CN113298689A - Large-capacity image steganography method - Google Patents
Large-capacity image steganography method Download PDFInfo
- Publication number
- CN113298689A CN113298689A CN202110693663.6A CN202110693663A CN113298689A CN 113298689 A CN113298689 A CN 113298689A CN 202110693663 A CN202110693663 A CN 202110693663A CN 113298689 A CN113298689 A CN 113298689A
- Authority
- CN
- China
- Prior art keywords
- image
- network
- secret
- layers
- output
- 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
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
- G06N3/084—Backpropagation, e.g. using gradient descent
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T5/00—Image enhancement or restoration
- G06T5/50—Image enhancement or restoration by the use of more than one image, e.g. averaging, subtraction
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Biophysics (AREA)
- Evolutionary Computation (AREA)
- Artificial Intelligence (AREA)
- Biomedical Technology (AREA)
- Health & Medical Sciences (AREA)
- Computational Linguistics (AREA)
- Data Mining & Analysis (AREA)
- Life Sciences & Earth Sciences (AREA)
- General Health & Medical Sciences (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- General Engineering & Computer Science (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Image Processing (AREA)
Abstract
The invention belongs to the technical field of image steganography, and particularly relates to a high-capacity image steganography method. Inputting the carrier image and the two secret images into a trained image steganography model to obtain a carrier image and two extracted images; the image steganography model comprises a hiding stage and an extracting stage; the hiding stage comprises a carrier image preprocessing network, a secret image preprocessing network and a hiding network; the hidden network is used for processing the output of the cascaded carrier image preprocessing network and the output of the secret image preprocessing network to obtain a carrier image containing two secret images. The method can embed the two secret images into the carrier image to obtain the carrier secret image, so that the visual effect of the carrier secret image is consistent with that of the carrier image, the steganography capacity is obviously improved, and the imperceptibility and the anti-detection performance of the carrier secret image are maintained; moreover, the invention can also extract two extracted images, and the two extracted images and the two secret images are basically consistent in visual effect.
Description
Technical Field
The invention belongs to the technical field of image steganography, and particularly relates to a high-capacity image steganography method.
Background
With the rapid development of communication technology in recent years, information security issues accompanying communication technology have attracted attention of individuals, enterprises, governments, and related research fields. In the communication process of some individuals or organizations, the confidential information transmitted by the individuals or organizations is not desired to be intercepted by a third party or even sensed by the third party to be existed, and meanwhile, the holders are desired to protect the copyright owned by the digital media works generated in the modern network. The existing cryptography technology cannot be well met aiming at the requirements. Steganography (Steganography), which is an information hiding technique that pursues imperceptibility, embeds confidential information into the carrier while maintaining its original perceptual properties, and also hides the communication behavior itself, makes Steganography well suited to meet the above requirements. Nowadays, digital images on the internet are seen everywhere, and due to the characteristics of low acquisition difficulty and large redundancy, the digital images are very suitable to be used as carriers of Steganography, so that Image Steganography (Image Steganography) becomes a large research hotspot in the field of information hiding.
Image steganography generally has three metrics, capacity, imperceptibility, and resistance to detection. Capacity: the length of the secret information embedded in the carrier image; imperceptibility: the perception characteristic of the secret image is kept, namely the similarity of the secret image and the original carrier image in visual effect is kept; resistance to detection: that is, the method should have a certain anti-detection property for detection means such as steganalysis that may be encountered in the communication process.
The traditional image steganography is based on a cost function designed by professionals, and the embedding position of secret information in a spatial domain or a frequency domain of a carrier image is determined by calculating single-point loss and global loss through the cost function so as to ensure that the influence on the perception characteristic of the carrier image is minimum. However, the traditional steganalysis method has the disadvantages of low capacity, poor visual effect, difficulty in resisting steganalysis and the like.
