CN111681154B - Color image steganography distortion function design method based on generation countermeasure network - Google Patents

Color image steganography distortion function design method based on generation countermeasure network Download PDF

Info

Publication number
CN111681154B
CN111681154B CN202010517997.3A CN202010517997A CN111681154B CN 111681154 B CN111681154 B CN 111681154B CN 202010517997 A CN202010517997 A CN 202010517997A CN 111681154 B CN111681154 B CN 111681154B
Authority
CN
China
Prior art keywords
color image
modification
color
matrix
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.)
Active
Application number
CN202010517997.3A
Other languages
Chinese (zh)
Other versions
CN111681154A (en
Inventor
廖鑫
唐志强
胡娟
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Hunan University
Original Assignee
Hunan University
Priority date (The priority date 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 date listed.)
Filing date
Publication date
Application filed by Hunan University filed Critical Hunan University
Priority to CN202010517997.3A priority Critical patent/CN111681154B/en
Publication of CN111681154A publication Critical patent/CN111681154A/en
Application granted granted Critical
Publication of CN111681154B publication Critical patent/CN111681154B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

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
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/047—Probabilistic or stochastic 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
    • H—ELECTRICITY
    • H04—ELECTRIC COMMUNICATION TECHNIQUE
    • H04N—PICTORIAL COMMUNICATION, e.g. TELEVISION
    • H04N1/00—Scanning, transmission or reproduction of documents or the like, e.g. facsimile transmission; Details thereof
    • H04N1/32—Circuits or arrangements for control or supervision between transmitter and receiver or between image input and image output device, e.g. between a still-image camera and its memory or between a still-image camera and a printer device
    • H04N1/32101—Display, printing, storage or transmission of additional information, e.g. ID code, date and time or title
    • H04N1/32144—Display, printing, storage or transmission of additional information, e.g. ID code, date and time or title embedded in the image data, i.e. enclosed or integrated in the image, e.g. watermark, super-imposed logo or stamp
    • H04N1/32149—Methods relating to embedding, encoding, decoding, detection or retrieval operations
    • H04N1/32267—Methods relating to embedding, encoding, decoding, detection or retrieval operations combined with processing of the image
    • H04N1/32272—Encryption or ciphering

Landscapes

  • Engineering & Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • Theoretical Computer Science (AREA)
  • General Physics & Mathematics (AREA)
  • General Health & Medical Sciences (AREA)
  • General Engineering & Computer Science (AREA)
  • Biophysics (AREA)
  • Computational Linguistics (AREA)
  • Data Mining & Analysis (AREA)
  • Evolutionary Computation (AREA)
  • Artificial Intelligence (AREA)
  • Molecular Biology (AREA)
  • Computing Systems (AREA)
  • Biomedical Technology (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Mathematical Physics (AREA)
  • Software Systems (AREA)
  • Health & Medical Sciences (AREA)
  • Probability & Statistics with Applications (AREA)
  • Multimedia (AREA)
  • Signal Processing (AREA)
  • Image Processing (AREA)
  • Editing Of Facsimile Originals (AREA)

Abstract

The invention relates to a color image steganography distortion function design method based on a generation countermeasure network, which is used for color image steganography communication. The invention mainly comprises the following steps: (1) a generator network structure for generating color image modification probability is provided; (2) a color image steganography distortion function framework is designed based on a generation countermeasure network. Compared with the prior art, the color image steganography distortion function design method based on the generation countermeasure network, provided by the invention, is used for carrying out secret information steganography transmission on the airspace color image with wider application. The method is feasible and effective, and the steganography distortion function generated by the trained model can better resist the existing color image steganography analyzer by combining with the coding technologies such as the Syndrome-Trellis Code and the like, so that the method has higher safety.

