CN110458185A - Image-recognizing method and device, storage medium, computer equipment - Google Patents
Image-recognizing method and device, storage medium, computer equipment Download PDFInfo
- Publication number
- CN110458185A CN110458185A CN201910559070.3A CN201910559070A CN110458185A CN 110458185 A CN110458185 A CN 110458185A CN 201910559070 A CN201910559070 A CN 201910559070A CN 110458185 A CN110458185 A CN 110458185A
- Authority
- CN
- China
- Prior art keywords
- image
- network model
- differentiation
- trained
- sample set
- 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
- 238000000034 method Methods 0.000 title claims abstract description 38
- 238000003860 storage Methods 0.000 title claims abstract description 18
- 230000004069 differentiation Effects 0.000 claims abstract description 100
- 238000012549 training Methods 0.000 claims description 27
- 238000005242 forging Methods 0.000 claims description 11
- 238000010276 construction Methods 0.000 claims description 8
- 238000004590 computer program Methods 0.000 claims description 7
- 238000003909 pattern recognition Methods 0.000 claims description 5
- 238000005516 engineering process Methods 0.000 description 21
- 230000004913 activation Effects 0.000 description 17
- 230000006870 function Effects 0.000 description 12
- 238000004422 calculation algorithm Methods 0.000 description 6
- 238000010586 diagram Methods 0.000 description 4
- 238000003475 lamination Methods 0.000 description 4
- 238000004364 calculation method Methods 0.000 description 3
- 238000004891 communication Methods 0.000 description 3
- 241001269238 Data Species 0.000 description 2
- 241000209149 Zea Species 0.000 description 2
- 235000005824 Zea mays ssp. parviglumis Nutrition 0.000 description 2
- 235000002017 Zea mays subsp mays Nutrition 0.000 description 2
- 230000008901 benefit Effects 0.000 description 2
- 230000008859 change Effects 0.000 description 2
- 235000005822 corn Nutrition 0.000 description 2
- 238000009826 distribution Methods 0.000 description 2
- 238000007477 logistic regression Methods 0.000 description 2
- 238000005457 optimization Methods 0.000 description 2
- 230000008569 process Effects 0.000 description 2
- 238000013527 convolutional neural network Methods 0.000 description 1
- 230000006735 deficit Effects 0.000 description 1
- 230000001419 dependent effect Effects 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 230000000694 effects Effects 0.000 description 1
- 239000011159 matrix material Substances 0.000 description 1
- 230000001537 neural effect Effects 0.000 description 1
- 238000007781 pre-processing Methods 0.000 description 1
- 230000003362 replicative effect Effects 0.000 description 1
- 238000011160 research Methods 0.000 description 1
- 238000012795 verification Methods 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F21/00—Security arrangements for protecting computers, components thereof, programs or data against unauthorised activity
- G06F21/60—Protecting data
- G06F21/64—Protecting data integrity, e.g. using checksums, certificates or signatures
-
- 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
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- General Engineering & Computer Science (AREA)
- Health & Medical Sciences (AREA)
- Life Sciences & Earth Sciences (AREA)
- Artificial Intelligence (AREA)
- Software Systems (AREA)
- Evolutionary Computation (AREA)
- General Health & Medical Sciences (AREA)
- Computing Systems (AREA)
- Computer Security & Cryptography (AREA)
- Computational Linguistics (AREA)
- Mathematical Physics (AREA)
- Biophysics (AREA)
- Biomedical Technology (AREA)
- Molecular Biology (AREA)
- Bioinformatics & Computational Biology (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Evolutionary Biology (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Bioethics (AREA)
- Computer Hardware Design (AREA)
- Image Analysis (AREA)
Abstract
This application discloses image-recognizing method and device, storage medium, computer equipments, are related to image identification technical field, can promote image recognition accuracy.Wherein method is included: and is fought to generate trained generation network model in network model using depth convolution, is generated according to altered data and is forged image;Trained differentiation network model in generation network model is fought to depth convolution using the image discriminating sample set being made of forgery image generated and preset true picture to be trained, and obtains final differentiation network model;Target image is identified using final differentiation network model, determines that the target image is to forge image or true picture.The application is suitable for being public security, and the image forensics of the departments such as court provide higher reliability.
Description
Technical field
This application involves image identification technical field, particularly with regard to image-recognizing method and device, storage medium and
Computer equipment.
Background technique
With the fast development of computer technology, computer software can make or be spliced into that details is life-like, level
Clearly demarcated forgery image, it is extremely similar to the true picture that digital camera is shot, visually it is difficult to distinguish.And it forges
Image occurs gradually over the every field such as politics, military affairs, the news of society, brings great harm to society.Therefore, to image
The true and false evidence obtaining research it is particularly significant.
In traditional image forensics technology, verification information is added in advance in active forensic technologies needs in the picture, and right
Prior information is not contained in the image that most of application scenarios are got, therefore active forensic technologies have biggish limitation
Property;Existing passive blind forensic technologies, rely primarily on image statistics or shallow-layer characteristic information, such as gray value, grey scale change
Deng existing passive blind forensic technologies are very dependent on the selection of shallow-layer feature, and the quality of shallow-layer feature is to image recognition result
Accuracy be affected, additionally due to passive blind forensic technologies need it is a large amount of forge sample, and forge the foundation of sample set
It generally requires and is accomplished manually, take a substantial amount of time and energy, cost of labor are higher.