With the development of deep learning, the fields of convolutional neural network target classification, image segmentation, scene recognition and the like are fully applied, and fruitful results are obtained. Meanwhile, an image steganography scheme taking a deep convolutional neural network as an implementation means is also rapidly developed. The steganography and the extraction of the image are realized by extracting and fusing the features of the image by utilizing a deep convolutional neural network. In the training process of the network model, by means of the constraint of a loss function, the network weight is updated through a back propagation algorithm, and a final hidden network and an extracted network are obtained after a loss value is converged, but the problem of low steganography capacity still exists.
Disclosure of Invention
The invention provides a high-capacity image steganography method, which is used for solving the problem of low capacity of the steganography method in the prior art.
In order to solve the technical problem, the technical scheme of the invention comprises the following steps:
the invention discloses a steganography method for a large-capacity image, which comprises the following steps of:
inputting the carrier image and the two secret images into a trained image steganography model to obtain a secret-carrying image and two extracted images, wherein the secret-carrying image and the carrier image are kept consistent in visual effect, and the two extracted images are respectively kept consistent in visual effect with the two secret images;
wherein the image steganography model comprises a hiding phase and an extracting phase;
the hiding stage comprises a carrier image preprocessing network, a secret image preprocessing network and a hiding network; the carrier image preprocessing network is used for preprocessing an input carrier image; the secret image preprocessing network is used for preprocessing the two cascaded secret images; the hidden network is used for processing the output of the cascaded carrier image preprocessing network and the output of the secret image preprocessing network to obtain a carrier image containing two secret images;
the extraction phase comprises two extraction networks; the two extraction networks are used for respectively processing the input secret-carrying images to obtain corresponding extraction images.
The beneficial effects of the above technical scheme are: the image steganography model constructed by the invention can embed two secret images into the carrier image to obtain the secret carrying image, so that the visual effect of the secret carrying image is consistent with that of the carrier image, the steganography capacity is obviously improved, and the imperceptibility and the detectability resistance of the secret carrying image are kept; moreover, the invention can also extract two extracted images, and the two extracted images and the two secret images are basically consistent in visual effect.
Further, in order to extract high-frequency texture features of the carrier image to help the hiding network to embed secret information into a high-frequency texture area of the carrier to enhance the visual effect of the carrier image, the carrier image preprocessing network comprises a jump connection, and a convolution layer and a maximum pooling layer which are connected in sequence; the input of the convolution layer of the carrier image preprocessing network is a carrier image; the jump connection is used for cascading the output of the maximum pooling layer with the carrier image, and the result after cascading is the output of the carrier image preprocessing network.
Further, in order to preserve the global features of the secret image to maintain the integrity of the secret image in the feature extraction of the subsequent network, the secret image preprocessing network comprises a convolution layer and an average pooling layer which are connected in sequence, the input of the convolution layer of the secret image preprocessing network is two secret images after concatenation, and the output of the average pooling layer is the output of the secret image preprocessing network.
Further, in order to keep the secret-carrying image consistent with the carrier image in visual effect, the hidden network comprises a down-sampling module, a pyramid pooling module and an up-sampling module; the down-sampling module of the hidden network comprises four convolution layers which are sequentially connected, wherein the input of the first convolution layer in the four convolution layers is the input of the hidden network, and the output of the last convolution layer is the input of the pyramid pooling module; the up-sampling module of the hidden network comprises five convolution layers and four jump connections, wherein the four jump connections are used for cascading the outputs of the first four convolution layers in the five convolution layers with the outputs of the convolution layers with the corresponding sizes in the down-sampling module, the input of the first convolution layer is the output of the pyramid pooling module, the inputs of the last four convolution layers are the outputs of the four jump connections respectively, and the output of the last convolution layer is the output of the hidden network.
Further, in order to be suitable for image steganography, the pyramid pooling module of the hidden network comprises five pooling layers, five convolution layers and five upsampling layers; the inputs of the five pooling layers are all the outputs of the down-sampling module of the hidden network; the inputs of the five convolution layers are the outputs of the five pooling layers respectively; the inputs of the five upsampling layers are the outputs of the five convolutional layers, and the output is the output of the pyramid pooling module of the hidden network.
