CN107066941A - A kind of face identification method and system - Google Patents
A kind of face identification method and system Download PDFInfo
- Publication number
- CN107066941A CN107066941A CN201710117801.XA CN201710117801A CN107066941A CN 107066941 A CN107066941 A CN 107066941A CN 201710117801 A CN201710117801 A CN 201710117801A CN 107066941 A CN107066941 A CN 107066941A
- Authority
- CN
- China
- Prior art keywords
- mtd
- face
- msub
- mtr
- layer
- 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
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V40/00—Recognition of biometric, human-related or animal-related patterns in image or video data
- G06V40/10—Human or animal bodies, e.g. vehicle occupants or pedestrians; Body parts, e.g. hands
- G06V40/16—Human faces, e.g. facial parts, sketches or expressions
- G06V40/168—Feature extraction; Face representation
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
-
- 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
- G06N3/06—Physical realisation, i.e. hardware implementation of neural networks, neurons or parts of neurons
- G06N3/061—Physical realisation, i.e. hardware implementation of neural networks, neurons or parts of neurons using biological neurons, e.g. biological neurons connected to an integrated circuit
-
- 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
- G06N3/08—Learning methods
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V40/00—Recognition of biometric, human-related or animal-related patterns in image or video data
- G06V40/10—Human or animal bodies, e.g. vehicle occupants or pedestrians; Body parts, e.g. hands
- G06V40/16—Human faces, e.g. facial parts, sketches or expressions
- G06V40/161—Detection; Localisation; Normalisation
- G06V40/166—Detection; Localisation; Normalisation using acquisition arrangements
Abstract
The present invention provides a kind of face identification method and system, and its method includes:Registration process is carried out to the face test image in the face sample image and test library in Sample Storehouse respectively;Multilayer feature fusion convolutional neural networks are built, merge convolutional neural networks by the multilayer feature carries out face characteristic recognition training to the face sample image after alignment, the multilayer feature fusion convolutional neural networks after being trained;Recognition of face is carried out in the multilayer feature fusion convolutional neural networks that face test image input after alignment is completed to training, so as to identify the face in face test image.The present invention can merge convolutional neural networks by multilayer feature and extract face shallow-layer feature, then merge to carry out recognition of face with further feature by it, all significantly lifted in terms of identification accuracy and reliability.
Description
Technical field
The invention mainly relates to image identification technical field, and in particular to a kind of face identification method and system.
Background technology
Recognition of face is as a kind of one of important biological identification technology, with cost is low, user is easily accepted by, reliability high
The advantages of, it is with a wide range of applications in fields such as authentication, safety monitoring, man-machine interactions.Convolutional neural networks are one
It is individual inspired by biological vision, multilayer perceptron for the purpose of most simplified pretreatment operation is deformed, and is a kind of efficient identification
Method, was applied to by deep learning and progressively every field, convolutional neural networks are entered by successively convolution dimensionality reduction in recent years
Row feature extraction, then maps by multilayered nonlinear, allows network according to the characteristics of training sample, and automatic study forms suitable
Close the feature extractor and grader of the identification mission.
With the development of society, the progress of technology, influence of the science and technology to people is increasing, people are for face detection
Demand is more and more stronger, but under different scenes, angle is different, illumination under resolution of video camera difference and different scenes
Inconsistent the problems such as, same pedestrian is caused to be easy to be mistaken for same a group traveling together.Traditional feature extraction can not be extracted
Enough finer characteristic informations.
The content of the invention
The technical problems to be solved by the invention are
System.
The technical scheme that the present invention solves above-mentioned technical problem is as follows:A kind of face identification method, comprises the following steps:
Step S1:The face test image in the face sample image and test library in Sample Storehouse is carried out at alignment respectively
Reason;
Step S2:Multilayer feature fusion convolutional neural networks are built, convolutional neural networks are merged by the multilayer feature
Face characteristic recognition training is carried out to the face sample image after alignment, the multilayer feature fusion convolutional Neural net after being trained
Network;
Step S3:In the multilayer feature fusion convolutional neural networks that face test image input after alignment is completed to training
Recognition of face is carried out, so as to identify the face in face test image.
The beneficial effects of the invention are as follows:Convolutional neural networks can be merged by multilayer feature and extract face shallow-layer feature,
Then it is merged to carry out recognition of face with further feature, all obtained largely in terms of identification accuracy and reliability
Lifting, the face test image in pending face sample image and test library is alignd by landmark methods
Processing, can accelerate to recognize the process of face characteristic.
On the basis of above-mentioned technical proposal, the present invention can also do following improvement.
