CN108090451A - A kind of face identification method and system - Google Patents
A kind of face identification method and system Download PDFInfo
- Publication number
- CN108090451A CN108090451A CN201711384016.7A CN201711384016A CN108090451A CN 108090451 A CN108090451 A CN 108090451A CN 201711384016 A CN201711384016 A CN 201711384016A CN 108090451 A CN108090451 A CN 108090451A
- Authority
- CN
- China
- Prior art keywords
- image
- identified
- facial image
- dimensional
- feature vector
- 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
- G06F18/22—Matching criteria, e.g. proximity measures
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
Abstract
The invention discloses a kind of face identification method and systems.Method includes：First, using training set training convolutional neural networks, the convolutional neural networks after being trained；Secondly, the similarity between each facial image in facial image to be identified and default database is calculated using the convolutional neural networks after training；And the facial image of similarity maximum is chosen as preliminary identification image；Then, the three-dimensional feature vector sum of the calculating facial image to be identified tentatively identifies Chebyshev's distance between the three-dimensional feature vector of image；Finally judge whether Chebyshev's distance is more than the first predetermined threshold value, if, then the facial image to be identified and the preliminary identification image are not the images of same person, the facial image to be identified is not in presetting database, if not, the then facial image to be identified and the image that the preliminary identification image is same person, realize in the database and not database unknown face accurate identification.
Description
Technical field
The present invention relates to field of image recognition, more particularly to a kind of face identification method and system.
Background technology
With the development of depth learning technology, the face recognition technology based on deep learning is gradually being applied increasingly
Extensively.Recognition of face is a kind of biological identification technology that the facial feature information based on people carries out identification.Face identification method
It is integrated with a variety of professional techniques such as machine learning, pattern-recognition and Digital Image Processing.Recognition of face key has at 2 points：(1) people
The extraction of face feature；(2) identification of feature is compared.Deep learning realizes the automation of image characteristics extraction and identification, pole
Big improves accuracy of identification.The design of deep learning network structure often directly influences face used in recognition of face
The effect of identification.Therefore design a kind of suitable deep learning network structure be improve recognition of face precision vital task it
One.Although higher to the known recognition of face precision in database based on the face recognition technology of deep learning, for not
It is easily judged by accident in the unknown recognition of face of lane database.
The content of the invention
The object of the present invention is to database neutralization is not identified accurately in the unknown face of database in order to realize,
A kind of face identification method and system are provided.
To achieve the above object, the present invention provides following schemes：
A kind of face identification method, the recognition methods include the following steps：
Multiple facial images are extracted from default database, form training set；
According to the training set training convolutional neural networks, the convolutional neural networks after being trained；
Obtain facial image to be identified；
Each face figure in facial image to be identified and default database is calculated using the convolutional neural networks after training
Similarity as between；
The facial image of similarity maximum is chosen as preliminary identification image；
The three-dimensional feature vector sum for obtaining facial image to be identified respectively tentatively knows the three-dimensional feature vector of image；
The three-dimensional feature vector sum for calculating the facial image to be identified is tentatively identified between the three-dimensional feature vector of image
Chebyshev's distance；
Judge whether Chebyshev distance is more than the first predetermined threshold value, if so, the facial image to be identified with
The preliminary identification image is not the image of same person, and the facial image to be identified is not in presetting database, if it is not, then
The facial image to be identified and the image that the preliminary identification image is same person.
Optionally, it is described according to the training set training convolutional neural networks, the convolutional neural networks after being trained, tool
Body includes：
Training set is formed into triplet sets；
Triplet sets are sent into convolutional neural networks, obtain the loss function of convolutional neural networks；
The training convolutional neural networks make the value of the loss function be less than the second predetermined threshold value, the volume after being trained
Product neutral net.
Optionally, it is described that training set is formed into triplet sets, it specifically includes：
Three facial images are chosen from training set, three facial images include two and come from same person difference shape
The facial image of state, a facial image from another person form a triple；
Facial image in training set is formed triplet sets by the step of repeating to form a triple.
Optionally, the loss function is：
Wherein, i=1,2 ... ..., N represent i-th of triple, and N represents that three-number set concentrates the quantity of three-number set,It is the feature representation that same people's different conditions face two opens image in i-th of triple,It is i-th
The feature representation of the image of another people in triple, when+the value for representing in [] is more than zero, value in [] is actual value, less than zero
When, the value in [] is that zero, α isWithThe distance between andWithThe distance between minimal difference.
