CN110084734A - A kind of big data ownership guard method being locally generated confrontation network based on object - Google Patents
A kind of big data ownership guard method being locally generated confrontation network based on object Download PDFInfo
- Publication number
- CN110084734A CN110084734A CN201910340335.0A CN201910340335A CN110084734A CN 110084734 A CN110084734 A CN 110084734A CN 201910340335 A CN201910340335 A CN 201910340335A CN 110084734 A CN110084734 A CN 110084734A
- Authority
- CN
- China
- Prior art keywords
- image
- network
- steganography
- network module
- characteristic pattern
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Granted
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T1/00—General purpose image data processing
- G06T1/0021—Image watermarking
- G06T1/005—Robust watermarking, e.g. average attack or collusion attack resistant
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Biophysics (AREA)
- Evolutionary Computation (AREA)
- Artificial Intelligence (AREA)
- Biomedical Technology (AREA)
- Health & Medical Sciences (AREA)
- Computational Linguistics (AREA)
- Data Mining & Analysis (AREA)
- Life Sciences & Earth Sciences (AREA)
- General Health & Medical Sciences (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- General Engineering & Computer Science (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Image Analysis (AREA)
Abstract
The present invention is a kind of big data ownership guard method that confrontation network is locally generated based on object, and on training set, convolutional network of the carrier image Jing Guo multilayer obtains multiple groups characteristic pattern, be successively passed to carrier reconstructed module.Generating network further includes multilayer deconvolution network, is successively passed to carrier reconstructed module in the characteristic pattern that formation condition obtains after multilayer deconvolution.The destination carrier image of generation is input to steganography network, carry out amplification and the steganography secret information in steganography region, then the image of generation is input in more granularity arbiters, it is differentiated on image Weak characteristic and picture quality respectively, update the parameter for differentiating network, by repeatedly modifying, obtaining network model parameter and saving.Sender and recipient are generated containing secret image using above-mentioned network model parameter or restore steganography region and extract secret information.This patent is realized to be hidden based on the generation confrontation network information, is solved the problems, such as to generate the steganography in the steganography of confrontation network, is improved safety.
Description
Technical field
The invention belongs to big data technical treatment technical fields, specifically a kind of to be locally generated confrontation net based on object
The big data ownership guard method of network.
Background technique
In big data era, data are the drivings of scientific research.The rapid development of the subjects such as deep learning, artificial intelligence from
The development of data science is not opened.Internet user is using the artificial intelligence skill such as similar image search, proposed algorithm, speech recognition
While art is brought convenience, it be unable to do without various data sets.In the research of deep learning, around with Special Significance
Data set training one has the model of specific function, is the development process of most of intellectual product instantly.However, in big data
Under the background of cloud computing, data are stored by cloud service provider.User is produced in the software or service provided using cloud service provider
In the server of raw private data storage and cloud service provider.User can only default trust cloud to use the software or service
Service provider.However, simple trust can't bring eternal safety, the thing for often thering are the data of cloud service provider to be stolen in recent years
Part occurs.Therefore, for the user for having security needs, the safety of data cannot be affected under the present circumstances, number
According to ownership needs ensured.This method utilizes Information Hiding Techniques, by the sensitive information of user and ownership Information hiding in
In the day regular data (such as image) at family, these day regular data itself is meaningful, expresses the information of the reception and registration of user.But
In Information hiding field, the steganographic algorithm of hiding information is more likely to the complex texture region hiding information of selection image, user
Daily image data may not have enough complex texture regions.Therefore, this method is made in the daily image data of user
On the basis of background, using the foreground object for fighting network generation and meeting current context is generated, which has complexity
The characteristics of texture, therefore the secret information of user can be hidden with the prospect of generation, achieve the purpose that promote user information safety.
Under the tide of deep learning, confrontation network (Generative Adversarial Networks, GAN) is generated
It is rapidly developed, is proposed so far from GAN [1], fight thought using it, emerged in large numbers various outstanding derivative algorithms and skill
Art.Such as WGAN [2] improves the effect of GAN；Conditional GAN [3] increases generation image on the basis of GAN
Formation condition；Stack GAN [4] is realized from text generation image, which meets text description.