Further, the step S1 is specifically included:
Step S101:Using regression tree set Ensemble of Regression Trees algorithms to face sample image
Face datection is carried out with face test image, then human face characteristic point Landmarks detections are carried out to the face detected, is obtained respectively
Much the Landmark points of individual face sample image and the Landmark points of multiple face test images;
Step S102:When alignment face sample image, selected at random from the Landmark points of multiple face sample images
Default n Landmark points and the Landmark points of default standard faces template are taken, the affine square of structure is substituted into the lump
In battle array H, new affine matrix is obtained by conversion, align face sample image according to the transformed value of new affine matrix;
Step S103:When alignment face test image, selected at random from the Landmark points of multiple face test images
Default n Landmark points and the Landmark points of default standard faces template are taken, the affine square of structure is substituted into the lump
In battle array H, new affine matrix is obtained by conversion, face test image of being alignd according to the transformed value of new affine matrix.
Further, the affine matrix H is
If the coordinate of each Landmark points of affine matrix H is (x, y), after affine matrix H conversion, obtain each
The new coordinate (x', y') of Landmark points
Beneficial effect using above-mentioned further scheme is:By landmark methods to face sample image and test library
In face test image pre-processed, can make multilayer feature merge convolutional neural networks more rapidly more accurately recognize people
Face feature.
Further, the multilayer feature fusion convolutional neural networks include convolutional layer Conv1, the pond layer being sequentially connected
Pool1, convolutional layer Conv2, pond layer Pool2, convolutional layer Conv3, pond layer Pool3 and convolutional layer Conv4;Also include and institute
State convolutional layer Conv6 and convolutional layer Conv7 that pond layer Pool1 is sequentially connected, the convolutional layer being connected with the pond layer Pool2
Conv5;The convolutional layer Conv4 is connected by full articulamentum Fc1 with articulamentum Concat, and the pond layer Pool3 passes through complete
Articulamentum Fc2 is connected with articulamentum Concat, and the convolutional layer Conv5 is connected by full articulamentum Fc3 with articulamentum Concat,
The convolutional layer Conv7 is connected by full articulamentum Fc4 with articulamentum Concat;Also include classification layer softmax, the classification
Layer softmax is connected by full articulamentum Fc5 with articulamentum Concat.
Further, the convolutional layer Conv carries out convolution algorithm to the image of input, and passes through neuron activation functions meter
Calculate the output of convolution;
The image that the pond layer Pool is exported to convolutional layer carries out down-sampling to reduce the size of image;
All neuron nodes of the full articulamentum Fc by any one neuron node of last layer all with next layer
Connection;
The articulamentum Concat is by full articulamentum Fc1, full articulamentum Fc2, full articulamentum Fc3 and full articulamentum Fc4
The neuron node of output carries out Fusion Features;
The classification layer softmax carries out tagsort to the input value of each neuron node.
Beneficial effect using above-mentioned further scheme is:Compared with traditional face identification method, using multilayer feature
The method of fusion convolutional neural networks embodies more preferable recognition accuracy, and the number of plies of convolutional neural networks is deeper, its mark sheet
It is better up to power, it is contemplated that the output of the detail textures of shallow-layer feature, shallow-layer feature and further feature are blended, make identification accurate
Property and reliability in terms of all significantly lifted.
Another technical scheme that the present invention solves above-mentioned technical problem is as follows:A kind of face identification system, including:
Image alignment module, for respectively to the face test image in the face sample image and test library in Sample Storehouse
Carry out registration process;
Training module, for building multilayer feature fusion convolutional neural networks, convolution god is merged by the multilayer feature
Face characteristic recognition training is carried out to the face sample image after alignment through network, the multilayer feature fusion convolution after being trained
Neutral net;
Identification module, the multilayer feature for the face test image input after alignment to be completed to training merges convolutional Neural
Recognition of face is carried out in network, so as to identify the face in face test image.
On the basis of above-mentioned technical proposal, the present invention can also do following improvement.
Further, described image alignment module includes:
Landmark point detection units, for utilizing regression tree set Ensemble of Regression Trees algorithms
Face datection is carried out to face sample image and face test image, then human face characteristic point is carried out to the face detected
Landmarks is detected, the Landmark points and multiple face test images of multiple face sample images are obtained respectively
Landmark points;
Sample image alignment unit, for when align face sample image when, at random from multiple face sample images
Default n Landmark points and the Landmark points of default standard faces template are chosen in Landmark points, in the lump generation
In the affine matrix H for entering structure, new affine matrix is obtained by conversion, align face sample according to the transformed value of new affine matrix
This image;
Test image alignment unit, for when alignment face test image, at random from multiple face test images
Default n Landmark points and the Landmark points of default standard faces template are chosen in Landmark points, in the lump generation
In the affine matrix H for entering structure, new affine matrix is obtained by conversion, the face that alignd according to the transformed value of new affine matrix is surveyed
Attempt picture.