Optionally, the three-dimensional feature vector sum for obtaining facial image to be identified respectively tentatively knows the three-dimensional feature of image
Vector specifically includes：
Based on deep neural network, according to the facial image to be identified and the preliminary identification image, three-dimensional shaped is established
Varying model, the three-dimensional deformation model include the three-dimensional deformation model of facial image to be identified and the three-dimensional shaped of preliminary identification image
Varying model；
Weak perspective projection, the three-dimensional deformation model after being projected are carried out to the three-dimensional deformation model；
Three-dimensional deformation model after the projection is subjected to 3D mesh generations, normalized and adjusts three-dimensional deformation model
Anchor point, obtain treated three-dimensional deformation model；
Facial trend fitting, the three-dimensional deformation model after being fitted are carried out to treated the three-dimensional deformation model；
Three-dimensional feature vector is obtained according to the three-dimensional deformation model after fitting；The three-dimensional feature vector includes people to be identified
The three-dimensional feature vector sum of face image tentatively identifies the three-dimensional feature vector of image.
A kind of face identification system, the identifying system include：
Training set acquisition module for extracting multiple facial images from default database, forms training set；
Convolutional neural networks training module, for according to the training set training convolutional neural networks, after being trained
Convolutional neural networks；
Facial image acquisition module to be identified, for obtaining facial image to be identified；
Similarity calculation module, for calculating facial image to be identified and default using the convolutional neural networks after training
The similarity between each facial image in database；
Preliminary identification image chooses module, for choosing the facial image of similarity maximum as preliminary identification image；
Three-dimensional feature vector acquisition module, the three-dimensional feature vector sum for obtaining facial image to be identified respectively are tentatively known
The three-dimensional feature vector of image；
Chebyshev's distance calculation module, the three-dimensional feature vector sum for calculating the facial image to be identified are tentatively known
Chebyshev's distance between the three-dimensional feature vector of other image；
Judgment module, for judging whether Chebyshev's distance is more than the first predetermined threshold value, if so, described wait to know
Others' face image and the image that the preliminary identification image is not same person, the facial image to be identified is not in preset data
In storehouse, if it is not, the then facial image to be identified and the image that the preliminary identification image is same person.
Optionally, the convolutional neural networks training module, specifically includes：
Triplet sets setting up submodule, for training set to be formed triplet sets；
Loss function acquisition submodule for triplet sets to be sent into convolutional neural networks, obtains convolutional neural networks
Loss function；
Convolutional neural networks train submodule, and the value for making the loss function for training the convolutional neural networks is less than
Second predetermined threshold value, the convolutional neural networks after being trained.
Optionally, the triplet sets setting up submodule, specifically includes：
Triple establishes unit, and for choosing three facial images from training set, three facial images include two
The facial image from same person different conditions is opened, a facial image from another person forms a triple；
The step of triple combination establishes unit, a triple is formed for repetition, by the facial image in training set
Form triplet sets.
Optionally, the loss function in loss function acquisition submodule is：
Wherein, i=1,2 ... ..., N represent i-th of triple, and N represents that three-number set concentrates the quantity of three-number set,It is the feature representation that same people's different conditions face two opens image in i-th of triple,It is i-th
The feature representation of the image of another people in triple, when+the value for representing in [] is more than zero, value in [] is actual value, less than zero
When, the value in [] is that zero, α isWithThe distance between andWithThe distance between minimal difference.
Optionally, the three-dimensional feature vector acquisition module, specifically includes：
Three-dimensional deformation model foundation submodule, for the facial image to be identified and the preliminary identification image to be inputted
Deep neural network, establishes three-dimensional deformation model, and the three-dimensional deformation model includes the three-dimensional deformation mould of facial image to be identified
The three-dimensional deformation model of type and preliminary identification image；
Weak perspective projection submodule, for carrying out weak perspective projection to the three-dimensional deformation model, three after being projected
Tie up deformation model；
Submodule is handled, for the three-dimensional deformation model after the projection to be carried out 3D mesh generations, normalized simultaneously
The anchor point of three-dimensional deformation model is adjusted, obtains treated three-dimensional deformation model；
Facial trend fitting submodule for carrying out facial trend fitting to treated the three-dimensional deformation model, obtains
Three-dimensional deformation model after to fitting；
Three-dimensional feature vector acquisition submodule, for obtaining three-dimensional feature vector according to the three-dimensional deformation model after fitting；
The three-dimensional feature vector sum that the three-dimensional feature vector includes facial image to be identified tentatively identifies the three-dimensional feature vector of image.
The specific embodiment provided according to the present invention, the invention discloses following technique effects：
The invention discloses a kind of face identification method and systems, and first, convolutional neural networks are trained, utilize instruction
Convolutional neural networks after white silk identify the facial image in facial image to be identified and default database, obtain preliminary identification figure
Picture improves the precision of recognition of face；Then, facial image to be identified and the Qie Bixue of preliminary identification image are further passed through
Whether husband's Distance Judgment facial image to be identified and preliminary identification image are same person, and then judge that facial image to be identified is
It is no in default database, realization accurately identified in the unknown face of database in the database and not.
