CN108334816A - The Pose-varied face recognition method of network is fought based on profile symmetry constraint production - Google Patents
The Pose-varied face recognition method of network is fought based on profile symmetry constraint production Download PDFInfo
- Publication number
- CN108334816A CN108334816A CN201810033455.1A CN201810033455A CN108334816A CN 108334816 A CN108334816 A CN 108334816A CN 201810033455 A CN201810033455 A CN 201810033455A CN 108334816 A CN108334816 A CN 108334816A
- Authority
- CN
- China
- Prior art keywords
- network
- image
- face
- pose
- output
- 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
- 238000000034 method Methods 0.000 title claims abstract description 37
- 238000004519 manufacturing process Methods 0.000 title claims abstract description 18
- 238000012549 training Methods 0.000 claims abstract description 14
- 230000001815 facial effect Effects 0.000 claims abstract description 13
- 230000006870 function Effects 0.000 claims description 15
- 230000004069 differentiation Effects 0.000 claims description 14
- 238000012360 testing method Methods 0.000 claims description 12
- 238000005286 illumination Methods 0.000 claims description 11
- 239000011159 matrix material Substances 0.000 claims description 9
- 230000008569 process Effects 0.000 claims description 7
- 230000004913 activation Effects 0.000 claims description 6
- 230000000644 propagated effect Effects 0.000 claims description 6
- 238000013527 convolutional neural network Methods 0.000 claims description 5
- 238000013507 mapping Methods 0.000 claims description 3
- 230000009467 reduction Effects 0.000 claims description 3
- 239000000284 extract Substances 0.000 abstract description 2
- 230000036544 posture Effects 0.000 description 15
- 238000010586 diagram Methods 0.000 description 3
- 238000005516 engineering process Methods 0.000 description 3
- 230000000007 visual effect Effects 0.000 description 3
- 230000008859 change Effects 0.000 description 2
- 238000000605 extraction Methods 0.000 description 2
- 210000000887 face Anatomy 0.000 description 2
- 238000012545 processing Methods 0.000 description 2
- 241001269238 Data Species 0.000 description 1
- 238000013528 artificial neural network Methods 0.000 description 1
- 238000013135 deep learning Methods 0.000 description 1
- 230000007812 deficiency Effects 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 235000013399 edible fruits Nutrition 0.000 description 1
- 210000004709 eyebrow Anatomy 0.000 description 1
- 238000010606 normalization Methods 0.000 description 1
- 230000003287 optical effect Effects 0.000 description 1
- 210000004279 orbit Anatomy 0.000 description 1
- 238000003909 pattern recognition Methods 0.000 description 1
- 230000008447 perception Effects 0.000 description 1
- 230000000750 progressive effect Effects 0.000 description 1
- 238000011160 research Methods 0.000 description 1
- 238000012216 screening Methods 0.000 description 1
- 230000016776 visual perception Effects 0.000 description 1
- 239000011800 void material Substances 0.000 description 1
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
- G06F18/213—Feature extraction, e.g. by transforming the feature space; Summarisation; Mappings, e.g. subspace methods
- G06F18/2132—Feature extraction, e.g. by transforming the feature space; Summarisation; Mappings, e.g. subspace methods based on discrimination criteria, e.g. discriminant analysis
- G06F18/21322—Rendering the within-class scatter matrix non-singular
-
- 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 discloses the Pose-varied face recognition methods that network is fought based on profile symmetry constraint production, characterized in that includes the following steps：1）Data prediction；2）Profile constraint generates network；3）Symmetry constraint fights network；4）Balance training network；5）It rebuilds and identifies.The attitude angle deflection that this method can effectively solve facial image influences, extracts the feature that face has more robustness under multi-pose, especially global quality and local detail are mutually constrained under wide-angle posture reconstruction, the contour feature information for maintaining positive face can meet the requirements for high precision to Pose-varied face recognition in practical application.
Description
Technical field
The present invention relates to intelligent image processing and area of pattern recognition, and in particular to one kind is generated based on profile symmetry constraint
Formula fights network (Contour Symmetry Constraint Generative Adversarial Network, abbreviation SC-
GAN Pose-varied face recognition method).
Background technology
Pose-varied face recognition (Multi-pose Face recognition) is the heat of machine vision research in recent years
The it is proposed and rise of point, especially deep learning so that the great progress that the technology of recognition of face is all obtained in various fields
With quick development.However in reality, facial image is vulnerable to many factors such as varying environment, illumination, expression, posture, shadow
The precision of recognition of face is rung, human face posture is an extremely challenging problem among these.
