Secret graph open transmission method and device based on image steganography and storage medium

Abstract

The invention discloses a secret image open transmission method and device based on image steganography and a storage medium, and belongs to the technical field of information security, and the method comprises the steps that a sender obtains a secret image to be transmitted, inputs the secret image, a carrier image and a random noise tensor into a reversible steganography network model, obtains a secret carrying image and forward output loss information, and transmits the secret carrying image to a storage medium; sending the secret-carrying graph to a receiver; the receiver receives the secret carrying graph and inputs the secret carrying graph and backward input loss information into the reversible steganography network model to obtain a recovered secret graph; wherein the reversible steganography network model comprises the following steps: generating a noise fuzzy mechanism or removing adaptive noise to guide fusion or coupling of a secret graph and a carrier graph; the secret key steganography mechanism is used for respectively generating a public key and a public key label according to corresponding noise of forward input and backward output; the method is also used for generating a private key label and a private key respectively according to corresponding lost information of forward output and backward input, so that the concealment and the recovery quality of image steganography are remarkably improved.

Description

Technical Field

[0001] The present invention belongs to the field of information security technology, and in particular relates to a secret image open transmission method, device and storage medium based on image steganography. Background Art

[0002] In recent years, with the rapid development of the internet and multimedia technologies, the field of information technology has been undergoing unprecedented rapid change. However, this has been accompanied by increasingly severe information security challenges. Various security issues are not only becoming more frequent but also increasingly complex, placing higher and more comprehensive demands on information security protection systems. Against this backdrop, image steganography, as a highly effective means of information security protection, enables secret communications through highly covert means. It has been widely used in key areas such as national defense, military, healthcare, and commercial finance, and occupies a pivotal position in both theoretical research and practical applications of information security.

[0003] To ensure the quality of recovered secret images, embedded-carrier secret image steganography methods often embed high-frequency information from the secret image into the carrier image to generate a payload image. This payload image is then transmitted to achieve covert communication of the secret information. However, when the payload image is transmitted in an open environment, third parties may also have access to it, placing higher demands on its security. Since steganalysis methods in passive attacks identify the payload image from natural images by comparing the carrier image with the payload image, the embedded secret information in the payload image poses a serious security risk to covert communication. Experimental analysis using the UDH method reveals that during secret recovery, the carrier image information interferes with the recovery of the secret. In other words, for embedded-carrier secret image steganography methods, secret recovery is highly dependent on the secret information in the payload image. Therefore, improving the stealth of steganography while ensuring recovery quality remains a challenge.

[0004] Although existing steganography methods can generate a carrier image that visually resembles the original carrier image, and the recovered secret image is relatively similar to the original secret image, when the carrier image and the carrier image are subtracted to obtain a carrier difference image, the carrier difference image produced by existing methods contains a large amount of secret image information. This means that the carrier image used for transmission contains a large amount of secret image information, which not only hinders the improvement of the carrier image quality but also poses a serious transmission security risk. In particular, when the carrier image is accessible to a third party, a passive attack can be used to obtain a large amount of secret image information from the intercepted carrier image and carrier image, thereby posing a risk of secret information leakage. Furthermore, a visual comparison of the secret error images generated by existing methods shows that the quality of these recovered secret images is clearly suboptimal and cannot meet the requirements for covert communication. Existing methods have never actively studied the protection of the embedded secret information, making it difficult to ensure the security of embedded steganography methods. Summary of the Invention

[0005] The purpose of the present invention is to provide a secret graph open transmission method, device and storage medium based on image steganography, which combines the noise obfuscation mechanism with the key steganography mechanism, guides the fusion or decoupling of secrets and carriers through noise, and establishes soft consistency between the sending and receiving ends through private keys and private key tags, thereby significantly improving the concealment and recovery quality of image steganography.

[0006] To achieve the above object, the present invention is implemented by adopting the following technical solutions:

[0007] In a first aspect, the present invention provides a secret image open transmission method based on image steganography, characterized by comprising:

[0008] The sender obtains the secret image to be transmitted, and inputs the secret image, the prepared carrier image, and the random noise tensor into a pre-trained reversible steganography network model to obtain the secret image and forward output loss information, and then sends the secret image to the receiver;

[0009] The receiver receives the steganographic image and inputs the steganographic image and the backward input loss information into the pre-trained reversible steganographic network model to obtain the recovered secret image.

[0010] Among them, the reversible steganography network model includes a noise obfuscation mechanism and a key steganography mechanism; the noise obfuscation mechanism is used to generate a fusion of an adaptive noise-guided secret graph and a carrier graph based on a random noise tensor, and to remove the decoupling of the adaptive noise-guided secret graph and the carrier graph; the key steganography mechanism is used to generate a public key and a public key label according to the corresponding noise of the forward input and the backward output, and reduce the steganography loss between the public key and the public key label; it is also used to generate a private key label and a private key according to the corresponding lost information of the forward output and the backward input, and reduce the steganography loss between the private key and the private key label.