Description of the drawings
It in order to illustrate more clearly about the embodiment of the present invention or technical scheme of the prior art, below will be to institute in embodiment
Attached drawing to be used is needed to be briefly described, it should be apparent that, the accompanying drawings in the following description is only some implementations of the present invention
Example, for those of ordinary skill in the art, without having to pay creative labor, can also be according to these attached drawings
Obtain other attached drawings.
Fig. 1 is a kind of flow chart of face identification method provided by the invention；
Fig. 2 is a kind of structure diagram of face identification system provided by the invention.
Specific embodiment
The object of the present invention is to provide a kind of face identification method and systems, and database is neutralized not in database with realizing
Unknown face accurately identified.
In order to make the foregoing objectives, features and advantages of the present invention clearer and more comprehensible, it is below in conjunction with the accompanying drawings and specific real
Mode is applied to be described in further detail invention.
As shown in Figure 1, the present invention provides a kind of face identification method, the recognition methods includes the following steps：
Step 101, multiple facial images are extracted from default database, forms training set.
Step 102, according to the training set training convolutional neural networks, the convolutional neural networks after being trained.
Step 103, facial image to be identified is obtained.
Step 104, calculated using the convolutional neural networks after training in facial image to be identified and default database
Similarity between each facial image.
Step 105, the facial image of similarity maximum is chosen as preliminary identification image.
Step 106, obtain respectively facial image to be identified three-dimensional feature vector sum tentatively know the three-dimensional feature of image to
Amount.
Step 107, the three-dimensional feature vector sum for calculating the facial image to be identified tentatively identifies the three-dimensional feature of image
Chebyshev's distance between vector.
Step 108, judge whether Chebyshev's distance is more than the first predetermined threshold value, if so, the people to be identified
Face image and the image that the preliminary identification image is not same person, the facial image to be identified is not in presetting database
In, if it is not, the then facial image to be identified and the image that the preliminary identification image is same person.
Specifically, according to the training set training convolutional neural networks, before the convolutional neural networks after being trained, also
Including：
Convolutional network parameter is initialized, the convolutional network parameter includes learning rate learning_rate, convolutional Neural member
Network number of plies layer and every layer of convolution kernel size and number, training sample input in batches, and every batch of training sample number is denoted as
Batchsize, canonical index L2_penalty, maximum training algebraically maxstep, per generation comprising batch number step_size, figure
The size image_size of picture, the number people_per_batch of each batch, everyone is how many pictures images_per_
person；Specifically, learning rate learning_rate=0.1, the convolutional Neural metanetwork number of plies and every layer of convolution kernel size and
Number can directly invoke inception_resnet_v1 modules, and training sample inputs in batches, and every batch of training sample number is denoted as
Batch_size=45, canonical index L2_penalty=1e-4, maximum train algebraically maxstep=2000, per the batch in generation
Number step_size=2000, the size image_size=160 of image, the number people_per_batch=of each batch
45, everyone is how many pictures images_per_person=40.
Optionally, according to the training set training convolutional neural networks described in step 102, the convolutional Neural after being trained
Network specifically includes：
Training set is formed into triplet sets.
Triplet sets are sent into convolutional neural networks, obtain the loss function of convolutional neural networks.
The training convolutional neural networks make the value of the loss function be less than the second predetermined threshold value, the volume after being trained
Product neutral net.
Optionally, it is described that training set is formed into triplet sets, it specifically includes：
Three facial images are chosen from training set, three facial images include two and come from same person difference shape
The facial image of state, a facial image from another person form a triple.
Facial image in training set is formed triplet sets by the step of repeating to form a triple.
Optionally, the loss function is：
Wherein, i=1,2 ... ..., N represent i-th of triple, and N represents that three-number set concentrates the quantity of three-number set,It is the feature representation that same people's different conditions face two opens image in i-th of triple,It is i-th
The feature representation of the image of another people in triple, when+the value for representing in [] is more than zero, value in [] is actual value, less than zero
When, the value in [] is that zero, α isWithThe distance between andWithThe distance between minimal difference.
Optionally, the three-dimensional feature vector sum for obtaining facial image to be identified described in step 106 respectively tentatively knows image
Three-dimensional feature vector, specifically includes：
Gather the human face characteristic point of facial image to be identified and preliminary identification image.
Based on deep neural network, according to the facial image to be identified and the preliminary identification image, three-dimensional shaped is established
Varying model, the three-dimensional deformation model include the three-dimensional deformation model of facial image to be identified and the three-dimensional shaped of preliminary identification image
Varying model；The deep neural network is by face's basic configuration model (Basel Face Model), face's basic facial expression mould
Type (Face Warehouse) and non-rigid closest approach iteration three-dimensional face Registration of Measuring Data algorithm (Nonrigid ICP) form one
Kind deep neural network.