Furthermore, in order to keep the extracted image consistent with the secret image in visual effect, the two extraction networks have the same structure, and each extraction network comprises a down-sampling module, a pyramid pooling module and an up-sampling module; the down-sampling module of the extraction network comprises four convolution layers which are sequentially connected, wherein the input of the first convolution layer in the four convolution layers is the input of the extraction network, and the output of the last convolution layer is the input of the pyramid pooling module; the up-sampling module of the extraction network comprises five convolution layers and four jump connections, wherein the four jump connections are used for cascading the outputs of the first four convolution layers in the five convolution layers with the outputs of the convolution layers with the corresponding sizes in the down-sampling module, the input of the first convolution layer is the output of the pyramid pooling module, the inputs of the last four convolution layers are the outputs of the four jump connections respectively, and the output of the last convolution layer is the output of the extraction network.
Further, in order to be suitable for image steganography, the pyramid pooling module of the extraction network comprises five pooling layers, five convolution layers and five upsampling layers; the inputs of the five pooling layers are all the outputs of the down-sampling module of the extraction network; the inputs of the five convolution layers are the outputs of the five pooling layers respectively; the inputs of the five upsampling layers are the outputs of the five convolutional layers, and the output is the output of the pyramid pooling module of the extraction network.
Further, in order to obtain a better trained image steganography model, in the process of training the image steganography model, the total loss function adopted is as follows:
therein, LossSumRepresenting the total loss function, LossHiRepresenting the hidden loss function, LossE, of each groupiRepresents the extraction loss function for each group, phi is a weighting value, and n represents the total number of groups.
Further, the loss value is calculated using the mean square error as a concealment loss function.
Further, the loss value is calculated using the mean square error as an extraction loss function.
Drawings
FIG. 1 is an overall framework diagram of the image steganography model of the present invention;
FIG. 2 is an architectural diagram of the carrier image pre-processing network of the present invention;
FIG. 3 is an architecture diagram of the secret image pre-processing network of the present invention;
FIG. 4 is an architecture diagram of the pyramid pooling module of the present invention;
FIG. 5 is an architecture diagram of the hidden network of the present invention;
FIG. 6 is a block diagram of the extraction network of the present invention;
FIG. 7 is a sample graph of the present invention for image steganography and extraction;
FIG. 8 is a graph showing the results of the assay for resistance to detection according to the present invention.
Detailed Description
According to the high-capacity image steganography method, two secret images can be embedded into a carrier image through a trained image steganography model to obtain a secret-carrying image, so that the secret-carrying image and the carrier image are consistent in visual effect, and imperceptibility and anti-detection performance of the secret-carrying image are kept; and the secret-carrying images can be extracted, two extracted images are successfully extracted from the secret-carrying images, and the two extracted images are consistent with the two secret images embedded in advance in visual effect. The method is described in detail below with reference to the accompanying drawings.
Step one, constructing an image steganography model. In this embodiment, the image steganography model is based on a deep convolutional neural network model, and the basic architecture thereof is shown in fig. 1. The image steganography model includes a hiding phase and an extraction phase.
The hiding phase comprises a carrier image preprocessing network, a secret image preprocessing network and a hiding network. In fig. 2 to 6, (D, H, W), D represents the number of channels of the tensor (also referred to as the number of convolution kernels of the convolution layer that outputs the tensor), and H and W represent the height and width of the tensor, respectively; the rectangles in the figure represent tensors, and the differently shaped arrows represent corresponding convolution, pooling, upsampling, and concatenation operations.
The carrier image preprocessing network is used for preprocessing an input carrier image, extracting a feature tensor containing high-frequency features of the carrier image, cascading the feature tensor containing the high-frequency features of the carrier image with the carrier image, and outputting a cascaded result as the carrier image preprocessing network. As shown in fig. 2, in the present embodiment, the carrier image preprocessing network includes a jump connection, and a convolution layer and a max-pooling layer connected in sequence. The maximum pooling layer can extract high-frequency texture features of the carrier image, and is beneficial to embedding secret information into a high-frequency texture area of the carrier image by a hidden network, so that the visual effect of the carrier image is enhanced. Considering that the final goal of the hiding stage is to generate a carrier image which is similar to the carrier image as much as possible, a jump connection is added in the carrier image preprocessing network, and the carrier image and the feature tensor output by the maximum pooling layer are cascaded to reduce the loss of the carrier image features in the subsequent feature extraction.
