CN107423701A - The non-supervisory feature learning method and device of face based on production confrontation network - Google Patents

The non-supervisory feature learning method and device of face based on production confrontation network Download PDF

Info

Publication number
CN107423701A
CN107423701A CN201710581981.7A CN201710581981A CN107423701A CN 107423701 A CN107423701 A CN 107423701A CN 201710581981 A CN201710581981 A CN 201710581981A CN 107423701 A CN107423701 A CN 107423701A
Authority
CN
China
Prior art keywords
face
network
generation
image
target
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
Application number
CN201710581981.7A
Other languages
Chinese (zh)
Other versions
CN107423701B (en
Inventor
王栋
杨东
周孺
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Athena Eyes Co Ltd
Original Assignee
Athena Eyes Science & Technology Co Ltd
Priority date (The priority date 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 date listed.)
Filing date
Publication date
Application filed by Athena Eyes Science & Technology Co Ltd filed Critical Athena Eyes Science & Technology Co Ltd
Priority to CN201710581981.7A priority Critical patent/CN107423701B/en
Publication of CN107423701A publication Critical patent/CN107423701A/en
Application granted granted Critical
Publication of CN107423701B publication Critical patent/CN107423701B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

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
    • G06V40/161—Detection; Localisation; Normalisation
    • G06V40/165—Detection; Localisation; Normalisation using facial parts and geometric relationships
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
    • G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/045—Combinations of networks
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/048—Activation functions
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V10/00—Arrangements for image or video recognition or understanding
    • G06V10/20—Image preprocessing
    • G06V10/32—Normalisation of the pattern dimensions

Landscapes

  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • General Physics & Mathematics (AREA)
  • Health & Medical Sciences (AREA)
  • Data Mining & Analysis (AREA)
  • General Engineering & Computer Science (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Artificial Intelligence (AREA)
  • Evolutionary Computation (AREA)
  • General Health & Medical Sciences (AREA)
  • Computational Linguistics (AREA)
  • Computing Systems (AREA)
  • Molecular Biology (AREA)
  • Biophysics (AREA)
  • Mathematical Physics (AREA)
  • Software Systems (AREA)
  • Biomedical Technology (AREA)
  • Multimedia (AREA)
  • Oral & Maxillofacial Surgery (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Evolutionary Biology (AREA)
  • Geometry (AREA)
  • Bioinformatics & Computational Biology (AREA)
  • Human Computer Interaction (AREA)
  • Bioinformatics & Cheminformatics (AREA)
  • Image Processing (AREA)
  • Image Analysis (AREA)

Abstract

The invention discloses a kind of non-supervisory feature learning method and device of face based on production confrontation network, by being pre-processed to the original facial image collected, to be converted into the face training image being sized；The target generation network in the depth convolution generation confrontation network of structure is trained using the face training image changed as training data；The random vector collection of generation is input in the target generation network trained, obtains the generation image set corresponding with random vector collection；In the depth Recurrent networks for the depth convolutional neural networks that obtained generation image set is input into structure, depth Recurrent networks are trained, the face feature vector of extraction generation image set.Face non-supervisory feature learning method and apparatus provided by the invention based on production confrontation network, unsupervised learning is done by the way of DCGAN and DCNN are combined, network is generated to learn a reverse target using depth Recurrent networks, results of learning are good, and accuracy of identification is high.