Weak perspective projection, the three-dimensional deformation model after being projected are carried out to the three-dimensional deformation model.
Three-dimensional deformation model after the projection is subjected to 3D mesh generations, normalized and adjusts three-dimensional deformation model
Anchor point, obtain treated three-dimensional deformation model.
Facial trend fitting, the three-dimensional deformation model after being fitted are carried out to treated the three-dimensional deformation model.
Three-dimensional feature vector is obtained according to the three-dimensional deformation model after fitting；The three-dimensional feature vector includes people to be identified
The three-dimensional feature vector sum of face image tentatively identifies the three-dimensional feature vector of image.
As shown in Fig. 2, the present invention also provides a kind of face identification system, the identifying system includes：
Training set acquisition module 201 for extracting multiple facial images from default database, forms training set.
Convolutional neural networks training module 202, for according to the training set training convolutional neural networks, after being trained
Convolutional neural networks.
Facial image acquisition module 203 to be identified, for obtaining facial image to be identified.
Similarity calculation module 204, for using the convolutional neural networks calculating facial image to be identified after training and in advance
If database in each facial image between similarity.
Preliminary identification image chooses module 205, for choosing the facial image of similarity maximum as preliminary identification image.
Three-dimensional feature vector acquisition module 206, at the beginning of obtaining the three-dimensional feature vector sum of facial image to be identified respectively
Step knows the three-dimensional feature vector of image.
Chebyshev's distance calculation module 207, at the beginning of calculating the three-dimensional feature vector sum of the facial image to be identified
Chebyshev's distance between the three-dimensional feature vector of step identification image.
Judgment module 208, for judging whether Chebyshev's distance is more than the first predetermined threshold value, if so, described
Facial image to be identified and the image that the preliminary identification image is not same person, the facial image to be identified is not default
In database, if it is not, the then facial image to be identified and the image that the preliminary identification image is same person.
Optionally, the convolutional neural networks training module 202, specifically includes：
Triplet sets setting up submodule, for training set to be formed triplet sets.
Loss function acquisition submodule for triplet sets to be sent into convolutional neural networks, obtains convolutional neural networks
Loss function.
Convolutional neural networks train submodule, and the value for making the loss function for training the convolutional neural networks is less than
Second predetermined threshold value, the convolutional neural networks after being trained.
Optionally, the triplet sets setting up submodule, specifically includes：
Triple establishes unit, and for choosing three facial images from training set, three facial images include two
The facial image from same person different conditions is opened, a facial image from another person forms a triple.
The step of triple combination establishes unit, a triple is formed for repetition, by the facial image in training set
Form triplet sets.
Optionally, the loss function in loss function acquisition submodule is：
Wherein, i=1,2 ... ..., N represent i-th of triple, and N represents that three-number set concentrates the quantity of three-number set,It is the feature representation that same people's different conditions face two opens image in i-th of triple,It is i-th
The feature representation of the image of another people in triple, when+the value for representing in [] is more than zero, the value of loss function is Y, less than zero
When, the value of loss function is that zero, α isWithThe distance between andWithThe distance between minimal difference.
Optionally, the three-dimensional feature vector acquisition module 206, specifically includes：
Three-dimensional deformation model foundation submodule, for the facial image to be identified and the preliminary identification image to be inputted
Deep neural network, establishes three-dimensional deformation model, and the three-dimensional deformation model includes the three-dimensional deformation mould of facial image to be identified
The three-dimensional deformation model of type and preliminary identification image.
Weak perspective projection submodule, for carrying out weak perspective projection to the three-dimensional deformation model, three after being projected
Tie up deformation model.
Submodule is handled, for the three-dimensional deformation model after the projection to be carried out 3D mesh generations, normalized simultaneously
The anchor point of three-dimensional deformation model is adjusted, obtains treated three-dimensional deformation model.
Facial trend fitting submodule for carrying out facial trend fitting to treated the three-dimensional deformation model, obtains
Three-dimensional deformation model after to fitting.
Three-dimensional feature vector acquisition submodule, for obtaining three-dimensional feature vector according to the three-dimensional deformation model after fitting；
The three-dimensional feature vector sum that the three-dimensional feature vector includes facial image to be identified tentatively identifies the three-dimensional feature vector of image.