The secret image preprocessing network is used for preprocessing the two cascaded secret images and extracting a feature tensor containing global features of the two secret images. As shown in fig. 3, in the present embodiment, the secret image preprocessing network includes a convolution layer and an average pooling layer connected in sequence. The average pooling layer may preserve global features of the secret image, thereby preserving the integrity of the secret image in feature extraction of subsequent networks.
The hidden network is used for processing the output of the cascaded secret image preprocessing network and the output of the carrier image preprocessing network, namely: and performing feature extraction/fusion processing on the feature tensor containing the global features of the secret image and the feature tensor containing the high-frequency features of the carrier image to obtain a secret-carrying image embedded with two secret images, wherein the secret-carrying image and the carrier image are consistent in visual effect.
As shown in fig. 5, in the present embodiment, the hidden network includes a downsampling module, a pyramid pooling module, and an upsampling module. The down-sampling module comprises four convolution layers which are sequentially connected, the input of the first convolution layer in the four convolution layers is the input of the hidden network, the output of the last convolution layer is the input of the pyramid pooling module, the continuous four convolution layers reduce the size of the input tensor by half, and meanwhile, the number of channels is increased. The up-sampling module of the hidden network comprises five convolution layers and four jump connections, wherein the four jump connections are used for cascading the outputs of the first four convolution layers in the five convolution layers with the outputs of the convolution layers with the corresponding sizes in the down-sampling module, the input of the first convolution layer is the output of the pyramid pooling module, the inputs of the last four convolution layers are the outputs of the four jump connections respectively, and the output of the last convolution layer is the output of the hidden network. The function is as follows: the first four convolutional layers double the size of the input tensor, reduce the number of channels at the same time, which can be regarded as the inverse process of down-sampling, and finally reduce the number of channels to 3 through one convolutional layer to output a secret-carrying image; furthermore, the added four-hop connection can cascade down-sampled tensors into up-sampling steps of corresponding size, which helps to reduce the loss of important features.
As shown in fig. 4, the pyramid pooling module includes five pooling layers, five convolution layers and five upsampling layers; the inputs of the five pooling layers are all the outputs of the down-sampling module of the hidden network; the inputs of the five convolution layers are the outputs of the five pooling layers respectively; the inputs of the five upsampling layers are the outputs of the five convolutional layers, and the input of the upsampling module of the hidden network is the output of the pyramid pooling module. The function is as follows: the pyramid pooling module performs downsampling on the input tensor with different sizes through five parallel convolution layers, then expands the input tensor to the same size through five upsampling layers and then performs cascade output.
The extraction stage comprises two extraction networks, the two extraction networks have the same structure, and the two extraction networks are used for respectively processing the input secret-carrying images to obtain corresponding extraction images. As shown in fig. 6, in this embodiment, the structure of each extraction network (including the pyramid pooling module therein) is substantially the same as that of the hidden network, and the difference is the number of convolution kernels of the input layer, where the hidden network is 21 and the extraction network is 3. Furthermore, the two extraction networks have the same structure, but they are two networks with different parameter values.
And step two, after the image steganography model is constructed, the image steganography model needs to be trained to obtain the trained image steganography model. The specific process is as follows:
1. 40,000 images were randomly selected from ImageNet and formed into a training set, a validation set, and a test set, respectively, in a number ratio of 6:1: 1. The training set is used for training the network, the verification set verifies the performance of the network model obtained after each round of training is finished, and the test set is used for testing the performance of the obtained optimal network model after the whole training process is finished.
2. All images of the training set are randomly divided into two groups of carrier images and secret images, the number of the two groups of images is the same, and only one carrier image and two secret images are input each time.
