CN110222774A - Illegal image discrimination method, device, content safety firewall and storage medium - Google Patents
Illegal image discrimination method, device, content safety firewall and storage medium Download PDFInfo
- Publication number
- CN110222774A CN110222774A CN201910498093.8A CN201910498093A CN110222774A CN 110222774 A CN110222774 A CN 110222774A CN 201910498093 A CN201910498093 A CN 201910498093A CN 110222774 A CN110222774 A CN 110222774A
- Authority
- CN
- China
- Prior art keywords
- image
- illegal
- sample
- model
- confrontation network
- 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
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
-
- 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)
- Data Mining & Analysis (AREA)
- Evolutionary Computation (AREA)
- Life Sciences & Earth Sciences (AREA)
- Artificial Intelligence (AREA)
- General Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- General Health & Medical Sciences (AREA)
- Software Systems (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- Biophysics (AREA)
- Biomedical Technology (AREA)
- Mathematical Physics (AREA)
- Computational Linguistics (AREA)
- Health & Medical Sciences (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Bioinformatics & Computational Biology (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Evolutionary Biology (AREA)
- Image Analysis (AREA)
Abstract
The embodiment of the present invention proposes a kind of illegal image discrimination method, device, content safety firewall and computer readable storage medium.Wherein illegal image discrimination method includes: to carry out image procossing to illegal image sample, is generated corresponding to resisting sample；Using illegal image sample and corresponding to resisting sample, training condition generates confrontation network；Confrontation network is generated using trained condition, by the first image restoring at the second image, the first image is image to be identified after image procossing, and the second image is obtained by the first image restoring without the image by image procossing；Second image is passed into illegal image and identifies service, so that illegal image identifies whether service the second image of identification is illegal image.The embodiment of the present invention only need to generate it is a small amount of to resisting sample can training condition generate confrontation network, resist the black box attack for identifying service for illegal image, and need not the original illegal image of re -training identify service model.
Description
Technical field
The present invention relates to information technology field more particularly to a kind of illegal image discrimination method, device, content safety fire prevention
Wall and computer readable storage medium.
Background technique
A large amount of Internet application allows user to upload the image informations such as head portrait, photo at present.Content of the country to image
With stringent regulation, forbid uploading, share illegal image, such as forbids uploading, shares yellow image, i.e., obscene pornographic figure
Picture.AI (Artificial Intelligence, artificial intelligence) company, cloud service manufacturer and the service security manufacturer of mainstream
API (Application Programming Interface, application programming interface) service is both provided, for detecting
The image of upload whether illegal image.For example, the illegal image for detecting the whether obscene pornographic image of the image uploaded identifies
The yellow service of service referred to as mirror.Usual illegal image identifies service and is all based on deep learning model.Part black industry uses
To resisting sample (Adversarial examples) technology, pass through FGSM (Fast Gradient Sign Method, quick ladder
Spend symbol) algorithm, C/W (Carlini&Wagner) method, JSMA (Jacobian Saliency Map Approach, it is refined can
Ratio characteristic mapping method) etc. confrontation sample algorithm, be superimposed certain disturbance on illegal image, generate to resisting sample, then around
It has crossed illegal image and has identified service.
It has been observed that whitepack can be carried out when black industry gets illegal image and identifies the deep learning model of service
Attack, generates to resisting sample, identifies service around illegal image.Illegal image, which is not got, in black industry identifies service
When deep learning model, usually carried out by modes such as noise, affine variation, filtered blurry, brightness change and monochromatizations black
Box attack attempts the detection for identifying service around illegal image, threatens to content safety.In response to this, at present
Illegal image identifies service and needs to pass through noise, affine variation, filtered blurry, brightness for existing yellow image sample
The modes such as variation and monochromatization are handled, and are generated largely to resisting sample, and then the original illegal image of re -training reflects again
It does not service.It is all comparatively laborious to generate the process for largely identifying to resisting sample and re -training illegal image and servicing, leads to figure
As the efficiency identified is lower.
Summary of the invention
The embodiment of the present invention provides a kind of illegal image discrimination method, device, content safety firewall and computer-readable
Storage medium, to solve one or more technical problems in the prior art.
In a first aspect, the embodiment of the invention provides a kind of illegal image discrimination methods, comprising:
Image procossing is carried out to illegal image sample, is generated corresponding to resisting sample；
Using the illegal image sample and described corresponding to resisting sample, training condition generates confrontation network；
Confrontation network is generated using the trained condition, by the first image restoring at the second image, first figure
It seem image to be identified after described image is handled, second image is not having of being restored by the first image
The image handled by described image；
Second image is passed into illegal image and identifies service, so that the illegal image identifies described in service identification
Whether the second image is illegal image.
In one embodiment, image procossing is carried out to illegal image sample, generated corresponding to resisting sample, comprising:
By at least one of noise, affine transformation, filtered blurry, brightness change and monochromatization mode to illegal figure
Decent progress image procossing.
In one embodiment, image procossing is carried out to illegal image sample, generated corresponding to resisting sample, comprising:
Image procossing is carried out to illegal image sample using open source computer vision library and/or image processing software.