In order to solve to change in the class that attitudes vibration in recognition of face is brought, researcher have been achieved for it is certain at
Fruit, current main technology are divided into two class method of 2D and 3D.2D methods mainly have：Stack step own coding (Stacked
Progressive Auto-encoders, abbreviation SPAE), non-face image is gradually reconstructed into using shallow-layer stepping own coding
Face image；Depth convolutional network (Deep Convolution Neural Network, abbreviation DCNN), to different postures and
The face extraction of illumination identity keeping characteristics, go out the face image of normal illumination using feature reconstruction.Although these methods are easy
In implementation, and the stronger feature of robustness can be extracted, but local grain information is lost too much, is rebuild under face image quality
Drop, influences subsequent recognition performance.In 3D methods, the main method for rebuilding difference with assessment depth information and minimum
The corresponding 3D faceforms of 2D faces are fitted, then unified face view, such as subjective epimorph are reconstructed by normalization
Type (View-Based Active Appearance, abbreviation VAAM) is shifted from 3D human face datas to generate the void of test pictures
Intend angle and is compared with the face image synthesized.Such methods need a large amount of depth informations, fit procedure and calculating process
It is excessively difficult.In recent years, production confrontation network obtains numerous researchers with its outstanding performance in visual perception task
Fervent concern.A kind of production confrontation network (Generative adversarial of the propositions such as Goodfellow
Networks, abbreviation GAN), which is made of generator and arbiter, and generator captures truthful data sample and generates newly
Sample data, arbiter differentiate the data of authentic specimen and generation to reach the balance of genuine/counterfeit discriminating, and production confrontation network is
A kind of model of unsupervised learning, a series of problems that can be generated with effective solution data, and have more in feature extraction
Add abundant ability of self-teaching.Because production fights network, to restore rich and saturability ability to image stronger, is
It solves the positive face of face that posture influences and provides new thinking.
Invention content
The purpose of the present invention is in view of the deficiencies of the prior art, and provides and a kind of fought based on profile symmetry constraint production
The Pose-varied face recognition method of network.The attitude angle deflection that this method can effectively solve facial image is influenced, is extracted
Face has more the feature of robustness under multi-pose, especially under wide-angle posture reconstruction that global quality and local detail is mutual
Constraint, maintains the contour feature information of positive face, can meet the requirements for high precision to Pose-varied face recognition in practical application.
Realizing the technical solution of the object of the invention is：
The Pose-varied face recognition method that network is fought based on profile symmetry constraint production, is included the following steps：
1) data prediction：In order to establish the positive face template characteristic of better face, multi-pose Face database is divided into instruction
Practice image and test image, and training image and test image are normalized；
2) profile constraint generates network：The image of any attitude under normal illumination is handled by convolution pondization first, it is defeated
The face image gone out is as the output for generating network, if it is arbitrary under w × h normal illuminations to generate the facial image inputted in network
Posture is x0, generating network is made of the convolutional neural networks of two convolutional layers, two pond layers, Wi 1It is produced for first layer convolution
Raw weight matrix mappings characteristics figure,For the characteristic pattern that second layer convolution generates, V1、V2It is the one or two layer of matrix pool respectively
Change is handled, so image x0The characteristic pattern obtained by two layers of convolutional network isRELU functions are selected, then：
Wherein, σ indicates activation primitive, and in generating the complete image of network reconfiguration, positive face profile histogram is added to constrain
The quality of global characteristics indicates the center of image with the arbitrary pixel (i, j) of face image f (x, y), calculate separately to
Determine gradient of the center pixel of window on the directions m and n, the edge of image coordinate is obtained by θAnd with volume
The image convolution of product network output, obtains the calculating y for reconstructing positive facei：
It is the positive face of reconstruct based on convolutional network in generation network, is declined using gradient and network parameter is constantly updated
Backpropagation, i.e., newer parameterΔ indicates intermediate variable,WhereinFor convolutional network
Backpropagation formula, contain reverse propagated error item eiWith characteristic variable xi-1,Before wherein characteristic variable is
One feature, it is hereby achieved that reverse propagated error eiIn the reversed error term of each layer；
3) symmetry constraint fights network：The differentiation network that network is a similar contrast device is fought, is generated by step 2)
The reconstructed image of network outputWith the positive face data of real human face as desired outputCarry out genuine/counterfeit discriminating cost functionIn order to make network progressively reach the phase of corresponding sample
