CN109214327A - A kind of anti-face identification method based on PSO - Google Patents
A kind of anti-face identification method based on PSO Download PDFInfo
- Publication number
- CN109214327A CN109214327A CN201810992234.7A CN201810992234A CN109214327A CN 109214327 A CN109214327 A CN 109214327A CN 201810992234 A CN201810992234 A CN 201810992234A CN 109214327 A CN109214327 A CN 109214327A
- Authority
- CN
- China
- Prior art keywords
- pso
- target
- face
- label
- particle
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Granted
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V40/00—Recognition of biometric, human-related or animal-related patterns in image or video data
- G06V40/10—Human or animal bodies, e.g. vehicle occupants or pedestrians; Body parts, e.g. hands
- G06V40/16—Human faces, e.g. facial parts, sketches or expressions
- G06V40/172—Classification, e.g. identification
-
- 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
- G06N3/004—Artificial life, i.e. computing arrangements simulating life
- G06N3/006—Artificial life, i.e. computing arrangements simulating life based on simulated virtual individual or collective life forms, e.g. social simulations or particle swarm optimisation [PSO]
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V40/00—Recognition of biometric, human-related or animal-related patterns in image or video data
- G06V40/10—Human or animal bodies, e.g. vehicle occupants or pedestrians; Body parts, e.g. hands
- G06V40/16—Human faces, e.g. facial parts, sketches or expressions
- G06V40/168—Feature extraction; Face representation
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Health & Medical Sciences (AREA)
- General Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- Oral & Maxillofacial Surgery (AREA)
- General Health & Medical Sciences (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Life Sciences & Earth Sciences (AREA)
- General Engineering & Computer Science (AREA)
- Multimedia (AREA)
- Human Computer Interaction (AREA)
- Evolutionary Computation (AREA)
- Artificial Intelligence (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Bioinformatics & Computational Biology (AREA)
- Evolutionary Biology (AREA)
- Biomedical Technology (AREA)
- Biophysics (AREA)
- Computational Linguistics (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Image Analysis (AREA)
Abstract
A kind of anti-face identification method based on PSO, comprising the following steps: S1: data set is added in the pretreatment of attacker's human face data;S2: mixed data set training face classifier is utilized;S3: setting PSO parameter and confrontation attack parameter;It is optimal to disturbance rejection to start the searching of iterative evolution process for S4:PSO initialization;S5: extracting and material objectization fights accessory, carries out physical attacks test.The present invention is directed to black box face identification system, it is generated by PSO evolution strategy optimal to disturbance rejection, the material objectization to disturbance rejection is realized using facial accessory, can there are better model generalization ability and practical application value compared to other white-box attack strategies in number and the effective anti-recognition of face of realization in physical environment.
Description
Technical field
The present invention relates to computer visions, machine learning field, more particularly to a kind of anti-recognition of face side based on PSO
Method.
Background technique
With the fast development of artificial intelligence, template matching, PCA principal component point of the research method of recognition of face from early stage
Artificial extraction feature is analysed, then the deep learning of mainstream by now, this technology have moved to maturity and be successfully applied to market.
However there are natural defects for researcher's discovery neural network, pass through doing to resisting sample for confrontation attack strategies generation
Disturb precision of the deep learning model in the tasks such as image recognition.Attacker is generated by confrontation attack strategies with to disturbance rejection
Data sample, this will lead to model and predict to change to it to resisting sample input neural network, be reflected in recognition of face
It is attacker by can be in change of status under face identification system to personation to resisting sample in scene.
At present for recognition of face to attack resistance, i.e., the research of anti-recognition of face is mainly based upon whitepack recognition of face
The image attack of model.Such research is mainly manifested in and has ignored: can not obtain people in 1. reality scenes there are certain drawback
The internal information of face identifying system;2. picture, which locally or globally disturbs, can not be deployed in real scene;3. white-box attack seriously according to
Rely model, generalization ability is poor;4. harm caused by physical attacks is bigger.
