CN115731089A - Component energy-based double-task image steganography method - Google Patents
Component energy-based double-task image steganography method Download PDFInfo
- Publication number
- CN115731089A CN115731089A CN202211619646.9A CN202211619646A CN115731089A CN 115731089 A CN115731089 A CN 115731089A CN 202211619646 A CN202211619646 A CN 202211619646A CN 115731089 A CN115731089 A CN 115731089A
- Authority
- CN
- China
- Prior art keywords
- image
- energy
- steganographically
- secret
- steganographic
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Granted
Links
- 238000000034 method Methods 0.000 title claims abstract description 76
- 238000000605 extraction Methods 0.000 claims abstract description 30
- 238000013528 artificial neural network Methods 0.000 claims abstract description 22
- 230000002441 reversible effect Effects 0.000 claims abstract description 21
- 230000008569 process Effects 0.000 claims description 30
- 230000006870 function Effects 0.000 claims description 24
- 238000005457 optimization Methods 0.000 claims description 7
- 239000000126 substance Substances 0.000 claims description 6
- 238000011176 pooling Methods 0.000 claims description 4
- 238000012804 iterative process Methods 0.000 claims 1
- 238000006243 chemical reaction Methods 0.000 description 5
- 238000013527 convolutional neural network Methods 0.000 description 4
- 238000003062 neural network model Methods 0.000 description 4
- 230000005540 biological transmission Effects 0.000 description 3
- 238000004891 communication Methods 0.000 description 3
- 230000008878 coupling Effects 0.000 description 3
- 238000010168 coupling process Methods 0.000 description 3
- 238000005859 coupling reaction Methods 0.000 description 3
- 125000004122 cyclic group Chemical group 0.000 description 2
- 238000010586 diagram Methods 0.000 description 2
- 239000000284 extract Substances 0.000 description 2
- 230000006872 improvement Effects 0.000 description 2
- 238000012549 training Methods 0.000 description 2
- 230000003044 adaptive effect Effects 0.000 description 1
- 230000001174 ascending effect Effects 0.000 description 1
- 230000002457 bidirectional effect Effects 0.000 description 1
- 230000008859 change Effects 0.000 description 1
- 238000012512 characterization method Methods 0.000 description 1
- 238000000354 decomposition reaction Methods 0.000 description 1
- 238000013136 deep learning model Methods 0.000 description 1
- 230000009977 dual effect Effects 0.000 description 1
- 230000010365 information processing Effects 0.000 description 1
- 238000010801 machine learning Methods 0.000 description 1
- 238000013507 mapping Methods 0.000 description 1
- 230000006386 memory function Effects 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 238000011084 recovery Methods 0.000 description 1
- 238000006467 substitution reaction Methods 0.000 description 1
- 239000000758 substrate Substances 0.000 description 1
- 230000001131 transforming effect Effects 0.000 description 1
- 230000000007 visual effect Effects 0.000 description 1
Images
Landscapes
- Image Processing (AREA)
Abstract
The disclosure belongs to the field of image encryption, and particularly relates to a component energy-based double-task image steganography method, which comprises the following steps: acquiring a data pair of an image to be steganographically, wherein the data pair of the image to be steganographically comprises a secret image to be steganographically and an overlay image to be steganographically; respectively obtaining first energy hidden variable pairs corresponding to the to-be-hidden-writing energy data pairs on the basis of the to-be-hidden-writing image data pairs through a component energy extraction model; obtaining hidden variable pairs based on the first energy hidden variable pairs through a reversible neural network; the hidden variable pair obtains a hidden image through an image generation module; acquiring a noise image; obtaining a second energy hidden variable pair based on the steganographic image and the noise image through a component energy extraction model; obtaining a revealing hidden variable pair based on a second energy hidden variable pair through a reversible neural network; and revealing the hidden variable pair to obtain a secret image and an overlay image through an image generation module so as to accurately realize steganography and revealing.
Description
Technical Field
The disclosure belongs to the field of image encryption, and particularly relates to a component energy-based double-task image steganography method.