Further, the affine matrix H is
If the coordinate of each Landmark points of affine matrix H is (x, y), after affine matrix H conversion, obtain each
The new coordinate (x', y') of Landmark points
Further, the multilayer feature fusion convolutional neural networks include convolutional layer Conv1, the pond layer being sequentially connected
Pool1, convolutional layer Conv2, pond layer Pool2, convolutional layer Conv3, pond layer Pool3 and convolutional layer Conv4;Also include and institute
State convolutional layer Conv6 and convolutional layer Conv7 that pond layer Pool1 is sequentially connected, the convolutional layer being connected with the pond layer Pool2
Conv5;The convolutional layer Conv4 is connected by full articulamentum Fc1 with articulamentum Concat, and the pond layer Pool3 passes through complete
Articulamentum Fc2 is connected with articulamentum Concat, and the convolutional layer Conv5 is connected by full articulamentum Fc3 with articulamentum Concat,
The convolutional layer Conv7 is connected by full articulamentum Fc4 with articulamentum Concat;Also include classification layer softmax, the classification
Layer softmax is connected by full articulamentum Fc5 with articulamentum Concat.
Further, the convolutional layer Conv, carries out convolution algorithm, and pass through neuronal activation letter for the image to input
Number calculates the output of convolution;
The pond layer Pool, the image for being exported to convolutional layer carries out down-sampling to reduce the size of image;
The full articulamentum Fc, for all neurons by any one neuron node of last layer all with next layer
Node is connected;
The articulamentum Concat, for by full articulamentum Fc1, full articulamentum Fc2, full articulamentum Fc3 and full articulamentum
The neuron node exported in Fc4 carries out Fusion Features;
The classification layer softmax, tagsort is carried out for the input value to each neuron node.
Brief description of the drawings
Fig. 1 is the method flow diagram of face identification method provided in an embodiment of the present invention;
Fig. 2 is the module frame chart of face identification system provided in an embodiment of the present invention;
Fig. 3 is the structure chart that multilayer feature provided in an embodiment of the present invention merges convolutional neural networks.
Embodiment
The principle and feature of the present invention are described below in conjunction with accompanying drawing, the given examples are served only to explain the present invention, and
It is non-to be used to limit the scope of the present invention.
Fig. 1 is the method flow diagram of face identification method provided in an embodiment of the present invention;
As shown in figure 1, a kind of face identification method, comprises the following steps:
Step S1:The face test image in the face sample image and test library in Sample Storehouse is carried out at alignment respectively
Reason;
Step S2:Multilayer feature fusion convolutional neural networks are built, convolutional neural networks are merged by the multilayer feature
Face characteristic recognition training is carried out to the face sample image after alignment, the multilayer feature fusion convolutional Neural net after being trained
Network;
Step S3:In the multilayer feature fusion convolutional neural networks that face test image input after alignment is completed to training
Recognition of face is carried out, so as to identify the face in face test image.
Specifically, test library can disclose storehouse as face test image sample set using CASIA.
In above-mentioned implementation, convolutional neural networks can be merged by multilayer feature and extract face shallow-layer feature, then by it
Merge to carry out recognition of face with further feature, all significantly lifted in terms of identification accuracy and reliability,
Face test image in pending face sample image and test library is carried out by registration process, energy by landmark methods
Enough processes for accelerating identification face characteristic.
Alternatively, as one embodiment of the present of invention, the step S1 is specifically included:
Step S101:Using regression tree set Ensemble of Regression Trees algorithms to face sample image
Face datection is carried out with face test image, the face detected is entered further according to human face characteristic point Landmarks detection methods
Row feature point detection, obtains the Landmark points of multiple face sample images and the Landmark of multiple face test images respectively
Point;
Step S102:When alignment face sample image, selected at random from the Landmark points of multiple face sample images
Default n Landmark points and the Landmark points of default standard faces template are taken, the affine square of structure is substituted into the lump
In battle array H, new affine matrix is obtained by conversion, align face sample image according to the transformed value of new affine matrix;Wherein n >=
3;
Step S103:When alignment face test image, selected at random from the Landmark points of multiple face test images
Default n Landmark points and the Landmark points of default standard faces template are taken, the affine square of structure is substituted into the lump
In battle array H, new affine matrix is obtained by conversion, face test image of being alignd according to the transformed value of new affine matrix.Alternatively,
As one embodiment of the present of invention, the affine matrix H is
If the coordinate of each Landmark points of affine matrix H is (x, y), after affine matrix H conversion, obtain each
The new coordinate (x', y') of Landmark points
In above-described embodiment, the face test image in face sample image and test library is entered by landmark methods
Row pretreatment, can make multilayer feature merge convolutional neural networks and more rapidly more accurately recognize face characteristic.