In view of above research blind spot, the present invention is generated and is known for black box face by the evolution strategy of particle group optimizing
Other model to resisting sample, while using facial accessory constraint and physico to disturbance rejection, make it possible to be deployed in real scene,
Reach preferable physical attacks effect.
Summary of the invention
In order to overcome existing recognition of face to have ignored physical realizability to disturbance rejection, attack strategies to attack resistance
The problems such as generalization ability, the present invention provides a kind of anti-face identification method based on PSO consider not only setting black-box model
Simulation of real scenes, while also being realized using facial accessory to the physico of disturbance rejection.
In order to solve the above technical problem, the present invention provides the following technical solutions:
A kind of anti-face identification method based on PSO, includes the following steps:
S1: data set pretreatment: the facial image for testing the attacker of physical attacks is pre-processed, according to being chosen
The input requirements of face identification model network carry out cutting alignment to data;Number is added in pretreated attacker's human face data
It is mixed according to library and existing data set, for training face classifier;
S2: training face classifier: using the characteristic model of selected face identification system pre-training, to pretreated data
Collection is trained to obtain face classification device, and utilizes test set testing classification device precision;
S3: parameter setting: parameter and the parameter to attack resistance needed for setting PSO;
S4: evolution optimizing: progress PSO initialization first generates a certain number of different pure color face particles;PSO's
In each iteration, recognition of face is carried out to whole particles and obtains corresponding label confidence level ranking, according to the mark of each particle
Label confidence level ranking and facial accessory Pixel Information calculate the fitness of each particle, the individual optimal and population of more new particle
It is optimal, the speed and location information of each particle of final updating, iteration until reach setting maximum number of iterations or
Population's fitness convergence;
S5: physical attacks test: if successfully obtaining extracting to the confrontation accessory in resisting sample, printing is simultaneously resisting sample
It is worn on corresponding experimenter, tested for the physical attacks of the human face recognition model.
Further, in the step S1, existing data set is the part labels chosen from LFW data set in database
The Sub Data Set of composition can be used for providing target labels not can contact label.The experimenter of test physical attacks is that can connect
The accessible label of touching, wearable facial accessory, being different from LFW data set not can contact label.Prototype network it is defeated
Enter the size that requirement is facial image.
Further, in the step S2, the characteristic model of pre-training be it is that official provides, on other large data collection
Carry out the model that feature extraction was trained.Meanwhile pretreated data set is divided into 80% training set automatically in the training process,
20% test set.
In the step S3, the adjusted addition inertial factor of PSO evolution strategy, using the speed of formula (1) more new particle
Information:
vi=ω × vi+c1×rand()×(pbesti-xi)+c2×rand()×(gbest-xi) (1)
Wherein, vi、xiThe speed of respectively i-th particle and position, ω are inertial factor, and rand () is between (0,1)
Between random number, c1、c2For Studying factors, pbestiFor the history optimal location of i-th of particle, gbest is kind of a group discovery
Global optimum position;
Using the location information of formula (2) more new particle:
xi=xi+vi (2)
Inertial factor is updated using formula (3):
ω(t)=(ωini-ωend)(Gk-g)/Gk+ωend (3)
Wherein GkFor maximum number of iterations, ωiniFor the initial inertia factor, ωendWhen for iteration to maximum evolutionary generation
Inertial factor;
To the parameter of attack resistance are as follows: the true tag label of attacker, the target labels target for pretending to be object and confrontation
The smoothing factor κ of disturbance.