Background
The image steganography is a process of hiding a secret image containing sensitive information into an overlay image without the sensitive information to generate a steganographic image, and only allowing an informed receiver to obtain the overlay image and the secret image through image disclosure. For information communication security, it is generally required that the steganographic image and the overlay image visually look indistinguishable. Unlike image encryption and traditional steganography, image steganography has higher requirements on the information capacity of steganographic images, the invisibility of secret images and the security of information processing procedures, and thus the task is more challenging. The image steganography has high practicability, and requires that the steganography image containing secret image information is difficult to distinguish from the coverage image, so that the image steganography is more suitable for encrypted communication and ensures the security of secret data communication.
The traditional image steganography method generally consists of two sub-functional modules, namely image steganography and image revealing. The method is mainly realized through the following two technical means, wherein the secret image information is hidden in the overlay image through a complex replacement and conversion algorithm, but with the improvement of computing resource performance, the replacement and conversion algorithm cannot guarantee that the steganographic image has reasonable resolution and capacity, and the data security cannot be guaranteed. And secondly, the dual-function modules based on the machine learning deep neural network are in loose coupling connection, and the two function modules are respectively provided with independent parameter sets, so that data quality problems such as image color distortion and texture flaws can be caused. Furthermore, conventional image steganography methods have little to no security concerns, such that the hidden information is easily detected.
The method based on the reversible neural network respectively completes the image steganography process and the image revealing process by using the same deep neural network functional module to perform bidirectional action. The image revealing is a reverse process of the image hiding, so that the functional module can obtain all network parameters required for completing the image hiding and image revealing functions only by training once, which is the fundamental difference between the method and the traditional method. At present, the method reaches the most advanced level in the aspects of secret image recovery accuracy, hidden security and invisibility. However, this method has high model coupling degree, weak expansibility, and limited input image characterization capability.
Disclosure of Invention
The present disclosure is made based on the above-mentioned needs of the prior art, and an object of the present disclosure is to provide a component energy-based dual-task image steganography method to accurately steganography and reveal a secret image.
In order to solve the above problem, the technical solution provided by the present disclosure includes:
a component energy-based dual-task image steganography method is provided, which comprises the following steps: acquiring a data pair of an image to be steganographically, wherein the data pair of the image to be steganographically comprises a secret image to be steganographically and an overlay image to be steganographically; respectively obtaining component energy pairs of the to-be-steganographic energy data pairs based on the to-be-steganographic image data pairs through a component energy extraction model, wherein the energy extraction model comprises a two-dimensional convolution layer, a full connection layer and a global average pooling layer, and the component energy of the to-be-steganographic energy data pairs comprises the component energy of the to-be-steganographic secret image and the component energy of the to-be-steganographic coverage image; obtaining hidden variable pairs through a reversible neural network based on component energy pairs of the energy data pairs to be hidden; the hidden variable pair obtains a hidden image through an image generation module; acquiring a noise image; obtaining component energies of the steganographic image and the noise image respectively based on the steganographic image and the noise image through the component energy extraction model; obtaining a revealing hidden variable pair based on the component energy of the steganographic image and the component energy of the noise image through a reversible neural network; and the revealing hidden variable pair obtains a secret image and an overlay image through the image generation module.
Preferably, when acquiring the pair of to-be-steganographic image data, the physical scale information of the image in the pair of to-be-steganographic image data is acquired at the same time.
Preferably, the parameters of the component energy extraction model are self-adjusted according to the input physical scale information of the image to be steganographically.
Preferably, the secret image to be steganographed obtains an energy steganography variable set of the secret image to be steganographed through a component energy extraction modelZ n ={z 1 ，z 2 ，…，z n And (c) the step of (c) in which,z i =Enc θ (x secreti )，z i a hidden variable representing the ith secret image to be steganographically,Enc θ () Representing the encoding process of the secret image to be steganographically treated,x secreti is shown asiA secret image to be steganographically; the image to be steganographically covered is subjected to component energy extraction model to obtain an energy steganography variable set of the image to be steganographically coveredZ' n ={z' 1 ，z' 2 ，…，z' n And (c) the step of (c) in which,z' i =Enc θ '(x coveri )，z' i a hidden variable representing the ith to-be-steganographically overlaid image,Enc θ '() Representing the encoding process for the to-be-steganographically overlaid image,x coveri is shown asiAnd covering the image to be steganographically.