Summary of the invention
In view of this, this application provides image-recognizing method and device, storage medium, computer equipment, main purpose
It is that solving existing passive blind forensic technologies excessively relies on image statistics or shallow-layer characteristic information, the standard of image recognition result
Exactness is lower, and constructs the corresponding technical problem with high labor costs for forging sample set.
According to the one aspect of the application, a kind of image-recognizing method is provided, this method comprises:
It is fought using depth convolution and generates trained generation network model in network model, generated according to altered data pseudo-
Make image;
Using the image discriminating sample set being made of forgery image generated and preset true picture to depth convolution
Confrontation generates trained differentiation network model in network model and is trained, and obtains final differentiation network model;
Target image is identified using final differentiation network model, determine the target image be forge image or
Person's true picture.
According to the another aspect of the application, a kind of pattern recognition device is provided, which includes:
Generation module, for generating trained generation network model in network model using the confrontation of depth convolution, according to
Altered data, which generates, forges image;
Training module, for utilizing the image discriminating sample being made of forgery image generated and preset true picture
Collection fights trained differentiation network model in generation network model to depth convolution and is trained, and obtains final differentiation network
Model;
Identification module determines the target figure for identifying using final differentiation network model to target image
It seem to forge image or true picture.
According to the application another aspect, a kind of storage medium is provided, computer program, described program are stored thereon with
Above-mentioned image-recognizing method is realized when being executed by processor.
According to the application another aspect, a kind of computer equipment is provided, including storage medium, processor and be stored in
On storage medium and the computer program that can run on a processor, the processor realize above-mentioned image when executing described program
Recognition methods.
By above-mentioned technical proposal, image-recognizing method and device provided by the present application, storage medium, computer equipment,
With existing based on active forensic technologies, compared with the technical solution of passive blind forensic technologies identification image true-false, the application is using deeply
Spend convolution confrontation generate network model in trained generation network model, according to altered data generate forge image, using by
The image discriminating sample set generated for forging image and preset true picture composition is to depth convolution confrontation generation network mould
Trained differentiation network model is trained in type, obtains final differentiation network model, to utilize final differentiation net
Network model identifies target image, determines that the target image is to forge image or true picture.As it can be seen that passing through training
Good generation network model generates the forgery image for meeting image discriminating sample distribution, is generated with will pass through a small amount of forgery image
It is a large amount of to forge image, it better solves foundation and forges sample set technical problem with high labor costs;In addition, being rolled up using depth
Product confrontation generates differentiation network model recognition target image final in network model, can better solve passive blind evidence obtaining skill
Art excessively relies on the technical problems such as robustness is poor of image shallow-layer characteristic information and network model, is effectively ensured and final sentences
The robustness of the accuracy of other network model identification image true-false and final differentiation network model.
Above description is only the general introduction of technical scheme, in order to better understand the technological means of the application,
And it can be implemented in accordance with the contents of the specification, and in order to allow above and other objects, features and advantages of the application can
It is clearer and more comprehensible, below the special specific embodiment for lifting the application.
Detailed description of the invention
The drawings described herein are used to provide a further understanding of the present application, constitutes part of this application, this Shen
Illustrative embodiments and their description please are not constituted an undue limitation on the present application for explaining the application.In the accompanying drawings:
Fig. 1 shows a kind of flow diagram of image-recognizing method provided by the embodiments of the present application;
Fig. 2 shows the flow diagrams of another image-recognizing method provided by the embodiments of the present application;
Fig. 3 shows a kind of structural schematic diagram of pattern recognition device provided by the embodiments of the present application.
Specific embodiment
The application is described in detail below with reference to attached drawing and in conjunction with the embodiments.It should be noted that not conflicting
In the case of, the features in the embodiments and the embodiments of the present application can be combined with each other.
For existing based on active forensic technologies, during passive blind forensic technologies identification image true-false, actively collect evidence
There are the limitation that the image got does not contain prior information, and passive blind forensic technologies, excessively dependence image system in technology
Characteristic or shallow-layer characteristic information are counted, is affected to the accuracy of image recognition result, and constructs corresponding forgery sample set
Technical problem with high labor costs.A kind of image-recognizing method is present embodiments provided, can effectively be avoided existing passive blind
It causes the accuracy of image recognition result lower during forensic technologies identification image, and constructs corresponding forgery sample set
Technical problem with high labor costs, so that the accuracy of the image recognition true and false is effectively promoted, as shown in Figure 1, this method comprises:
101, it is fought using depth convolution and generates trained generation network model in network model, it is raw according to altered data
At forgery image.
The confrontation of depth convolution generates network (DCGAN:Deep Convolutional Generative Adversarial
It Networks) include generating network model and differentiating network model, while training generates network model and differentiates network model, one
Aspect generates network model keeps the gap between its forgery image generated and true picture as small as possible by training, to take advantage of
Deceive differentiation network model;On the other hand, differentiate that network model makes its target figure for differentiating input accurate as much as possible by training
The true and false of picture.
In the present embodiment, it is reversed convolutional Neural that the confrontation of depth convolution, which generates and generates network model in network DCGAN,
Network model, totally 5 layers, specifically:
1) first layer is input layer, is Normal Distribution, the data dimension holding one of input layer quantity and input
It causes.For example, input data is 100 dimension datas, input layer quantity is also 100.
2) second layer is warp lamination, and input data is the output of first layer as a result, setting its convolution kernel size as 4*
4, filter is 64*8, is input in activation primitive after carrying out batch regularization, and activation primitive is ReLU function.
3) third layer is warp lamination, and input data is the output of the second layer as a result, setting its convolution kernel size as 4*
4, step-length 2*2, filter are 64*4, are input in activation primitive after carrying out batch regularization, and activation primitive is ReLU letter
Number.