The specific embodiment provided according to the present invention, the invention discloses following technique effects：
The invention discloses a kind of face identification method and systems, and first, convolutional neural networks are trained, utilize instruction
Convolutional neural networks after white silk identify the facial image in facial image to be identified and default database, obtain preliminary identification figure
Picture improves the precision of recognition of face；Then, facial image to be identified and the Qie Bixue of preliminary identification image are further passed through
Whether husband's Distance Judgment facial image to be identified and preliminary identification image are same person, and then judge that facial image to be identified is
It is no in default database, realization accurately identified in the unknown face of database in the database and not.
The method and system of the present invention, have the object not in storehouse a quite reliable judgment, and due to can be
Illumination condition, face posture and facial expression are adjusted in a certain range, the requirement to target acquisition reduces, and precision greatly improves.
Each embodiment is described by the way of progressive in this specification, the highlights of each of the examples are with other
The difference of embodiment, just to refer each other for identical similar portion between each embodiment.For system disclosed in embodiment
For, since it is corresponded to the methods disclosed in the examples, so description is fairly simple, related part is said referring to method part
It is bright.
Specific case used herein is set forth the principle and embodiment of invention, the explanation of above example
It is only intended to help the method and its core concept for understanding the present invention, described embodiment is only that the part of the present invention is real
Example is applied, instead of all the embodiments, based on the embodiments of the present invention, those of ordinary skill in the art are not making creation
Property work under the premise of all other embodiments obtained, belong to the scope of protection of the invention.
Claims (10)
1. a kind of face identification method, which is characterized in that the recognition methods includes the following steps：
Multiple facial images are extracted from default database, form training set；
According to the training set training convolutional neural networks, the convolutional neural networks after being trained；
Obtain facial image to be identified；
Using the convolutional neural networks after training calculate each facial image in facial image to be identified and default database it
Between similarity；
The facial image of similarity maximum is chosen as preliminary identification image；
The three-dimensional feature vector sum for obtaining facial image to be identified respectively tentatively knows the three-dimensional feature vector of image；
The three-dimensional feature vector sum for calculating the facial image to be identified tentatively identifies cutting between the three-dimensional feature vector of image
Than avenging husband's distance；
Judge whether Chebyshev distance is more than the first predetermined threshold value, if so, the facial image to be identified with it is described
Preliminary identification image is not the image of same person, and the facial image to be identified is not in presetting database, if it is not, then described
Facial image to be identified and the image that the preliminary identification image is same person.
2. face identification method according to claim 1, which is characterized in that described according to training set training convolutional god
Through network, the convolutional neural networks after being trained specifically include：
Training set is formed into triplet sets；
Triplet sets are sent into convolutional neural networks, obtain the loss function of convolutional neural networks；
The training convolutional neural networks make the value of the loss function be less than the second predetermined threshold value, the convolution god after being trained
Through network.
3. face identification method according to claim 2, which is characterized in that it is described that training set is formed into triplet sets,
It specifically includes：
Three facial images are chosen from training set, three facial images include two from same person different conditions
Facial image, a facial image from another person form a triple；
Facial image in training set is formed triplet sets by the step of repeating to form a triple.
4. face identification method according to claim 2, which is characterized in that the loss function is：
<mrow>
<mi>Y</mi>
<mo>=</mo>
<munderover>
<mo>&Sigma;</mo>
<mi>i</mi>
<mi>N</mi>
</munderover>
<msub>
<mrow>
<mo>&lsqb;</mo>
<mo>|</mo>
<mo>|</mo>
<mi>f</mi>
<mrow>
<mo>(</mo>
<msubsup>
<mi>x</mi>
<mi>i</mi>
<mi>a</mi>
</msubsup>
<mo>)</mo>
</mrow>
<mo>-</mo>
<mi>f</mi>
<mrow>
<mo>(</mo>
<msubsup>
<mi>x</mi>
<mi>i</mi>
<mi>p</mi>
</msubsup>
<mo>)</mo>
</mrow>
<mo>|</mo>
<msubsup>
<mo>|</mo>
<mn>2</mn>
<mn>2</mn>
</msubsup>
<mo>-</mo>
<mo>|</mo>
<mo>|</mo>
<mi>f</mi>
<mrow>
<mo>(</mo>
<msubsup>
<mi>x</mi>
<mi>i</mi>
<mi>a</mi>
</msubsup>
<mo>)</mo>
</mrow>
<mo>-</mo>
<mi>f</mi>
<mrow>
<mo>(</mo>
<msubsup>
<mi>x</mi>
<mi>i</mi>
<mi>n</mi>
</msubsup>
<mo>)</mo>
</mrow>
<mo>|</mo>
<msubsup>
<mo>|</mo>
<mn>2</mn>
<mn>2</mn>
</msubsup>
<mo>+</mo>
<mi>&alpha;</mi>
<mo>&rsqb;</mo>
</mrow>
<mo>+</mo>
</msub>
</mrow>
Wherein, i=1,2 ... ..., N represent i-th of triple, and N represents that three-number set concentrates the quantity of three-number set,It is the feature representation that same people's different conditions face two opens image in i-th of triple,It is i-th
The feature representation of the image of another people in triple ,+represent [] in value be more than zero when, Y be loss function value, less than zero
When, Y zero, α areWithThe distance between andWithThe distance between minimal difference.
