CN114897658A - Image steganography model protection method based on extraction network parameter scrambling - Google Patents
Image steganography model protection method based on extraction network parameter scrambling Download PDFInfo
- Publication number
- CN114897658A CN114897658A CN202210458370.4A CN202210458370A CN114897658A CN 114897658 A CN114897658 A CN 114897658A CN 202210458370 A CN202210458370 A CN 202210458370A CN 114897658 A CN114897658 A CN 114897658A
- Authority
- CN
- China
- Prior art keywords
- scrambling
- image
- network model
- extraction network
- secret
- 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
- 238000000605 extraction Methods 0.000 title claims abstract description 86
- 238000000034 method Methods 0.000 title claims abstract description 56
- 239000000284 extract Substances 0.000 claims abstract description 9
- 239000010410 layer Substances 0.000 claims description 131
- 230000000739 chaotic effect Effects 0.000 claims description 22
- 239000011229 interlayer Substances 0.000 claims description 22
- 238000005070 sampling Methods 0.000 claims description 21
- 238000004458 analytical method Methods 0.000 claims description 8
- 238000011176 pooling Methods 0.000 claims description 6
- 230000001537 neural effect Effects 0.000 claims description 4
- 238000013135 deep learning Methods 0.000 claims description 3
- 230000005540 biological transmission Effects 0.000 abstract description 8
- 230000008569 process Effects 0.000 abstract description 7
- 230000002708 enhancing effect Effects 0.000 abstract description 3
- 230000000694 effects Effects 0.000 description 11
- 238000010586 diagram Methods 0.000 description 7
- 230000006854 communication Effects 0.000 description 6
- 238000005516 engineering process Methods 0.000 description 6
- 238000004891 communication Methods 0.000 description 5
- 230000003044 adaptive effect Effects 0.000 description 3
- 239000000969 carrier Substances 0.000 description 3
- 230000006870 function Effects 0.000 description 3
- 230000004913 activation Effects 0.000 description 2
- 238000012360 testing method Methods 0.000 description 2
- 238000012549 training Methods 0.000 description 2
- 230000000007 visual effect Effects 0.000 description 2
- 230000005856 abnormality Effects 0.000 description 1
- 230000008859 change Effects 0.000 description 1
- 238000013527 convolutional neural network Methods 0.000 description 1
- 230000006872 improvement Effects 0.000 description 1
- 238000013507 mapping Methods 0.000 description 1
- 239000011159 matrix material Substances 0.000 description 1
- 238000005259 measurement Methods 0.000 description 1
- 238000005312 nonlinear dynamic Methods 0.000 description 1
- 238000010606 normalization Methods 0.000 description 1
- 230000008447 perception Effects 0.000 description 1
- 238000007781 pre-processing Methods 0.000 description 1
- 238000002360 preparation method Methods 0.000 description 1
- 239000002356 single layer Substances 0.000 description 1
- 238000012795 verification Methods 0.000 description 1
- 230000003245 working effect Effects 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T1/00—General purpose image data processing
- G06T1/0021—Image watermarking
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T9/00—Image coding
- G06T9/002—Image coding using neural networks
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Evolutionary Computation (AREA)
- Artificial Intelligence (AREA)
- General Health & Medical Sciences (AREA)
- General Engineering & Computer Science (AREA)
- Data Mining & Analysis (AREA)
- Biophysics (AREA)
- Biomedical Technology (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- Computational Linguistics (AREA)
- Life Sciences & Earth Sciences (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Health & Medical Sciences (AREA)
- Multimedia (AREA)
- Computer And Data Communications (AREA)
- Data Exchanges In Wide-Area Networks (AREA)
Abstract
The invention provides an image steganography model protection method based on extraction network parameter scrambling, and belongs to the technical field of image encryption. Since the sender may be stolen by a third party when sending the extracting network to the receiver for the first time, the secret image of the subsequent transmission is leaked. For this, the sender scrambles the parameters of the extracted network model by a scrambling algorithm agreed with the receiver, and then reconstructs a new extracted network model by extracting the parameters of the network model after scrambling. When the sender sends the new extraction network and the secret-carrying image to the receiver, even if the extraction network model is acquired by a third party in the transmission process, the similarity between the secret image extracted from the secret-carrying image and the original secret image is low, so that the secret image is prevented from being leaked. And the receiver can restore the new extraction network model to the extraction network model before scrambling according to the agreed scrambling algorithm, and extract the secret image from the secret image, thereby enhancing the safety of the image steganography process.