Fig. 3 is the structure chart that multilayer feature provided in an embodiment of the present invention merges convolutional neural networks;
Alternatively, as one embodiment of the present of invention, as shown in figure 3, the multilayer feature merges convolutional neural networks
Including the convolutional layer Conv1 being sequentially connected, pond layer Pool1, convolutional layer Conv2, pond layer Pool2, convolutional layer Conv3, pond
Change layer Pool3 and convolutional layer Conv4;Also include the convolutional layer Conv6 and convolutional layer being sequentially connected with the pond layer Pool1
Conv7, the convolutional layer Conv5 being connected with the pond layer Pool2;The convolutional layer Conv4 is by full articulamentum Fc1 with being connected
Layer Concat connections, the pond layer Pool3 is connected by full articulamentum Fc2 with articulamentum Concat, the convolutional layer Conv5
It is connected by full articulamentum Fc3 with articulamentum Concat, the convolutional layer Conv7 passes through full articulamentum Fc4 and articulamentum
Concat connections;Also include classification layer softmax, the classification layer softmax and pass through full articulamentum Fc5 and articulamentum Concat
Connection.
Alternatively, as one embodiment of the present of invention, the convolutional layer Conv carries out convolution algorithm to the image of input,
And the output of convolution is calculated by neuron activation functions;
The image that the pond layer Pool is exported to convolutional layer carries out down-sampling to reduce the size of image;
The full articulamentum Fc, for all neurons by any one neuron node of last layer all with next layer
Node is connected;
The articulamentum Concat is by full articulamentum Fc1, full articulamentum Fc2, full articulamentum Fc3 and full articulamentum Fc4
The neuron node of output carries out Fusion Features;
The classification layer softmax carries out tagsort to the input value of each neuron node.
In above-described embodiment, compared with traditional face identification method, convolutional neural networks are merged using multilayer feature
Method embodies more preferable recognition accuracy, and the number of plies of convolutional neural networks is deeper, and its feature representation power is better, it is contemplated that shallow-layer
The output of the detail textures of feature, shallow-layer feature and further feature are blended, and make identification accuracy and reliability aspect all
Significantly lifted.
Alternatively, as one embodiment of the present of invention, according to mini-batch gradient descent algorithms in the step S2
Multilayer feature fusion convolutional neural networks are restrained, to improve the recognition correct rate to facial image feature;Specifically, false
If face test image training set has m sample, each mini-batch (a subset of training set) has b sample, then,
Whole training set is segmented into m/b mini-batch, and a mini-batch is represented with ω, is represented with Ω j in jth wheel iteration
All mini-batch set, have:Ω={ ωk:K=1,2...m/b }.
Wherein, mini-batch GD algorithm flows are as follows:
repeate{
repente{
foreachωkinΩ:
For (k=1,2...m/b)
}
Fig. 2 is the module frame chart of face identification system provided in an embodiment of the present invention;
Alternatively, as one embodiment of the present of invention, as shown in Fig. 2 a kind of face identification system, including:
Image alignment module, for respectively to the face test image in the face sample image and test library in Sample Storehouse
Carry out registration process;
Training module, for building multilayer feature fusion convolutional neural networks, convolution god is merged by the multilayer feature
Face characteristic recognition training is carried out to the face sample image after alignment through network, the multilayer feature fusion convolution after being trained
Neutral net;
Identification module, the multilayer feature for the face test image input after alignment to be completed to training merges convolutional Neural
Recognition of face is carried out in network, so as to identify the face in face test image.
Alternatively, as one embodiment of the present of invention, described image alignment module includes:
Landmark point detection units, for utilizing regression tree set Ensemble of Regression Trees algorithms
Face datection is carried out to face sample image and face test image, further according to facial feature points detection Landmarks methods pair
The face detected carries out feature point detection, and Landmark points and the test of multiple faces of multiple face sample images are obtained respectively
The Landmark points of image;
Sample image alignment unit, for when align face sample image when, at random from multiple face sample images
Default n Landmark points and the Landmark points of default standard faces template are chosen in Landmark points, in the lump generation
In the affine matrix H for entering structure, new affine matrix is obtained by conversion, align face sample according to the transformed value of new affine matrix
This image;Wherein n >=3;
Test image alignment unit, for when alignment face test image, at random from multiple face test images
Default n Landmark points and the Landmark points of default standard faces template are chosen in Landmark points, in the lump generation
In the affine matrix H for entering structure, new affine matrix is obtained by conversion, the face that alignd according to the transformed value of new affine matrix is surveyed
Attempt picture.