5. face identification method according to claim 1, which is characterized in that described to obtain facial image to be identified respectively
Three-dimensional feature vector sum tentatively knows the three-dimensional feature vector of image, specifically includes：
Based on deep neural network, according to the facial image to be identified and the preliminary identification image, three-dimensional deformation mould is established
Type, the three-dimensional deformation model include the three-dimensional deformation model of facial image to be identified and the three-dimensional deformation mould of preliminary identification image
Type；
Weak perspective projection, the three-dimensional deformation model after being projected are carried out to the three-dimensional deformation model；
Three-dimensional deformation model after the projection is subjected to 3D mesh generations, normalized and the anchor for adjusting three-dimensional deformation model
Point obtains treated three-dimensional deformation model；
Facial trend fitting, the three-dimensional deformation model after being fitted are carried out to treated the three-dimensional deformation model；
Three-dimensional feature vector is obtained according to the three-dimensional deformation model after fitting；The three-dimensional feature vector includes face figure to be identified
The three-dimensional feature vector sum of picture tentatively identifies the three-dimensional feature vector of image.
6. a kind of face identification system, which is characterized in that the identifying system includes：
Training set acquisition module for extracting multiple facial images from default database, forms training set；
Convolutional neural networks training module, for according to the training set training convolutional neural networks, the convolution after being trained
Neutral net；
Facial image acquisition module to be identified, for obtaining facial image to be identified；
Similarity calculation module, for calculating facial image to be identified and default data using the convolutional neural networks after training
The similarity between each facial image in storehouse；
Preliminary identification image chooses module, for choosing the facial image of similarity maximum as preliminary identification image；
Three-dimensional feature vector acquisition module, the three-dimensional feature vector sum for obtaining facial image to be identified respectively tentatively know image
Three-dimensional feature vector；
Chebyshev's distance calculation module, for calculating the three-dimensional feature vector sum of the facial image to be identified tentatively identification figure
Chebyshev's distance between the three-dimensional feature vector of picture；
Judgment module, for judging whether Chebyshev's distance is more than the first predetermined threshold value, if so, the people to be identified
Face image and the image that the preliminary identification image is not same person, the facial image to be identified is not in presetting database
In, if it is not, the then facial image to be identified and the image that the preliminary identification image is same person.
7. face identification system according to claim 6, which is characterized in that the convolutional neural networks training module, tool
Body includes：
Triplet sets setting up submodule, for training set to be formed triplet sets；
Loss function acquisition submodule for triplet sets to be sent into convolutional neural networks, obtains the damage of convolutional neural networks
Lose function；
Convolutional neural networks train submodule, and the value for making the loss function for training the convolutional neural networks is less than second
Predetermined threshold value, the convolutional neural networks after being trained.
8. face identification system according to claim 7, which is characterized in that the triplet sets setting up submodule, tool
Body includes：
Triple establishes unit, and for choosing three facial images from training set, three facial images come including two
From the facial image of same person different conditions, a facial image from another person forms a triple；
The step of triple combination establishes unit, a triple is formed for repetition, the facial image in training set is formed
Triplet sets.
9. face identification system according to claim 7, which is characterized in that the loss letter in loss function acquisition submodule
Number is：
<mrow>
<mi>Y</mi>
<mo>=</mo>
<munderover>
<mo>&Sigma;</mo>
<mi>i</mi>
<mi>N</mi>
</munderover>
<msub>
<mrow>
<mo>&lsqb;</mo>
<mo>|</mo>
<mo>|</mo>
<mi>f</mi>
<mrow>
<mo>(</mo>
<msubsup>
<mi>x</mi>
<mi>i</mi>
<mi>a</mi>
</msubsup>
<mo>)</mo>
</mrow>
<mo>-</mo>
<mi>f</mi>
<mrow>
<mo>(</mo>
<msubsup>
<mi>x</mi>
<mi>i</mi>
<mi>p</mi>
</msubsup>
<mo>)</mo>
</mrow>
<mo>|</mo>
<msubsup>
<mo>|</mo>
<mn>2</mn>
<mn>2</mn>
</msubsup>
<mo>-</mo>
<mo>|</mo>
<mo>|</mo>
<mi>f</mi>
<mrow>
<mo>(</mo>
<msubsup>
<mi>x</mi>
<mi>i</mi>
<mi>a</mi>
</msubsup>
<mo>)</mo>
</mrow>
<mo>-</mo>
<mi>f</mi>
<mrow>
<mo>(</mo>
<msubsup>
<mi>x</mi>
<mi>i</mi>
<mi>n</mi>
</msubsup>
<mo>)</mo>
</mrow>
<mo>|</mo>
<msubsup>
<mo>|</mo>
<mn>2</mn>
<mn>2</mn>
</msubsup>
<mo>+</mo>
<mi>&alpha;</mi>
<mo>&rsqb;</mo>
</mrow>
<mo>+</mo>
</msub>
</mrow>
Wherein, i=1,2 ... ..., N represent i-th of triple, and N represents that three-number set concentrates the quantity of three-number set,It is the feature representation that same people's different conditions face two opens image in i-th of triple,It is i-th
The feature representation of the image of another people in triple, when+the value for representing in [] is more than zero, value in [] is actual value, less than zero
When, the value in [] is that zero, α isWithThe distance between andWithThe distance between minimal difference.