Information hiding is the important branch of information security field, it is expected that utilizing use in relatively unsafe transmission
Information, the hiding users such as the data (such as text, image, video) that family generates are not intended to disclosed data.Wherein, comparison basis
Information Hiding Algorithms, such as least significant bit (Least Significant Bit, LSB) steganographic algorithm passes through modification pixel
Least significant bit, carry out embedding information, complete Information hiding.Its principle according to human eye to the insensitivity of small pixel value, i.e.,
Pixel least significant bit is modified, the identifiable change of human eye will not be caused to image itself.For example, common single channel figure
As each pixel is made of 82 binary digits, after being converted into 10 systems, pixel value section is 0 to 255, least significant bit
Modification is so that the change section of pixel value is -1,1 or 0.Since this algorithm can not resist simple image statistics analysis, because
This, on the basis of LSB steganography, the improvement direction of steganographic algorithm is adaptive and anti-statistical analysis.Such as UNIWARD [5],
WOW [6] this kind of algorithm reaches adaptive and anti-by reducing distortion rate by defining the cost function calculation distortion rate of steganography
The purpose of statistical analysis.
Summary of the invention
The technical problem to be solved in the present invention is to provide a kind of big data ownerships that confrontation network is locally generated based on object
Guard method, this method increase steganalysis in differentiating network and differentiate by increasing steganography network module in generating network
Network.While Background Reconstruction, the relative complex object of texture is generated as safety zone in foreground area, it is hidden to carry out information
Hiding.Meanwhile image quality analysis and steganalysis are carried out in the generating process of stego-image.
In order to solve the above technical problems, the technical solution adopted by the present invention are as follows:
A kind of big data ownership guard method being locally generated confrontation network based on object, it is characterized in that: specific step is as follows:
Step 1, it pre-processes；Initial carrier image is inputted, initial carrier image passes through four layers of volume in pretreatment network module
Product network, successively obtains four groups of various sizes of characteristic patterns；
Step 2, image generates；Input random noise z and formation condition c, which enters, generates network module as formation condition, passes through
Dimensional variation and the scale of the last layer characteristic pattern of pretreatment network module output are consistent, and input Background Reconstruction together
Module successively passes through four Background Reconstruction modules respectively in connection with the characteristic pattern of correspondingly-sized, it is newly-generated to generate network module output
Carrier image and exposure mask, the carrier image be based on initial carrier image have generate object image, institute
The exposure mask stated is used to indicate the position that object is generated in carrier image；
Step 3, information steganography；Carrier image and exposure mask are input in steganography network module, inputted into steganography network module
Secret information, the steganography network module, which is used to pass through, carries out secret in the generation object area that steganographic algorithm indicates exposure mask
Information hiding exports stego-image；
Step 4, as a result differentiate；Exposure mask and stego-image are input to by the feature that down-sampling extracts and differentiated in network module,
The differentiation network module is used to judge whether the stego-image of output to generate the object for meeting training image, and output
Stego-image whether contain secret information；
Step 5, repetition training；Error loss after differentiating network module output comparison, according to the gradient direction of deviation, update is sentenced
The parameter of other network module, the parameter that adjustment generates in network module finally obtain accurate network by multiple training process
Model parameter；
Step 6, sender and recipient share parameter model network structure corresponding with its of Weak characteristic convolutional neural networks,
Initial carrier image is passed to after pretreatment network module pretreatment and generates network module by sender, exports stego-image,
Recipient utilizes parameter model network structure corresponding with its, restores all steganography regions, recycles the reversible of steganographic algorithm
Extraction algorithm extracts secret information.
Four groups of characteristic pattern scales obtained in the step 1 are respectively different, and each characteristic pattern has initial carrier figure
The feature of picture.
The random noise z is the random sequence that one section of dimension is 100 dimensions, and the random noise z is used for as life
At the initial distribution of image, the formation condition c is the one-hot coding of one section of 16 dimension, and the formation condition c is for referring to
Show the image type for generating object.
The input that network module is differentiated in the step 4 is the stego-image generated and the true picture of non-generation, institute
For the differentiation network module stated for differentiating picture quality and steganalysis, the differentiation network module is different rulers by structure
The artificial convolutional neural networks of very little convolution kernel carry out differentiation picture quality, and the differentiation network module passes through steganalysis network
Differentiate the steganography distortion of the stego-image of generation, Weak characteristic can be extracted by being provided in the steganalysis network
Convolution kernel.
In the Background Reconstruction module, the m dimensional feature figure of preceding layer convolutional layer in network module is made a living into input, is passed through
The depth of expansion figure be 2m, wherein the characteristic pattern of m depth after sigmoid is activated with background image dot product, Zhi Houyu
The characteristic pattern of another m depth is added, and the depth of the characteristic pattern of output is m.