Alternatively, as one embodiment of the present of invention, the affine matrix H is
If the coordinate of each Landmark points of affine matrix H is (x, y), after affine matrix H conversion, obtain each
The new coordinate (x', y') of Landmark points
Alternatively, as one embodiment of the present of invention, the multilayer feature fusion convolutional neural networks include connecting successively
The convolutional layer Conv1 that connects, pond layer Pool1, convolutional layer Conv2, pond layer Pool2, convolutional layer Conv3, pond layer Pool3 and
Convolutional layer Conv4;Also include the convolutional layer Conv6 and convolutional layer Conv7 that are sequentially connected with the pond layer Pool1, it is and described
The convolutional layer Conv5 of pond layer Pool2 connections;The convolutional layer Conv4 is connected by full articulamentum Fc1 and articulamentum Concat
Connect, the pond layer Pool3 is connected by full articulamentum Fc2 with articulamentum Concat, the convolutional layer Conv5 by connecting entirely
Layer Fc3 is connected with articulamentum Concat, and the convolutional layer Conv7 is connected by full articulamentum Fc4 with articulamentum Concat;Also wrap
Classification layer softmax is included, the classification layer softmax is connected by full articulamentum Fc5 with articulamentum Concat.
Alternatively, as one embodiment of the present of invention, the convolutional layer Conv carries out convolution for the image to input
Computing, and pass through the output of neuron activation functions calculating convolution;
The pond layer Pool, the image for being exported to convolutional layer carries out down-sampling to reduce the size of image;
The full articulamentum Fc, for all neurons by any one neuron node of last layer all with next layer
Node is connected;
The articulamentum Concat, for by full articulamentum Fc1, full articulamentum Fc2, full articulamentum Fc3 and full articulamentum
The neuron node exported in Fc4 carries out Fusion Features;
The classification layer softmax, tagsort is carried out for the input value to each neuron node.
The method embodiment of the invention for merging convolutional neural networks using multilayer feature compared with traditional face identification method
Go out more preferable recognition accuracy, the number of plies of convolutional neural networks is deeper, its feature representation power is better, it is contemplated that shallow-layer feature it is thin
The output of texture is saved, shallow-layer feature and further feature are blended, makes all to obtain very big in terms of identification accuracy and reliability
The lifting of degree.
The foregoing is only presently preferred embodiments of the present invention, be not intended to limit the invention, it is all the present invention spirit and
Within principle, any modification, equivalent substitution and improvements made etc. should be included in the scope of the protection.
Claims (10)
1. a kind of face identification method, it is characterised in that comprise the following steps:
Step S1:Registration process is carried out to the face test image in the face sample image and test library in Sample Storehouse respectively;
Step S2:Multilayer feature fusion convolutional neural networks are built, convolutional neural networks are merged to right by the multilayer feature
Face sample image after neat carries out face characteristic recognition training, the multilayer feature fusion convolutional neural networks after being trained;
Step S3:Carried out in the multilayer feature fusion convolutional neural networks that face test image input after alignment is completed to training
Recognition of face, so as to identify the face in face test image.
2. a kind of face identification method according to claim 1, it is characterised in that the step S1 is specifically included:
Step S101:Using regression tree set Ensemble of Regression Trees algorithms to face sample image and people
Face test image carries out Face datection, the face detected is carried out further according to facial feature points detection Landmarks methods special
A detection is levied, the Landmark points of multiple face sample images and the Landmark points of multiple face test images are obtained respectively;
Step S102:When alignment face sample image, choose pre- from the Landmark points of multiple face sample images at random
If n Landmark point and default standard faces template Landmark points, the affine matrix H of structure is substituted into the lump
In, new affine matrix is obtained by conversion, align face sample image according to the transformed value of new affine matrix;
Step S103:When alignment face test image, choose pre- from the Landmark points of multiple face test images at random
If n Landmark point and default standard faces template Landmark points, the affine matrix H of structure is substituted into the lump
In, new affine matrix is obtained by conversion, face test image of being alignd according to the transformed value of new affine matrix.