Description

The non-supervisory feature learning method and device of face based on production confrontation network
Technical field
The present invention relates to technical field of face recognition, especially, it is non-to be related to a kind of face based on production confrontation network Supervise feature learning method and device.
Background technology
With the development of deep learning, the accumulation of internet big data and the development of hardware, current recognition of face skill Art compared with 10 years before have qualitative leap, be widely used in the field of authentication such as security protection, finance.In public data collection IFW In (Information Frame Work, information describing framework), ERR (mistake) can be also reduced to 1% by most of company Within.But the face recognition algorithms for being currently based on deep learning are all based on supervised learning greatly, it is necessary to the sample largely marked, Such as data set as 20,000 people everyone more than 50.The collection of data expends substantial amounts of manpower and materials and financial resources, so technology Barrier become data.And the face recognition algorithms based on unsupervised learning do not obtain preferable effect always.
Therefore, the existing face recognition algorithms using supervised learning are, it is necessary to substantial amounts of sample and the substantial amounts of manpower of consuming Material resources, and do not obtain preferable effect always using the face recognition algorithms based on unsupervised learning, it is one urgently to be resolved hurrily Technical problem.
The content of the invention
The invention provides a kind of non-supervisory feature learning method and device of face based on production confrontation network, with solution The certainly existing face recognition algorithms using supervised learning are, it is necessary to substantial amounts of sample and expend substantial amounts of manpower and materials, and use Face recognition algorithms based on unsupervised learning do not obtain the technical problem of preferable effect always.
The technical solution adopted by the present invention is as follows：
According to an aspect of the present invention, there is provided a kind of non-supervisory feature learning side of face based on production confrontation network Method, including step：
The original facial image collected is pre-processed, to be converted into the face training image being sized；
The mesh in network is resisted using the face training image changed as depth convolution generation of the training data to structure Mark generation network is trained；
The random vector collection of generation is input in the target generation network trained, obtained corresponding with random vector collection Generation image set；
In the depth Recurrent networks for the depth convolutional neural networks that obtained generation image set is input into structure, to depth Recurrent networks are trained, the face feature vector of extraction generation image set.
Further, the original facial image collected is pre-processed, trained with the face for being converted into being sized The step of image, includes：
Face datection is carried out to the original facial image collected, to detect the eyes coordinates of facial image；
The face in original facial image is alignd using eyes coordinates and normalized, to be converted into setting chi Very little face training image.
Further, using depth convolution generation confrontation net of the face training image changed as training data to structure The step of target generation network in network is trained includes：
The network structure that confrontation network is generated to former depth convolution is improved, and is built and new is used to generate human face target figure The target generation network of picture and the target-recognition network for being differentiated to the human face target image of generation；
Depth convolutional layer is added in the target generation network of structure, makes the random vector conversion of input target generation network Human face target image to be sized is exported.
Further, structure target generation network in add depth convolutional layer, make input target generation network with Machine vector includes after being converted to the step of human face target image being sized is exported：
The human face target image of target generation network output is differentiated with target-recognition network, determines face mesh Logo image and face true picture.
Further, obtained generation image set is input to the depth Recurrent networks of the depth convolutional neural networks of structure In, depth Recurrent networks are trained, extraction generation image set face feature vector the step of include：
Image set will be generated as the inputs of depth convolutional neural networks, random vector collection as depth convolutional neural networks Supervisory signals and depth Recurrent networks are trained using Euler's distance function as excitation function；
By the face feature vector of the depth Recurrent networks extraction generation image set trained, to identify face to be identified Face characteristic in image.
According to another aspect of the present invention, there is provided a kind of non-supervisory feature learning dress of face based on production confrontation network Put, including：
Pretreatment module, for being pre-processed to the original facial image collected, to be converted into the people being sized Face training image；
First training module, the face training image for that will change are given birth to as training data to the depth convolution of structure It is trained into the target generation network in confrontation network；
Acquisition module, for the random vector collection of generation is input to train target generation network in, obtain with The corresponding generation image set of machine vector set；
Second training module, the depth of the depth convolutional neural networks for obtained generation image set to be input to structure In Recurrent networks, depth Recurrent networks are trained, the face feature vector of extraction generation image set.
Further, pretreatment module includes detection unit and converting unit,
Detection unit, for carrying out Face datection to the original facial image collected, to detect the eye of facial image Eyeball coordinate；
Converting unit, for being alignd using eyes coordinates to the face in original facial image and normalized, To be converted into the face training image being sized.
Further, acquisition module includes construction unit and adding device,
Construction unit, the network structure for generating confrontation network to former depth convolution are improved, and build new be used for Generate the target generation network of human face target image and the target-recognition net for being differentiated to the human face target image of generation Network；
Adding device, for adding depth convolutional layer in generating network in the target of structure, input target is set to generate network Random vector be converted to the human face target image being sized and exported.
Further, acquisition module also includes judgement unit,
Judgement unit, for sentencing with target-recognition network to the human face target image of target generation network output Not, human face target image and face true picture are determined.
Further, the second training module includes training unit and extraction unit,
Training unit, for image set will to be generated as the inputs of depth convolutional neural networks, random vector collection as depth Spend the supervisory signals of convolutional neural networks and depth Recurrent networks are instructed using Euler's distance function as excitation function Practice；
Extraction unit, the face feature vector of image set is generated for the depth Recurrent networks extraction by training, with Identify the face characteristic in facial image to be identified.
The invention has the advantages that：
Face non-supervisory feature learning method and apparatus provided by the invention based on production confrontation network, are used The mode that DCGAN and DCNN are combined does unsupervised learning, is given birth to using depth Recurrent networks to learn a reverse target Into network, so as to achieve effect more more preferable than general unsupervised learning.It is provided by the invention that network is resisted based on production The non-supervisory feature learning method and apparatus of face, results of learning are good, and accuracy of identification is high.
In addition to objects, features and advantages described above, the present invention also has other objects, features and advantages. Below with reference to figure, the present invention is further detailed explanation.
Brief description of the drawings
The accompanying drawing for forming the part of the application is used for providing a further understanding of the present invention, schematic reality of the invention Apply example and its illustrate to be used to explain the present invention, do not form inappropriate limitation of the present invention.In the accompanying drawings：
Fig. 1 is the flow of face non-supervisory feature learning method preferred embodiment of the present invention based on production confrontation network Schematic diagram；
Fig. 2 is the schematic network structure of target generation network G in DCGAN networks in the prior art；
Fig. 3 is the schematic network structure of target generation network G in DCGAN networks of the present invention；
Fig. 4 is the schematic network structure of depth Recurrent networks E in DCNN networks of the present invention；
Fig. 5 is that the original facial image collected is pre-processed in Fig. 1, is trained with the face for being converted into being sized The refinement schematic flow sheet of the step of image；
Fig. 6 be the random vector collection of generation is input in Fig. 1 train target generation network in, obtain with random to Corresponding refinement schematic flow sheet the step of generating image set of quantity set；
Fig. 7 is that the depth that obtained generation image set is input to the depth convolutional neural networks of structure in Fig. 1 returns net In network, depth Recurrent networks are trained, extraction generation image set face feature vector the step of refinement flow signal Figure；
Fig. 8 is the function of face non-supervisory feature learning device preferred embodiment of the present invention based on production confrontation network Block diagram；
Fig. 9 is the high-level schematic functional block diagram of pretreatment module preferred embodiment in Fig. 8；
Figure 10 is the high-level schematic functional block diagram of acquisition module preferred embodiment in Fig. 8；
Figure 11 is the high-level schematic functional block diagram of the second training module preferred embodiment in Fig. 8.
Drawing reference numeral explanation：
10th, pretreatment module；20th, the first training module；30th, acquisition module；40th, the second training module；11st, detection is single Member；12nd, converting unit；31st, construction unit；32nd, adding device；33rd, judgement unit；41st, training unit；42nd, extraction unit.
Embodiment
It should be noted that in the case where not conflicting, the feature in embodiment and embodiment in the application can phase Mutually combination.Describe the present invention in detail below with reference to the accompanying drawings and in conjunction with the embodiments.
Reference picture 1, the preferred embodiments of the present invention provide a kind of non-supervisory spy of face based on production confrontation network Levy learning method, including step：
Step S100, the original facial image collected is pre-processed, trained with the face for being converted into being sized Image.
The original facial image collected is pre-processed, the face that original facial image is converted into being sized is instructed Practice image, so that changing good face training image meets the needs of recognition of face.In the present embodiment, it is original by what is collected Facial image is converted into the face training image being sized, to meet in recognition of face the needs of to resolution ratio.