In one embodiment, it includes generating model and discrimination model that the condition, which generates confrontation network,；Using described
Illegal image sample and described corresponding to resisting sample, training condition generates confrontation network, comprising:
The generation model is inputted to resisting sample and random noise by described, original image is gone back by generation model generation；
By the illegal image sample, described to resisting sample and the original image of going back is input to the discrimination model, by institute
It states and goes back the authenticity probability that original image is the illegal image sample described in discrimination model generation；
According to the authenticity probability, pass through the Game Learning of the generation model and the discrimination model training item
Part generates confrontation network.
Second aspect, the embodiment of the invention provides a kind of illegal image identification devices, comprising:
Generation unit is used for: being carried out image procossing to illegal image sample, is generated corresponding to resisting sample；
Training unit is used for: using the illegal image sample and described corresponding to resisting sample, training condition is generated pair
Anti- network；
Reduction unit is used for: confrontation network is generated using the trained condition, by the first image restoring at the second figure
Picture, the first image are images to be identified after described image is handled, and second image is by the first image
Restore the obtained image without passing through described image processing；
Transfer unit is used for: second image being passed to illegal image and identifies service, so as to illegal image mirror
It Fu Wu not identify whether second image is illegal image.
In one embodiment, the generation unit is used for:
By at least one of noise, affine transformation, filtered blurry, brightness change and monochromatization mode to illegal figure
Decent progress image procossing.
In one embodiment, the generation unit is used for:
Image procossing is carried out to illegal image sample using open source computer vision library and/or image processing software.
In one embodiment, it includes generating model and discrimination model that the condition, which generates confrontation network,；The training
Unit is used for:
The generation model is inputted to resisting sample and random noise by described, original image is gone back by generation model generation；
By the illegal image sample, described to resisting sample and the original image of going back is input to the discrimination model, by institute
It states and goes back the authenticity probability that original image is the illegal image sample described in discrimination model generation；
According to the authenticity probability, pass through the Game Learning of the generation model and the discrimination model training item
Part generates confrontation network.
The third aspect, the embodiment of the invention provides a kind of illegal image identification device, the function of described device can lead to
Hardware realization is crossed, corresponding software realization can also be executed by hardware.The hardware or software include it is one or more with it is upper
State the corresponding module of function.
It include processor and memory in the structure of described device in a possible design, the memory is used for
Storage supports described device to execute the program of above-mentioned illegal image discrimination method, the processor is configured to described for executing
The program stored in memory.Described device can also include communication interface, be used for and other equipment or communication.
Fourth aspect, the embodiment of the invention provides a kind of content safety firewalls characterized by comprising
One or more processors；
Storage device, for storing one or more programs；
The content safety firewall is used for: being restored image to be identified using above-mentioned illegal image identification device
Processing.
5th aspect, the embodiment of the invention provides a kind of computer readable storage mediums, for storing illegal image mirror
Computer software instructions used in other device comprising for executing program involved in above-mentioned illegal image discrimination method.
Above-mentioned technical proposal has the following advantages that or the utility model has the advantages that only needing to generate a small amount of can train resisting sample
Condition generates confrontation network, resists the black box attack for identifying service for illegal image, and this method does not need re -training original
Some illegal images identify service model.
Above-mentioned general introduction is merely to illustrate that the purpose of book, it is not intended to be limited in any way.Except foregoing description
Schematical aspect, except embodiment and feature, by reference to attached drawing and the following detailed description, the present invention is further
Aspect, embodiment and feature, which will be, to be readily apparent that.
Detailed description of the invention
In the accompanying drawings, unless specified otherwise herein, otherwise indicate the same or similar through the identical appended drawing reference of multiple attached drawings
Component or element.What these attached drawings were not necessarily to scale.It should be understood that these attached drawings depict only according to the present invention
Disclosed some embodiments, and should not serve to limit the scope of the present invention.
Fig. 1 shows the flow chart of illegal image discrimination method according to an embodiment of the present invention.
Fig. 2 shows the processes that the training condition of illegal image discrimination method according to an embodiment of the present invention generates confrontation network
Figure.
The condition that Fig. 3 shows illegal image discrimination method according to an embodiment of the present invention generates confrontation network diagram.
Fig. 4 shows the flow chart of illegal image discrimination method according to an embodiment of the present invention.
Fig. 5 shows the flow chart of illegal image discrimination method according to an embodiment of the present invention.
Fig. 6 shows the structural block diagram of illegal image identification device according to an embodiment of the present invention.
Fig. 7 shows the structural block diagram of illegal image identification device according to an embodiment of the present invention.
Specific embodiment
Hereinafter, certain exemplary embodiments are simply just described.As one skilled in the art will recognize that
Like that, without departing from the spirit or scope of the present invention, described embodiment can be modified by various different modes.
Therefore, attached drawing and description are considered essentially illustrative rather than restrictive.
Fig. 1 shows the flow chart of illegal image discrimination method according to an embodiment of the present invention.As shown in Figure 1, the illegal figure
As discrimination method includes:
Step S110 carries out image procossing to illegal image sample, generates corresponding to resisting sample；
Step S120, using the illegal image sample and described corresponding to resisting sample, training condition generates confrontation net
Network；
Step S130 generates confrontation network using the trained condition, by the first image restoring at the second image, institute
Stating the first image is image to be identified after described image is handled, and second image is restored by the first image
The image without passing through described image processing arrived；
Second image is passed to illegal image and identifies service by step S140, is taken so that the illegal image identifies
Business identifies whether second image is illegal image.
