CN114338945A - Color image steganography method and device based on frequency domain component selection - Google Patents
Color image steganography method and device based on frequency domain component selection Download PDFInfo
- Publication number
- CN114338945A CN114338945A CN202210005481.XA CN202210005481A CN114338945A CN 114338945 A CN114338945 A CN 114338945A CN 202210005481 A CN202210005481 A CN 202210005481A CN 114338945 A CN114338945 A CN 114338945A
- Authority
- CN
- China
- Prior art keywords
- image
- secret
- steganography
- frequency domain
- color image
- 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.)
- Pending
Links
Images
Landscapes
- Image Processing (AREA)
Abstract
The invention relates to a color image steganography method based on frequency domain component selection. The color image steganography method based on frequency domain component selection comprises the following steps: constructing a color image steganography model selected based on the frequency domain components; training a color image steganography model; inputting the carrier image and the secret image into a color image steganography model to generate a carrier image and a reconstructed secret image to complete image steganography; the color image steganography model comprises a frequency component selection module, a characteristic data extraction module, a secret-carrying image generation module and a secret image reconstruction module; the frequency component selection module is used for receiving the carrier image and generating first input data; the characteristic data extraction module is used for extracting second input data; the secret-carrying image generation module is used for receiving the first input data and the second input data and generating a secret-carrying image; the secret image reconstruction module is used for generating a reconstructed secret image. The steganography method has good imperceptibility and high information transmission safety.
Description
Technical Field
The invention relates to the technical field of image steganography, in particular to a color image steganography method and a color image steganography device based on frequency domain component selection.
Background
Currently, many image technologies have been developed and successfully applied to various fields such as finance, medical treatment, communication, and the like. The image steganography technology is a current research and application hotspot, and plays an important role in the application of intellectual property protection, secret communication and the like by hiding information into an image to realize secret transfer of the information.
An excellent image steganography technique should have both large steganography capacity and good imperceptibility. But there is a strong correlation between the capacity and imperceptibility, and an increase in steganographic capacity in an image tends to result in a decrease in imperceptibility. Currently, most image steganography techniques have difficulty in simultaneously combining these two characteristics.
In the traditional image steganography technology, an image is used as a carrier, and steganography steps are designed by a manual design method, so that secret information such as bit streams and texts is embedded into the image. Although conventional image steganography techniques are currently well developed, their capacities are generally small. The performance of the traditional image steganography technology is difficult to be further improved by a manual design method. The introduction of deep learning techniques breaks this impasse. The image steganography technology based on Deep learning proposed at present can complete the steganography process, for example, Hayes GAN proposed by Hayes [1] "Generation structural images via adaptive tracing." NIPS (2017) and the like, Zhu [2] "HiDDeN: Hiding Data With Deep networks." ECCV (2018) and the like are typical image steganography technology based on Deep learning. However, when the hidden information capacity is too large, the generated image still generates modification marks visible to human eyes, so that the imperceptibility of the steganography is reduced. For example, the Deep steganography model proposed by Baluja [3] "high Images in Plain Sight: Deep Steganography. In practical applications, the obvious steganography can possibly cause the suspicion of a third party, thereby influencing the secret transmission of information.
Based on the existing research results, most of the current image steganography technologies based on deep learning are steganography in the spatial domain. Spatial image information is information that is directly visible to the human eye, and therefore, direct modification of spatial domain information is prone to produce changes in color or texture, i.e., steganographic traces.
Disclosure of Invention
Based on the depth learning technology and the image frequency domain processing method, the invention aims to provide the color image steganography technology and the color image steganography model based on frequency domain component selection. The technology is used for solving the problem of the invisible trace which is obviously visible to human eyes in the image steganography, so that the imperceptibility of the steganography technology is improved, and the information safety of transmitted data is ensured.
Based on this, the invention aims to provide a color image steganography method based on frequency domain component selection, which combines a depth learning technology and an image frequency domain processing method, avoids the problem that steganography traces are too obvious in image steganography in a space domain, effectively improves the imperceptibility of the steganography method, and ensures the information security of transmission data.
The invention is realized by the following technical scheme:
a color image steganography method based on frequency domain component selection comprises the following steps: constructing a color image steganography model selected based on the frequency domain components; training the color image steganography model; inputting the color carrier image and the gray secret image into the color image steganography model to generate a carrier secret image and a reconstructed secret image to complete image steganography;
the color image steganography model comprises a frequency component selection module, a characteristic data extraction module, a secret-carrying image generation module and a secret image reconstruction module; the frequency component selection module is used for receiving the carrier image and generating a single-channel diagonal high-frequency component of the carrier image as first input data; the characteristic data extraction module is used for receiving the secret image and extracting the characteristic data of the secret image as second input data; the secret-carrying image generation module is used for receiving the first input data and the second input data and generating the secret-carrying image; the secret image reconstruction module is used for receiving the secret image and generating the reconstructed secret image.
