point sequence S is completely constructed; the specific steps of the sequence generation process can be found in Figure 2 .

[0109] Furthermore, the posture-guided character image generation model is specifically:

[0110] A sequence of key points of human posture is generated driven by secret information, and the sequence is drawn as a human key point connectivity graph. Then, the application uses the trained DPTN-TA model to synthesize the obtained human key point connectivity graph with any human image to obtain a human image with a new posture, which is transmitted as a secret image on a public channel.

[0111] The core of the DPTN-TA model is to introduce an auxiliary task, namely the source-to-source task, to assist the learning of the source-to-target task through the twin network structure. The two branch tasks share some weights and communicate through the Pose Transformer Module (PTM). The model structure is as follows Figure 3 shown.

[0112] Embodiment 3:

[0113] like Figure 4 As shown, the present invention provides a technical solution:

[0114] A stego information generating device, the stego information generating device comprising:

[0115] A secret information acquisition module 1, wherein the secret information acquisition module 1 is used to acquire a secret information bit stream;

[0116] Encryption module 2, wherein the encryption module 2 uses a key shared by both communicating parties to encrypt the secret information bit stream through a predetermined encryption algorithm to obtain an encrypted bit stream;

[0117] A sequence generation module 3, wherein the sequence generation module 3 drives a pre-trained sequence generation model to generate a human posture key point sequence based on the encrypted bit stream;

[0118] A key point connected graph drawing module 4, wherein the key point connected graph drawing module 4 draws the generated human posture key point sequence into a key point connected graph;

[0119] A posture-guided image generation module 5, wherein the posture-guided image generation module 5 inputs the key point connectivity graph and any character image into a pre-trained posture-guided character image generation model to generate a character image with a target posture;

[0120] The image transmission module 6 uploads the image of the person with the target posture through a public channel to transmit secret information.

[0121] Embodiment 4:

[0122] like Figure 5 As shown, the present invention provides a technical solution:

[0123] A method for extracting steganalytic information, applied to a receiving end, the method comprising:

[0124] Acquire a person image with a target pose from a public channel;

[0125] Extracting a human body posture key point sequence from the human image with the target posture using a posture estimation model;

[0126] Recovering an encrypted bit stream from the extracted human posture key point sequence using a sequence generation model;

[0127] The original secret information bit stream is restored based on the restored encrypted bit stream and the key shared by the communicating parties.

[0128] Embodiment 5:

[0129] On the basis of Example 4, Figure 6 As shown, in order to realize the information extraction process in the proposed steganography method, this application adopts YOLOv8n-Pose as the basic framework of the detection algorithm; that is, YOLOv8n-Pose is used as a posture estimation model to realize key point extraction;

[0130] Using YOLOv8n-Pose as the pose estimation model, the accuracy of extracting the key point sequence of human pose is high, reducing information extraction errors.

[0131] Furthermore, the proposed pose key point sequence generation model and the pose key point extraction network YOLOv8n-Pose are stored on the local device of the receiving end. When the receiving end obtains a set of secret images with multiple poses in the public channel, these two networks are used to recover the secret information. The specific steps are as follows:

[0132] 1. Extract key points of human body posture;

[0133] Input a encrypted image into the human body posture key point extraction network to obtain the key point coordinate sequence S;

[0134] 2. Recover the encrypted bitstream;

[0135] Get the initial element s1 from the sequence S, get the corresponding binary code according to the mapping relationship, and insert the code into the encrypted bit stream M'; Assume that the sequence S has n elements, and start from the second element and execute for each subsequent element: Input the first i-1 elements in S into the sequence generation model in turn, and get s i The probability distribution Pr(si |s1,s2,…,s i-1 ); Sort these coordinates in descending order of their probability, and filter out those coordinates below the threshold τ to initially form a candidate set Set i ; Then, select the top K=2 with the highest probability Li coordinates as the final candidate set Set i , and construct its mapping relationship with the binary code; get element s from sequence S i , obtain the corresponding binary code according to the mapping relationship, and insert the code into M';

[0136] Repeat steps 1 and 2 to extract the bit streams embedded in multiple encrypted images to obtain a complete encrypted bit stream M';

[0137] 3. Decode to get secret information;

[0138] According to the encrypted bit stream M' and the key Key held by both communicating parties, the secret information bit stream M is obtained by the operation M'⊕Key.

[0139] Comprehensive Examples 1-2 and Examples 4-5, such as Figure 8 The overall framework shown in this application proposes a generative steganography method based on posture key point encoding:

[0140] First, the sender generates a sequence of posture key points through a sequence generation network driven by secret information and draws a key point connection graph. Then, this graph and any character image are input into the posture-guided character image generation model to obtain a character image with the target posture. The sender uploads the image through a public channel to transmit secret information.