[0011] Optionally, the noise blurring mechanism comprises a plurality of noise blurring blocks connected in series;

[0012] When the reversible steganalysis network model is used for forward hiding, the noise blurring block includes multiple weighted multiplication and addition operations;

[0013] When the reversible steganograph network model is used for backward recovery, the noise blurring block includes multiple weighted division and subtraction operations.

[0014] Optionally, the key steganography mechanism includes a public key generation module and a private key generation module that are independent of each other;

[0015] The public key generation module includes a public key generator and a public key label generator that are independent of each other, and the private key generation module includes a private key generator and a private key label generator that are independent of each other;

[0016] The public key generator, public key label generator, private key generator and private key label generator each include multiple convolution blocks.

[0017] Optionally, during the forward hiding process, the noise blurring mechanism performs the following operations:

[0018] Each noise fuzzy block guides the input secret image and carrier image to be fused using weighted multiplication and addition operations according to the input noise tensor, and iteratively updates the noise tensor during the fusion process of the secret image and carrier image. Among them, the input of the first noise fuzzy block is the secret image to be transmitted, the prepared carrier image and random Gaussian noise, and the input of other noise fuzzy blocks is the fused secret image and carrier image output by the previous noise fuzzy block and the iteratively updated adaptive noise. The output of the forward hidden of the noise blurred block is as follows:

[0019] ,

[0020] ,

[0021] ,

[0022] ,

[0023] Where, Indicates the The output of the noise-blurred block forward hidden, Indicates the The vector map of the forward hidden output of the noise blurred block, Indicates the The secret image of the forward hidden output of the noise-blurred block, Indicates the The adaptive noise of the output is forward hidden by a noise blurring block, Indicates the The vector map of the forward hidden output of the noise blurred block, Indicates the The secret image of the forward hidden output of the noise-blurred block, Indicates the The noise blurring blocks forward hide the adaptive noise of the output, represents the Hadamard product, and represent the exponential function and the Sigmoid function respectively, 、 、 、 、 、 、 and They represent the conversion functions, 、 、 、 For carrier secret fusion, 、 、 and Guided fusion for adaptive noise;

[0024] After each noise fuzzy block completes forward hiding, the secret image, carrier image, and pre-generated public key output by the last noise fuzzy block are concatenated in the channel dimension to output the concatenated carrier image and forward output loss information. The calculation formula is as follows:

[0025] ,

[0026] Where, Represents a connection operation on a channel, Indicates a secret map. Indicates that the forward output loses information, represents the inverse compression operation, represents the product of mappings, represents the index of the noise blurred block, represents the number of noise blurred blocks, Indicates compression operation, Represents the prepared vector map, represents the secret graph to be transmitted, Represents a public key.

[0027] Optionally, during the backward recovery process, the noise blurring mechanism performs the following operations:

[0028] Each noise fuzzy block uses weighted division and subtraction operations to decouple the input secret graph and carrier graph according to the input noise tensor, and iteratively recovers the noise tensor during the decoupling process of the secret graph and carrier graph. Among them, the input of the last noise fuzzy block is the received secret graph and the pre-generated private key spliced ​​in the channel dimension, and the input of other noise fuzzy blocks is the decoupled secret graph and carrier graph output by the next noise fuzzy block and the iteratively recovered adaptive noise. The output of the backward recovery of the noise blurred block is as follows:

[0029] ,

[0030] ,

[0031] ,

[0032] ,

[0033] Where, Indicates the The output of the noise-blurred block is backward restored, Indicates the The noise blurred block is used to recover the output carrier map. Indicates the The secret image of the noise blurred block is recovered backward, Indicates the The noise blurring block backward recovers the adaptive noise of the output, Indicates the The noise blurred block is used to recover the output carrier map. Indicates the The secret image of the noise blurred block is recovered backward, Indicates the The noise blurring block backward recovers the adaptive noise of the output, represents the Hadamard product, and represent the exponential function and the Sigmoid function respectively, 、 、 、 、 、 、 and They represent the conversion functions, 、 、 、 For carrier secret decoupling, 、 、 and Guided decoupling for adaptive noise;

[0034] After the backward recovery of each noise fuzzy block is completed, the secret image output by the first noise fuzzy block is obtained as the recovered secret image. The calculation formula is as follows:

[0035] ,

[0036] Where, Represents a connection operation on a channel, Represents the recovered vector graph, represents the recovered secret graph, represents the recovered noise, represents the inverse compression operation, represents the product of mappings, represents the index of the noise blurred block, represents the number of noise blurred blocks, Indicates the received encrypted image. Represents a private key.

[0037] Optionally, the key steganography mechanism performs the following operations:

[0038] In the forward hidden process, the forward output loses information Enter the private key label generator to generate a private key label , and the noise tensor output by the last noise blur block Enter the public key generator to generate a public key ;

[0039] During the backward recovery process, the untransmitted random tensors The lost information is input as the backward input to the private key generator to generate the private key , and the restored noise Enter the public key label generator to generate a public key label .

