CN110517227A - Method, apparatus, computer equipment and the storage medium of image concealing infomation detection - Google Patents
Method, apparatus, computer equipment and the storage medium of image concealing infomation detection Download PDFInfo
- Publication number
- CN110517227A CN110517227A CN201910688378.8A CN201910688378A CN110517227A CN 110517227 A CN110517227 A CN 110517227A CN 201910688378 A CN201910688378 A CN 201910688378A CN 110517227 A CN110517227 A CN 110517227A
- Authority
- CN
- China
- Prior art keywords
- image
- privacy
- training
- neural networks
- convolutional neural
- 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
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
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/0002—Inspection of images, e.g. flaw detection
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2201/00—General purpose image data processing
- G06T2201/005—Image watermarking
- G06T2201/0065—Extraction of an embedded watermark; Reliable detection
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/20—Special algorithmic details
- G06T2207/20081—Training; Learning
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/20—Special algorithmic details
- G06T2207/20084—Artificial neural networks [ANN]
Abstract
The present invention relates to the method, apparatus of image concealing infomation detection, computer equipment and storage mediums, are applied to technical field of image detection.The described method includes: obtaining image to be detected；Image to be detected is inputted into trained anti-privacy and hides detection model；It is the convolutional neural networks model based on deep learning that privacy, which hides detection model,；The output that detection model is hidden according to anti-privacy, obtains the testing result of the hiding information of image to be detected；Wherein, anti-privacy is hidden detection model and is obtained by following steps training: every N in presetting database image patterns being set as a batch, N is the integer more than or equal to 2；Described image sample is the image comprising hiding containing information；Image pattern is inputted into convolutional neural networks model to be trained by batch and participates in training, until meeting training termination condition.The present invention is conducive to improve the detection accuracy of privacy information in image.
Description
Technical field
The present invention relates to technical field of image detection, method, image concealing more particularly to image concealing infomation detection
Device, computer equipment and the computer readable storage medium of infomation detection.
Background technique
Image privacy concealing technology be it is a kind of privacy information is embedded in image, to reach the skill of secret communication purpose
Art.Privacy information is hidden by it using structures such as image border points, the image comprising privacy information by common signal channel into
Row transmission, recipient extract privacy information according to key and specific extracting method from image.Image privacy is hidden
The characteristics of technology, is that it is possible to transmit privacy information in the case where not found by third party.However, image privacy concealing technology
Also it is utilized by certain criminals, they transmit illegal Information hiding by privacy concealing technology in the picture, are society
Meeting safety belt carrys out huge challenge.
Image privacy concealing technology usually utilizes marginal texture of image etc. to be embedded in privacy information；It is embedded in positions such as edges
Privacy information is visually difficult to find the difference for carrying close image and original image.In practice, the anti-privacy concealing technology of image passes through to figure
As piecemeal, the research of block structure is carried out, the privacy information being embedded in image is detected.
However, in the implementation of the present invention, following problem exists in the prior art in inventor: based on image point
The image privacy detection method of block is performed poor, it is difficult to hidden in image at effective detection when image insertion privacy information is less
The privacy information of hiding.
Summary of the invention
Based on this, it is necessary to for existing way when image insertion privacy information is less, it is difficult to image at effective detection
In the privacy information problem hidden, the method, apparatus of image concealing infomation detection a kind of, computer equipment and storage are provided and are situated between
Matter.
On the one hand, the embodiment of the present invention provides a kind of method of image concealing infomation detection, comprising:
Obtain image to be detected；
Described image to be detected is inputted into trained anti-privacy and hides detection model；The anti-privacy hides detection mould
Type is the convolutional neural networks model based on deep learning；
The output that detection model is hidden according to the anti-privacy, obtains the detection knot of the hiding information of described image to be detected
Fruit；
Wherein, the anti-privacy is hidden detection model and is obtained by following steps training:
Every N in presetting database image patterns are set as a batch, N is the integer more than or equal to 2；The figure
Decent is the image comprising hiding containing information；
Image pattern is inputted into convolutional neural networks model to be trained by batch and participates in training, is terminated until meeting training
Condition；
Image patterns all in the database are participated in into the convolutional neural networks model training and are once used as a training in rotation
Practice, every trained M wheel, the learning rate of the convolutional neural networks model is reduced according to setting ratio；M is whole more than or equal to 2
Number.
Include: in one of the embodiments,
The learning rate of the convolutional neural networks is initialized as 0.0001；Every training 10 is taken turns, the convolutional neural networks mould
The learning rate of type halves.
In one of the embodiments, further include:
Collect image pattern；
Described image sample is pre-processed, including by described image samples normalization be 256*256 pixel, Yi Jishe
The dimension for setting described image sample is 3 dimensions；
Privacy insertion processing is carried out to the image pattern after pretreatment, obtains carrying close image pattern；
According to multiple close image patterns of load, the database is constructed.