Description

Color image steganography distortion function design method based on generation countermeasure network
Technical Field
The invention relates to the technical field of multimedia security, in particular to a color image steganography distortion function design method based on a generation countermeasure network.
Background
The rapid development and application of internet technology enable people to transmit information on the internet more efficiently and more quickly, and meanwhile, the security of network information is more and more tested. The privacy information such as personal identity, password and the like is revealed, and the confidentiality in the fields of national politics, military, commerce and the like is monitored and stolen by hostile force, so that the security transmission of the information is guaranteed to be crucial to the healthy development of the network. Although the cryptographic technology can ensure the security of information, the encrypted information usually looks disordered and easily attracts the attention of an attacker, so that the possibility of being cracked is increased, and in addition, even if the attacker cannot crack the encrypted information, the information can be blocked from being spread, so that the communication fails. The information hiding technology adopts a concept different from cryptography, and the key point of the information hiding technology is that an attacker cannot crack the information, but the secret information is skillfully hidden and is not discovered by the attacker. Since human perception organs have a redundant characteristic in the perception of multimedia information, information hiding technology uses multimedia information as a carrier of secret information. After an attacker intercepts the embedded secret carrier, the attacker is difficult to perceive the existence of the secret information, and is more difficult to extract the secret information from the embedded secret carrier.
The traditional steganography technology needs to select a proper coding mode and design a minimized distortion function. Due to the great breakthrough of coding techniques, the focus of recent research is to design a suitable distortion minimization function. Steganalysis is a detection technology for steganography, and detects whether multimedia information carries secret information or not by carrying out statistical analysis on the multimedia information. The information hiding technique and the steganalysis technique compete with each other and promote each other, which is very similar to the generation of the competition model. Documents "w.tang, s.tan, b.li, and j.huang.automatic hierarchical rendering using a generic additive network.ieee Signal Processing Letters, vol.24, No.10, pp.1547-1551, oct.2017" first propose designing a gray picture steganography distortion function frame based on a generated countermeasure network, obtaining positive and negative modification probabilities and matrices of a gray picture through a generator network, obtaining a modification matrix after the positive and negative modification probabilities and matrices are simulated and embedded through a trained sub-network, obtaining a gray scale secret image by adding the gray scale image and the modification matrix, and training the carrier image and the secret image together as input of a steganography analyzer. The document "j.yang, d.ruan, j.huang, x.kang, and y.shi.an Embedding code Learning frame Using gan.ieee Transactions on Information requirements and Security, 2019" designs a double-tank function for analog Embedding, and uses the unet network as a steganography distortion function of the generator design airspace grayscale image, and the performance exceeds the existing method for manually designing the steganography distortion function. The document "Jianhua Yang, Danyan Ruan, Xiangui Kang, Yun-Qing Shi. Towards Automatic Embedding Cost Learning for JPEG Steganographic. IH & MMSec,2019: 37-46" designs an inverse discrete cosine transform module, successfully using to generate steganographic distortion functions against a network design frequency domain grayscale image.
In real life, color images are more widely applied, however, the current schemes for generating the anti-network-design steganography distortion function are designed for gray images, and compared with gray images, color image channels have correlation. In consideration of practical application problems, the invention aims to provide a steganographic distortion function design method based on a generation countermeasure network for a specific carrier of a color image.
Disclosure of Invention
The invention provides a color image steganography distortion function design method based on a generation countermeasure network, which is used for color image steganography communication and mainly comprises two contents:
(1) a generator network structure for generating color image modification probability is provided;
(2) a color image steganography distortion function framework is designed based on a generation countermeasure network.
The specific contents are as follows:
(1) a generator network structure for generating color image modification probabilities is proposed: the network structure is shown in fig. 1, a generator adopts two unet sub-network structures, a first unet sub-network inputs a carrier picture to generate positive and negative modification probabilities and a matrix, a second unet sub-network generates a positive modification probability ratio matrix according to the input carrier picture and the output of the first unet sub-network, a negative modification probability ratio matrix is obtained by subtracting the positive modification probability ratio matrix from the negative modification probability ratio matrix through 1, and then the negative modification probability ratio matrix and the positive modification probability ratio matrix are multiplied to respectively obtain the positive modification probability matrix and the negative modification probability matrix. After the secret information is embedded into the carrier image, part of the pixel values are modified, and the pixel values at different positions of the carrier image are different from the original image after modification, wherein the difference is called distortion. Generally, the distortion generated by modifying the image in the place where the texture of the carrier image is complicated is smaller than that of modifying the image in the place where the texture is smooth. The steganographic distortion function is used for modeling and quantifying the difference size of the modified carrier image, and has a negative correlation relation with the modification probability, namely the smaller the distortion value is, the larger the modification probability value is. Because the modification probability value is used in the simulation embedding stage, the generator directly generates the modification probability instead of the distortion value, and the modification probability is converted into the distortion value when the network is actually used after training is completed.
Figure BDA0002530893890000021
Wherein
Figure BDA0002530893890000022
Representing the probability of a forward modification of the k-th channel at a color image point (i, j),
Figure BDA0002530893890000023
representing the probability of negative modification of the k-th channel at the color image point (i, j),
Figure BDA0002530893890000024
representing the probability that the k-th channel at the color image point (i, j) is not modified,