Hope output, this part is according to arbiter reconstructed sample Pixel-levelWith true picture Pixel-levelDifferentiation loss introduce LrMelt
Close pixel loss function：
And according to this feature of facial symmetry, one side of something is covered to the width of the image of all inputs, gradually by the right coordinate
It blocks, by the absolute value of reconstructed sample image subtraction, gradually Expressive Features point, introducing are entangled for attitude reconstruction process septum reset
Positive Symmetric Loss LslSolve the symmetry of visible part and covering part reconstruct：
So final loss function is weighted formula (3), formula (4)：Lsyn=Lr+λ1Lsl+λ2Lcee, wherein LceeIt is
Cross entropy loss function, for limiting hiding activation primitive, λ1And λ2It is the coefficient for balancing penalty term, defines final loss
After function, using the more newly-generated network of Back Propagation Algorithm alternating and confrontation network weightAnd deviationTradeoff generates network
With the parameter of confrontation network
4) balance training network：After above-mentioned 3 step process, the independent alternating iteration undated parameter of network, first
Sequence is to fix to generate network G, training differentiation network D, so that the accuracy rate of differentiation is maximized, the sequence of next is fixed differentiation net
Network D, training generate network G, and the accuracy rate of differentiation is made to minimize, until determining and truthful dataImage about the same,
I.e. no matter for true and false sample, confrontation network all converts the two-dimensional matrix value of output to a probability value p；
5) it rebuilds and identifies：Network is fought by balancing production, will be inputted with the test image of different attitude angles
Into the network tested, the output image of generator is obtainedFor the image after reconstruction, the face face image that will have been rebuild
Fisher face, i.e. LDA methods dimensionality reduction is used to extract the face with identification respectively with network highest hidden layer feature
Feature is used in combination nearest neighbor classifier to complete recognition of face.
This method is using the Nonlinear Modeling ability of convolutional neural networks as the generator of network, by multi-pose
Visual angle successively positive twist, allow each corresponding posture feature variation to have more robustness, and use the positive face histogram constraint of face
The edge of profile, ensure that the quality of multi-pose multi-angle image rebuilding.And distribution is modeled based on production confrontation network
Ability, be used as arbiter by the positive face data of real human face to carry out genuine/counterfeit discriminating, due to introducing Symmetric Loss constraint, make
It obtains network and real human face information is had more to the positive face feature of the face rebuild in generator, improve the face number of screening resolved reconstruction
According to making the face image of reconstruction protrude more minutia information, whole smooth and noise spot is few also can be improved largely
Recognition efficiency.
The attitude angle deflection that this method efficiently solves facial image influences, and extracts face and is had more under multi-pose
The feature of robustness especially mutually constrains global quality and local detail under wide-angle posture reconstruction, maintains positive face
Contour feature information meets the requirements for high precision to Pose-varied face recognition in practical application.
Description of the drawings
Fig. 1 is the method flow schematic diagram of embodiment；
Fig. 2 is to fight network entire block diagram based on profile symmetry constraint production in embodiment；
Fig. 3 be embodiment in+75 ° on Multi-PIE data sets -- the partial reconstitution face image between 75 °；
Fig. 4 is the comparison diagram at 75 ° under difference method for reconstructing in embodiment.
Specific implementation mode
The content of present invention is described in further detail with reference to the accompanying drawings and examples, but is not the limit to the present invention
It is fixed.
Embodiment：
Referring to Fig.1, Fig. 2 fights the Pose-varied face recognition method of network based on profile symmetry constraint production, including such as
Lower step：
1) data prediction：In order to establish the positive face template characteristic of better face, multi-pose Face database is divided into instruction
Practice image and test image, and training image and test image are normalized, this example is in Multi-PIE facial images
The validity of the technical program is demonstrated on library, the database contain 337 people have altogether 754204 different postures, illumination,
The face picture of expression, this example take one of subgraph image set, contain 11 kinds of postures in -75 ° -+75 ° angular ranges,
With 15 ° for interval between posture, selected human face data is the normal expression under normal illumination, and the image in database is big
Small alignment cuts out and is set as 64x64, and as training image, remaining 127 people selects just in test set as test image preceding 210 people
Face image (0 °) is as the reference picture differentiated in network, as shown in Figure 3；
2) profile constraint generates network：The image of any attitude under normal illumination is handled by convolution pondization first, it is defeated
The face image gone out is as the output for generating network, if it is any attitude under normal illumination to generate the facial image inputted in network