Preferably, the loss function of the method comprises an energy loss function and a generation loss function.
Preferably, the energy loss function is expressed as:
wherein the content of the first and second substances,
a latent variable representation of the foreground information energy representing the steganographic image,
a background information energy latent variable representation representing a steganographic image,
foreground information energy latent variable representation representing a secret image,
a background information energy latent variable representation representing the secret image.
Preferably, the generation loss function is expressed as:
wherein the content of the first and second substances,Ɛ、μand
are the weights in the optimization process and,x cover representing the image to be steganographically overlaid,x' cover a representation of the overlay image is shown,x secret representing the secret image to be steganographically represented,x' secret a representation of the secret image is shown,Mrepresenting the total number of frames of images of each input model in the algorithm model loss iteration process,
is the firstiThe frame input is the average of the overlaid images,
is a disclosing processiThe frame generates an average value of the overlaid image,
is the firstiThe frame input covers the variance of the image,
is a process of disclosureiThe frame generates a variance that covers the image,
is shown asiFrame input overlay image and uncovering processiThe frame generation covers the covariance between the images,
is the firstiThe average value of the frame input secret image,
is a process of disclosureiThe frame generates an average value of the secret image,
is the firstiThe variance of the frame input secret image is
Disclosure of ProcessiThe frame generates the variance of the secret image,
is shown asiFrame input secret image and revealing processiThe frame generates a covariance between the secret images,c 1 andc 2 is a constant for ensuring learning fairness.
Compared with the prior art, the invention provides a component energy-based dual-task deep neural network model, which consists of three parts: the system comprises a component energy extraction model, a reversible neural network and an image generation module. The component energy extraction model encodes the input image into combined energy components representing the foreground and the background respectively by using an energy equation. The reversible neural network is composed of a deep neural network functional module composed of a cyclic neural network with a memory function and a one-dimensional convolutional neural network layer, so that the combined energy information is hidden, and the safety of the hidden information is further enhanced. And the image generation module respectively completes the decoding process of secret energy obtained by secret transmission aiming at the functional requirements of image steganography and image revelation and outputs the target image. The data information transmitted among the three modules is the image characteristics containing high-level semantic information. And secondly, the energy information dimensionality generated by the component energy extraction model can be adaptively adjusted according to the physical scale of the input image, the function of the image generation module depends on an energy component decomposition network, and the model structure and the function coupling degree are enhanced. And finally, the constructed reversible neural network function is realized mainly depending on a one-dimensional convolutional neural network layer, the structure is simple, the calculated amount is low, the initial layer structure can be adaptively adjusted along with the change of the image resolution, and the reversible network is information-lossless, so that the model can keep the detail information of input data, and the adaptive adjustment energy information representation capability is strong.
Drawings
In order to more clearly illustrate the embodiments of the present specification or the technical solutions in the prior art, the drawings needed to be used in the description of the embodiments or the prior art will be briefly described below, it is obvious that the drawings in the following description are only some embodiments described in the embodiments of the present specification, and other drawings can be obtained by those skilled in the art according to the drawings.
FIG. 1 is a flow chart illustrating the steps of a dual-task image steganography method based on component energy in accordance with the present invention;
FIG. 2 is a schematic diagram of steganography process of a component energy-based dual-task image steganography method according to the present invention;
FIG. 3 is a schematic diagram illustrating a process of exposing a dual-task image steganography method based on component energy according to the present invention.
Detailed Description
In order to make the objects, technical solutions and advantages of the embodiments of the present application clearer, the technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the drawings in the embodiments of the present application, and it is obvious that the described embodiments are some embodiments of the present application, but not all embodiments. All other embodiments, which can be derived by a person skilled in the art from the embodiments given herein without making any creative effort, shall fall within the protection scope of the present application.
In the description of the embodiments of the present disclosure, it should be noted that, unless otherwise explicitly stated or limited, the term "connected" should be interpreted broadly, and may be, for example, a fixed connection, a detachable connection, or an integral connection, which may be a mechanical connection, an electrical connection, which may be a direct connection, or an indirect connection via an intermediate medium. The specific meaning of the above terms in the present disclosure can be understood by those of ordinary skill in the art as appropriate.