Figure BDA0002530893890000025
a distortion value representing a forward modification of the k-th channel at a color image point (i, j),
Figure BDA0002530893890000026
a distortion value representing the negative modification of the k-th channel at a color image point (i, j).
The color image comprises three channels of red, green and blue, correlation exists among the channels, the content-based adaptive algorithm can more select to embed information in a place with high texture complexity, and due to the similarity of the three channels of the color image, if the three channels are simply taken as three gray images to carry out secret information embedding, the positions of the three channels of the same pixel tend to be modified. The existing gray level picture generates a steganography distortion function frame, the positive modification probability and the negative modification probability of each position are simply set to be equal, the modification directions of three channel positions of the same pixel are completely random, no correlation exists, and the correlation among color image channels is easier to destroy. In the designed generator network structure, a new unet sub-network is added to learn the adjustment of the positive and negative modification probability, and the sub-network is used for learning the distribution of the positive and negative modification probability, so that the modification at the positions of three channels of the color image conforms to a certain special rule, and the damage to the correlation among the color image channels is reduced.
(2) A color image steganography distortion function framework is designed based on a generation countermeasure network, and comprises the following steps: the overall framework is as shown in fig. 3, the color carrier picture obtains a positive modification probability matrix and a negative modification probability matrix through a generator (fig. 1), the positive modification probability matrix and the negative modification probability matrix pass through a simulation embedding module to obtain a modification matrix, the color image and the modification matrix are added to obtain a color steganography image, the carrier image and the steganography image are used as the input of a color steganography analyzer (fig. 2) for training, and in the training of continuous confrontation with the generator, the generator is prompted to continuously learn and design a color image steganography distortion function.
The document "g.xu, h.wu, and y.shi.structural design for hierarchical analysis. ieee Signal Processing Letters, vol.23, No.5, pp.708-712,2016" is a classical grayscale image steganography analysis network that needs to be modified accordingly in order for the generator to be able to take channel correlation into account during the process of fighting with steganography analyzers. The network structure of the improved color image steganography analyzer is shown in fig. 2, and in the preprocessing stage, in order to better maintain the signal-to-noise ratio, the color image is firstly divided into three gray images which are respectively convoluted with 30 fixed SRM filtering kernels
1st＝[-1 +1] (2)
2nd＝[+1 -2 +1] (3)
3rd＝[+1 -3 +3 -1] (4)
Figure BDA0002530893890000031
Figure BDA0002530893890000032
Figure BDA0002530893890000033
Figure BDA0002530893890000034
Wherein, 8, 4 and 8 different filter kernels can be obtained by rotating the '1 st', '2 nd' and '3 rd' by 45 degrees respectively, and 4 different filter kernels can be obtained by rotating the 'EDGE 3X 3' and 'EDGE 5X 5' by 90 degrees respectively.
After convolution, 3 feature maps with the channel number of 30 are obtained, and then a feature map with the channel number of 90 is synthesized. The pooling layer may reduce the number of calculations while preserving the main properties, preventing overfitting, but at the same time the pooling layer may lose some information. In order to retain the advantages of the pooling layer and reduce the loss of effective information, the color image steganalysis network structure is not simply a pooling layer, but is replaced by a b structure shown in fig. 2, and the structure is divided into two parallel branches which are respectively added and converged after passing through the pooling layer of 5 × 5 and the convolution kernel of 5 × 5.
According to the existing work, the simulation embedding can be used for replacing the Code embedding of the real Syndrome-Trellis Code and the like to generate the secret-carrying image. Given the probability of forward modification
Figure BDA0002530893890000041
Probability of negative modification
Figure BDA0002530893890000042
And a random number ni,j,kCorrespondingly modify the value mi,j,kIs composed of
Figure BDA0002530893890000043
Since the function is a step function and cannot propagate the gradient in the opposite direction, a truncated linear function is proposed to replace the step function
Figure BDA0002530893890000044
Wherein the function f (x) shows that ax is limited to the range of [ -1,1], a is a hyperparameter and needs to be tried in experiments, and generally, a is taken as 100. The simulation function is linear and has low computational complexity.
Loss function for color steganalyser
Figure BDA0002530893890000045
Wherein y is1And y2Is the output of the discriminator D after the Softmax activation function, y1'and y'2Corresponding to the true tag value.
The loss function of the generator comprises two parts, the first part is used for competing with the discriminator and is set as the inverse number of the loss function of the discriminator
Figure BDA0002530893890000046
The second part is used for controlling the embedding density capacity, the textures of the three channels of the color image are different, the capacities suitable for embedding the color image are different, the color image is not suitable for being embedded into the same capacity as three gray-scale images simply, therefore, the total capacity is controlled to be constant by summing the embedding capacities of the three channels of the color image, and the generator can learn to allocate the suitable capacities to the different channels in the counterstudy process. Color image embedding density capacity calculation (bit unit)
Figure BDA0002530893890000047
Figure BDA0002530893890000051
Wherein
Figure BDA0002530893890000052
Representing the probability of a forward modification of the k-th channel at a color image point (i, j),
Figure BDA0002530893890000053
representing the probability of negative modification of the k-th channel at the color image point (i, j),
Figure BDA0002530893890000054
indicates the probability, cap, that the k-th channel at the color image point (i, j) is not modifiedkRepresenting the embeddability of the color image channel k.
The second partial loss function of the generator is
Figure BDA0002530893890000055
Where H denotes the height of the color image, W denotes the width of the color image, and q denotes the set embedding ratio (bpp). The generator will let in the training process
Figure BDA0002530893890000056
The embedding capacity is reduced so as to approach the expected value H multiplied by W multiplied by q. Since the sum of three channels is controlled, the specific embedding capacity cap of each channelkThere are no constraints, so the generator can automatically learn how to perform the inter-channel capacity allocation in the course of learning against the steganalyser.