Description
Technical Field
The invention relates to an image steganography model protection method based on extraction network parameter scrambling, and belongs to the technical field of image encryption.
Background
In daily work, people can not transmit information without a computer, because the computer can meet the daily life and work needs of people. In order to ensure that secret information can be safely transmitted in a network, people usually select an encryption technology to protect the information, such as file encryption and picture encryption, and the encrypted file has an obvious disadvantage that the encrypted file cannot be normally opened, so that the encrypted file is easily doubted by others in the transmission process. The information hiding can well solve the problems encountered above, and the information hiding refers to hiding secret information to be transmitted in a media which can be disclosed, and people can hardly find the existence of the secret information through intuitive hearing and vision. There are many carriers used for hiding information, such as images, sound, video, documents, etc.
The digital image is one of the most common carriers for information hiding, and the digital image is selected as the carrier mainly for the following 3 reasons: (1) under the background of the information technology era, a large amount of picture information exists in the internet, and people can easily obtain a carrier; (2) the semantic content contained in the image is very rich, and different information can be expressed; (3) digital images have a larger storage capacity and a larger data redundancy than text and audio, and are well suited as carriers for information hiding, so that image steganography has emerged. The image steganography technique refers to: the secret information to be transmitted is embedded into the carrier image by a sender through a certain technical means, then the secret image containing the secret image is sent to a receiver through an open network channel, and the secret information is extracted from the carrier image by a decryption technology after the receiver receives the secret image, so that the perception of a third party is avoided.
The image steganography goes through roughly 3 stages. In phase 1, researchers implement image steganography by modifying pixel values of the image. Such as LSB steganography, the purpose of information hiding is achieved by modifying or replacing the least significant bits. The method cannot cause great influence on the overall visual effect of the image, but the method is easy to detect by a steganalysis technology, and the LSB steganalysis technology is difficult to meet the requirement of modern communication safety. In the 2 nd stage, researchers consider that the image has high complexity and more redundancies, so that the content of the image is modified by a content adaptive steganography method to realize information hiding, people are difficult to visually perceive, and abnormality is difficult to detect by a steganography analysis technical means. Common adaptive steganography methods include HUGO, WOW, UNIWARD and HILL, but with further improvement of steganography analysis technology, the SRM can analyze and detect the adaptive steganography. In the 3 rd stage, with the rise of Deep Convolutional Neural Networks (DCNN) in various fields, researchers introduce DCNN into the image steganography field to achieve better image steganography effect. For example, the encoder-decoder image steganography framework proposed by Baluja et al includes an image preprocessing stage, a hidden network and an extraction network, which can ensure that a secret-carrying image and a carrier image are consistent, and the extracted secret image and an original secret image are virtually undistorted and have certain steganography analysis resistance.
The image steganography network model based on deep learning mainly comprises a hidden network and an extraction network, firstly, a sender selects a picture data set of the sender to train so as to obtain the image steganography network model, and then the sender and a receiver carry out covert communication through the trained steganography network model. When a sender and a receiver communicate for the first time, the sender needs to send the trained parameters of the extracted network model to the receiver, so that preparation work of covert communication is established, the receiver can establish the extracted network model only by taking the parameters of the extracted network model, and further secret information hidden in a secret-carrying image is extracted through the extracted network model. However, when the two parties transmit and extract the parameters of the network model for the first time, certain leakage risks exist, the safety degree is low, and if the parameters are intercepted by the third party in the transmission process, the subsequent communication contents can be obtained by the third party through the extraction network.
Disclosure of Invention
The invention aims to provide an image steganography model protection method based on extraction network parameter scrambling, which is used for solving the problem that in the image steganography method, the safety degree is low when a sender sends an extraction network model in an image steganography network to a receiver.
In order to achieve the above object, the present invention provides an image steganography model protection method based on extraction network parameter scrambling, which comprises the following steps:
s1, hiding the secret image to be sent into a carrier image by a sender through the trained hidden network model to obtain a corresponding secret-carrying image, and obtaining a trained extraction network model, wherein the hidden network model and the extraction network model jointly form an image steganography network for deep learning;
s2, the sender scrambles the trained parameters of the extraction network model through an agreed scrambling algorithm, and reconstructs the parameters according to the scrambled parameters of the extraction network model to obtain a new extraction network model; the sender sends the secret-carrying image and the new extraction network model to the receiver; the promised scrambling algorithm satisfies the following conditions: for the same secret-carrying image, the similarity of the secret image extracted by the extraction network model before scrambling and the secret image extracted by the extraction network model after scrambling is lower than a set threshold value;
and S3, after receiving the new extracted network model, the receiver scrambles and decrypts the new extracted network model according to the agreed scrambling algorithm to obtain the extracted network model before scrambling, and extracts the received secret-carrying image according to the extracted network model before scrambling to obtain the secret image.