4) the 4th layer is warp lamination, and input data is the output of third layer as a result, setting its convolution kernel size as 4*
4, step-length 2*2, filter are 64, are input in activation primitive after carrying out batch regularization, and activation primitive is ReLU function.
5) layer 5 is warp lamination, and output result is used to construct the image discriminating sample set for differentiating network model, if
Its fixed convolution kernel size is 4*4, and step-length 2*2, filter is 64, is input in activation primitive, and activation primitive is Tanh letter
Number.
102, using the image discriminating sample set being made of forgery image generated and preset true picture to depth
Convolution confrontation generates trained differentiation network model in network model and is trained, and obtains final differentiation network model.
In the present embodiment, the confrontation of depth convolution, which generates, differentiates that network model is convolutional neural networks mould in network DCGAN
Type, totally 5 layers, specifically:
1) first layer is input layer, sets the matrix specification of the data vector of its input as 64*64*3, convolution kernel size is
4*4, activation primitive LeakyReLU.Wherein, the calculation formula of activation primitive LeakyReLU specifically:
Wherein, xiFor the data vector of input, yiFor the data that obtain that treated of output after being calculated via activation primitive to
Amount, aiIt is the preset parameter in the section (1 ,+∞).
2) second layer is convolutional layer, and input data is the output of first layer as a result, setting its convolution kernel size as 4*4,
Filter is 64*2, is input in activation primitive after carrying out batch standardization, activation primitive LeakyReLU.
3) third layer is convolutional layer, and input data is the output of the second layer as a result, setting its convolution kernel size as 4*4,
Step-length is 2*2, and filter is 64*4, is input in activation primitive after carrying out batch standardization, activation primitive is
LeakyReLU。
4) the 4th layer is convolutional layer, and input data is the output of third layer as a result, setting its convolution kernel size as 4*4,
Step-length is 2*2, and filter is 64*8, is input in activation primitive after carrying out batch standardization, activation primitive is
LeakyReLU。
5) layer 5 is convolutional layer, sets its convolution kernel size as 4*4, and filter is 1, is obtained after carrying out smooth operation
Export result.
103, target image is identified using final differentiation network model, determines that the target image is forgery figure
Picture or true picture.
In the present embodiment, target image is inputted to final differentiation network model, if output result is infinitely close to 0,
Differentiate the target image then to forge image;If output result is infinitely close to 1, differentiate that the target image is true picture.
In the scene of practical application, set and forge discriminant value as a, if export result (0, a] in range, then differentiate the target image
To forge image;If export result [b, 1) in range, then differentiate that the target image is true picture, forgery do not sentenced herein
The progress of not value and true discriminant value specifically limits.
The present embodiment can be fought using depth convolution according to above scheme and generate trained life in network model
At network model, is generated according to altered data and forge image, using by forgery image generated and preset true picture structure
At image discriminating sample set to depth convolution fight generate network model in trained differentiation network model be trained, obtain
The mesh is determined to final differentiation network model to identify using final differentiation network model to target image
Logo image is to forge image or true picture, identifies image based on active forensic technologies, passive blind forensic technologies with existing
The technical solution of the true and false is compared, and the present embodiment makes to differentiate that network model has preferable differentiation energy by the learning training of early period
Power still can individually be trained differentiation network model, in the case where generation network model remains unchanged to differentiate
Network model is adaptively from image discriminating sample focusing study its internal statistical rule, to improve final differentiation network mould
The generalization ability of type.
Further, as the refinement and extension of above-described embodiment specific embodiment, in order to completely illustrate the present embodiment
Specific implementation process, provide another image-recognizing method, as shown in Fig. 2, this method comprises:
201, differentiate that sample set fights depth convolution using first be made of noise variance and true picture and generate net
Initial differentiation network model in network model is trained, and obtains the first differentiation network model.
In the present embodiment, differentiate that network model is trained to initial, obtain the first differentiation network model, it is specific to wrap
Include: using noise variance and true picture as the initial input data for differentiating network model, and using obtained output result as
The input data of logistic regression output function;Further, the penalty values d_ of true picture is obtained using first-loss function
Loss_real, and utilize gradient ascent algorithm training initial network parameter θd, so that output result is infinitely close to 1, thus
Network model is differentiated to first.
Wherein, first-loss function are as follows:
Wherein, xiAnd ziRespectively true picture and noise variance, m are the first differentiation sample size, D (xi) it is initially to sentence
Other network model, D (G (zi)) it is to be initially generated network model.
Utilize gradient ascent algorithm training initial network parameter θdCalculation formula are as follows:
When output result is infinitely close to 1, using the initial network parameter after optimization as first network parameter.
202, differentiate that sample set differentiates network mould to described first using by the second of noise variance and forgery image construction
Type is trained, and obtains the second differentiation network model.
In the present embodiment, network model, which is trained, to be differentiated to initialization, obtains the first differentiation network model, it is specific to wrap
It includes: using noise variance and forging image as the input data of the first differentiation network model, and result will be exported and returned as logic
Return the input data of output function;Further, obtain forging the penalty values d_loss_ of image using the second loss function
Fake, and utilize gradient descent algorithm training first network parameter θd, so that output result is infinitely close to 0, so that it is determined that the
Two differentiate the second network parameter θ of network modeldAnd second differentiate network model.
Wherein, the second loss function are as follows:
Wherein, yiTo forge image, m is the second differentiation sample size, D (xi) it is the first differentiation network model, D (G (zi))
To be initially generated network model.