The generator needs to take the tasks of confrontation with the steganographic analyzer and certain embedding capacity limitation into consideration, so that two parts of loss functions are added according to weight
Figure BDA0002530893890000057
Where α and β are hyper-parameters, need to be tried in experiments.
Compared with the prior art, the technical scheme at least has the following beneficial effects:
1. the invention provides a color image steganography distortion function design method based on a generation countermeasure network. The generator can learn to make the modification directions at the three channel positions of the color image accord with a certain special rule by counterstudy with the steganalyser, thereby reducing the damage to the correlation among the color image channels. Compared with the existing CMDC of the traditional color image steganography distortion function design scheme, the CMDC is designed artificially, in the scheme, the work of distributing and adjusting the positive distortion value and the negative distortion value is handed to the network, and better design is automatically learned from mass data by continuously carrying out countertraining with a steganography analyzer.
2. The invention provides a color image steganography distortion function design method based on a generation countermeasure network, which is characterized in that the total steganography capacity of three channels of a color image is controlled in a loss function of a generator, and the capacity distribution of the three channels of the color image is automatically learned through the network for the first time. Because the textures of the three channels of the color image are different and the capacities suitable for embedding are also different, compared with an equal distribution scheme of the three channels of the color image, the method can better resist the steganalysis method of the color image. The proposed linear simulation embedding function (formula 2) not only can well fit the original step function (formula 1), but also has the advantage of small calculation amount.
Drawings
FIG. 1 is a block diagram of an overall framework of a generator for generating color image modification probabilities in accordance with the present invention;
FIG. 2 is a schematic diagram of a network structure for steganalysis of color images according to the present invention;
FIG. 3 is a schematic diagram of a color image steganography distortion function framework designed based on a generation countermeasure network according to the present invention;
FIG. 4 is a schematic diagram of a generator network generating positive and negative modification probability matrices according to an embodiment of the present invention;
FIG. 5 is a flow chart of the present invention for performing covert communications;
Detailed Description
The invention relates to a color image steganography distortion function design method based on a generation countermeasure network. The following describes the specific embodiment of the present invention by way of example with the embedding rate q being 1.2bpp, which is the classic Syndrome-trellis code encoding. It should be understood by those skilled in the art that these embodiments are only for explaining the technical principle of the present invention and are not intended to limit the application scope of the present invention.
The development language is python, the deep learning framework is tensorflow, and the specific steps are as follows:
step 1: a training data set is prepared. 40000 color pictures were obtained from the image database and the center of the picture was cut to 256 × 256 pictures to fit the size of the web input.
Step 2: and (5) building a network structure. The network is constructed as shown in fig. 3, a positive modification probability matrix and a negative modification probability matrix are obtained from a color carrier picture through a generator (fig. 1) with a double-unit structure, the positive modification probability matrix and the negative modification probability matrix are subjected to analog embedding through a formula (10) to obtain a modification matrix, a color secret-carrying image is obtained by adding the color image and the modification matrix, and the carrier image and the secret-carrying image are used as the input of a color image steganography analyzer (fig. 2).
And step 3: and (5) network training. Setting the learning rates of the generator and the steganalyser to be 0.0001, training by adopting an Adam optimizer, firstly fixing the parameters of the generator during training, updating the parameters of the steganalyser through a loss function of an Adam optimizer optimization formula (11), then fixing the parameters of steganalysis, and updating the parameters of the generator through a loss function of an Adam optimizer optimization formula (16), wherein H in a formula (15) is 256, W is 256, and q is 1.2. And alternately updating the parameters of the steganalyser and the generator once every iteration, storing the training parameters once every 10000 times of iteration, and finishing the training when the iteration is 120000 times.
And 4, step 4: and obtaining a color image distortion value matrix and the respective embedding capacity of the three channels. For a color picture, it is first cut into 256 × 256 pictures. In this embodiment, a color image is input into a trained generator network to output a positive and negative modified probability matrix, and the modified probability matrix is converted into a positive and negative distortion value matrix through a formula (1). The capacity of each of the three channels to embed secret information is calculated by equation (14). As shown in fig. 4, the three channel allocation capacities of the color image are 0.74bpp, 0.20bpp, and 0.26bpp, respectively, and the generator can automatically learn how to perform the inter-channel capacity allocation in the process of counterlearning with the steganalyser.
And 5: and carrying out secret communication. As shown in fig. 5, in order to further ensure the security of information, the information to be transmitted is encrypted before being embedded into the color carrier image by using the Syndrome-Trellis Code, and then the encrypted secret information is embedded into the three channels of the color carrier image by using the Syndrome-Trellis Code in combination with the positive and negative distortion value matrices and the embedding capacity of each of the three channels to obtain the secret-carrying image. The sender transmits the secret-carrying picture to the receiver through the network medium, the receiver receives the picture and then decodes the picture by using the Syndrome-Trellis Code to obtain the encrypted secret information, and then decrypts the encrypted information to finally complete one-time communication.
In summary, for a special carrier of color images, the invention designs a color image steganography distortion function framework based on generation countermeasure network design. For color pictures from different sources, the method has robustness and can generate a steganographic distortion function of the color pictures.
It will be appreciated by persons skilled in the art that the scope of the present invention is not limited to the specific embodiments described. Equivalent changes or substitutions of related technical features can be made by those skilled in the art without departing from the principle of the invention, and it is noted that the technical solutions after the changes or substitutions will fall within the protection scope of the invention.