Since the sender may be stolen by a third party when sending the extracting network to the receiver for the first time, the secret image of the subsequent transmission is leaked. For this, the sender scrambles the parameters of the extracted network model by a scrambling algorithm agreed with the receiver, and then reconstructs a new extracted network model by extracting the parameters of the network model after scrambling. When the sender sends the new extraction network and the secret-carrying image to the receiver, even if the extraction network model is obtained by a third party in the transmission process, the secret image extracted from the secret-carrying image has a larger difference with the original secret image and has lower similarity, thereby avoiding the leakage of the secret image. And the receiver can restore the new extraction network model to the extraction network model before scrambling according to the agreed scrambling algorithm, and extract the secret-carrying image to obtain an effective secret image, thereby enhancing the safety of the image steganography process.
Further, in the method, the hidden network model and the extracted network model both adopt a deep neural convolutional network, and the scrambling operation refers to performing in-layer scrambling on the positions of convolutional kernels in one or more convolutional layers in the trained extracted network model, or performing interlayer scrambling on the positions of convolutional kernels in different convolutional layers; the position of the convolution kernel is changed only in the convolution layer when the layers are scrambled, and the position of the convolution kernel is changed in different convolution layers when the layers are scrambled.
The extraction network model employs a deep neural convolutional network, which includes a plurality of convolutional layers. When scrambling operation is carried out, carrying out in-layer scrambling or interlayer scrambling on the trained extraction network model, wherein the method for in-layer scrambling comprises the following steps: changing the position of a convolution kernel in the convolution layer; the interlayer scrambling method comprises the following steps: the positions of the convolution kernels of the different convolution layers are changed. By changing the position of the convolution layer, the convolution operation sequence after the secret-carrying image is input into the extraction network model is changed, thereby realizing the encryption of the extraction network model and enhancing the safety of the steganography process of the image.
Further, in the above method, the agreed scrambling algorithm employs a hyper Lorenz chaotic system.
Aiming at the appointed scrambling algorithm, a specific implementation method is provided, and the sender and the receiver realize the appointment of the scrambling algorithm through mutually appointing the control parameters of the hyper Lorenz chaotic system and the scrambling operation process.
Further, in the above method, the similarity is determined by using a structural similarity analysis method SSIM.
For the same secret-carrying image, the similarity of the secret image extracted by the extraction network model before scrambling and the secret image extracted by the extraction network model after scrambling needs to be analyzed, and a specific method is provided for determining the similarity, so that the method is convenient to implement.
Further, in the above method, the extraction network model includes an upsampling module, a pooling layer, a downsampling module, and a convolutional layer, which are connected to each other; the upsampling module and the downsampling module each include 4 convolutional layers.
A specific abstraction network model is provided to facilitate the implementation of the present invention.
Further, in the above method, when the scrambling operation is performed in step S2, the 1 st, 2 nd, or 4 th convolutional layer in the up-sampling module is subjected to the intra-layer scrambling, or the 3 rd or 4 th convolutional layer of the down-sampling module is subjected to the intra-layer scrambling.
For the above-mentioned extraction network model, a new extraction network model with a good effect can be reconstructed by performing in-layer scrambling on the 1 st, 2 nd or 4 th convolution layer in the up-sampling module or performing in-layer scrambling on the 3 rd or 4 th convolution layer in the down-sampling module, and the operation is simple.
Further, in the above method, when the scrambling operation is performed in step S2, the 1 st, 2 nd, and 4 th convolutional layers in the up-sampling block, and the 3 rd and 4 th convolutional layers in the down-sampling block are subjected to interlayer scrambling.
For the above-mentioned extraction network model, a new extraction network model with good effect can be reconstructed by performing interlayer scrambling on the 1 st, 2 nd and 4 th convolution layers in the up-sampling module and the 3 rd and 4 th convolution layers in the down-sampling module, so that the implementation of the invention is facilitated, and the safety degree is high.
Further, in the above method, when the scrambling operation is performed in step S2, the inter-layer scrambling is performed on the 4 convolutional layers of the upsampling module.