3. And inputting the carrier image and the two secret images into an image steganography model so as to finally obtain two different extracted images. The specific process is as follows: inputting the carrier image into a corresponding carrier preprocessing network to obtain a result of cascade connection of a characteristic tensor of high-frequency characteristics of the carrier image and the carrier image; two secret images are cascaded and then input to a corresponding secret image preprocessing network, and a feature tensor containing the global features of the secret images is extracted; the output of the carrier preprocessing network and the output of the secret image preprocessing network are cascaded and input to the hidden network to obtain a secret image containing two secret images; and respectively and simultaneously inputting the secret-carrying images into two extraction networks with the same structure, and extracting two different extraction images.
4. Judging whether all the images of the training set are input into the image steganography model: if so, judging that the training of the current round is finished, and continuing the next step; if so, returning to the step 3 to continue the training of the current round.
5. After the training of the current round is finished, images of the verification set are randomly divided into two groups, namely a carrier image and a secret image, the two groups are input into the current model, and a corresponding carrier image and an extraction image are obtained.
6. After all the images of the verification set are input, the mean square error is used as a hiding loss function to calculate hiding loss values between each group of secret images and the original secret images, and the mean square error is used as an extraction loss function to calculate extraction loss values between each group of extracted images and the corresponding original secret images. The mean square error is calculated as follows:
where X represents the original image and Y represents the modified image.
7. And weighting the hidden loss value and the extracted loss value of each group to obtain a corresponding total loss value, and then calculating the average value to be used as the total loss value of the current round. The total loss value is calculated as follows:
therein, LossSumRepresenting the total loss function, LossHiRepresenting the hidden loss function, LossE, of each groupiRepresents the extraction loss function of each group, phi is a weighted value, in this embodiment, phi is 0.5, and n represents the total number of groups.
8. Comparing whether the total loss value of the previous round is less than the previous minimum total loss value: if yes, the weight value of the current model is saved, and if not, the weight value is not saved.
9. Judging whether the number of training rounds reaches a threshold value: if so, ending the training, otherwise returning to the step 3 to continue the next round of training.
10. And after training is finished, testing the images of the test set by using the stored multiple groups of models. And measuring the obtained image by using indexes such as peak signal-to-noise ratio, structural similarity and the like, wherein a group of models with the optimal result is the final result, namely the trained image steganography model is obtained.
And step three, after the trained image steganography model is obtained, inputting one color carrier image and two color secret images into the trained image steganography model, thereby obtaining a secret carrier image which is embedded with the two color secret images and has no difference with the carrier image in visual effect, and obtaining two color extraction images which are extracted from the secret carrier image and have no difference with the two secret images in visual effect.
In order to verify the effectiveness of the method, the method is trained and verified on data sets such as ImageNet, COCO, CeleA and the like, and a remote sensing image is added for testing, wherein a sample is shown in FIG. 7, and the original carrier image and the secret image, the original secret image and the extracted image can be observed to be consistent in visual effect. After the model is trained, the performance of the model is tested by using a test set (the test set is the ImageNet test set mentioned in the second part 1 of the specification), and after the test is finished, a secret image and an original carrier image are obtained. From the test results, 350 secret images and 50 carrier images were randomly selected and analyzed using steganalysis software (i.e., the software attempted to distinguish which secret images were carrier images) to obtain a graph of the results of the anti-detection analysis as shown in fig. 8. The meaning of fig. 8 is: the closer the zigzag line is to the dotted line, the better the detection resistance of the steganographic model. As can be seen from this figure, the method of the present invention is more resistant to detection.
In summary, the large-capacity image steganography method of the present invention has the following characteristics:
1. the method can hide two secret images with the same size at the same time, improves the steganographic capacity from 1 byte/pixel to 2bytes/pixel, obviously improves the steganographic capacity, and simultaneously keeps the imperceptibility and the anti-detection performance of steganography.
2. The invention only uses the hidden network and the extraction network to carry out synchronous training, and has high convergence speed.