[0040] Optionally, the key steganography mechanism further performs the following operations:

[0041] In the Private Key tab and private key A private key steganographic loss is established between them to supervise the iterative learning of the private key. The calculation formula of the private key steganographic loss is as follows:

[0042] ,

[0043] Where, Indicates the loss of private key steganography, represents the index of a private key or private key label, represents the number of private keys or private key labels, represents the L2 norm, Indicates the A private key, Indicates the a private key tag;

[0044] In the public key and public key tags A public key steganography loss is established between them to supervise the iterative learning of the public key. The calculation formula of the public key steganography loss is as follows:

[0045] ,

[0046] Where, Indicates the public key steganographic loss, represents the index of a public key or public key label, represents the number of public keys or public key labels, Indicates the A public key, Indicates the A public key tag.

[0047] Optionally, the training method of the reversible steganographic network model includes:

[0048] Obtain historical secret map and carrier map datasets;

[0049] The corresponding secret graphs and carrier graphs in the historical secret graph and carrier graph dataset and the random noise vector are combined into a sample set;

[0050] The model parameters of the reversible steganographic network model are iteratively updated using the sample set until the loss function is minimized, thereby obtaining a trained reversible steganographic network model. The calculation formula of the loss function includes:

[0051] ,

[0052] ,

[0053] ,

[0054] Where, represents the loss function for reversible steganography network model training, represents the weight of the carrier graph steganographic loss, represents the carrier graph steganographic loss, represents the weight of the secret graph steganography loss, represents the secret graph steganography loss, represents the weight of the private key steganographic loss, Indicates the loss of private key steganography, represents the weight of the public key steganography loss, Indicates the public key steganographic loss, represents the index of the input vector map or the output vector map, represents the number of input carrier maps or output carrier maps or input secret maps or output secret maps, represents the L2 norm, Indicates the Output vector graph, Indicates the Input vector graph, represents the index of the input secret graph or the output secret graph, Indicates the output secret graph, Indicates the Input secret graph.

[0055] In a second aspect, the present invention provides a secret image open transmission device based on image steganography, comprising:

[0056] The sender module is used to obtain the secret image to be transmitted, and input the secret image, the prepared carrier image, and the random noise tensor into the pre-trained reversible steganography network model to obtain the secret image and forward output loss information, and then send the secret image to the receiver;

[0057] Receiver module: used to receive the secret image and input the secret image and backward input loss information into the pre-trained reversible steganography network model to obtain the recovered secret image;

[0058] Among them, the reversible steganography network model includes a noise obfuscation mechanism and a key steganography mechanism; the noise obfuscation mechanism is used to generate a fusion of an adaptive noise-guided secret graph and a carrier graph based on a random noise tensor, and to remove the decoupling of the adaptive noise-guided secret graph and the carrier graph; the key steganography mechanism is used to generate a public key and a public key label according to the corresponding noise of the forward input and the backward output, and reduce the steganography loss between the public key and the public key label; it is also used to generate a private key label and a private key according to the corresponding lost information of the forward output and the backward input, and reduce the steganography loss between the private key and the private key label.

[0059] In a third aspect, the present invention provides a computer storage medium having a computer program stored thereon. When the computer program is executed by a processor, the method for open transmission of a secret image based on image steganography as described in any one of the first aspects is implemented.

[0060] Compared with the prior art, the present invention achieves the following beneficial effects: combining the noise obfuscation mechanism with the key steganography mechanism, actively introducing adaptive noise in the hiding process, obfuscating the secret information in the carrier-secret map, and then combining the design of a public key generation module to generate a specific public key for secret hiding, thereby improving the concealment and security of secret communication. Moreover, since the noise obfuscation mechanism has the reversible characteristic, combined with the design of a private key generation module, a decryption private key is generated for the sender, thereby improving the similarity between the carrier-secret map and the secret-recovered secret. The recovered secret map can be gradually decoupled from the carrier-secret map along the reverse flow of the noise obfuscation module. A high-quality recovered secret map can still be reconstructed from the received carrier-secret map during the recovery process, effectively ensuring the quality of secret recovery. BRIEF DESCRIPTION OF THE DRAWINGS

[0061] Figure 1 The figure shows a structural diagram of a reversible steganographic network model in one embodiment of the present invention;

[0062] Figure 2 The following is a diagram showing the first step in the secret hiding process in one embodiment of the present invention. Forward flow framework diagram of NBM blocks;

[0063] Figure 3 The following is a diagram showing the first step in the secret recovery process in one embodiment of the present invention. Backward flow framework diagram of NBM blocks;

[0064] Figure 4 FIG2 is a schematic diagram showing a comparison of the secret recovery quality of the reversible steganographic network model of the present invention and other steganographic models in an embodiment of the present invention. DETAILED DESCRIPTION

[0065] The present invention will be further described below in conjunction with the accompanying drawings. The following embodiments are only used to more clearly illustrate the technical solutions of the present invention and are not intended to limit the scope of protection of the present invention.