For x0, generate the convolutional neural networks that network is convolutional layer by two 5 × 5, two 3 × 3 pond layers, Wi 1For first layer convolution
The weight matrix mappings characteristics figure of generation,For the characteristic pattern that second layer convolution generates, V1、V2It is the one or two layer of matrix respectively
Pondization processing, so image x0As the characteristic pattern obtained by two layers of convolutional network isRELU functions are selected, then：
Wherein, σ indicates activation primitive, and in generating the complete image of network reconfiguration, positive face profile histogram is added to constrain
The quality of global characteristics indicates the center of image with the arbitrary pixel (i, j) of face image f (x, y), calculate separately to
Determine gradient of the center pixel of window on the directions m and n, the edge of image coordinate is obtained by θAnd with volume
The image convolution of product network output obtains reconstructing positive face calculating yi：
It is the positive face of reconstruct based on convolutional network in generation network, is declined using gradient and network parameter is constantly updated
Backpropagation, i.e., newer parameterΔ indicates intermediate variable,WhereinFor convolutional network
Backpropagation formula, contain reverse propagated error item eiWith characteristic variable xi-1,Before wherein characteristic variable is
One feature, it is hereby achieved that reverse propagated error eiIn the reversed error term of each layer；
3) symmetry constraint fights network：The differentiation network that network is a similar contrast device is fought, is generated by step 2)
The reconstructed image of network outputWith the true positive face data as desired outputInto
Row genuine/counterfeit discriminating cost functionIn order to make network progressively reach the desired output of corresponding sample, this portion
Divide according to arbiter reconstructed sample Pixel-levelWith true picture Pixel-levelDifferentiation loss introduce LrMerge pixel loss letter
Number：
And according to this feature of facial symmetry, one side of something is covered to the width of the image of all inputs, gradually by the right coordinate
It blocks, by the absolute value of reconstructed sample image subtraction, gradually Expressive Features point, introducing are entangled for attitude reconstruction process septum reset
Positive Symmetric Loss LslSolve the symmetry of visible part and covering part reconstruct：
So final loss function is weighted formula (3), formula (4)：Lsyn=Lr+λ1Lsl+λ2Lcee, wherein LceeIt is
Cross entropy loss function, for limiting hiding activation primitive, λ1And λ2It is the coefficient for balancing penalty term, defines final loss
After function, using the more newly-generated network of Back Propagation Algorithm alternating and confrontation network weightAnd deviationTradeoff generates network
With the parameter of confrontation network
4) balance training network：After above-mentioned 3 step process, the independent alternating iteration undated parameter of network, first
Sequence is to fix to generate network G, training differentiation network D, so that the accuracy rate of differentiation is maximized, the sequence of next is fixed differentiation net
Network D, training generate network G, and the accuracy rate of differentiation is made to minimize, until determining and truthful dataImage about the same,
I.e. no matter for true and false sample, confrontation network all converts the two-dimensional matrix value of output to a probability value p；
5) it rebuilds and identifies：Network is fought by balancing production, will be inputted with the test image of different attitude angles
Into the network tested, the output image of generator is obtainedFor the image after reconstruction, the face face image that will have been rebuild
Fisher face, i.e. LDA methods dimensionality reduction is used to extract the face with identification respectively with network highest hidden layer feature
Feature is used in combination nearest neighbor classifier to complete recognition of face.
, can be by the human face rebuilding of different postures at face image using the above embodiments method, it can be intuitive by Fig. 3
Find out that the face image reconstructed with the method for the technical program can preferably recover the texture information of original image, wherein
1,3,5 rows are different faces from+75 ° -- 75 ° of different posture artworks；2,4,6 rows are this paper when corresponding different angle
The face face image that method reconstructs, it can be seen that method of the invention under ± 75 ° of wide-angle posture reconstructs
The positive face of face still keep the symmetry and optical clarity of original image；In Fig. 4, when taking human face posture angle at 75 °,
Illustrate the technical program method and GAN (production confrontation network), DCNN (depth convolutional neural networks), MVP (various visual angles senses
Perception model) even depth learning method comparison, from visual observation can intuitively find out the technical program rebuild wide-angle appearance
When state face, the better than other methods done on Facial symmetry, the technical program makes facial information more naturally, especially
The comparison that network is fought with common production also demonstrates the importance that symmetry loss function is added in the technical program, and at some
The face detail information detailed information that for example eyebrow, eye socket, the feature on beard retain compared with DCNN and MVP models more fully,
The reconstruction of face is also more abundant, ensure that more face characteristic.
