CN110428006A - The detection method of computer generated image, system, device - Google Patents
The detection method of computer generated image, system, device Download PDFInfo
- Publication number
- CN110428006A CN110428006A CN201910706992.2A CN201910706992A CN110428006A CN 110428006 A CN110428006 A CN 110428006A CN 201910706992 A CN201910706992 A CN 201910706992A CN 110428006 A CN110428006 A CN 110428006A
- Authority
- CN
- China
- Prior art keywords
- computer generated
- image
- generated image
- detection
- model
- 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
- 238000001514 detection method Methods 0.000 title claims abstract description 110
- 238000012549 training Methods 0.000 claims abstract description 84
- 238000000034 method Methods 0.000 claims abstract description 44
- 238000012360 testing method Methods 0.000 claims description 62
- 238000001914 filtration Methods 0.000 claims description 40
- 230000008569 process Effects 0.000 claims description 19
- 230000006870 function Effects 0.000 claims description 13
- 238000012545 processing Methods 0.000 claims description 5
- 238000004364 calculation method Methods 0.000 claims description 4
- ORILYTVJVMAKLC-UHFFFAOYSA-N Adamantane Natural products C1C(C2)CC3CC1CC2C3 ORILYTVJVMAKLC-UHFFFAOYSA-N 0.000 claims description 3
- 238000005457 optimization Methods 0.000 claims description 2
- 230000000694 effects Effects 0.000 description 5
- 238000013527 convolutional neural network Methods 0.000 description 4
- 238000010586 diagram Methods 0.000 description 4
- 230000008859 change Effects 0.000 description 3
- 238000011161 development Methods 0.000 description 3
- 238000005516 engineering process Methods 0.000 description 3
- 230000001815 facial effect Effects 0.000 description 3
- 238000002474 experimental method Methods 0.000 description 2
- 239000000463 material Substances 0.000 description 2
- 238000005259 measurement Methods 0.000 description 2
- 230000004913 activation Effects 0.000 description 1
- 230000006978 adaptation Effects 0.000 description 1
- 238000013528 artificial neural network Methods 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 238000006243 chemical reaction Methods 0.000 description 1
- 239000003086 colorant Substances 0.000 description 1
- 230000000052 comparative effect Effects 0.000 description 1
- 238000013480 data collection Methods 0.000 description 1
- 230000001419 dependent effect Effects 0.000 description 1
- 238000013461 design Methods 0.000 description 1
- 235000013399 edible fruits Nutrition 0.000 description 1
- 238000003384 imaging method Methods 0.000 description 1
- 230000006872 improvement Effects 0.000 description 1
- 230000010365 information processing Effects 0.000 description 1
- 238000007689 inspection Methods 0.000 description 1
- 238000007781 pre-processing Methods 0.000 description 1
- 230000001737 promoting effect Effects 0.000 description 1
- 230000000306 recurrent effect Effects 0.000 description 1
- 238000011160 research Methods 0.000 description 1
- 230000026676 system process Effects 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—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
Abstract
The invention belongs to computer generated image, computer vision and image forensics fields, more particularly to the detection method, system, device of a kind of computer generated image, it is intended to it is low to solve the problems, such as that active computer generates the image detection model Data Detection accuracy rate non-homogeneous to the training set of model.This system method includes obtaining image to be detected, as input picture；Based on input picture, image detection result is obtained by computer generated image detection model；Wherein, computer generated image detection model uses the CNN network struction based on DCGAN discrimination model.Invention enhances the generalization ability of computer generated image detection model, the accuracy rate of its Data Detection non-homogeneous to the training set of model is improved.
Description
Technical field
The invention belongs to computer generated image, computer vision and image forensics fields, and in particular to a kind of computer
Generate detection method, the system, device of image.
Background technique
In the intelligentized epoch, digital picture plays very important role.Have benefited from generating confrontation network (GAN)
Fast development, the various technologies based on GAN are more and more, wherein PGGAN and styleGAN is more outstanding in GAN development course
Generation model out can be generated human eye almost and true and false high-resolution facial image cannot be distinguished.However, some illegal points
Son carrys out registered face identifying system using the facial image that GAN is generated, and then the safety of face identification system is made to receive prestige
The side of body.In addition to this, the software of changing face based on GAN can accomplish effect of changing face very true to nature, this is directly affected dependent on image
Credible industry, such as press and publication sector, court's evidence obtaining, insurance etc..For the generation and replacement of facial image, there is
Very big risk, this makes the authenticity of image become the problem of being concerned, therefore digital image evidence collecting is increasingly becoming one
A important project.