Utilize gradient descent algorithm training first network parameter θdCalculation formula are as follows:
In the scene of practical application, the second obtained differentiation network model can be used as trained differentiation network mould
Type, so as to the image discriminating sample forging image and preset true picture and constituting generated using trained generation network model
This collection further trains the trained differentiation network model, so that final differentiation network model is obtained, with reality
The now identification to image and true picture is forged.
203, differentiate that sample set differentiates network mould to described second using the third being made of noise variance and true picture
Type is trained, and is obtained third and is differentiated network model.
204, differentiate that sample set differentiates network mould to the third using by the 4th of noise variance and forgery image construction
Type is trained, and obtains trained differentiation network model.
In the present embodiment, third differentiates that sample set can be identical with the first differentiation sample set, can also be according to actually answering
Needs adjust accordingly;Correspondingly, the 4th differentiates that sample set differentiates that sample set can be identical with second, can also basis
The needs of practical application adjust accordingly, and, first differentiates that sample size, second differentiate that sample size, third differentiate sample
This quantity, the 4th differentiation sample size can also adjust accordingly according to the needs of practical application, not differentiate herein to third
Sample set and first differentiates that sample set and the 4th differentiation sample set and second differentiate sample set and the first differentiation sample number
Amount, the second differentiation sample size, third differentiate that sample size, the 4th differentiation sample size are specifically limited.
205, sample set is generated using first be made of noise variance to fight in generation network model depth convolution
It is initially generated network model to be trained, obtains trained generation network model.
In the present embodiment, be trained to being initially generated network model, obtain it is trained at network model, it is specific to wrap
It includes: will be used to train the noise variance for generating network model as the input data for being initially generated network model, for example, noise becomes
Amount is 100 dimension datas, and using obtained output result as the input data of logistic regression output function;Further, it utilizes
The loss function for generating network model obtains forging image impairment value d_loss, and utilizes gradient descent algorithm, passes through minimum
It is initially generated the penalty values g_loss of network model, training obtains the trained network parameter θ for generating network modelg, so as to
The forgery image of output is input to trained differentiation network model, and obtained output result is infinitely close to 1, to be instructed
The generation network model perfected, for reducing the trained discriminating power for differentiating network model.
Wherein, the loss function of network model is generated are as follows:
Utilize gradient descent algorithm training network parameter θgFormula are as follows:
206, it is fought using depth convolution and generates trained generation network model in network model, it is raw according to altered data
At forgery image.
It, can be right in order to make to differentiate that the discriminating power of network model reaches better effect in the scene of practical application
Trained generation network model is further optimized.For example, using altered data to trained generation network model
Further optimization training is carried out, the generation network model optimized is obtained, is forged to further be generated according to altered data
Image constructs image discriminating sample set, generates trained differentiation network mould in network model to realize to fight depth convolution
Type advanced optimizes.
207, using the image discriminating sample set being made of forgery image generated and preset true picture to depth
Convolution confrontation generates trained differentiation network model in network model and is trained, and obtains final differentiation network model.
In the present embodiment, trained generation network model or the puppet for generating network model and generating optimized are utilized
Image is made, and the true picture got, constructs image discriminating sample set.Using constructed image discriminating sample set to instruction
The differentiation network model perfected is trained, and by minimizing the trained penalty values d_loss for differentiating network model, is obtained
The network parameter θ of final differentiation network modeld, to obtain final differentiation network model.
208, the target signature in the images to be recognized got is identified and is intercepted, obtain corresponding to the target spy
The target image of sign.
209, the image further feature of the target image is obtained.
In the present embodiment, the images to be recognized got is pre-processed, specifically, to the mesh in images to be recognized
Mark feature is identified, is intercepted to the target signature recognized, and carry out ruler according to a certain percentage to the image being truncated to
Very little adjustment obtains the target image for characterizing target signature.Wherein, according to the needs of practical application scene, target image
Image further feature can be profile, texture, light and shade, color and combinations thereof and corresponding high-level semantic and combinations thereof.
210, the target image is identified according to the deep layer characteristics of image got, determines that the target image is
Forge image or true picture.
In order to illustrate the specific embodiment of step 210, as a kind of preferred embodiment, step 210 be can specifically include:
If the altered data is that duplication paste type image data, fuzzy retouching types of image data or computer generate type
Image data correspondingly identifies target image using final differentiation network model, determines that the target image is pseudo-
Image is made, then corresponding forgery image type is respectively to replicate paste type image, fuzzy retouching types of image or computer
Generate types of image.
In the present embodiment, if differentiating that the noise variance of network model does not set data type for training, only to training
Good generation network model or the altered data of the generation network model optimized input set data type and paste as duplication
Type or fuzzy retouching type or computer generate type, then final differentiation network model is for determining target image
To forge image or true picture, and for determining that target image be that forge the image type of image be respectively to replicate stickup
Types of image, fuzzy retouching types of image or computer generate types of image.
According to the needs of practical application scene, the data class of the noise variance of training differentiation network model can also will be used for
Type is set as replicating paste type or fuzzy retouching type or computer generates type, so that is made final sentences
Other network model is more stable, rapidly carries out genuine/counterfeit discriminating to target image, is public security, the image forensics of the departments such as court mention
For higher reliability.
Further, since duplication paste type image, fuzzy retouching types of image or computer generate types of image and exist
Shared deep layer characteristics of image, therefore, noise variance type and trained generation net for training differentiation network model
Network model or the altered data of the generation network model optimized input can not also set data type, final differentiation net
Network model also can be used in determining target image to forge image or true picture, and for determining target image for forgery
The image type of image is respectively to replicate paste type image, fuzzy retouching types of image or computer to generate types of image.