3. a kind of face identification method according to claim 2, it is characterised in that the affine matrix H is
<mrow>
<mi>H</mi>
<mo>=</mo>
<mfenced open = "[" close = "]">
<mtable>
<mtr>
<mtd>
<msub>
<mi>m</mi>
<mn>00</mn>
</msub>
</mtd>
<mtd>
<msub>
<mi>m</mi>
<mn>01</mn>
</msub>
</mtd>
<mtd>
<msub>
<mi>m</mi>
<mn>02</mn>
</msub>
</mtd>
</mtr>
<mtr>
<mtd>
<msub>
<mi>m</mi>
<mn>10</mn>
</msub>
</mtd>
<mtd>
<msub>
<mi>m</mi>
<mn>11</mn>
</msub>
</mtd>
<mtd>
<msub>
<mi>m</mi>
<mn>12</mn>
</msub>
</mtd>
</mtr>
<mtr>
<mtd>
<mn>0</mn>
</mtd>
<mtd>
<mn>0</mn>
</mtd>
<mtd>
<mn>1</mn>
</mtd>
</mtr>
</mtable>
</mfenced>
</mrow>
If the coordinate of each Landmark points of affine matrix H is (x, y), after affine matrix H conversion, each Landmark points are obtained
New coordinate (x', y')
<mrow>
<mfenced open = "[" close = "]">
<mtable>
<mtr>
<mtd>
<msup>
<mi>x</mi>
<mo>&prime;</mo>
</msup>
</mtd>
</mtr>
<mtr>
<mtd>
<msup>
<mi>y</mi>
<mo>&prime;</mo>
</msup>
</mtd>
</mtr>
<mtr>
<mtd>
<mn>1</mn>
</mtd>
</mtr>
</mtable>
</mfenced>
<mo>=</mo>
<mfenced open = "[" close = "]">
<mtable>
<mtr>
<mtd>
<msub>
<mi>m</mi>
<mn>00</mn>
</msub>
</mtd>
<mtd>
<msub>
<mi>m</mi>
<mn>01</mn>
</msub>
</mtd>
<mtd>
<msub>
<mi>m</mi>
<mn>02</mn>
</msub>
</mtd>
</mtr>
<mtr>
<mtd>
<msub>
<mi>m</mi>
<mn>10</mn>
</msub>
</mtd>
<mtd>
<msub>
<mi>m</mi>
<mn>11</mn>
</msub>
</mtd>
<mtd>
<msub>
<mi>m</mi>
<mn>12</mn>
</msub>
</mtd>
</mtr>
<mtr>
<mtd>
<mn>0</mn>
</mtd>
<mtd>
<mn>0</mn>
</mtd>
<mtd>
<mn>1</mn>
</mtd>
</mtr>
</mtable>
</mfenced>
<mfenced open = "[" close = "]">
<mtable>
<mtr>
<mtd>
<mi>x</mi>
</mtd>
</mtr>
<mtr>
<mtd>
<mi>y</mi>
</mtd>
</mtr>
<mtr>
<mtd>
<mn>1</mn>
</mtd>
</mtr>
</mtable>
</mfenced>
<mo>.</mo>
</mrow>
4. a kind of face identification method according to claim 1, it is characterised in that the multilayer feature merges convolutional Neural net
Network include be sequentially connected convolutional layer Conv1, pond layer Pool1, convolutional layer Conv2, pond layer Pool2, convolutional layer Conv3,
Pond layer Pool3 and convolutional layer Conv4;Also include the convolutional layer Conv6 and convolutional layer being sequentially connected with the pond layer Pool1
Conv7, the convolutional layer Conv5 being connected with the pond layer Pool2;The convolutional layer Conv4 is by full articulamentum Fc1 with being connected
Layer Concat connections, the pond layer Pool3 is connected by full articulamentum Fc2 with articulamentum Concat, the convolutional layer Conv5
It is connected by full articulamentum Fc3 with articulamentum Concat, the convolutional layer Conv7 passes through full articulamentum Fc4 and articulamentum
Concat connections;Also include classification layer softmax, the classification layer softmax and pass through full articulamentum Fc5 and articulamentum Concat
Connection.
5. a kind of face identification method according to claim 4, it is characterised in that
The convolutional layer Conv carries out convolution algorithm to the image of input, and calculates the defeated of convolution by neuron activation functions
Go out;
The image that the pond layer Pool is exported to convolutional layer carries out down-sampling to reduce the size of image;
All neuron nodes of the full articulamentum Fc by any one neuron node of last layer all with next layer are connected;
The articulamentum Concat will be exported in full articulamentum Fc1, full articulamentum Fc2, full articulamentum Fc3 and full articulamentum Fc4
Neuron node carry out Fusion Features;
The classification layer softmax carries out tagsort to the input value of each neuron node.
6. a kind of face identification system, it is characterised in that including:
Image alignment module, for being carried out respectively to the face test image in the face sample image and test library in Sample Storehouse
Registration process;
Training module, for building multilayer feature fusion convolutional neural networks, convolutional Neural net is merged by the multilayer feature
Network carries out face characteristic recognition training to the face sample image after alignment, the multilayer feature fusion convolutional Neural after being trained
Network;
Identification module, the multilayer feature for the face test image input after alignment to be completed to training merges convolutional neural networks
Middle carry out recognition of face, so as to identify the face in face test image.
7. a kind of face identification system according to claim 6, it is characterised in that described image alignment module includes:
Landmark point detection units, for utilizing regression tree set Ensemble of Regression Trees algorithms to people
Face sample image and face test image carry out Face datection, further according to facial feature points detection Landmarks methods to detection
The face arrived carries out feature point detection, and the Landmark points and multiple face test images of multiple face sample images are obtained respectively
Landmark points;
Sample image alignment unit, for when align face sample image when, at random from multiple face sample images
Default n Landmark points and the Landmark points of default standard faces template are chosen in Landmark points, in the lump generation
In the affine matrix H for entering structure, new affine matrix is obtained by conversion, align face sample according to the transformed value of new affine matrix
This image;
Test image alignment unit, for when alignment face test image, at random from multiple face test images
Default n Landmark points and the Landmark points of default standard faces template are chosen in Landmark points, in the lump generation
In the affine matrix H for entering structure, new affine matrix is obtained by conversion, the face that alignd according to the transformed value of new affine matrix is surveyed
Attempt picture.