According to the color image steganography method based on frequency domain component selection, through frequency component selection operation, a steganography mode of a map is achieved, steganography capacity is expanded, a secret-carrying image with better structural image quality and a reconstructed secret image can be generated, the limitation that space domain image steganography is poor in imperceptibility is avoided, and information transmission safety is improved.
Further, the frequency component selection module includes a channel separation unit, a first frequency domain conversion unit, and a first frequency component selection unit; the channel separation unit is used for separating single-channel information of the carrier image; the first frequency domain conversion unit is used for converting single-channel information of the carrier image into frequency domain information of the carrier image by adopting discrete wavelet transform; the first frequency component selection unit is configured to select a single-channel diagonal high-frequency component of the carrier image as the first input data.
Further, the feature data extraction module comprises a second frequency domain conversion unit and a preprocessing network; the second frequency domain conversion unit is used for converting the spatial domain information of the secret image into the frequency domain information of the secret image by adopting discrete wavelet transform, and transmitting the frequency domain information of the secret image to the preprocessing network; the preprocessing network is used for receiving the frequency domain information of the secret image and extracting the characteristic data of the secret image as the second input data.
Further, the preprocessing network sequentially stacks a first input layer, a first convolution group, a first connection layer, a second convolution group and a second connection layer; the first convolution group and the second convolution group each include a plurality of convolution layers.
Further, the secret-carrying image generation module comprises an encoding network and a first spatial domain conversion unit; the coding network is used for receiving the first input data and the second input data, hiding the second input data into the first input data and generating a single-channel diagonal high-frequency component of the secret-carrying image; the first spatial domain conversion unit is used for receiving a single-channel diagonal high-frequency component of the secret-carrying image and generating the secret-carrying image by adopting inverse discrete wavelet transform.
Further, the coding network sequentially stacks a second input layer, a third connection layer, a third convolution group, a fourth connection layer and a first output layer; the third convolution group includes a plurality of convolution layers.
Further, the secret image reconstruction module includes a second frequency component selection unit, a decoding network, and a second spatial-domain conversion unit; the second frequency component selection unit is used for receiving the secret-carrying image and generating a single-channel diagonal high-frequency component of the secret-carrying image; the decoding network is used for receiving a single-channel diagonal high-frequency component of the secret image and generating frequency domain information of the reconstructed secret image; the second spatial domain conversion unit is used for receiving the frequency domain information of the reconstructed secret image and generating the reconstructed secret image by adopting inverse discrete wavelet transform.
Further, the decoding network sequentially stacks a third input layer, a fourth convolution group, a fifth connection layer and a second output layer; the fourth convolution group includes a plurality of convolution layers.
Further, the training process of the color image steganography model comprises the following steps: dividing an LFW data set into a training set, a verification set and a test set at a ratio of 10:1:1 at random; dividing the training set, the verification set and the test set into a carrier image set and a secret image set according to the proportion of 1:1 respectively; uniformly carrying out gray processing on the images in the secret image set; training the color image steganography model by using the training set, and iteratively updating parameters until the color image steganography model converges; in the iteration process, loss calculation is carried out by using a mean square error function; verifying the color image steganography model in training using the verification set every 20 iterations; and after the color image steganography model converges, testing the performance of the color image steganography model by using the test set.
The invention also provides a color image steganography device based on frequency domain component selection, which comprises a model construction module, a color image steganography module and a color image steganography module, wherein the model construction module is used for constructing a color image steganography model based on frequency domain component selection; the model training module is used for training the color image steganography model; the image steganography module is used for inputting the color carrier image and the gray secret image into the trained color image steganography model and generating a secret carrier image and a reconstructed secret image;
the color image steganography model comprises a frequency component selection module, a characteristic data extraction module, a secret-carrying image generation module and a secret image reconstruction module; the frequency component selection module is used for receiving the carrier image and generating a single-channel diagonal high-frequency component of the carrier image as first input data; the characteristic data extraction module is used for receiving the secret image and extracting the characteristic data of the secret image as second input data; the secret-carrying image generation module is used for receiving the first input data and the second input data and generating the secret-carrying image; the secret image reconstruction module is used for receiving the secret image and generating the reconstructed secret image.