Step S200, using depth convolution generation confrontation net of the face training image changed as training data to structure Target generation network in network is trained.
As shown in figure 3, the face training image changed is inputted to DCGAN (the Deep Convolutional of structure Generative Adverserial Networks, depth convolution generation confrontation network) in G networks (target generation network) In, G networks are trained using face training image as training data.
GAN (Generative Adverserial Networks, depth convolution generation confrontation network) is Ian Goodfellow 2014 propose a kind of unsupervised-learning algorithm, the main thought of algorithm is to pass through target in training process Generate network G and Target Countermeasure network D carries out 0-1 games confrontation study, final convergence obtains a target generation network G.It is defeated The true D grader that match in excellence or beauty can be obtained and be judged as really counting to target generation network G by entering a random vector rand_vec According to.
DCGAN (Deep Convolutional Generative Adverserial Networks, the generation of depth convolution Resist network) it is Alec Radford et al. by GAN thought and the CNN (Convolutional suitable for field of machine vision Generative Adverserial Networks, convolutional neural networks) combine, carrying out training image using GAN thought generates Device G..In the present embodiment, unsupervised learning is realized using DCGAN, but the branch for also having other GAN can reach this mesh , the algorithm used is not limited to use DCGAN, extends to the GAN algorithms of other generation images.It is a kind of common DCGAN G network architecture diagrams, as shown in Fig. 2 the stochastic variable of one 100 dimension of input, final by several layers of DCONV layers To the image of 64x64 RGB3 passages.
Step S300, the random vector collection of generation is input to train target generation network in, obtain with random to The corresponding generation image set of quantity set.
Random vector collection Z, random vector collection Z are generated by setting quantity and the random pictures group of dimension using random generator Conjunction forms, and is stored in database.In the present embodiment, the random vector collection of generation is input in the G networks trained, Obtain the generation image set P corresponding with random vector collection Z.Wherein, the quantity of random pictures is 1,000,000 in random vector collection Z Individual, dimension is 100 dimensions.
Step S400, obtained generation image set is input to the depth Recurrent networks of the depth convolutional neural networks of structure In, depth Recurrent networks are trained, the face feature vector of extraction generation image set.
See Fig. 4, obtained generation image set is input to DCNN (the Deep Convolutional Neural of structure Network, depth convolutional neural networks) E networks (depth Recurrent networks) in, E networks are trained, E networks is passed through G The generation image set P of network generation carrys out reversion choice G networks, obtains the random vector of the setting dimension of G network inputs, finally will The random vector of the setting dimension of G network inputs is as algorithm characteristics, extraction generation image set P face feature vector.
CNN (Convolutional Neural Network, convolutional neural networks), it is hand-written in processing by Yann LeCun A kind of convolution algorithm for stacking that character recognition proposes.DCNN refers to the technology expanded in CNN technologies, is that a kind of network depth is big In the convolutional neural networks equal to 3.
The non-supervisory feature learning method of face based on production confrontation network that the present embodiment provides, compared to existing skill Art, unsupervised learning is done using GAN technologies, network is generated to learn a reverse target using depth Recurrent networks, from And achieve effect more more preferable than general unsupervised learning.Face provided by the invention based on production confrontation network is non-supervisory Feature learning method, results of learning are good, and accuracy of identification is high.
Preferably, as shown in figure 5, the non-supervisory feature learning of face based on production confrontation network that the present embodiment provides Method, step S100 include：
Step S110, Face datection is carried out to the original facial image collected, to detect that the eyes of facial image are sat Mark.
Original facial image is collected, Face datection is carried out to the original facial image of 2,000,000 collected, to detect Eyes and eyes coordinates in facial image.
Step S120, using eyes coordinates the face in original facial image is alignd and normalized, to turn Change the face training image being sized into.
Horizontal alignment is carried out to the face in original facial image using the eyes coordinates detected, at face normalization Manage and be sized size, the face training image that original facial image is converted into being sized, so as to beneficial to face Identification.In the present embodiment, in order to meet in recognition of face the needs of to resolution ratio, DCGAN primitive networks is improved, increased Add one layer of DCONV layer, export 128x128 RGB image.
The non-supervisory feature learning method of face based on production confrontation network that the present embodiment provides, by having collected Original facial image carry out Face datection, to detect the eyes coordinates of facial image；Using eyes coordinates to original face Face in image is alignd and normalized, to be converted into the face training image being sized.The present embodiment provides Based on production confrontation network the non-supervisory feature learning method of face, the people that original facial image is converted into being sized Face training image, so as to the identification beneficial to face, face accuracy of identification is improved, and obtain more preferably results of learning.
Preferably, as shown in fig. 6, the non-supervisory feature learning of face based on production confrontation network that the present embodiment provides Method, step S300 include：
Step S310, the network structure that confrontation network is generated to former depth convolution is improved, and is built and new is used to generate The target generation network of human face target image and the target-recognition network for being differentiated to the human face target image of generation.
Former DCGAN network structure is improved, builds the new G network (targets for being used to generate human face target image Generate network) and the D networks for being differentiated to the human face target image of generation (target-recognition network).
Step S320, structure target generation network in add depth convolutional layer, make input target generation network with Machine vector is converted to the human face target image being sized and exported.
As shown in figure 3, for the new G networks of structure, depth convolutional layer is added on the basis of former G networks, in conv3 One layer of conv4 of upper addition, finally output is the human face target image being sized, to meet that recognition of face is wanted to resolution ratio Ask.In the present embodiment, the human face target image of output is 128x128x3 RGB image.
The non-supervisory feature learning method of face based on production confrontation network that the present embodiment provides, to former depth convolution The network structure of generation confrontation network is improved, and builds new the target generation network and use that are used to generate human face target image In the target-recognition network that the human face target image to generation is differentiated；Depth volume is added in the target generation network of structure Lamination, the random vector of input target generation network is converted to the human face target image being sized and exported.This implementation The non-supervisory feature learning method of face based on production confrontation network that example provides, added in the target generation network of structure Depth convolutional layer, to meet requirement of the recognition of face to resolution ratio, so that depth Recurrent networks learn reverse target generation Results of learning are good during network, and accuracy of identification is high.
Preferably, see Fig. 6, the non-supervisory feature learning side of face based on production confrontation network of the present embodiment offer Method, step S320 include afterwards：
Step S330, the human face target image of target generation network output is differentiated with target-recognition network, sentenced Do not go out human face target image and face true picture.
The human face target image that G networks generate is differentiated with D networks, it is true with face to determine human face target image Real image.The target of D networks is just to try to the human face target image and face true picture of generation to be distinguished from.Final mesh Mark is exactly to input random vector, generates the human face target image close to face true picture.In the present embodiment, D networks are preferred The GoogleNet for 128x128RGB images is inputted as two sorter networks.Certainly, the selection input of D networks, is not limited to The GoogleNet networks, it is only the achievable scheme of reference herein.
The non-supervisory feature learning method of face based on production confrontation network that the present embodiment provides, with target-recognition Network differentiates to the human face target image of target generation network output, is truly schemed with face with determining human face target image Picture.The non-supervisory feature learning method of face based on production confrontation network that the present embodiment provides, target generation network are autonomous Study, results of learning are good, and accuracy of identification is high.
Preferably, as shown in fig. 7, the non-supervisory feature learning of face based on production confrontation network that the present embodiment provides Method, step S400 include：
Step S410, image set will be generated as the inputs of depth convolutional neural networks, random vector collection as depth volume Accumulate the supervisory signals of neutral net and depth Recurrent networks are trained using Euler's distance function as excitation function.
When training using obtained generation image set P as the inputs of E networks, random vector collection Z as E networks prison Signal and Euler's distance function is superintended and directed to be trained E networks as excitation function.
Step S420, by the face feature vector of the depth Recurrent networks extraction generation image set trained, with identification Face characteristic in facial image to be identified.
The generation image set P that the E networks trained are generated using G networks obtains G network inputs come reversion choice G networks 100 dimension float random vectors, finally regard the 100 of G network inputs dimension float random vectors as algorithm characteristics, extraction generates Image set P face feature vector, to identify the face characteristic in facial image to be identified.
The non-supervisory feature learning method of face based on production confrontation network that the present embodiment provides, will generate image set Input, random vector collection as depth convolutional neural networks are as the supervisory signals of depth convolutional neural networks and by Europe Distance function is drawn to be trained as excitation function to depth Recurrent networks；Generation is extracted by the depth Recurrent networks trained The face feature vector of image set, to identify the face characteristic in facial image to be identified.The present embodiment provide based on generation Formula resists the non-supervisory feature learning method of face of network, and the depth Recurrent networks trained utilize target generation network generation Generation image set comes reversion choice target generation network, the face feature vector of extraction generation image set, to identify people to be identified Face characteristic in face image, results of learning are good, accuracy of identification is high.