The terms "top," "bottom," "above … …," "down," and "above … …" as used throughout the description are relative positions with respect to components of the device, such as the relative positions of the top and bottom substrates inside the device. It will be appreciated that the devices are multifunctional, regardless of their orientation in space.
For the purpose of facilitating understanding of the embodiments of the present application, the following description will be made in terms of specific embodiments with reference to the accompanying drawings, which are not intended to limit the embodiments of the present application.
The embodiment provides a component energy-based dual-task image steganography method, as shown in fig. 1-3.
As shown in fig. 1, the component energy-based dual task image steganography method includes:
s1, acquiring a data pair of an image to be steganographically, wherein the data pair of the image to be steganographically comprises a secret image to be steganographically and an overlay image to be steganographically.
The secret image to be steganographically and the covered image to be steganographically are respectively from a secret image data set and a covered image data set. Specifically, one image is randomly extracted from the secret image data set and the coverage image data set respectively to form a pair of data pairs of images to be steganographically, and the data pairs of images to be steganographically comprise secret images to be steganographicallyx secret And to-be-steganographically overlay imagesx cover . In addition, in the process, the physical scale information of the secret image to be steganographically and the covered image to be steganographically is acquired simultaneously.
And S2, respectively obtaining component energy pairs of the to-be-steganographic energy data pairs based on the to-be-steganographic image data pairs through a component energy extraction model, wherein the energy extraction model comprises a two-dimensional convolution layer, a full connection layer and a global average pooling, and the component energy of the to-be-steganographic energy data pairs comprises the component energy of the to-be-steganographic secret image and the component energy of the to-be-steganographic coverage image.
Inputting the acquired data pairs of the image to be steganographically into the component energyExtracting the secret image to be steganographed from the model Ex secret And to-be-steganographically overlay imagesx cover The component energy of (1). The component energy extraction model E comprises a neural network model, and the neural network model comprises a two-dimensional convolution layer, a full connection layer and a global average pooling layer. And the component energy extraction model adjusts the parameters of the component energy extraction model E according to the information of the input image, wherein the information comprises the previously acquired physical scale information. Further, when the input of the component energy extraction model is a secret image to be steganographically, the component energy extraction model is E 1 (ii) a When the input of the component energy extraction model is an image to be steganographically covered, the component energy extraction model is E 2 。E 1 And E 2 And respectively carrying out self-adaptive adjustment according to the obtained physical scale information of the secret image to be steganographed and the physical scale information of the covered image to be steganographed, so that the dimension of the energy variable representing the foreground information and the background information extracted from each image is the same as the dimension of the image.
The component energy extraction model can realize the function of a learnable encoder, and encodes the input secret image to be steganographically into a set of steganographically hidden variable energy of the secret image to be steganographically and capable of representing component energy of the secret image to be steganographicallyZ n ={z 1 ，z 2 ，…，z n And (c) the step of (c) in which,z i = Enc θ (x secreti )，z i a hidden variable representing the ith secret image to be steganographically,Enc θ () Representing the encoding process of the secret image to be steganographically treated,x secreti is shown asiA secret image to be steganographically; coding an input image to be steganographically covered into a set of steganographically covered image energy steganographic variables capable of representing component energy of the image to be steganographically coveredZ' n ={z' 1 ，z' 2 ，…，z' n -means for, among other things,z' i =Enc θ '(x coveri )，z' i a hidden variable representing the ith image to be steganographically overlaid,Enc θ '() Representing the encoding process for the to-be-steganographically overlaid image,x coveri is shown asiThe overlay image is to be steganographically. Combining implicit variablesZ n Conversion into component energy of secret image to be steganographedEn 1 Combining the hidden variablesZ' n Conversion to component energy of an overlay image to be steganographically representedEn 2 。
Followed by unsupervised learningL 1 Parameter optimization procedure for norm minimizes model parameter θ:
L 1 (θ) For the norm of the minimization of the parameter theta,arg x min() In order to minimize the function for the parameter,E θ () The parameter set within which the model is extracted for the component energy isθAs a function of the time of day,x i for the ith secret image to be steganographically oriAnd covering the image to be steganographically.