Compared with the prior art, the color image steganography method and the color image steganography device based on frequency component selection combine an image frequency domain processing technology with a coding-decoding network in a deep learning technology, realize a steganography mode of a hidden image, and remarkably improve steganography capacity; meanwhile, the steganography method can generate a secret-carrying image and a reconstructed secret image which are similar in structure and good in image quality, and the imperceptibility of the steganography method and the safety of information transmission are guaranteed.
For a better understanding and practice, the invention is described in detail below with reference to the accompanying drawings.
Drawings
Fig. 1 is a flowchart illustrating a color image steganography method based on frequency domain component selection according to an embodiment of the present invention;
FIG. 2 is a network architecture diagram of a color image steganography model based on frequency domain component selection according to an embodiment of the present invention;
fig. 3 is a schematic flow chart of a frequency component selection module according to an embodiment of the present invention;
FIG. 4 is a network architecture diagram of a preprocessing network according to an embodiment of the present invention;
FIG. 5 is a network structure diagram of a coding network according to an embodiment of the present invention;
fig. 6 is a network structure diagram of a decoding network according to an embodiment of the present invention;
FIG. 7 is a flowchart illustrating the steps of training a steganography model for color images according to an embodiment of the present invention;
fig. 8 is a steganography effect diagram of the color image steganography model with the B channel of the carrier image as the embedded domain according to an embodiment of the present invention.
Detailed Description
In order to make the objects, technical solutions and advantages of the present application more clear, embodiments of the present application will be described in further detail below with reference to the accompanying drawings.
It should be understood that the embodiments described are only some embodiments of the present application, and not all embodiments. All other embodiments obtained by a person of ordinary skill in the art based on the embodiments in the present application without any creative effort belong to the protection scope of the embodiments in the present application.
The terminology used in the embodiments of the present application is for the purpose of describing particular embodiments only and is not intended to be limiting of the embodiments of the present application. As used in the examples of this application and the appended claims, the singular forms "a", "an", and "the" are intended to include the plural forms as well, unless the context clearly indicates otherwise. It should also be understood that the term "and/or" as used herein refers to and encompasses any and all possible combinations of one or more of the associated listed items.
When the following description refers to the accompanying drawings, like numbers in different drawings represent the same or similar elements unless otherwise indicated. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with the present application. Rather, they are merely examples of apparatus and methods consistent with certain aspects of the application, as detailed in the appended claims. In the description of the present application, it is to be understood that the terms "first," "second," "third," and the like are used solely to distinguish one from another and are not necessarily used to describe a particular order or sequence, nor are they to be construed as indicating or implying relative importance. The specific meaning of the above terms in the present application can be understood by those of ordinary skill in the art as appropriate.
Further, in the description of the present application, "a plurality" means two or more unless otherwise specified. "and/or" describes the association relationship of the associated objects, meaning that there may be three relationships, e.g., a and/or B, which may mean: a exists alone, A and B exist simultaneously, and B exists alone. The character "/" generally indicates that the former and latter associated objects are in an "or" relationship.
Please refer to fig. 1, which is a flowchart illustrating a color image steganography method based on frequency domain component selection according to this embodiment. The method comprises the following steps:
s1, constructing a color image steganography model selected on the basis of the frequency domain components;
s2, training a color image steganography model;
and S3, inputting the color carrier image C and the gray secret image S into a color image steganography model, and generating a carrier image C 'and a reconstructed secret image S' to finish image steganography.
Please refer to fig. 2, which is a network structure diagram of a color image steganography model based on frequency domain component selection according to the present embodiment. In step S1, the color image steganography model includes a frequency component selection module, a feature data extraction module, a secret image generation module, and a secret image reconstruction module.
The frequency component selection module is used for receiving the carrier image C and generating a single-channel diagonal high-frequency component of the carrier image C as first input data. The frequency component selection module includes a channel separation unit, a first frequency domain conversion unit, and a first frequency component selection unit. The channel separation unit is used for separating single-channel information of the carrier image C. The first frequency domain conversion unit is configured to convert single-channel information of the carrier image C into frequency domain information of the carrier image C using Discrete Wavelet Transform (DWT). The first frequency component selection unit is used for selecting a single-channel diagonal high-frequency component of the carrier image C as first input data of the carrier image generation module.
Please refer to fig. 3, which is a flowchart illustrating a frequency component selection module according to the present embodiment. In the example of fig. 3, a carrier image C in RGB format is input to a channel separation unit, and B-channel information C of the carrier image C is separatedB(ii) a C is to beBAs an embedded field of secret information, without any modification of the R-channel and G-channel information. Then C is mixedBInputting into a first frequency domain conversion unit, converting C by DWTBConversion into frequency domain information CLL,CLH,CHL,CHH](ii) a Wherein, CLLThe low frequency component and the other three components are high frequency components. Information of frequency domain [ C ]LL,CLH,CHL,CHH]Input to a first frequency component selection unit to select a diagonal high-frequency component CHHAs the first input data of the secret image generation module, other components are not modified.