8. a kind of face identification system according to claim 7, it is characterised in that the affine matrix H is
<mrow>
<mi>H</mi>
<mo>=</mo>
<mfenced open = "[" close = "]">
<mtable>
<mtr>
<mtd>
<msub>
<mi>m</mi>
<mn>00</mn>
</msub>
</mtd>
<mtd>
<msub>
<mi>m</mi>
<mn>01</mn>
</msub>
</mtd>
<mtd>
<msub>
<mi>m</mi>
<mn>02</mn>
</msub>
</mtd>
</mtr>
<mtr>
<mtd>
<msub>
<mi>m</mi>
<mn>10</mn>
</msub>
</mtd>
<mtd>
<msub>
<mi>m</mi>
<mn>11</mn>
</msub>
</mtd>
<mtd>
<msub>
<mi>m</mi>
<mn>12</mn>
</msub>
</mtd>
</mtr>
<mtr>
<mtd>
<mn>0</mn>
</mtd>
<mtd>
<mn>0</mn>
</mtd>
<mtd>
<mn>1</mn>
</mtd>
</mtr>
</mtable>
</mfenced>
</mrow>
If the coordinate of each Landmark points of affine matrix H is (x, y), after affine matrix H conversion, each Landmark points are obtained
New coordinate (x', y')
<mrow>
<mfenced open = "[" close = "]">
<mtable>
<mtr>
<mtd>
<msup>
<mi>x</mi>
<mo>&prime;</mo>
</msup>
</mtd>
</mtr>
<mtr>
<mtd>
<msup>
<mi>y</mi>
<mo>&prime;</mo>
</msup>
</mtd>
</mtr>
<mtr>
<mtd>
<mn>1</mn>
</mtd>
</mtr>
</mtable>
</mfenced>
<mo>=</mo>
<mfenced open = "[" close = "]">
<mtable>
<mtr>
<mtd>
<msub>
<mi>m</mi>
<mn>00</mn>
</msub>
</mtd>
<mtd>
<msub>
<mi>m</mi>
<mn>01</mn>
</msub>
</mtd>
<mtd>
<msub>
<mi>m</mi>
<mn>02</mn>
</msub>
</mtd>
</mtr>
<mtr>
<mtd>
<msub>
<mi>m</mi>
<mn>10</mn>
</msub>
</mtd>
<mtd>
<msub>
<mi>m</mi>
<mn>11</mn>
</msub>
</mtd>
<mtd>
<msub>
<mi>m</mi>
<mn>12</mn>
</msub>
</mtd>
</mtr>
<mtr>
<mtd>
<mn>0</mn>
</mtd>
<mtd>
<mn>0</mn>
</mtd>
<mtd>
<mn>1</mn>
</mtd>
</mtr>
</mtable>
</mfenced>
<mfenced open = "[" close = "]">
<mtable>
<mtr>
<mtd>
<mi>x</mi>
</mtd>
</mtr>
<mtr>
<mtd>
<mi>y</mi>
</mtd>
</mtr>
<mtr>
<mtd>
<mn>1</mn>
</mtd>
</mtr>
</mtable>
</mfenced>
<mo>.</mo>
</mrow>
9. a kind of face identification system according to claim 6, it is characterised in that the multilayer feature merges convolutional Neural net
Network include be sequentially connected convolutional layer Conv1, pond layer Pool1, convolutional layer Conv2, pond layer Pool2, convolutional layer Conv3,
Pond layer Pool3 and convolutional layer Conv4;Also include the convolutional layer Conv6 and convolutional layer being sequentially connected with the pond layer Pool1
Conv7, the convolutional layer Conv5 being connected with the pond layer Pool2;The convolutional layer Conv4 is by full articulamentum Fc1 with being connected
Layer Concat connections, the pond layer Pool3 is connected by full articulamentum Fc2 with articulamentum Concat, the convolutional layer Conv5
It is connected by full articulamentum Fc3 with articulamentum Concat, the convolutional layer Conv7 passes through full articulamentum Fc4 and articulamentum
Concat connections;Also include classification layer softmax, the classification layer softmax and pass through full articulamentum Fc5 and articulamentum Concat
Connection.