Usual illegal image identifies service and is all based on deep learning model.Illegal image is not got in black industry
When identifying the deep learning model of service, usually generated by way of carrying out image procossing to the image of upload to resisting sample,
Black box attack is carried out, the detection for identifying service around illegal image is attempted.Wherein, to resisting sample be one by small sample perturbations or
Adjustment can allow the input sample of deep learning algorithm output error result.For generating the side to the image procossing of resisting sample
Formula may include noise, affine variation, filtered blurry, brightness change and monochromatization etc..
The illegal image discrimination method of the embodiment of the present invention generates a small amount of confrontation sample using existing illegal image sample
This, generates confrontation network (CGAN, Conditional Generative using a small amount of confrontation sample training condition
Adversarial Nets, also referred to as pix2pix).Then use condition generates confrontation network, passes through figure what user uploaded
As the image of processing, it is reduced into normal image.Wherein, the image that user uploads may be the confrontation sample by image procossing
This.The processing that confrontation network is generated by condition makes to be eliminated disturbance increased in resisting sample, to make to lose resisting sample
Effect.Then the illegal image for passing to rear end again identifies service and is analyzed, so that illegal image identification service can be more quasi-
Really identify image.
Specifically, in step s 110, existing illegal image sample is obtained, can be bypassed by image procossing generation non-
Method image authentication service to resisting sample.That is, for each illegal image sample, can generate one it is right therewith
Answer to resisting sample.In the step s 120, the illegal image sample that generates in step S110 and described corresponding right is used
Resisting sample, training condition generate confrontation network so that the trained condition generate confrontation network can will be to resisting sample
It is reduced into illegal image sample.In step s 130, after user uploads image, confrontation net is generated using the trained condition
The image that network uploads user is handled.If the image that user uploads is treated to resisting sample, condition is generated
Fighting network can be by the image restoring at the image for not passing through image procossing.Wherein, the image that user uploads is known as the first figure
Picture.It is known as the second image without the image by described image processing by what the first image restored.In step S140
In, the second image of reduction is passed into illegal image identification service and is identified.Since the second image has eliminated first
Increased disturbance in image, so that illegal image, which identifies service, can more accurately identify image.
In one embodiment, image procossing is carried out to illegal image sample, generated corresponding to resisting sample, comprising:
By at least one of noise, affine transformation, filtered blurry, brightness change and monochromatization mode to illegal figure
Decent progress image procossing.
Wherein, affine transformation is also known as affine maps, refers in geometry, and a vector space carries out once linear transformation simultaneously
A translation is connected, another vector space is transformed to.The treatment process of filtered blurry mainly carries out at denoising image
Reason.At least one of smothing filtering, gaussian filtering, median filtering mode can be used to be filtered Fuzzy processing.
In one embodiment, image procossing is carried out to illegal image sample, generated corresponding to resisting sample, comprising:
Image procossing is carried out to illegal image sample using open source computer vision library and/or image processing software.
In one example, OpenCV (Open source Computer Vision Library, open source meter can be used
Calculation machine vision library) image procossing is carried out to illegal image sample.OpenCV is one for image procossing, analysis, machine vision
The open source function library of aspect.It realizes many general-purpose algorithms in terms of image procossing and computer vision.That is, it is
API (the Application Programming Interface, using journey of a set of open source code about computer vision
Sequence programming interface) function library.Common image procossing mode can include: removal or reduce noise, unitary of illumination, brightness are returned
The operations such as one changes, is blurred, sharpening, expanding, burn into is opened and closed.And for these operations, OpenCV each provides corresponding API
Function.
In another example, PS (Photoshop) can be used to carry out image procossing to illegal image sample.
Photoshop is the image processing software of a profession.It mainly handles the digital picture constituted with pixel.It uses
Photoshop it is numerous compile and drawing tool, image processing work can be effectively performed.
Fig. 2 shows the processes that the training condition of illegal image discrimination method according to an embodiment of the present invention generates confrontation network
Figure.As shown in Fig. 2, in one embodiment, it includes generating model and discrimination model that the condition, which generates confrontation network,；Fig. 1
In step S120, using the illegal image sample and described corresponding to resisting sample, training condition generates confrontation network, tool
Body can include:
Step S210 inputs the generation model to resisting sample and random noise for described, is generated by the generation model
Also original image；
Step S220, by the illegal image sample, described to resisting sample and the original image of going back is input to the differentiation
Model goes back the authenticity probability that original image is the illegal image sample as described in discrimination model generation；
Step S230 passes through the Game Learning for generating model and the discrimination model according to the authenticity probability
The training condition generates confrontation network.
Generating confrontation network (GAN, Generative Adversarial Networks) is a kind of deep learning model.