Claims (1)
1. fighting the Pose-varied face recognition method of network based on profile symmetry constraint production, characterized in that including walking as follows
Suddenly：
1) data prediction：In order to establish the positive face template characteristic of better face, multi-pose Face database is divided into trained figure
Picture and test image, and training image and test image are normalized；
2) profile constraint generates network：The image of any attitude under normal illumination is handled by convolution pondization first, output
Face image is as the output for generating network, if it is any attitude under w × h normal illuminations to generate the facial image inputted in network
For x0, generating network is made of the convolutional neural networks of two convolutional layers, two pond layers, Wi 1It is generated for first layer convolution
Weight matrix mappings characteristics figure,For the characteristic pattern that second layer convolution generates, V1、V2It is at the one or two layer of matrix pool respectively
Reason, so image x0As the characteristic pattern obtained by two layers of convolutional network isRELU functions are selected, then：
Wherein, σ indicates activation primitive, and in generating the complete image of network reconfiguration, positive face profile histogram is added to constrain the overall situation
The quality of feature indicates the center of image with the arbitrary pixel (i, j) of face image f (x, y), calculates separately given window
Gradient of the center pixel of mouth on the directions m and n, the edge of image coordinate is obtained by θAnd with convolution net
The image convolution of network output, obtains the calculating y for reconstructing positive facei：
It is the positive face of reconstruct based on convolutional network in generation network, is declined using gradient and network parameter is constantly updated reversely
It propagates, i.e., newer parameterΔ indicates intermediate variable,WhereinFor the anti-of convolutional network
To propagation formula, reverse propagated error item e is containediWith characteristic variable xi-1,Wherein characteristic variable is previous item
Feature, it is hereby achieved that reverse propagated error eiIn the reversed error term of each layer；
3) symmetry constraint fights network：The differentiation network that network is a similar contrast device is fought, network is generated by step 2)
The reconstructed image of outputWith the positive face data of real human face as desired outputInto
Row genuine/counterfeit discriminating cost functionIn order to make network progressively reach the desired output of corresponding sample, this portion
Divide according to arbiter reconstructed sample Pixel-levelWith true picture Pixel-levelDifferentiation loss introduce LrMerge pixel loss
Function：
And according to this feature of facial symmetry, one side of something is covered to the width of the image of all inputs, the right coordinate is gradually blocked,
By the absolute value of reconstructed sample image subtraction, gradually Expressive Features point, introducing are directed to pair that attitude reconstruction process septum reset is corrected
Claim loss LslSolve the symmetry of visible part and covering part reconstruct：
So final loss function is weighted formula (3), formula (4)：Lsyn=Lr+λ1Lsl+λ2Lcee, wherein LceeIt is to intersect
Entropy loss function, for limiting hiding activation primitive, λ1And λ2It is the coefficient for balancing penalty term, defines final loss function
Afterwards, using the more newly-generated network of Back Propagation Algorithm alternating and confrontation network weightAnd deviationTradeoff generate network and
Fight the parameter of network
4) balance training network：After above-mentioned 3 step process, the independent alternating iteration undated parameter of network, sequence first
Be it is fixed generate network G, training differentiates network D, so that the accuracy rate of differentiation is maximized, the sequence of next be fixed differentiation network D,
Training generates network G, and the accuracy rate of differentiation is made to minimize, until determining and truthful dataImage about the same, i.e., without
By for true and false sample, confrontation network all converts the two-dimensional matrix value of output to a probability value p；
5) it rebuilds and identifies：Network is fought by balancing production, the test image with different attitude angles is input to survey
In the network tried, the output image of generator is obtainedFor the image after reconstruction, most by the face image rebuild and network
High hidden layer feature uses Fisher face, i.e. LDA methods dimensionality reduction to extract the face characteristic with identification, be used in combination respectively