Preferably, as shown in figure 8, the present embodiment also provides a kind of non-supervisory feature of face based on production confrontation network Learning device, including：Pretreatment module 10, for being pre-processed to the original facial image collected, to be converted into setting The face training image of size；First training module 20, for the face training image that will change as training data to structure Target generation network in the depth convolution generation confrontation network built is trained；Acquisition module 30, for by the random of generation Vector set is input in the target generation network trained, obtains the generation image set corresponding with random vector collection；Second instruction Practice module 40, in the depth Recurrent networks for the depth convolutional neural networks that obtained generation image set is input into structure, Depth Recurrent networks are trained, the face feature vector of extraction generation image set.
Pretreatment module 10 is pre-processed to the original facial image collected, and original facial image is converted into setting The face training image of size, so that changing good face training image meets the needs of recognition of face.In the present embodiment, will The original facial image collected is converted into the face training image being sized, to meet in recognition of face to the need of resolution ratio Ask.
First training module 20 inputs the face training image changed the DCGAN (Deep of structure Convolutional Generative Adverserial Networks, depth convolution generation confrontation network) in G networks In (target generation network), G networks are trained using face training image as training data.
Random vector collection Z, random vector collection Z are generated by setting quantity and the random pictures group of dimension using random generator Conjunction forms, and is stored in database.In the present embodiment, the random vector collection of generation is input to and trained by acquisition module 30 G networks in, obtain the generation image set P corresponding with random vector collection Z.Wherein, in random vector collection Z random pictures number Measure as 1,000,000, dimension is 100 dimensions.
Obtained generation image set is input to DCNN (the Deep Convolutional of structure by the second training module 40 Neural Network, depth convolutional neural networks) E networks (depth Recurrent networks) in, E networks are trained, make E nets The generation image set P that network is generated by G networks come reversion choice G networks, obtain G network inputs setting dimension it is random to Amount, finally using the random vector of the setting dimension of G network inputs as algorithm characteristics, extraction generation image set P face characteristic Vector.
The non-supervisory feature learning device of face based on production confrontation network that the present embodiment provides, compared to existing skill Art, unsupervised learning is done using GAN technologies, network is generated to learn a reverse target using depth Recurrent networks, from And achieve effect more more preferable than general unsupervised learning.Face provided by the invention based on production confrontation network is non-supervisory Feature learning device, results of learning are good, and accuracy of identification is high.
Preferably, as shown in figure 9, the present embodiment also provides a kind of non-supervisory feature of face based on production confrontation network Learning device, pretreatment module 10 include detection unit 11 and converting unit 12, detection unit 11, for original to what is collected Facial image carries out Face datection, to detect the eyes coordinates of facial image；Converting unit 12, for using eyes coordinates pair Face in original facial image is alignd and normalized, to be converted into the face training image being sized.
Detection unit 11 collects original facial image, and Face datection is carried out to the 2000000 original facial images collected, To detect the eyes and eyes coordinates in facial image.
Converting unit 12 carries out horizontal alignment using the eyes coordinates detected to the face in original facial image, by people Face normalized is to being sized size, the face training image that original facial image is converted into being sized, so as to Beneficial to the identification of face.In the present embodiment, in order to meet in recognition of face the needs of to resolution ratio, by DCGAN primitive networks It is improved, increases by one layer of DCONV layer, exports 128x128 RGB image.
The non-supervisory feature learning method of face based on production confrontation network that the present embodiment provides, by having collected Original facial image carry out Face datection, to detect the eyes coordinates of facial image；Using eyes coordinates to original face Face in image is alignd and normalized, to be converted into the face training image being sized.The present embodiment provides Based on production confrontation network the non-supervisory feature learning method of face, the people that original facial image is converted into being sized Face training image, so as to the identification beneficial to face, face accuracy of identification is improved, and obtain more preferably results of learning.
Preferably, as shown in Figure 10, the present embodiment also provides a kind of non-supervisory spy of face based on production confrontation network Learning device is levied, acquisition module 30 includes construction unit 31 and adding device 32, construction unit 31, for being given birth to former depth convolution Network structure into confrontation network is improved, and is built the new target generation network for being used to generate human face target image and is used for The target-recognition network differentiated to the human face target image of generation；Adding device 32, net is generated for the target in structure Depth convolutional layer is added in network, the random vector of input target generation network is converted to the human face target image being sized and enters Row output.
Construction unit 31 is improved to former DCGAN network structure, builds the new G for being used to generate human face target image Network (target generation network) and the D networks for being differentiated to the human face target image of generation (target-recognition network).
Adding device 32 adds depth convolutional layer, in conv3 for the new G networks of structure on the basis of former G networks One layer of conv4 of upper addition, finally output is the human face target image being sized, to meet that recognition of face is wanted to resolution ratio Ask.In the present embodiment, the human face target image of output is 128x128x3 RGB image.
The non-supervisory feature learning method of face based on production confrontation network that the present embodiment provides, to former depth convolution The network structure of generation confrontation network is improved, and builds new the target generation network and use that are used to generate human face target image In the target-recognition network that the human face target image to generation is differentiated；Depth volume is added in the target generation network of structure Lamination, the random vector of input target generation network is converted to the human face target image being sized and exported.This implementation The non-supervisory feature learning method of face based on production confrontation network that example provides, added in the target generation network of structure Depth convolutional layer, to meet requirement of the recognition of face to resolution ratio, so that depth Recurrent networks learn reverse target generation Results of learning are good during network, and accuracy of identification is high.
Preferably, as shown in Figure 10, the present embodiment also provides a kind of non-supervisory spy of face based on production confrontation network Learning device is levied, acquisition module 30 also includes judgement unit 33, judgement unit 33, for being given birth to target-recognition network to target Human face target image into network output is differentiated, determines human face target image and face true picture.
Judgement unit 33 differentiates with D networks to the human face target image that G networks generate, and determines human face target figure Picture and face true picture.The target of D networks is just to try to the human face target image and face true picture of generation to open respectively Come.Final target is exactly to input random vector, generates the human face target image close to face true picture.In the present embodiment In, the preferred input of D networks is the GoogleNet of 128x128RGB images as two sorter networks.Certainly, the selection of D networks is defeated Enter, be not limited to the GoogleNet networks, be only the achievable scheme of reference herein.
The non-supervisory feature learning device of face based on production confrontation network that the present embodiment provides, with target-recognition Network differentiates to the human face target image of target generation network output, is truly schemed with face with determining human face target image Picture.The non-supervisory feature learning device of face based on production confrontation network that the present embodiment provides, allow target generation network certainly Primary learning, results of learning are good, and accuracy of identification is high.
Preferably, as shown in figure 11, the present embodiment also provides a kind of non-supervisory spy of face based on production confrontation network Learning device is levied, the second training module 40 includes training unit 41 and extraction unit 42, training unit 41, for that will generate image Collect the supervisory signals and general of the input as depth convolutional neural networks, random vector collection as depth convolutional neural networks Euler's distance function is trained as excitation function to depth Recurrent networks；Extraction unit 42, for the depth by training The face feature vector of Recurrent networks extraction generation image set is spent, to identify the face characteristic in facial image to be identified.
Using inputs of the obtained generation image set P as E networks, random vector collection Z works when training unit 41 is trained Supervisory signals and Euler's distance function for E networks are trained as excitation function to E networks.
The generation image set P that extraction unit 42 is generated using G networks obtains G network inputs come reversion choice G networks 100 dimension float random vectors, finally using 100 dimension float random vectors of G network inputs as algorithm characteristics, extraction generation figure Image set P face feature vector, to identify the face characteristic in facial image to be identified.
The non-supervisory feature learning device of face based on production confrontation network that the present embodiment provides, will generate image set Input, random vector collection as depth convolutional neural networks are as the supervisory signals of depth convolutional neural networks and by Europe Distance function is drawn to be trained as excitation function to depth Recurrent networks；Generation is extracted by the depth Recurrent networks trained The face feature vector of image set, to identify the face characteristic in facial image to be identified.The present embodiment provide based on generation Formula resists the non-supervisory feature learning device of face of network, and the depth Recurrent networks trained utilize target generation network generation Generation image set comes reversion choice target generation network, the face feature vector of extraction generation image set, to identify people to be identified Face characteristic in face image, results of learning are good, accuracy of identification is high.
The preferred embodiments of the present invention are the foregoing is only, are not intended to limit the invention, for the skill of this area For art personnel, the present invention can have various modifications and variations.Within the spirit and principles of the invention, that is made any repaiies Change, equivalent substitution, improvement etc., should be included in the scope of the protection.