And S3, obtaining hidden variable pairs through a reversible neural network based on component energy pairs of the to-be-hidden-writing energy data pairs.
The component energy of the secret image to be steganographically displayedEn 1 And the component energy of the image to be steganographically overlaidEn 2 Simultaneously input into a reversible neural network Inver, and pass through supervised model parameters
The forward energy hiding correlation transmission is carried out by the minimized energy constraint learning to obtain a hiding variable pairR n AndR' n expressed as:
wherein the content of the first and second substances,
indicating a hidden relationiZhang Daiyin writes the foreground energy of the secret image or the overlay image to be steganographically,
indicating that the concealment associated corresponds to foreground energyiZhang Daiyin writes the background energy of the secret image or the overlay image to be steganographically covered. Since two images are included after the hidden association, the images are concealed and relatediIs 1 or 2.
Through optimization, the model is forced to be biased to learn the foreground characteristics of the image, and energy is concentrated in foreground information. In particular, reversible neural networksInverByMAnd the hiding sub-function modules are formed to complete the energy hiding process. For the firstjA hidden sub-function module which inputsjComponent energy of a secret image to be steganographically
And a firstjComponent energy of an overlay image to be steganographically represented
Output is
And
expressed as:
wherein the content of the first and second substances,βis a sigmoidThe function is multiplied by a constant factor,
the point is shown as being an operation,ρ()，
and
and generating a residual error model ResNet functional block proposed in the countermeasure network by adopting the enhanced super-resolution. The output of the last hidden sub-function module is
And
and inputting the image into an image generator to finish the image generation. In order to simplify the presentation of the presentation,
，
。
furthermore, the neural network model comprises 8 hiding sub-function algorithm modules with the same structure, and the 8 algorithm modules are arranged in an ascending order during image steganography, namely [ y 1 ；y 2 ；…；y 8 ]. In addition, each ResNet functional component comprises 5 one-dimensional convolutional neural network layers and 1 layer of cyclic neural network layer LSTM, each functional component extracts high-level energy characteristics with different visual receptive field scales respectively, the characteristics of the 5 one-dimensional convolutional neural network layers are cascaded, the receptive field of the energy characteristics is enhanced, the association granularity of the energy characteristics is finer, nonlinearity is further increased from an algorithm level, and the energy characteristic expression capability of the reversible neural network is improved.
And S4, obtaining a steganographic image by the hidden variable pair through an image generation module.
Will be at the topHidden variables obtained in the processR n AndR' n input image generation moduleGIn generating steganographic imagesx stego Sum and difference imagex error To implement image steganography. When the input image is an energy hidden variable of a secret image to be hidden, the image generation module G 1 (ii) a When the input image is an energy hidden variable of an image to be hidden-written and covered, the image generation module G 2 Two hidden variables pass through G respectively 1 And G 2 Obtaining a steganographic image and a difference image, and G 1 And G 2 Are all realized by a supervised model parameter optimization process through random gradient descent.
S5, acquiring a noise image.
Obtaining a noise image randomly generated with the same scale as the steganographic image obtained abovex noise 。
And S6, respectively obtaining the component energy of the steganographic image and the noise image based on the steganographic image and the noise image through the component energy extraction model.
Inputting the steganographic image and the noise image into the component energy extraction model E to respectively extract steganographic imagesx stego And noisy imagesx noise Is used to determine the energy implicit variable. And the component energy extraction model performs self-adaptive adjustment according to the physical scales of the steganographic image and the noise image, namely performs component energy mapping in the same energy conversion mode as S2. Finally, the component energy of the steganographic image and the noise image is obtainedEn' 1 AndEn' 2 。
and S7, obtaining a revealing hidden variable pair based on the component energy of the steganographic image and the component energy of the noise image through a reversible neural network.
Transforming the component energies of the steganographic imageEn' 1 And component energy of noisy imagesEn' 2 Simultaneously input into a reversible neural network inverter, and perform reverse energy hiding correlation transmission by the same energy hiding method as S3And obtaining a revealing implicit variable pair. The 8 algorithm blocks in the current reversible neural network are arranged in descending order, i.e. [ y ] 8 ；y 7 ；…；y 1 ]. The revealed hidden variable pair is obtained through the processT n AndT' n 。
s8, the revealing hidden variable pair obtains a secret image and a coverage image through the image generation module.
