CN109919093A - A kind of face identification method, device, equipment and readable storage medium storing program for executing - Google Patents
A kind of face identification method, device, equipment and readable storage medium storing program for executing Download PDFInfo
- Publication number
- CN109919093A CN109919093A CN201910173706.0A CN201910173706A CN109919093A CN 109919093 A CN109919093 A CN 109919093A CN 201910173706 A CN201910173706 A CN 201910173706A CN 109919093 A CN109919093 A CN 109919093A
- Authority
- CN
- China
- Prior art keywords
- face
- facial image
- characteristic
- sample
- personnel
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
Landscapes
- Image Analysis (AREA)
- Image Processing (AREA)
Abstract
The invention discloses a kind of face identification methods, method includes the following steps: obtaining the facial image of personnel to be identified, facial image is input in target network model, obtain the target complex class label and characteristic of face；The facial image characteristic with target complex class label and marked personnel identity information is read from database；Characteristic is compared with facial image characteristic, and determines the identity information of personnel to be identified according to comparison result.This method can reduce comparison number, shorten and compare duration, further promote recognition of face efficiency.The invention also discloses a kind of face identification device, equipment and readable storage medium storing program for executing, have corresponding technical effect.
Description
Technical field
The present invention relates to technical field of computer vision, more particularly to a kind of face identification method, device, equipment and can
Read storage medium.
Background technique
Video monitoring is quickly popularized, and there is an urgent need to a kind of remote, non-cooperations of user for numerous video surveillance applications
Quick identity recognizing technology under state realizes intelligent early-warning in the hope of remote quickly confirmation personnel identity.Face recognition technology
Undoubtedly optimal selection, using human face detection tech can from monitor video image real-time searching face, and with face number
It is compared in real time according to library, to realize quick identification.
It is controlling at present under matching requirements, recognition of face can obtain relatively high discrimination.But in monitoring scene
Under, it is influenced since the great varieties such as movement, illumination, posture and clarity are low etc., discrimination sharply declines！Personnel's bayonet simultaneously
What is stored in analysis system is large-scale face database, so the comparison in order to reach real-time identification, to face
Speed has tightened up requirement.
In conclusion the problems such as how effectively promoting the comparison speed in face recognition process, is current this field skill
Art personnel technical problem urgently to be solved.
Summary of the invention
The object of the present invention is to provide a kind of face identification method, device, equipment and readable storage medium storing program for executing, to improve face
Comparison speed in identification process, further speeds up recognition of face speed.
In order to solve the above technical problems, the invention provides the following technical scheme:
A kind of face identification method, comprising:
The facial image is input in target network model, described in acquisition by the facial image for obtaining personnel to be identified
The target complex class label and characteristic of face；
Reading from database has the facial image of the target complex class label and marked personnel identity information special
Levy data；
The characteristic is compared with the facial image characteristic, and according to comparison result determine it is described to
The identity information of identification personnel.
Preferably, before the facial image for obtaining personnel to be identified, further includes:
Obtain the facial image sample set of the marked personnel identity information；
It is special using the face of each facial image sample in facial image sample set described in the target network model extraction
Levy data and face realm distinguishing label；
The face characteristic data are stored in the database respectively according to the face realm distinguishing label.
Preferably, before the facial image for obtaining personnel to be identified, further includes:
Initial Face training set of images is obtained, and randomly chooses destination subset from the Initial Face training set；
Using in the Initial Face training set of images sample size and the destination subset in face characteristic data it is true
The quantity that centers and initial center；
Using IndexIVFKmeans method, model training is carried out in conjunction with the centric quantity and the initial center, is obtained
Group's label markup model of face realm distinguishing label must be used to mark；
The Initial Face training set of images is input in group's label markup model, is obtained for training the mesh
Mark the target facial image training set of network model.
Preferably, described to utilize IndexIVFKmeans method, it is carried out in conjunction with the centric quantity and the initial center
Model training, comprising:
When occurring advantage point and isolated point at the same time, using the shortest route of the surging point and the isolated point as diameter
Two point substitution surging points of random selection and the isolated point in circle.
Preferably, the training process of the target network, comprising:
Default network is trained using first-loss function, obtains initial network model data；
Regularized learning algorithm rate, and add the second loss function on the basis of the initial network model data and carry out network instruction
Practice, obtains target network model data.
Preferably, the second loss function is added on the basis of the initial network model data and carries out network training, is obtained
Obtain target network model data, comprising:
Sample is divided into difficult sample and general sample, and the first weight is set for the difficult sample, is described one
As sample be arranged the second weight；Wherein, the difficult sample is that the distance between similar sample is greater than or equal to inhomogeneity sample
Between distance sample；
Second loss function is adjusted using first weight and second weight, obtains third loss
Function；
The initial network is trained using the third loss function, obtains the target network.
Preferably, the facial image for obtaining personnel to be identified, comprising:
The face location and face size of the personnel to be identified are detected from monitoring scene, and utilize the face location
Original facial image is determined with the face size；
Crucial point location is carried out to the original facial image, and is normalized according to positioning result, the people is obtained
Face image.
A kind of face identification device, comprising:
The facial image is input to mesh for obtaining the facial image of personnel to be identified by image information acquisition module
It marks in network model, obtains the target complex class label and characteristic of the face；
Comparison data screening module has the target complex class label and marked personnel for reading from database
The facial image characteristic of identity information；
Matching identification module, for the characteristic to be compared with the facial image characteristic, and according to
Comparison result determines the identity information of the personnel to be identified.