Current digital image evidence obtaining field has had certain methods to be suggested, and has CNN- for the detection technique for video of changing face
RNN network detects automatically, video trace detection etc. of changing face.It can refer to document: (1) David G ¨ uera and Edward J
Delp,“Deepfake video detection using recurrent neural networks,”in IEEE
International Conference on Advanced Video and Signalbased Surveillance(to
appear),2018.(2)Yuezun Li and Siwei Lyu,“Exposing deepfake videos by
Detecting face warping artifacts, " arXiv preprint arXiv:1811.00656,2018. is directed to
The detection that GAN generates image has the detection model of image conversion, the detection model of Ensemble classifier, the inspection based on Weakly supervised study
Model etc. is surveyed, can refer to document: (1) Francesco Marra, Diego Gragnaniello, Davide Cozzolino,
and Luisa Verdoliva,“Detection of gangenerated fake images over social
networks,”in 2018IEEE Conference on Multimedia Information Processing and
Retrieval(MIPR).IEEE,2018,pp.384–389.(2)Shahroz Tariq,Sangyup Lee,Hoyoung
Kim,Youjin Shin,and Simon S Woo,“Detecting both machine and human created
fake face images in the wild,”in Proceedings of the 2nd International
Workshop on Multimedia Privacy and Security.ACM,2018,pp.81–87.(3)Davide
Cozzolino,Justus Thies,Andreas R¨ossler,Christian Riess,Matthias Nieβner,and
Luisa Verdoliva,“Forensictransfer:Weakly-supervised domain adaptation for
forgery detection,”arXiv preprint arXiv:1812.02510,2018.
Although having existed some methods for detecting computer generated image, these existing methods are nearly all
Specifically for a kind of detection of computer generated image, simultaneously for other detection effects for generating images non-homogeneous with training set
It is unknowable.Therefore it is proposed that a kind of detection method of computer generated image, can effectively promote detection model pair and instruction
Practice the detection performance for the computer generated image for collecting non-homogeneous.
Summary of the invention
In order to solve the above problem in the prior art, image detection model is generated to mould in order to solve active computer
The low problem of the non-homogeneous Data Detection accuracy rate of the training set of type, first aspect present invention propose a kind of computer generation
The method of the detection of image, this method comprises:
Step S10 obtains image to be detected, as input picture；
Step S20 is based on the input picture, obtains image detection result by computer generated image detection model；
Wherein,
The computer generated image detection model uses the CNN network struction based on DCGAN discrimination model；The model
Its training method are as follows:
Step A10 obtains training set of images；Described image training set includes computer generated image, true picture；
Step A20, is respectively adopted Gaussian noise, gaussian filtering pre-processes described image training set, obtains Gauss
Noisy operation training set and gaussian filtering operation training collection；
Step A30 is based on training set of images, Gaussian noise operation training collection, gaussian filtering operation training collection, respectively to institute
Computer generated image detection model is stated to be trained.
In some preferred embodiments, " Gaussian noise is respectively adopted, gaussian filtering carries out described image training set
Pretreatment ", method are as follows:
The range of the filtering core of standard deviation, gaussian filtering based on preset Gaussian noise carries out standard deviation, filtering core
It randomly selects；
Gaussian noise is respectively adopted after selection, gaussian filtering pre-processes described image training set.
In some preferred embodiments, its selection section of the standard deviation of the Gaussian noise is [0,5].
In some preferred embodiments, its selection collection of the filtering core of the gaussian filtering is combined into { 1,3,5,7 }.
In some preferred embodiments, the computer generated image detection model uses two points in the training process
Class cross entropy loss function carries out parameter optimization using Adam algorithm.
In some preferred embodiments, the computer generated image detection model presets four layers of convolutional layer, each
The step-length of convolutional layer is 2, be filled with 1, convolution kernel size is 4 × 4.
The second aspect of the present invention proposes a kind of system of the detection of computer generated image, which includes obtaining
Module, detection module；
The acquisition module is configured to obtain image to be detected, as input picture；
The detection module is configured to the input picture, is obtained by computer generated image detection model
Image detection result；
Wherein,
The computer generated image detection model uses the CNN network struction based on DCGAN discrimination model；The model
Its training method are as follows:
Step A10 obtains training set of images；Described image training set includes computer generated image, true picture；
Step A20, is respectively adopted Gaussian noise, gaussian filtering pre-processes described image training set, obtains Gauss
Noisy operation training set and gaussian filtering operation training collection；
Step A30 is based on training set of images, Gaussian noise operation training collection, gaussian filtering operation training collection, respectively to institute
Computer generated image detection model is stated to be trained.
The third aspect of the present invention proposes a kind of storage device, wherein be stored with a plurality of program, described program apply by
Processor loads and executes the detection method to realize above-mentioned computer generated image.