Nearest neighbor classifier completes recognition of face.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810033455.1A CN108334816B (en) | 2018-01-15 | 2018-01-15 | Multi-pose face recognition method based on contour symmetric constraint generation type countermeasure network |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810033455.1A CN108334816B (en) | 2018-01-15 | 2018-01-15 | Multi-pose face recognition method based on contour symmetric constraint generation type countermeasure network |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN108334816A true CN108334816A (en) | 2018-07-27 |
| CN108334816B CN108334816B (en) | 2021-11-23 |
Family
ID=62924212
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201810033455.1A Expired - Fee Related CN108334816B (en) | 2018-01-15 | 2018-01-15 | Multi-pose face recognition method based on contour symmetric constraint generation type countermeasure network |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN108334816B (en) |
Cited By (24)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109146868A (en) * | 2018-08-27 | 2019-01-04 | 北京青燕祥云科技有限公司 | 3D Lung neoplasm generation method, device and electronic equipment |
| CN109255831A (en) * | 2018-09-21 | 2019-01-22 | 南京大学 | The method that single-view face three-dimensional reconstruction and texture based on multi-task learning generate |
| CN109344706A (en) * | 2018-08-28 | 2019-02-15 | 杭州电子科技大学 | It is a kind of can one man operation human body specific positions photo acquisition methods |
| CN109543827A (en) * | 2018-12-02 | 2019-03-29 | 清华大学 | Production fights network equipment and training method |
| CN109671084A (en) * | 2018-11-15 | 2019-04-23 | 华东交通大学 | A kind of measurement method of workpiece shapes |
| CN109684973A (en) * | 2018-12-18 | 2019-04-26 | 哈尔滨工业大学 | The facial image fill system of convolutional neural networks based on symmetrical consistency |
| CN109711386A (en) * | 2019-01-10 | 2019-05-03 | 北京达佳互联信息技术有限公司 | Obtain method, apparatus, electronic equipment and the storage medium of identification model |
| CN109815928A (en) * | 2019-01-31 | 2019-05-28 | 中国电子进出口有限公司 | A kind of face image synthesis method and apparatus based on confrontation study |
| CN109919018A (en) * | 2019-01-28 | 2019-06-21 | 浙江英索人工智能科技有限公司 | Image eyes based on reference picture automatically open method and device |
| CN110135336A (en) * | 2019-05-14 | 2019-08-16 | 腾讯科技（深圳）有限公司 | Training method, device and the storage medium of pedestrian's generation model |
| CN110321849A (en) * | 2019-07-05 | 2019-10-11 | 腾讯科技（深圳）有限公司 | Image processing method, device and computer readable storage medium |
| CN110427864A (en) * | 2019-07-29 | 2019-11-08 | 腾讯科技（深圳）有限公司 | A kind of image processing method, device and electronic equipment |
| CN110751098A (en) * | 2019-10-22 | 2020-02-04 | 中山大学 | Face recognition method for generating confrontation network based on illumination and posture |
| CN110796080A (en) * | 2019-10-29 | 2020-02-14 | 重庆大学 | Multi-pose pedestrian image synthesis algorithm based on generation of countermeasure network |
| CN111046707A (en) * | 2018-10-15 | 2020-04-21 | 天津大学青岛海洋技术研究院 | Face restoration network in any posture based on facial features |
| CN111523497A (en) * | 2020-04-27 | 2020-08-11 | 深圳市捷顺科技实业股份有限公司 | Face correction method and device and electronic equipment |
| CN111652798A (en) * | 2020-05-26 | 2020-09-11 | 浙江大华技术股份有限公司 | Human face pose migration method and computer storage medium |
| CN111680566A (en) * | 2020-05-11 | 2020-09-18 | 东南大学 | Hand sample face recognition method based on sliding block generation countermeasure network |
| CN111861949A (en) * | 2020-04-21 | 2020-10-30 | 北京联合大学 | Multi-exposure image fusion method and system based on generation countermeasure network |
| CN111860093A (en) * | 2020-03-13 | 2020-10-30 | 北京嘀嘀无限科技发展有限公司 | Image processing method, device, equipment and computer readable storage medium |
| CN112884030A (en) * | 2021-02-04 | 2021-06-01 | 重庆邮电大学 | Cross reconstruction based multi-view classification system and method |
| CN113111776A (en) * | 2021-04-12 | 2021-07-13 | 京东数字科技控股股份有限公司 | Method, device and equipment for generating countermeasure sample and storage medium |
| CN113140015A (en) * | 2021-04-13 | 2021-07-20 | 杭州欣禾圣世科技有限公司 | Multi-view face synthesis method and system based on generation countermeasure network |
| US11475608B2 (en) | 2019-09-26 | 2022-10-18 | Apple Inc. | Face image generation with pose and expression control |
Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106951867A (en) * | 2017-03-22 | 2017-07-14 | 成都擎天树科技有限公司 | Face identification method, device, system and equipment based on convolutional neural networks |
| CN107154023A (en) * | 2017-05-17 | 2017-09-12 | 电子科技大学 | Face super-resolution reconstruction method based on generation confrontation network and sub-pix convolution |