For the above-mentioned extraction network model, a new extraction network model with a good effect can be reconstructed by performing interlayer scrambling on the 4 convolution layers of the up-sampling module, so that the implementation of the invention is facilitated, and the safety degree is high.
Further, in the above method, when the scrambling operation is performed in step S2, the inter-layer scrambling is performed on the 4 convolution layers of the downsampling block.
For the extracted network model, a new extracted network model with good effect can be reconstructed by carrying out interlayer scrambling on 4 convolution layers of the down-sampling module, so that the implementation of the invention is facilitated, and the safety degree is high.
Further, in the above method, when the scrambling operation is performed in step S2, the inter-layer scrambling is performed on the 9 convolutional layers in the extracted network model.
For the extracted network model, a new extracted network model with good effect can be reconstructed by carrying out interlayer scrambling on 9 convolutional layers in the extracted network model, so that the implementation of the method is facilitated, and the safety degree is high.
Drawings
FIG. 1 is a block diagram of a flow of an image steganography model protection method based on extraction network parameter scrambling in an embodiment of the method of the present invention;
FIG. 2 is a schematic diagram of the working process of the image steganography network model in the embodiment of the method of the present invention;
FIG. 3 is a schematic diagram illustrating the working effect of the steganographic network model in the embodiment of the method of the present invention;
FIG. 4 is a schematic diagram illustrating the effect of extracting a single-layer convolutional layer scrambling scheme in a network according to an embodiment of the present invention;
FIG. 5 is a schematic diagram illustrating the effect of extracting the multilayer convolutional layer scrambling scheme in the network according to the embodiment of the present invention;
FIG. 6 is a diagram illustrating the effect of scrambling the convolution kernel positions in the last convolutional layer according to an embodiment of the present invention.
Detailed Description
In order to make the objects, technical solutions and advantages of the present invention more apparent, the present invention is further described in detail below with reference to the accompanying drawings and embodiments.
The method comprises the following steps:
the extraction network model is scrambled through the chaotic system encryption algorithm, even if the extraction network model is obtained by a third party in the transmission process, a useful secret image cannot be extracted from the secret-carrying image through the extraction network model, and the receiver can restore the scrambled extraction network model into a correct extraction network according to the pre-defined scrambling algorithm, so that the safety and the reliability of the communication process of the sender and the receiver are improved.
As shown in FIG. 2, the present invention utilizes the image steganography model framework of the deep neural convolutional network and the concepts of the encoder and decoder to construct the image steganography network model. The image steganography Network model comprises a hidden Network model (mapping Network) and an Extraction Network model (Extraction Network), wherein the hidden Network model connects a Secret image (Secret image) and a carrier image (Cover image) into a 6-channel tensor by using a coding function, then outputs a 3-channel Secret image (Container image), and the Extraction Network model can extract Secret information in the Secret image so as to recover the Secret image (Recovered image) in the Secret image.
As shown in fig. 1, the image steganography model protection method based on extraction network parameter scrambling of the present invention includes the following steps:
1. and constructing a hidden network model and extracting a network model.
The hidden network model and the extracted network model are structurally as shown in the following table 1, the extracted network model mainly comprises 3 parts of content, namely an up-sampling Module, a Pooling Module (Pooling Module), a down-sampling Module and a convolution Module, the up-sampling Module comprises 4 layers of convolution layers Conv and BN (batch normalization) which are connected with each other, the Pooling Module comprises 1 layer of Pooling layer, the down-sampling Module comprises 4 layers of deconvolution layers Deconv and BN layers which are connected with each other, and the convolution Module comprises 1 layer of convolution layers Conv and BN layers which are connected with each other. The up-sampling module and the down-sampling module adopt an activation function ReLU, and the convolution module adopts an activation function Sigmoid.
TABLE 1 hidden network model and extract network model internal structure table
As can be seen from table 1, the number of convolution kernels in the first 4 convolution layers in the extracted network model is 32, 64, 128 and 256, respectively, and the sizes of the convolution kernels are 3 × 3, 4 × 4 and 4 × 4, respectively; the number of convolution kernels of the last 4 deconvolution layers is 128, 64, 32 and 16 respectively, and the sizes of the convolution kernels are 4 × 4, 4 × 4 and 3 × 3 respectively; the number of convolution kernels of the last 1 convolution layer is 3, and the size of the convolution kernels is 3 x 3. The first 4 layers of convolutional layers sequentially perform convolution operation on the input secret image (Container image), the secret image is converted into a feature map of 256 channels from 3 channels, the last 4 layers of deconvolution layers perform deconvolution operation, and finally 1 convolutional layer generates a secret image of 3 channels.