[0066] Example 1

[0067] This embodiment establishes a carrier-secret interaction mechanism between the adaptive noise generation and carrier-secret decoupling processes at different depths in the network, so that a high-quality recovered secret graph can be gradually generated along the reverse flow of the noise obfuscation mechanism with the assistance of guided noise, effectively improving the concealment of secret communication and the quality of secret graph recovery.

[0068] This embodiment proposes a reversible steganonetwork model for open transmission scenarios to improve the quality and security of the payload image and prevent the loss of secret information in the recovered secret image. The secret image open transmission process based on the reversible steganonetwork model consists of two steps: a forward hiding process and a backward recovery process. During the forward hiding process, the noise blur module (NBM) in the reversible steganonetwork model gradually injects adaptive noise into the carrier and secret images at different network depths. Through an interactive mechanism, it guides the progressive fusion of the carrier and secret images, thereby blurring the secret information in the payload image and improving its security. Furthermore, the present invention designs a key steganography module (KSM) within the reversible steganonetwork model. Leveraging the consistent input and output dimensions of reversible neural networks, the KSM establishes soft consistency between the input and output in the forward and reverse directions of the network to generate adaptive public and private keys, further improving the quality of the payload image and recovered secret image. During the backward recovery process, due to the shared weights during the carrier secret fusion and separation process, the recovered secret image can be gradually decoupled into a high-quality payload image along the reverse direction of the NBM. The overall framework of the present invention is shown in FIG. Figure 1 As shown, the structure and mechanism of the noise fuzzy block NBM and the key steganography module KSM in the reversible steganography network model are described in detail below.

[0069] 1. Noise Blurring Mechanism

[0070] In the forward and backward flows of the noise blurring block, adaptive noise is gradually generated or removed during the carrier-secret fusion and decoupling processes to guide the hiding and recovery processes, respectively. In addition, the NBM block fuses the information of the carrier map and the secret map through adaptive noise weighting, which can dynamically guide the hiding process at different depths. The detailed structures of the NBM forward and backward flows are shown in Figure 2. Figure 2 、 3 shown.

[0071] In the forward hiding process, the carrier image, secret image and noise tensor are fused through weighted addition and multiplication operations. The information fusion direction is from the first to the second. NBM blocks.

[0072] Assume that The input to the forward NBM block is , , then the output yes 、 and The channel dimension is connected. That is, The forward NBM block generates a vector, secret, and noise tensor. The adaptive noise generated from the previous NBM block adjusts and guides the fusion of the current vector graph and the secret graph using a weighted operation. At the same time, the current noise is also adjusted by introducing the current vector and secret information to the subsequent NBM block. This mechanism in which noise guides the fusion process and the fusion process reacts to noise is called an interactive mechanism. The interaction mechanism in each NBM block can be expressed as follows:

[0073] ,

[0074] ,

[0075] ,

[0076] ,

[0077] Where, Indicates the The output of the noise-blurred block forward hidden, Indicates the The vector map of the forward hidden output of the noise blurred block, Indicates the The secret image of the forward hidden output of the noise-blurred block, Indicates the The noise blurring blocks forward hide the adaptive noise of the output, Indicates the The vector map of the forward hidden output of the noise blurred block, Indicates the The secret image of the forward hidden output of the noise-blurred block, Indicates the The noise blurring blocks forward hide the adaptive noise of the output, represents the Hadamard product, and represent the exponential function and the Sigmoid function respectively, 、 、 、 、 、 、 and They represent the conversion functions, 、 、 、 For carrier secret fusion, 、 、 and Guided fusion for adaptive noise. In addition, 、 By weighted adjustment and , accordingly, and is used by introducing and The information is used to adjust the adaptive noise of the next NBM block , No. The output of the NBM block yes 、 and Connected in the channel dimension, its size is same.

[0078] In the reverse recovery process, the weighted subtraction and division operations are performed from the encrypted image. The carrier graph, secret graph and noise vector are gradually decoupled in the process. The direction of information decoupling is opposite to that of information fusion, i.e. to the first NBM block.

[0079] Assume that The input of the reverse NBM block is , then the output yes 、 and The channel dimension connection. The calculation process of a reverse NBM block is as follows:

[0080] ,

[0081] ,

[0082] ,

[0083] ,

[0084] Where, Indicates the The output of the noise-blurred block is backward restored, Indicates the The noise blurred block is used to recover the output carrier map. Indicates the The secret image of the noise blurred block is recovered backward, Indicates the The noise blurring block backward recovers the adaptive noise of the output, Indicates the The noise blurred block is used to recover the output carrier map. Indicates the The secret image of the noise blurred block is recovered backward, Indicates the The noise blurring block backward recovers the adaptive noise of the output, 、 、 、 、 、 、 and Shared in the forward hiding and backward recovery process. Accordingly, 、 and Used to Decoupling Due to the shared parameters of the transformation function, the recovered secret image can be separated from the carried secret image with high quality in the reverse process of the hiding process, thus ensuring the effectiveness of secret communication.