10. face identification system according to claim 6, which is characterized in that the three-dimensional feature vector acquisition module, tool
Body includes：
Three-dimensional deformation model foundation submodule, for the facial image to be identified and the preliminary identification image to be inputted depth
Neutral net, establishes three-dimensional deformation model, the three-dimensional deformation model include facial image to be identified three-dimensional deformation model and
The three-dimensional deformation model of preliminary identification image；
Weak perspective projection submodule, for carrying out weak perspective projection, the three-dimensional shaped after being projected to the three-dimensional deformation model
Varying model；
Submodule is handled, for the three-dimensional deformation model after the projection to be carried out 3D mesh generations, normalized and is adjusted
The anchor point of three-dimensional deformation model obtains treated three-dimensional deformation model；
Facial trend fitting submodule for carrying out facial trend fitting to treated the three-dimensional deformation model, is intended
Three-dimensional deformation model after conjunction；
Three-dimensional feature vector acquisition submodule, for obtaining three-dimensional feature vector according to the three-dimensional deformation model after fitting；It is described
The three-dimensional feature vector sum that three-dimensional feature vector includes facial image to be identified tentatively identifies that the three-dimensional feature of image is vectorial.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201711384016.7A CN108090451B (en) | 2017-12-20 | 2017-12-20 | Face recognition method and system |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201711384016.7A CN108090451B (en) | 2017-12-20 | 2017-12-20 | Face recognition method and system |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN108090451A true CN108090451A (en) | 2018-05-29 |
| CN108090451B CN108090451B (en) | 2019-12-24 |
Family
ID=62176230
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201711384016.7A Expired - Fee Related CN108090451B (en) | 2017-12-20 | 2017-12-20 | Face recognition method and system |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN108090451B (en) |
Cited By (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108921106A (en) * | 2018-07-06 | 2018-11-30 | 重庆大学 | A kind of face identification method based on capsule |
| CN109034131A (en) * | 2018-09-03 | 2018-12-18 | 福州海景科技开发有限公司 | A kind of semi-automatic face key point mask method and storage medium |
| CN109299643A (en) * | 2018-07-17 | 2019-02-01 | 深圳职业技术学院 | A kind of face identification method and system based on big attitude tracking |
| CN109558798A (en) * | 2018-10-23 | 2019-04-02 | 广东工业大学 | One kind being based on the matched face identification method of convolution characteristic pattern and system |
| CN109583332A (en) * | 2018-11-15 | 2019-04-05 | 北京三快在线科技有限公司 | Face identification method, face identification system, medium and electronic equipment |
| CN110245645A (en) * | 2019-06-21 | 2019-09-17 | 北京字节跳动网络技术有限公司 | Face vivo identification method, device, equipment and storage medium |
Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN105930832A (en) * | 2016-05-18 | 2016-09-07 | 成都芯软科技发展有限公司 | Identity recognition system and method |
| CN106022317A (en) * | 2016-06-27 | 2016-10-12 | 北京小米移动软件有限公司 | Face identification method and apparatus |
| CN106203533A (en) * | 2016-07-26 | 2016-12-07 | 厦门大学 | The degree of depth based on combined training study face verification method |
| US20170061246A1 (en) * | 2015-09-02 | 2017-03-02 | Fujitsu Limited | Training method and apparatus for neutral network for image recognition |
| CN106845330A (en) * | 2016-11-17 | 2017-06-13 | 北京品恩科技股份有限公司 | A kind of training method of the two-dimension human face identification model based on depth convolutional neural networks |
-
2017
- 2017-12-20 CN CN201711384016.7A patent/CN108090451B/en not_active Expired - Fee Related
Patent Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20170061246A1 (en) * | 2015-09-02 | 2017-03-02 | Fujitsu Limited | Training method and apparatus for neutral network for image recognition |
| CN105930832A (en) * | 2016-05-18 | 2016-09-07 | 成都芯软科技发展有限公司 | Identity recognition system and method |
| CN106022317A (en) * | 2016-06-27 | 2016-10-12 | 北京小米移动软件有限公司 | Face identification method and apparatus |
| CN106203533A (en) * | 2016-07-26 | 2016-12-07 | 厦门大学 | The degree of depth based on combined training study face verification method |