Described image to be detected is being inputted into the hiding detection model of trained anti-privacy in one of the embodiments,
Before, further includes:
Described image to be detected is pre-processed, its size and dimension one with image pattern in the database are made
It causes.
The anti-privacy hides detection model and includes: at least two convolutional layers, is truncated linearly in one of the embodiments,
Module, at least two regularization layers, at least two active coatings, at least two pond layers and 1 full articulamentum；
The convolution kernel size of at least two convolutional layer is 3*3；The interceptive value that linear block is truncated is 1.5；Swash
Layer living uses ReLU activation primitive；Pond layer is using average pond.
It in one of the embodiments, include: that the hiding detection model of the anti-privacy includes 5 convolutional layers, 5 convolutional layers
Intrinsic dimensionality be respectively 3,32,64,128,144；
And/or the full articulamentum uses 144 neurons.
Include: in one of the embodiments, the anti-privacy hide detection model include 4 regularization layers, to 4 activation
Layer, 4 pond layers.
On the other hand, the embodiment of the present invention provides a kind of device of hiding detection of image privacy, comprising:
Image collection module, for obtaining image to be detected；
Image detection module hides detection model for described image to be detected to be inputted trained anti-privacy；Institute
Stating anti-privacy to hide detection model is the convolutional neural networks model based on deep learning；
Privacy hides detection module, for hiding the output of detection model according to the anti-privacy, obtains described to be detected
The privacy of image hides testing result；
Wherein, the anti-privacy is hidden detection model and is obtained by following steps training:
Every N in presetting database image patterns are set as a batch, N is the integer more than or equal to 2；
Image pattern is inputted into convolutional neural networks model to be trained by batch and participates in training, is terminated until meeting training
Condition；
Image patterns all in the database are participated in into the convolutional neural networks model training and are once used as a training in rotation
Practice, every trained M wheel, the learning rate of the convolutional neural networks model is reduced according to setting ratio；M is whole more than or equal to 2
Number.
In another aspect, the embodiment of the present invention provides a kind of computer equipment, including memory and processor, the memory
It is stored with computer program, the processor realizes the image as described in above-mentioned any embodiment when executing the computer program
The method of steganographic detection.
In another aspect, the embodiment of the present invention provides a kind of computer storage medium, it is stored thereon with computer program, the journey
The method of the image concealing infomation detection as described in above-mentioned any embodiment is realized when sequence is executed by processor.
A technical solution in above-mentioned technical proposal has the following advantages that or the utility model has the advantages that for image to be detected, first
First based on the convolutional neural networks of deep learning, the survey model of inspection is hidden in building for anti-privacy, including will be in presetting database
Every N image patterns be set as a batch, N is the integer more than or equal to 2；Image pattern inputted by batch to be trained
Convolutional neural networks model participates in training, until meeting training termination condition；Image patterns all in the database are participated in
The convolutional neural networks model training is once used as a wheel to train, every trained M wheel, the study of the convolutional neural networks model
Rate is reduced according to setting ratio；M is the integer more than or equal to 2；The survey mould of inspection is hidden using the anti-privacy of aforesaid way building
Type detects the privacy information of image to be detected, can effectively detect the privacy information in image, especially scheme
When the privacy information being embedded in as in is less, the anti-privacy based on deep learning hides detection model, and deep learning is extracting weak spy
Sign aspect performance is good, thus significantly improves the detection accuracy for carrying privacy information in close image.
Detailed description of the invention
Fig. 1 is the schematic flow chart of the method for the image concealing infomation detection of an embodiment；
Fig. 2 is image to be detected of an embodiment and its schematic diagram of corresponding characteristic pattern；
Fig. 3 is that the anti-privacy of training of an embodiment hides the schematic flow chart of detection model；
Fig. 4 is that the anti-privacy of an embodiment hides the structural schematic diagram of detection model；
Fig. 5 is the schematic diagram of the device of the image concealing infomation detection of an embodiment；
Fig. 6 is the internal structure chart of computer equipment in one embodiment.
Specific embodiment
In order to make the objectives, technical solutions, and advantages of the present invention clearer, with reference to the accompanying drawings and embodiments, right
The present invention is further elaborated.It should be appreciated that the specific embodiments described herein are merely illustrative of the present invention, and
It is not used in the restriction present invention.
Referenced herein " embodiment " is it is meant that a particular feature, structure, or characteristic described can wrap in conjunction with the embodiments
It is contained at least one embodiment of the application.Each position in the description occur the phrase might not each mean it is identical
Embodiment, nor the independent or alternative embodiment with other embodiments mutual exclusion.Those skilled in the art explicitly and
Implicitly understand, embodiment described herein can be combined with other embodiments.
The method of image concealing infomation detection provided by the present application, can be applied to computer equipment.Wherein, computer is set
It is standby can include but is not limited to terminal device or server, the terminal device be various personal computers, laptop,
Smart phone, tablet computer and portable wearable device, server can use independent server either multiple servers
The server cluster of composition is realized.
In one embodiment, as shown in Figure 1, providing a kind of method of image concealing infomation detection, including following step
It is rapid:
S110 obtains image to be detected.