Herein without specifically limiting.
Technical solution by applying this embodiment is fought using depth convolution and generates trained generation in network model
Network model generates according to altered data and forges image, constitutes using by forgery image generated and preset true picture
Image discriminating sample set to depth convolution fight generate network model in trained differentiation network model be trained, obtain
Final differentiation network model determines the target to identify using final differentiation network model to target image
Image is to forge image or true picture.Image true-false is identified based on active forensic technologies, passive blind forensic technologies with existing
Technical solution compare, the present embodiment lead to too small amount of forgery image generate it is a large amount of forge image, it is pseudo- to better solve foundation
Sample set technical problem with high labor costs is made, and generates differentiation net final in network model using the confrontation of depth convolution
Network model recognition target image can effectively ensure that the final accuracy for differentiating network model identification image true-false and final
Differentiation network model robustness.
Further, the specific implementation as Fig. 1 method, the embodiment of the present application provide a kind of pattern recognition device, such as
Shown in Fig. 3, which includes: generation module 35, training module 36, identification module 37.
Generation module 35 can be used for generating trained generation network mould in network model using the confrontation of depth convolution
Type generates according to altered data and forges image;The generation module 35 be the present apparatus identify images to be recognized be forge image or
The basic module of true picture.
Training module 36 can be used for sentencing using the image being made of forgery image generated and preset true picture
Other sample set, which fights depth convolution, to be generated trained differentiation network model in network model and is trained, and obtains final sentencing
Other network model;The training module 36 is that the present apparatus identifies that images to be recognized is to forge the major function of image or true picture
The corn module of module and the present apparatus.
Identification module 37 can be used for identifying target image using final differentiation network model, described in determination
Target image is to forge image or true picture;The identification module 37 be the present apparatus identify images to be recognized be forge image or
The main functional modules of person's true picture and the corn module of the present apparatus.
It further include the first discriminative training module 31 or the second discriminative training module 32 in specific application scenarios,
One discriminative training module 31 can be used for differentiating that sample set rolls up depth using first be made of noise variance and true picture
The initial differentiation network model that product confrontation generates in network model is trained, and obtains the first differentiation network model;And it utilizes
The first differentiation network model is trained by noise variance and the second differentiation sample set for forging image construction, is instructed
The differentiation network model perfected.
Second discriminative training module 32 can be used for differentiating sample using first be made of noise variance and true picture
Collect and the initial differentiation network model in depth convolution confrontation generation network model is trained, obtains the first differentiation network mould
Type;And using by noise variance and forge image construction second differentiate sample set to it is described first differentiate network model into
Row training, obtains the second differentiation network model;And sample set is differentiated using the third being made of noise variance and true picture
The second differentiation network model is trained, third is obtained and differentiates network model;And using by noise variance and forgery
4th differentiation sample set of image construction differentiates that network model is trained to the third, obtains trained differentiation network mould
Type.
Further include the first generation training module 33 in specific application scenarios, can be used for using by noise variance structure
At first generation sample set to depth convolution fight generate network model in the network model that is initially generated be trained, obtain
Trained generation network model.
In specific application scenarios, further includes preprocessing module 34, can be used for in the images to be recognized got
Target signature identified and intercepted, obtain the target image for corresponding to the target signature.
In specific application scenarios, if the altered data is duplication paste type image data, fuzzy retouching type
Image data or computer generate types of image data, accordingly, using final differentiation network model to target image into
Row identification, determine the target image be forge image, then corresponding forgerys image type be respectively duplication paste type image,
Fuzzy retouching types of image or computer generate types of image.
In specific application scenarios, identification module 37 specifically can be used for obtaining the image deep layer of the target image
Feature;The target image is identified according to the deep layer characteristics of image got, determines that the target image is forgery figure
Picture or true picture.
It should be noted that other of each functional unit involved by a kind of pattern recognition device provided by the embodiments of the present application
Corresponding description, can be with reference to the corresponding description in Fig. 1 and Fig. 2, and details are not described herein.
Based on above-mentioned method as depicted in figs. 1 and 2, correspondingly, the embodiment of the present application also provides a kind of storage medium,
On be stored with computer program, which realizes above-mentioned image-recognizing method as depicted in figs. 1 and 2 when being executed by processor.
Based on this understanding, the technical solution of the application can be embodied in the form of software products, which produces
Product can store in a non-volatile memory medium (can be CD-ROM, USB flash disk, mobile hard disk etc.), including some instructions
With so that computer equipment (can be personal computer, server or the network equipment an etc.) execution the application is each
Method described in implement scene.
It is above-mentioned in order to realize based on above-mentioned method as shown in Figure 1 and Figure 2 and virtual bench embodiment shown in Fig. 3
Purpose, the embodiment of the present application also provides a kind of computer equipments, are specifically as follows personal computer, server, the network equipment
Deng the entity device includes storage medium and processor;Storage medium, for storing computer program;Processor, for executing
Computer program is to realize above-mentioned image-recognizing method as depicted in figs. 1 and 2.
Optionally, which can also include user interface, network interface, camera, radio frequency (Radio
Frequency, RF) circuit, sensor, voicefrequency circuit, WI-FI module etc..User interface may include display screen
(Display), input unit such as keyboard (Keyboard) etc., optional user interface can also connect including USB interface, card reader
Mouthful etc..Network interface optionally may include standard wireline interface and wireless interface (such as blue tooth interface, WI-FI interface).