[0141] Finally, the receiver obtains the secret image from the public channel, uses the posture estimation model to extract the posture key points, and then uses the sequence generation model to extract the secret information.

[0142] Embodiment 6:

[0143] like Figure 7 As shown, the present invention provides a technical solution:

[0144] A stego information extraction device, the stego information extraction device comprising:

[0145] An image acquisition module 7, wherein the image acquisition module 7 is used to acquire a person image with a target posture from a public channel;

[0146] A posture estimation module 8, wherein the posture estimation module 8 uses a posture estimation model to extract a human posture key point sequence from the human image having the target posture;

[0147] A bitstream recovery module 9, wherein the bitstream recovery module 9 recovers the encrypted bitstream from the extracted human posture key point sequence using a sequence generation model;

[0148] The data decryption module 10 restores the original secret information bit stream according to the restored encrypted bit stream and the key shared by the communicating parties.

[0149] Embodiment 7:

[0150] The present invention provides a technical solution:

[0151] A device for generating stego information comprises: a processor, a memory and a program stored in the memory and executable on the processor, wherein the program implements the aforementioned method for generating stego information when executed by the processor.

[0152] Embodiment 8:

[0153] The present invention provides a technical solution:

[0154] A device for extracting stego information comprises: a processor, a memory and a program stored in the memory and executable on the processor, wherein the program implements the aforementioned method for extracting stego information when executed by the processor.

[0155] Embodiment 9:

[0156] The present invention provides a technical solution:

[0157] A readable storage medium stores a program, and when the program is executed by a processor, the program implements the aforementioned stego information generation method or the aforementioned stego information extraction method.

[0158] In the several embodiments provided in the present application, it should be understood that the disclosed systems, devices and methods can be implemented in other ways. The device embodiments described above are merely schematic. For example, the division of the units is only a logical function division. There may be other division methods in actual implementation. For example, multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be through some communication interfaces, and the indirect coupling or communication connection of devices or units can be electrical, mechanical or other forms.

[0159] The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed on multiple network units. Some or all of the units may be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0160] In addition, each functional unit in each embodiment of the present invention may be integrated into one processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit.

[0161] Although example embodiments have been described herein with reference to the accompanying drawings, it should be understood that the above example embodiments are merely exemplary and are not intended to limit the scope of the present application to this. Those of ordinary skill in the art may make various changes and modifications therein without departing from the scope and spirit of the present application. All these changes and modifications are intended to be included within the scope of the present application as required by the appended claims.

[0162] Similarly, it should be understood that in order to streamline the present application and help understand one or more of the various application aspects, in the description of the exemplary embodiments of the present application, the various features of the present application are sometimes grouped together into a single embodiment, figure, or description thereof. However, the method of the present application should not be interpreted as reflecting the following intention: the claimed application requires more features than the features clearly stated in each claim. More specifically, as reflected in the corresponding claims, the application point is that the corresponding technical problem can be solved with less than all the features of a single disclosed embodiment. Therefore, the claims following the specific embodiment are hereby explicitly incorporated into the specific embodiment, wherein each claim itself serves as a separate embodiment of the present application.

[0163] In addition, those skilled in the art will appreciate that, although some embodiments described herein include certain features included in other embodiments but not other features, the combination of features of different embodiments is meant to be within the scope of the present application and form different embodiments. For example, in the claims, any one of the claimed embodiments can be used in any combination.

[0164] It should be noted that the above embodiments illustrate rather than limit the present application, and that those skilled in the art may devise alternative embodiments without departing from the scope of the appended claims. In the claims, any reference symbol between brackets should not be construed as a limitation to the claims. The use of the words first, second, and third, etc. does not indicate any order. These words may be interpreted as names.

Claims

1. A method for generating stego information, characterized in that: Applied to the sending end, the generation method includes: Obtain secret information bit stream; Using a key shared by both communicating parties, encrypting the secret information bit stream by a predetermined encryption algorithm to obtain an encrypted bit stream; Based on the encrypted bit stream, driving a pre-trained sequence generation model to generate a human body posture key point sequence; Drawing the generated human body posture key point sequence into a key point connected graph; Inputting the key point connectivity graph and any character image into a pre-trained posture-guided character image generation model to generate a character image with a target posture; The person image with the target posture is uploaded through a public channel to transmit secret information.