The fourth aspect of the present invention proposes a kind of processing setting, including processor, storage device；Processor is suitable for
Execute each program；Storage device is suitable for storing a plurality of program；Described program is suitable for being loaded by processor and being executed with reality
The detection method of existing above-mentioned computer generated image.
Beneficial effects of the present invention:
The present invention improves the standard of the computer generated image detection model Data Detection non-homogeneous to the training set of model
True rate.The present invention destroys by using gaussian filtering and gaussian noise image pretreatment operation and inhibits image, semantic level
Evidence obtaining clue, improves the statistical nature similitude between true picture and computer generated image on pixel layer, forces
The study of computer generated image detection model is to more inherent, significant features, and non-image content itself, enhances
It is non-to model training collection same to improve computer generated image detection model for the generalization ability of computer generated image detection model
The accuracy rate of the Data Detection in source.
Detailed description of the invention
By reading the detailed description done to non-limiting embodiment done referring to the following drawings, the application other
Feature, objects and advantages will become more apparent upon.
Fig. 1 is the flow diagram of the detection method of the computer generated image of an embodiment of the present invention；
Fig. 2 is the block schematic illustration of the detection method of the computer generated image of an embodiment of the present invention；
Fig. 3 is the exemplary diagram for the image that the PGGAN of an embodiment of the present invention is generated；
Fig. 4 is the exemplary diagram that the image of an embodiment of the present invention is compared through the front and back of Gaussian noise pretreatment operation；
Fig. 5 is the exemplary diagram that the image of an embodiment of the present invention is compared through the front and back of gaussian filtering pretreatment operation；
Fig. 6 is that the computer generated image detection model of an embodiment of the present invention is trained and the signal of the process of test
Figure.
Specific embodiment
To make the object, technical solutions and advantages of the present invention clearer, below in conjunction with attached drawing to the embodiment of the present invention
In technical solution be clearly and completely described, it is clear that described embodiments are some of the embodiments of the present invention, without
It is whole embodiments.Based on the embodiments of the present invention, those of ordinary skill in the art are not before making creative work
Every other embodiment obtained is put, shall fall within the protection scope of the present invention.
The application is described in further detail with reference to the accompanying drawings and examples.It is understood that this place is retouched
The specific embodiment stated is only used for explaining related invention, rather than the restriction to the invention.It also should be noted that in order to just
Part relevant to related invention is illustrated only in description, attached drawing.
It should be noted that in the absence of conflict, the features in the embodiments and the embodiments of the present application can phase
Mutually combination.
The detection method of computer generated image of the invention, as shown in Figure 1, comprising the following steps:
Step S10 obtains image to be detected, as input picture；
Step S20 is based on the input picture, obtains image detection result by computer generated image detection model；
Wherein,
The computer generated image detection model generates the discrimination model structure of confrontation network DCGAN based on depth convolution
It builds；Its training method of the model are as follows:
Step A10 obtains training set of images；Described image training set includes computer generated image, true picture；
Step A20, is respectively adopted Gaussian noise, gaussian filtering pre-processes described image training set, obtains Gauss
Noisy operation training set and gaussian filtering operation training collection；
Step A30 is based on training set of images, Gaussian noise operation training collection, gaussian filtering operation training collection, respectively to institute
Computer generated image detection model is stated to be trained.
In order to which the detection method more clearly to computer generated image of the present invention is illustrated, with reference to the accompanying drawing to this
Each step carries out expansion detailed description in a kind of embodiment of inventive method.
Hereafter in preferred embodiment, first computer generated image detection model is described in detail, then to raw using computer
The detection method that the computer generated image of the testing result of image to be detected is obtained at image detection model is described in detail.
1, the training of computer generated image detection model
1.1 building training samples
In the present invention, collect or generate three kinds of computer generated image data set PGGAN (Fpg)、DCGAN(Fdc)、
WGANGP(Fwg) and real image data collection celebA-HQ (Rcel), Fig. 3 shows the image generated by PGGAN, in human eye
Observation is lower to be almost beyond recognition true and false, these dummy's faces generated by GAN in real life and are not present.
Wherein FpgAnd RcelIt is the data set downloading original author and providing, FdcAnd FwgIt is to reappear out according to original author's document
The data set generated after image composer.
By RcelIt is divided into training set Rcel_trainWith test set Rcel_test, by FpgIt is divided into training set Fpg_trainAnd test set
Fpg_test.Choose Fpg_trainAnd Rcel_trainAs training set, F is chosenpg_test、Fdc、FwgAnd Rcel_testAs test set,
Fpg_train、Fpg_test、Fdc、FwgAnd Rcel_train、Rcel_testData set includes 10000 images.