It will be understood by those skilled in the art that a kind of computer equipment structure provided in this embodiment is not constituted to the reality
The restriction of body equipment may include more or fewer components, perhaps combine certain components or different component layouts.
It can also include operating system, network communication module in storage medium.Operating system is that management computer equipment is hard
The program of part and software resource supports the operation of message handling program and other softwares and/or program.Network communication module is used
Communication between each component in realization storage medium inside, and communicated between other hardware and softwares in the entity device.
Through the above description of the embodiments, those skilled in the art can be understood that the application can borrow
It helps software that the mode of necessary general hardware platform is added to realize, hardware realization can also be passed through.Pass through the skill of application the application
Art scheme, with existing based on active forensic technologies, compared with the technical solution of passive blind forensic technologies identification image true-false, this implementation
Example can lead to too small amount of forgery image and generate a large amount of forgery image;And it is fought using depth convolution and generates network model
In final differentiation network model recognition target image, can effectively ensure that and final differentiate network model identification image true-false
The robustness of accuracy and final differentiation network model.
It will be appreciated by those skilled in the art that the accompanying drawings are only schematic diagrams of a preferred implementation scenario, module in attached drawing or
Process is not necessarily implemented necessary to the application.It will be appreciated by those skilled in the art that the mould in device in implement scene
Block can according to implement scene describe be distributed in the device of implement scene, can also carry out corresponding change be located at be different from
In one or more devices of this implement scene.The module of above-mentioned implement scene can be merged into a module, can also be into one
Step splits into multiple submodule.
Above-mentioned the application serial number is for illustration only, does not represent the superiority and inferiority of implement scene.Disclosed above is only the application
Several specific implementation scenes, still, the application is not limited to this, and the changes that any person skilled in the art can think of is all
The protection scope of the application should be fallen into.
Claims (10)
1. a kind of image-recognizing method characterized by comprising
It is fought using depth convolution and generates trained generation network model in network model, generated according to altered data and forge figure
Picture;
Depth convolution is fought using the image discriminating sample set being made of forgery image generated and preset true picture
It generates trained differentiation network model in network model to be trained, obtains final differentiation network model;
Target image is identified using final differentiation network model, determines that the target image is to forge image or true
Real image.
2. the method according to claim 1, wherein described generated in network model using the confrontation of depth convolution is instructed
The generation network model perfected generates before forging image according to altered data, and the method is specific further include:
Differentiate that sample set fights depth convolution using first be made of noise variance and true picture to generate in network model
Initial differentiation network model be trained, obtain the first differentiation network model;
The first differentiation network model is instructed using the second differentiation sample set by noise variance and forgery image construction
Practice, obtains trained differentiation network model.
3. the method according to claim 1, wherein described generated in network model using the confrontation of depth convolution is instructed
The generation network model perfected generates before forging image according to altered data, and the method is specific further include:
Differentiate that sample set fights depth convolution using first be made of noise variance and true picture to generate in network model
Initial differentiation network model be trained, obtain the first differentiation network model;
The first differentiation network model is instructed using the second differentiation sample set by noise variance and forgery image construction
Practice, obtains the second differentiation network model;
Differentiate that sample set instructs the second differentiation network model using the third being made of noise variance and true picture
Practice, obtains third and differentiate network model;
Network model, which is instructed, to be differentiated to the third using by noise variance and the 4th differentiation sample set of forgery image construction
Practice, obtains trained differentiation network model.
4. the method according to claim 1, wherein described generated in network model using the confrontation of depth convolution is instructed
The generation network model perfected generates before forging image according to altered data, and the method is specific further include:
Depth convolution is fought using the first generation sample set being made of noise variance and generates being initially generated in network model
Network model is trained, and obtains trained generation network model.
5. the method according to claim 1, wherein it is described using final differentiation network model to target image
It is identified, determines that the target image is before forging image or true picture, specifically further include:
Target signature in the images to be recognized got is identified and intercepted, the target for corresponding to the target signature is obtained
Image.
6. the method according to claim 1, wherein if the altered data is duplication paste type picture number
It generates types of image data according to, fuzzy retouching types of image data or computer and accordingly utilizes final differentiation network
Model identifies target image, determines that the target image is to forge image, then corresponding forgery image type is respectively
It replicates paste type image, fuzzy retouching types of image or computer and generates types of image.
7. -6 any method according to claim 1, which is characterized in that it is described using final differentiation network model, it is right
Target image is identified, is determined that the target image is to forge image or true picture, is specifically included:
Obtain the image further feature of the target image;
The target image is identified according to the deep layer characteristics of image got, determines that the target image is to forge image
Or true picture.
8. a kind of pattern recognition device characterized by comprising
Generation module, for generating trained generation network model in network model using the confrontation of depth convolution, according to distorting
Data, which generate, forges image;
Training module, for utilizing the image discriminating sample set pair being made of forgery image generated and preset true picture
The confrontation of depth convolution generates trained differentiation network model in network model and is trained, and obtains final differentiation network mould
Type;
Identification module determines that the target image is for identifying using final differentiation network model to target image
Forge image or true picture.
9. a kind of storage medium, is stored thereon with computer program, which is characterized in that realization when described program is executed by processor
Image-recognizing method described in any one of claims 1 to 7.
10. a kind of computer equipment, including storage medium, processor and storage can be run on a storage medium and on a processor
Computer program, which is characterized in that the processor is realized described in any one of claims 1 to 7 when executing described program
Image-recognizing method.