[0085] 2. Key Steganography Mechanism

[0086] In response to the high requirements for steganographic security in open transmission scenarios and the potential information leakage risks of embedded carrier steganography methods, the present invention designs a key steganography mechanism to generate specific public and private keys for the sender and receiver for hiding and recovering secret information, thereby improving the security and recovery quality of steganography.

[0087] For reversible neural networks, due to their reversibility and parameter sharing properties, image reconstruction is optimal when the forward output and backward input remain consistent. However, for image steganography tasks, only the secret image in the forward output is transmitted to the receiver, and the backward input is used to initialize the untransmitted portion of the tensor to complete secret recovery. Therefore, information loss occurs between the forward output and backward input of a reversible neural network, making it difficult to decouple a satisfactory recovered secret image from the image steganography recovery process.

[0088] To address these issues, this chapter proposes a key steganography mechanism. This involves designing a private key generation module that establishes soft consistency between the forward output and the backward input by constructing a private key generator and a private key label generator. This approach indirectly approximates the forward output and the backward input without excessively restricting the forward hiding and information fusion processes. Furthermore, this approach provides a specific key for secret recovery, further enhancing its security.

[0089] like Figure 1 As shown, the present invention discards the forward output The private key tag is used as input to the private key tag generator. , when inputting backward, initialize the random tensor that is not transmitted Get the private key as input to the private key generator By using the private key and private key tags Establish the private key steganography loss to supervise the iterative learning of the private key. The loss can be expressed by the following formula:

[0090]

[0091] in, Indicates the loss of private key steganography, represents the index of a private key or private key label, represents the number of private keys or private key labels, represents the L2 norm, Indicates the A private key, Indicates the Private key tags

[0092] In order to further improve the steganographic security of the model, the present invention designs a public key generation module, which uses the public key to further learn the noise of the forward input as a hiding process. As input to the public key generator to obtain the final public key , and the recovered noise of the backward output The public key tag is used as input to the public key tag generator. .

[0093] By using the public key tag and public key The public key steganography loss is established between the two to supervise the iterative learning of the public key. The loss can be expressed by the following formula:

[0094]

[0095] in, Indicates the public key steganographic loss, represents the index of a public key or public key label, represents the number of public keys or public key labels, Indicates the A public key, Indicates the A public key tag.

[0096] The public key label generation module, public key generation module, private key label generation module, and private key generation module share the same architecture, consisting of six convolutional blocks. The first five convolutional blocks contain three layers: a 3×3 convolutional layer, a batch normalization layer, and a Reluctant Unified Unit (ReLU) activation layer. The sixth convolutional block consists of a 3×3 convolutional layer and a sigmoid activation layer. The input and output channels of the convolutional layers in different convolutional blocks are {12, 64, 128, 256, 128, 64} and {64, 128, 256, 128, 64, 12}, respectively.

[0097] 3. Secret Image Steganography

[0098] The method for performing secret graph steganography using a reversible steganographic network model specifically includes the following steps:

[0099] 1. Secret forward hiding process

[0100] First, along the forward flow of NBM, the carrier graph and the secret graph are fused by noise guidance to generate the secret graph. Then, KSM generates an adaptive public key by establishing soft consistency between the forward input and the backward output, which further improves the quality and security of the secret graph. The input of the forward hiding process is the carrier graph. , Secret Picture and a random noise tensor ,in, and Represents height and width respectively. and Represent the carrier and secret channel respectively, and and denote the number of carriers and secret graphs respectively. In addition, when the number of hidden secret graphs When , the secret graph is connected in the channel dimension. It is worth noting that the input random noise tensor of the first layer in NBCS-Net From Gaussion noise , while in other NBM blocks, noise can be adaptively generated through iterative learning updates. The forward process of NBM in the forward hidden process can be expressed as follows:

[0101] ,

[0102] in, Represents a connection operation on a channel, Indicates a secret map. Indicates that the forward output loses information, represents the inverse compression operation, represents the product of mappings, represents the index of the noise blurred block, represents the number of noise blurred blocks, Indicates compression operation, Represents the prepared vector map, represents the secret graph to be transmitted, represents the public key, represents the number of noise-blurred blocks in NBM, and ,Similar to ISN and DEEPMIH, compression and inverse compression operations are used to preprocess input and output information,compression operation Enter The size from Convert to , accordingly, the inverse compression operation The opposite of compression will output The size from Convert to The output of the forward hiding process is the encrypted image and lost information The concatenation of the two (the remaining information of the hidden output) is performed, and only the cryptogram will be transmitted for covert communication.

[0103] During the transmission of the stegosaurus, a third party can intercept it and determine whether it contains secret information, or even extract the secret information by comparing it with the carrier image. This poses a significant threat to the security of secret communications. To address this challenge, the present invention proposes KSM, which establishes soft consistency between the network input and output, further improving the similarity between the carrier image and the stegosaurus, thereby enhancing the security of steganography.