1.2 carry out pretreatment operation to training set of images using gaussian filtering, Gaussian noise
Although computer generated image is very true to nature, still there is very big difference with true picture, especially in certain systems
Count feature.In order to change the statistical nature of image, the present invention is using gaussian filtering, Gaussian noise to true picture and calculating
Machine generates image and carries out pretreatment operation.
Gaussian noise is a kind of noise with Gaussian Profile probability density function.Wherein Gaussian Profile can be indicated such as public affairs
Shown in formula (1):
Wherein, z indicates that the gray value of image, u indicate the desired value of gray value, and σ indicates the standard deviation of gray value, σ2It indicates
The variance of gray value.
The gaussian filtering process of image is exactly that convolution is done in image and normal distribution.It is such as public in the mathematical expression of two-dimensional space
Shown in formula (2):
Wherein, u2+v2=r2, r is the radius of filtering core, and the value range of u, v are [- r, r].
Fig. 4 illustrates the contrast effect of Gaussian noise before and after the processing, and Fig. 5 illustrates the comparison effect before and after gaussian filtering process
Fruit.It can be seen that the statistical nature for the picture material level that image pretreatment operation inhibits low level unstable, at the same again by
In true picture and computer generated image image pretreatment operation all having the same, which increase between true and false image
Statistical nature similitude on pixel layer, and then the deeper time that can force detection model study to more inherences is significant
Feature, promote the Generalization Capability of detection model.
In order to increase the randomness of pretreatment operation, and then increase the diversity of model training collection, in the present invention, Gauss filter
The filtering core of wave is randomly choosed from 1,3,5,7 at random, the standard deviation of Gaussian noise from 0 to 5 between randomly choose.Data it is more
Sample can make model learning to more general feature, and then carry out the Generalization Capability of lift scheme.
In the present invention, three kinds of training sets are used, the training set F of pretreatment operation is respectively not usedpg_trainWith
Rcel_train；The training set F for having used gaussian filtering to operatepg_train_GBAnd Rcel-_train_GB；The instruction for having used Gaussian noise to operate
Practice collection Fpg_train_GNAnd Rcel_train_GN。
1.3 construct training computer generated image detection model
In the present invention, the discrimination model (CNN network) for generating confrontation network DCGAN based on depth convolution is used as computer
Generate image detection model.The input of model is the true picture or computer generated image that image size is 128 × 128, mould
Result 0 or 1 of the output of type to image resolution (0 is computer generated image, i.e. fault image, and 1 is true image).Model is as one
A two classifier is equipped with four layers of convolutional layer, and the step-length of all convolutional layers is 2, be filled with 1, convolution kernel size is 4 × 4.Model
During training, batch standardization and nonlinear activation function are introduced, while having used two classification cross entropy loss functions
As the loss function of computer generated image detection model, optimized using parameter of the Adam algorithm to model.
Training set data (F after the different images pretreatment operation obtained according to 1.2pg_trainAnd Rcel_train、
Fpg_train_GBAnd Rcel_train_GB、Fpg_train_GNAnd Rcel_train_GN) input as model, finally obtain three models difference
It is M, MGBAnd MGN, three models are all computer generated image detection models, are divided into three here, main convenient for below
It is tested.
1.4 refer to the performance of the model to computer generated image detection model, and according to testing result based on test set
Mark is analyzed
In the present invention, it is based on the trained computer generated image detection model of above-mentioned training process, chooses test set,
It is input to and obtains testing result in the model and analyzed, as shown in Figure 6.To ensure image pretreatment operation to lift scheme
The validity of Generalization Capability generates the test set of image detection model not by image pretreatment operation for measuring and calculation machine
Reason.
Each test set (Fpg_test、Fdc、FwgAnd Rcel_test) data set include 10000 images, by test set group
It is combined into three kinds of modes and is separately input to three kinds of models, Fpg_test、Fdc、FwgAnd Rcel_testThese three combinations are respectively
Rcel_testAnd Fpg_test、Rcel_testAnd Fdc、Rcel_testAnd Fwg.Wherein Rcel_testAnd Fpg_testInput as model measurement is
In order to be used as comparative experiments, the basic computer generated image detectability of Forensics Model is tested；Rcel_testAnd Fdc、
Rcel_testAnd FwgInput as model measurement is in order to examine the promotion effect of detection model Generalization Capability.