In the step S4, when PSO is initialized, for the original facial image of attacker, upper different pure color faces will be rendered
Initial position square of the rgb value as particle to resisting sample as different particles, at all pixels point of disturbance after portion's accessory
Battle array x, rate matrices v of the color changing rate as particle;The pattern of facial accessory includes faces' ornaments such as spectacle-frame;
In each iteration, algorithm predicts all particles input black box face identification system, exports to predict first three
Label confidence level ranking, be expressed as Top-3, while also obtaining the disturbance Pixel Information of each particle;According to more mesh of design
Scalar functions calculate the fitness fitness of each particle, and multiple objective function is the linear combination of the antagonism and flatness of disturbance;
Using the antagonism of formula (4) calculation perturbation:
Wherein, scorelabelIndicate the confidence of true class label when true tag is in Top-3, scoretopTable
The confidence for showing class top ranked in Top-3, when label is not in Top-3, the just mark top ranked with confidence level
It signs top and replaces label;scorecurr_targetIt is the confidence of current goal, rank indicates current goal curr_target
Ranking in Top-3.When target is not in Top-3, PSO operates in intermediate simulation process, and function should increase MF, MF
It is set to a sufficiently large penalty term;
Intermediate simulation process solves the PSO when target labels target is not in Top-3 can not be defeated using model prediction
Information calculates the problem of particle fitness out;The process introduces current goal curr_target as intermediate variable, and guidance solution is empty
Between be returned to solution space from Top-3 to target it is mobile.Curr_target is defined as follows: for the prediction knot of an image
Fruit:
If a. target is in Top-3, current goal curr_target is the target of attacker's setting;
If b. target is not in Top-3, using the class of the second high confidence score as current goal curr_
target;If the new class not occurred in an iteration prediction before occurring, using new class as current goal curr_target;
Using the flatness of formula (5), (6) calculation perturbation:
Wherein ri,jIt is being averaged for the RGB triple channel pixel value of the disturbance pixel at coordinate (i, j);
Multiple objective function is defined using formula (7):
Wherein κ is the smoothing factor of multiple objective function, and x is the original image of attack.
It, in digital environment can be by the face identification system directly using obtaining to resisting sample in the step S5
Mistake is classified as the target labels of setting;Before physical attacks test, by extracting to the confrontation accessory in resisting sample, pass through rotation
Its size adjusting is to adapt to the size of attacker's face by the operations such as correction, amplification;The confrontation accessory of material objectization is printed and wears,
The facial image that attacker is obtained by camera, inputs the human face recognition model and is predicted, obtains attack effect.
The present invention is directed to black box face identification system, without obtaining model inner parameter information, not by PSO evolution strategy
Break and adjust input to change prediction output, finally searching obtains optimal to disturbance rejection, wants so that the prediction of model reaches attack
It asks.And then the confrontation accessory by wearing material objectization, attacker can pretend to be him in change of status label under face identification system
People.
The invention has the benefit that the evolution strategy based on PSO can help attacker not knowing about recognition of face mould
In the case where type internal information, the confrontation face accessory for hiding personal identification is generated, protects individual privacy to a certain extent.Finally
Confrontation effect influenced by printing device performance, but be able to satisfy the requirement of actual use substantially.
Detailed description of the invention
Fig. 1 is a kind of algorithm flow chart of anti-face identification method based on PSO of the embodiment of the present invention.
Fig. 2 is the part attack effect schematic diagram of the embodiment of the present invention.
Specific embodiment
A specific embodiment of the invention is described in further detail with reference to the accompanying drawings of the specification.
Referring to Figures 1 and 2, a kind of anti-face identification method based on PSO, present invention use contain in LFW data set
The mixed data set of part labels is generated for black box face identification system by PSO evolution strategy to resisting sample, in number
Face identification system is attacked in environment and physical environment.
The present invention the following steps are included:
S1: data set pretreatment: the facial image for the experimenter (i.e. attacker) for testing physical attacks is pre-processed,
According to the input requirements of selected human face recognition model network, cutting alignment is carried out to data;By pretreated attacker's face
Database and the mixing of existing data set is added in data, for training face classifier;
S2: training face classifier: using the characteristic model of selected face identification system pre-training, to pretreated data
Collection is trained to obtain face classification device, and utilizes test set testing classification device precision;
S3: parameter setting: parameter and the parameter to attack resistance needed for setting PSO;
S4: evolution optimizing: progress PSO initialization first generates a certain number of different pure color face particles;PSO's
In each iteration, recognition of face is carried out to whole particles and obtains corresponding label confidence level ranking, according to the mark of each particle
Label confidence level ranking and facial accessory Pixel Information calculate the fitness of each particle, the individual optimal and population of more new particle
It is optimal, the speed and location information of each particle of final updating, iteration until reach setting maximum number of iterations or
Population's fitness convergence;
S5: physical attacks test: if successfully obtaining extracting to the confrontation accessory in resisting sample, printing is simultaneously resisting sample
It is worn on corresponding experimenter, tested for the physical attacks of the human face recognition model.