Claims (10)

  1. A kind of 1. non-supervisory feature learning method of face based on production confrontation network, it is characterised in that including step：
    The original facial image collected is pre-processed, to be converted into the face training image being sized；
    The mesh in network is resisted using the face training image changed as depth convolution generation of the training data to structure Mark generation network is trained；
    The random vector collection of generation is input in the target generation network trained, obtained and the random vector collection phase Corresponding generation image set；
    In the depth Recurrent networks for the depth convolutional neural networks that the obtained generation image set is input into structure, to described Depth Recurrent networks are trained, and extract the face feature vector of the generation image set.
  2. 2. the face non-supervisory feature learning method according to claim 1 based on production confrontation network, its feature exist In the original facial image to having collected pre-processes, the step of to be converted into the face training image being sized Including：
    Face datection is carried out to the original facial image collected, to detect the eyes coordinates of the facial image；
    The face in the original facial image is alignd using the eyes coordinates and normalized, to be converted into setting The face training image being sized.
  3. 3. the face non-supervisory feature learning method according to claim 1 based on production confrontation network, its feature exist In described to be resisted the face training image changed as depth convolution generation of the training data to structure in network The step of target generation network is trained includes：
    The network structure that confrontation network is generated to former depth convolution is improved, and is built and new is used to generate human face target image Target generates network and the target-recognition network for being differentiated to the human face target image of generation；
    Depth convolutional layer is added in the target generation network of structure, makes the random vector of the input target generation network The human face target image being sized is converted to be exported.
  4. 4. the face non-supervisory feature learning method according to claim 3 based on production confrontation network, its feature exist In, it is described to add depth convolutional layer in the target generation network of structure, make the random of the input target generation network Vector includes after being converted to the step of human face target image being sized is exported：
    The human face target image of target generation network output is differentiated with target-recognition network, determines institute State human face target image and face true picture.
  5. 5. the face non-supervisory feature learning side according to any one of claim 1 to 4 based on production confrontation network Method, it is characterised in that
    It is right in the depth Recurrent networks of the depth convolutional neural networks that the obtained generation image set is input into structure The depth Recurrent networks are trained, extract it is described generation image set face feature vector the step of include：
    Using the generation image set as the input of the depth convolutional neural networks, the random vector collection is as the depth The supervisory signals of convolutional neural networks, the depth convolutional neural networks are instructed using Euler's distance function as excitation function Practice；
    The face feature vector of the generation image set is extracted by the depth convolutional neural networks trained, is treated with identification Identify the face characteristic in facial image.
  6. A kind of 6. non-supervisory feature learning device of face based on production confrontation network, it is characterised in that including：
    Pretreatment module (10), for being pre-processed to the original facial image collected, to be converted into the people being sized Face training image；
    First training module (20), for being rolled up using the face training image changed as training data to the depth of structure Target generation network in product generation confrontation network is trained；
    Acquisition module (30), for being input to the random vector collection of generation in the target trained generation network, obtain The generation image set corresponding with the random vector collection；
    Second training module (40), for the obtained generation image set to be input to the depth convolutional neural networks of structure In depth Recurrent networks, the depth Recurrent networks are trained, extract the face feature vector of the generation image set.
  7. 7. the face non-supervisory feature learning device according to claim 6 based on production confrontation network, its feature exist In,
    The pretreatment module (10) includes detection unit (11) and converting unit (12),
    Detection unit (11), for carrying out Face datection to the original facial image collected, to detect the facial image Eyes coordinates；
    Converting unit (12), for being alignd using the eyes coordinates to the face in the original facial image and normalizing Change is handled, to be converted into the face training image being sized.
  8. 8. the face non-supervisory feature learning device according to claim 6 based on production confrontation network, its feature exist In,
    The acquisition module (30) includes construction unit (31) and adding device (32),
    Construction unit (31), the network structure for generating confrontation network to former depth convolution are improved, and build new be used for The target generation network of generation human face target image and the target for being differentiated to the human face target image of generation are sentenced Other network；
    Adding device (32), for adding depth convolutional layer in generating network in the target of structure, make the input target The random vector of generation network is converted to the human face target image being sized and exported.
  9. 9. the face non-supervisory feature learning device according to claim 6 based on production confrontation network, its feature exist In,
    The acquisition module (30) also includes judgement unit (33),
    Judgement unit (33), for the human face target image with target-recognition network to target generation network output Differentiated, determine the human face target image and face true picture.
  10. 10. the non-supervisory feature learning dress of the face based on production confrontation network according to any one of claim 6 to 9 Put,
    Characterized in that,
    Second training module (40) includes training unit (41) and extraction unit (42),
    Training unit (41), it is described random for the input using the generation image set as the depth convolutional neural networks Supervisory signals of the vector set as the depth convolutional neural networks, using Euler's distance function as excitation function to the depth Convolutional neural networks are trained；
    Extraction unit (42), the face of the generation image set is extracted for the depth convolutional neural networks by training Characteristic vector, to identify the face characteristic in facial image to be identified.
CN201710581981.7A 2017-07-17 2017-07-17 Face unsupervised feature learning method and device based on generative confrontation network Active CN107423701B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN201710581981.7A CN107423701B (en) 2017-07-17 2017-07-17 Face unsupervised feature learning method and device based on generative confrontation network

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN201710581981.7A CN107423701B (en) 2017-07-17 2017-07-17 Face unsupervised feature learning method and device based on generative confrontation network

Publications (2)

Publication Number Publication Date
CN107423701A true CN107423701A (en) 2017-12-01
CN107423701B CN107423701B (en) 2020-09-01

Family

ID=60429985

Family Applications (1)

Application Number Title Priority Date Filing Date
CN201710581981.7A Active CN107423701B (en) 2017-07-17 2017-07-17 Face unsupervised feature learning method and device based on generative confrontation network

Country Status (1)

Country Link
CN (1) CN107423701B (en)