100000 pictures are randomly selected from the ImageNet data set, wherein 98000 pictures are used as a training set, 1000 pictures are used as a verification set, and the rest 1000 pictures are used as a test set. And obtaining a steganographic network model of the image steganographic network model and an extraction network model through 200 epochs training and testing.
As shown in fig. 3, the first line in the figure is the carrier image, the second line is the secret image after hiding the secret image, the third line is the secret image that the sender needs to send to the receiver, and the fourth line is the secret image that the receiver extracts from the secret image through the extraction network model. As can be seen from the content shown in fig. 3, there is almost no difference between the carrier image and the secret image in terms of visual effect, and no loss is observed between the extracted secret image and the original hidden secret image, which indicates that the trained image steganography network model has small error loss and good hiding effect and extraction effect.
2. And the sender scrambles the trained parameters of the extracted network model through an agreed scrambling algorithm, reconstructs the parameters according to the scrambled parameters of the extracted network model to obtain a new extracted network model, and then sends the secret-carrying image and the new extracted network model to the receiver.
The specific method comprises the following steps: dividing parameters of the extracted network model into 9 parameter files according to the number of the convolutional layers and the anti-convolutional layers in the extracted network model, regarding a convolutional kernel as a minimum operation unit in each parameter file, scrambling and encrypting the positions of the convolutional kernels in each convolutional layer and the anti-convolutional layer by using a hyper-Lorenz chaotic system scrambling algorithm in an MATLAB (matrix laboratory), thereby obtaining a scrambled parameter file, and reconstructing a new extracted network model by using the scrambled parameter file.
The hyper-Lorenz chaotic system has bright nonlinear dynamics characteristics and is sensitive to initial values, so that the hyper-Lorenz chaotic system is widely applied to digital image encryption. In general, a shuffle system scrambles image pixel positions to encrypt an image, but this method is similar to file encryption, and when the encrypted image is transmitted through a public channel, suspicion is easily caused, so that a third party may prevent communication between the two parties, and a receiving party may not receive information of a transmitting party at all.
The system model of the scrambling algorithm of the hyper-Lorenz chaotic system is as follows:
in the formula, x, y, z and w are state variables of the hyper Lorenz chaotic system, and a, b, c and r are control parameters of the hyper Lorenz chaotic system. When the control parameter a of the chaotic system is 10, b is 8/3, c is 28, r is more than-1.52 and less than-0.06, the system is in a hyperchaotic state. A hyper-chaotic state is a chaotic, unpredictable, chaotic state.
The state variables x, y, z and w of the hyper-Lorenz chaotic system are secret keys for scrambling encryption, so that a one-dimensional chaotic sequence is generated. Therefore, the scrambling encryption process is: firstly, a one-dimensional chaotic sequence is generated through a hyper-Lorenz chaotic system, after a convolution kernel of each layer is loaded to the hyper-Lorenz chaotic system, scrambling operation is carried out on the position of the convolution kernel in a parameter file through the one-dimensional chaotic sequence, so that a scrambled parameter file is obtained, and a new extraction network model is reconstructed by utilizing the scrambled parameter file. The method for scrambling the positions of the convolution kernels comprises the following two methods:
first, layer scrambling. Scrambling the positions of the convolution kernels in each convolution layer, namely: the positions of the convolution kernels in each convolution layer and each deconvolution layer are randomly arranged without repetition. For example, when the positions of 32 convolution kernels in the 1 st convolutional layer are disturbed and a convolution operation is performed on the 1 st convolutional layer after a secret image is input, the output 32-channel feature map is different from the 32-channel feature map output by the original 1 st convolutional layer. Similarly, if the positions of convolution kernels in other convolution layers and in the deconvolution layer are disturbed, the output characteristic diagrams are different when convolution operation and deconvolution operation are performed.
As shown in fig. 4, the first graph from left to right in the drawing is a secret image extracted by an original extraction network model (i.e., an extraction network model before scrambling encryption), the second graph to the fifth graph from left to right in the first row are secret images extracted by an extraction network obtained by performing in-layer scrambling on the first 4 convolutional layers, respectively, and the five graphs from left to right in the second row are secret images extracted by an extraction network obtained by performing in-layer scrambling on the last 4 deconvolution layers and the last convolutional layer, respectively.