Priority Applications (2)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201910559070.3A CN110458185A (en) | 2019-06-26 | 2019-06-26 | Image-recognizing method and device, storage medium, computer equipment |
PCT/CN2019/118187 WO2020258667A1 (en) | 2019-06-26 | 2019-11-13 | Image recognition method and apparatus, and non-volatile readable storage medium and computer device |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201910559070.3A CN110458185A (en) | 2019-06-26 | 2019-06-26 | Image-recognizing method and device, storage medium, computer equipment |
Publications (1)
Publication Number | Publication Date |
---|---|
CN110458185A true CN110458185A (en) | 2019-11-15 |
Family
ID=68481088
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201910559070.3A Pending CN110458185A (en) | 2019-06-26 | 2019-06-26 | Image-recognizing method and device, storage medium, computer equipment |
Country Status (2)
Country | Link |
---|---|
CN (1) | CN110458185A (en) |
WO (1) | WO2020258667A1 (en) |
Cited By (9)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN111581671A (en) * | 2020-05-11 | 2020-08-25 | 笵成科技南京有限公司 | Digital passport protection method combining deep neural network and block chain |
CN112001785A (en) * | 2020-07-21 | 2020-11-27 | 小花网络科技(深圳)有限公司 | Network credit fraud identification method and system based on image identification |
CN112116592A (en) * | 2020-11-19 | 2020-12-22 | 北京瑞莱智慧科技有限公司 | Image detection method, training method, device and medium of image detection model |
CN112132133A (en) * | 2020-06-16 | 2020-12-25 | 杭州中科睿鉴科技有限公司 | Identification image data enhancement method and authenticity intelligent identification method |
CN112766189A (en) * | 2021-01-25 | 2021-05-07 | 北京有竹居网络技术有限公司 | Depth forgery detection method, device, storage medium, and electronic apparatus |
CN112801281A (en) * | 2021-03-22 | 2021-05-14 | 东南大学 | Countermeasure generation network construction method based on quantization generation model and neural network |
CN113542221A (en) * | 2021-06-15 | 2021-10-22 | 四川英得赛克科技有限公司 | Method and system for judging tampering of sensor data of intelligent substation, electronic equipment and storage medium |
WO2022073414A1 (en) * | 2020-10-09 | 2022-04-14 | 腾讯科技(深圳)有限公司 | Image recognition method and apparatus, computing device and computer-readable storage medium |
WO2022151655A1 (en) * | 2021-01-18 | 2022-07-21 | 深圳市商汤科技有限公司 | Data set generation method and apparatus, forgery detection method and apparatus, device, medium and program |
Families Citing this family (9)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN112686331B (en) * | 2021-01-11 | 2022-09-09 | 中国科学技术大学 | Forged image recognition model training method and forged image recognition method |
CN113052203B (en) * | 2021-02-09 | 2022-01-18 | 哈尔滨工业大学(深圳)(哈尔滨工业大学深圳科技创新研究院) | Anomaly detection method and device for multiple types of data |
CN113379600A (en) * | 2021-05-26 | 2021-09-10 | 北京邮电大学 | Short video super-resolution conversion method, device and medium based on deep learning |
CN113822160B (en) * | 2021-08-20 | 2023-09-19 | 西安交通大学 | Evaluation method, system and equipment of depth counterfeiting detection model |
CN114298137A (en) * | 2021-11-12 | 2022-04-08 | 广州辰创科技发展有限公司 | Tiny target detection system based on countermeasure generation network |
CN114841236B (en) * | 2022-03-28 | 2024-04-23 | 中国科学院宁波材料技术与工程研究所 | Method for identifying flexible pressure sensing array map based on deep learning |
CN115270614B (en) * | 2022-07-18 | 2024-05-28 | 郑州轻工业大学 | Visual generation method for digital twin bodies of multiple physical fields of muddy water circulating system |
CN115308799B (en) * | 2022-09-05 | 2023-07-25 | 中国地质科学院地质力学研究所 | Method and system for identifying free gas structure of seismic imaging |
CN117593311B (en) * | 2024-01-19 | 2024-06-21 | 浙江大学 | Depth synthetic image detection enhancement method and device based on countermeasure generation network |
Citations (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN107016406A (en) * | 2017-02-24 | 2017-08-04 | 中国科学院合肥物质科学研究院 | The pest and disease damage image generating method of network is resisted based on production |
CN108921220A (en) * | 2018-06-29 | 2018-11-30 | 国信优易数据有限公司 | Image restoration model training method, device and image recovery method and device |
US20190087648A1 (en) * | 2017-09-21 | 2019-03-21 | Baidu Online Network Technology (Beijing) Co., Ltd | Method and apparatus for facial recognition |
CN109543740A (en) * | 2018-11-14 | 2019-03-29 | 哈尔滨工程大学 | A kind of object detection method based on generation confrontation network |
Family Cites Families (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US10127428B2 (en) * | 2014-08-26 | 2018-11-13 | Gingy Technology Inc. | Fingerprint identification method and fingerprint identification device |
CN108197700A (en) * | 2018-01-12 | 2018-06-22 | 广州视声智能科技有限公司 | A kind of production confrontation network modeling method and device |
CN109033940B (en) * | 2018-06-04 | 2019-07-05 | 上海依图网络科技有限公司 | A kind of image-recognizing method, calculates equipment and storage medium at device |
CN109784384B (en) * | 2018-12-28 | 2023-04-07 | 佛山科学技术学院 | Method and device for automatically distinguishing authenticity of trademark |
-
2019
- 2019-06-26 CN CN201910559070.3A patent/CN110458185A/en active Pending
- 2019-11-13 WO PCT/CN2019/118187 patent/WO2020258667A1/en active Application Filing