Claims (3)

1. A method for designing a steganography distortion function of a color image based on a generation countermeasure network, the method comprising:
the device comprises a generator, an analog embedding module, a color steganalyser and a loss function; the color carrier image is processed by a generator to obtain a positive modification probability matrix and a negative modification probability matrix; the positive and negative modification probability matrix is subjected to a simulation embedding module to obtain a modification matrix; adding the color carrier image and the modification matrix to obtain a color secret-carrying image, and inputting the carrier image and the secret-carrying image into a color steganography analyzer to detect whether the secret-carrying image carries secret information; under the constraint control of a designed loss function, alternately training a generator and a color steganalyser; after the network training is finished, converting the modification probability into a distortion value through a mutual conversion formula of the failure value and the modification probability;
the generator adopts a double UNet structure; inputting a carrier picture into a first UNet sub-network to generate positive and negative modification probability and a matrix; the second UNet sub-network generates a forward modification probability ratio matrix according to the input carrier picture and the output of the first UNet sub-network; the negative modification probability ratio matrix is obtained by subtracting the positive modification probability ratio matrix from 1; finally, multiplying the positive and negative modification probability matrix by the matrix to respectively obtain a positive modification probability matrix and a negative modification probability matrix;
aiming at the carrier characteristics of the color image, the color steganalyser is improved on the basis of a classical gray-scale image steganalyser; the improvement is divided into the following two points, firstly, an input color image is divided into three gray level images, the three gray level images are respectively convolved with 30 fixed SRM filtering kernels, 3 characteristic graphs with the channel number of 30 are obtained after convolution, and a characteristic graph with the channel number of 90 is synthesized; and designing an improved pooling layer structure in the color image steganalysis network, wherein the improved pooling layer structure comprises 5-by-5 pooling layers and 5-by-5 convolution kernels which are connected in parallel, and adding and merging operations.
2. The method for designing a color image steganography distortion function based on a generative countermeasure network as claimed in claim 1, wherein the simulation embedding module specifically comprises:
simulating real Syndrome-Trellis Code embedding by using an embedding function, and constructing a truncated linear function to replace a commonly used step function to finish the back propagation of the gradient when simulating the embedding, wherein the truncated linear function is
Figure FDA0003633886400000011
Wherein the f (x) function represents limiting ax to [ -1,1 [ ]]Range, a is a hyperparameter, requiring trial and error in experiments, mi,j,kRepresenting the modification value, n, corresponding to the k-th channel at a color image point (i, j)i,j,kIs a random number, and is a random number,
Figure FDA0003633886400000012
is the probability of a forward modification,
Figure FDA0003633886400000013
is the probability of a negative modification.
3. The method for designing a color image steganography distortion function based on a generative countermeasure network as claimed in claim 1, wherein the loss function specifically comprises:
color steganalysis loss and generation loss; the color steganalysis loss adopts the commonly used binary cross entropy in the binary problem, and the loss function is
Figure FDA0003633886400000014
Wherein y is1And y2Is the output of the discriminator D after the Softmax activation function, y1'and y'2Corresponding to the true tag value; the loss function of the generator G comprises two parts, the first part being intended to compete with the arbiter, set it as the inverse of the arbiter loss function,
Figure FDA0003633886400000015
the second part is used for controlling the embedding densityCapacity, controlled by calculating the square of the difference between the sum of the steganographic capacities of the three channels of the color image and the desired value, with a loss function of
Figure FDA0003633886400000021
Wherein capacity is the embedding density capacity of the color image, H denotes the height of the color image, W denotes the width of the color image, and q denotes the set embedding density; the two-part loss function is weighted and summed,
Figure FDA0003633886400000022
where α and β are hyper-parameters, need to be tried in experiments.
CN202010517997.3A 2020-06-09 2020-06-09 Color image steganography distortion function design method based on generation countermeasure network Active CN111681154B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202010517997.3A CN111681154B (en) 2020-06-09 2020-06-09 Color image steganography distortion function design method based on generation countermeasure network