Accuracy rate (Accuracy) ACC, kidney-Yang rate (True Postive Rate) TPR and Kidney-Yin rate (True
Negative Rate) TNR is used as three performance indicators of test result, and test result is as shown in table 1:
Table 1
| No | Detector model | Testing set | ACC (%) | TPR (%) | TNR (%) |
| 1 | M | Fpg_test+Rcel_test | 95.45 | 95.12 | 95.77 |
| 2 | MGB | Fpg_test+Rcel_test | 94.28 | 94.28 | 95.47 |
| 3 | MGN | Fpg_test+Rcel_test | 95.02 | 94.65 | 95.38 |
| 4 | M | Fwg+Rcel_test | 64.62 | 95.12 | 34.12 |
| 5 | MGB | Fwg+Rcel_test | 68.07 | 93.08 | 43.06 |
| 6 | MGN | Fwg+Rcel_test | 68.28 | 94.65 | 41.91 |
| 7 | M | Fdc+Rcel_test | 60.55 | 95.12 | 25.98 |
| 8 | MGB | Fdc+Rcel_test | 64.05 | 93.08 | 35.02 |
| 9 | MGN | Fdc+Rcel_test | 66.38 | 94.65 | 38.11 |
As shown in table 1, experiment is divided into three parts by different test data sets.No is number, the 1st row to the 3rd row
Test data set (Testing set) is Fpg_testAnd Rcel_test, the test data set of the 4th row to the 6th row is FwgWith
Rcel_test, the test data set of the 7th row to the 9th row is FdcAnd Rcel_test.Comparing the 1st row does not have the M mould of image pretreatment operation
Type and the 2nd row, the 3rd row have image pretreatment operation M respectivelyGBAnd MGNModel, ACC, TPR and TNR are increasing image preprocessing
Later substantially without variation, it means that the image pretreatment operation proposed will not damage detection model (Detector
Model) stability.From first three rows it can also be seen that may be implemented very for carrying out test with the homologous data set of training set
High classification performance.
Can be seen that in the test data set homologous with training dataset from the data of the 1st row, detect ACC, TPR and
TNR is above the 95%, but the 4th row and the ACC and TNR of the 7th row are significantly lower than training dataset.This result means to detect
For model in the detection test image non-homogeneous with training set, Generalization Capability performance is very poor.
The 4th row and the 5th, 6 rows in comparison sheet 1, test data set is FwgAnd Rcel-_test, it can be seen that increasing image
After pretreatment operation, the TNR of model improves about 10%, and whole ACC is also improved.Similarly, compare the 7th row and
8th, 9 rows, TNR also have about 10% improvement.This shows image pretreatment operation proposed by the present invention for promoting computer
It is effective for generating the generalization ability of image detection model.
2, the detection method of computer generated image
Step S10 obtains image to be detected, as input picture.
The extreme similitude of computer generated image and true picture, meets the needs of socio-economic development, but with
This also brings some negative impacts to society simultaneously.Therefore carrying out authentication checks to the authenticity of computer generated image is
The important research scope in current computer generation technique field.
In the present embodiment, image to be detected is obtained by network or other means, as input picture.
Step S20 is based on the input picture, obtains image detection result by computer generated image detection model.
True picture refers to the image of the real world obtained by imaging devices such as digital camera, scanners, picture material
Reflect real world.And computer generated image (CG) refer to by computer by image procossing generate it is similar with real world
Image, the two smoothness, number of colours, histogram continuity and it is tiny in terms of be very different.
How does is to identify an image true picture or computer generated image in the present embodiment, by training
Computer generated image detection model judge input picture for true picture or computer generated image.I.e. by training
Computer generated image detection model, obtain the image detection of input picture as a result, if image detection result is 0, it is described
Input picture is computer generated image；If described image testing result is 1, the input picture is true picture.
A kind of detection system of computer generated image of second embodiment of the invention, as shown in Figure 2, comprising: obtain mould
Block 100, detection module 200；
Module 100 is obtained, is configured to obtain image to be detected, as input picture；
Detection module 200 is configured to the input picture, obtains image by computer generated image detection model
Testing result；
Wherein,
The computer generated image detection model uses the CNN network struction based on DCGAN discrimination model；The model
Its training method are as follows:
Step A10 obtains training set of images；Described image training set includes computer generated image, true picture；
Step A20, is respectively adopted Gaussian noise, gaussian filtering pre-processes described image training set, obtains Gauss
Noisy operation training set and gaussian filtering operation training collection；
Step A30 is based on training set of images, Gaussian noise operation training collection, gaussian filtering operation training collection, respectively to institute
Computer generated image detection model is stated to be trained.
The technical personnel in the technical field can be clearly understood that, for convenience and simplicity of description, foregoing description
The specific course of work of system and related explanation, can be no longer superfluous herein with reference to the corresponding process in signature embodiment of the method
It states.