The characteristic data extraction module is used for receiving the secret image S and extracting the characteristic data X of the secret image S as second input data. The feature data extraction module comprises a second frequency domain conversion unit and a preprocessing network. The second frequency domain conversion unit is used for converting the spatial domain information of the secret image S into the frequency domain information of the secret image S by adopting DWT (discrete wavelet transform) conversion and transmitting the frequency domain information of the secret image S to the preprocessing network; the preprocessing network is used for receiving the frequency domain information of the secret image S and extracting the characteristic data X of the secret image S as second input data.
Inputting the original data of the gray secret image S into a second frequency domain conversion unit, and obtaining the frequency domain information [ S ] of the S through DWT conversionLL,SLH,SHL,SHH](ii) a Information of frequency domain SLL,SLH,SHL,SHH]And inputting the data into a preprocessing network, extracting the characteristic data X of the S, and using the characteristic data X as second input data of the secret-carrying image generation module.
Please refer to fig. 4, which is a network structure diagram of the preprocessing network according to the present embodiment. The preprocessing network has stacked in order a first input layer, a first convolution group, a first connection layer, a second convolution group, and a second connection layer. Wherein, the first input layer (i.e. input layer 1) is used for receiving the input of the secret image S frequency domain information; the first convolution group and the second convolution group respectively comprise a plurality of convolution layers and are used for extracting the characteristic data X of the S frequency domain information; the first connection layer (i.e., connection layer 1) and the second connection layer (i.e., connection layer 2) perform a connection operation on outputs of the plurality of convolution layers of the first convolution group and the second convolution group, respectively.
Optionally, the first convolution group includes convolution layers with different convolution kernel structures, and the convolution kernel structures are expressed in a form of (number, size), so that convolution kernel structures of convolution layers 1 to 3 are (50,3 × 3), (10,4 × 4), and (5,5 × 5), respectively; the step sizes of all the convolution layers are set to be 1, and the ReLu function is adopted as the activation function. The second convolution groups (convolution layers 4-6) have the same structure as the first convolution groups (convolution layers 1-3).
The secret image generation module comprises an encoding network and a first spatial domain conversion unit. The encoding network is used for receiving the first input data and the second input data, hiding the second input data into the first input data and generating a single-channel diagonal high-frequency component of the secret image C'; the first spatial domain conversion unit is used for receiving the single-channel diagonal high-frequency components of the secret-carrying image C 'and generating the secret-carrying image C' by adopting inverse discrete wavelet transform (namely, inverse DWT transform).
Will be firstAn input data CHHAnd second input data X is input into the encoding network to generate a B channel diagonal high frequency component C 'of the secret image C'HH(ii) a C'HHInputting into a first space domain converting means of C'HHAnd fusing the frequency domain information of the C of the carrier image and generating a carrier image C' by using DWT inverse transformation.
Please refer to fig. 5, which is a network structure diagram of the coding network according to the present embodiment. The coding network has stacked in order a second input layer, a third connection layer, a third convolution group, a fourth connection layer, and a first output layer. Wherein the second input layer (i.e. input layer 2) is arranged to receive the first input data CHH(ii) a The third connection layer (i.e. connection layer 3) is for receiving and connecting the first input data CHHAnd second input data X; the third convolution group comprises a plurality of convolution layers for pair CHHAnd X is coded; the fourth connection layer (i.e., connection layer 4) performs a connection operation on the outputs of the plurality of convolution layers of the third convolution group; the first output layer (i.e., output layer 1) is for extracting the features of the connected data, generating and outputting the single-channel diagonal high-frequency component C 'of the secret-code image'HH。
Optionally, the structure of the third convolution group is the same as that of the first convolution group, that is, the convolution kernel structures of the convolution layers 7 to 9 are (50,3 × 3), (10,4 × 4), and (5,5 × 5), the step lengths are all set to 1, and the ReLu functions are all used as the activation functions.
The secret image reconstruction module includes a second frequency component selection unit, a decoding network, and a second spatial-domain conversion unit. The second frequency component selection unit is used for receiving the secret image C ' and generating a single-channel diagonal high-frequency component C ' of the secret image C 'HH(ii) a The decoding network is used for receiving the single-channel diagonal high-frequency component C 'of the secret image C'HHGenerating spatial domain information of the reconstructed secret image S'; the second spatial-domain conversion unit is used for receiving the spatial-domain information of the reconstructed secret image S 'and generating the reconstructed secret image S' by adopting DWT inverse transformation.