In the step S1, existing data set is that the part labels chosen from LFW data set are constituted in database
Sub Data Set can be used for providing target labels not can contact label.Test physical attacks experimenter be can contact,
The accessible label of wearable face accessory, being different from LFW data set not can contact label.The input requirements of prototype network
For the size of facial image.
In the step S2, the characteristic model of pre-training be it is that official provides, on other large data collection carry out feature
Extract the model trained.Meanwhile pretreated data set is divided into 80% training set, 20% test automatically in the training process
Collection.
In the step S3, the adjusted addition inertial factor of PSO evolution strategy, using the speed of formula (1) more new particle
Information:
vi=ω × vi+c1×rand()×(pbesti-xi)+c2×rand()×(gbest-xi) (1)
Wherein, vi、xiThe speed of respectively i-th particle and position, ω are inertial factor, and rand () is between (0,1)
Between random number, c1、c2For Studying factors, pbestiFor the history optimal location of i-th of particle, gbest is kind of a group discovery
Global optimum position;
Using the location information of formula (2) more new particle:
xi=xi+vi (2)
Inertial factor is updated using formula (3):
ω(t)=(ωini-ωend)(Gk-g)/Gk+ωend (3)
Wherein GkFor maximum number of iterations, ωiniFor the initial inertia factor, ωendWhen for iteration to maximum evolutionary generation
Inertial factor;
To the parameter of attack resistance are as follows: the true tag label of attacker, the target labels target for pretending to be object and confrontation
The smoothing factor κ of disturbance.
In the step S4, when PSO is initialized, for the original facial image of attacker, upper different pure color faces will be rendered
Initial position square of the rgb value as particle to resisting sample as different particles, at all pixels point of disturbance after portion's accessory
Battle array x, rate matrices v of the color changing rate as particle;The pattern of facial accessory includes faces' ornaments such as spectacle-frame;
In each iteration, algorithm predicts all particles input black box face identification system, exports to predict first three
Label confidence level ranking, be expressed as Top-3, while also obtaining the disturbance Pixel Information of each particle;According to more mesh of design
Scalar functions calculate the fitness fitness of each particle, and multiple objective function is the linear combination of the antagonism and flatness of disturbance;
Using the antagonism of formula (4) calculation perturbation:
Wherein, scorelabelIndicate the confidence of true class label when true tag is in Top-3, scoretopTable
The confidence for showing class top ranked in Top-3, when label is not in Top-3, the just mark top ranked with confidence level
It signs top and replaces label;scorecurr_targetIt is the confidence of current goal, rank indicates current goal curr_target
Ranking in Top-3.When target is not in Top-3, PSO operates in intermediate simulation process, and function should increase MF, MF
It is set to a sufficiently large penalty term;
Intermediate simulation process solves the PSO when target labels target is not in Top-3 can not be defeated using model prediction
Information calculates the problem of particle fitness out;The process introduces current goal curr_target as intermediate variable, and guidance solution is empty
Between be returned to solution space from Top-3 to target it is mobile.Curr_target is defined as follows: for the prediction knot of an image
Fruit:
If a. target is in Top-3, current goal curr_target is the target of attacker's setting;
If b. target is not in Top-3, using the class of the second high confidence score as current goal curr_
target;If the new class not occurred in an iteration prediction before occurring, using new class as current goal curr_target;
Using the flatness of formula (5), (6) calculation perturbation:
Wherein ri,jIt is being averaged for the RGB triple channel pixel value of the disturbance pixel at coordinate (i, j);
Multiple objective function is defined using formula (7):
Wherein κ is the smoothing factor of multiple objective function, and x is the original image of attack.