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202010517997.3A CN111681154B (en) 2020-06-09 2020-06-09 Color image steganography distortion function design method based on generation countermeasure network

Publications (2)

Publication Number Publication Date
CN111681154A CN111681154A (en) 2020-09-18
CN111681154B true CN111681154B (en) 2022-06-21

Family

ID=72454174

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202010517997.3A Active CN111681154B (en) 2020-06-09 2020-06-09 Color image steganography distortion function design method based on generation countermeasure network

Country Status (1)

Country Link
CN (1) CN111681154B (en)

Families Citing this family (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN112199693A (en) * 2020-09-30 2021-01-08 东南数字经济发展研究院 Steganography method based on cartoon image
CN112634118B (en) * 2020-12-24 2022-09-30 中国科学技术大学 Anti-batch steganography method
CN112767226B (en) * 2021-01-15 2023-09-12 南京信息工程大学 Image steganography method and system based on automatic learning distortion of GAN network structure
CN113076549B (en) * 2021-04-08 2023-05-12 上海电力大学 Novel U-Net structure generator-based countermeasures network image steganography method
CN115348360B (en) * 2022-08-11 2023-11-07 国家电网有限公司大数据中心 GAN-based self-adaptive embedded digital tag information hiding method
CN115695673B (en) * 2022-12-30 2023-04-21 中国科学技术大学 Image anti-steganography method and device based on joint distortion

Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
WO2018212811A1 (en) * 2017-05-19 2018-11-22 Google Llc Hiding information and images via deep learning
CN109587372A (en) * 2018-12-11 2019-04-05 北京邮电大学 A kind of invisible image latent writing art based on generation confrontation network
CN110334805A (en) * 2019-05-05 2019-10-15 中山大学 A kind of JPEG domain image latent writing method and system based on generation confrontation network
CN111222583A (en) * 2020-01-15 2020-06-02 北京中科研究院 Image steganalysis method based on confrontation training and key path extraction

Family Cites Families (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US11645835B2 (en) * 2017-08-30 2023-05-09 Board Of Regents, The University Of Texas System Hypercomplex deep learning methods, architectures, and apparatus for multimodal small, medium, and large-scale data representation, analysis, and applications

Patent Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
WO2018212811A1 (en) * 2017-05-19 2018-11-22 Google Llc Hiding information and images via deep learning
CN109587372A (en) * 2018-12-11 2019-04-05 北京邮电大学 A kind of invisible image latent writing art based on generation confrontation network
CN110334805A (en) * 2019-05-05 2019-10-15 中山大学 A kind of JPEG domain image latent writing method and system based on generation confrontation network
CN111222583A (en) * 2020-01-15 2020-06-02 北京中科研究院 Image steganalysis method based on confrontation training and key path extraction

Non-Patent Citations (4)

* Cited by examiner, † Cited by third party
Title
A New Payload Partition Strategy in Color Image Steganography;Xin Liao; Yingbo Yu; Bin Li;《 IEEE Transactions on Circuits and Systems for Video Technology》;20190131;第30卷(第3期);685 - 696 *
Clustering Steganographic Modification Directions for Color Components;Tang, Weixuan,etc;《IEEE Signal Processing Letters》;20160228;第23卷(第2期);197-201 *
基于卷积神经网络的图像隐写分析;魏伟航;《中国优秀博硕士学位论文全文数据库(硕士)信息科技辑》;20190715(第07期);I138-31 *
基于生成对抗网络的图像信息隐藏研究;董士琪;《中国优秀博硕士学位论文全文数据库(硕士)信息科技辑》;20190915(第09期);I138-132 *

Also Published As

Publication number Publication date
CN111681154A (en) 2020-09-18

Similar Documents

Publication Publication Date Title
CN111681154B (en) Color image steganography distortion function design method based on generation countermeasure network
CN108346125B (en) Airspace image steganography method and system based on generation countermeasure network
Abdelhakim et al. A quality guaranteed robust image watermarking optimization with Artificial Bee Colony
CN111523668B (en) Training method and device of data generation system based on differential privacy
CN109993678B (en) Robust information hiding method based on deep confrontation generation network
CN109639479B (en) Network traffic data enhancement method and device based on generation countermeasure network
CN112634117B (en) End-to-end JPEG domain image steganography method based on generation of countermeasure network
El-Emam et al. New steganography algorithm to conceal a large amount of secret message using hybrid adaptive neural networks with modified adaptive genetic algorithm
Hsu et al. A high-capacity QRD-based blind color image watermarking algorithm incorporated with AI technologies
CN105404817A (en) (k,n) threshold based user-friendly visual secret sharing method
Wang et al. HidingGAN: High capacity information hiding with generative adversarial network
US20240104681A1 (en) Image steganography utilizing adversarial perturbations
Li et al. AdvSGAN: Adversarial image Steganography with adversarial networks
Li et al. Privacy protection method based on multidimensional feature fusion under 6G networks
CN111768326A (en) High-capacity data protection method based on GAN amplification image foreground object
Sharma et al. A new optimization based color image watermarking using non-negative matrix factorization in discrete cosine transform domain
CN117079083B (en) Image restoration model training method and device, electronic equipment and storage medium
Liu et al. Image processing method based on chaotic encryption and wavelet transform for planar design
Jin et al. Minimizing Turtle-Shell Matrix Based Stego Image Distortion Using Particle Swarm Optimization.
Li et al. A new robust watermarhing scheme based on shuffled frog leaping algorithm
Devi et al. A robust and optimized 3D red-cyan anaglyph blind image watermarking in the DWT domain
CN114900586B (en) Information steganography method and device based on DCGAN
Jaya Prakash et al. Improved reversible data hiding scheme employing dual image-based least significant bit matching for secure image communication using style transfer
Sun et al. Optimization of MSFs for watermarking using DWT-DCT-SVD and fish migration optimization with QUATRE
Li et al. A robust watermarking scheme with high security and low computational complexity

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