[0104] 2. Secret Backward Recovery Process

[0105] Thanks to the reversible nature of Invertible Neural Networks (INNs), the recovered secret image can be gradually separated from the secret image along the reverse direction of NBM without retraining a new recovery network. The input of the backward recovery process is the received secret image. and The connection in the channel dimension, so the backward recovery process can be expressed as follows:

[0106] ,

[0107] in, Represents the recovered vector graph, represents the recovered secret graph, represents the recovered noise, Indicates the received encrypted image. represents the private key. Since the NBM parameters are shared in both the forward and reverse processes, the independence of the carrier and the secret information is well maintained during the hiding process. This facilitates the high-quality decoupling of the recovered secret image from the carrier image.

[0108] like Figure 4 As shown, using the same carrier graph and secret graph, Weng, UDH, ISN, DEEPMIH and the steganographic network model of the present invention are used to perform secret steganographic tasks respectively, and the carrier graph, secret graph, carrier graph, carrier difference graph and recovered secret graph are visualized respectively. From the visualization results, it can be seen that the carrier difference graph of the reversible steganographic network model of the present invention has the least information and the recovered secret graph is the clearest.

[0109] 3. The training process of the reversible steganography network model is as follows:

[0110] Step 1: Initialize parameters: carrier graph sample , Secret Image Sample and noisy images ; Number of secret images and the number of carrier images ; Number of noise blur blocks ; Weights in the loss function 、 as well as ; Initial learning rate ; Number of batch samples ; Maximum number of iterations .

[0111] Step 2: Repeat the following steps (steps within an epoch) until the network converges:

[0112] Generate the encrypted image using the following formula and discard information :

[0113] ;

[0114] Generate public key by public key generator ;

[0115] Generate private key label by private key label generator ;

[0116] Sender uploads stego map to cloud server or other open transmission scenarios ;

[0117] Receiver receives stego map from open transmission scenarios ;

[0118] Generate private key by private key generator ;

[0119] Generate recovery secret map by the following formula :

[0120] ;

[0121] Generate public key label by public key label generator ;

[0122] Calculate the overall loss according to the following formula, and update the model parameters by gradient descent algorithm.

[0123] ,

[0124] ,

[0125] ,

[0126] ,

[0127] ,

[0128] wherein, represents the loss function of reversible steganography network model training, represents the weight of carrier map steganography loss, represents the carrier map steganography loss, represents the weight of secret map steganography loss, represents the secret map steganography loss, represents the weight of private key steganography loss, represents the weight of public key steganography loss, represents the index of input carrier map or output carrier map, represents the number of input carrier maps or output carrier maps or input secret maps or output secret maps, Indicates the Output vector graph, Indicates the Input vector graph, represents the index of the input secret graph or the output secret graph, Indicates the output secret graph, Indicates the Input secret graph.

[0129] Example 2

[0130] This embodiment provides a secret image open transmission device based on image steganography, comprising:

[0131] The sender module is used to obtain the secret image to be transmitted, and input the secret image, the prepared carrier image, and the random noise tensor into the pre-trained reversible steganography network model to obtain the secret image and forward output loss information, and then send the secret image to the receiver;

[0132] Receiver module: used to receive the secret image and input the secret image and backward input loss information into the pre-trained reversible steganography network model to obtain the recovered secret image;

[0133] Among them, the reversible steganography network model includes a noise obfuscation mechanism and a key steganography mechanism; the noise obfuscation mechanism is used to generate a fusion of an adaptive noise-guided secret graph and a carrier graph based on a random noise tensor, and to remove the decoupling of the adaptive noise-guided secret graph and the carrier graph; the key steganography mechanism is used to generate a public key and a public key label according to the corresponding noise of the forward input and the backward output, and reduce the steganography loss between the public key and the public key label; it is also used to generate a private key label and a private key according to the corresponding lost information of the forward output and the backward input, and reduce the steganography loss between the private key and the private key label.

[0134] The device provided in this embodiment can execute the method provided in any step of Example 1, and has the corresponding functional modules and beneficial effects of the execution method.

[0135] Example 3

[0136] This embodiment provides a computer storage medium having a computer program stored thereon. When the computer program is executed by a processor, the secret image open transmission method based on image steganography as described in any step of Embodiment 1 is implemented.

[0137] Those skilled in the art will appreciate that the embodiments of the present application may be provided as methods, systems, or computer program products. Therefore, the present application may take the form of an entirely hardware embodiment, an entirely software embodiment, or an embodiment combining software and hardware. Furthermore, the present application may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to magnetic disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0138] The present application is described with reference to the flowcharts and / or block diagrams of the methods, devices (systems), and computer program products according to the embodiments of the present application. It should be understood that each process and / or block in the flowchart and / or block diagram, as well as the combination of processes and / or blocks 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 data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. 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.