The secret information is hidden to be generated in network by the way that reversible steganography function to be integrated into, and steganography is realized, in information
When extraction, the position of hiding information is found using mask image, available all steganography regions recycle steganography function corresponding
Extraction function, extract secret information.
This kind based on object be locally generated confrontation network the guard method of big data ownership can the beneficial effects are as follows:
This method makes it that foreground object not only can be generated when generating by modification generator, can also rebuild background, and realization is being carried on the back
Prospect is generated on scape.Background image is the carrier image of user, and the image of generation is final destination carrier image, foreground object
It is that background semantic is combined to generate.Meanwhile foreground object is more complicated than background image texture, appoints to be more applicable for steganography
Business；By modifying arbiter, steganalysis network is added, realizes differentiation and analysis to visual characteristic and steganography characteristic.Pass through
Dual training between generator and arbiter reaches the validity and robustness of steganography.Carrier image is utilized in this patent, and
It does not directly generate, on the one hand ensure that the original meaning of carrier image；On the other hand, sentence since this patent introduces more granularities
Other network generates picture quality and is improved.In terms of extraction, compared to the above method, method presents explicitly mention
Scheme is taken, realization efficiently extracts secret information.
Detailed description of the invention
Fig. 1 is a kind of flow chart for the big data ownership guard method that confrontation network is locally generated based on object of the present invention.
Fig. 2 is to pre-process net in a kind of big data ownership guard method for being locally generated confrontation network based on object of the present invention
Network module and the structural schematic diagram for generating network module.
Fig. 3 is to differentiate network in a kind of big data ownership guard method for being locally generated confrontation network based on object of the present invention
The structural schematic diagram of module.
Fig. 4, which is that the present invention is a kind of, is locally generated Background Reconstruction in the big data ownership guard method of confrontation network based on object
The structural schematic diagram of module.
Fig. 5 be invent it is a kind of based on object be locally generated confrontation network big data ownership guard method carrier in image with
The example diagram of destination carrier image after reconstruct.
Specific embodiment
Below in conjunction with Figure of description and specific preferred embodiment, the invention will be further described.
A kind of big data ownership guard method being locally generated confrontation network based on object, it is characterized in that: specific steps are such as
Under:
Step 1, it pre-processes；Initial carrier image is inputted, initial carrier image passes through four layers of volume in pretreatment network module
Product network, successively obtains four groups of various sizes of characteristic patterns；
Step 2, image generates；Input random noise z and formation condition c, which enters, generates network module as formation condition, passes through
Dimensional variation and the scale of the last layer characteristic pattern of pretreatment network module output are consistent, and input Background Reconstruction together
Module successively passes through four Background Reconstruction modules respectively in connection with the characteristic pattern of correspondingly-sized, it is newly-generated to generate network module output
Carrier image and exposure mask, the carrier image be based on initial carrier image have generate object image, institute
The exposure mask stated is used to indicate the position that object is generated in carrier image；
Step 3, information steganography；Carrier image and exposure mask are input in steganography network module, inputted into steganography network module
Secret information, the steganography network module, which is used to pass through, carries out secret in the generation object area that steganographic algorithm indicates exposure mask
Information hiding exports stego-image；
Step 4, as a result differentiate；Exposure mask and stego-image are input to by the feature that down-sampling extracts and differentiated in network module,
The differentiation network module is used to judge whether the stego-image of output to generate the object for meeting training image, and output
Stego-image whether contain secret information；
Step 5, repetition training；Error loss after differentiating network module output comparison, according to the gradient direction of deviation, update is sentenced
The parameter of other network module, the parameter that adjustment generates in network module finally obtain accurate network by multiple training process
Model parameter；
Step 6, sender and recipient share parameter model network structure corresponding with its of Weak characteristic convolutional neural networks,
Initial carrier image is passed to after pretreatment network module pretreatment and generates network module by sender, exports stego-image,
Recipient utilizes parameter model network structure corresponding with its, restores all steganography regions, recycles the reversible of steganographic algorithm
Extraction algorithm extracts secret information.
As shown in Figure 1, input is made of four parts, input 1 is formation condition c, and instruction generates the type of object；Input
2 be random automatic initial noise, is used as the initial value of generating process；Input 3 is initial carrier image；Input 4 is to turn
It is changed to the secret information of 2 systems.Output is stego-image, contains secret information in object masked areas.Wherein, 3 conducts are inputted
The characteristic pattern for respectively obtaining 4 groups of different scales, these features are extracted in the input for pre-processing network by four layers of convolutional neural networks
Image strip has the feature of initial carrier image, makes a living into network completion increase object on the basis of carrier image and provides possibility.