Cited By (39)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN108090905A (en) * 2018-01-05 2018-05-29 清华大学 The determination methods and system of producing line exception
CN108090465A (en) * 2017-12-29 2018-05-29 国信优易数据有限公司 A kind of dressing effect process model training method and dressing effect processing method
CN108122209A (en) * 2017-12-14 2018-06-05 浙江捷尚视觉科技股份有限公司 A kind of car plate deblurring method based on confrontation generation network
CN108121975A (en) * 2018-01-04 2018-06-05 中科汇通投资控股有限公司 A kind of face identification method combined initial data and generate data
CN108133201A (en) * 2018-01-17 2018-06-08 百度在线网络技术（北京）有限公司 Face character recognition methods and device
CN108171266A (en) * 2017-12-25 2018-06-15 中国矿业大学 A kind of learning method of multiple target depth convolution production confrontation network model
CN108229349A (en) * 2017-12-21 2018-06-29 中国科学院自动化研究所 Reticulate pattern facial image identification device
CN108305239A (en) * 2018-03-06 2018-07-20 陕西师范大学 A kind of restorative procedure for the Bridge Crack image fighting network based on production
CN108333183A (en) * 2018-01-31 2018-07-27 西安工程大学 A kind of color based on DCGAN and DCNN knits shirt cut-parts defect inspection method
CN108334847A (en) * 2018-02-06 2018-07-27 哈尔滨工业大学 A kind of face identification method based on deep learning under real scene
CN108416324A (en) * 2018-03-27 2018-08-17 百度在线网络技术（北京）有限公司 Method and apparatus for detecting live body
CN108460760A (en) * 2018-03-06 2018-08-28 陕西师范大学 A kind of Bridge Crack image discriminating restorative procedure fighting network based on production
CN108537152A (en) * 2018-03-27 2018-09-14 百度在线网络技术（北京）有限公司 Method and apparatus for detecting live body
CN108710896A (en) * 2018-04-24 2018-10-26 浙江工业大学 The field learning method of learning network is fought based on production
CN108921123A (en) * 2018-07-17 2018-11-30 重庆科技学院 A kind of face identification method based on double data enhancing
CN109063776A (en) * 2018-08-07 2018-12-21 北京旷视科技有限公司 Image identifies network training method, device and image recognition methods and device again again
CN109145938A (en) * 2018-06-26 2019-01-04 阿里巴巴集团控股有限公司 A kind of method and apparatus classified using classifier
CN109522857A (en) * 2018-11-26 2019-03-26 山东大学 A kind of Population size estimation method based on production confrontation network model
CN109871898A (en) * 2019-02-27 2019-06-11 南京中设航空科技发展有限公司 A method of deposit training sample is generated using confrontation network is generated
CN109993177A (en) * 2017-12-29 2019-07-09 华为技术有限公司 A kind of image processing method and device
CN110097086A (en) * 2019-04-03 2019-08-06 平安科技（深圳）有限公司 Image generates model training method, image generating method, device, equipment and storage medium
CN110135574A (en) * 2018-02-09 2019-08-16 北京世纪好未来教育科技有限公司 Neural network training method, image generation method and computer storage medium
CN110378979A (en) * 2019-07-04 2019-10-25 公安部第三研究所 The method automatically generated based on the generation confrontation customized high-resolution human face picture of network implementations
CN110516544A (en) * 2019-07-19 2019-11-29 平安科技（深圳）有限公司 Face identification method, device and computer readable storage medium based on deep learning
CN110598595A (en) * 2019-08-29 2019-12-20 合肥工业大学 Multi-attribute face generation algorithm based on face key points and postures
CN111370020A (en) * 2020-02-04 2020-07-03 清华珠三角研究院 Method, system, device and storage medium for converting voice into lip shape
CN111436929A (en) * 2019-01-17 2020-07-24 复旦大学 A method for generating and identifying neurophysiological signals
CN111652827A (en) * 2020-04-24 2020-09-11 山东大学 A method and system for frontal face synthesis based on generative adversarial network
CN111724310A (en) * 2019-03-21 2020-09-29 马上消费金融股份有限公司 Training method of image restoration model, image restoration method and device
CN111832404A (en) * 2020-06-04 2020-10-27 中国科学院空天信息创新研究院 A small sample remote sensing feature classification method and system based on feature generation network
CN112308113A (en) * 2020-09-23 2021-02-02 济南浪潮高新科技投资发展有限公司 Target identification method, device and medium based on semi-supervision
CN112330526A (en) * 2019-08-05 2021-02-05 深圳Tcl新技术有限公司 Training method of face conversion model, storage medium and terminal equipment
CN113344792A (en) * 2021-08-02 2021-09-03 浙江大华技术股份有限公司 Image generation method and device and electronic equipment
CN113378660A (en) * 2021-05-25 2021-09-10 广州紫为云科技有限公司 Low-data-cost face recognition method and device
US11250329B2 (en) 2017-10-26 2022-02-15 Nvidia Corporation Progressive modification of generative adversarial neural networks
CN114078082A (en) * 2020-08-10 2022-02-22 北京达佳互联信息技术有限公司 A training and image generation method and device of a character image gender conversion model
US11263525B2 (en) 2017-10-26 2022-03-01 Nvidia Corporation Progressive modification of neural networks
US11475608B2 (en) 2019-09-26 2022-10-18 Apple Inc. Face image generation with pose and expression control
US12136139B2 (en) * 2022-03-31 2024-11-05 Konica Minolta Business Solutions U.S.A., Inc. Method and system for automated generation of representative icons from images

Citations (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US8582807B2 (en) * 2010-03-15 2013-11-12 Nec Laboratories America, Inc. Systems and methods for determining personal characteristics
US20150347820A1 (en) * 2014-05-27 2015-12-03 Beijing Kuangshi Technology Co., Ltd. Learning Deep Face Representation
CN106845471A (en) * 2017-02-20 2017-06-13 深圳市唯特视科技有限公司 A kind of vision significance Forecasting Methodology based on generation confrontation network
CN106874883A (en) * 2017-02-27 2017-06-20 中国石油大学(华东) A kind of real-time face detection method and system based on deep learning
CN106909905A (en) * 2017-03-02 2017-06-30 中科视拓（北京）科技有限公司 A kind of multi-modal face identification method based on deep learning
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

Patent Citations (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US8582807B2 (en) * 2010-03-15 2013-11-12 Nec Laboratories America, Inc. Systems and methods for determining personal characteristics
US20150347820A1 (en) * 2014-05-27 2015-12-03 Beijing Kuangshi Technology Co., Ltd. Learning Deep Face Representation
CN106845471A (en) * 2017-02-20 2017-06-13 深圳市唯特视科技有限公司 A kind of vision significance Forecasting Methodology based on generation confrontation network
CN106874883A (en) * 2017-02-27 2017-06-20 中国石油大学(华东) A kind of real-time face detection method and system based on deep learning
CN106909905A (en) * 2017-03-02 2017-06-30 中科视拓（北京）科技有限公司 A kind of multi-modal face identification method based on deep learning
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