Hiding hidden variables obtained in the processT n AndT' n input image generation moduleGIn (1), generating a secret imagex' secret And overlay imagesx' cover To complete the reverse image disclosure. The image generation module in this step shares the same functional module with the image generation module in S3.
Through the process, as shown in fig. 2 and fig. 3, the same complete method model is used for realizing image steganography and image revelation, and the parameters of each functional module of the algorithm model are adjusted through deep learning model training, so that the functions of the model are gradually optimized, and the double tasks of image steganography and image revelation are completed.
Further, in the above process, the energy generation loss function isL EG And constraining end-to-end overall adjustment and optimization of the parameters of the model, wherein the energy generation loss function comprises an energy loss function and a generation loss function, and is represented as:
L
EG
=L
energy
+L
generation
wherein the content of the first and second substances,Ɛ、μand
are the weights in the optimization process and,x cover representing the image to be steganographically overlaid,x' cover the representation of the overlay image is shown,x secret representing the secret image to be steganographically represented,x' secret a representation of the secret image is shown,Mrepresenting the total number of frames of images of each input model in the algorithm model loss iteration process,
is the firstiThe frame input is the average of the overlaid images,
is a process of disclosureiThe frame generates an average value of the overlaid image,
is the firstiThe frame input covers the variance of the image,
is a process of disclosureiThe frame generates a variance that covers the image,
is shown asiFrame input overlay image and uncovering processiThe frame generation covers the covariance between the images,
is the firstiThe average value of the frame input secret image,
is a disclosing processiThe frame generates an average value of the secret image,
is the firstiThe variance of the frame input secret image is
Disclosure of ProcessiThe frame generates the variance of the secret image,
is shown asiFrame input secret image and revealing processiThe frame generates a covariance between the secret images,c 1 andc 2 is a constant for ensuring learning fairness.
The above-mentioned embodiments, objects, technical solutions and advantages of the present application are described in further detail, it should be understood that the above-mentioned embodiments are merely exemplary embodiments of the present application, and are not intended to limit the scope of the present application, and any modifications, equivalent substitutions, improvements and the like made within the spirit and principle of the present application should be included in the scope of the present application.
Claims (7)
1. A component energy-based double-task image steganography method is characterized by comprising the following steps:
acquiring a data pair of an image to be steganographically, wherein the data pair of the image to be steganographically comprises a secret image to be steganographically and an overlay image to be steganographically;
respectively obtaining component energy pairs of the to-be-steganographic energy data pairs based on the to-be-steganographic image data pairs through a component energy extraction model, wherein the energy extraction model comprises a two-dimensional convolution layer, a full connection layer and a global average pooling layer, and the component energy of the to-be-steganographic energy data pairs comprises the component energy of the to-be-steganographic secret image and the component energy of the to-be-steganographic coverage image;
obtaining hidden variable pairs through a reversible neural network based on component energy pairs of energy data pairs to be hidden;
the hidden variable pair obtains a hidden image through an image generation module;
acquiring a noise image;
obtaining component energies of the steganographic image and the noise image respectively based on the steganographic image and the noise image through the component energy extraction model;
obtaining a revealing hidden variable pair based on the component energy of the steganographic image and the component energy of the noise image through a reversible neural network;
and the revealing hidden variable pair obtains a secret image and an overlay image through the image generation module.
2. The component energy-based multitask image steganography method according to claim 1, characterized in that when obtaining the pair of image data to be steganography, the physical dimension information of the image in the pair of image data to be steganography is obtained at the same time.
3. The method as claimed in claim 2, wherein the parameters of the component energy extraction model are self-adjusted according to the input physical scale information of the image to be steganographically.