A kind of face recognition device, comprising:
Memory, for storing computer program；
Processor, the step of above-mentioned face identification method is realized when for executing the computer program.
A kind of readable storage medium storing program for executing is stored with computer program, the computer program quilt on the readable storage medium storing program for executing
The step of processor realizes above-mentioned face identification method when executing.
Using method provided by the embodiment of the present invention, the facial image of personnel to be identified is obtained, facial image is inputted
Into target network model, the target complex class label and characteristic of face are obtained；Reading from database has target complex
The facial image characteristic of class label and marked personnel identity information；By characteristic and facial image characteristic into
Row compares, and the identity information of personnel to be identified is determined according to comparison result.
Before carrying out face alignment, the facial image first with target network model extraction personnel to be identified is corresponding
The characteristic and target complex class label of face.Then, only reading in the database has target complex class label and has marked
Remember the facial image characteristic of personnel identity information.Characteristic is compared with facial image characteristic, Bian Kegen
The identity information of face to be identified is determined according to comparison result.It will be in the facial image and database of personnel to be identified compared to directly
Face images characteristic be compared, this method can be by the difference of face realm distinguishing label, from database only
It reads with the facial image characteristic with target complex class label, that is to say, that by face realm distinguishing label to data
A large amount of facial image characteristic is screened in library, using only the face of the higher identical face realm distinguishing label of similarity
Image feature data compares.Comparison number can be so reduced, shortens and compares duration, further promotes recognition of face effect
Rate.
Correspondingly, the embodiment of the invention also provides face identification device corresponding with above-mentioned face identification method, set
Standby and readable storage medium storing program for executing, has above-mentioned technique effect, and details are not described herein.
Detailed description of the invention
In order to more clearly explain the embodiment of the invention or the technical proposal in the existing technology, to embodiment or will show below
There is attached drawing needed in technical description to be briefly described, it should be apparent that, the accompanying drawings in the following description is only this
Some embodiments of invention for those of ordinary skill in the art without creative efforts, can be with
It obtains other drawings based on these drawings.
Fig. 1 is a kind of implementation flow chart of face identification method in the embodiment of the present invention；
Fig. 2 is a kind of specific implementation flow chart of face identification method in the embodiment of the present invention；
Fig. 3 is the schematic illustration of the residual unit in convolutional neural networks；
Fig. 4 is a kind of structural schematic diagram of face identification device in the embodiment of the present invention；
Fig. 5 is a kind of structural schematic diagram of face recognition device in the embodiment of the present invention；
Fig. 6 is a kind of concrete structure schematic diagram of face recognition device in the embodiment of the present invention.
Specific embodiment
In order to enable those skilled in the art to better understand the solution of the present invention, with reference to the accompanying drawings and detailed description
The present invention is described in further detail.Obviously, the described embodiment is only a part of the embodiment of the present invention, rather than complete
The embodiment in portion.Based on the embodiments of the present invention, those of ordinary skill in the art are without making creative work
Every other embodiment obtained, shall fall within the protection scope of the present invention.
Embodiment one:
Referring to FIG. 1, Fig. 1 is a kind of flow chart of face identification method in the embodiment of the present invention, this method includes following
Step:
S101, the facial image for obtaining personnel to be identified, facial image is input in target network model, obtains face
Target complex class label and characteristic.
Wherein, personnel to be identified can be the personnel to be identified in video, photo.For example, what is occurred in monitor video is all
Such as a suspect of thief.There are many obtaining in the mode of facial image, mode one directly reads face figure from storage equipment
Picture；Mode two carries out Face datection to video data, obtains facial image；Mode three carries out face to the image with personage
Detection, obtains facial image.Below to carry out Face datection from real time monitoring scene, the process for obtaining facial image is carried out in detail
It describes in detail bright.The process that the facial image of personnel to be identified is obtained from monitoring scene includes:
Step 1: detecting the face location and face size of personnel to be identified from monitoring scene, and utilize face location
Original facial image is determined with face size；
Step 2: carrying out crucial point location to original facial image, and it is normalized according to positioning result, obtains face
Image.
It is illustrated for ease of description, below combining above-mentioned two step.
Monitoring scene, that is, monitor video as referred to herein, can be used method for detecting human face detected from monitoring scene it is to be identified
The face location and face size of personnel.Then, original facial image is determined based on the face location and face size.Then
Crucial point location is carried out to original facial image, and is normalized according to positioning result, facial image is obtained.Wherein face is examined
Survey can refer to ViolaJones method for detecting human face, and this method is a kind of based on integrogram, cascade detectors and AdaBoost
A kind of method of (Adaptive Boosting, machine learning method) algorithm, method frame includes following three parts:
One, using Harr-like character representation face, " integrogram " is used to realize the quick calculating of character numerical value；
Two, some rectangular characteristics (Weak Classifier) that can most represent face are picked out using Adaboost algorithm, according to adding
Weak Classifier is configured to a strong classifier by the mode of power ballot；
Three, several strong classifiers that training obtains are composed in series to the cascade filtering of a cascade structure, cascade structure
The detection speed of classifier can be effectively improved.
Certainly, it can also take other people face detection algorithms to obtain in other embodiments of the invention and carry out Face datection.
Wherein, the positioning of the key points such as crucial point location, that is, face key point location, such as eyes, nose, mouth, can be with deeply
Crucial point location is realized in degree study, and this is no longer going to repeat them.
After obtaining facial image, facial image can be input in target network model and be handled, obtain the people
The characteristic and target complex class label of face.Specifically, the quantity of face realm distinguishing label is up to thousands of, specifically for
For one database, the quantity of face realm distinguishing label is more, and the speed for carrying out face alignment is faster, correspondingly, for wait know
The matched accuracy of the facial image of others' member will be declined.Therefore, most beautiful woman can be determined according to practical application request
The quantity of face realm distinguishing label.
Wherein, target network model can be specially that convolutional neural networks etc. pass through with the model of deep learning ability
It trains and obtains.And before carrying out network training, it also needs to obtain and be instructed for the target facial image of training objective network model
Practice collection.Target facial image training set is the training set marked after face realm distinguishing label to original facial image training set.It can
Using IndexIVFKmeans method to the face realm of each Initial Face image pattern in original facial image training set
Distinguishing label, to obtain the target facial image training set.Wherein, IndexIVFKmeans method can return for one kind
The indexing means of IndexIVFFlat (inverted file or inverted index) object；IndexIVFFlat is specially to use K-means
Establish cluster centre, then by inquiring nearest cluster centre, then compare institute's directed quantity in cluster obtain it is similar to
Amount.Specific implementation process:
Step 1: obtaining Initial Face training set of images, and destination subset is randomly choosed from Initial Face training set；
Step 2: using in Initial Face training set of images sample size and destination subset in face characteristic data determine
Centric quantity and initial center；
Step 3: carrying out model training using IndexIVFKmeans method in conjunction with centric quantity and initial center, obtaining
For marking group's label markup model of face realm distinguishing label；
Step 4: Initial Face training set of images is input in group label markup model, obtains and be used for training objective net
The target facial image training set of network model.
It is illustrated for ease of description, below combining aforementioned four step.
Since Initial Face training set of images reaches ten million rank, therefore first randomly select a part of human face data (i.e. target
Subset), then use the characteristic of the neural network model extraction face an of benchmark as IndexIVFKmeans method (one
Kind of indexing means) input, the number of central point calculated with following formula:
Wherein, the number put centered on ncentroids, a are one
Coefficient, value can be in the number of samples in the destination subset that [4,16], ntrainsamnum are training group's label markup model.Its
In, central point is initialized as the mean value (i.e. the mean value of sample in destination subset) of the random Initial Face training set of images in part.
Central point has multiple, and each central point is multidimensional data, and dimension is identical with the dimension of human face data.Initial Face training set of images
Mean value i.e. by multiple face fusions together, calculate average face (i.e. each dimension corresponding face characteristic mean value).Instruction
It, can be defeated by Initial Face training set of images after getting the group's label markup model that can be used for marking face realm distinguishing label
Enter into group's label markup model, it is final to obtain the target facial image training set that can be used for training objective network model.
Preferably, it is contemplated that, may in the cluster process in IndexIVFKmeans method, especially cluster training process
It will appear that certain points are too surging, and some points are too isolated (such as faces domain there are its one, face in domain in addition to it at this
Except do not belong to the point of set).Therefore, most short with isolated point with surging point when occurring surging point and isolated point at the same time
Line is the surging point of two point substitutions of random selection and isolated point in the circle of diameter.Specifically, can be by surging point and isolated point
The shortest route (direct-connected line) draws circle circle1 as diameter, then draws the concentric circles circle2 that a diameter is its half again,
Two o'clock is randomly selected in circle2 replaces surging point and isolated point.Certainly, it can also directly take surging point most short with isolated point
The surging point of the central point substitution of line and isolated point, can also directly abandon surging point or isolated point.
After obtaining the target facial image training set for training objective network model, the target face figure can be utilized
As training set carries out network training, final acquisition target network model.In order to improve training effect, the side of classification training can be taken
Formula carries out, and for ease of description, sample involved in training process includes the sample in target facial image training set, should
Training process can include:
First order training: being trained default network using first-loss function, obtains initial network model data；
Second level training: regularized learning algorithm rate, and add on the basis of initial network model data the second loss function into
Row network training obtains target network model data.
For ease of description, the training of above-mentioned two-stage is combined below and is illustrated.
It is the target network model data of determining final target network model to the process that default network is trained.
Network training accuracy rate can be promoted by way of classification, multitask training.Specifically,
Preferably, in order to improve accuracy rate, can also by setting weight balancing difficulty sample (difficult sample be similar sample
The distance between this is greater than or equal to the sample of distance between inhomogeneity sample) and general sample.Specifically, can be instructed in the second level
When practicing, sample is divided into difficult sample and general sample, and the first weight is set for difficult sample, for general sample setting the
Two weights；The second loss function is adjusted using the first weight and the second weight, obtains third loss function；Utilize third
Loss function is trained initial network, obtains target network.
Wherein, first-loss function:
N is face sample number, and i is i-th of sample, and yi is classification belonging to i-th of sample, and s is zoom factor, and θ is power
Angle interval between weight vector sum sampling feature vectors, t is angled edge.
Second loss function:Wherein, N is sample
Number,For the characteristic of sample i,For with the same category of sample characteristics data of sample i,For and sample
The inhomogeneous sample characteristics data of this i.
Third loss function:Wherein, N is sample
Number,For the characteristic of sample i,For with the same category of sample characteristics data of sample iFor and sample
The inhomogeneous sample characteristics data of this i work as a, and when this group of sample of p, n is difficult sample, m is just m1 (with the first weight), otherwise
For m2 (with the second weight).It is so-called difficulty sample refer to that distance is farther out between similar sample, between inhomogeneity sample distance compared with
Close sample.Wherein, regularized learning algorithm rate specifically can refer to reduce learning rate, such as be adjusted to learning rate when the first order is trained very
One of.
S102, the facial image spy with target complex class label and marked personnel identity information is read from database
Levy data.
In embodiments of the present invention, before carrying out recognition of face, i.e., before the facial image for obtaining personnel to be identified,
It also needs that the facial image characteristic in database is marked, is marked in different face realm distinguishing labels respectively.Specifically
The realization process includes:
Step 1: obtaining the facial image sample set of marked personnel identity information；
Step 2: utilizing the face characteristic of each facial image sample in target network model extraction facial image sample set
Data and face realm distinguishing label；
Step 3: face characteristic data are stored in database respectively according to face realm distinguishing label.
For ease of description, above three step is combined below and is illustrated.
Wherein, the personnel identity information that each facial image sample is marked in facial image sample set can handmarking.
Then the face characteristic data of each facial image sample and face realm distinguishing label in each facial image sample set are extracted.
In the database be arranged respective numbers face realm distinguishing label, and by each facial image sample classification deposit and division result
In matched face realm distinguishing label.For example, 3500 are arranged in the database if division result is to have 3500 central points
Then which group face realm distinguishing label is respectively belonging to according to each facial image sample, then belongs to the facial image
In corresponding face realm distinguishing label, the determination and division of the face realm distinguishing label of facial image sample are so just completed, i.e.,
Complete the face realm distinguishing label label of facial image characteristic.
In this way, can determine just to know facial image with after the matched target complex class label of facial image and return
The similitude for belonging to the facial image characteristic of other target realms is less than facial image and belongs to target complex class label
Facial image characteristic similitude.Therefore, it is special that the non-facial image for belonging to target complex class label can be excluded first
Levy data.The face belonged to target complex class label and marked personnel identity information can be read directly from database at this time
Image feature data.
S103, characteristic is compared with facial image characteristic, and people to be identified is determined according to comparison result
The identity information of member.
After reading facial image characteristic, can by the characteristic of the face and facial image characteristic into
Row compares, and the identity information of personnel to be identified is determined according to comparison result.
Wherein, comparison process can be specially to calculate the similitude of characteristic and facial image characteristic, the phase one by one
It can be calculated with specific reference to face characteristic data like property, similitude is then selected to be greater than a facial image spy of similar threshold value
Personnel identity information of the personnel identity information that sign data are marked as facial image.Also the maximum face of similitude may be selected
Personnel identity information of the personnel identity information that image feature data is marked as facial image.Wherein personnel identity information is
The identity that personnel to be identified can be uniquely determined, if personnel identity information can be ID number or name+age+address.Such as
This, just completes recognition of face.
Using method provided by the embodiment of the present invention, the facial image of personnel to be identified is obtained, facial image is inputted
Into target network model, the target complex class label and characteristic of face are obtained；Reading from database has target complex
The facial image characteristic of class label and marked personnel identity information；By characteristic and facial image characteristic into
Row compares, and the identity information of personnel to be identified is determined according to comparison result.
Before carrying out face alignment, the facial image first with target network model extraction personnel to be identified is corresponding
The characteristic and target complex class label of face.Then, only reading in the database has target complex class label and has marked
Remember the facial image characteristic of personnel identity information.Characteristic is compared with facial image characteristic, Bian Kegen
The identity information of face to be identified is determined according to comparison result.It will be in the facial image and database of personnel to be identified compared to directly
Face images characteristic be compared, this method can be by the difference of face realm distinguishing label, from database only
It reads with the facial image characteristic with target complex class label, that is to say, that by face realm distinguishing label to data
A large amount of facial image characteristic is screened in library, using only the face of the higher identical face realm distinguishing label of similarity
Image feature data compares.Comparison number can be so reduced, shortens and compares duration, further promotes recognition of face effect
Rate.
Embodiment two:
Face identification method provided by the embodiment of the present invention is better understood for the ease of those skilled in the art, below
By taking specific application scenarios as an example, it is provided for the embodiments of the invention face identification method and is described in detail,
Referring to FIG. 2, Fig. 2 is a kind of specific implementation flow chart of face identification method in the embodiment of the present invention.Firstly, searching
Collect the face sample of separate sources, the order of magnitude is that millions are other, manually demarcates the class label of these samples, can also obtain one
The public collection marked a bit.Face datection and face key point location are done to these faces, and it is same to normalize to 112x112
Size.
Face sample after normalizing above randomly chooses a part of sample, the instruction for IndexIVFKmeans method
Practice, the order of magnitude is 100,000 grades.The network model of benchmark is used to extract its characteristic as the defeated of IndexIVFKmeans method
Enter.The number of its central point is usedIt calculates, a takes 10.
With the demographic categories label of the model data automatic Calibration face sample obtained after training.
Network structure is constructed, the convolutional neural networks that can be constructed, which includes input layer (Input), convolutional layer
(Conv), nonlinear transformation layer (PRelu), pond layer (Pooling), residual unit (Resblock, as shown in Figure 3), Quan Lian
It connects layer (fc), loss function layer (ArcFace and improved TripletLoss) is described as follows shown in table:
| Layer | Description |
| Input layer | Human face data |
| Convolutional layer | 64 map outputs |
| Pond layer | Down-sampling |
| Resblock1x3 | Continuous 3 residual units are connected |
| Resblock2x4 | Continuous 4 residual units are connected |
| Resblock3x14 | Continuous 14 residual units are connected |
| Resblock4x3 | Continuous 3 residual units are connected |
| Connect layer entirely | Linear weighted function summation |
| ArcFace (first-loss function) | Loss function 1, weight 1 |
| Improved Triletloss (third loss function) | Loss function 2, weight 2 |
Convolutional neural networks describe table
First order training, the loss function of network is first-loss function (ArcFace), and the learning rate of network is set as 0.1.
When network is no longer restrained, deconditioning.Wherein network no longer restrain i.e. network penalty values variation it is smaller, can specifically set
Corresponding threshold value is set to carry out judging whether to reach the lesser requirement of variation.
Second level training, training, combines classification and the face group of face in the trained network models of the first order
Body classification is trained.The loss function of network is ArcFace and third loss function, i.e., improved Tripletloss, net
The learning rate of network is set as 0.01.When network is no longer restrained, stop its training.
The face characteristic data and face realm distinguishing label for needing to be put in storage with trained model extraction, by different faces group
The face characteristic data of class label are stored in different face databases, are labeled differentiation with face realm distinguishing label.
The face characteristic data of personnel to be identified and affiliated face realm distinguishing label in monitoring scene are extracted, it then will be special
The face characteristic data of same face realm distinguishing label are compared in sign data and database, export result.
Embodiment three:
Corresponding to above method embodiment, the embodiment of the invention also provides a kind of face identification devices, are described below
Face identification device can correspond to each other reference with above-described face identification method.
Shown in Figure 4, which comprises the following modules:
Facial image is input to target for obtaining the facial image of personnel to be identified by image information acquisition module 101
In network model, the target complex class label and characteristic of face are obtained；
Comparison data screening module 102 has target complex class label and marked personnel for reading from database
The facial image characteristic of identity information；
Matching identification module 103 is tied for characteristic to be compared with facial image characteristic, and according to comparing
Fruit determines the identity information of personnel to be identified.
Using device provided by the embodiment of the present invention, the facial image of personnel to be identified is obtained, facial image is inputted
Into target network model, the target complex class label and characteristic of face are obtained；Reading from database has target complex
The facial image characteristic of class label and marked personnel identity information；By characteristic and facial image characteristic into
Row compares, and the identity information of personnel to be identified is determined according to comparison result.
Before carrying out face alignment, the facial image first with target network model extraction personnel to be identified is corresponding
The characteristic and target complex class label of face.Then, only reading in the database has target complex class label and has marked
Remember the facial image characteristic of personnel identity information.Characteristic is compared with facial image characteristic, Bian Kegen
The identity information of face to be identified is determined according to comparison result.It will be in the facial image and database of personnel to be identified compared to directly
Face images characteristic be compared, which can be by the difference of face realm distinguishing label, from database only
It reads with the facial image characteristic with target complex class label, that is to say, that by face realm distinguishing label to data
A large amount of facial image characteristic is screened in library, using only the face of the higher identical face realm distinguishing label of similarity
Image feature data compares.Comparison number can be so reduced, shortens and compares duration, further promotes recognition of face effect
Rate.
In a kind of specific embodiment of the invention, further includes:
The face realm distinguishing label setup module of comparison data, for before the facial image for obtaining personnel to be identified,
Obtain the facial image sample set of marked personnel identity information；Using each in target network model extraction facial image sample set
The face characteristic data and face realm distinguishing label of a facial image sample；By face characteristic data according to face realm distinguishing label
It is stored in database respectively.
In a kind of specific embodiment of the invention, further includes:
Target facial image training set obtains module, for obtaining just before the facial image for obtaining personnel to be identified
Beginning facial image training set, and destination subset is randomly choosed from Initial Face training set；Utilize Initial Face training set of images
In sample size and destination subset in face characteristic data determine centric quantity and initial center；Utilize IndexIVFKmeans
Method carries out model training in conjunction with centric quantity and initial center, obtains group's label mark for marking face realm distinguishing label
Remember model；Initial Face training set of images is input in group label markup model, is obtained for training objective network model
Target facial image training set.
In a kind of specific embodiment of the invention, the face realm distinguishing label setup module of comparison data is specific to use
When occurring advantage point and isolated point at the same time, using the shortest route of surging point and isolated point to randomly choose two in the circle of diameter
A point substitution advantage point and isolated point.
In a kind of specific embodiment of the invention, further includes:
Target network training module obtains initial network for being trained using first-loss function to default network
Model data；Regularized learning algorithm rate, and add the second loss function on the basis of initial network model data and carry out network training,
Obtain target network model data.
In a kind of specific embodiment of the invention, target network training module is tired specifically for sample to be divided into
Difficult sample and general sample, and the first weight is set for difficult sample, for general sample, the second weight is set；Wherein, difficult sample
This is the sample that the distance between similar sample is greater than or equal to distance between inhomogeneity sample；It is weighed using the first weight and second
The second loss function is adjusted again, obtains third loss function；Initial network is trained using third loss function,
Obtain target network.
In a kind of specific embodiment of the invention, image information acquisition module 101 is detected from monitoring scene wait know
The face location and face size of others' member, and original facial image is determined using face location and face size；To primitive man
Face image carries out crucial point location, and is normalized according to positioning result, obtains facial image.
Example IV:
Corresponding to above method embodiment, the embodiment of the invention also provides a kind of face recognition devices, are described below
A kind of face recognition device can correspond to each other reference with a kind of above-described face identification method.
Shown in Figure 5, which includes:
Memory D1, for storing computer program；
Processor D2, when for executing computer program the step of the face identification method of realization above method embodiment.
Specifically, referring to FIG. 6, Fig. 6 be a kind of concrete structure schematic diagram of face recognition device provided in this embodiment,
The face recognition device can generate bigger difference because configuration or performance are different, may include one or more processing
Device (central processing units, CPU) 322 (for example, one or more processors) and memory 332, one
(such as one or more mass memories of storage medium 330 of a or more than one storage application program 342 or data 344
Equipment).Wherein, memory 332 and storage medium 330 can be of short duration storage or persistent storage.It is stored in storage medium 330
Program may include one or more modules (diagram does not mark), and each module may include in data processing equipment
Series of instructions operation.Further, central processing unit 322 can be set to communicate with storage medium 330, in recognition of face
The series of instructions operation in storage medium 330 is executed in equipment 301.
Face recognition device 301 can also include one or more power supplys 326, one or more wired or nothings
Wired network interface 350, one or more input/output interfaces 358, and/or, one or more operating systems 341.
For example, Windows ServerTM, Mac OS XTM, UnixTM, LinuxTM, FreeBSDTM etc..
Step in face identification method as described above can be realized by the structure of face recognition device.
Embodiment five:
Corresponding to above method embodiment, the embodiment of the invention also provides a kind of readable storage medium storing program for executing, are described below
A kind of readable storage medium storing program for executing can correspond to each other reference with a kind of above-described face identification method.
A kind of readable storage medium storing program for executing is stored with computer program on readable storage medium storing program for executing, and computer program is held by processor
The step of face identification method of above method embodiment is realized when row.
The readable storage medium storing program for executing be specifically as follows USB flash disk, mobile hard disk, read-only memory (Read-Only Memory,
ROM), the various program storage generations such as random access memory (Random Access Memory, RAM), magnetic or disk
The readable storage medium storing program for executing of code.
Professional also can be further appreciated that, unit described in conjunction with the examples disclosed in the embodiments of the present disclosure and
Algorithm steps can be realized with electronic hardware, computer software, or a combination of the two, in order to clearly demonstrate hardware and soft
The interchangeability of part generally describes each exemplary composition and step according to function in the above description.These function
It can be implemented in hardware or software actually, the specific application and design constraint depending on technical solution.Professional skill
Art personnel can use different methods to achieve the described function each specific application, but this realization should not be recognized
It is beyond the scope of this invention.
Claims (10)
1. a kind of face identification method characterized by comprising
The facial image is input in target network model by the facial image for obtaining personnel to be identified, obtains the face
Target complex class label and characteristic；
The facial image characteristic with the target complex class label and marked personnel identity information is read from database
According to；
The characteristic is compared with the facial image characteristic, and is determined according to comparison result described to be identified
The identity information of personnel.
2. face identification method according to claim 1, which is characterized in that in the face figure for obtaining personnel to be identified
Before picture, further includes:
Obtain the facial image sample set of the marked personnel identity information；
Utilize the face characteristic number of each facial image sample in facial image sample set described in the target network model extraction
According to face realm distinguishing label；
The face characteristic data are stored in the database respectively according to the face realm distinguishing label.
3. face identification method according to claim 2, which is characterized in that in the face figure for obtaining personnel to be identified
Before picture, further includes:
Initial Face training set of images is obtained, and randomly chooses destination subset from the Initial Face training set；
Using in the Initial Face training set of images sample size and the destination subset in face characteristic data determine in
Calculation amount and initial center；
Using IndexIVFKmeans method, model training is carried out in conjunction with the centric quantity and the initial center, is used
In group's label markup model of label face realm distinguishing label；
The Initial Face training set of images is input in group's label markup model, is obtained for training the target network
The target facial image training set of network model.
4. face identification method according to claim 2, which is characterized in that it is described to utilize IndexIVFKmeans method,
Model training is carried out in conjunction with the centric quantity and the initial center, comprising:
When occurring advantage point and isolated point at the same time, using the shortest route of the surging point and the isolated point as in the circle of diameter
Randomly choose two point substitution surging points and the isolated point.
5. face identification method according to claim 1, which is characterized in that the training process of the target network, comprising:
Default network is trained using first-loss function, obtains initial network model data；
Regularized learning algorithm rate, and add the second loss function on the basis of the initial network model data and carry out network training,
Obtain target network model data.
6. face identification method according to claim 5, which is characterized in that on the basis of the initial network model data
The second loss function of upper addition carries out network training, obtains target network model data, comprising:
Sample is divided into difficult sample and general sample, and the first weight is set for the difficult sample, is the general sample
The second weight of this setting；Wherein, the difficult sample is that the distance between similar sample is greater than or equal between inhomogeneity sample
The sample of distance；
Second loss function is adjusted using first weight and second weight, third is obtained and loses letter
Number；
The initial network is trained using the third loss function, obtains the target network.
7. face identification method according to any one of claims 1 to 5, which is characterized in that described to obtain personnel to be identified
Facial image, comprising:
The face location and face size of the personnel to be identified are detected from monitoring scene, and utilize the face location and institute
It states face size and determines original facial image；
Crucial point location is carried out to the original facial image, and is normalized according to positioning result, the face figure is obtained
Picture.
8. a kind of face identification device characterized by comprising
The facial image is input to target network for obtaining the facial image of personnel to be identified by image information acquisition module
In network model, the target complex class label and characteristic of the face are obtained；
Comparison data screening module has the target complex class label and marked personnel identity for reading from database
The facial image characteristic of information；
Matching identification module, for the characteristic to be compared with the facial image characteristic, and according to comparison
As a result the identity information of the personnel to be identified is determined.
9. a kind of face recognition device characterized by comprising
Memory, for storing computer program；
Processor, realizing the face identification method as described in any one of claim 1 to 7 when for executing the computer program
Step.