| CN107239766A (en) * | 2017-06-08 | 2017-10-10 | 深圳市唯特视科技有限公司 | A kind of utilization resists network and the significantly face of three-dimensional configuration model ajusts method |
| CN107292813A (en) * | 2017-05-17 | 2017-10-24 | 浙江大学 | A kind of multi-pose Face generation method based on generation confrontation network |
| CN107437077A (en) * | 2017-08-04 | 2017-12-05 | 深圳市唯特视科技有限公司 | A kind of method that rotation face based on generation confrontation network represents study |
| US20170372505A1 (en) * | 2016-06-23 | 2017-12-28 | LoomAi, Inc. | Systems and Methods for Generating Computer Ready Animation Models of a Human Head from Captured Data Images |
-
2018
- 2018-01-15 CN CN201810033455.1A patent/CN108334816B/en not_active Expired - Fee Related
Patent Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20170372505A1 (en) * | 2016-06-23 | 2017-12-28 | LoomAi, Inc. | Systems and Methods for Generating Computer Ready Animation Models of a Human Head from Captured Data Images |
| CN106951867A (en) * | 2017-03-22 | 2017-07-14 | 成都擎天树科技有限公司 | Face identification method, device, system and equipment based on convolutional neural networks |
| CN107154023A (en) * | 2017-05-17 | 2017-09-12 | 电子科技大学 | Face super-resolution reconstruction method based on generation confrontation network and sub-pix convolution |
| CN107292813A (en) * | 2017-05-17 | 2017-10-24 | 浙江大学 | A kind of multi-pose Face generation method based on generation confrontation network |
| CN107239766A (en) * | 2017-06-08 | 2017-10-10 | 深圳市唯特视科技有限公司 | A kind of utilization resists network and the significantly face of three-dimensional configuration model ajusts method |
| CN107437077A (en) * | 2017-08-04 | 2017-12-05 | 深圳市唯特视科技有限公司 | A kind of method that rotation face based on generation confrontation network represents study |
Cited By (35)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109146868A (en) * | 2018-08-27 | 2019-01-04 | 北京青燕祥云科技有限公司 | 3D Lung neoplasm generation method, device and electronic equipment |
| CN109344706A (en) * | 2018-08-28 | 2019-02-15 | 杭州电子科技大学 | It is a kind of can one man operation human body specific positions photo acquisition methods |
| CN109255831A (en) * | 2018-09-21 | 2019-01-22 | 南京大学 | The method that single-view face three-dimensional reconstruction and texture based on multi-task learning generate |
| CN109255831B (en) * | 2018-09-21 | 2020-06-12 | 南京大学 | Single-view face three-dimensional reconstruction and texture generation method based on multi-task learning |
| CN111046707A (en) * | 2018-10-15 | 2020-04-21 | 天津大学青岛海洋技术研究院 | Face restoration network in any posture based on facial features |
| CN109671084A (en) * | 2018-11-15 | 2019-04-23 | 华东交通大学 | A kind of measurement method of workpiece shapes |
| US11574199B2 (en) | 2018-12-02 | 2023-02-07 | Tsinghua University | Generative adversarial network device and training method thereof |
| CN109543827B (en) * | 2018-12-02 | 2020-12-29 | 清华大学 | Generating type confrontation network device and training method |
| CN109543827A (en) * | 2018-12-02 | 2019-03-29 | 清华大学 | Production fights network equipment and training method |
| CN109684973B (en) * | 2018-12-18 | 2023-04-07 | 哈尔滨工业大学 | Face image filling system based on symmetric consistency convolutional neural network |
| CN109684973A (en) * | 2018-12-18 | 2019-04-26 | 哈尔滨工业大学 | The facial image fill system of convolutional neural networks based on symmetrical consistency |
| CN109711386A (en) * | 2019-01-10 | 2019-05-03 | 北京达佳互联信息技术有限公司 | Obtain method, apparatus, electronic equipment and the storage medium of identification model |
| CN109919018A (en) * | 2019-01-28 | 2019-06-21 | 浙江英索人工智能科技有限公司 | Image eyes based on reference picture automatically open method and device |
| CN109815928A (en) * | 2019-01-31 | 2019-05-28 | 中国电子进出口有限公司 | A kind of face image synthesis method and apparatus based on confrontation study |
| CN110135336B (en) * | 2019-05-14 | 2023-08-25 | 腾讯科技（深圳）有限公司 | Training method, device and storage medium for pedestrian generation model |
| CN110135336A (en) * | 2019-05-14 | 2019-08-16 | 腾讯科技（深圳）有限公司 | Training method, device and the storage medium of pedestrian's generation model |
| CN110321849B (en) * | 2019-07-05 | 2023-12-22 | 腾讯科技（深圳）有限公司 | Image data processing method, device and computer readable storage medium |
| CN110321849A (en) * | 2019-07-05 | 2019-10-11 | 腾讯科技（深圳）有限公司 | Image processing method, device and computer readable storage medium |
| CN110427864A (en) * | 2019-07-29 | 2019-11-08 | 腾讯科技（深圳）有限公司 | A kind of image processing method, device and electronic equipment |
| US11475608B2 (en) | 2019-09-26 | 2022-10-18 | Apple Inc. | Face image generation with pose and expression control |
| CN110751098A (en) * | 2019-10-22 | 2020-02-04 | 中山大学 | Face recognition method for generating confrontation network based on illumination and posture |