It should be noted that the detection system of computer generated image provided by the above embodiment, only with above-mentioned each function
The division of module carries out for example, in practical applications, can according to need and by above-mentioned function distribution by different functions
Module is completed, i.e., by the embodiment of the present invention module or step decompose or combine again, for example, the mould of above-described embodiment
Block can be merged into a module, can also be further split into multiple submodule, to complete whole described above or portion
Divide function.For module involved in the embodiment of the present invention, the title of step, it is only for distinguish modules or step
Suddenly, it is not intended as inappropriate limitation of the present invention.
A kind of storage device of third embodiment of the invention, wherein be stored with a plurality of program, described program be suitable for by
Reason device loads and realizes the detection method of above-mentioned computer generated image.
A kind of processing unit of fourth embodiment of the invention, including processor, storage device；Processor is adapted for carrying out each
Program；Storage device is suitable for storing a plurality of program；Described program be suitable for by processor load and execute by realize it is above-mentioned in terms of
The detection method of calculation machine generation image.
The technical personnel in the technical field can be clearly understood that is do not described is convenienct and succinct, foregoing description
The specific work process and related explanation of storage device, processing unit, can be with reference to the corresponding process in signature method example, In
This is repeated no more.
Those skilled in the art should be able to recognize that, mould described in conjunction with the examples disclosed in the embodiments of the present disclosure
Block, method and step, can be realized with electronic hardware, computer software, or a combination of the two, software module, method and step pair
The program answered can be placed in random access memory (RAM), memory, read-only memory (ROM), electrically programmable ROM, electric erasable and can compile
Any other form of storage well known in journey ROM, register, hard disk, moveable magnetic disc, CD-ROM or technical field is situated between
In matter.In order to clearly demonstrate the interchangeability of electronic hardware and software, in the above description according to function generally
Describe each exemplary composition and step.These functions are executed actually with electronic hardware or software mode, depend on technology
The specific application and design constraint of scheme.Those skilled in the art can carry out using distinct methods each specific application
Realize described function, but such implementation should not be considered as beyond the scope of the present invention.
Term " first ", " second " etc. are to be used to distinguish similar objects, rather than be used to describe or indicate specific suitable
Sequence or precedence.
Term " includes " or any other like term are intended to cover non-exclusive inclusion, so that including a system
Process, method, article or equipment/device of column element not only includes those elements, but also including being not explicitly listed
Other elements, or further include the intrinsic element of these process, method, article or equipment/devices.
So far, it has been combined preferred embodiment shown in the drawings and describes technical solution of the present invention, still, this field
Technical staff is it is easily understood that protection scope of the present invention is expressly not limited to these specific embodiments.Without departing from this
Under the premise of the principle of invention, those skilled in the art can make equivalent change or replacement to the relevant technologies feature, these
Technical solution after change or replacement will fall within the scope of protection of the present invention.
Claims (9)
1. a kind of detection method of computer generated image, which is characterized in that this method comprises:
Step S10 obtains image to be detected, as input picture；
Step S20 is based on the input picture, obtains image detection result by computer generated image detection model；
Wherein,
The computer generated image detection model uses the CNN network struction based on DCGAN discrimination model；Its instruction of the model
Practice method are as follows:
Step A10 obtains training set of images；Described image training set includes computer generated image, true picture；
Step A20, is respectively adopted Gaussian noise, gaussian filtering pre-processes described image training set, obtains Gaussian noise
Operation training collection and gaussian filtering operation training collection；
Step A30 is based on training set of images, Gaussian noise operation training collection, gaussian filtering operation training collection, respectively to the meter
Calculation machine generates image detection model and is trained.
2. the detection method of computer generated image according to claim 1, which is characterized in that " Gauss is respectively adopted to make an uproar
Sound, gaussian filtering pre-process described image training set ", method are as follows:
The range of the filtering core of standard deviation, gaussian filtering based on preset Gaussian noise, carry out standard deviation, filtering core it is random
It chooses；
Gaussian noise is respectively adopted after selection, gaussian filtering pre-processes described image training set.
3. the detection method of computer generated image according to claim 2, which is characterized in that the mark of the Gaussian noise
Its selection section of quasi- difference is [0,5].
4. the detection method of computer generated image according to claim 2, which is characterized in that the filter of the gaussian filtering
Its selection collection of wave core is combined into { 1,3,5,7 }.
5. the detection method of computer generated image according to claim 1, which is characterized in that the computer generates figure
As detection model uses two classification cross entropy loss functions in the training process, Adam algorithm is used to carry out parameter optimization.
6. the detection method of computer generated image according to claim 1, which is characterized in that the computer generates figure
As detection model presets four layers of convolutional layer, the step-length of each convolutional layer is 2, be filled with 1, convolution kernel size is 4 × 4.