The frame of this model includes: to generate model (Generative Model, abbreviation G) and discrimination model (Discriminative
Modell, abbreviation D).Fairly good output is generated by the mutual Game Learning of two models.By taking image processing process as an example,
The effect for generating model is to generate image.That is, a width will be exported by generating model after inputting a random noise z
Image G (z) automatically generate, false.Discrimination model is for judging, it receives the image for generating model output as defeated
Enter, then judges the true and false of diagram picture.For example, 1 is exported if the image is really, the output if the image is false
0。
In traditional generation confrontation network, a random noise is inputted, a width random image will be exported.But user
The output image of demand is corresponding with input picture, related.Such as the sketch of one cat of input, the same shape of desired output
The true picture of the cat of state.It is wherein a kind of embodiment of user's control to the requirement of form.
Condition generates confrontation network and has done a small change to traditional generation confrontation network, generates in confrontation network in condition
Input is the image for needing to convert.Wherein, the image for needing to convert generates the condition of confrontation network as condition.Using condition
Confrontation network is generated, input picture can be established and exports the corresponding relationship of image.Condition generates confrontation network and refers to generating
Condition is added in confrontation network, the effect of condition is prefect into confrontation network.After generating model generation output image, it will generate
Input picture of the input picture and output image of model together as discrimination model.Condition generate confrontation network need use at
Pair data be trained.In embodiments of the present invention, using the illegal image sample and described corresponding to resisting sample, instruction
The condition of white silk generates confrontation network.
Many problems in image procossing be all by one input image be changed into a corresponding output image, such as
Grayscale image, gradient map, conversion between cromogram etc..Each usual problem all uses specific algorithm.The sheet of these methods
Matter is all the mapping from pixel to pixel in fact.Condition generate confrontation network can be used to solve the problems, such as it is this kind of, for completing into
Pair image conversion.For example, the carriageway image under normal weather is converted into the carriageway image under the conditions of sleet.
The condition that Fig. 3 shows illegal image discrimination method according to an embodiment of the present invention generates confrontation network diagram.Such as
Shown in Fig. 3, image x generates the condition of confrontation network as condition, needs to be input in generation model G and discrimination model D.It generates
The input of model G is { x, z }.Wherein, x is the image for needing to convert；Z is random noise.The output for generating model G is to generate
Image G (x, z).Discrimination model D then needs to tell { x, G (x, z) } and { x, y }, that is, it is raw for telling the image of input
The image or true picture generated at model G.Wherein, y indicates true picture.
Fig. 4 shows the flow chart of illegal image discrimination method according to an embodiment of the present invention.As shown in figure 4, first to original
Beginning image (illegal image sample) carries out image procossing, and treatment process includes superimposed noise, the affine variation of superposition, superposition filtering mould
Gelatinization variation, superposition brightness change, superposition tone color variation.It is generated after above-mentioned processing corresponding to resisting sample, sample will be fought
Originally it is input to condition and generates confrontation network.It includes generating model G and discrimination model D that condition, which generates confrontation network,.Condition generation pair
Anti- network will be converted into resisting sample to go back original image.Using the illegal image sample and described corresponding to resisting sample, pass through
Loss function and optimizer training condition generate confrontation network.
Fig. 5 shows the flow chart of illegal image discrimination method according to an embodiment of the present invention.As shown in figure 5, can be in content
The illegal image discrimination method of the embodiment of the present invention is realized in security firewall.By taking the yellow service of mirror as an example, which is located at mirror
Before Huang service, when user uploads head portrait or picture, head portrait or picture are handled and then are transmitted by content safety firewall
To reflecting, yellow service is identified.One illustrative image authentication process is as follows:
First step: using the sample of existing yellow picture, become by noise, affine variation, filtered blurry, brightness
Change and the modes such as monochromatization are handled, generates new sample.
Confrontation network C GAN is generated to condition using new sample to be trained.Relative to general deep learning model,
Generating confrontation network can use the training process of a small amount of sample completion network and over-fitting does not occur.After training
CGAN can be by processed yellow maps of modes such as noise, affine variation, filtered blurry, brightness change and monochromatizations
Piece is reduced into normal yellow picture.
Second step: when user uploads head portrait or picture, head portrait or picture need to handle by content safety firewall.
The core function of content safety firewall be using CGAN, will by noise, affine variation, filtered blurry, brightness change and
The processed yellow picture of the modes such as monochromatization, is reduced into normal yellow picture.
If be passed to content safety firewall in this step is normal picture, any processing not will do it.
Third step: it is identified using the yellow service of original mirror.
Above-mentioned technical proposal has the following advantages that or the utility model has the advantages that only needing to generate a small amount of can train resisting sample
Condition generates confrontation network, resists the black box attack for identifying service for illegal image, and this method does not need re -training original
Some illegal images identify service model.
Fig. 6 shows the structural block diagram of illegal image identification device according to an embodiment of the present invention.As shown in fig. 6, of the invention
The illegal image identification device of embodiment includes:
Generation unit 100, is used for: carrying out image procossing to illegal image sample, generates corresponding to resisting sample；
Training unit 200, is used for: using the illegal image sample and described corresponding to resisting sample, training condition is raw
At confrontation network；
Reduction unit 300, is used for: confrontation network is generated using the trained condition, by the first image restoring at the
Two images, the first image are images to be identified after described image is handled, and second image is by described first
The image without passing through described image processing that image restoring obtains；
Transfer unit 400, is used for: second image being passed to illegal image and identifies service, so as to the illegal figure
Identify whether second image is illegal image as identifying service.