Input 1 and input 2 are input to as the primary condition for generating network to be generated in network, defeated by dimensional variation and pretreatment network
The scale of the last layer out is consistent, and inputs Background Reconstruction module together.Successively after 4 Background Reconstruction modules,
Image and corresponding exposure mask that network output generates object on the basis of initial carrier are generated, which indicates the carrier of generation
In image, the position of object is newly increased.Next, the carrier image of generation and exposure mask are input in steganography network, use is hidden
Algorithm is write to Information hiding is carried out in the generation object area of exposure mask, exports stego-image.Later, exposure mask and stego-image pass through
Down-sampling is input in differentiation network after extracting feature, differentiates that network judges whether the image of output generates respectively and meets training
Whether the image of the object output of image contains secret information.Error loss is obtained after judgement, which is passed into life again
At network, primary training is completed.
In the present embodiment, the parameter of the hardware platform of use are as follows: Intel i9 CPU, 32GB running memory, NVIDIA
The memory space of 1080Ti GPU and 2TB.
As shown in Fig. 2, generator includes pretreatment network module and image generation module, the input of generator include original
Carrier image cover and random noise z and formation condition c.Reshape represents dimension variation.Deconv represents warp lamination,
Conv represents convolutional layer.Destination carrier is four-way, wherein before triple channel as color image tri- channels RBG another lead to
Road is as mask.Wherein, random noise is the random sequence that one section of dimension is 100 dimensions, as the initial distribution for generating image.It is raw
It is solely hot at the one-hot(that condition is one section of 16 dimension) coding, indicate the type for generating subject image.For example,
0000000000000001 representation type dog, 0000000000000010 representation type bird.Initial carrier image cover successively leads to
Four layers of convolutional layer are crossed, the effect of convolutional layer is the local feature for extracting image, these local features represent image in different layers
The characteristic pattern of characteristic on secondary, the characteristic pattern of these different levels and level same in following generating process, which blends, to be conducive to
The synthesis for generating image is rebuild.Obtained characteristic pattern is separately input to respective Background Reconstruction module R1 to R4.Meanwhile at random
Noise z and formation condition c is input to as formation condition and generates in network, passes sequentially through four layers of warp lamination, warp lamination
Effect is the scaling up for the characteristic pattern that will be inputted, simultaneously because deconvolution operates, the characteristic pattern of output is made to continue to keep part
Feature.Obtained characteristic pattern is separately input to Background Reconstruction module R1 to R4, the convolution characteristic pattern with initial carrier image cover
Carry out Background Reconstruction.Finally, the characteristic pattern tensor that network exports one 4 dimension is generated, secondary 4 channel images generated are represented.By
In characteristic pattern tensor between -1 to 1 floating number numerical intervals, need to be converted into the image pixel numerical value of 8bit depth
The integer in section, image pixel numerical value section are 0 to 255.After processing, preceding 3 dimension forms destination carrier figure as R, G, channel B
Picture, foreground object mask of the 4th dimension as destination carrier image.The effect of the foreground object mask of destination carrier image be rear
Continuous steganography function provides preliminary steganography region.The acquisition process of Mask is by etching operation, so that the value of Mask tensor is divided into 0
Or 1, wherein 0 represents steganography region, and the pixel value after converting is 0；1 represents non-steganography region, the pixel value after conversion
It is 255.
As shown in figure 3, differentiate network module input be generate stego-image and non-generation it is true, by differentiation net
Network.The effect for differentiating network is to differentiate picture quality and steganalysis.It is the differentiation network of picture quality on the right side of Fig. 3, structure is
The artificial convolutional neural networks of different size convolution kernels, task are to differentiate the picture quality for generating stego-image.Pass through 3 groups respectively
Different convolutional coding structures, it is therefore an objective to different convolution features is extracted using different size of convolution kernel, for example, (RGB) is tieed up by 3,
Length and width are respectively that the image of 64 and 64 pixels obtains 32 dimensions after the first layer of first group of convolutional neural networks, and length and width are respectively
The characteristic pattern of 32 and 32 pixels continues on through convolutional layer processing and goes down, finally obtains 256 dimensions.Length and width are respectively 4 and 4 characteristic pattern.