The secret image C ' is input to a second frequency component selection unit, and the diagonal high-frequency component C ' of the B channel is output 'HH(ii) a C'HHInput decoding networkAnd outputs frequency domain information [ S ] of the reconstructed secret image S'LL,S’LH,S’HL,S’HH](ii) a The frequency domain information is input to a second spatial domain conversion unit, and S' is generated by inverse DWT.
Please refer to fig. 6, which is a network structure diagram of the decoding network according to the present embodiment. The decoding network sequentially stacks a third input layer, a fourth convolution group, a fifth connection layer, and a second output layer. Wherein the third input layer (i.e. input layer 3) is for receiving the secret image C'HH(ii) a A fourth convolution group comprising a plurality of convolution layers for extracting C'HHThe feature data X of the secret image in (1); the fifth connection layer (i.e., connection layer 5) performs a connection operation on the outputs of the plurality of convolution layers of the fourth convolution group; the second output layer extracts the characteristics of the connection data and outputs frequency domain information [ S 'of the reconstructed secret image S'LL,S’LH,S’HL,S’HH]。
Please refer to fig. 7, which is a flowchart illustrating a step of training a color image steganography model based on frequency domain component selection according to this embodiment. In step S2, the training process for the color image steganography model further includes the following sub-steps:
s21, preparing a data set: randomly dividing an outdoor face recognition (The laboratory Faces in The Wild, LFW) database into a training set, a verification set and a test set in a ratio of 10:1: 1; then, respectively dividing the training set, the verification set and the test set into 1:1, dividing the image into a carrier image set and a secret image set, and uniformly carrying out gray processing on the images of the secret image set;
s22, training a color image steganography model: training the color image steganography model by using the training set in the S21, and iteratively updating parameters until the color image steganography model converges;
s23, verification of the color image steganography model: every 20 iterations, the color image steganography model under training is verified using the verification set in S21;
s24, testing of the color image steganography model: after the color image steganography model converges, the performance of the color image steganography model is tested using the test set in S21.
In step S22, the present embodiment uses mean-square error (MSE) as a loss function and updates the network weights by Adam optimizer; the learning rate is set to 0.001 during training and is reduced according to certain iteration; the batch size was set to 16; training was performed for a total of 400 iterations. Wherein the loss function is defined as follows:
Loss(C,C',S,S')＝||C-C'||2+β|S-S'||2
in the above formula, C represents carrier image data, and C' represents secret image data; s denotes secret image data, and S' denotes reconstructed secret image data. Beta in the loss function is an artificially set constant used for adjusting the weight of the loss term.
In step S3, the carrier image C and the secret image S are input to the trained color image steganography model, and a carrier image C 'and a reconstructed secret image S' are generated. As can be seen from fig. 7, it is difficult for the human eye to distinguish the difference between the generated secret image and the carrier image, and the reconstructed secret image and the secret image, and the invisibility of the steganographic method is good. Meanwhile, the color image steganography model is based on frequency component selection, and the steganography capacity is improved to 8bpp (namely Pixel depth Bits Per Pixel) in a steganography mode of a map.
Based on the color image steganography method based on frequency component selection, the invention also provides a color image steganography device based on frequency component selection. The device comprises a model building module, a model training module and an image steganography module. The model construction module is used for constructing a color image steganography model selected on the basis of the frequency domain components; the model training module is used for training a color image steganography model; the image steganography module is used for inputting the color carrier image and the gray secret image into the trained color image steganography model and generating a secret carrier image and a reconstructed secret image;
the color image steganography model comprises a frequency component selection module, a characteristic data extraction module, a secret image generation module and a secret image reconstruction module.
The frequency component selection is used to receive the carrier image C and generate a single-channel diagonal high-frequency component of the carrier image C as first input data of the secret image generation module. The frequency component selection module includes a channel separation unit, a first frequency domain conversion unit, and a first frequency component selection unit. The channel separation unit is used for separating single-channel information of the carrier image C. The first frequency domain conversion unit is configured to convert the single-channel information of the carrier image C into frequency domain information of the carrier image C using a DWT transform. The first frequency component selection unit is used for selecting a single-channel diagonal high-frequency component of the carrier image C as first input data of the carrier image generation module.