It may include the privacy informations such as text, number, symbol in the embodiment of the present invention, in image to be detected, these privacies
The position of information in the picture can be diversified, therefore can both have independent privacy insertion area in image to be detected
Domain can also have the privacy information being superimposed upon on Background.
Described image to be detected is inputted trained anti-privacy and hides detection model by S120；The anti-privacy is hidden
Detection model is the convolutional neural networks model based on deep learning.
In one embodiment, the anti-privacy is hidden detection model and is specifically used for: carrying out to input image to be detected more
Secondary process of convolution obtains the corresponding characteristic pattern of image to be detected, and identifies privacy information therein based on characteristic pattern.Characteristic pattern is
Refer to the medium object that neural network learns image, such as neural network includes multilayer neural network structure, feature
Figure refers to the image information that middle layer neural network structure obtains.It is shown in Figure 2, for image to be detected (a) of input,
Characteristic pattern shown in network structure available (b) in neural network.
S130 hides the output of detection model according to the anti-privacy, obtains the hiding information of described image to be detected
Testing result.
In the embodiment of the present invention, privacy information can be preset character information or graphical information, concrete form and
Content can be set according to actual scene.
In the embodiment of the present invention, the anti-privacy is hidden detection model and is obtained by following steps training:
Every N in presetting database image patterns are set as a batch, N is the integer more than or equal to 2；If last
The image pattern of one batch is less than N, little for trained influence.Described image sample is comprising hiding containing information
Image；By batch N image patterns are inputted into convolutional neural networks model to be trained together and participate in training, trained until meeting
Termination condition；In other cases, N image patterns can also be inputted into convolutional neural networks to be trained several times by batch
Model participates in training, until meeting training termination condition.And it in the training process, can also be by figures all in the database
Decent participates in the convolutional neural networks model training and a wheel is once used as to train, every trained M wheel, the convolutional neural networks
The learning rate of model is reduced according to setting ratio；M is the integer more than or equal to 2.
In the embodiment of the present invention, anti-privacy hide detection model testing result, may include in image to be detected whether
It is concealed with privacy information, can also include the location information of hiding privacy information in the picture, further, due to described
Anti- privacy hides the convolutional neural networks model that detection model is deep learning, therefore hides the detection of detection model in anti-privacy
It as a result can also include the particular content for the privacy information hidden in image in.
It in one embodiment, can also include that building is used for before hiding detection model to anti-privacy and being trained
The step of trained database, it can specifically include following steps:
S31 collects image pattern；
The image pattern of the collection step can be arbitrary image, and the content and pixel of image are not construed as limiting.
S32 pre-processes described image sample, including by described image samples normalization be 256*256 pixel, with
And the dimension of setting described image sample is 3 dimensions.
For color image sample, 3 dimensions respectively correspond tri- channels RGB, decent for gray level image sample and binary map
This, then can replicate pixel becomes 3 parts, thus sets 3 dimensions for the dimension of image pattern.
It is by this step, the size and dimension that are used for trained image pattern is unified, to eliminate different sizes and dimension
Brought training error.
S33 carries out privacy insertion processing to the image pattern after pretreatment, obtains carrying close image pattern.
The purpose of this step is to be purposefully added privacy information into image pattern, the content and form of the privacy information of addition
It is unlimited.Also, privacy insertion processing in embodiments of the present invention, is carried out to image pattern, obtains carrying the specific of close image pattern
Means are also not construed as limiting, including but not limited to FusedDM method.
S34 constructs the database according to multiple close image patterns of load.
The database obtained through the foregoing embodiment, the size for carrying close image pattern and dimension therein are unified, and carry
In close image pattern known to privacy information (content, insertion form, in embedded location at least one of known to), therefore be based on data
Effective training that anti-privacy hides detection model may be implemented in library.
Further, in some embodiments, after obtaining the database, further includes:
Every 20 image patterns in presetting database are set as a batch by S35；
Image pattern is inputted convolutional neural networks model to be trained by batch and participates in training, trained until meeting by S36
Termination condition；
In addition, being directed to convolutional neural networks model to be trained, anti-privacy can also be hidden the corresponding volume of detection model
The learning rate of product neural network is initialized as 0.0001；Every training 10 is taken turns, and the learning rate of convolutional neural networks model halves, thus
The training that anti-privacy hides the corresponding convolutional neural networks of detection model can be rapidly completed, shorten the model training time, and
Obtain more accurate result.The value of learning rate initialization is larger, is conducive to accelerate convergence, during subsequent school, learning rate by
It is decrescence small, be conducive to obtain more accurate result.
In some embodiments, the specific structure that anti-privacy hides detection model includes: at least two convolutional layers, transversal
Property module, at least two regularization layers, at least two active coatings, at least two pond layers and 1 full articulamentum.