As can be seen from fig. 4, the secret images extracted by the extraction network model obtained by performing the intra-layer scrambling on the 1 st, 2 nd, 4 th, 7 th and 8 th convolutional layers are greatly changed. For this purpose, a structural Similarity analysis method SSIM (structural Similarity Index measurement) is introduced to analyze the Similarity between images. The similarity between the secret image extracted by the extraction network model obtained by performing the intra-layer scrambling on the 1 st, 2 nd, 4 th, 7 th and 8 th convolution layers and the original secret image (namely, the secret image extracted by the original extraction network) is lower than a set threshold k, and the similarity between the secret image extracted by the extraction network model obtained by performing the intra-layer scrambling on the 3 rd, 5 th, 6 th and 9 th convolution layers and the original secret image is higher than the set threshold k. Therefore, when the transmission side transmits to the reception side the extracted network model obtained by performing the layer scrambling on the 1 st, 2 nd, 4 th, 7 th or 8 th convolution layer, even if the third side acquires the extracted network model, it is difficult to acquire a perfect secret image from the secret image, and these extracted network models can be used as the extracted network to be transmitted to the reception side. In this embodiment, the threshold k is set to 0.1.
And secondly, carrying out interlayer scrambling. Scrambling the positions of the convolution kernels in different convolution layers, namely: and randomly selecting a plurality of convolution layers, and randomly arranging the positions of convolution kernels in each convolution layer without repetition, wherein the position change of the convolution kernels not only occurs in the layers, but also changes to other convolution layers or deconvolution layers.
In this embodiment, 4 secret images extracted by the extraction network model after interlayer scrambling are provided for display, which are respectively: 1) selecting convolution kernels of the 1 st, 2 nd, 4 th, 7 th and 8 th convolution layers for interlayer scrambling; 2) selecting the first 4 convolutional layers for interlayer scrambling; 3) selecting the last 4 deconvolution layers to carry out interlayer scrambling; 4) all convolutional layers and deconvolution layers are selected to be subjected to interlayer scrambling.
As shown in fig. 5, the first image is an original secret image, and the second to 5 th images are secret images extracted from the secret image by the extraction network model obtained by the above-described 4-layer interlamination. And similarly, a structural similarity analysis method SSIM is adopted for similarity analysis, and the similarity between each secret image and the original secret image is smaller than a set threshold k. Therefore, when the transmitting side transmits the extracted network model obtained by the 4-layer interspersing to the receiving side, even if the third side acquires the extracted network model, it is difficult to acquire a perfect secret image from the secret image, and these extracted network models can be used as the extracted network model to be transmitted to the receiving side.
Furthermore, as can be seen from the last image in fig. 4, when the last 1 convolutional layer is subjected to intra-layer scrambling, the obtained extraction network model may only affect the color of the image when extracting the secret image. For this reason, the last 1 convolutional layer is subjected to intra-layer scrambling separately, the last 1 convolutional layer has only 3 convolutional kernels, the initial position is 123, and the positions of the convolutional kernels after intra-layer scrambling are only 5 cases: 132. 213, 231, 312 and 321. As shown in fig. 6, since the positions of the convolution kernels in the last convolution layer are individually subjected to the in-layer scrambling, only the color of the secret image extracted by the extraction network model is different from that of the original secret image, and the information in the secret image is still exposed, the extraction network model obtained by individually subjecting the positions of the convolution kernels in the last convolution layer to the in-layer scrambling is not adopted.
The sender sends control parameters a of the hyper-chaotic scrambling system to be 10, b to be 8/3, c to be 28, r to be 1.52 and r to be less than or equal to-0.06 and state variables to the receiver, and scrambles which convolutional layers or deconvolution layers in a mode of scrambling algorithm appointed with the receiver, thereby realizing advanced appointment of the scrambling algorithm.
As another embodiment, a simpler scrambling method may be adopted by the agreement, and in this case, the sender and the receiver do not need to agree on the control parameters in advance. For example, for layer scrambling, the positions of convolution kernels in a certain convolutional layer are exchanged according to an agreed rule, such as head-to-tail exchange or symmetric exchange about the middle position of the convolution kernels; for the interlayer scrambling, the positions of convolution kernels at the same position in some two layers of convolution layers are exchanged.
In order to improve the security of the subsequent image steganography process, the sender can also periodically re-agree with the receiver on the scrambling algorithm.