The characteristic data extraction module is used for receiving the secret image S and extracting the characteristic data X of the secret image S as second input data of the secret image generation module. The feature data extraction module comprises a second frequency domain conversion unit and a preprocessing network. The second frequency domain conversion unit is used for converting the spatial domain information of the secret image S into the frequency domain information of the secret image S by adopting DWT (discrete wavelet transform) conversion and transmitting the frequency domain information of the secret image S to the preprocessing network; the preprocessing network is used for receiving the frequency domain information of the secret image S and extracting the characteristic data X of the secret image S as second input data.
The secret image generation module comprises an encoding network and a first spatial domain conversion unit. The encoding network is used for receiving the first input data and the second input data, hiding the second input data into the first input data and generating a single-channel diagonal high-frequency component of the secret image C'; the first spatial domain conversion unit is used for receiving the single-channel diagonal high-frequency components of the secret-carrying image C 'and generating the secret-carrying image C' by adopting DWT inverse transformation.
The secret image reconstruction module includes a second frequency component selection unit, a decoding network, and a second spatial-domain conversion unit. The second frequency component selection unit is used for receiving the secret image C ' and generating a single-channel diagonal high-frequency component C ' of the secret image C 'HH(ii) a The decoding network is used for receiving the single-channel diagonal high-frequency component C 'of the secret image C'HHGenerating spatial domain information of the reconstructed secret image S'; a second spatial domain conversion unit for receiving the reconstructionAnd generating the reconstructed secret image S 'by using the spatial domain information of the secret image S' and adopting DWT inverse transformation.
Please refer to fig. 8, which is a steganography effect diagram of the color image steganography model provided in this embodiment with the B channel of the carrier image as the embedded domain. In other embodiments, the R channel or the G channel of the carrier image may also be employed as the embedded domain of the secret image. Please refer to table 1, which shows the comparison of the embedding results of the steganographic model provided by the present invention based on different channels of the RGB format carrier image. As can be seen from table 1, when any channel of RGB format is used as the embedded domain of the secret image, the quality of the steganographic result is maintained at the same level, and when the channel B is used as the embedded domain, the Peak Signal to Noise Ratio (PSNR) and the Structural Similarity (SSIM) of the generated secret image C 'and the reconstructed secret image S' are the best as a whole.
Table 1. comparison of the results of embedding of the steganographic model of the present invention based on different channels of the RGB format carrier image.
Referring to table 2, compared with the two steganography methods of the anti model and ISGAN in the prior art, the secret-carrying image C 'and the reconstructed secret image S' generated by the color image steganography model provided by the present invention have higher PNSR and SSIM as a whole, which shows that the secret-carrying image C 'and the reconstructed secret image S' with more similar structures and better image quality can be generated by the color image steganography model provided by the present invention.
TABLE 2 comparison of results for steganography methods of the present invention with steganography methods of the same type
In summary, compared with the prior art, the color image steganography method and device based on frequency component selection provided by the invention combine the image frequency domain processing technology and the coding-decoding network in the deep learning technology, so that a steganography mode of a hidden image is realized, and the steganography capacity is remarkably improved; meanwhile, the steganography method can generate a secret-carrying image and a reconstructed secret image which are similar in structure and good in image quality, and the imperceptibility of the steganography method and the safety of information transmission are guaranteed.
The above-mentioned embodiments only express several embodiments of the present invention, and the description thereof is more specific and detailed, but not construed as limiting the scope of the invention. It should be noted that, for a person skilled in the art, several variations and modifications can be made without departing from the inventive concept, which falls within the scope of the present invention.
Claims (10)
1. A color image steganography method based on frequency domain component selection is characterized by comprising the following steps:
constructing a color image steganography model selected based on the frequency domain components;
training the color image steganography model;
inputting the color carrier image and the gray secret image into the trained color image steganography model to generate a carrier image and a reconstructed secret image to complete image steganography;
the color image steganography model comprises a frequency component selection module, a characteristic data extraction module, a secret-carrying image generation module and a secret image reconstruction module; the frequency component selection module is used for receiving the carrier image and generating a single-channel diagonal high-frequency component of the carrier image as first input data; the characteristic data extraction module is used for receiving the secret image and extracting the characteristic data of the secret image as second input data; the secret-carrying image generation module is used for receiving the first input data and the second input data and generating the secret-carrying image; the secret image reconstruction module is used for receiving the secret image and generating the reconstructed secret image.
2. A color image steganography method based on frequency domain component selection as recited in claim 1, wherein:
the frequency component selection module comprises a channel separation unit, a first frequency domain conversion unit and a first frequency component selection unit;
the channel separation unit is used for separating single-channel information of the carrier image;
the first frequency domain conversion unit is used for converting single-channel information of the carrier image into frequency domain information of the carrier image by adopting discrete wavelet transform;
the first frequency component selection unit is configured to select a single-channel diagonal high-frequency component of the carrier image as the first input data.