| CN110751098B (en) * | 2019-10-22 | 2022-06-14 | 中山大学 | Face recognition method for generating confrontation network based on illumination and posture |
| CN110796080A (en) * | 2019-10-29 | 2020-02-14 | 重庆大学 | Multi-pose pedestrian image synthesis algorithm based on generation of countermeasure network |
| CN111860093A (en) * | 2020-03-13 | 2020-10-30 | 北京嘀嘀无限科技发展有限公司 | Image processing method, device, equipment and computer readable storage medium |
| CN111861949A (en) * | 2020-04-21 | 2020-10-30 | 北京联合大学 | Multi-exposure image fusion method and system based on generation countermeasure network |
| CN111861949B (en) * | 2020-04-21 | 2023-07-04 | 北京联合大学 | Multi-exposure image fusion method and system based on generation countermeasure network |
| CN111523497A (en) * | 2020-04-27 | 2020-08-11 | 深圳市捷顺科技实业股份有限公司 | Face correction method and device and electronic equipment |
| CN111523497B (en) * | 2020-04-27 | 2024-02-27 | 深圳市捷顺科技实业股份有限公司 | Face correction method and device and electronic equipment |
| CN111680566A (en) * | 2020-05-11 | 2020-09-18 | 东南大学 | Hand sample face recognition method based on sliding block generation countermeasure network |
| CN111652798B (en) * | 2020-05-26 | 2023-09-29 | 浙江大华技术股份有限公司 | Face pose migration method and computer storage medium |
| CN111652798A (en) * | 2020-05-26 | 2020-09-11 | 浙江大华技术股份有限公司 | Human face pose migration method and computer storage medium |
| CN112884030A (en) * | 2021-02-04 | 2021-06-01 | 重庆邮电大学 | Cross reconstruction based multi-view classification system and method |
| CN113111776A (en) * | 2021-04-12 | 2021-07-13 | 京东数字科技控股股份有限公司 | Method, device and equipment for generating countermeasure sample and storage medium |
| CN113140015A (en) * | 2021-04-13 | 2021-07-20 | 杭州欣禾圣世科技有限公司 | Multi-view face synthesis method and system based on generation countermeasure network |
| CN113140015B (en) * | 2021-04-13 | 2023-03-14 | 杭州欣禾圣世科技有限公司 | Multi-view face synthesis method and system based on generation countermeasure network |
Also Published As
| Publication number | Publication date |
|---|---|
| CN108334816B (en) | 2021-11-23 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN108334816A (en) | The Pose-varied face recognition method of network is fought based on profile symmetry constraint production | |
| CN108537743B (en) | Face image enhancement method based on generation countermeasure network | |
| CN109255831B (en) | Single-view face three-dimensional reconstruction and texture generation method based on multi-task learning | |
| Hu et al. | Pose-guided photorealistic face rotation | |
| Cao et al. | 3D aided duet GANs for multi-view face image synthesis | |
| CN110348330A (en) | Human face posture virtual view generation method based on VAE-ACGAN | |
| CN107423678A (en) | A kind of training method and face identification method of the convolutional neural networks for extracting feature | |
| CN109886881B (en) | Face makeup removal method | |
| CN110852941B (en) | Neural network-based two-dimensional virtual fitting method | |
| CN110751098B (en) | Face recognition method for generating confrontation network based on illumination and posture | |
| CN110211045A (en) | Super-resolution face image method based on SRGAN network | |
| CN105373777B (en) | A kind of method and device for recognition of face | |
| CN109815826A (en) | The generation method and device of face character model | |
| CN108932536A (en) | Human face posture method for reconstructing based on deep neural network | |
| Shu et al. | Kinship-guided age progression | |
| Yampolskiy et al. | Face recognition in the virtual world: recognizing avatar faces | |
| CN110211035A (en) | Merge the image super-resolution method of the deep neural network of mutual information | |
| CN111428667A (en) | Human face image correcting method for generating confrontation network based on decoupling expression learning | |
| CN108846343B (en) | Multi-task collaborative analysis method based on three-dimensional video | |
| CN110660020B (en) | Image super-resolution method of antagonism generation network based on fusion mutual information | |
| CN111476241B (en) | Character clothing conversion method and system | |
| CN102163330B (en) | Multi-view face synthesis method based on tensor resolution and Delaunay triangulation | |
| CN107371016A (en) | Based on asymmetric distortion without with reference to 3D stereo image quality evaluation methods | |
| CN105117712A (en) | Single-sample human face recognition method compatible for human face aging recognition | |
| Zhai et al. | Asian female facial beauty prediction using deep neural networks via transfer learning and multi-channel feature fusion |
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 | ||
| CF01 | Termination of patent right due to non-payment of annual fee |
Granted publication date: 20211123 |
|
| CF01 | Termination of patent right due to non-payment of annual fee |