4. The component energy-based duplex image steganography method as recited in claim 1,
the secret image to be steganographed obtains an energy latent variable set of the secret image to be steganographed through a component energy extraction modelZ n ={z 1 ，z 2 ，…，z n And (c) the step of (c) in which,z i =Enc θ (x secreti )，z i a hidden variable representing the ith secret image to be steganographically,Enc θ () Representing the encoding process of the secret image to be steganographically treated,x secreti denotes the firstiA secret image to be steganographically;
the image to be steganographically covered is subjected to component energy extraction model to obtain an energy steganography variable set of the image to be steganographically coveredZ' n ={z' 1 ，z' 2 ，…，z' n And (c) the step of (c) in which,z' i =Enc θ '(x coveri )，z' i a hidden variable representing the ith to-be-steganographically overlaid image,Enc θ '() Representing the encoding process for the to-be-steganographically overlaid image,x coveri denotes the firstiAnd covering the image to be steganographically.
5. The method according to claim 1, wherein the loss function of the method comprises an energy loss function and a generation loss function.
6. The method according to claim 5, wherein the energy loss function is expressed as:
wherein, the first and the second end of the pipe are connected with each other,
a latent variable representation of the foreground information energy representing the steganographic image,
a background information energy latent variable representation representing a steganographic image,
foreground information energy latent variable representation representing a secret image,
a background information energy latent variable representation representing the secret image.
7. The method of claim 5, wherein the generation penalty function is expressed as:
wherein the content of the first and second substances,Ɛ、μand
are the weights in the optimization process and,x cover representing the image to be steganographically overlaid,x' cover the representation of the overlay image is shown,x secret representing the secret image to be steganographically represented,x' secret a representation of the secret image is shown,Mrepresents the total number of frames of images of each input model in the algorithm model loss iterative process,
is the firstiThe frame input is the average of the overlaid images,
is a process of disclosureiThe frame generates an average value of the overlaid image,
is the firstiThe frame input covers the variance of the image,
is a process of disclosureiThe frame generates a variance that covers the image,
is shown asiFrame input overlay image and uncovering processiThe frame generation covers the covariance between the images,
is the firstiThe average value of the frame input secret image,
is a process of disclosureiThe frame generates an average value of the secret image,
is the firstiThe variance of the frame input secret image is
Disclosure of ProcessiThe frame generates the variance of the secret image,
is shown asiFrame input secret image and revealing processiThe frame generates a covariance between the secret images,c 1 andc 2 is a constant for ensuring learning fairness.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202211619646.9A CN115731089B (en) | 2022-12-16 | 2022-12-16 | Dual-task image steganography method based on component energy |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202211619646.9A CN115731089B (en) | 2022-12-16 | 2022-12-16 | Dual-task image steganography method based on component energy |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN115731089A true CN115731089A (en) | 2023-03-03 |
| CN115731089B CN115731089B (en) | 2024-02-23 |
Family
ID=85301438
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202211619646.9A Active CN115731089B (en) | 2022-12-16 | 2022-12-16 | Dual-task image steganography method based on component energy |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN115731089B (en) |
Citations (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107992466A (en) * | 2017-10-09 | 2018-05-04 | 中国电子科技集团公司第二十八研究所 | A kind of such poems of the Song Dynasty carrier text information hiding method based on clause control, system |
| CN109859091A (en) * | 2018-12-24 | 2019-06-07 | 中国人民解放军国防科技大学 | Image steganography detection method based on Gabor filtering and convolutional neural network |
| US20200184592A1 (en) * | 2017-05-19 | 2020-06-11 | Google Llc | Hiding Information and Images via Deep Learning |
| CN111598762A (en) * | 2020-04-21 | 2020-08-28 | 中山大学 | Generating type robust image steganography method |
| CN112581345A (en) * | 2020-12-04 | 2021-03-30 | 华南理工大学 | Image block-based image steganalysis method, system, device and medium |