In a concrete application, the structure that anti-privacy hides detection model be may refer to shown in Fig. 4, including 5 convolution
Layer, 1 truncation linear block, 4 regularization layers, 4 active coatings, 4 pond layers and 1 full articulamentum, the spy of 5 convolutional layers
Levying dimension can be respectively 3,32,64,128,144, and convolution kernel size is 3*3.In one example, first convolutional layer
Weight initialization is the filtering core of high-pass filter, and high pass filter filters core K1, K2, K3 of first convolutional layer specifically can be with
Are as follows:
The interceptive value that linear block is wherein truncated is 1.5, and truncation linear block can be expressed as:
Wherein T is interceptive value, can be with given threshold for 1.5.
Activation primitive is provided in active coating, common activation primitive includes sigmoid, tanh, ReLU, by activating letter
Number is to introduce non-linear factor.In one embodiment, the activation primitive of active coating can use ReLU function, i.e. line rectification
Function, also known as amendment linear unit, are common activation primitives in a kind of neural network, have that convergence is fast, asks gradient simple
Feature, calculation formula is also very simple, and for the negative of input, output is all 0, for the positive value of input, then exports as former state.
The activation primitive of gradient disappearance problem can be eliminated and corrected for other.
Pond layer main function is to compress the characteristic pattern inputted from upper one layer, on the one hand makes characteristic information quantitative change
It is small, simplify network query function complexity；On the one hand characteristic pattern compression is carried out, main feature is extracted；Common pond mode includes most
Great Chiization, summation pond and average pond.In one embodiment, the pond layer that anti-privacy is hidden in detection model can use
Average pond, pond layer convolution kernel size is 3*3 in addition to the last one pond layer, the last one pond layer convolution kernel is spy
Figure size is levied, global pool is carried out, pond layer step-length can be 2, and marginal portion is expanded by the way of duplication.
Each neuron in full articulamentum is connect entirely with all neurons of its preceding layer.Full articulamentum can be former
With the local message of class discrimination in one layer；In order to promote network performance, the excitation function of the complete each neuron of articulamentum
Generally use ReLU function, naturally it is also possible to can be avoided and correct the excitation function of gradient disappearance problem using other.This hair
In bright embodiment, full articulamentum may include 144 neurons, for identification the privacy in the corresponding characteristic pattern of image to be detected
Characteristic information.
Output layer, the privacy feature information for being obtained based on full articulamentum export the privacy letter of described image to be detected
Cease testing result.
In the embodiment of the present invention, the privacy information testing result of described image to be detected includes but is not limited to following at least one
: whether include privacy information in described image to be detected, the location information of privacy information in described image to be detected, it is described to
The embedded mode of privacy information in detection image, the content of privacy information in described image to be detected.
The method of the image concealing infomation detection of above-described embodiment is primarily based on deep learning for image to be detected
The survey model of inspection is hidden in convolutional neural networks, building for anti-privacy, including setting every N in presetting database image patterns
For a batch, N is the integer more than or equal to 2；Image pattern is inputted to convolutional neural networks model to be trained by batch
Training is participated in, until meeting training termination condition；Image patterns all in the database are participated in into the convolutional neural networks
Model training is once used as a wheel to train, and every trained M wheel, the learning rate of the convolutional neural networks model is dropped according to setting ratio
It is low；M is the integer more than or equal to 2；The survey model that inspection is hidden using the anti-privacy of aforesaid way building, to image to be detected
Privacy information detected, can effectively detect the privacy information in image, the privacy that is especially embedded in the picture letter
When ceasing less, the anti-privacy based on deep learning hides detection model, significantly improves the detection for carrying privacy information in close image
Precision.
It should be understood that for the various method embodiments described above, although each step in flow chart is according to arrow
Instruction is successively shown, but these steps are not that the inevitable sequence according to arrow instruction successively executes.Unless having herein bright
True explanation, there is no stringent sequences to limit for the execution of these steps, these steps can execute in other order.And
And at least part step in the flow chart of embodiment of the method may include multiple sub-steps or multiple stages, this is a little
Step or stage are not necessarily to execute completion in synchronization, but can execute at different times, these sub-steps
Perhaps the execution sequence in stage be also not necessarily successively carry out but can with the sub-step of other steps or other steps or
At least part in person's stage executes in turn or alternately.
Based on thought identical with the method for image concealing infomation detection in above-described embodiment, it is hidden that image is also provided herein
Hide the device of infomation detection.
In one embodiment, as shown in figure 5, the device of the image concealing infomation detection of the present embodiment includes:
Image collection module 501, for obtaining image to be detected；
Image detection module 502 hides detection model for described image to be detected to be inputted trained anti-privacy；
It is the convolutional neural networks model based on deep learning that the anti-privacy, which hides detection model,；
Privacy hides detection module 503, for hiding the output of detection model according to the anti-privacy, obtains described to be checked
The privacy of altimetric image hides testing result；
Wherein, the anti-privacy is hidden detection model and is obtained by following steps training:
Every N in presetting database image patterns are set as a batch, N is the integer more than or equal to 2；
Image pattern is inputted into convolutional neural networks model to be trained by batch and participates in training, is terminated until meeting training
Condition；
Image patterns all in the database are participated in into the convolutional neural networks model training and are once used as a training in rotation
Practice, every trained M wheel, the learning rate of the convolutional neural networks model is reduced according to setting ratio；M is whole more than or equal to 2
Number.