It, in digital environment can be by the face identification system directly using obtaining to resisting sample in the step S5
Mistake is classified as the target labels of setting;Before physical attacks test, by extracting to the confrontation accessory in resisting sample, pass through rotation
Its size adjusting is to adapt to the size of attacker's face by the operations such as correction, amplification;The confrontation accessory of material objectization is printed and wears,
The facial image that attacker is obtained by camera, inputs the human face recognition model and is predicted, obtains attack effect.
It is directed to black box face identification system for the present invention as described above, anti-recognition of face pair is generated by PSO evolution strategy
The embodiment introduction of anti-accessory.Final testing result, as shown in table 1,
Table 1
The present invention is based on evolution strategies, may be inferior to other confrontation attack algorithms on time complexity, but in digital rings
Success attack rate in border is suitable with other algorithms, and can will realize the confrontation in physical environment to disturbance rejection material object
Attack.It is merely illustrative and not restrictive for the invention.Those skilled in the art understand that in invention claim
Defined by many changes, modifications, and even equivalents may be made in spirit and scope, but fall within protection model of the invention
In enclosing.
Claims (6)
1. a kind of anti-face identification method based on PSO, it is characterised in that: described method includes following steps:
S1: data set pretreatment: the facial image for testing the attacker of physical attacks is pre-processed, and is known according to selected face
The input requirements of other prototype network carry out cutting alignment to data;Database is added in pretreated attacker's human face data,
It is mixed with existing data set, for training face classifier;
S2: training face classifier: using selected face identification system pre-training characteristic model, to pretreated data set into
Row training obtains face classification device, and utilizes test set testing classification device precision;
S3: parameter setting: parameter and the parameter to attack resistance needed for setting PSO;
S4: evolution optimizing: progress PSO initialization first generates a certain number of different pure color face particles;In each of PSO
In iteration, recognition of face is carried out to whole particles and obtains corresponding label confidence level ranking, is set according to the label of each particle
Reliability ranking and facial accessory Pixel Information calculate the fitness of each particle, and the individual optimal and population of more new particle is most
Excellent, the speed and location information of each particle of final updating, iteration is until reach the maximum number of iterations or kind of setting
Group's fitness convergence;
S5: physical attacks test: if successfully obtaining extracting to the confrontation accessory in resisting sample to resisting sample, printing and wear
On corresponding experimenter, test for the physical attacks of the human face recognition model.
2. a kind of anti-face identification method based on PSO as described in claim 1, it is characterised in that: in the step S1, number
It is the Sub Data Set that the part labels chosen from LFW data set are constituted according to data set existing in library, not can contact label,
It can be used for providing target labels.The experimenter of test physical attacks is the accessible mark of accessible, wearable facial accessory
Label, being different from LFW data set not can contact label, and the input requirements of prototype network are the size of facial image.
3. a kind of anti-face identification method based on PSO as claimed in claim 1 or 2, it is characterised in that: the step S2
In, the characteristic model of pre-training is model that official provides, that progress feature extraction was trained on other large data collection;Together
When, pretreated data set is divided into 80% training set, 20% test set automatically in the training process.
4. a kind of anti-face identification method based on PSO as claimed in claim 1 or 2, it is characterised in that: the step S3
In, the adjusted addition inertial factor of PSO evolution strategy, using the velocity information of formula (1) more new particle:
vi=ω × vi+c1×rand()×(pbesti-xi)+c2×rand()×(gbest-xi) (1)
Wherein, vi、xiThe speed of respectively i-th particle and position, ω are inertial factor, and rand () is between (0,1)
Random number, c1、c2For Studying factors, pbestiFor the history optimal location of i-th of particle, gbest be kind of group discovery it is global most
Excellent position;
Using the location information of formula (2) more new particle:
xi=xi+vi (2)
Inertial factor is updated using formula (3):
ω(t)=(ωini-ωend)(Gk-g)/Gk+ωend (3)
Wherein GkFor maximum number of iterations, ωiniFor the initial inertia factor, ωendInertia when for iteration to maximum evolutionary generation because
Son;
To the parameter of attack resistance are as follows: the true tag label of attacker, pretend to be the target labels target of object and to disturbance rejection
Smoothing factor κ.