3. A color image steganography method based on frequency domain component selection as recited in claim 1, wherein:
the characteristic data extraction module comprises a second frequency domain conversion unit and a preprocessing network;
the second frequency domain conversion unit is used for converting the spatial domain information of the secret image into the frequency domain information of the secret image by adopting discrete wavelet transform, and transmitting the frequency domain information of the secret image to the preprocessing network;
the preprocessing network is used for receiving the frequency domain information of the secret image and extracting the characteristic data of the secret image as the second input data.
4. A color image steganography method based on frequency domain component selection as recited in claim 3, wherein:
the preprocessing network is sequentially stacked with a first input layer, a first convolution group, a first connection layer, a second convolution group and a second connection layer; wherein the first convolution group and the second convolution group each include a plurality of convolution layers.
5. A color image steganography method based on frequency domain component selection as recited in claim 1, wherein:
the secret-carrying image generation module comprises an encoding network and a first spatial domain conversion unit;
the coding network is used for receiving the first input data and the second input data, hiding the second input data into the first input data and generating a single-channel diagonal high-frequency component of the secret-carrying image;
the first spatial domain conversion unit is used for receiving a single-channel diagonal high-frequency component of the secret-carrying image and generating the secret-carrying image by adopting inverse discrete wavelet transform.
6. A color image steganography method based on frequency domain component selection as recited in claim 5, wherein:
the coding network is sequentially stacked with a second input layer, a third connection layer, a third convolution group, a fourth connection layer and a first output layer; wherein the third convolution group includes a plurality of convolution layers.
7. A color image steganography method based on frequency domain component selection as recited in claim 5, wherein:
the secret image reconstruction module comprises a second frequency component selection unit, a decoding network and a second spatial domain conversion unit;
the second frequency component selection unit is used for receiving the secret-carrying image and generating a single-channel diagonal high-frequency component of the secret-carrying image;
the decoding network is used for receiving a single-channel diagonal high-frequency component of the secret image and generating frequency domain information of the reconstructed secret image;
the second spatial domain conversion unit is used for receiving the frequency domain information of the reconstructed secret image and generating the reconstructed secret image by adopting inverse discrete wavelet transform.
8. A color image steganography method based on frequency domain component selection as recited in claim 7, wherein:
the decoding network is sequentially stacked with a third input layer, a fourth convolution group, a fifth connection layer and a second output layer; wherein the fourth convolution group includes a plurality of convolution layers.
9. The color image steganography method based on frequency domain component selection as claimed in claim 1, wherein the training process of the color image steganography model comprises the following steps:
dividing an LFW data set into a training set, a verification set and a test set at a ratio of 10:1:1 at random; dividing the training set, the verification set and the test set into a carrier image set and a secret image set according to the proportion of 1:1 respectively; uniformly carrying out gray processing on the images in the secret image set;
training the color image steganography model by using the training set, and iteratively updating parameters until the color image steganography model converges; in the iteration process, loss calculation is carried out by using a mean square error function;
verifying the color image steganography model in training using the verification set every 20 iterations;
and after the color image steganography model converges, testing the performance of the color image steganography model by using the test set.