These last characteristic patterns combine the input for being deformed into anticipation function softmax function, and the output of prediction is as judgement
It is whether qualified as a result, being eventually converted into the penalty values of network.It is steganalysis network on the left of Fig. 3, task is to differentiate life
It is distorted at the steganography of stego-image.Its result is similar to the right side, but increases the convolution kernel that can extract Weak characteristic, purpose
It is to extract bring minor alteration after steganography.
The image of generation, which is input to, to be differentiated in network module, is sentenced on image Weak characteristic and picture quality respectively to it
Not, according to the gradient direction of deviation, the parameter that more granularities differentiate network is updated；Again by adjusting the parameter for generating network, update
Generate the parameter of network.So far, a training process is completed.Reach 1,000 time in frequency of training, wherein each training contains
After having 1,000 iteration, terminates training, obtain network model parameter and save.
As shown in figure 4, the m dimensional feature figure of preceding layer convolutional layer in network is made a living into input when carrier is rebuild, pass through expansion
Increase characteristic pattern depth be 2m, wherein the characteristic pattern of m depth after sigmoid is activated with background image dot product, the back
Scape image is carrier image, is added later with the characteristic pattern of another m depth, the depth of the characteristic pattern of output is m.Obtaining steganography
Region part passes through the destination carrier image and mask image hiding information of generation.In Information hiding part, by will be reversible hidden
It writes function to be integrated into generation network, realizes steganography.In information extraction part, the position of hiding information is found using mask image
It sets, available all steganography regions, recycles the corresponding extraction function of steganography function, extract secret information.
In the present embodiment, size adjusting is able to carry out when carrier is rebuild, the function of the rescaling is for adjusting
The scale of the foreground object of whole generation merges it reasonably with background.
As shown in figure 5, being initial carrier image on the left of Fig. 5, right side is to generate destination carrier stego-image.Recipient utilizes
Parameter model network structure corresponding with its, restores all steganography regions；Finally, the reversible extraction using steganographic algorithm is calculated
Method extracts secret information.
This method generates foreground object on background image by generating confrontation network come composograph as new target
Secret information is hidden into foreground object, since foreground object passes through confrontation by carrier by generating the steganography module in network
It generates, is particularly suited for steganography.It is loaded into network structure when extraction using trained model before, extracts secret letter
Breath.Therefore, whole process improves the safety of carrier image, is easily achieved simultaneously.
The above is only the preferred embodiment of the present invention, protection scope of the present invention is not limited merely to above-described embodiment,
All technical solutions belonged under thinking of the present invention all belong to the scope of protection of the present invention.It should be pointed out that for the art
For those of ordinary skill, several improvements and modifications without departing from the principles of the present invention should be regarded as protection of the invention
Range.
Claims (6)
1. a kind of big data ownership guard method that confrontation network is locally generated based on object, it is characterized in that: specific step is as follows:
Step 1, it pre-processes；Initial carrier image is inputted, initial carrier image passes through four layers of volume in pretreatment network module
Product network, successively obtains four groups of various sizes of characteristic patterns；
Step 2, image generates；Input random noise z and formation condition c, which enters, generates network module as formation condition, passes through
Dimensional variation and the scale of the last layer characteristic pattern of pretreatment network module output are consistent, and input Background Reconstruction together
Module successively passes through four Background Reconstruction modules respectively in connection with the characteristic pattern of correspondingly-sized, it is newly-generated to generate network module output
Carrier image and exposure mask, the carrier image be based on initial carrier image have generate object image, institute
The exposure mask stated is used to indicate the position that object is generated in carrier image；
Step 3, information steganography；Carrier image and exposure mask are input in steganography network module, inputted into steganography network module
Secret information, the steganography network module, which is used to pass through, carries out secret in the generation object area that steganographic algorithm indicates exposure mask
Information hiding exports stego-image；
Step 4, as a result differentiate；Exposure mask and stego-image are input to by the feature that down-sampling extracts and differentiated in network module,
The differentiation network module is used to judge whether the stego-image of output to generate the object for meeting training image, and output
Stego-image whether contain secret information；
Step 5, repetition training；Error loss after differentiating network module output comparison, according to the gradient direction of deviation, update is sentenced
The parameter of other network module, the parameter that adjustment generates in network module finally obtain accurate network by multiple training process
Model parameter；
Step 6, sender and recipient share parameter model network structure corresponding with its of Weak characteristic convolutional neural networks,
Initial carrier image is passed to after pretreatment network module pretreatment and generates network module by sender, exports stego-image,
Recipient utilizes parameter model network structure corresponding with its, restores all steganography regions, recycles the reversible of steganographic algorithm
Extraction algorithm extracts secret information.