5. a kind of anti-face identification method based on PSO as claimed in claim 4, it is characterised in that: in the step S4,
It, will be initial to resisting sample after the upper different pure color face accessories of rendering for the original facial image of attacker when PSO is initialized
As different particles, location matrix x of the rgb value as particle at all pixels point of disturbance, color changing rate is as grain
The rate matrices v of son;The pattern of facial accessory includes faces' ornaments such as spectacle-frame;
In each iteration, algorithm predicts all particles input black box face identification system, exports the mark to predict first three
Confidence level ranking is signed, is expressed as Top-3, while also obtaining the disturbance Pixel Information of each particle;According to the multiple target letter of design
Number calculates the fitness fitness of each particle, and multiple objective function is the linear combination of the antagonism and flatness of disturbance;
Using the antagonism of formula (4) calculation perturbation:
Wherein, scorelabelIndicate the confidence of true class label when true tag is in Top-3, scoretopIt indicates
The confidence of top ranked class in Top-3, when label is not in Top-3, the just label top ranked with confidence level
Top replaces label;scorecurr_targetIt is the confidence of current goal, rank indicates that current goal curr_target exists
Ranking in Top-3.When target is not in Top-3, PSO operates in intermediate simulation process, and function should increase MF, MF quilt
It is set as a sufficiently large penalty term;
Current goal curr_target is introduced as intermediate variable, the solution for guiding solution space to be returned to Top-3 to target is empty
Between move, curr_target is defined as follows:
For the prediction result of an image:
If a. target is in Top-3, current goal curr_target is the target of attacker's setting;
If b. target is not in Top-3, using the class of the second high confidence score as current goal curr_target;If
The new class not occurred in an iteration prediction before occurring, then using new class as current goal curr_target;
Using the flatness of formula (5), (6) calculation perturbation:
Wherein ri,jIt is being averaged for the RGB triple channel pixel value of the disturbance pixel at coordinate (i, j);
Multiple objective function is defined using formula (7):
Wherein κ is the smoothing factor of multiple objective function, and x is the original image of attack.
6. a kind of anti-face identification method based on PSO as claimed in claim 1 or 2, it is characterised in that: the step S5
In, directly utilize the mesh for obtaining that resisting sample can be classified as by the face identification system mistake in digital environment setting
Mark label;It, will by operations such as rotational correction, amplifications by extracting to the confrontation accessory in resisting sample before physical attacks test
Its size adjusting is to adapt to the size of attacker's face;The confrontation accessory for printing and wearing material objectization is attacked by camera acquisition
The facial image for the person of hitting inputs the human face recognition model and is predicted, obtains attack effect.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201810992234.7A CN109214327B (en) | 2018-08-29 | 2018-08-29 | Anti-face recognition method based on PSO |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201810992234.7A CN109214327B (en) | 2018-08-29 | 2018-08-29 | Anti-face recognition method based on PSO |
Publications (2)
Publication Number | Publication Date |
---|---|
CN109214327A true CN109214327A (en) | 2019-01-15 |
CN109214327B CN109214327B (en) | 2021-08-03 |
Family
ID=64985514
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201810992234.7A Active CN109214327B (en) | 2018-08-29 | 2018-08-29 | Anti-face recognition method based on PSO |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN109214327B (en) |