Patent Citations (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN107016406A (en) * | 2017-02-24 | 2017-08-04 | 中国科学院合肥物质科学研究院 | The pest and disease damage image generating method of network is resisted based on production |
US20190087648A1 (en) * | 2017-09-21 | 2019-03-21 | Baidu Online Network Technology (Beijing) Co., Ltd | Method and apparatus for facial recognition |
CN108921220A (en) * | 2018-06-29 | 2018-11-30 | 国信优易数据有限公司 | Image restoration model training method, device and image recovery method and device |
CN109543740A (en) * | 2018-11-14 | 2019-03-29 | 哈尔滨工程大学 | A kind of object detection method based on generation confrontation network |
Non-Patent Citations (1)
Title |
---|
杨露菁等: "《智能图像处理及应用》", 31 March 2019, 中国铁道出版社, pages: 96 - 97 * |
Cited By (13)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN111581671A (en) * | 2020-05-11 | 2020-08-25 | 笵成科技南京有限公司 | Digital passport protection method combining deep neural network and block chain |
CN112132133A (en) * | 2020-06-16 | 2020-12-25 | 杭州中科睿鉴科技有限公司 | Identification image data enhancement method and authenticity intelligent identification method |
CN112132133B (en) * | 2020-06-16 | 2023-11-17 | 中国科学院计算技术研究所数字经济产业研究院 | Identification image data enhancement method and true-false intelligent identification method |
CN112001785A (en) * | 2020-07-21 | 2020-11-27 | 小花网络科技(深圳)有限公司 | Network credit fraud identification method and system based on image identification |
WO2022073414A1 (en) * | 2020-10-09 | 2022-04-14 | 腾讯科技(深圳)有限公司 | Image recognition method and apparatus, computing device and computer-readable storage medium |
CN112116592A (en) * | 2020-11-19 | 2020-12-22 | 北京瑞莱智慧科技有限公司 | Image detection method, training method, device and medium of image detection model |
CN112116592B (en) * | 2020-11-19 | 2021-04-02 | 北京瑞莱智慧科技有限公司 | Image detection method, training method, device and medium of image detection model |
WO2022151655A1 (en) * | 2021-01-18 | 2022-07-21 | 深圳市商汤科技有限公司 | Data set generation method and apparatus, forgery detection method and apparatus, device, medium and program |
CN112766189B (en) * | 2021-01-25 | 2023-08-08 | 北京有竹居网络技术有限公司 | Deep forgery detection method and device, storage medium and electronic equipment |
CN112766189A (en) * | 2021-01-25 | 2021-05-07 | 北京有竹居网络技术有限公司 | Depth forgery detection method, device, storage medium, and electronic apparatus |
CN112801281A (en) * | 2021-03-22 | 2021-05-14 | 东南大学 | Countermeasure generation network construction method based on quantization generation model and neural network |
CN113542221A (en) * | 2021-06-15 | 2021-10-22 | 四川英得赛克科技有限公司 | Method and system for judging tampering of sensor data of intelligent substation, electronic equipment and storage medium |
CN113542221B (en) * | 2021-06-15 | 2023-11-03 | 四川英得赛克科技有限公司 | Method and system for judging falsification of sensor data of intelligent substation, electronic equipment and storage medium |
Also Published As
Publication number | Publication date |
---|---|
WO2020258667A1 (en) | 2020-12-30 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN110458185A (en) | Image-recognizing method and device, storage medium, computer equipment | |
CN109657554B (en) | Image identification method and device based on micro expression and related equipment | |
CN105138993B (en) | Establish the method and device of human face recognition model | |
CN109492627B (en) | Scene text erasing method based on depth model of full convolution network | |
CN109583449A (en) | Character identifying method and Related product | |
CN109840467A (en) | A kind of in-vivo detection method and system | |
CN106599872A (en) | Method and equipment for verifying living face images | |
CN109361934A (en) | Image processing method, device, equipment and storage medium | |
CN111860055B (en) | Face silence living body detection method, device, readable storage medium and equipment | |
CN110245621A (en) | Face identification device and image processing method, Feature Selection Model, storage medium | |
Kim et al. | Classifying genuine face images from disguised face images | |
Pishori et al. | Detecting deepfake videos: An analysis of three techniques | |
Arora et al. | A review of techniques to detect the GAN-generated fake images | |
WO2022156214A1 (en) | Liveness detection method and apparatus | |
CN113240050B (en) | Metal printing molten pool detection method with adjustable feature fusion weight | |
CN107369086A (en) | A kind of identity card stamp system and method | |
CN111091055A (en) | Face shape recognition method, device, equipment and computer readable storage medium | |
Hebda et al. | A compact deep convolutional neural network architecture for video based age and gender estimation | |
Grinchuk et al. | 3d mask presentation attack detection via high resolution face parts | |
CN115731620A (en) | Method for detecting counter attack and method for training counter attack detection model | |
CN113327212B (en) | Face driving method, face driving model training device, electronic equipment and storage medium | |
CN109815796B (en) | Method and device for testing influence factors of face recognition passing rate | |
Nachiappan et al. | Artificial and Deceitful Faces Detection Using Machine Learning;[Detección De Rostros Artificiales Y Engañosos Mediante Aprendizaje Automático] | |
Dong | 3D face recognition neural network for digital human resource management | |
Zhou et al. | Learning parallel and hierarchical mechanisms for edge detection |
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 |