3. The hidden stage and the extraction stage in the image steganography model provided by the invention have small parameter quantity and low calculation quantity, and the requirement on hardware during training is not high (the hardware equipment in the embodiment of the invention is only NVIDIA GeForce GTX 1080-8 GB).
Claims (10)
1. A steganography method for a large-capacity image is characterized by comprising the following steps:
inputting the carrier image and the two secret images into a trained image steganography model to obtain a secret-carrying image and two extracted images, wherein the secret-carrying image and the carrier image are kept consistent in visual effect, and the two extracted images are respectively kept consistent in visual effect with the two secret images;
wherein the image steganography model comprises a hiding phase and an extracting phase;
the hiding stage comprises a carrier image preprocessing network, a secret image preprocessing network and a hiding network; the carrier image preprocessing network is used for preprocessing an input carrier image; the secret image preprocessing network is used for preprocessing the two cascaded secret images; the hidden network is used for processing the output of the cascaded carrier image preprocessing network and the output of the secret image preprocessing network to obtain a carrier image containing two secret images;
the extraction phase comprises two extraction networks; the two extraction networks are used for respectively processing the input secret-carrying images to obtain corresponding extraction images.
2. A large-capacity image steganography method as claimed in claim 1, wherein the carrier image preprocessing network comprises a hopping connection, and a convolution layer and a max-pooling layer connected in sequence; the input of the convolution layer of the carrier image preprocessing network is a carrier image; the jump connection is used for cascading the output of the maximum pooling layer with the carrier image, and the result after cascading is the output of the carrier image preprocessing network.
3. A large-capacity image steganography method according to claim 1, wherein the secret image preprocessing network comprises a convolution layer and an average pooling layer which are connected in sequence, the input of the convolution layer of the secret image preprocessing network is two secret images after concatenation, and the output of the average pooling layer is the output of the secret image preprocessing network.
4. A large-capacity image steganography method as claimed in claim 1, wherein the hidden network comprises a down-sampling module, a pyramid pooling module and an up-sampling module; the down-sampling module of the hidden network comprises four convolution layers which are sequentially connected, wherein the input of the first convolution layer in the four convolution layers is the input of the hidden network, and the output of the last convolution layer is the input of the pyramid pooling module; the up-sampling module of the hidden network comprises five convolution layers and four jump connections, wherein the four jump connections are used for cascading the outputs of the first four convolution layers in the five convolution layers with the outputs of the convolution layers with the corresponding sizes in the down-sampling module, the input of the first convolution layer is the output of the pyramid pooling module, the inputs of the last four convolution layers are the outputs of the four jump connections respectively, and the output of the last convolution layer is the output of the hidden network.
5. A large-capacity image steganography method according to claim 4, wherein the pyramid pooling module of the hidden network comprises five pooling layers, five convolution layers and five upsampling layers; the inputs of the five pooling layers are all the outputs of the down-sampling module of the hidden network; the inputs of the five convolution layers are the outputs of the five pooling layers respectively; the inputs of the five upsampling layers are the outputs of the five convolutional layers, and the output is the output of the pyramid pooling module of the hidden network.
6. A large-capacity image steganography method as claimed in claim 1, wherein the two extraction networks have the same structure, and each extraction network comprises a down-sampling module, a pyramid pooling module and an up-sampling module; the down-sampling module of the extraction network comprises four convolution layers which are sequentially connected, wherein the input of the first convolution layer in the four convolution layers is the input of the extraction network, and the output of the last convolution layer is the input of the pyramid pooling module; the up-sampling module of the extraction network comprises five convolution layers and four jump connections, wherein the four jump connections are used for cascading the outputs of the first four convolution layers in the five convolution layers with the outputs of the convolution layers with the corresponding sizes in the down-sampling module, the input of the first convolution layer is the output of the pyramid pooling module, the inputs of the last four convolution layers are the outputs of the four jump connections respectively, and the output of the last convolution layer is the output of the extraction network.