Cited By (12)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN109815887A (en) * | 2019-01-21 | 2019-05-28 | 浙江工业大学 | A kind of classification method of complex illumination servant's face image based on Multi-Agent Cooperation |
CN110175513A (en) * | 2019-04-15 | 2019-08-27 | 浙江工业大学 | A kind of guideboard identification attack defense method based on the optimization of multiple target road |
CN110516619A (en) * | 2019-08-29 | 2019-11-29 | 河南中原大数据研究院有限公司 | A kind of cos-attack recognition of face attack algorithm |
CN110795703A (en) * | 2019-09-20 | 2020-02-14 | 华为技术有限公司 | Data anti-theft method and related product |
CN111027628A (en) * | 2019-12-12 | 2020-04-17 | 支付宝(杭州)信息技术有限公司 | Model determination method and system |
CN111461239A (en) * | 2020-04-03 | 2020-07-28 | 成都考拉悠然科技有限公司 | White box attack method of CTC scene character recognition model |
CN111583146A (en) * | 2020-04-30 | 2020-08-25 | 济南博观智能科技有限公司 | Face image deblurring method based on improved multi-scale circulation network |
WO2020233564A1 (en) * | 2019-05-21 | 2020-11-26 | 华为技术有限公司 | Method and electronic device for detecting adversarial example |
CN113283377A (en) * | 2021-06-10 | 2021-08-20 | 重庆师范大学 | Face privacy protection method, system, medium and electronic terminal |
WO2021212675A1 (en) * | 2020-04-21 | 2021-10-28 | 清华大学 | Method and apparatus for generating adversarial sample, electronic device and storage medium |
CN114240732A (en) * | 2021-06-24 | 2022-03-25 | 中国人民解放军陆军工程大学 | Anti-patch generation method for attacking face verification model |
CN115171196A (en) * | 2022-08-25 | 2022-10-11 | 北京瑞莱智慧科技有限公司 | Face image processing method, related device and storage medium |
Citations (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN105513221A (en) * | 2015-12-30 | 2016-04-20 | 四川川大智胜软件股份有限公司 | ATM (Automatic Teller Machine) cheat-proof device and system based on three-dimensional human face identification |
CN108446765A (en) * | 2018-02-11 | 2018-08-24 | 浙江工业大学 | The multi-model composite defense method of sexual assault is fought towards deep learning |
-
2018
- 2018-08-29 CN CN201810992234.7A patent/CN109214327B/en active Active
Patent Citations (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN105513221A (en) * | 2015-12-30 | 2016-04-20 | 四川川大智胜软件股份有限公司 | ATM (Automatic Teller Machine) cheat-proof device and system based on three-dimensional human face identification |
CN108446765A (en) * | 2018-02-11 | 2018-08-24 | 浙江工业大学 | The multi-model composite defense method of sexual assault is fought towards deep learning |
Non-Patent Citations (2)
Title |
---|
佚名: "反人脸识别花样多 实际应用路还远", 《今日科技》 * |
李晓鹏: "基于PSO训练SVM的人脸识别", 《电脑知识与技术》 * |
Cited By (17)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN109815887B (en) * | 2019-01-21 | 2020-10-16 | 浙江工业大学 | Multi-agent cooperation-based face image classification method under complex illumination |
CN109815887A (en) * | 2019-01-21 | 2019-05-28 | 浙江工业大学 | A kind of classification method of complex illumination servant's face image based on Multi-Agent Cooperation |
CN110175513A (en) * | 2019-04-15 | 2019-08-27 | 浙江工业大学 | A kind of guideboard identification attack defense method based on the optimization of multiple target road |
WO2020233564A1 (en) * | 2019-05-21 | 2020-11-26 | 华为技术有限公司 | Method and electronic device for detecting adversarial example |
CN110516619A (en) * | 2019-08-29 | 2019-11-29 | 河南中原大数据研究院有限公司 | A kind of cos-attack recognition of face attack algorithm |
CN110795703A (en) * | 2019-09-20 | 2020-02-14 | 华为技术有限公司 | Data anti-theft method and related product |
CN110795703B (en) * | 2019-09-20 | 2024-04-16 | 华为技术有限公司 | Data theft prevention method and related product |
CN111027628A (en) * | 2019-12-12 | 2020-04-17 | 支付宝(杭州)信息技术有限公司 | Model determination method and system |
CN111027628B (en) * | 2019-12-12 | 2022-03-11 | 支付宝(杭州)信息技术有限公司 | Model determination method and system |