Non-Patent Citations (3)

* Cited by examiner, † Cited by third party
Title
ALEC RADFORD ET AL: "Unsupervised representation learning with deep convolutional generative adversarial networks", 《ARXIV PREPRINT ARXIV:1511.06343》 *
C.DONG ET AL: "Learning a deep convolutional network for image super-resolution", 《COMPUTER VISION-ECCV 2014》 *
王坤峰等: "生成式对抗网络GAN的研究进展与展望", 《自动化学报》 *

Cited By (63)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US11763168B2 (en) 2017-10-26 2023-09-19 Nvidia Corporation Progressive modification of generative adversarial neural networks
US11250329B2 (en) 2017-10-26 2022-02-15 Nvidia Corporation Progressive modification of generative adversarial neural networks
US11263525B2 (en) 2017-10-26 2022-03-01 Nvidia Corporation Progressive modification of neural networks
CN108122209B (en) * 2017-12-14 2020-05-15 浙江捷尚视觉科技股份有限公司 License plate deblurring method based on countermeasure generation network
CN108122209A (en) * 2017-12-14 2018-06-05 浙江捷尚视觉科技股份有限公司 A kind of car plate deblurring method based on confrontation generation network
CN108229349B (en) * 2017-12-21 2020-09-01 中国科学院自动化研究所 Reticulated face image recognition device
CN108229349A (en) * 2017-12-21 2018-06-29 中国科学院自动化研究所 Reticulate pattern facial image identification device
CN108171266A (en) * 2017-12-25 2018-06-15 中国矿业大学 A kind of learning method of multiple target depth convolution production confrontation network model
CN109993177A (en) * 2017-12-29 2019-07-09 华为技术有限公司 A kind of image processing method and device
CN108090465B (en) * 2017-12-29 2020-05-01 国信优易数据有限公司 Dressing effect treatment model training method and dressing effect treatment method
CN108090465A (en) * 2017-12-29 2018-05-29 国信优易数据有限公司 A kind of dressing effect process model training method and dressing effect processing method
CN108121975B (en) * 2018-01-04 2022-04-19 中科汇通投资控股有限公司 Face recognition method combining original data and generated data
CN108121975A (en) * 2018-01-04 2018-06-05 中科汇通投资控股有限公司 A kind of face identification method combined initial data and generate data
CN108090905A (en) * 2018-01-05 2018-05-29 清华大学 The determination methods and system of producing line exception
CN108133201A (en) * 2018-01-17 2018-06-08 百度在线网络技术（北京）有限公司 Face character recognition methods and device
CN108333183A (en) * 2018-01-31 2018-07-27 西安工程大学 A kind of color based on DCGAN and DCNN knits shirt cut-parts defect inspection method
CN108334847A (en) * 2018-02-06 2018-07-27 哈尔滨工业大学 A kind of face identification method based on deep learning under real scene
CN108334847B (en) * 2018-02-06 2019-10-22 哈尔滨工业大学 A Face Recognition Method Based on Deep Learning in Real Scenes
CN110135574A (en) * 2018-02-09 2019-08-16 北京世纪好未来教育科技有限公司 Neural network training method, image generation method and computer storage medium
CN108460760A (en) * 2018-03-06 2018-08-28 陕西师范大学 A kind of Bridge Crack image discriminating restorative procedure fighting network based on production
CN108305239B (en) * 2018-03-06 2020-08-11 陕西师范大学 Bridge crack image repairing method based on generation type countermeasure network
CN108460760B (en) * 2018-03-06 2020-08-25 陕西师范大学 A method for discriminative repair of bridge crack images based on generative adversarial network
CN108305239A (en) * 2018-03-06 2018-07-20 陕西师范大学 A kind of restorative procedure for the Bridge Crack image fighting network based on production
CN108537152B (en) * 2018-03-27 2022-01-25 百度在线网络技术（北京）有限公司 Method and apparatus for detecting living body
CN108537152A (en) * 2018-03-27 2018-09-14 百度在线网络技术（北京）有限公司 Method and apparatus for detecting live body
CN108416324B (en) * 2018-03-27 2022-02-25 百度在线网络技术（北京）有限公司 Method and apparatus for detecting living body
CN108416324A (en) * 2018-03-27 2018-08-17 百度在线网络技术（北京）有限公司 Method and apparatus for detecting live body
CN108710896A (en) * 2018-04-24 2018-10-26 浙江工业大学 The field learning method of learning network is fought based on production
CN108710896B (en) * 2018-04-24 2021-10-29 浙江工业大学 Domain learning method based on generative adversarial learning network
CN109145938A (en) * 2018-06-26 2019-01-04 阿里巴巴集团控股有限公司 A kind of method and apparatus classified using classifier
CN108921123A (en) * 2018-07-17 2018-11-30 重庆科技学院 A kind of face identification method based on double data enhancing
CN109063776B (en) * 2018-08-07 2021-08-10 北京旷视科技有限公司 Image re-recognition network training method and device and image re-recognition method and device
CN109063776A (en) * 2018-08-07 2018-12-21 北京旷视科技有限公司 Image identifies network training method, device and image recognition methods and device again again
CN109522857A (en) * 2018-11-26 2019-03-26 山东大学 A kind of Population size estimation method based on production confrontation network model
CN109522857B (en) * 2018-11-26 2021-04-23 山东大学 A population estimation method based on generative adversarial network model
CN111436929A (en) * 2019-01-17 2020-07-24 复旦大学 A method for generating and identifying neurophysiological signals
CN111436929B (en) * 2019-01-17 2021-06-01 复旦大学 A method for generating and identifying neurophysiological signals
CN109871898A (en) * 2019-02-27 2019-06-11 南京中设航空科技发展有限公司 A method of deposit training sample is generated using confrontation network is generated
CN111724310B (en) * 2019-03-21 2023-08-15 马上消费金融股份有限公司 Training method of image restoration model, image restoration method and device
CN111724310A (en) * 2019-03-21 2020-09-29 马上消费金融股份有限公司 Training method of image restoration model, image restoration method and device
CN110097086A (en) * 2019-04-03 2019-08-06 平安科技（深圳）有限公司 Image generates model training method, image generating method, device, equipment and storage medium
CN110378979A (en) * 2019-07-04 2019-10-25 公安部第三研究所 The method automatically generated based on the generation confrontation customized high-resolution human face picture of network implementations
CN110378979B (en) * 2019-07-04 2022-12-23 公安部第三研究所 Method for realizing automatic generation of user-defined high-resolution face picture based on generation countermeasure network
CN110516544B (en) * 2019-07-19 2024-04-09 平安科技（深圳）有限公司 Face recognition method and device based on deep learning and computer readable storage medium
CN110516544A (en) * 2019-07-19 2019-11-29 平安科技（深圳）有限公司 Face identification method, device and computer readable storage medium based on deep learning
CN112330526B (en) * 2019-08-05 2024-02-09 深圳Tcl新技术有限公司 Training method of face conversion model, storage medium and terminal equipment
CN112330526A (en) * 2019-08-05 2021-02-05 深圳Tcl新技术有限公司 Training method of face conversion model, storage medium and terminal equipment
CN110598595B (en) * 2019-08-29 2022-03-18 合肥工业大学 A multi-attribute face generation algorithm based on face key points and poses
CN110598595A (en) * 2019-08-29 2019-12-20 合肥工业大学 Multi-attribute face generation algorithm based on face key points and postures
US11475608B2 (en) 2019-09-26 2022-10-18 Apple Inc. Face image generation with pose and expression control
CN111370020A (en) * 2020-02-04 2020-07-03 清华珠三角研究院 Method, system, device and storage medium for converting voice into lip shape
CN111370020B (en) * 2020-02-04 2023-02-14 清华珠三角研究院 Method, system, device and storage medium for converting voice into lip shape
CN111652827A (en) * 2020-04-24 2020-09-11 山东大学 A method and system for frontal face synthesis based on generative adversarial network
CN111652827B (en) * 2020-04-24 2023-04-18 山东大学 Front face synthesis method and system based on generation countermeasure network
CN111832404A (en) * 2020-06-04 2020-10-27 中国科学院空天信息创新研究院 A small sample remote sensing feature classification method and system based on feature generation network
CN114078082A (en) * 2020-08-10 2022-02-22 北京达佳互联信息技术有限公司 A training and image generation method and device of a character image gender conversion model
CN114078082B (en) * 2020-08-10 2024-12-03 北京达佳互联信息技术有限公司 A training and image generation method and device for a gender conversion model of a person image
CN112308113A (en) * 2020-09-23 2021-02-02 济南浪潮高新科技投资发展有限公司 Target identification method, device and medium based on semi-supervision
CN113378660A (en) * 2021-05-25 2021-09-10 广州紫为云科技有限公司 Low-data-cost face recognition method and device
CN113378660B (en) * 2021-05-25 2023-11-07 广州紫为云科技有限公司 Face recognition method and device with low data cost
WO2023010701A1 (en) * 2021-08-02 2023-02-09 Zhejiang Dahua Technology Co., Ltd. Image generation method, apparatus, and electronic device
CN113344792A (en) * 2021-08-02 2021-09-03 浙江大华技术股份有限公司 Image generation method and device and electronic equipment
US12136139B2 (en) * 2022-03-31 2024-11-05 Konica Minolta Business Solutions U.S.A., Inc. Method and system for automated generation of representative icons from images