2. a kind of big data ownership guard method that confrontation network is locally generated based on object according to claim 1,
Be characterized in that: four groups of characteristic pattern scales obtained in the step 1 are respectively different, and each characteristic pattern has initial carrier figure
The feature of picture.
3. a kind of big data ownership guard method that confrontation network is locally generated based on object according to claim 1,
Be characterized in that: the random noise z is the random sequence that one section of dimension is 100 dimensions, and the random noise z is used for as life
At the initial distribution of image, the formation condition c is the one-hot coding of one section of 16 dimension, and the formation condition c is for referring to
Show the image type for generating object.
4. a kind of big data ownership guard method that confrontation network is locally generated based on object according to claim 1,
Be characterized in that: the input that network module is differentiated in the step 4 is the stego-image generated and the true picture of non-generation, institute
For the differentiation network module stated for differentiating picture quality and steganalysis, the differentiation network module is different rulers by structure
The artificial convolutional neural networks of very little convolution kernel carry out differentiation picture quality, and the differentiation network module passes through steganalysis network
Differentiate the steganography distortion of the stego-image of generation, Weak characteristic can be extracted by being provided in the steganalysis network
Convolution kernel.
5. a kind of big data ownership guard method that confrontation network is locally generated based on object according to claim 1,
Be characterized in that: in the Background Reconstruction module, the m dimensional feature figure of preceding layer convolutional layer in network module is made a living into input, is led to
Cross expansion figure depth be 2m, wherein the characteristic pattern of m depth after sigmoid is activated with background image dot product, later
It is added with the characteristic pattern of another m depth, the depth of the characteristic pattern of output is m.
6. a kind of big data ownership guard method that confrontation network is locally generated based on object according to claim 5,
Be characterized in that: the secret information is hidden to be generated in network by the way that reversible steganography function to be integrated into, and steganography is realized, in information
When extraction, the position of hiding information is found using mask image, available all steganography regions recycle steganography function corresponding
Extraction function, extract secret information.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910340335.0A CN110084734B (en) | 2019-04-25 | 2019-04-25 | Big data ownership protection method based on object local generation countermeasure network |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910340335.0A CN110084734B (en) | 2019-04-25 | 2019-04-25 | Big data ownership protection method based on object local generation countermeasure network |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN110084734A true CN110084734A (en) | 2019-08-02 |
| CN110084734B CN110084734B (en) | 2023-02-14 |
Family
ID=67416832
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201910340335.0A Active CN110084734B (en) | 2019-04-25 | 2019-04-25 | Big data ownership protection method based on object local generation countermeasure network |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN110084734B (en) |
Cited By (15)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110648289A (en) * | 2019-08-29 | 2020-01-03 | 腾讯科技（深圳）有限公司 | Image denoising processing method and device |
| CN111242133A (en) * | 2020-01-14 | 2020-06-05 | 济南浪潮高新科技投资发展有限公司 | GAN hidden layer unit and method and system for generating object correlation in image |
| CN111507933A (en) * | 2019-12-16 | 2020-08-07 | 曲阜师范大学 | DIBR synthetic image quality evaluation method based on cavity and contour amplification |
| CN111640052A (en) * | 2020-05-22 | 2020-09-08 | 南京信息工程大学 | Robust high-capacity digital watermarking method based on mark code |
| CN111711821A (en) * | 2020-06-15 | 2020-09-25 | 南京工程学院 | Information hiding method based on deep learning |
| CN111768326A (en) * | 2020-04-03 | 2020-10-13 | 南京信息工程大学 | High-capacity data protection method based on GAN amplification image foreground object |