[0139] These computer program instructions may also be stored in a computer readable memory that can direct a computer or other programmable data processing device to work in a specific manner, so that the instructions stored in the computer readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 a process or multiple processes and / or boxes Figure 1 The function specified in one or more boxes.

[0140] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operational steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing the instructions executed on the computer or other programmable device for implementing the process. Figure 1 a process or multiple processes and / or boxes Figure 1 A step that specifies a function in one or more boxes.

[0141] The embodiments of the present invention are described above in conjunction with the accompanying drawings, but the present invention is not limited to the above-mentioned specific implementation methods. The above-mentioned specific implementation methods are merely illustrative and not restrictive. Under the guidance of the present invention, ordinary technicians in this field can also make many forms without departing from the scope of protection of the purpose of the present invention and the claims, which are all protected by the present invention.

Claims

1. A secret image open transmission method based on image steganography, characterized in that: include: The sender obtains the secret image to be transmitted, and inputs the secret image, the prepared carrier image, and the random noise tensor into a pre-trained reversible steganography network model to obtain the secret image and forward output loss information, and then sends the secret image to the receiver; The receiver receives the steganographic image and inputs the steganographic image and the backward input loss information into the pre-trained reversible steganographic network model to obtain the recovered secret image. The reversible steganographic network model includes a noise obfuscation mechanism and a key steganography mechanism; the noise obfuscation mechanism is used to generate a fusion of an adaptive noise-guided secret graph and a carrier graph based on a random noise tensor, and to decouple the adaptive noise-guided secret graph and the carrier graph; the key steganography mechanism is used to generate a public key and a public key label based on the corresponding noise of the forward input and backward output, respectively, and to reduce the steganalysis loss between the public key and the public key label; It is also used to generate a private key tag and a private key according to the corresponding loss information of the forward output and the backward input, respectively, and to reduce the steganographic loss between the private key and the private key tag.

2. The secret graph open transmission method based on image steganography according to claim 1 is characterized in that: The noise blurring mechanism includes a plurality of noise blurring blocks connected in series; When the reversible steganalysis network model is used for forward hiding, the noise blurring block includes multiple weighted multiplication and addition operations; When the reversible steganograph network model is used for backward recovery, the noise blurring block includes multiple weighted division and subtraction operations.

3. The secret graph open transmission method based on image steganography according to claim 2 is characterized in that: The key steganography mechanism includes a public key generation module and a private key generation module which are independent of each other; The public key generation module includes a public key generator and a public key label generator that are independent of each other, and the private key generation module includes a private key generator and a private key label generator that are independent of each other; The public key generator, public key label generator, private key generator and private key label generator each include multiple convolution blocks.

4. The secret graph open transmission method based on image steganography according to claim 2 is characterized in that: During the forward hiding process, the noise blurring mechanism performs the following operations: Each noise fuzzy block guides the input secret image and carrier image to be fused using weighted multiplication and addition operations according to the input noise tensor, and iteratively updates the noise tensor during the fusion process of the secret image and carrier image. Among them, the input of the first noise fuzzy block is the secret image to be transmitted, the prepared carrier image and random Gaussian noise, and the input of other noise fuzzy blocks is the fused secret image and carrier image output by the previous noise fuzzy block and the iteratively updated adaptive noise. The output of the forward hidden of the noise blurred block is as follows: ， ， ， ， Where, Indicates the The output of the noise-blurred block forward hidden, Indicates the The vector map of the forward hidden output of the noise blurred block, Indicates the The secret image of the forward hidden output of the noise-blurred block, Indicates the The noise blurring blocks forward hide the adaptive noise of the output, Indicates the The vector map of the forward hidden output of the noise blurred block, Indicates the The secret image of the forward hidden output of the noise-blurred block, Indicates the The noise blurring blocks forward hide the adaptive noise of the output, represents the Hadamard product, and Represent the exponential function and Sigmoid function respectively, 、 、 、 、 、 、 and They represent the conversion functions, 、 、 、 For carrier secret fusion, 、 、 and Guided fusion for adaptive noise; After each noise fuzzy block completes forward hiding, the secret image, carrier image, and pre-generated public key output by the last noise fuzzy block are concatenated in the channel dimension to output the concatenated carrier image and forward output loss information. The calculation formula is as follows: ， Where, Represents a connection operation on a channel, Indicates a secret map. Indicates that the forward output loses information, represents the inverse compression operation, represents the product of mappings, represents the index of the noise blurred block, represents the number of noise blurred blocks, Indicates compression operation, Represents the prepared vector map, represents the secret graph to be transmitted, Represents a public key.