10. a kind of readable storage medium storing program for executing, which is characterized in that be stored with computer program, the meter on the readable storage medium storing program for executing
It is realized when calculation machine program is executed by processor as described in any one of claim 1 to 7 the step of face identification method.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910173706.0A CN109919093A (en) | 2019-03-07 | 2019-03-07 | A kind of face identification method, device, equipment and readable storage medium storing program for executing |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910173706.0A CN109919093A (en) | 2019-03-07 | 2019-03-07 | A kind of face identification method, device, equipment and readable storage medium storing program for executing |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN109919093A true CN109919093A (en) | 2019-06-21 |
Family
ID=66963802
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201910173706.0A Pending CN109919093A (en) | 2019-03-07 | 2019-03-07 | A kind of face identification method, device, equipment and readable storage medium storing program for executing |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN109919093A (en) |
Cited By (14)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110348387A (en) * | 2019-07-12 | 2019-10-18 | 腾讯科技（深圳）有限公司 | A kind of image processing method, device and computer readable storage medium |
| CN110765933A (en) * | 2019-10-22 | 2020-02-07 | 山西省信息产业技术研究院有限公司 | Dynamic portrait sensing comparison method applied to driver identity authentication system |
| CN111009065A (en) * | 2019-12-09 | 2020-04-14 | 金现代信息产业股份有限公司 | Face recognition access control system optimization method and system based on clustering algorithm |
| CN111325156A (en) * | 2020-02-24 | 2020-06-23 | 北京沃东天骏信息技术有限公司 | Face recognition method, device, equipment and storage medium |
| CN111507238A (en) * | 2020-04-13 | 2020-08-07 | 三一重工股份有限公司 | Face data screening method and device and electronic equipment |
| CN111563086A (en) * | 2020-01-13 | 2020-08-21 | 杭州海康威视系统技术有限公司 | Information association method, device, equipment and storage medium |
| CN111626091A (en) * | 2020-03-09 | 2020-09-04 | 咪咕文化科技有限公司 | Face image annotation method and device and computer readable storage medium |
| CN111968152A (en) * | 2020-07-15 | 2020-11-20 | 桂林远望智能通信科技有限公司 | Dynamic identity recognition method and device |
| CN112597803A (en) * | 2020-11-26 | 2021-04-02 | 深圳泰首智能技术有限公司 | Face recognition method, device and system and electronic equipment |
| CN112766164A (en) * | 2021-01-20 | 2021-05-07 | 深圳力维智联技术有限公司 | Face recognition model training method, device and equipment and readable storage medium |
| CN113095110A (en) * | 2019-12-23 | 2021-07-09 | 浙江宇视科技有限公司 | Method, device, medium and electronic equipment for dynamically warehousing face data |
| CN113221088A (en) * | 2021-06-15 | 2021-08-06 | 中国银行股份有限公司 | User identity identification method and device |
| CN113313034A (en) * | 2021-05-31 | 2021-08-27 | 平安国际智慧城市科技股份有限公司 | Face recognition method and device, electronic equipment and storage medium |
| CN115410265A (en) * | 2022-11-01 | 2022-11-29 | 合肥的卢深视科技有限公司 | Model training method, face recognition method, electronic device and storage medium |
Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107273796A (en) * | 2017-05-05 | 2017-10-20 | 珠海数字动力科技股份有限公司 | A kind of fast face recognition and searching method based on face characteristic |
| CN108492194A (en) * | 2018-03-06 | 2018-09-04 | 平安科技（深圳）有限公司 | Products Show method, apparatus and storage medium |
| CN109086669A (en) * | 2018-06-29 | 2018-12-25 | 汉王科技股份有限公司 | Recognition of face auth method, device, electronic equipment |
| CN109214360A (en) * | 2018-10-15 | 2019-01-15 | 北京亮亮视野科技有限公司 | A kind of construction method of the human face recognition model based on ParaSoftMax loss function and application |
| CN109359575A (en) * | 2018-09-30 | 2019-02-19 | 腾讯科技（深圳）有限公司 | Method for detecting human face, method for processing business, device, terminal and medium |
-
2019
- 2019-03-07 CN CN201910173706.0A patent/CN109919093A/en active Pending
Patent Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107273796A (en) * | 2017-05-05 | 2017-10-20 | 珠海数字动力科技股份有限公司 | A kind of fast face recognition and searching method based on face characteristic |
| CN108492194A (en) * | 2018-03-06 | 2018-09-04 | 平安科技（深圳）有限公司 | Products Show method, apparatus and storage medium |
| CN109086669A (en) * | 2018-06-29 | 2018-12-25 | 汉王科技股份有限公司 | Recognition of face auth method, device, electronic equipment |
| CN109359575A (en) * | 2018-09-30 | 2019-02-19 | 腾讯科技（深圳）有限公司 | Method for detecting human face, method for processing business, device, terminal and medium |
| CN109214360A (en) * | 2018-10-15 | 2019-01-15 | 北京亮亮视野科技有限公司 | A kind of construction method of the human face recognition model based on ParaSoftMax loss function and application |
Cited By (22)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110348387A (en) * | 2019-07-12 | 2019-10-18 | 腾讯科技（深圳）有限公司 | A kind of image processing method, device and computer readable storage medium |
| CN110765933A (en) * | 2019-10-22 | 2020-02-07 | 山西省信息产业技术研究院有限公司 | Dynamic portrait sensing comparison method applied to driver identity authentication system |
| CN111009065A (en) * | 2019-12-09 | 2020-04-14 | 金现代信息产业股份有限公司 | Face recognition access control system optimization method and system based on clustering algorithm |
| CN113095110B (en) * | 2019-12-23 | 2024-03-08 | 浙江宇视科技有限公司 | Method, device, medium and electronic equipment for dynamically warehousing face data |
| CN113095110A (en) * | 2019-12-23 | 2021-07-09 | 浙江宇视科技有限公司 | Method, device, medium and electronic equipment for dynamically warehousing face data |
| CN111563086A (en) * | 2020-01-13 | 2020-08-21 | 杭州海康威视系统技术有限公司 | Information association method, device, equipment and storage medium |
| CN111563086B (en) * | 2020-01-13 | 2023-09-19 | 杭州海康威视系统技术有限公司 | Information association method, device, equipment and storage medium |
| CN111325156B (en) * | 2020-02-24 | 2023-08-11 | 北京沃东天骏信息技术有限公司 | Face recognition method, device, equipment and storage medium |
| CN111325156A (en) * | 2020-02-24 | 2020-06-23 | 北京沃东天骏信息技术有限公司 | Face recognition method, device, equipment and storage medium |
| CN111626091B (en) * | 2020-03-09 | 2023-09-22 | 咪咕文化科技有限公司 | Face image labeling method and device and computer readable storage medium |
| CN111626091A (en) * | 2020-03-09 | 2020-09-04 | 咪咕文化科技有限公司 | Face image annotation method and device and computer readable storage medium |
| CN111507238A (en) * | 2020-04-13 | 2020-08-07 | 三一重工股份有限公司 | Face data screening method and device and electronic equipment |
| CN111968152B (en) * | 2020-07-15 | 2023-10-17 | 桂林远望智能通信科技有限公司 | Dynamic identity recognition method and device |
| CN111968152A (en) * | 2020-07-15 | 2020-11-20 | 桂林远望智能通信科技有限公司 | Dynamic identity recognition method and device |
| CN112597803A (en) * | 2020-11-26 | 2021-04-02 | 深圳泰首智能技术有限公司 | Face recognition method, device and system and electronic equipment |
| CN112766164A (en) * | 2021-01-20 | 2021-05-07 | 深圳力维智联技术有限公司 | Face recognition model training method, device and equipment and readable storage medium |
| CN113313034A (en) * | 2021-05-31 | 2021-08-27 | 平安国际智慧城市科技股份有限公司 | Face recognition method and device, electronic equipment and storage medium |
| CN113313034B (en) * | 2021-05-31 | 2024-03-22 | 平安国际智慧城市科技股份有限公司 | Face recognition method and device, electronic equipment and storage medium |
| CN113221088B (en) * | 2021-06-15 | 2022-08-19 | 中国银行股份有限公司 | User identity identification method and device |
| CN113221088A (en) * | 2021-06-15 | 2021-08-06 | 中国银行股份有限公司 | User identity identification method and device |
| CN115410265A (en) * | 2022-11-01 | 2022-11-29 | 合肥的卢深视科技有限公司 | Model training method, face recognition method, electronic device and storage medium |
| CN115410265B (en) * | 2022-11-01 | 2023-01-31 | 合肥的卢深视科技有限公司 | Model training method, face recognition method, electronic device and storage medium |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109919093A (en) | A kind of face identification method, device, equipment and readable storage medium storing program for executing | |
| CN110263774B (en) | A kind of method for detecting human face | |
| CN107194341B (en) | Face recognition method and system based on fusion of Maxout multi-convolution neural network | |
| CN111723786B (en) | Method and device for detecting wearing of safety helmet based on single model prediction | |
| CN103136533B (en) | Based on face identification method and the device of dynamic threshold | |
| CN106897738B (en) | A kind of pedestrian detection method based on semi-supervised learning | |
| CN109214373B (en) | Face recognition system and method for attendance checking | |
| CN102682309B (en) | Face feature registering method and device based on template learning | |
| CN110163258A (en) | A kind of zero sample learning method and system reassigning mechanism based on semantic attribute attention | |
| CN102938065B (en) | Face feature extraction method and face identification method based on large-scale image data | |
| WO2019179403A1 (en) | Fraud transaction detection method based on sequence width depth learning | |
| CN105447532A (en) | Identity authentication method and device | |
| CN111325115A (en) | Countermeasures cross-modal pedestrian re-identification method and system with triple constraint loss | |
| CN112926405A (en) | Method, system, equipment and storage medium for detecting wearing of safety helmet | |
| CN109284675A (en) | A kind of recognition methods of user, device and equipment | |
| CN102902980B (en) | A kind of biometric image analysis based on linear programming model and recognition methods | |
| CN110503000B (en) | Teaching head-up rate measuring method based on face recognition technology | |
| CN110598535A (en) | Face recognition analysis method used in monitoring video data | |
| CN105184266B (en) | A kind of finger venous image recognition methods | |
| CN112183438B (en) | Image identification method for illegal behaviors based on small sample learning neural network | |
| CN109948616A (en) | Image detecting method, device, electronic equipment and computer readable storage medium | |
| CN110688888B (en) | Pedestrian attribute identification method and system based on deep learning | |
| CN110827432B (en) | Class attendance checking method and system based on face recognition | |
| CN105956570B (en) | Smiling face's recognition methods based on lip feature and deep learning | |
| CN106096517A (en) | A kind of face identification method based on low-rank matrix Yu eigenface |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| RJ01 | Rejection of invention patent application after publication | ||
| RJ01 | Rejection of invention patent application after publication |
Application publication date: 20190621 |