The device of the image concealing infomation detection of above-described embodiment, is primarily based on the convolutional neural networks of deep learning, structure
The survey model that inspection is hidden for anti-privacy is built, including every N in presetting database image patterns are set as a batch, N is
Integer more than or equal to 2；Image pattern is inputted into convolutional neural networks model to be trained by batch and participates in training, Zhi Daoman
Foot training termination condition；Image patterns all in the database are participated in into the primary conduct of convolutional neural networks model training
One wheel training, every trained M wheel, the learning rate of the convolutional neural networks model are reduced according to setting ratio；M be more than or equal to
2 integer；The survey model that inspection is hidden using the anti-privacy of aforesaid way building, examines the privacy information of image to be detected
It surveys, can effectively detect the privacy information in image, when the privacy information being especially embedded in the picture is less, be based on depth
The anti-privacy of study hides detection model, significantly improves the detection accuracy for carrying privacy information in close image.
In other embodiments, the learning rate of the above-mentioned convolutional neural networks is initialized as 0.0001；Every training 10
Wheel, the learning rate of the convolutional neural networks model halve.
In other embodiments, the device of above-mentioned image concealing infomation detection can also include: database sharing module,
The module includes:
Sample collection unit, for collecting image pattern；
Described image samples normalization for pre-processing to described image sample, including is by pretreatment unit
256*256 pixel, and the dimension of setting described image sample is 3 dimensions；
Close unit is carried, for carrying out privacy insertion processing to the image pattern after pretreatment, obtains carrying close image pattern；
And database sharing unit, for constructing the database according to multiple close image patterns of load.
In other embodiments, the device of above-mentioned image concealing infomation detection can also include:
Image processing module, in image detection module 502 that the input of described image to be detected is trained anti-hidden
Privacy hide detection model before, described image to be detected is pre-processed, make its in the database image pattern it is big
It is small consistent with dimension.
In other embodiments, it includes: at least two convolutional layers, the linear mould of truncation that above-mentioned anti-privacy, which hides detection model,
Block, at least two regularization layers, at least two active coatings, at least two pond layers and 1 full articulamentum；Described at least two
The convolution kernel size of convolutional layer is 3*3；The interceptive value that linear block is truncated is 1.5；Active coating uses ReLU activation primitive；
Pond layer is using average pond.Specific structure may refer to Fig. 4 and its corresponding embodiment, not repeat them here.
The specific restriction of device about image concealing infomation detection may refer to examine above for image concealing information
The restriction of the method for survey, details are not described herein.Modules in the device of above-mentioned image concealing infomation detection can whole or portion
Divide and is realized by software, hardware and combinations thereof.Above-mentioned each module can be embedded in the form of hardware or independently of computer equipment
In processor in, can also be stored in a software form in the memory in computer equipment, in order to processor calling hold
The corresponding operation of the above modules of row.
In addition, the logic of each program module is drawn in the embodiment of the device of the image concealing infomation detection of above-mentioned example
Divide and be merely illustrative of, can according to need in practical application, such as the configuration requirement of corresponding hardware or the reality of software
Existing convenient consideration is completed above-mentioned function distribution by different program modules, i.e., by the dress of described image steganographic detection
The internal structure set is divided into different program modules, to complete all or part of the functions described above.
In one embodiment, a kind of computer equipment is provided, which can be terminal, be also possible to take
Business device, internal structure chart can be as shown in Figure 6.The computer equipment includes the processor connected by system bus, storage
Device, network interface and database.Wherein, the processor of the computer equipment is for providing calculating and control ability.The computer
The memory of equipment includes non-volatile memory medium, built-in storage.The non-volatile memory medium is stored with operating system, meter
Calculation machine program and database.The built-in storage is that the operation of the operating system and computer program in non-volatile memory medium mentions
For environment.The database of the computer equipment is for storing image sensitive information detection device data.The net of the computer equipment
Network interface is used to communicate with external terminal by network connection.To realize a kind of figure when the computer program is executed by processor
As the method for steganographic detection.
It will be understood by those skilled in the art that structure shown in Fig. 6, only part relevant to application scheme is tied
The block diagram of structure does not constitute the restriction for the computer equipment being applied thereon to application scheme, specific computer equipment
It may include perhaps combining certain components or with different component layouts than more or fewer components as shown in the figure.
In one embodiment, a kind of computer equipment, including memory and processor are provided, is stored in memory
Computer program, the processor perform the steps of acquisition image to be detected when executing computer program；By the mapping to be checked
Detection model is hidden as inputting trained anti-privacy；It is the convolution based on deep learning that the anti-privacy, which hides detection model,
Neural network model；The output that detection model is hidden according to the anti-privacy, obtains the hiding information of described image to be detected
Testing result；
Wherein, the anti-privacy is hidden detection model and is obtained by following steps training: by every N in presetting database
Image pattern is set as a batch, and N is the integer more than or equal to 2；Described image sample is comprising hiding containing information
Image；Image pattern is inputted into convolutional neural networks model to be trained by batch and participates in training, terminates item until meeting training
Part.
Image patterns all in the database are participated in into the convolutional neural networks model training and are once used as a training in rotation
Practice, every trained M wheel, the learning rate of the convolutional neural networks model is reduced according to setting ratio；M is whole more than or equal to 2
Number.
In one embodiment, other above-mentioned image privacy information detection sides are also realized when processor executes computer program
Step in method embodiment.
The computer equipment of above-described embodiment is primarily based on the convolutional neural networks of deep learning for image to be detected,
The survey model of inspection is hidden in building for anti-privacy, including every N in presetting database image patterns are set as a batch, N
For the integer more than or equal to 2；Image pattern is inputted into convolutional neural networks model to be trained by batch and participates in training, until
Meet training termination condition；Image patterns all in the database are participated in the convolutional neural networks model training once to make
For a wheel training, every trained M wheel, the learning rate of the convolutional neural networks model is reduced according to setting ratio；M is to be greater than or wait
In 2 integer；The survey model that inspection is hidden using the anti-privacy of aforesaid way building carries out the privacy information of image to be detected
Detection, can effectively detect the privacy information in image, when the privacy information being especially embedded in the picture is less, based on deep
The anti-privacy of degree study hides detection model, significantly improves the detection accuracy for carrying privacy information in close image.
In one embodiment, a kind of computer readable storage medium is provided, computer program is stored thereon with, is calculated
Machine program performs the steps of acquisition image to be detected when being executed by processor；Described image to be detected is inputted by training
Anti- privacy hide detection model；It is the convolutional neural networks model based on deep learning that the anti-privacy, which hides detection model,；
The output that detection model is hidden according to the anti-privacy, obtains the testing result of the hiding information of described image to be detected；
Wherein, the anti-privacy is hidden detection model and is obtained by following steps training: by every N in presetting database
Image pattern is set as a batch, and N is the integer more than or equal to 2；Described image sample is comprising hiding containing information
Image；Image pattern is inputted into convolutional neural networks model to be trained by batch and participates in training, terminates item until meeting training
Part；
Image patterns all in the database are participated in into the convolutional neural networks model training and are once used as a training in rotation
Practice, every trained M wheel, the learning rate of the convolutional neural networks model is reduced according to setting ratio；M is whole more than or equal to 2
Number.
In one embodiment, above-mentioned other image privacy informations detection is also realized when computer program is executed by processor
Step in embodiment of the method.
The computer readable storage medium of above-described embodiment is primarily based on the convolution of deep learning for image to be detected
The survey model of inspection is hidden in neural network, building for anti-privacy, including every N in presetting database image patterns are set as one
A batch, N are the integer more than or equal to 2；Image pattern convolutional neural networks model to be trained is inputted by batch to participate in
Training, until meeting training termination condition；Image patterns all in the database are participated in into the convolutional neural networks model
Training is once trained as a wheel, and every trained M wheel, the learning rate of the convolutional neural networks model is reduced according to setting ratio；M
For the integer more than or equal to 2；The survey model that inspection is hidden using the anti-privacy of aforesaid way building, to the hidden of image to be detected
Personal letter breath is detected, and can effectively detect the privacy information in image, the privacy information being especially embedded in the picture compared with
When few, the anti-privacy based on deep learning hides detection model, significantly improves the detection accuracy for carrying privacy information in close image.
Those of ordinary skill in the art will appreciate that realizing all or part of the process in above-described embodiment method, being can be with
Relevant hardware is instructed to complete by computer program, the computer program can be stored in a non-volatile computer
In read/write memory medium, the computer program is when being executed, it may include such as the process of the embodiment of above-mentioned each method.Wherein,
To any reference of memory, storage, database or other media used in each embodiment provided herein,
Including non-volatile and/or volatile memory.Nonvolatile memory may include read-only memory (ROM), programming ROM
(PROM), electrically programmable ROM (EPROM), electrically erasable ROM (EEPROM) or flash memory.Volatile memory may include
Random access memory (RAM) or external cache.By way of illustration and not limitation, RAM is available in many forms,
Such as static state RAM (SRAM), dynamic ram (DRAM), synchronous dram (SDRAM), double data rate sdram (DDRSDRAM), enhancing
Type SDRAM (ESDRAM), synchronization link (Synchlink) DRAM (SLDRAM), memory bus (Rambus) direct RAM
(RDRAM), direct memory bus dynamic ram (DRDRAM) and memory bus dynamic ram (RDRAM) etc..
Each technical characteristic of above embodiments can be combined arbitrarily, for simplicity of description, not to above-described embodiment
In each technical characteristic it is all possible combination be all described, as long as however, the combination of these technical characteristics be not present lance
Shield all should be considered as described in this specification.
In the above-described embodiments, it all emphasizes particularly on different fields to the description of each embodiment, there is no the portion being described in detail in some embodiment
Point, it may refer to the associated description of other embodiments.
The term " includes " of embodiment hereof and " having " and their any deformations, it is intended that cover non-exclusive packet
Contain.Such as contain series of steps or the process, method, system, product or equipment of (module) unit are not limited to arrange
Out the step of or unit, but optionally further comprising the step of not listing or unit, or optionally further comprising for these mistakes
The intrinsic other step or units of journey, method, product or equipment.
Referenced herein " multiple " refer to two or more."and/or", the association for describing affiliated partner are closed
System indicates may exist three kinds of relationships, for example, A and/or B, can indicate: individualism A exists simultaneously A and B, individualism
These three situations of B.Character "/" typicallys represent the relationship that forward-backward correlation object is a kind of "or".
Referenced herein " first second " is used only as distinguishing, be not used to size, successively, the limit of subordinate etc.
It is fixed.It is to be appreciated that specific sequence or precedence can be interchanged in " first second " in the case where permission, alternatively, " the
One the second " distinguish object be interchangeable under appropriate circumstances so that the embodiments described herein can be in addition to scheming herein
Those of show or describe that the mode other than embodiment is implemented.
The several embodiments of the application above described embodiment only expresses, the description thereof is more specific and detailed, but simultaneously
It cannot therefore be construed as limiting the scope of the patent.It should be pointed out that coming for those of ordinary skill in the art
It says, without departing from the concept of this application, various modifications and improvements can be made, these belong to the protection of the application
Range.Therefore, the scope of protection shall be subject to the appended claims for the application patent.
Claims (10)
1. a kind of method of image concealing infomation detection characterized by comprising
Obtain image to be detected；
Described image to be detected is inputted into trained anti-privacy and hides detection model；The anti-privacy hides detection model
Convolutional neural networks model based on deep learning；
The output that detection model is hidden according to the anti-privacy, obtains the testing result of the hiding information of described image to be detected；
Wherein, the anti-privacy is hidden detection model and is obtained by following steps training:
Every N in presetting database image patterns are set as a batch, N is the integer more than or equal to 2；Described image sample
This is the image comprising hiding containing information；
Image pattern is inputted into convolutional neural networks model to be trained by batch and participates in training, terminates item until meeting training
Part；
Image patterns all in the database, which are participated in the convolutional neural networks model training, is once used as a wheel to train, often
Training M wheel, the learning rate of the convolutional neural networks model are reduced according to setting ratio；M is the integer more than or equal to 2.
2. the method as described in claim 1 characterized by comprising
The learning rate of the convolutional neural networks is initialized as 0.0001；
Every training 10 is taken turns, and the learning rate of the convolutional neural networks model halves.
3. the method as described in claim 1, which is characterized in that further include:
Collect image pattern；
Described image sample is pre-processed, including by described image samples normalization be 256*256 pixel, and setting institute
The dimension for stating image pattern is 3 dimensions；
Privacy insertion processing is carried out to the image pattern after pretreatment, obtains carrying close image pattern；
According to multiple close image patterns of load, the database is constructed.
4. method as claimed in claim 3, which is characterized in that described image to be detected is being inputted trained anti-privacy
Before hiding detection model, further includes:
Described image to be detected is pre-processed, keeps it consistent with the size of image pattern in the database and dimension.
5. such as the described in any item methods of Claims 1-4, which is characterized in that the anti-privacy hide detection model include: to
Few two convolutional layers, truncation linear block, at least two regularization layers, at least two active coatings, at least two pond layers and 1
A full articulamentum；
The convolution kernel size of at least two convolutional layer is 3*3；The interceptive value that linear block is truncated is 1.5；Active coating
Use ReLU activation primitive；Pond layer is using average pond.
6. method as claimed in claim 5 characterized by comprising it includes 5 convolution that the anti-privacy, which hides detection model,
Layer, the intrinsic dimensionality of 5 convolutional layers is respectively 3,32,64,128,144；
And/or the full articulamentum uses 144 neurons.
7. the method as described in claim 1 characterized by comprising it includes 4 regularizations that the anti-privacy, which hides detection model,
Layer, to 4 active coatings, 4 pond layers.
8. the device that a kind of image privacy hides detection characterized by comprising
Image collection module, for obtaining image to be detected；
Image detection module hides detection model for described image to be detected to be inputted trained anti-privacy；It is described anti-
It is the convolutional neural networks model based on deep learning that privacy, which hides detection model,；
Privacy hides detection module, for hiding the output of detection model according to the anti-privacy, obtains described image to be detected
Privacy hide testing result；
Wherein, the anti-privacy is hidden detection model and is obtained by following steps training:
Every N in presetting database image patterns are set as a batch, N is the integer more than or equal to 2；
Image pattern is inputted into convolutional neural networks model to be trained by batch and participates in training, terminates item until meeting training
Part；
Image patterns all in the database, which are participated in the convolutional neural networks model training, is once used as a wheel to train, often
Training M wheel, the learning rate of the convolutional neural networks model are reduced according to setting ratio；M is the integer more than or equal to 2.