In one embodiment, the generation unit 100 is used for:
By at least one of noise, affine transformation, filtered blurry, brightness change and monochromatization mode to illegal figure
Decent progress image procossing.
In one embodiment, the generation unit 100 is used for:
Image procossing is carried out to illegal image sample using open source computer vision library and/or image processing software.
In one embodiment, it includes generating model and discrimination model that the condition, which generates confrontation network,；The training
Unit 200 is used for:
The generation model is inputted to resisting sample and random noise by described, original image is gone back by generation model generation；
By the illegal image sample, described to resisting sample and the original image of going back is input to the discrimination model, by institute
It states and goes back the authenticity probability that original image is the illegal image sample described in discrimination model generation；
According to the authenticity probability, pass through the Game Learning of the generation model and the discrimination model training item
Part generates confrontation network.
The function of each unit in illegal image identification device of the embodiment of the present invention may refer to the correspondence in the above method
Description, details are not described herein.
Fig. 7 shows the structural block diagram of illegal image identification device according to an embodiment of the present invention.As shown in fig. 7, the device
Include: memory 910 and processor 920, the computer program that can be run on processor 920 is stored in memory 910.Institute
State the illegal image discrimination method realized in above-described embodiment when processor 920 executes the computer program.The memory
910 and processor 920 quantity can for one or more.
The device further include:
Communication interface 930 carries out data interaction for being communicated with external device.
Memory 910 may include high speed RAM memory, it is also possible to further include nonvolatile memory (non-
Volatile memory), a for example, at least magnetic disk storage.
If memory 910, processor 920 and the independent realization of communication interface 930, memory 910,920 and of processor
Communication interface 930 can be connected with each other by bus and complete mutual communication.The bus can be Industry Standard Architecture
Structure (ISA, Industry Standard Architecture) bus, external equipment interconnection (PCI, Peripheral
Component Interconnect) bus or extended industry-standard architecture (EISA, Extended Industry
Standard Architecture) bus etc..The bus can be divided into address bus, data/address bus, control bus etc..For
Convenient for indicating, only indicated with a thick line in Fig. 7, it is not intended that an only bus or a type of bus.
Optionally, in specific implementation, if memory 910, processor 920 and communication interface 930 are integrated in one piece of core
On piece, then memory 910, processor 920 and communication interface 930 can complete mutual communication by internal interface.
The embodiment of the invention provides a kind of content safety firewalls, comprising: one or more processors；Storage device,
For storing one or more programs；The content safety firewall is used for: will be wait reflect using above-mentioned illegal image identification device
Other image carries out reduction treatment.
The embodiment of the invention provides a kind of computer readable storage mediums, are stored with computer program, the program quilt
Processor realizes any method in above-described embodiment when executing.
In the description of this specification, reference term " one embodiment ", " some embodiments ", " example ", " specifically show
The description of example " or " some examples " etc. means specific features, structure, material or spy described in conjunction with this embodiment or example
Point is included at least one embodiment or example of the invention.Moreover, particular features, structures, materials, or characteristics described
It may be combined in any suitable manner in any one or more of the embodiments or examples.In addition, without conflicting with each other, this
The technical staff in field can be by the spy of different embodiments or examples described in this specification and different embodiments or examples
Sign is combined.
In addition, term " first ", " second " are used for descriptive purposes only and cannot be understood as indicating or suggesting relative importance
Or implicitly indicate the quantity of indicated technical characteristic." first " is defined as a result, the feature of " second " can be expressed or hidden
It include at least one this feature containing ground.In the description of the present invention, the meaning of " plurality " is two or more, unless otherwise
Clear specific restriction.
Any process described otherwise above or method description are construed as in flow chart or herein, and expression includes
It is one or more for realizing specific logical function or process the step of executable instruction code module, segment or portion
Point, and the range of the preferred embodiment of the present invention includes other realization, wherein can not press shown or discussed suitable
Sequence, including according to related function by it is basic simultaneously in the way of or in the opposite order, Lai Zhihang function, this should be of the invention
Embodiment person of ordinary skill in the field understood.
Expression or logic and/or step described otherwise above herein in flow charts, for example, being considered use
In the order list for the executable instruction for realizing logic function, may be embodied in any computer-readable medium, for
Instruction execution system, device or equipment (such as computer based system, including the system of processor or other can be held from instruction
The instruction fetch of row system, device or equipment and the system executed instruction) it uses, or combine these instruction execution systems, device or set
It is standby and use.For the purpose of this specification, " computer-readable medium ", which can be, any may include, stores, communicates, propagates or pass
Defeated program is for instruction execution system, device or equipment or the dress used in conjunction with these instruction execution systems, device or equipment
It sets.The more specific example (non-exhaustive list) of computer-readable medium include the following: there is the electricity of one or more wirings
Interconnecting piece (electronic device), portable computer diskette box (magnetic device), random access memory (RAM), read-only memory
(ROM), erasable edit read-only storage (EPROM or flash memory), fiber device and portable read-only memory
(CDROM).In addition, computer-readable medium can even is that the paper that can print described program on it or other suitable Jie
Matter, because can then be edited, be interpreted or when necessary with other for example by carrying out optical scanner to paper or other media
Suitable method is handled electronically to obtain described program, is then stored in computer storage.