| CN106845330A (en) * | 2016-11-17 | 2017-06-13 | 北京品恩科技股份有限公司 | A kind of training method of the two-dimension human face identification model based on depth convolutional neural networks |
Cited By (10)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108921106A (en) * | 2018-07-06 | 2018-11-30 | 重庆大学 | A kind of face identification method based on capsule |
| CN108921106B (en) * | 2018-07-06 | 2021-07-06 | 重庆大学 | Capsule-based face recognition method |
| CN109299643A (en) * | 2018-07-17 | 2019-02-01 | 深圳职业技术学院 | A kind of face identification method and system based on big attitude tracking |
| CN109299643B (en) * | 2018-07-17 | 2020-04-14 | 深圳职业技术学院 | Face recognition method and system based on large-posture alignment |
| CN109034131A (en) * | 2018-09-03 | 2018-12-18 | 福州海景科技开发有限公司 | A kind of semi-automatic face key point mask method and storage medium |
| CN109034131B (en) * | 2018-09-03 | 2021-10-26 | 福建海景科技开发有限公司 | Semi-automatic face key point marking method and storage medium |
| CN109558798A (en) * | 2018-10-23 | 2019-04-02 | 广东工业大学 | One kind being based on the matched face identification method of convolution characteristic pattern and system |
| CN109583332A (en) * | 2018-11-15 | 2019-04-05 | 北京三快在线科技有限公司 | Face identification method, face identification system, medium and electronic equipment |
| CN109583332B (en) * | 2018-11-15 | 2021-07-27 | 北京三快在线科技有限公司 | Face recognition method, face recognition system, medium, and electronic device |
| CN110245645A (en) * | 2019-06-21 | 2019-09-17 | 北京字节跳动网络技术有限公司 | Face vivo identification method, device, equipment and storage medium |
Also Published As
| Publication number | Publication date |
|---|---|
| CN108090451B (en) | 2019-12-24 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN108090451A (en) | A kind of face identification method and system | |
| CN108334816B (en) | Multi-pose face recognition method based on contour symmetric constraint generation type countermeasure network | |
| CN106127170B (en) | A kind of training method, recognition methods and system merging key feature points | |
| CN107403168B (en) | Face recognition system | |
| CN108038420B (en) | Human behavior recognition method based on depth video | |
| CN108764041B (en) | Face recognition method for lower shielding face image | |
| CN105956560A (en) | Vehicle model identification method based on pooling multi-scale depth convolution characteristics | |
| CN110738161A (en) | face image correction method based on improved generation type confrontation network | |
| CN107944435A (en) | A kind of three-dimensional face identification method, device and processing terminal | |
| CN107808129A (en) | A kind of facial multi-characteristic points localization method based on single convolutional neural networks | |
| CN103810490A (en) | Method and device for confirming attribute of face image | |
| CN104615996B (en) | A kind of various visual angles two-dimension human face automatic positioning method for characteristic point | |
| CN106446862A (en) | Face detection method and system | |
| CN106778785A (en) | Build the method for image characteristics extraction model and method, the device of image recognition | |
| CN110135277B (en) | Human behavior recognition method based on convolutional neural network | |
| CN112613097A (en) | BIM rapid modeling method based on computer vision | |
| CN112686191B (en) | Living body anti-counterfeiting method, system, terminal and medium based on three-dimensional information of human face | |
| CN105631852A (en) | Depth image contour line-based indoor human body detection method | |
| CN109784297A (en) | A kind of Three-dimensional target recognition based on deep learning and Optimal Grasp method | |
| CN104598872A (en) | Face comparison method, apparatus and face recognition method, system | |
| CN105956570A (en) | Lip characteristic and deep learning based smiling face recognition method | |
| CN103544478A (en) | All-dimensional face detection method and system | |
| CN110163175A (en) | A kind of gait recognition method and system based on improvement VGG-16 network | |
| CN106909874B (en) | A kind of nose localization method and device | |
| CN113705466B (en) | Face five sense organ shielding detection method for shielding scene, especially under high imitation shielding |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant | ||
| CF01 | Termination of patent right due to non-payment of annual fee | ||
| CF01 | Termination of patent right due to non-payment of annual fee |
Granted publication date: 20191224 Termination date: 20201220 |