Also Published As

Publication number Publication date
CN107423701B (en) 2020-09-01

Similar Documents

Publication Publication Date Title
CN107423701A (en) The non-supervisory feature learning method and device of face based on production confrontation network
CN105518709B (en) The method, system and computer program product of face for identification
CN115100574B (en) Action recognition method and system based on fusion graph convolutional network and Transformer network
Feng et al. Robust and efficient algorithms for separating latent overlapped fingerprints
CN109522966A (en) A kind of object detection method based on intensive connection convolutional neural networks
CN110516616A (en) A kind of double authentication face method for anti-counterfeit based on extensive RGB and near-infrared data set
CN106469302A (en) A kind of face skin quality detection method based on artificial neural network
CN106326874A (en) Method and device for recognizing iris in human eye images
CN109902715A (en) A Context Aggregation Network Based Infrared Small and Small Target Detection Method
CN110097029B (en) Identity authentication method based on high way network multi-view gait recognition
CN109190643A (en) Based on the recognition methods of convolutional neural networks Chinese medicine and electronic equipment
CN111259814B (en) Living body detection method and system
US20230053911A1 (en) Detecting an object in an image using multiband and multidirectional filtering
CN105303163B (en) A kind of method and detection device of target detection
CN106919921A (en) With reference to sub-space learning and the gait recognition method and system of tensor neutral net
CN106855944A (en) Pedestrian's Marker Identity method and device
CN113221655A (en) Face spoofing detection method based on feature space constraint
CN110222604A (en) Target identification method and device based on shared convolutional neural networks
Hebbale et al. Real time COVID-19 facemask detection using deep learning
Sakthimohan et al. Detection and Recognition of Face Using Deep Learning
CN109614866A (en) Method for detecting human face based on cascade deep convolutional neural networks
CN108921071A (en) Human face in-vivo detection method, device, storage medium and processor
CN106951844A (en) A method and system for classification of EEG signals based on deep learning machine
CN113033305A (en) Living body detection method, living body detection device, terminal equipment and storage medium
CN104008374B (en) Miner's detection method based on condition random field in a kind of mine image

Legal Events

Date Code Title Description
PB01 Publication
PB01 Publication
SE01 Entry into force of request for substantive examination
CB02 Change of applicant information
CB02 Change of applicant information

Address after: 100097 Beijing Haidian District Kunming Hunan Road 51 C block two floor 207.

Applicant after: BEIJING ATHENA EYES SCIENCE & TECHNOLOGY CO.,LTD.

Address before: 100193 4, 403, block A, 14 building, 10 East North Road, Haidian District, Beijing.

Applicant before: BEIJING ATHENA EYES SCIENCE & TECHNOLOGY CO.,LTD.

CB02 Change of applicant information
CB02 Change of applicant information

Address after: 410205 14 Changsha Zhongdian Software Park Phase I, 39 Jianshan Road, Changsha High-tech Development Zone, Yuelu District, Changsha City, Hunan Province

Applicant after: Wisdom Eye Technology Co.,Ltd.

Address before: 100097 2nd Floor 207, Block C, 51 Hunan Road, Kunming, Haidian District, Beijing

Applicant before: BEIJING ATHENA EYES SCIENCE & TECHNOLOGY CO.,LTD.

GR01 Patent grant
GR01 Patent grant
PE01 Entry into force of the registration of the contract for pledge of patent right
PE01 Entry into force of the registration of the contract for pledge of patent right

Denomination of invention: Unsupervised face feature learning method and device based on generative countermeasure network

Effective date of registration: 20221205

Granted publication date: 20200901

Pledgee: Agricultural Bank of China Limited Hunan Xiangjiang New Area Branch

Pledgor: Wisdom Eye Technology Co.,Ltd.

Registration number: Y2022430000107

PC01 Cancellation of the registration of the contract for pledge of patent right
PC01 Cancellation of the registration of the contract for pledge of patent right

Date of cancellation: 20231220

Granted publication date: 20200901

Pledgee: Agricultural Bank of China Limited Hunan Xiangjiang New Area Branch

Pledgor: Wisdom Eye Technology Co.,Ltd.

Registration number: Y2022430000107

CP03 Change of name, title or address
CP03 Change of name, title or address

Address after: No. 205, Building B1, Huigu Science and Technology Industrial Park, No. 336 Bachelor Road, Bachelor Street, Yuelu District, Changsha City, Hunan Province, 410000

Patentee after: Wisdom Eye Technology Co.,Ltd.

Address before: 410205 building 14, phase I, Changsha Zhongdian Software Park, No. 39, Jianshan Road, Changsha high tech Development Zone, Yuelu District, Changsha City, Hunan Province

Patentee before: Wisdom Eye Technology Co.,Ltd.