| US20210150312A1 (en) * | 2019-11-15 | 2021-05-20 | Booz Allen Hamilton Inc. | System and method for digital image steganography detection using an ensemble of neural spatial rich models |
| CN112884630A (en) * | 2021-02-04 | 2021-06-01 | 南开大学 | Reversible neural network-based high-capacity image steganography and recovery method |
| CN114140309A (en) * | 2021-12-03 | 2022-03-04 | 中国人民武装警察部队工程大学 | Novel image steganography method and system based on NICE model |
-
2022
- 2022-12-16 CN CN202211619646.9A patent/CN115731089B/en active Active
Patent Citations (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20200184592A1 (en) * | 2017-05-19 | 2020-06-11 | Google Llc | Hiding Information and Images via Deep Learning |
| CN107992466A (en) * | 2017-10-09 | 2018-05-04 | 中国电子科技集团公司第二十八研究所 | A kind of such poems of the Song Dynasty carrier text information hiding method based on clause control, system |
| CN109859091A (en) * | 2018-12-24 | 2019-06-07 | 中国人民解放军国防科技大学 | Image steganography detection method based on Gabor filtering and convolutional neural network |
| US20210150312A1 (en) * | 2019-11-15 | 2021-05-20 | Booz Allen Hamilton Inc. | System and method for digital image steganography detection using an ensemble of neural spatial rich models |
| CN111598762A (en) * | 2020-04-21 | 2020-08-28 | 中山大学 | Generating type robust image steganography method |
| CN112581345A (en) * | 2020-12-04 | 2021-03-30 | 华南理工大学 | Image block-based image steganalysis method, system, device and medium |
| CN112884630A (en) * | 2021-02-04 | 2021-06-01 | 南开大学 | Reversible neural network-based high-capacity image steganography and recovery method |
| CN114140309A (en) * | 2021-12-03 | 2022-03-04 | 中国人民武装警察部队工程大学 | Novel image steganography method and system based on NICE model |
Non-Patent Citations (3)
| Title |
|---|
| 彭伟;胡宁;胡??;: "图像隐写分析算法研究概述", 计算机科学, no. 1 * |
| 李祁云等: "《一种基于ＦａｓｔＩＣＡ的图像隐写提取算法》", 《计算机应用》, vol. 27, no. 5 * |
| 苏日成: "《基于深度学习的构造式信息隐藏方法研究》", 《中国优秀硕士学位论文全文库》 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN115731089B (en) | 2024-02-23 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN110084734B (en) | Big data ownership protection method based on object local generation countermeasure network | |
| Emad et al. | A secure image steganography algorithm based on least significant bit and integer wavelet transform | |
| CN109587372B (en) | Invisible image steganography based on generation of countermeasure network | |
| CN111768325B (en) | Security improvement method based on generation of countermeasure sample in big data privacy protection | |
| CN111243066A (en) | Facial expression migration method based on self-supervision learning and confrontation generation mechanism | |
| CN110706302A (en) | System and method for text synthesis image | |
| EP2449524A1 (en) | Contrast enhancement | |
| CN111681154B (en) | Color image steganography distortion function design method based on generation countermeasure network | |
| CN112115490B (en) | GAN-based carrier image synthesis steganography method | |
| Peng et al. | A robust coverless steganography based on generative adversarial networks and gradient descent approximation | |
| CN112560753A (en) | Face recognition method, device and equipment based on feature fusion and storage medium | |
| CN113658040A (en) | Face super-resolution method based on prior information and attention fusion mechanism | |
| Sisaudia et al. | Copyright protection using KELM-PSO based multi-spectral image watermarking in DCT domain with local texture information based selection | |
| CN115758422A (en) | File encryption method and system | |
| CN115115500A (en) | Watermark embedding method combined with underwater image enhancement | |
| US20240104681A1 (en) | Image steganography utilizing adversarial perturbations | |
| CN111726472A (en) | Image anti-interference method based on encryption algorithm | |
| CN117079083B (en) | Image restoration model training method and device, electronic equipment and storage medium | |
| CN115731089A (en) | Component energy-based double-task image steganography method | |
| Rana et al. | Image steganography method based on kohonen neural network | |
| CN115879516A (en) | Data evidence obtaining method | |
| Aberna et al. | Optimal Semi-Fragile Watermarking based on Maximum Entropy Random Walk and Swin Transformer for Tamper Localization | |
| Dhawan et al. | High‐quality steganography scheme using hybrid edge detector and Vernam algorithm based on hybrid fuzzy neural network | |
| Liu et al. | Dual-channel and two-stage dehazing network for promoting ship detection in visual perception system | |
| CN114494387A (en) | Data set network generation model and fog map generation method |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |