CN108932459A - Face recognition model training method and device and recognition algorithms - Google Patents
Face recognition model training method and device and recognition algorithms Download PDFInfo
- Publication number
- CN108932459A CN108932459A CN201710384934.3A CN201710384934A CN108932459A CN 108932459 A CN108932459 A CN 108932459A CN 201710384934 A CN201710384934 A CN 201710384934A CN 108932459 A CN108932459 A CN 108932459A
- Authority
- CN
- China
- Prior art keywords
- face
- ethnic
- neural network
- training
- data collection
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Granted
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V40/00—Recognition of biometric, human-related or animal-related patterns in image or video data
- G06V40/10—Human or animal bodies, e.g. vehicle occupants or pedestrians; Body parts, e.g. hands
- G06V40/16—Human faces, e.g. facial parts, sketches or expressions
-
- 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
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- General Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- Life Sciences & Earth Sciences (AREA)
- Evolutionary Biology (AREA)
- Evolutionary Computation (AREA)
- Bioinformatics & Computational Biology (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Artificial Intelligence (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Health & Medical Sciences (AREA)
- General Health & Medical Sciences (AREA)
- Oral & Maxillofacial Surgery (AREA)
- Human Computer Interaction (AREA)
- Multimedia (AREA)
- Image Analysis (AREA)
- Image Processing (AREA)
Abstract
The present invention relates to face recognition model training method and device and recognition algorithms.The training method includes：Neural network model is trained using the first ethnic face data collection；Transformation, which is carried out, based on the first ethnic face data collection and the second ethnic face threedimensional model obtains similar second ethnic face data collection；And neural network model is further trained using similar second ethnic face data collection, the weight of neural network is finely adjusted, uses trained obtained neural network model as face recognition model.According to the method for the present invention with device without devoting a tremendous amount of time, manpower and resource collect face data, and the training across ethnic face recognition model can be carried out.
Description
Technical field
The present invention relates to field of image processing, relate more specifically to a kind of face recognition model training method and device and face
Portion's recognition methods.
Background technique
In recent years, face recognition has obtained extensive concern in academia and industry and has achieved many progress, leads
It wants the reason is that large-scale face data is collected and the development of convolutional neural networks (CNN).Currently, the face recognition across race is difficult
Spend it is larger, it is good to specific ethnic recognition performance, and poor to other ethnic recognition performances.For example the training set of model is with some kind
Based on race's data, then this face identification system can preferably identify this race, it is not very to other race's recognition performances
It is good.Currently, there is the face data much increased income artificially to lead with west, and Asian, African face data are less.And convolutional Neural
Network training needs a large amount of face data, and a large amount of time, manpower and resource can be wasted by collecting a large amount of face data.
Therefore, it is necessary to it is a kind of be able to solve the above problem, across race face recognition model training method and device and
Recognition algorithms.
Summary of the invention
Brief summary of the present invention is given below, in order to provide the basic reason about certain aspects of the invention
Solution.It should be appreciated that this summary is not an exhaustive overview of the invention.It is not intended to determine key of the invention
Or pith, nor is it intended to limit the scope of the present invention.Its purpose only provides certain concepts in simplified form, with
This is as the preamble in greater detail discussed later.
A primary object of the present invention is, provides a kind of training side of face recognition model neural network based
Method, including：Neural network model is trained using the first ethnic face data collection；Based on the first ethnic face data collection and second
Ethnic face's threedimensional model carries out transformation and obtains similar second ethnic face data collection；And use similar second ethnic face number
Neural network model is further trained according to collection, the weight of neural network is finely adjusted, the neural network mould obtained with training
Type is as face recognition model.
According to an aspect of the present invention, a kind of recognition algorithms are provided, including：Data to be identified input is trained
To face recognition model in calculated；The feature being calculated is obtained similarity feature with training to be compared；And
The recognition result of data to be identified is determined based on comparative result.
According to a further aspect of the invention, a kind of training device of face recognition model neural network based is provided,
Including：First nerves network model training unit is configured with the first ethnic face data collection to train neural network mould
Type；Similar second ethnic face data collection obtaining unit, is configured as based on the first ethnic face data collection and the second ethnic face
Portion's threedimensional model is converted to obtain similar second ethnic face data collection；And nervus opticus network model training unit, quilt
Similar second ethnic face data collection is configured so that further to train neural network model, thus to the weight of neural network
It is finely adjusted, uses trained obtained neural network model as face recognition model.
In addition, the embodiments of the present invention also provide the computer programs for realizing the above method.
In addition, the embodiments of the present invention also provide the computer program product of at least computer-readable medium form,
Upper record has the computer program code for realizing the above method.
By the detailed description below in conjunction with attached drawing to highly preferred embodiment of the present invention, these and other of the invention is excellent
Point will be apparent from.
Detailed description of the invention
Below with reference to the accompanying drawings illustrate embodiments of the invention, the invention will be more easily understood it is above and its
Its objects, features and advantages.Component in attached drawing is intended merely to show the principle of the present invention.In the accompanying drawings, identical or similar
Technical characteristic or component will be indicated using same or similar appended drawing reference.
Fig. 1 shows the overall framework schematic diagram of trained face recognition model；
Fig. 2 shows the training methods of face recognition model neural network based according to an embodiment of the invention
The flow chart of 200 example process；
Fig. 3 is the flow chart for showing a kind of example process of the step S204 in Fig. 2；
Fig. 4 shows the schematic diagram that topography of face block is extracted from face image；
Fig. 5 shows training and obtains the schematic diagram of multiple convolutional neural networks；
Fig. 6 is to show the process for the example process that face recognition is carried out using the face recognition model that training obtains
Figure；
Fig. 7 is the training cartridge for showing face recognition model neural network based according to another embodiment of the invention
Set the block diagram of 700 exemplary configuration；
Fig. 8 is the frame for showing a kind of exemplary configuration of the similar second ethnic face data collection obtaining unit 704 in Fig. 7
Figure；And
Fig. 9 is to show the training method and dress that can be used for implementing face recognition model neural network based of the invention
The exemplary block diagram for the calculating equipment set.
Specific embodiment
Exemplary embodiment of the invention is described hereinafter in connection with attached drawing.For clarity and conciseness,
All features of actual implementation mode are not described in the description.It should be understood, however, that developing any this actual implementation
Much decisions specific to embodiment must be made during example, to realize the objectives of developer, for example, symbol
Restrictive condition those of related to system and business is closed, and these restrictive conditions may have with the difference of embodiment
Changed.In addition, it will also be appreciated that although development is likely to be extremely complex and time-consuming, to having benefited from the disclosure
For those skilled in the art of content, this development is only routine task.
Here, and also it should be noted is that, in order to avoid having obscured the present invention because of unnecessary details, in the accompanying drawings
Illustrate only with closely related device structure and/or processing step according to the solution of the present invention, and be omitted and the present invention
The little other details of relationship.
The invention proposes a kind of recognition algorithms across race.By using existing ethnic face data collection and
Some target races face threedimensional model obtains similar target race face data collection largely with target race face feature
As training data.And the choosing method for proposing a kind of new topography of face block, chooses the office with complementary feature
Then portion's image block carries out the training of neural network model using topography's block.During training, existing kind is first used
The multiple neural network models of race's face data training, then finely tune multiple neural network models using target race face data,
Measuring similarity can also be trained for being tested, finally obtain the model across ethnic face recognition.
The face recognition model neural network based of embodiment according to the present invention is described in detail with reference to the accompanying drawing
Training method and device.It is discussed below to carry out in the following order：
1. the training method of face recognition model neural network based
2. the training device of face recognition model neural network based
3. to the calculating equipment for implementing the present processes and device
[training methods of 1. face recognition models neural network based]
Fig. 1 shows training according to the method for the present invention and obtains the overall framework schematic diagram of face recognition model.In Fig. 1
It is illustrated by taking training convolutional neural networks as an example.
As shown in Figure 1, in the training process of neural network, first using the first multiple volumes of ethnic face data collection training
Product neural network model, the power for multiple convolutional neural networks that then training is obtained using similar second ethnic face data collection
It is finely adjusted again, finally trains measuring similarity with multiple convolutional neural networks models that training obtains, trained process is just
It is so that losing ever-reduced process.In identification process, use that the second ethnic face data input training obtains as face
Multiple convolutional neural networks of portion's identification model are compared with the similarity feature trained, and may finally obtain identification knot
Fruit.
Fig. 2 shows the training methods of face recognition model neural network based according to an embodiment of the invention
The flow chart of 200 example process.The training of face recognition model neural network based is illustrated below in conjunction with Fig. 2
The process of method 200.
Firstly, training neural network model using the first ethnic face data collection in step S202.
First ethnic face data collection for example can be existing WebFace database, be instructed using WebFace database
Practice neural network model.
Then, in step S204, become based on the first ethnic face data collection and the second ethnic face threedimensional model
Change the similar second ethnic face data collection of acquisition.
By applying transform method to second ethnic face's threedimensional model, the second ethnic face two dimensional image, then benefit are obtained
The data of the largely feature with second of clansman are generated with the first ethnic database.
Fig. 3 is the step S204 that shows in Fig. 2 (three-dimensional namely based on the first ethnic face data collection and the second ethnic face
Model is converted to obtain similar second ethnic face data collection) a kind of example process flow chart.
As shown in figure 3, extracting second ethnic face's two dimension based on second ethnic face's threedimensional model in step S2042
Image.
Then, in step S2044, the first ethnic face data is concentrated respectively first ethnic face's two dimensional image and
Extracted second ethnic face's two dimensional image carries out face's critical point detection.
Then, in step S2046, the face's key point that will test is aligned.For example, camera calibration can be passed through
Face's key point will test is aligned, thus by first ethnic face's two dimensional image and the second ethnic face X-Y scheme
As alignment.
Finally, generating similar second ethnic face data in step S2048.
After obtaining similar second ethnic face data collection in step S204, in step S206, use is generated
Neural network model is further trained similar to the second ethnic face data collection, the weight of neural network is finely adjusted, finally
The neural network model that training obtains is the face recognition model for being used for face recognition.
In the present invention, it proposes a kind of pair of face data collection and carries out pretreated method.Preferably, the first is being used
Race's face data collection will first pre-process the first ethnic face data collection before neural network model to train, and make
It before also will be to similar second ethnic face number come further training neural network model with similar second ethnic face data collection
It is pre-processed according to collection.
Specifically, pretreatment includes：Firstly, choosing M face's key point on face's two dimensional image；Then, based on M
Face's key point samples N number of topography's block from face's two dimensional image, wherein M and N is positive integer.
The existing method for choosing topography's block is usually a large amount of topography's block of stochastical sampling, and in the present invention
In, propose a kind of method that a small amount of topography's block is sampled based on multiple face's key points.
In one example, multiple face's key points may include two eyes, nose and two corners of the mouth this five keys
Point, i.e. M=5.These face's key points facilitate the complementary information provided include in topography's block.
In one example, the selection process of topography's block of face is as follows.
Firstly, calculating the height and width of topography of face block using following formula：
xmin={ x1,x2,x3,x4,x5}
xmax={ x1,x2,x3,x4,x5}
ymin={ y1,y2,y3,y4,y5}
ymax={ y1,y2,y3,y4,y5}
Wherein, (x1,y1)、(x2,y2)、(x3,y3)、(x4,y4)、(x4,y4) be respectively left eye, right eye, nose, the left corners of the mouth,
The coordinate of this five key points of the right corners of the mouth.xminIndicate x1To x5In minimum value, xmaxIndicate maximum value of the x1 into x5, ymin
There is similar meaning with ymax.HbaseAnd WbaseIt is the basic height and width of topography of face block respectively, and Hbase=Wbase。
Then, according to the basic height and width of topography of face block, the height of all topographies is determined according to predetermined ratio
And width.
For example, in one example, taking N=6, that is, sampling 6 topography's blocks, then can be calculated with following equation each
The height and width of a topography of face.
H1=0.6*Hbase
H2=0.5*Hbase
H3=0.6*Hbase
H4=Hbase
H5=Hbase
H6=Hbase
Wherein, H1、H2、H3、H4、H5And H6The height of respectively 6 topography of face blocks, topography of face block width and
It is high equal.The central point of this six topography of face blocks is respectively (x1,y1)、(x3,y3)、(0.5*(x4+x5),y4)、(0.5*
(x1+x2),y1)、(0.5*(x4+x5),y4)、(x3,y3).By central point, width and height, as shown in the lower part of Figure 4 six can be extracted
A topography of face block.
Preferably, when carrying out the training of neural network model, first with above-mentioned preprocess method respectively to the first race
The image that the image and similar second ethnic face data that face data is concentrated are concentrated is pre-processed, and then uses the respectively again
The complete face image and train to obtain by pre-processing obtained N number of topography of face block that one ethnic face data is concentrated
N+1 neural network model, and using the complete face image of similar second ethnic face data concentration and pass through pretreatment
Obtained N number of topography's block further trains neural network model, carries out to the weight of N+1 neural network model micro-
It adjusts.
Preferably, the neural network model that training obtains is convolutional neural networks model (MCNN).
Fig. 5 shows the schematic diagram using complete face image and local image block training convolutional neural networks.In Fig. 5
In, it is respectively trained to obtain multiple (N+1 using complete face image (shown in the uppermost arrow of Fig. 5) and local image block first
It is a) CNN (convolutional neural networks) model, different CNN is for learning the complementary feature of face.
After training obtains multiple convolutional neural networks models, for predetermined face image, instruction can be based further on
Multiple convolutional neural networks models for getting train to obtain similarity feature, are convenient for face recognition.
Specifically, it for predetermined face image, is primarily based on the N+1 convolutional neural networks model that training obtains and obtains respectively
To N+1 feature, these features can indicate the complementary characteristic of face.
Then, N+1 obtained feature is connected together as similarity feature, to be used for face recognition.
In the present invention, six topography of face blocks are obtained by pretreatment.It will be understood by those skilled in the art that
Topography's block of other quantity can be obtained by pretreatment, and be not limited to 6.
In one example, the first race is west race, and the second race is east race.Those skilled in the art can be with
Understand, the first race and the second race are without being limited thereto, are also possible to except other above races.
The invention proposes a kind of recognition algorithms across race.By using existing ethnic face data collection and
Some target races face threedimensional model obtains similar target race face data collection largely with target race face feature
As training data.And the choosing method for proposing a kind of new topography of face block, chooses the office with complementary feature
Then portion's image block carries out the training of neural network model using topography's block.During training, existing kind is first used
The multiple neural network models of race's face data training, then finely tune multiple neural network models using target race face data,
Measuring similarity can also be trained for being tested, finally obtained across ethnic face recognition model.
Fig. 6 is to show to obtain using the training of training method 200 of above-mentioned face recognition model neural network based
Face recognition model carries out the flow chart of the example process of face recognition.Face recognition side is illustrated below in conjunction with Fig. 6
The process of method 600.
Firstly, data to be identified to be input to the face obtained using the training of above-mentioned training method 200 in step S602
It is calculated in identification model.
Then, in step s 604, the feature being calculated similarity feature is obtained with training to be compared.
Finally, determining the recognition result of data to be identified based on comparative result in step S606.
2. language model training device neural network based
Fig. 7 is the training cartridge for showing face recognition model neural network based according to another embodiment of the invention
Set the block diagram of 700 exemplary configuration.
As shown in fig. 7, the training device 700 of face recognition model neural network based includes first nerves network model
Training unit 702, similar second ethnic face data collection obtaining unit 704 and nervus opticus network model training unit 706.
Wherein, first nerves network model training unit 702 is configured with the first ethnic face data collection to train
Neural network model.
Similar second ethnic face data collection obtaining unit 704 is configured as based on the first ethnic face data collection and second
Ethnic face's threedimensional model is converted to obtain similar second ethnic face data collection.
Nervus opticus network model training unit 706, which is configured with similar second ethnic face data collection, to be come further
Training neural network model, is finely adjusted the weight of neural network, and the neural network model for using training to obtain is known as face
Other model.
Fig. 8 is the frame for showing a kind of exemplary configuration of the similar second ethnic face data collection obtaining unit 704 in Fig. 7
Figure.
As shown in fig. 7, similar second ethnic face data collection obtaining unit 704 includes：Two dimensional image extracts subelement
7042, critical point detection subelement 7044, alignment subelement 7046 and data generate subelement 7048.
Two dimensional image extracts subelement 7042 and is configured as extracting the second ethnic face based on second ethnic face's threedimensional model
Portion's two dimensional image.
Critical point detection subelement 7044 is configured to the first ethnic face that the ethnic face data of detection first is concentrated
Face's key point of portion's two dimensional image and extracted second ethnic face two dimensional image.
Face's key point that alignment subelement 7046 is configured as will test is aligned.
Data generate subelement 7048 and are configurable to generate similar second ethnic face data.
In one example, the training device 700 of face recognition model neural network based further includes pretreatment unit
(not shown).Pretreatment unit is configured as：M face's key point is chosen on face's two dimensional image, M is positive integer；
And N number of topography's block is sampled from face's two dimensional image based on M face's key point, N is positive integer.
Pretreatment unit is used to train in first nerves network model training unit using the first ethnic face data collection
Before neural network model and nervus opticus network model training unit using similar second ethnic face data collection come into one
The first ethnic face data collection and similar second ethnic face data collection are carried out respectively before walking training neural network model pre-
Processing.
Wherein, M face's key point includes：Two eyes, nose and two corners of the mouths.
Wherein, first nerves network model training unit 702 is configured to using the first ethnic face data collection
In complete face image and by pre-process obtain N number of topography's block training obtain N+1 neural network model.
Wherein, nervus opticus network model training unit 706 is configured to using similar second ethnic face number
It is trained according to the complete face image of concentration and by pre-processing obtained N number of topography's block, to N+1 neural network mould
The weight of type is finely adjusted.
Wherein, the training device 700 of face recognition model neural network based further includes similarity feature training unit
(not shown).Similarity feature training unit is configured as：For predetermined face image, the nerve net obtained based on training
Network model trains to obtain similarity feature.
Wherein, similarity feature training unit is configured to：N+ is respectively obtained based on N+1 neural network model
1 feature；And N+1 feature is connected together as similarity feature.
Wherein, N=6.
Wherein, the first race is west race, and the second race is east race.
Wherein neural network is convolutional neural networks.
The operations and functions of the various pieces of training device 700 about face recognition model neural network based it is thin
Section is referred to the implementation of the training method for the face recognition model neural network based of the invention for combining Fig. 1-5 to describe
Example, is not detailed herein.
It should be noted that the training device 700 of face recognition model neural network based shown in Fig. 7-8 and
The structure of its component units is only exemplary, and those skilled in the art can according to need to structural frames shown in Fig. 6-7
Figure is modified.
The invention proposes a kind of extending methods across ethnic face data.In the feelings of the face data without certain race
Under condition, according to the method for the present invention without devoting a tremendous amount of time, manpower and resource collect face data, and can carry out across
Ethnic face recognition.
[the 3. calculating equipment to implement the present processes and device]
Basic principle of the invention is described in conjunction with specific embodiments above, however, it is desirable to, it is noted that this field
For those of ordinary skill, it is to be understood that the whole or any steps or component of methods and apparatus of the present invention, Ke Yi
Any computing device (including processor, storage medium etc.) perhaps in the network of computing device with hardware, firmware, software or
Their combination is realized that this is that those of ordinary skill in the art use them in the case where having read explanation of the invention
Basic programming skill can be achieved with.
Therefore, the purpose of the present invention can also by run on any computing device a program or batch processing come
It realizes.The computing device can be well known fexible unit.Therefore, the purpose of the present invention can also include only by offer
The program product of the program code of the method or device is realized to realize.That is, such program product is also constituted
The present invention, and the storage medium for being stored with such program product also constitutes the present invention.Obviously, the storage medium can be
Any well known storage medium or any storage medium that developed in the future.
In the case where realizing the embodiment of the present invention by software and/or firmware, from storage medium or network to having
The computer of specialized hardware structure, such as the installation of general purpose computer shown in Fig. 9 900 constitute the program of the software, the computer
When being equipped with various programs, it is able to carry out various functions etc..
In Fig. 9, central processing unit (CPU) 901 is according to the program stored in read-only memory (ROM) 902 or from depositing
The program that storage part 908 is loaded into random access memory (RAM) 903 executes various processing.In RAM 903, also according to need
Data ^CPU 901, ROM 902 and the RAM 903 required when CPU 901 executes various processing etc. are stored via bus
904 links each other.Input/output interface 905 also link to bus 904.
Components described below link is to input/output interface 905：Importation 906 (including keyboard, mouse etc.), output section
Divide 907 (including display, such as cathode-ray tube (CRT), liquid crystal display (LCD) etc. and loudspeakers etc.), storage section
908 (including hard disks etc.), communications portion 909 (including network interface card such as LAN card, modem etc.).Communications portion 909
Communication process is executed via network such as internet.As needed, driver 910 can also link to input/output interface 905.
Detachable media 911 such as disk, CD, magneto-optic disk, semiconductor memory etc. is installed in driver 910 as needed
On, so that the computer program read out is mounted to as needed in storage section 908.
It is such as removable from network such as internet or storage medium in the case where series of processes above-mentioned by software realization
Unload the program that the installation of medium 911 constitutes software.
It will be understood by those of skill in the art that this storage medium be not limited to it is shown in Fig. 9 be wherein stored with program,
Separately distribute with equipment to provide a user the detachable media 911 of program.The example of detachable media 911 includes disk
(including floppy disk (registered trademark)), CD (including compact disc read-only memory (CD-ROM) and digital versatile disc (DVD)), magneto-optic disk
(including mini-disk (MD) (registered trademark)) and semiconductor memory.Alternatively, storage medium can be ROM 902, storage section
Hard disk for including in 908 etc., wherein computer program stored, and user is distributed to together with the equipment comprising them.
The present invention also proposes a kind of program product of instruction code for being stored with machine-readable.Instruction code is read by machine
When taking and executing, can be performed it is above-mentioned according to the method for the embodiment of the present invention.
Correspondingly, it is also wrapped for carrying the storage medium of the program product of the above-mentioned instruction code for being stored with machine-readable
It includes in disclosure of the invention.Storage medium includes but is not limited to floppy disk, CD, magneto-optic disk, storage card, memory stick etc..
It should be appreciated by those skilled in the art that being exemplary what this was enumerated, the present invention is not limited thereto.
In the present specification, the statements such as " first ", " second " and " n-th " be in order to by described feature in text
On distinguish, the present invention is explicitly described.Therefore, it should not serve to that there is any limited meaning.
As an example, each step of the above method and all modules and/or unit of above equipment can
To be embodied as software, firmware, hardware or combinations thereof, and as a part in relevant device.Each composition mould in above-mentioned apparatus
Block, unit when being configured by way of software, firmware, hardware or combinations thereof workable specific means or mode be ability
Known to field technique personnel, details are not described herein.
It as an example, can be from storage medium or network to having in the case where being realized by software or firmware
Computer (such as general purpose computer 900 shown in Fig. 9) installation of specialized hardware structure constitutes the program of the software, the computer
When being equipped with various programs, it is able to carry out various functions etc..
In the description above to the specific embodiment of the invention, for the feature a kind of embodiment description and/or shown
It can be used in one or more other embodiments in a manner of same or similar, with the feature in other embodiments
It is combined, or the feature in substitution other embodiments.
It should be emphasized that term "comprises/comprising" refers to the presence of feature, element, step or component when using herein, but simultaneously
It is not excluded for the presence or additional of other one or more features, element, step or component.
In addition, method of the invention be not limited to specifications described in time sequencing execute, can also according to it
His time sequencing, concurrently or independently execute.Therefore, the execution sequence of method described in this specification is not to this hair
Bright technical scope is construed as limiting.
The present invention and its advantage it should be appreciated that without departing from the essence of the invention being defined by the claims appended hereto
Various changes, substitution and transformation can be carried out in the case where mind and range.Moreover, the scope of the present invention is not limited only to specification institute
The specific embodiment of the process of description, equipment, means, method and steps.One of ordinary skilled in the art is from of the invention
Disclosure it will be readily understood that can be used according to the present invention execute the function essentially identical to corresponding embodiment in this or
Obtain the result essentially identical with it, existing and to be developed in the future process, equipment, means, method or step.Cause
This, the attached claims are intended in the range of them include such process, equipment, means, method or step.
Based on above explanation, it is known that open at least to disclose following technical scheme：
It is attached 1, a kind of training method of face recognition model neural network based, including：
Neural network model is trained using the first ethnic face data collection；
It carries out transformation based on the described first ethnic face data collection and the second ethnic face threedimensional model and obtains similar second
Ethnic face data collection；And
The neural network model is further trained using the similar second ethnic face data collection, to the nerve
The weight of network is finely adjusted, and uses trained obtained neural network model as face recognition model.
Note 2, the method according to note 1, based on the described first ethnic face data collection and the second ethnic face three
Dimension module is converted to obtain similar second ethnic face data collection：
The second ethnic face two dimensional image is extracted based on described second ethnic face's threedimensional model；
First ethnic face's two dimensional image and extracted second that the described first ethnic face data is concentrated is detected respectively
Face's key point of ethnic face's two dimensional image；
The face's key point that will test is aligned；And
Generate the similar second ethnic face data.
Note 3, the method according to note 1 are training neural network model using the first ethnic face data collection
The described first ethnic face data collection is pre-processed before, and use the similar second ethnic face data collection
The similar second ethnic face data collection is pre-processed before the neural network model further to train,
In, the pretreatment includes：
M face's key point is chosen on face's two dimensional image, M is positive integer；And
N number of topography's block is sampled from face's two dimensional image based on M face key point, N is positive integer.
Note 4, according to method described in note 3, wherein M face key point includes：Two eyes, nose and
Two corners of the mouths.
It is attached 5, according to method described in note 3, wherein
It includes using the described first ethnic face data that neural network model is trained using the first ethnic face data collection
The complete face image concentrated and the N number of topography's block training obtained by pretreatment obtain N+1 neural network model；With
And
Further training the neural network model using the similar second ethnic face data collection includes using institute
State the complete face image and instructed by pre-processing obtained N number of topography's block that similar second ethnic face data is concentrated
Practice, the weight of the N+1 neural network model is finely adjusted.
Note 6, the method according to note 5 further include：
For predetermined face image, train to obtain similarity feature based on the neural network model that training obtains.
Note 7, the method according to note 6, wherein for predetermined face image, the mind obtained based on training
Train to obtain similarity feature through network model include：
N+1 feature is respectively obtained based on the N+1 neural network model；And
The N+1 feature is connected together as the similarity feature.
It is attached 8, according to method described in note 3, wherein N=6.
Note 9, the method according to note 1, wherein first race is west race, and second race is
East race.
Note 10, the method according to note 1, wherein the neural network is convolutional neural networks.
11, a kind of recognition algorithms are attached, including：
Data to be identified are input to using in the face recognition model that method training obtains according to note 1-10
It is calculated；
The feature being calculated is obtained similarity feature with training to be compared；And
The recognition result of the data to be identified is determined based on comparative result.
12, a kind of training device of face recognition model neural network based are attached, including：
First nerves network model training unit is configured with the first ethnic face data collection to train neural network
Model；
Similar second ethnic face data collection obtaining unit is configured as based on the described first ethnic face data collection and the
Two ethnic face threedimensional models are converted to obtain similar second ethnic face data collection；And
Nervus opticus network model training unit is configured with the similar second ethnic face data collection and comes into one
The step training neural network model, is finely adjusted the weight of the neural network, the neural network model obtained with training
As face recognition model.
Note 13, the device according to note 12, wherein the similar second ethnic face data Ji Huo get Dan Yuanbao
It includes：
Two dimensional image extracts subelement, is configured as extracting the second ethnic face based on described second ethnic face's threedimensional model
Portion's two dimensional image；
Critical point detection subelement is configured to detect the first ethnic face that the described first ethnic face data is concentrated
Face's key point of portion's two dimensional image and extracted second ethnic face two dimensional image；
It is directed at subelement, the face's key point for being configured as will test is aligned；And
Data generate subelement, are configurable to generate the similar second ethnic face data.
Note 14, the device according to note 12, further include pretreatment unit, the pretreatment unit is configured as：
M face's key point is chosen on face's two dimensional image, M is positive integer；And
N number of topography's block is sampled from face's two dimensional image based on M face key point, N is positive integer,
Wherein, the pretreatment unit uses the first ethnic face data in the first nerves network model training unit
Collection is trained before neural network model and the nervus opticus network model training unit uses similar second race
Face data collection is further trained before the neural network model respectively to the described first ethnic face data collection and described
Similar second ethnic face data collection is pre-processed.
Note 15, the device according to note 14, wherein M face key point includes：Two eyes, noses
With two corners of the mouths.
Note 16, the device according to note 14, wherein
The first nerves network model training unit is configured to using the described first ethnic face data collection
In complete face image and by pre-process obtain N number of topography's block training obtain N+1 neural network model；And
Nervus opticus network model training unit is configured to using the similar second ethnic face data collection
In complete face image and be trained by pre-processing obtained N number of topography's block, to the N+1 neural network mould
The weight of type is finely adjusted.
Note 17, the device according to note 16, further include similarity feature training unit, the similarity feature instruction
Practice unit to be configured as：For predetermined face image, the neural network model obtained based on training is similar to train to obtain
Spend feature.
It is attached 18, according to device as stated in Note 17, wherein the similarity feature training unit is further configured
For：
N+1 feature is respectively obtained based on the N+1 neural network model；And
The N+1 feature is connected together as the similarity feature.
Note 19, the device according to note 14, wherein N=6.
Note 20, the device according to note 12, wherein first race is west race, second race
For east race.
Claims (10)
1. a kind of training method of face recognition model neural network based, including：
Neural network model is trained using the first ethnic face data collection；
Transformation, which is carried out, based on the described first ethnic face data collection and the second ethnic face threedimensional model obtains similar second race
Face data collection；And
The neural network model is further trained using the similar second ethnic face data collection, to the neural network
Weight be finely adjusted, use the obtained neural network model of training as face recognition model.
2. according to the method described in claim 1, wherein, being based on the described first ethnic face data collection and the second ethnic face three
Dimension module is converted to obtain similar second ethnic face data collection：
The second ethnic face two dimensional image is extracted based on described second ethnic face's threedimensional model；
The first ethnic face's two dimensional image and extracted second race that the described first ethnic face data is concentrated are detected respectively
Face's key point of face's two dimensional image；
The face's key point that will test is aligned；And
Generate the similar second ethnic face data.
3. according to the method described in claim 1, before training neural network model using the first ethnic face data collection
Described first ethnic face data collection is pre-processed, and using the similar second ethnic face data collection come into
The similar second ethnic face data collection is pre-processed before the one step training neural network model, wherein institute
Stating pretreatment includes：
M face's key point is chosen on face's two dimensional image, M is positive integer；And
N number of topography's block is sampled from face's two dimensional image based on M face key point, N is positive integer.
4. according to the method described in claim 3, wherein, M face key point includes：Two eyes, nose and two
The corners of the mouth.
5. according to the method described in claim 3, wherein,
It includes being concentrated using the described first ethnic face data that neural network model is trained using the first ethnic face data collection
Complete face image and by pre-process obtain N number of topography's block training obtain N+1 neural network model；And
Further training the neural network model using the similar second ethnic face data collection includes using the class
It is trained like the complete face image of the second ethnic face data concentration and by pre-processing obtained N number of topography's block,
The weight of the N+1 neural network model is finely adjusted.
6. according to the method described in claim 5, further including：
For predetermined face image, train to obtain similarity feature based on the neural network model that training obtains.
7. according to the method described in claim 6, wherein, for predetermined face image, the nerve net obtained based on training
Network model includes to train to obtain similarity feature：
N+1 feature is respectively obtained based on the N+1 neural network model；And
The N+1 feature is connected together as the similarity feature.
8. second race is east according to the method described in claim 1, wherein, first race is west race
Race.
9. a kind of recognition algorithms, including：
By data to be identified be input in the face recognition model obtained using the training of method described in -8 according to claim 1 into
Row calculates；
The feature being calculated is obtained similarity feature with training to be compared；And
The recognition result of the data to be identified is determined based on comparative result.
10. a kind of training device of face recognition model neural network based, including：
First nerves network model training unit is configured with the first ethnic face data collection to train neural network mould
Type；
Similar second ethnic face data collection obtaining unit, is configured as based on the described first ethnic face data collection and second
Face of race threedimensional model is converted to obtain similar second ethnic face data collection；And
Nervus opticus network model training unit is configured with the similar second ethnic face data collection further to instruct
Practice the neural network model, to be finely adjusted to the weight of the neural network, the neural network model obtained with training
As face recognition model.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201710384934.3A CN108932459B (en) | 2017-05-26 | 2017-05-26 | Face recognition model training method and device and face recognition method |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201710384934.3A CN108932459B (en) | 2017-05-26 | 2017-05-26 | Face recognition model training method and device and face recognition method |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN108932459A true CN108932459A (en) | 2018-12-04 |
| CN108932459B CN108932459B (en) | 2021-12-10 |
Family
ID=64450378
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201710384934.3A Active CN108932459B (en) | 2017-05-26 | 2017-05-26 | Face recognition model training method and device and face recognition method |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN108932459B (en) |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN112825122A (en) * | 2019-11-20 | 2021-05-21 | 北京眼神智能科技有限公司 | Ethnicity judgment method, ethnicity judgment device, ethnicity judgment medium and ethnicity judgment equipment based on two-dimensional face image |
| CN112990121A (en) * | 2021-04-25 | 2021-06-18 | 中国人民解放军国防科技大学 | Target detection method and device, electronic equipment and storage medium |
Citations (16)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101499132A (en) * | 2009-03-12 | 2009-08-05 | 广东药学院 | Three-dimensional transformation search method for extracting characteristic points in human face image |
| CN101877146A (en) * | 2010-07-15 | 2010-11-03 | 北京工业大学 | Method for extending three-dimensional face database |
| KR101212110B1 (en) * | 2011-06-22 | 2012-12-13 | 한국과학기술원 | Method for 2d and 3d virtual face modeling |
| US20130307848A1 (en) * | 2012-05-17 | 2013-11-21 | Disney Enterprises, Inc. | Techniques for processing reconstructed three-dimensional image data |
| CN103440479A (en) * | 2013-08-29 | 2013-12-11 | 湖北微模式科技发展有限公司 | Method and system for detecting living body human face |
| CN104036247A (en) * | 2014-06-11 | 2014-09-10 | 杭州巨峰科技有限公司 | Facial feature based face racial classification method |
| CN104376594A (en) * | 2014-11-25 | 2015-02-25 | 福建天晴数码有限公司 | Three-dimensional face modeling method and device |
| CN104850825A (en) * | 2015-04-18 | 2015-08-19 | 中国计量学院 | Facial image face score calculating method based on convolutional neural network |
| CN105631398A (en) * | 2014-11-24 | 2016-06-01 | 三星电子株式会社 | Method and apparatus for recognizing object, and method and apparatus for training recognizer |
| CN106067190A (en) * | 2016-05-27 | 2016-11-02 | 俞怡斐 | A kind of fast face threedimensional model based on single image generates and alternative approach |
| US20160335480A1 (en) * | 2015-05-15 | 2016-11-17 | Toshiba Tec Kabushiki Kaisha | Video Based Facial Recognition for Customer Verification at Touchless Checkout |
| CN106203248A (en) * | 2014-09-05 | 2016-12-07 | 三星电子株式会社 | Method and apparatus for face recognition |
| CN106250877A (en) * | 2016-08-19 | 2016-12-21 | 深圳市赛为智能股份有限公司 | Near-infrared face identification method and device |
| CN106295584A (en) * | 2016-08-16 | 2017-01-04 | 深圳云天励飞技术有限公司 | Depth migration study is in the recognition methods of crowd's attribute |
| CN106485192A (en) * | 2015-09-02 | 2017-03-08 | 富士通株式会社 | Training method for the neutral net of image recognition and device |
| CN106503684A (en) * | 2016-10-28 | 2017-03-15 | 厦门中控生物识别信息技术有限公司 | A kind of face image processing process and device |
-
2017
- 2017-05-26 CN CN201710384934.3A patent/CN108932459B/en active Active
Patent Citations (16)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101499132A (en) * | 2009-03-12 | 2009-08-05 | 广东药学院 | Three-dimensional transformation search method for extracting characteristic points in human face image |
| CN101877146A (en) * | 2010-07-15 | 2010-11-03 | 北京工业大学 | Method for extending three-dimensional face database |
| KR101212110B1 (en) * | 2011-06-22 | 2012-12-13 | 한국과학기술원 | Method for 2d and 3d virtual face modeling |
| US20130307848A1 (en) * | 2012-05-17 | 2013-11-21 | Disney Enterprises, Inc. | Techniques for processing reconstructed three-dimensional image data |
| CN103440479A (en) * | 2013-08-29 | 2013-12-11 | 湖北微模式科技发展有限公司 | Method and system for detecting living body human face |
| CN104036247A (en) * | 2014-06-11 | 2014-09-10 | 杭州巨峰科技有限公司 | Facial feature based face racial classification method |
| CN106203248A (en) * | 2014-09-05 | 2016-12-07 | 三星电子株式会社 | Method and apparatus for face recognition |
| CN105631398A (en) * | 2014-11-24 | 2016-06-01 | 三星电子株式会社 | Method and apparatus for recognizing object, and method and apparatus for training recognizer |
| CN104376594A (en) * | 2014-11-25 | 2015-02-25 | 福建天晴数码有限公司 | Three-dimensional face modeling method and device |
| CN104850825A (en) * | 2015-04-18 | 2015-08-19 | 中国计量学院 | Facial image face score calculating method based on convolutional neural network |
| US20160335480A1 (en) * | 2015-05-15 | 2016-11-17 | Toshiba Tec Kabushiki Kaisha | Video Based Facial Recognition for Customer Verification at Touchless Checkout |
| CN106485192A (en) * | 2015-09-02 | 2017-03-08 | 富士通株式会社 | Training method for the neutral net of image recognition and device |
| CN106067190A (en) * | 2016-05-27 | 2016-11-02 | 俞怡斐 | A kind of fast face threedimensional model based on single image generates and alternative approach |
| CN106295584A (en) * | 2016-08-16 | 2017-01-04 | 深圳云天励飞技术有限公司 | Depth migration study is in the recognition methods of crowd's attribute |
| CN106250877A (en) * | 2016-08-19 | 2016-12-21 | 深圳市赛为智能股份有限公司 | Near-infrared face identification method and device |
| CN106503684A (en) * | 2016-10-28 | 2017-03-15 | 厦门中控生物识别信息技术有限公司 | A kind of face image processing process and device |
Non-Patent Citations (8)
| Title |
|---|
| AMIN JOURABLOO等: "Large-pose Face Alignment via CNN-based Dense 3D Model Fitting", 《PROCEEDINGS OF THE IEEE CONFERENCE ON COMPUTER VISION AND PATTERN RECOGNITION(CVPR)》 * |
| KOICHIRO NIINUMA等: "Automatic Multi-view Face Recognition via 3D Model Based Pose Regularization", 《2013 IEEE SIXTH INTERNATIONAL CONFERENCE ON BIOMETRICS: THEORY, APPLICATIONS AND SYSTEMS(BTAS)》 * |
| NEERU NARANG 等: "Gender and Ethnicity Classification using Deep Learning in Heterogeneous Face Recognition", 《2016 INTERNATIONAL CONFERENCE ON BIOMETRICS(ICB)》 * |
| 宋顶利等: "关键点匹配三维人脸识别方法", 《计算机应用研究》 * |
| 李辉等: "基于卷积神经网络的人脸识别算法", 《软件导刊》 * |
| 蒋玉等: "一种侧视图的三维人脸重建方法", 《宁波大学学报(理工版)》 * |
| 邢健飞: "基于三维人脸建模的多视角人脸识别方法研究", 《中国优秀硕士学位论文全文数据库 信息科技辑》 * |
| 邱盛: "基于深度学习的人脸民族识别", 《中国优秀硕士学位论文全文数据库 信息科技辑》 * |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN112825122A (en) * | 2019-11-20 | 2021-05-21 | 北京眼神智能科技有限公司 | Ethnicity judgment method, ethnicity judgment device, ethnicity judgment medium and ethnicity judgment equipment based on two-dimensional face image |
| CN112990121A (en) * | 2021-04-25 | 2021-06-18 | 中国人民解放军国防科技大学 | Target detection method and device, electronic equipment and storage medium |
Also Published As
| Publication number | Publication date |
|---|---|
| CN108932459B (en) | 2021-12-10 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN111582009B (en) | Device and method for training classification model and device for classifying by using classification model | |
| US10706332B2 (en) | Analog circuit fault mode classification method | |
| US11915104B2 (en) | Normalizing text attributes for machine learning models | |
| CN108229651A (en) | Neural network model moving method and system, electronic equipment, program and medium | |
| CN111582008B (en) | Device and method for training classification model and device for classifying by using classification model | |
| CN112006654B (en) | Bone age assessment method based on convolutional neural network and multiple attention mechanism | |
| US12002289B2 (en) | Expression recognition method and apparatus, computer device, and readable storage medium | |
| CN112949408B (en) | Real-time identification method and system for target fish passing through fish channel | |
| CN113448843B (en) | Image recognition software test data enhancement method and device based on defect analysis | |
| CN111144215A (en) | Image processing method, image processing device, electronic equipment and storage medium | |
| CN111666846A (en) | Face attribute identification method and device | |
| Xiao et al. | Deterministic model fitting by local-neighbor preservation and global-residual optimization | |
| US20220092359A1 (en) | Image data classification method, device and system | |
| CN107291775A (en) | The reparation language material generation method and device of error sample | |
| CN108932459A (en) | Face recognition model training method and device and recognition algorithms | |
| CN107633058A (en) | A kind of data dynamic filtration system and method based on deep learning | |
| CN112839185B (en) | Method, apparatus, device and medium for processing image | |
| CN108334602A (en) | Data mask method and device, electronic equipment, computer storage media, program | |
| CN111666748A (en) | Construction method of automatic classifier and method for recognizing decision from software development text product | |
| CN110428012A (en) | Brain method for establishing network model, brain image classification method, device and electronic equipment | |
| CN109740750A (en) | Method of data capture and device | |
| CN109472276A (en) | The construction method and device and mode identification method of pattern recognition model | |
| CN108764068A (en) | A kind of image-recognizing method and device | |
| CN107463528A (en) | The gauss hybrid models split-and-merge algorithm examined based on KS | |
| Kazangirler et al. | UIBee: An improved deep instance segmentation and classification of UI elements in wireframes |
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 |