7. a kind of detection system of computer generated image, which is characterized in that the system includes obtaining module, detection module；
The acquisition module is configured to obtain image to be detected, as input picture；
The detection module is configured to the input picture, obtains image by computer generated image detection model
Testing result；
Wherein,
The computer generated image detection model uses the CNN network struction based on DCGAN discrimination model；Its instruction of the model
Practice method are as follows:
Step A10 obtains training set of images；Described image training set includes computer generated image, true picture；
Step A20, is respectively adopted Gaussian noise, gaussian filtering pre-processes described image training set, obtains Gaussian noise
Operation training collection and gaussian filtering operation training collection；
Step A30 is based on training set of images, Gaussian noise operation training collection, gaussian filtering operation training collection, respectively to the meter
Calculation machine generates image detection model and is trained.
8. a kind of storage device, wherein being stored with a plurality of program, which is characterized in that described program is applied and loaded and held by processor
It goes to realize the detection method of computer generated image described in any one of claims 1-6.
9. a kind of processing setting, including processor, storage device；Processor is adapted for carrying out each program；Storage device is fitted
For storing a plurality of program；It is characterized in that, described program is suitable for being loaded by processor and being executed to realize claim 1-6
The detection method of described in any item computer generated images.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910706992.2A CN110428006A (en) | 2019-08-01 | 2019-08-01 | The detection method of computer generated image, system, device |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910706992.2A CN110428006A (en) | 2019-08-01 | 2019-08-01 | The detection method of computer generated image, system, device |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN110428006A true CN110428006A (en) | 2019-11-08 |
Family
ID=68412082
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201910706992.2A Pending CN110428006A (en) | 2019-08-01 | 2019-08-01 | The detection method of computer generated image, system, device |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN110428006A (en) |
Cited By (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110942034A (en) * | 2019-11-28 | 2020-03-31 | 中国科学院自动化研究所 | Method, system and device for detecting multi-type depth network generated image |
| CN111291640A (en) * | 2020-01-20 | 2020-06-16 | 北京百度网讯科技有限公司 | Method and apparatus for recognizing gait |
| CN112949464A (en) * | 2021-02-26 | 2021-06-11 | 中国科学院自动化研究所 | Face changing counterfeiting detection method, system and equipment based on three-dimensional shape of human face |
Citations (9)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107025647A (en) * | 2017-03-09 | 2017-08-08 | 中国科学院自动化研究所 | Distorted image evidence collecting method and device |
| CN107194418A (en) * | 2017-05-10 | 2017-09-22 | 中国科学院合肥物质科学研究院 | A kind of Aphids in Rice Field detection method based on confrontation feature learning |
| CN108197525A (en) * | 2017-11-20 | 2018-06-22 | 中国科学院自动化研究所 | Face image synthesis method and device |
| CN108492265A (en) * | 2018-03-16 | 2018-09-04 | 西安电子科技大学 | CFA image demosaicing based on GAN combines denoising method |
| CN109118438A (en) * | 2018-06-29 | 2019-01-01 | 上海航天控制技术研究所 | A kind of Gaussian Blur image recovery method based on generation confrontation network |
| CN109190665A (en) * | 2018-07-30 | 2019-01-11 | 国网上海市电力公司 | A kind of general image classification method and device based on semi-supervised generation confrontation network |
| US20190080205A1 (en) * | 2016-09-16 | 2019-03-14 | Deep Learning Analytics, LLC | Systems and Methods for Deep Model Translation Generation |
| CN109543674A (en) * | 2018-10-19 | 2019-03-29 | 天津大学 | A kind of image copy detection method based on generation confrontation network |
| CN109767440A (en) * | 2019-01-11 | 2019-05-17 | 南京信息工程大学 | A kind of imaged image data extending method towards deep learning model training and study |
-
2019
- 2019-08-01 CN CN201910706992.2A patent/CN110428006A/en active Pending
Patent Citations (9)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20190080205A1 (en) * | 2016-09-16 | 2019-03-14 | Deep Learning Analytics, LLC | Systems and Methods for Deep Model Translation Generation |
| CN107025647A (en) * | 2017-03-09 | 2017-08-08 | 中国科学院自动化研究所 | Distorted image evidence collecting method and device |
| CN107194418A (en) * | 2017-05-10 | 2017-09-22 | 中国科学院合肥物质科学研究院 | A kind of Aphids in Rice Field detection method based on confrontation feature learning |
| CN108197525A (en) * | 2017-11-20 | 2018-06-22 | 中国科学院自动化研究所 | Face image synthesis method and device |
| CN108492265A (en) * | 2018-03-16 | 2018-09-04 | 西安电子科技大学 | CFA image demosaicing based on GAN combines denoising method |
| CN109118438A (en) * | 2018-06-29 | 2019-01-01 | 上海航天控制技术研究所 | A kind of Gaussian Blur image recovery method based on generation confrontation network |
| CN109190665A (en) * | 2018-07-30 | 2019-01-11 | 国网上海市电力公司 | A kind of general image classification method and device based on semi-supervised generation confrontation network |
| CN109543674A (en) * | 2018-10-19 | 2019-03-29 | 天津大学 | A kind of image copy detection method based on generation confrontation network |
| CN109767440A (en) * | 2019-01-11 | 2019-05-17 | 南京信息工程大学 | A kind of imaged image data extending method towards deep learning model training and study |
Non-Patent Citations (5)
| Title |
|---|
| ALEC RADFORD ET AL.: "Unsupervised Representation Learning with Deep Convolutional Generative Adversarial Networks", 《HTTPS://ARXIV.ORG/ABS/1511.06434》 * |
| XINSHENG XUAN ET AL.: "ON THE GENERALIZATION OF GAN IMAGE FORENSICS", 《HTTPS://ARXIV.ORG/ABS/1902.11153V1》 * |
| 王耀杰 等: "基于生成对抗网络的图像隐藏方案", 《信息网络安全》 * |
| 蔡利梅 等: "《数字图像处理》", 31 August 2014, 中国矿业大学出版社 * |
| 韩九强 等: "《数字图像处理 基于XAVIS组态软件》", 31 August 2018, 西安交通大学出版社 * |
Cited By (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110942034A (en) * | 2019-11-28 | 2020-03-31 | 中国科学院自动化研究所 | Method, system and device for detecting multi-type depth network generated image |
| CN111291640A (en) * | 2020-01-20 | 2020-06-16 | 北京百度网讯科技有限公司 | Method and apparatus for recognizing gait |
| CN111291640B (en) * | 2020-01-20 | 2023-02-17 | 北京百度网讯科技有限公司 | Method and apparatus for recognizing gait |
| CN112949464A (en) * | 2021-02-26 | 2021-06-11 | 中国科学院自动化研究所 | Face changing counterfeiting detection method, system and equipment based on three-dimensional shape of human face |
| CN112949464B (en) * | 2021-02-26 | 2021-12-14 | 中国科学院自动化研究所 | Face changing counterfeiting detection method, system and equipment based on three-dimensional shape of human face |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Kang et al. | Robust median filtering forensics using an autoregressive model | |
| Chang et al. | A forgery detection algorithm for exemplar-based inpainting images using multi-region relation | |
| Bayram et al. | Image manipulation detection | |
| CN107133948A (en) | Image blurring and noise evaluating method based on multitask convolutional neural networks | |
| Beheshtipour et al. | Deep learning for clustering of continuous gravitational wave candidates | |
| Shi et al. | A detection and classification approach for underwater dam cracks | |
| CN110428006A (en) | The detection method of computer generated image, system, device | |
| US20070223839A1 (en) | Signal noise estimation | |
| Chen et al. | An improved splicing localization method by fully convolutional networks | |
| CN107103595A (en) | Method, device, storage medium and the equipment of detection image change | |
| CN116168240A (en) | Arbitrary-direction dense ship target detection method based on attention enhancement | |
| Almawas et al. | Comparative performance study of classification models for image-splicing detection | |
| CN115393698A (en) | Digital image tampering detection method based on improved DPN network | |
| CN111368865A (en) | Method and device for detecting remote sensing image oil storage tank, readable storage medium and equipment | |
| Manu et al. | Tamper detection of social media images using quality artifacts and texture features | |
| van Werkhoven et al. | A Jungle Computing approach to common image source identification in large collections of images | |
| Bedi et al. | Estimating cover image for universal payload region detection in stego images | |
| Heng | Analysis of infrared images based on grey system and neural network | |
| KR20200124887A (en) | Method and Apparatus for Creating Labeling Model with Data Programming | |
| Zeng et al. | Exposing image splicing with inconsistent sensor noise levels | |
| Niskanen et al. | Experiments with SOM based inspection of wood | |
| Ahmed et al. | Blind copy-move forgery detection using SVD and KS test | |
| Mashaan et al. | Image Forgery Detection Using Low Dimensional Texture Feature Vector | |
| CN114842399B (en) | Video detection method, training method and device for video detection model | |
| CN117558011B (en) | Image text tampering detection method based on self-consistency matrix and multi-scale loss |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| RJ01 | Rejection of invention patent application after publication | ||
| RJ01 | Rejection of invention patent application after publication |
Application publication date: 20191108 |