CN111461239B (en) * | 2020-04-03 | 2023-05-09 | 成都考拉悠然科技有限公司 | CTC scene character recognition model optimization method based on white-box attack |
CN111461239A (en) * | 2020-04-03 | 2020-07-28 | 成都考拉悠然科技有限公司 | White box attack method of CTC scene character recognition model |
WO2021212675A1 (en) * | 2020-04-21 | 2021-10-28 | 清华大学 | Method and apparatus for generating adversarial sample, electronic device and storage medium |
CN111583146A (en) * | 2020-04-30 | 2020-08-25 | 济南博观智能科技有限公司 | Face image deblurring method based on improved multi-scale circulation network |
CN111583146B (en) * | 2020-04-30 | 2023-05-23 | 济南博观智能科技有限公司 | Face image deblurring method based on improved multi-scale circulation network |
CN113283377A (en) * | 2021-06-10 | 2021-08-20 | 重庆师范大学 | Face privacy protection method, system, medium and electronic terminal |
CN114240732A (en) * | 2021-06-24 | 2022-03-25 | 中国人民解放军陆军工程大学 | Anti-patch generation method for attacking face verification model |
CN115171196A (en) * | 2022-08-25 | 2022-10-11 | 北京瑞莱智慧科技有限公司 | Face image processing method, related device and storage medium |
Also Published As
Publication number | Publication date |
---|---|
CN109214327B (en) | 2021-08-03 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN109214327A (en) | A kind of anti-face identification method based on PSO | |
Zhu et al. | A three-pathway psychobiological framework of salient object detection using stereoscopic technology | |
Guo et al. | Automatic image cropping for visual aesthetic enhancement using deep neural networks and cascaded regression | |
Liu et al. | Person re-identification: What features are important? | |
CN105608447A (en) | Method for detecting human face smile expression depth convolution nerve network | |
CN107808132A (en) | A kind of scene image classification method for merging topic model | |
CN109196526A (en) | For generating the method and system of multi-modal digital picture | |
CN108229277A (en) | Gesture identification, control and neural network training method, device and electronic equipment | |
CN109977739A (en) | Image processing method, device, storage medium and electronic equipment | |
CN109284749A (en) | Refine image recognition | |
CN109325443A (en) | A kind of face character recognition methods based on the study of more example multi-tag depth migrations | |
CN109978918A (en) | A kind of trajectory track method, apparatus and storage medium | |
CN108052884A (en) | A kind of gesture identification method based on improvement residual error neutral net | |
Almogdady et al. | A flower recognition system based on image processing and neural networks | |
CN108710893B (en) | Digital image camera source model classification method based on feature fusion | |
CN108717524A (en) | It is a kind of based on double gesture recognition systems and method for taking the photograph mobile phone and artificial intelligence system | |
CN108090561A (en) | Storage medium, electronic device, the execution method and apparatus of game operation | |
CN107944459A (en) | A kind of RGB D object identification methods | |
Jalab et al. | Human computer interface using hand gesture recognition based on neural network | |
CN109145871A (en) | Psychology and behavior recognition methods, device and storage medium | |
CN108229503A (en) | A kind of feature extracting method for clothes photo | |
CN108073851A (en) | A kind of method, apparatus and electronic equipment for capturing gesture identification | |
CN110956099B (en) | Dynamic gesture instruction identification method | |
CN107463954A (en) | A kind of template matches recognition methods for obscuring different spectrogram picture | |
CN109685713A (en) | Makeup analog control method, device, computer equipment and storage medium |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
PB01 | Publication | ||
PB01 | Publication | ||
SE01 | Entry into force of request for substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
GR01 | Patent grant | ||
GR01 | Patent grant |