| CN111768325A (en) * | 2020-04-03 | 2020-10-13 | 南京信息工程大学 | Security improvement method based on generation of countermeasure sample in big data privacy protection |
| CN111859897A (en) * | 2019-10-16 | 2020-10-30 | 沈阳工业大学 | Text steganalysis method based on dynamic routing capsule network |
| CN111951149A (en) * | 2020-08-14 | 2020-11-17 | 中国人民武装警察部队工程大学 | Image information steganography method based on neural network |
| CN112767226A (en) * | 2021-01-15 | 2021-05-07 | 南京信息工程大学 | Image steganography method and system based on GAN network structure automatic learning distortion |
| CN112785478A (en) * | 2021-01-15 | 2021-05-11 | 南京信息工程大学 | Hidden information detection method and system based on embedded probability graph generation |
| CN112926569A (en) * | 2021-03-16 | 2021-06-08 | 重庆邮电大学 | Method for detecting natural scene image text in social network |
| CN113077377A (en) * | 2021-05-13 | 2021-07-06 | 海南大学 | Color image steganography method based on generation countermeasure network |
| CN113610878A (en) * | 2021-07-15 | 2021-11-05 | 南京信息工程大学 | Structural information hiding method based on automatic contour generation |
| CN116861013A (en) * | 2023-09-04 | 2023-10-10 | 深圳市易图资讯股份有限公司 | CIM data credibility improving method |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107563155A (en) * | 2017-08-08 | 2018-01-09 | 中国科学院信息工程研究所 | A kind of safe steganography method and device based on generation confrontation network |
| CN108346125A (en) * | 2018-03-15 | 2018-07-31 | 中山大学 | A kind of spatial domain picture steganography method and system based on generation confrontation network |
| CN108921764A (en) * | 2018-03-15 | 2018-11-30 | 中山大学 | A kind of image latent writing method and system based on generation confrontation network |
| CN109492416A (en) * | 2019-01-07 | 2019-03-19 | 南京信息工程大学 | A kind of guard method of big data image and system based on safety zone |
-
2019
- 2019-04-25 CN CN201910340335.0A patent/CN110084734B/en active Active
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107563155A (en) * | 2017-08-08 | 2018-01-09 | 中国科学院信息工程研究所 | A kind of safe steganography method and device based on generation confrontation network |
| CN108346125A (en) * | 2018-03-15 | 2018-07-31 | 中山大学 | A kind of spatial domain picture steganography method and system based on generation confrontation network |
| CN108921764A (en) * | 2018-03-15 | 2018-11-30 | 中山大学 | A kind of image latent writing method and system based on generation confrontation network |
| CN109492416A (en) * | 2019-01-07 | 2019-03-19 | 南京信息工程大学 | A kind of guard method of big data image and system based on safety zone |
Non-Patent Citations (2)
| Title |
|---|
| ZHUO ZHANG 等: "Generative Information Hiding Method Based on Adversarial Networks", 《CENET 2018》 * |
| 王耀杰 等: "基于生成对抗网络的信息隐藏方案", 《计算机应用》 * |
Cited By (26)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110648289B (en) * | 2019-08-29 | 2023-07-11 | 腾讯科技（深圳）有限公司 | Image noise adding processing method and device |
| CN110648289A (en) * | 2019-08-29 | 2020-01-03 | 腾讯科技（深圳）有限公司 | Image denoising processing method and device |
| CN111859897A (en) * | 2019-10-16 | 2020-10-30 | 沈阳工业大学 | Text steganalysis method based on dynamic routing capsule network |
| CN111507933A (en) * | 2019-12-16 | 2020-08-07 | 曲阜师范大学 | DIBR synthetic image quality evaluation method based on cavity and contour amplification |
| CN111507933B (en) * | 2019-12-16 | 2023-08-01 | 曲阜师范大学 | DIBR synthetic image quality evaluation method based on cavity and contour amplification |
| CN111242133A (en) * | 2020-01-14 | 2020-06-05 | 济南浪潮高新科技投资发展有限公司 | GAN hidden layer unit and method and system for generating object correlation in image |
| CN111242133B (en) * | 2020-01-14 | 2022-06-28 | 山东浪潮科学研究院有限公司 | Method and system for generating correlation of object in image and GAN hidden layer unit |
| CN111768325A (en) * | 2020-04-03 | 2020-10-13 | 南京信息工程大学 | Security improvement method based on generation of countermeasure sample in big data privacy protection |
| CN111768326B (en) * | 2020-04-03 | 2023-08-25 | 南京信息工程大学 | High-capacity data protection method based on GAN (gas-insulated gate bipolar transistor) amplified image foreground object |
| CN111768326A (en) * | 2020-04-03 | 2020-10-13 | 南京信息工程大学 | High-capacity data protection method based on GAN amplification image foreground object |
| CN111768325B (en) * | 2020-04-03 | 2023-07-25 | 南京信息工程大学 | Security improvement method based on generation of countermeasure sample in big data privacy protection |
| CN111640052A (en) * | 2020-05-22 | 2020-09-08 | 南京信息工程大学 | Robust high-capacity digital watermarking method based on mark code |