7. A large-capacity image steganography method as claimed in claim 6, wherein the pyramid pooling module of the extraction network comprises five pooling layers, five convolution layers and five upsampling layers; the inputs of the five pooling layers are all the outputs of the down-sampling module of the extraction network; the inputs of the five convolution layers are the outputs of the five pooling layers respectively; the inputs of the five upsampling layers of the extraction network are the outputs of the five convolutional layers, and the output is the output of the pyramid pooling module of the extraction network.
8. A large-capacity image steganography method according to any one of claims 1 to 7, wherein in the process of training the image steganography model, the total loss function is adopted as follows:
therein, LossSumRepresenting the total loss function, LossHiRepresenting the hidden loss function, LossH, of each groupiRepresents the extraction loss function for each group, phi is a weighting value, and n represents the total number of groups.
9. A large-capacity image steganography method as claimed in claim 8, wherein the loss value is calculated using a mean square error as a concealment loss function.
10. A large-capacity image steganography method as claimed in claim 8, wherein the loss value is calculated using a mean square error as an extraction loss function.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110693663.6A CN113298689B (en) | 2021-06-22 | 2021-06-22 | Large-capacity image steganography method |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110693663.6A CN113298689B (en) | 2021-06-22 | 2021-06-22 | Large-capacity image steganography method |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN113298689A true CN113298689A (en) | 2021-08-24 |
| CN113298689B CN113298689B (en) | 2023-04-18 |
Family
ID=77329119
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202110693663.6A Active CN113298689B (en) | 2021-06-22 | 2021-06-22 | Large-capacity image steganography method |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN113298689B (en) |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN116958006A (en) * | 2023-09-19 | 2023-10-27 | 湖北微模式科技发展有限公司 | Equal-size image superposition algorithm based on pixel bidirectional fusion |
| CN117132671A (en) * | 2023-10-27 | 2023-11-28 | 哈尔滨工业大学(深圳)(哈尔滨工业大学深圳科技创新研究院) | Multi-task steganography method, system and medium based on depth self-adaptive steganography network |
Citations (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US6557103B1 (en) * | 1998-04-13 | 2003-04-29 | The United States Of America As Represented By The Secretary Of The Army | Spread spectrum image steganography |
| US20180068429A1 (en) * | 2015-04-15 | 2018-03-08 | Institute Of Automation Chinese Academy Of Sciences | Image Steganalysis Based on Deep Learning |
| CN107845116A (en) * | 2017-10-16 | 2018-03-27 | 北京京东尚科信息技术有限公司 | The method and apparatus for generating the compressed encoding of plane picture |
| CN109492416A (en) * | 2019-01-07 | 2019-03-19 | 南京信息工程大学 | A kind of guard method of big data image and system based on safety zone |
| CN109934761A (en) * | 2019-01-31 | 2019-06-25 | 中山大学 | Jpeg image steganalysis method based on convolutional neural networks |
| CN112529758A (en) * | 2020-12-18 | 2021-03-19 | 海南大学 | Color image steganography method based on convolutional neural network |
| CN112950445A (en) * | 2021-03-11 | 2021-06-11 | 河南师范大学 | Compensation-based detection feature selection method in image steganalysis |
-
2021
- 2021-06-22 CN CN202110693663.6A patent/CN113298689B/en active Active
Patent Citations (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US6557103B1 (en) * | 1998-04-13 | 2003-04-29 | The United States Of America As Represented By The Secretary Of The Army | Spread spectrum image steganography |
| US20180068429A1 (en) * | 2015-04-15 | 2018-03-08 | Institute Of Automation Chinese Academy Of Sciences | Image Steganalysis Based on Deep Learning |
| CN107845116A (en) * | 2017-10-16 | 2018-03-27 | 北京京东尚科信息技术有限公司 | The method and apparatus for generating the compressed encoding of plane picture |
| CN109492416A (en) * | 2019-01-07 | 2019-03-19 | 南京信息工程大学 | A kind of guard method of big data image and system based on safety zone |