3. After receiving the extracted network model, the receiver reconstructs the chaotic scrambling system model according to the control parameter a of the chaotic scrambling system which is sent by the sender in advance, b is 8/3, c is 28, r is more than-1.52 and less than or equal to-0.06, then uses the received state variable as a secret key, uses the received extracted network as an object to be decrypted, inputs the state variable and the extracted network into the reconstructed chaotic scrambling system model, scrambles and decrypts the extracted network model through an agreed scrambling algorithm, and restores the extracted network model to the original extracted network model. Therefore, the receiving side can extract the secret image in which the secret image is hidden by using the original extraction network model, thereby obtaining the secret image.
By adopting the method and the device, the extracted network model in the image steganography network model can be encrypted, and the security of the image steganography is further improved.
Claims (10)
1. An image steganography model protection method based on extraction network parameter scrambling is characterized by comprising the following steps:
s1, hiding the secret image to be sent into a carrier image by a sender through the trained hidden network model to obtain a corresponding secret-carrying image, and obtaining a trained extraction network model, wherein the hidden network model and the extraction network model jointly form an image steganography network for deep learning;
s2, the sender scrambles the trained parameters of the extraction network model through an agreed scrambling algorithm, and reconstructs the parameters according to the scrambled parameters of the extraction network model to obtain a new extraction network model; the sender sends the secret-carrying image and the new extraction network model to the receiver; the promised scrambling algorithm satisfies the following conditions: for the same secret-carrying image, the similarity of the secret image extracted by the extraction network model before scrambling and the secret image extracted by the extraction network model after scrambling is lower than a set threshold value;
and S3, after receiving the new extracted network model, the receiver scrambles and decrypts the new extracted network model according to the agreed scrambling algorithm to obtain the extracted network model before scrambling, and extracts the received secret-carrying image according to the extracted network model before scrambling to obtain the secret image.
2. The image steganography model protection method based on extraction network parameter scrambling of claim 1, wherein the steganography network model and the extraction network model both adopt a deep neural convolution network, and the scrambling operation refers to performing in-layer scrambling on positions of convolution kernels in one or more convolutional layers in the trained extraction network model or performing interlayer scrambling on positions of convolution kernels in different convolutional layers; the position of the convolution kernel is changed only in the convolution layer when the layers are scrambled, and the position of the convolution kernel is changed in different convolution layers when the layers are scrambled.
3. The image steganography model protection method based on extraction network parameter scrambling of claim 2, wherein the agreed scrambling algorithm employs a hyper-Lorenz chaotic system.
4. The image steganography model protection method based on extraction network parameter scrambling as claimed in claim 1, wherein the similarity is determined by using structural similarity analysis method SSIM.
5. The image steganography model protection method based on extraction network parameter scrambling of claim 2, wherein the extraction network model comprises an upsampling module, a pooling layer, a downsampling module, and a convolution layer connected to each other; the upsampling module and the downsampling module each include 4 convolutional layers.
6. The image steganography model protection method based on extraction network parameter scrambling of claim 5, wherein in the scrambling operation in step S2, the 1 st, 2 nd or 4 th convolution layer of the up-sampling module is subjected to intra-layer scrambling, or the 3 rd or 4 th convolution layer of the down-sampling module is subjected to intra-layer scrambling.
7. The image steganography model protection method based on extraction network parameter scrambling of claim 5, wherein in the scrambling operation of step S2, the 1 st, 2 nd and 4 th convolution layers of the up-sampling module and the 3 rd and 4 th convolution layers of the down-sampling module are subjected to interlayer scrambling.
8. The image steganography model protection method based on extraction network parameter scrambling of claim 5, wherein in the scrambling operation in step S2, the 4 convolution layers of the upsampling module are subjected to interlayer scrambling.
9. The image steganography model protection method based on extracted network parameter scrambling as claimed in claim 5, wherein in step S2, when the scrambling operation is performed, the inter-layer scrambling is performed on 4 convolution layers of the down-sampling module.