9. a kind of computer equipment, including memory and processor, the memory are stored with computer program, feature exists
In when the processor executes described program the step of any one of realization claim 1 to 7 the method.
10. a kind of computer readable storage medium, is stored thereon with computer program, which is characterized in that the program is by processor
The step of any one of claim 1 to 7 the method is realized when execution.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910688378.8A CN110517227A (en) | 2019-07-29 | 2019-07-29 | Method, apparatus, computer equipment and the storage medium of image concealing infomation detection |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910688378.8A CN110517227A (en) | 2019-07-29 | 2019-07-29 | Method, apparatus, computer equipment and the storage medium of image concealing infomation detection |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN110517227A true CN110517227A (en) | 2019-11-29 |
Family
ID=68623555
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201910688378.8A Pending CN110517227A (en) | 2019-07-29 | 2019-07-29 | Method, apparatus, computer equipment and the storage medium of image concealing infomation detection |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN110517227A (en) |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20080190271A1 (en) * | 2007-02-14 | 2008-08-14 | Museami, Inc. | Collaborative Music Creation |
| US20120266740A1 (en) * | 2011-04-19 | 2012-10-25 | Nathan Hilbish | Optical electric guitar transducer and midi guitar controller |
| CN108764270A (en) * | 2018-04-03 | 2018-11-06 | 上海大学 | A kind of Information Hiding & Detecting method integrated using convolutional neural networks |
| CN109934761A (en) * | 2019-01-31 | 2019-06-25 | 中山大学 | Jpeg image steganalysis method based on convolutional neural networks |
-
2019
- 2019-07-29 CN CN201910688378.8A patent/CN110517227A/en active Pending
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20080190271A1 (en) * | 2007-02-14 | 2008-08-14 | Museami, Inc. | Collaborative Music Creation |
| US20120266740A1 (en) * | 2011-04-19 | 2012-10-25 | Nathan Hilbish | Optical electric guitar transducer and midi guitar controller |
| CN108764270A (en) * | 2018-04-03 | 2018-11-06 | 上海大学 | A kind of Information Hiding & Detecting method integrated using convolutional neural networks |
| CN109934761A (en) * | 2019-01-31 | 2019-06-25 | 中山大学 | Jpeg image steganalysis method based on convolutional neural networks |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109522872A (en) | A kind of face identification method, device, computer equipment and storage medium | |
| CN103116763B (en) | A kind of living body faces detection method based on hsv color Spatial Statistical Character | |
| CN110414670A (en) | A kind of image mosaic tampering location method based on full convolutional neural networks | |
| CN109492627B (en) | Scene text erasing method based on depth model of full convolution network | |
| CN109558832A (en) | A kind of human body attitude detection method, device, equipment and storage medium | |
| US11354797B2 (en) | Method, device, and system for testing an image | |
| CN112163638B (en) | Method, device, equipment and medium for defending image classification model back door attack | |
| CN107491771A (en) | Method for detecting human face and device | |
| CN111754396B (en) | Face image processing method, device, computer equipment and storage medium | |
| CN111639558B (en) | Finger vein authentication method based on ArcFace Loss and improved residual error network | |
| CN109816612A (en) | Image enchancing method and device, computer readable storage medium | |
| CN108596818A (en) | A kind of image latent writing analysis method based on multi-task learning convolutional neural networks | |
| Wang et al. | Image splicing detection based on convolutional neural network with weight combination strategy | |
| CN108154133B (en) | Face portrait-photo recognition method based on asymmetric joint learning | |
| CN109977832A (en) | A kind of image processing method, device and storage medium | |
| CN113011253B (en) | Facial expression recognition method, device, equipment and storage medium based on ResNeXt network | |
| CN113468996A (en) | Camouflage object detection method based on edge refinement | |
| CN115564993A (en) | Lip print image classification algorithm based on multi-scale feature fusion and attention mechanism | |
| Fu et al. | CCNet: CNN model with channel attention and convolutional pooling mechanism for spatial image steganalysis | |
| CN111476727B (en) | Video motion enhancement method for face-changing video detection | |
| CN114596608A (en) | Double-stream video face counterfeiting detection method and system based on multiple clues | |
| Han et al. | Spatial steganalysis based on non-local block and multi-channel convolutional networks | |
| Li et al. | High-capacity coverless image steganographic scheme based on image synthesis | |
| CN109801224A (en) | A kind of image processing method, device, server and storage medium | |
| CN112464873A (en) | Model training method, face living body recognition method, system, device and medium |
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 | ||
| RJ01 | Rejection of invention patent application after publication |
Application publication date: 20191129 |
|
| RJ01 | Rejection of invention patent application after publication |