5. The secret graph open transmission method based on image steganography according to claim 2 is characterized in that: During the backward recovery process, the noise blurring mechanism performs the following operations: Each noise fuzzy block uses weighted division and subtraction operations to decouple the input secret graph and carrier graph according to the input noise tensor, and iteratively recovers the noise tensor during the decoupling process of the secret graph and carrier graph. Among them, the input of the last noise fuzzy block is the received secret graph and the pre-generated private key spliced ​​in the channel dimension, and the input of other noise fuzzy blocks is the decoupled secret graph and carrier graph output by the next noise fuzzy block and the iteratively recovered adaptive noise. The output of the backward recovery of the noise blurred block is as follows: ， ， ， ， Where, Indicates the The output of the noise-blurred block is backward restored, Indicates the The noise blurred block is used to recover the output carrier map. Indicates the The secret image of the noise blurred block is recovered backward, Indicates the The noise blurring block backward recovers the adaptive noise of the output, Indicates the The noise blurred block is used to recover the output carrier map. Indicates the The secret image of the noise blurred block is recovered backward, Indicates the The noise blurring block backward recovers the adaptive noise of the output, represents the Hadamard product, and Represent the exponential function and Sigmoid function respectively, 、 、 、 、 、 、 and They represent the conversion functions, 、 、 、 For carrier secret decoupling, 、 、 and Guided decoupling for adaptive noise; After the backward recovery of each noise fuzzy block is completed, the secret image output by the first noise fuzzy block is obtained as the recovered secret image. The calculation formula is as follows: ， Where, Represents a connection operation on a channel, Represents the recovered vector graph, represents the recovered secret graph, represents the recovered noise, represents the inverse compression operation, represents the product of mappings, represents the index of the noise blurred block, represents the number of noise blurred blocks, Indicates the received encrypted image. Represents a private key.

6. The secret graph open transmission method based on image steganography according to claim 3 is characterized in that: The key steganography mechanism performs the following operations: In the forward hidden process, the forward output loses information Enter the private key label generator to generate a private key label , and the noise tensor output by the last noise blur block Enter the public key generator to generate a public key ; During the backward recovery process, the untransmitted random tensors The lost information is input as the backward input to the private key generator to generate the private key , and the restored noise Enter the public key label generator to generate a public key label .

7. The secret graph open transmission method based on image steganography according to claim 3 is characterized in that: The key steganography mechanism also performs the following operations: In the Private Key tab and private key A private key steganographic loss is established between them to supervise the iterative learning of the private key. The calculation formula of the private key steganographic loss is as follows: ， Where, Indicates the loss of private key steganography, represents the index of a private key or private key label, represents the number of private keys or private key labels, represents the L2 norm, Indicates the A private key, Indicates the a private key tag; In the public key and public key tags A public key steganography loss is established between them to supervise the iterative learning of the public key. The calculation formula of the public key steganography loss is as follows: ， Where, Indicates the public key steganographic loss, represents the index of a public key or public key label, represents the number of public keys or public key labels, Indicates the A public key, Indicates the A public key tag.

8. The secret graph open transmission method based on image steganography according to claim 1 is characterized in that: The training method of the reversible steganography network model includes: Obtain historical secret map and carrier map datasets; The corresponding secret graphs and carrier graphs in the historical secret graph and carrier graph dataset and the random noise vector are combined into a sample set; The model parameters of the reversible steganographic network model are iteratively updated using the sample set until the loss function is minimized, thereby obtaining a trained reversible steganographic network model. The calculation formula of the loss function includes: ， ， ， Where, represents the loss function for training the reversible steganography network model, represents the weight of the carrier graph steganographic loss, represents the carrier graph steganographic loss, represents the weight of the secret graph steganography loss, represents the secret graph steganography loss, represents the weight of the private key steganographic loss, Indicates the loss of private key steganography, represents the weight of the public key steganography loss, Indicates the public key steganographic loss, represents the index of the input vector map or the output vector map, represents the number of input carrier maps or output carrier maps or input secret maps or output secret maps, represents the L2 norm, Indicates the Output vector graph, Indicates the Input vector graph, represents the index of the input secret graph or the output secret graph, Indicates the output secret graph, Indicates the Input secret graph.

9. A secret image open transmission device based on image steganography, characterized in that: include: The sender module is used to obtain the secret image to be transmitted, and input the secret image, the prepared carrier image, and the random noise tensor into the pre-trained reversible steganography network model to obtain the secret image and forward output loss information, and then send the secret image to the receiver; Receiver module: used to receive the secret image and input the secret image and backward input loss information into the pre-trained reversible steganography network model to obtain the recovered secret image; The reversible steganographic network model includes a noise obfuscation mechanism and a key steganography mechanism; the noise obfuscation mechanism is used to generate a fusion of an adaptive noise-guided secret graph and a carrier graph based on a random noise tensor, and to decouple the adaptive noise-guided secret graph and the carrier graph; the key steganography mechanism is used to generate a public key and a public key label based on the corresponding noise of the forward input and backward output, respectively, and to reduce the steganalysis loss between the public key and the public key label; It is also used to generate a private key tag and a private key according to the corresponding loss information of the forward output and the backward input, respectively, and to reduce the steganographic loss between the private key and the private key tag.

10. A computer storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the secret image open transmission method based on image steganography according to any one of claims 1 to 8 is implemented.

Images