It should be appreciated that each section of the invention can be realized with hardware, software, firmware or their combination.Above-mentioned
In embodiment, software that multiple steps or method can be executed in memory and by suitable instruction execution system with storage
Or firmware is realized.It, and in another embodiment, can be under well known in the art for example, if realized with hardware
Any one of column technology or their combination are realized: having a logic gates for realizing logic function to data-signal
Discrete logic, with suitable combinational logic gate circuit specific integrated circuit, programmable gate array (PGA), scene
Programmable gate array (FPGA) etc..
Those skilled in the art are understood that realize all or part of step that above-described embodiment method carries
It suddenly is that relevant hardware can be instructed to complete by program, the program can store in a kind of computer-readable storage medium
In matter, which when being executed, includes the steps that one or a combination set of embodiment of the method.
It, can also be in addition, each functional unit in each embodiment of the present invention can integrate in a processing module
It is that each unit physically exists alone, can also be integrated in two or more units in a module.Above-mentioned integrated mould
Block both can take the form of hardware realization, can also be realized in the form of software function module.The integrated module is such as
Fruit is realized and when sold or used as an independent product in the form of software function module, also can store in a computer
In readable storage medium storing program for executing.The storage medium can be read-only memory, disk or CD etc..
The above description is merely a specific embodiment, but scope of protection of the present invention is not limited thereto, any
Those familiar with the art in the technical scope disclosed by the present invention, can readily occur in its various change or replacement,
These should be covered by the protection scope of the present invention.Therefore, protection scope of the present invention should be with the guarantor of the claim
It protects subject to range.
Claims (11)
1. a kind of illegal image discrimination method characterized by comprising
Image procossing is carried out to illegal image sample, is generated corresponding to resisting sample；
Using the illegal image sample and described corresponding to resisting sample, training condition generates confrontation network；
Confrontation network is generated using the trained condition, by the first image restoring at the second image, the first image is
The image to be identified after described image is handled, second image are restored by the first image without passing through
The image of described image processing；
Second image is passed into illegal image and identifies service, so that the illegal image identifies service identification described second
Whether image is illegal image.
2. generating and corresponding to the method according to claim 1, wherein carrying out image procossing to illegal image sample
To resisting sample, comprising:
By at least one of noise, affine transformation, filtered blurry, brightness change and monochromatization mode to illegal image sample
This progress image procossing.
3. according to the method described in claim 2, generation corresponds to it is characterized in that, carrying out image procossing to illegal image sample
To resisting sample, comprising:
Image procossing is carried out to illegal image sample using open source computer vision library and/or image processing software.
4. according to the method in any one of claims 1 to 3, which is characterized in that the condition generates confrontation network and includes
Generate model and discrimination model；Using the illegal image sample and described corresponding to resisting sample, training condition generates confrontation
Network, comprising:
The generation model is inputted to resisting sample and random noise by described, original image is gone back by generation model generation；
By the illegal image sample, described to resisting sample and the original image of going back is input to the discrimination model, sentenced by described
Other model goes back the authenticity probability that original image is the illegal image sample described in generating；
It is raw by the Game Learning of the generation model and the discrimination model training condition according to the authenticity probability
At confrontation network.
5. a kind of illegal image identification device characterized by comprising
Generation unit is used for: being carried out image procossing to illegal image sample, is generated corresponding to resisting sample；
Training unit is used for: using the illegal image sample and described corresponding to resisting sample, training condition generates confrontation net
Network；
Reduction unit is used for: confrontation network is generated using the trained condition, by the first image restoring at the second image,
The first image is image to be identified after described image is handled, and second image is restored by the first image
The obtained image without passing through described image processing；
Transfer unit is used for: second image being passed to illegal image and identifies service, is taken so that the illegal image identifies
Business identifies whether second image is illegal image.
6. device according to claim 5, which is characterized in that the generation unit is used for:
By at least one of noise, affine transformation, filtered blurry, brightness change and monochromatization mode to illegal image sample
This progress image procossing.
7. device according to claim 6, which is characterized in that the generation unit is used for:
Image procossing is carried out to illegal image sample using open source computer vision library and/or image processing software.
8. device according to any one of claims 5 to 7, which is characterized in that the condition generates confrontation network and includes
Generate model and discrimination model；The training unit is used for:
The generation model is inputted to resisting sample and random noise by described, original image is gone back by generation model generation；
By the illegal image sample, described to resisting sample and the original image of going back is input to the discrimination model, sentenced by described
Other model goes back the authenticity probability that original image is the illegal image sample described in generating；
It is raw by the Game Learning of the generation model and the discrimination model training condition according to the authenticity probability
At confrontation network.
9. a kind of illegal image identification device characterized by comprising
One or more processors；
Storage device, for storing one or more programs；
When one or more of programs are executed by one or more of processors, so that one or more of processors
Realize method according to any one of claims 1 to 4.