10. A color image steganography apparatus that selects based on frequency domain components, comprising:
the model building module is used for building a color image steganography model selected on the basis of the frequency domain components;
the model training module is used for training the color image steganography model;
the image steganography module is used for inputting the color carrier image and the gray secret image into the trained color image steganography model and generating a secret carrier image and a reconstructed secret image;
the color image steganography model comprises a frequency component selection module, a characteristic data extraction module, a secret-carrying image generation module and a secret image reconstruction module; the frequency component selection module is used for receiving the carrier image and generating a single-channel diagonal high-frequency component of the carrier image as first input data; the characteristic data extraction module is used for receiving the secret image and extracting the characteristic data of the secret image as second input data; the secret-carrying image generation module is used for receiving the first input data and the second input data and generating the secret-carrying image; the secret image reconstruction module is used for receiving the secret image and generating the reconstructed secret image.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202210005481.XA CN114338945A (en) | 2022-01-04 | 2022-01-04 | Color image steganography method and device based on frequency domain component selection |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202210005481.XA CN114338945A (en) | 2022-01-04 | 2022-01-04 | Color image steganography method and device based on frequency domain component selection |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN114338945A true CN114338945A (en) | 2022-04-12 |
Family
ID=81024179
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202210005481.XA Pending CN114338945A (en) | 2022-01-04 | 2022-01-04 | Color image steganography method and device based on frequency domain component selection |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN114338945A (en) |
Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20110194727A1 (en) * | 2010-02-11 | 2011-08-11 | National Taiwan University Of Science & Technology | Image data processig systems for hiding secret information and data hiding methods using the same |
| CN111028308A (en) * | 2019-11-19 | 2020-04-17 | 珠海涵辰科技有限公司 | Steganography and reading method for information in image |
| CN111311473A (en) * | 2020-01-21 | 2020-06-19 | 宁波大学 | Digital image steganography method and secret information extraction method |
| CN112465687A (en) * | 2020-11-17 | 2021-03-09 | 北京航空航天大学 | Image processing method and device |
| CN112597509A (en) * | 2020-12-03 | 2021-04-02 | 华南师范大学 | Information hiding method and system fusing wavelet and self-encoder |
| CN113077377A (en) * | 2021-05-13 | 2021-07-06 | 海南大学 | Color image steganography method based on generation countermeasure network |
-
2022
- 2022-01-04 CN CN202210005481.XA patent/CN114338945A/en active Pending
Patent Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20110194727A1 (en) * | 2010-02-11 | 2011-08-11 | National Taiwan University Of Science & Technology | Image data processig systems for hiding secret information and data hiding methods using the same |
| CN111028308A (en) * | 2019-11-19 | 2020-04-17 | 珠海涵辰科技有限公司 | Steganography and reading method for information in image |
| CN111311473A (en) * | 2020-01-21 | 2020-06-19 | 宁波大学 | Digital image steganography method and secret information extraction method |
| CN112465687A (en) * | 2020-11-17 | 2021-03-09 | 北京航空航天大学 | Image processing method and device |
| CN112597509A (en) * | 2020-12-03 | 2021-04-02 | 华南师范大学 | Information hiding method and system fusing wavelet and self-encoder |
| CN113077377A (en) * | 2021-05-13 | 2021-07-06 | 海南大学 | Color image steganography method based on generation countermeasure network |
Non-Patent Citations (2)
| Title |
|---|
| HAI SU等: "A Color Image Steganography Based on Frequency Sub-band Selection" * |
| 李庆忠;于琛;褚东升;: "一种DCT域稳健的彩色图像隐藏方法" * |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Huan et al. | Exploring stable coefficients on joint sub-bands for robust video watermarking in DT CWT domain | |
| Lu et al. | Bit-level optimized neural network for multi-antenna channel quantization | |
| CN111640444B (en) | CNN-based adaptive audio steganography method and secret information extraction method | |
| CN110738605A (en) | Image denoising method, system, device and medium based on transfer learning | |
| CN112634117B (en) | End-to-end JPEG domain image steganography method based on generation of countermeasure network | |
| CN110312138B (en) | High-embedding-capacity video steganography method and system based on time sequence residual convolution modeling | |
| CN112132737B (en) | Image robust steganography method without reference generation | |
| CN115131188A (en) | Robust image watermarking method based on generation countermeasure network | |
| CN112597509B (en) | Information hiding method and system integrating wavelet and self-encoder | |
| CN115272044A (en) | Depth image watermarking method based on attention of mixed frequency domain channel | |
| CN111292221A (en) | Safe and robust high-capacity image steganography method | |
| CN113362225A (en) | Multi-description compressed image enhancement method based on residual recursive compensation and feature fusion | |
| Wu et al. | DAC-Mobi: Data-assisted communications of mobile images with cloud computing support | |
| CN112308754A (en) | Multi-decoding-end parallel extraction mechanism design for high-capacity robust watermark application | |
| Xuan et al. | High capacity lossless data hiding based on integer wavelet transform | |
| CN116452401A (en) | Reversible robust watermark embedding and extraction model construction method for resisting image attack | |
| CN117078517A (en) | Image super-resolution steganography method based on reversible neural network | |
| CN114157773B (en) | Image steganography method based on convolutional neural network and frequency domain attention | |
| CN108521535B (en) | A kind of Information hiding transmission method based on image blend processing | |
| CN114338945A (en) | Color image steganography method and device based on frequency domain component selection | |
| Malik et al. | A high capacity data hiding scheme using modified AMBTC compression technique. | |
| Yang | Robust Watermarking Scheme Based on Radius Weight Mean and Feature‐Embedding Technique | |
| CN116029887A (en) | Image high-capacity robust watermarking method based on wavelet neural network | |
| CN114529442A (en) | Robust image watermarking method adopting two-stage precoding and wavelet network | |
| Zhong et al. | Enhanced Attention Mechanism-Based Image Watermarking With Simulated JPEG Compression |
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 |