| CN111711821B (en) * | 2020-06-15 | 2022-06-10 | 南京工程学院 | Information hiding method based on deep learning |
| CN111711821A (en) * | 2020-06-15 | 2020-09-25 | 南京工程学院 | Information hiding method based on deep learning |
| CN111951149B (en) * | 2020-08-14 | 2023-10-31 | 中国人民武装警察部队工程大学 | Image information steganography method based on neural network |
| CN111951149A (en) * | 2020-08-14 | 2020-11-17 | 中国人民武装警察部队工程大学 | Image information steganography method based on neural network |
| CN112767226A (en) * | 2021-01-15 | 2021-05-07 | 南京信息工程大学 | Image steganography method and system based on GAN network structure automatic learning distortion |
| CN112785478A (en) * | 2021-01-15 | 2021-05-11 | 南京信息工程大学 | Hidden information detection method and system based on embedded probability graph generation |
| 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 |
| CN112785478B (en) * | 2021-01-15 | 2023-06-23 | 南京信息工程大学 | Hidden information detection method and system based on generation of embedded probability map |
| CN112926569B (en) * | 2021-03-16 | 2022-10-18 | 重庆邮电大学 | Method for detecting natural scene image text in social network |
| CN112926569A (en) * | 2021-03-16 | 2021-06-08 | 重庆邮电大学 | Method for detecting natural scene image text in social network |
| CN113077377A (en) * | 2021-05-13 | 2021-07-06 | 海南大学 | Color image steganography method based on generation countermeasure network |
| CN113610878A (en) * | 2021-07-15 | 2021-11-05 | 南京信息工程大学 | Structural information hiding method based on automatic contour generation |
| CN116861013A (en) * | 2023-09-04 | 2023-10-10 | 深圳市易图资讯股份有限公司 | CIM data credibility improving method |
| CN116861013B (en) * | 2023-09-04 | 2023-12-19 | 深圳市易图资讯股份有限公司 | CIM data credibility improving method |
Also Published As
| Publication number | Publication date |
|---|---|
| CN110084734B (en) | 2023-02-14 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN110084734A (en) | A kind of big data ownership guard method being locally generated confrontation network based on object | |
| CN110458844B (en) | Semantic segmentation method for low-illumination scene | |
| Jin et al. | Generative adversarial network technologies and applications in computer vision | |
| Pan et al. | Loss functions of generative adversarial networks (GANs): Opportunities and challenges | |
| CN110084121A (en) | Implementation method based on the human face expression migration for composing normalized circulation production confrontation network | |
| Chen et al. | DARGS: Image inpainting algorithm via deep attention residuals group and semantics | |
| Wang et al. | TMS-GAN: A twofold multi-scale generative adversarial network for single image dehazing | |
| CN106845471A (en) | A kind of vision significance Forecasting Methodology based on generation confrontation network | |
| Zhong et al. | Generative adversarial networks with decoder–encoder output noises | |
| Johari et al. | Context-aware colorization of gray-scale images utilizing a cycle-consistent generative adversarial network architecture | |
| CN116563399A (en) | Image generation method based on diffusion model and generation countermeasure network | |
| CN110188791B (en) | Visual emotion label distribution prediction method based on automatic estimation | |
| Garau et al. | CapsulePose: A variational CapsNet for real-time end-to-end 3D human pose estimation | |
| Panda et al. | Modified ResNet-152 network with hybrid pyramidal pooling for local change detection | |
| Li et al. | Face Recognition Based on the Combination of Enhanced Local Texture Feature and DBN under Complex Illumination Conditions. | |
| Qin et al. | Virtual reality video image classification based on texture features | |
| Fang et al. | A New Method of Image Restoration Technology Based on WGAN. | |
| Zhou et al. | Image generation from text with entity information fusion | |
| Xie et al. | Design of painting art style rendering system based on convolutional neural network | |
| CN115862097A (en) | Method and device for identifying shielding face based on multi-attention and multi-scale feature learning | |
| Ye et al. | Interactive anime sketch colorization with style consistency via a deep residual neural network | |
| Wu et al. | Transformer Autoencoder for K-means Efficient clustering | |
| CN112581360B (en) | Method for enhancing aesthetic quality of multi-style image based on structural constraint | |
| Jiang et al. | Adaptive adversarial latent space for novelty detection | |
| Chen et al. | Infrared images augmentation based on images generation with generative adversarial networks |
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 |