10. a kind of content safety firewall characterized by comprising
One or more processors；
Storage device, for storing one or more programs；
The content safety firewall is used for: image to be identified being carried out reduction treatment using device as claimed in claim 9.
11. a kind of computer readable storage medium, is stored with computer program, which is characterized in that the program is held by processor
Method according to any one of claims 1 to 4 is realized when row.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910498093.8A CN110222774A (en) | 2019-06-10 | 2019-06-10 | Illegal image discrimination method, device, content safety firewall and storage medium |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910498093.8A CN110222774A (en) | 2019-06-10 | 2019-06-10 | Illegal image discrimination method, device, content safety firewall and storage medium |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN110222774A true CN110222774A (en) | 2019-09-10 |
Family
ID=67816166
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201910498093.8A Pending CN110222774A (en) | 2019-06-10 | 2019-06-10 | Illegal image discrimination method, device, content safety firewall and storage medium |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN110222774A (en) |
Cited By (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111783890A (en) * | 2020-07-02 | 2020-10-16 | 电子科技大学 | Small pixel countermeasure sample defense method for image recognition process |
| CN111783998A (en) * | 2020-06-30 | 2020-10-16 | 百度在线网络技术（北京）有限公司 | Illegal account recognition model training method and device and electronic equipment |
| CN112712002A (en) * | 2020-12-24 | 2021-04-27 | 深圳力维智联技术有限公司 | CGAN-based environment monitoring method, device, system and storage medium |
| CN112750067A (en) * | 2019-10-29 | 2021-05-04 | 爱思开海力士有限公司 | Image processing system and training method thereof |
| CN112801998A (en) * | 2021-02-05 | 2021-05-14 | 展讯通信（上海）有限公司 | Printed circuit board detection method and device, computer equipment and storage medium |
| CN113378118A (en) * | 2020-03-10 | 2021-09-10 | 百度在线网络技术（北京）有限公司 | Method, apparatus, electronic device, and computer storage medium for processing image data |
Citations (11)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN104484683A (en) * | 2014-12-31 | 2015-04-01 | 小米科技有限责任公司 | Porn picture detection method and device |
| CN106952239A (en) * | 2017-03-28 | 2017-07-14 | 厦门幻世网络科技有限公司 | image generating method and device |
| CN107274358A (en) * | 2017-05-23 | 2017-10-20 | 广东工业大学 | Image Super-resolution recovery technology based on cGAN algorithms |
| CN107730458A (en) * | 2017-09-05 | 2018-02-23 | 北京飞搜科技有限公司 | A kind of fuzzy facial reconstruction method and system based on production confrontation network |
| CN108304858A (en) * | 2017-12-28 | 2018-07-20 | 中国银联股份有限公司 | Fight specimen discerning model generating method, verification method and its system |
| CN108648188A (en) * | 2018-05-15 | 2018-10-12 | 南京邮电大学 | A kind of non-reference picture quality appraisement method based on generation confrontation network |
| CN108765295A (en) * | 2018-06-12 | 2018-11-06 | 腾讯科技（深圳）有限公司 | Image processing method, image processing apparatus and storage medium |
| CN109410184A (en) * | 2018-10-09 | 2019-03-01 | 天格科技（杭州）有限公司 | Live streaming pornographic image detection method based on dense confrontation network semi-supervised learning |
| CN109508643A (en) * | 2018-10-19 | 2019-03-22 | 北京陌上花科技有限公司 | Image processing method and device for porny |
| US20190096056A1 (en) * | 2017-09-25 | 2019-03-28 | United Technologies Corporation | Automated material characterization system including conditional generative adversarial networks |
| US20190188534A1 (en) * | 2017-12-14 | 2019-06-20 | Honda Motor Co., Ltd. | Methods and systems for converting a line drawing to a rendered image |
-
2019
- 2019-06-10 CN CN201910498093.8A patent/CN110222774A/en active Pending
Patent Citations (11)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN104484683A (en) * | 2014-12-31 | 2015-04-01 | 小米科技有限责任公司 | Porn picture detection method and device |
| CN106952239A (en) * | 2017-03-28 | 2017-07-14 | 厦门幻世网络科技有限公司 | image generating method and device |
| CN107274358A (en) * | 2017-05-23 | 2017-10-20 | 广东工业大学 | Image Super-resolution recovery technology based on cGAN algorithms |
| CN107730458A (en) * | 2017-09-05 | 2018-02-23 | 北京飞搜科技有限公司 | A kind of fuzzy facial reconstruction method and system based on production confrontation network |
| US20190096056A1 (en) * | 2017-09-25 | 2019-03-28 | United Technologies Corporation | Automated material characterization system including conditional generative adversarial networks |
| US20190188534A1 (en) * | 2017-12-14 | 2019-06-20 | Honda Motor Co., Ltd. | Methods and systems for converting a line drawing to a rendered image |
| CN108304858A (en) * | 2017-12-28 | 2018-07-20 | 中国银联股份有限公司 | Fight specimen discerning model generating method, verification method and its system |
| CN108648188A (en) * | 2018-05-15 | 2018-10-12 | 南京邮电大学 | A kind of non-reference picture quality appraisement method based on generation confrontation network |