2. The method for generating stego information according to claim 1, characterized in that: The step of encrypting the secret information bit stream comprises: Obtain the secret information bit stream M, obtain the key Key held by both parties in communication, and generate the encrypted bit stream M′ by calculating M′=M⊕Key; The training method of the sequence generation model includes: The MLE strategy is used to train the LSTM network, and the trained sequence generation model is obtained by optimizing the model parameters; In the LSTM network, the output of each time step t is associated with the information of all previous time steps; at time step t, the simplified output formula of the LSTM network is expressed as: h t =LSTM(x t ,h t-1 ;θ LSTM ) Output=softmax(h t ×W+b o ) In the above formula, h t is the hidden state at time step t; x t is the input at time step t; h t-1 is the hidden state at time step t-1; θ LSTM represents the weight parameter of the LSTM network; W represents the weight matrix of the output layer; b o Represents the bias in the output layer; Output represents the output result of the LSTM network at time step t, and the Softmax function is a function that converts the input vector into a probability distribution; The MLE strategy optimizes the model parameters by defining a loss function, which is expressed as: In the above formula, I is the length of the sequence; s i is the observed value at time step i.

3. The method for generating stego information according to claim 1, characterized in that: Based on the encrypted bit stream, driving a pre-trained sequence generation model to generate a human posture key point sequence including candidate set generation and secret drive coordinate selection; The candidate set generation includes: The first i-1 elements in the key point coordinate sequence S are input into the sequence generation model in sequence. At time step i, the output of the sequence generation model is s i The conditional probability distribution of i |s1,s2,…,s i-1 ); Will s i The probabilities of all legal coordinates are arranged in descending order, and the coordinates with probabilities lower than the preset threshold τi are screened out, and the remaining coordinates constitute the preliminary candidate set; To be able to work with binary encoding Bin i ={b1,b2,…,b K }Build a mapping relationship and select the top K=2 with the highest probability Li coordinates as the final candidate set Set i ={e1,e2,…,e K }; where L i Represents the constructed candidate set Set i The final embeddable hidden capacity is calculated by the following formula: L i =log2(size(Set i )) In the above formula, size(Set i ) represents the candidate set Set i The number of coordinates in the middle, log2 is the logarithmic function with base 2; Wherein, the secret drive coordinate selection includes: In generating the candidate set Set i Afterwards, based on the mapping relationship, a segment with a length of L is extracted from the encrypted bit stream. i If the length of the encrypted bit stream is greater than 0 and less than L i Then fill the header with zeros, select the coordinates as the element si through the mapping relationship and add it to the key point coordinate sequence S; The steps of candidate set generation and secret-driven coordinate selection are repeated until the human pose keypoint sequence is completely constructed.

4. A stego information generating device, characterized in that: The stego information generating device comprises: A secret information acquisition module, wherein the secret information acquisition module is used to acquire a secret information bit stream; An encryption module, wherein the encryption module uses a key shared by both communicating parties to encrypt the secret information bit stream through a predetermined encryption algorithm to obtain an encrypted bit stream; A sequence generation module, which drives a pre-trained sequence generation model to generate a human posture key point sequence based on the encrypted bit stream; A key point connected graph drawing module, wherein the key point connected graph drawing module draws the generated human posture key point sequence into a key point connected graph; A posture-guided image generation module, wherein the posture-guided image generation module inputs the key point connectivity graph and any character image into a pre-trained posture-guided character image generation model to generate a character image with a target posture; An image transmission module uploads the image of the person with the target posture through a public channel to transmit secret information.

5. A method for extracting steganalytic information, characterized in that: Applied to the receiving end, the extraction method includes: Acquire a person image with a target pose from a public channel; Extracting a human body posture key point sequence from the human image with the target posture using a posture estimation model; Recovering an encrypted bit stream from the extracted human posture key point sequence using a sequence generation model; The original secret information bit stream is restored based on the restored encrypted bit stream and the key shared by the communicating parties.

6. The method for extracting stegographic information according to claim 5, characterized in that: YOLOv8n-Pose is used as the pose estimation model to realize key point extraction.

7. A stego information extraction device, characterized in that: The steganalysis information extraction device comprises: An image acquisition module, the image acquisition module is used to acquire a person image with a target posture from a public channel; A posture estimation module, wherein the posture estimation module uses a posture estimation model to extract a human posture key point sequence from the human image having the target posture; A bitstream recovery module, wherein the bitstream recovery module uses a sequence generation model to recover an encrypted bitstream from the extracted human posture key point sequence; A data decryption module is used to restore the original secret information bit stream based on the restored encrypted bit stream and a key shared by both communicating parties.

8. A device for generating stego information, characterized in that: include: A processor, a memory, and a program stored in the memory and executable on the processor, wherein when the program is executed by the processor, the method for generating stego-information according to any one of claims 1 to 3 is implemented.

9. A device for extracting stego information, characterized in that: include: A processor, a memory, and a program stored in the memory and executable on the processor, wherein when the program is executed by the processor, the method for extracting stegographic information as claimed in any one of claims 5 to 6 is implemented.

10. A readable storage medium, characterized in that: The readable storage medium stores a program, and when the program is executed by a processor, the method for generating stego information as claimed in any one of claims 1 to 3 is implemented, or the method for extracting stego information as claimed in any one of claims 5 to 6 is implemented.

Images