10. The method for protecting an image steganography model based on extracted network parameter scrambling of claim 5, wherein in the scrambling operation in step S2, the layers of 9 convolution layers in the extracted network model are scrambled.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202210458370.4A CN114897658B (en) | 2022-04-27 | 2022-04-27 | Image steganography model protection method based on extracted network parameter scrambling |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202210458370.4A CN114897658B (en) | 2022-04-27 | 2022-04-27 | Image steganography model protection method based on extracted network parameter scrambling |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN114897658A true CN114897658A (en) | 2022-08-12 |
| CN114897658B CN114897658B (en) | 2024-08-16 |
Family
ID=82718787
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202210458370.4A Active CN114897658B (en) | 2022-04-27 | 2022-04-27 | Image steganography model protection method based on extracted network parameter scrambling |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN114897658B (en) |
Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO2016165082A1 (en) * | 2015-04-15 | 2016-10-20 | 中国科学院自动化研究所 | Image stego-detection method based on deep learning |
| CN112926607A (en) * | 2021-04-28 | 2021-06-08 | 河南大学 | Two-branch network image steganography framework and method based on convolutional neural network |
| CN113034336A (en) * | 2021-03-29 | 2021-06-25 | 大连海事大学 | Encryption and steganography dual-image protection method based on 2D-TSCC chaotic system |
| CN113298689A (en) * | 2021-06-22 | 2021-08-24 | 河南师范大学 | Large-capacity image steganography method |
| CN114049246A (en) * | 2021-11-10 | 2022-02-15 | 长春理工大学 | Information hiding method and system based on wavelet transformation |
-
2022
- 2022-04-27 CN CN202210458370.4A patent/CN114897658B/en active Active
Patent Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO2016165082A1 (en) * | 2015-04-15 | 2016-10-20 | 中国科学院自动化研究所 | Image stego-detection method based on deep learning |
| CN113034336A (en) * | 2021-03-29 | 2021-06-25 | 大连海事大学 | Encryption and steganography dual-image protection method based on 2D-TSCC chaotic system |
| CN112926607A (en) * | 2021-04-28 | 2021-06-08 | 河南大学 | Two-branch network image steganography framework and method based on convolutional neural network |
| CN113298689A (en) * | 2021-06-22 | 2021-08-24 | 河南师范大学 | Large-capacity image steganography method |
| CN114049246A (en) * | 2021-11-10 | 2022-02-15 | 长春理工大学 | Information hiding method and system based on wavelet transformation |
Non-Patent Citations (1)
| Title |
|---|
| 付章杰;王帆;孙星明;王彦;: "基于深度学习的图像隐写方法研究", 计算机学报, no. 09, 15 September 2020 (2020-09-15) * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN114897658B (en) | 2024-08-16 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Muhammad et al. | Image steganography using uncorrelated color space and its application for security of visual contents in online social networks | |
| Liu et al. | Reversible data-hiding in encrypted images by redundant space transfer | |
| Khan et al. | IMG‐forensics: Multimedia‐enabled information hiding investigation using convolutional neural network | |
| KR20230036100A (en) | Image distribution using synthetic re-encrypted images | |
| Kanzariya Nitin et al. | Comparison of various images steganography techniques | |
| Weng et al. | High capacity reversible data hiding in encrypted images using SIBRW and GCC | |
| Kolakalur et al. | Wavelet based color video steganography | |
| Wang et al. | A survey on digital image steganography | |
| Tan et al. | A survey on the new development of medical image security algorithms | |
| Abdullah et al. | Secure data transfer over internet using image steganography | |
| Wang et al. | TPE-ISE: approximate thumbnail preserving encryption based on multilevel DWT information self-embedding | |
| Li et al. | Fidelity preserved data hiding in encrypted highly autocorrelated data based on homomorphism and compressive sensing | |
| Duan et al. | DUIANet: A double layer U-Net image hiding method based on improved Inception module and attention mechanism | |
| Ch et al. | Robust steganographic framework for securing sensitive healthcare data of telemedicine using convolutional neural network | |
| Rizzi et al. | Digital watermarking for healthcare: a survey of ECG watermarking methods in telemedicine | |
| CN115632780B (en) | Use management system and method for seal of Internet of things | |
| CN114897658A (en) | Image steganography model protection method based on extraction network parameter scrambling | |
| CN116055648A (en) | Self-adaptive image steganography sending and receiving method | |
| Kich et al. | Image steganography scheme using dilated convolutional network | |
| Kaur et al. | XOR-EDGE based video steganography and testing against chi-square steganalysis | |
| Viji et al. | LSB Steganography in Color and Grayscale Images without using the Transformation | |
| CN113132550B (en) | Color image covert communication system based on FPGA | |
| Abd Aziz et al. | Image steganography:: Comparative analysis of their techniques, complexity and enhancements | |
| Zou et al. | A meaningful image encryption algorithm based on prediction error and wavelet transform | |
| Anumol et al. | Enhanced security in medical image steganography-a hybrid approach using spatial and transform domain |
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 |