| CN108765295A (en) * | 2018-06-12 | 2018-11-06 | 腾讯科技（深圳）有限公司 | Image processing method, image processing apparatus and storage medium |
| CN109410184A (en) * | 2018-10-09 | 2019-03-01 | 天格科技（杭州）有限公司 | Live streaming pornographic image detection method based on dense confrontation network semi-supervised learning |
| CN109508643A (en) * | 2018-10-19 | 2019-03-22 | 北京陌上花科技有限公司 | Image processing method and device for porny |
Non-Patent Citations (2)
| Title |
|---|
| YUYANG WANG 等: "Super-Resolution of Text Image Based on Conditional Generative Adversarial Network", 《PCM 2018: ADVANCES IN MULTIMEDIA INFORMATION PROCESSING-PCM 2018》 * |
| 脚踏实地飞翔者: "高达99.5%准确率，火眼金睛的"鉴黄系统"背后技术大揭秘", 《HTTP://WWW.360DOC.COM/CONTENT/17/1023/17/21779270_697476535.SHTML》 * |
Cited By (11)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN112750067A (en) * | 2019-10-29 | 2021-05-04 | 爱思开海力士有限公司 | Image processing system and training method thereof |
| CN112750067B (en) * | 2019-10-29 | 2024-05-07 | 爱思开海力士有限公司 | Image processing system and training method thereof |
| CN113378118A (en) * | 2020-03-10 | 2021-09-10 | 百度在线网络技术（北京）有限公司 | Method, apparatus, electronic device, and computer storage medium for processing image data |
| CN113378118B (en) * | 2020-03-10 | 2023-08-22 | 百度在线网络技术（北京）有限公司 | Method, apparatus, electronic device and computer storage medium for processing image data |
| CN111783998A (en) * | 2020-06-30 | 2020-10-16 | 百度在线网络技术（北京）有限公司 | Illegal account recognition model training method and device and electronic equipment |
| CN111783998B (en) * | 2020-06-30 | 2023-08-11 | 百度在线网络技术（北京）有限公司 | Training method and device for illegal account identification model and electronic equipment |
| CN111783890A (en) * | 2020-07-02 | 2020-10-16 | 电子科技大学 | Small pixel countermeasure sample defense method for image recognition process |
| CN111783890B (en) * | 2020-07-02 | 2022-06-03 | 电子科技大学 | Small pixel countermeasure sample defense method for image recognition process |
| CN112712002A (en) * | 2020-12-24 | 2021-04-27 | 深圳力维智联技术有限公司 | CGAN-based environment monitoring method, device, system and storage medium |
| CN112712002B (en) * | 2020-12-24 | 2024-05-14 | 深圳力维智联技术有限公司 | CGAN-based environment monitoring method, CGAN-based environment monitoring device, CGAN-based environment monitoring system and storage medium |
| CN112801998A (en) * | 2021-02-05 | 2021-05-14 | 展讯通信（上海）有限公司 | Printed circuit board detection method and device, computer equipment and storage medium |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN110222774A (en) | Illegal image discrimination method, device, content safety firewall and storage medium | |
| CN111275129B (en) | Image data augmentation policy selection method and system | |
| Barni et al. | CNN detection of GAN-generated face images based on cross-band co-occurrences analysis | |
| CN111414906B (en) | Data synthesis and text recognition method for paper bill pictures | |
| US20200380279A1 (en) | Method and apparatus for liveness detection, electronic device, and storage medium | |
| US20180357800A1 (en) | Multimodal style-transfer network for applying style features from multi-resolution style exemplars to input images | |
| CN109492627B (en) | Scene text erasing method based on depth model of full convolution network | |
| CN109191409B (en) | Image processing method, network training method, device, electronic equipment and storage medium | |
| CN111080628A (en) | Image tampering detection method and device, computer equipment and storage medium | |
| CN110473142B (en) | Single image super-resolution reconstruction method based on deep learning | |
| CN113128271A (en) | Counterfeit detection of face images | |
| JP6722351B2 (en) | Learning device, image recognition device, learning method and program | |
| CN111161178A (en) | Single low-light image enhancement method based on generation type countermeasure network | |
| Xiao et al. | Single image dehazing based on learning of haze layers | |
| KR102095443B1 (en) | Method and Apparatus for Enhancing Image using Structural Tensor Based on Deep Learning | |
| CN110443270B (en) | Chart positioning method, apparatus, computer device and computer readable storage medium | |
| CN107172354A (en) | Method for processing video frequency, device, electronic equipment and storage medium | |
| US20230142630A1 (en) | Image Enhancement in a Genealogy System | |
| CN109684853A (en) | For determining and providing the device and method of the anonymous content in image | |
| CN115358952B (en) | Image enhancement method, system, equipment and storage medium based on meta-learning | |
| CN112241741A (en) | Self-adaptive image attribute editing model and method based on classified countermeasure network | |
| Chin et al. | Facial skin image classification system using Convolutional Neural Networks deep learning algorithm | |
| US11875494B2 (en) | Artificial intelligence architectures for determining image authenticity | |
| Rukundo et al. | Advanced image enhancement method for distant vessels and structures in capsule endoscopy | |
| JP7437918B2 (en) | Information processing device, information processing method, and program |
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 |