| CN109934761A (en) * | 2019-01-31 | 2019-06-25 | 中山大学 | Jpeg image steganalysis method based on convolutional neural networks |
| CN112529758A (en) * | 2020-12-18 | 2021-03-19 | 海南大学 | Color image steganography method based on convolutional neural network |
| CN112950445A (en) * | 2021-03-11 | 2021-06-11 | 河南师范大学 | Compensation-based detection feature selection method in image steganalysis |
Non-Patent Citations (3)
| Title |
|---|
| XINTAO DUAN .ETAL: "SteganoCNN Image Steganography with Generalization Ability Based on Convolutional Neural Network", 《ENTROPY》 * |
| XINTAO DUAN .ETAL: "StegoPNet Image Steganography With Generalization Ability Based on Pyramid Pooling Module", 《IEEE ACCESS》 * |
| 孙树亮: "基于非下采样Contourlet变换和Hill编码的图像隐写", 《电信科学》 * |
Cited By (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN116958006A (en) * | 2023-09-19 | 2023-10-27 | 湖北微模式科技发展有限公司 | Equal-size image superposition algorithm based on pixel bidirectional fusion |
| CN116958006B (en) * | 2023-09-19 | 2024-01-02 | 湖北微模式科技发展有限公司 | Equal-size image superposition algorithm based on pixel bidirectional fusion |
| CN117132671A (en) * | 2023-10-27 | 2023-11-28 | 哈尔滨工业大学(深圳)(哈尔滨工业大学深圳科技创新研究院) | Multi-task steganography method, system and medium based on depth self-adaptive steganography network |
| CN117132671B (en) * | 2023-10-27 | 2024-02-23 | 哈尔滨工业大学(深圳)(哈尔滨工业大学深圳科技创新研究院) | Multi-task steganography method, system and medium based on depth self-adaptive steganography network |
Also Published As
| Publication number | Publication date |
|---|---|
| CN113298689B (en) | 2023-04-18 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Guan et al. | DeepMIH: Deep invertible network for multiple image hiding | |
| CN110349136A (en) | A kind of tampered image detection method based on deep learning | |
| CN112580782B (en) | Channel-enhanced dual-attention generation countermeasure network and image generation method | |
| CN113298689B (en) | Large-capacity image steganography method | |
| WO2022237078A1 (en) | Color image steganography method based on generative adversarial network | |
| CN108596818B (en) | Image steganalysis method based on multitask learning convolutional neural network | |
| CN108961137A (en) | A kind of image latent writing analysis method and system based on convolutional neural networks | |
| CN111861906B (en) | Pavement crack image virtual augmentation model establishment and image virtual augmentation method | |
| CN110968845B (en) | Detection method for LSB steganography based on convolutional neural network generation | |
| CN112150450A (en) | Image tampering detection method and device based on dual-channel U-Net model | |
| Wang et al. | HidingGAN: High capacity information hiding with generative adversarial network | |
| CN111222583B (en) | Image steganalysis method based on countermeasure training and critical path extraction | |
| Yuan et al. | GAN-based image steganography for enhancing security via adversarial attack and pixel-wise deep fusion | |
| CN115115500A (en) | Watermark embedding method combined with underwater image enhancement | |
| CN113887573A (en) | Human face forgery detection method based on visual converter | |
| CN116188439A (en) | False face-changing image detection method and device based on identity recognition probability distribution | |
| Shang et al. | Robust data hiding for JPEG images with invertible neural network | |
| CN112862655B (en) | JPEG image steganalysis method based on channel space attention mechanism | |
| CN116152061A (en) | Super-resolution reconstruction method based on fuzzy core estimation | |
| CN115311117A (en) | Image watermarking system and method for style migration depth editing | |
| CN116258973A (en) | Multi-scale feature combined remote sensing image scene classification method | |
| Golea et al. | A novel multi-objective genetic algorithm optimization for blind RGB color image watermarking | |
| Le et al. | Why do facial deepfake detectors fail? | |
| Benseddik et al. | Efficient interpolation-based reversible watermarking for protecting fingerprint images | |
| Li et al. | Image steganalysis via squeeze-and-excitation and separable convolutional neural network |
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 |