10. a kind of face identification system according to claim 9, it is characterised in that
The convolutional layer Conv, carries out convolution algorithm, and calculate convolution by neuron activation functions for the image to input
Output;
The pond layer Pool, the image for being exported to convolutional layer carries out down-sampling to reduce the size of image;
The full articulamentum Fc, for all neuron nodes by any one neuron node of last layer all with next layer
Connection;
The articulamentum Concat, for by full articulamentum Fc1, full articulamentum Fc2, full articulamentum Fc3 and full articulamentum Fc4
The neuron node of output carries out Fusion Features;
The classification layer softmax, tagsort is carried out for the input value to each neuron node.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201710117801.XA CN107066941A (en) | 2017-03-01 | 2017-03-01 | A kind of face identification method and system |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201710117801.XA CN107066941A (en) | 2017-03-01 | 2017-03-01 | A kind of face identification method and system |
Publications (1)
Publication Number | Publication Date |
---|---|
CN107066941A true CN107066941A (en) | 2017-08-18 |
Family
ID=59622777
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201710117801.XA Pending CN107066941A (en) | 2017-03-01 | 2017-03-01 | A kind of face identification method and system |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN107066941A (en) |
Cited By (12)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN107766794A (en) * | 2017-09-22 | 2018-03-06 | 天津大学 | The image, semantic dividing method that a kind of Fusion Features coefficient can learn |
CN107784284A (en) * | 2017-10-24 | 2018-03-09 | 哈尔滨工业大学深圳研究生院 | Face identification method and system |
CN107833328A (en) * | 2017-11-20 | 2018-03-23 | 北京奇虎科技有限公司 | Gate inhibition's verification method and device, computing device based on recognition of face |
CN108062545A (en) * | 2018-01-30 | 2018-05-22 | 北京搜狐新媒体信息技术有限公司 | A kind of method and device of face alignment |
CN108364000A (en) * | 2018-03-26 | 2018-08-03 | 南京大学 | A kind of similarity preparation method based on the extraction of neural network face characteristic |
CN108416327A (en) * | 2018-03-28 | 2018-08-17 | 京东方科技集团股份有限公司 | A kind of object detection method, device, computer equipment and readable storage medium storing program for executing |
CN109308681A (en) * | 2018-09-29 | 2019-02-05 | 北京字节跳动网络技术有限公司 | Image processing method and device |
CN110033420A (en) * | 2018-01-12 | 2019-07-19 | 北京京东金融科技控股有限公司 | A kind of method and apparatus of image co-registration |
CN111079514A (en) * | 2019-10-28 | 2020-04-28 | 湖北工业大学 | Face recognition method based on CLBP and convolutional neural network |
CN111241998A (en) * | 2020-01-09 | 2020-06-05 | 中移(杭州)信息技术有限公司 | Face recognition method and device, electronic equipment and storage medium |
US10956778B2 (en) | 2019-03-05 | 2021-03-23 | International Business Machines Corporation | Multi-level deep feature and multi-matcher fusion for improved image recognition |
CN115641637A (en) * | 2022-11-11 | 2023-01-24 | 杭州海量信息技术有限公司 | Face recognition method and system for mask |
Citations (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN105118023A (en) * | 2015-08-31 | 2015-12-02 | 电子科技大学 | Real-time video human face cartoonlization generating method based on human facial feature points |
CN105117692A (en) * | 2015-08-05 | 2015-12-02 | 福州瑞芯微电子股份有限公司 | Real-time face identification method and system based on deep learning |
US20160086337A1 (en) * | 2014-09-19 | 2016-03-24 | Samsung Electronics Co., Ltd. | Method for detecting object and object detecting apparatus |
CN106203395A (en) * | 2016-07-26 | 2016-12-07 | 厦门大学 | Face character recognition methods based on the study of the multitask degree of depth |
CN106203533A (en) * | 2016-07-26 | 2016-12-07 | 厦门大学 | The degree of depth based on combined training study face verification method |
-
2017
- 2017-03-01 CN CN201710117801.XA patent/CN107066941A/en active Pending
Patent Citations (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US20160086337A1 (en) * | 2014-09-19 | 2016-03-24 | Samsung Electronics Co., Ltd. | Method for detecting object and object detecting apparatus |
CN105117692A (en) * | 2015-08-05 | 2015-12-02 | 福州瑞芯微电子股份有限公司 | Real-time face identification method and system based on deep learning |
CN105118023A (en) * | 2015-08-31 | 2015-12-02 | 电子科技大学 | Real-time video human face cartoonlization generating method based on human facial feature points |
CN106203395A (en) * | 2016-07-26 | 2016-12-07 | 厦门大学 | Face character recognition methods based on the study of the multitask degree of depth |
CN106203533A (en) * | 2016-07-26 | 2016-12-07 | 厦门大学 | The degree of depth based on combined training study face verification method |
Non-Patent Citations (1)
Title |
---|
SNEN: "【原创】batch-GD,SGD,Mini-batch-GD,Stochastic GD,Online-GD--大数据背景下的梯度训练算法", 《HTTPS://WWW.CNBLOGS.COM/RICHQIAN/P/4549590.HTML》 * |
Cited By (19)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN107766794B (en) * | 2017-09-22 | 2021-05-14 | 天津大学 | Image semantic segmentation method with learnable feature fusion coefficient |
CN107766794A (en) * | 2017-09-22 | 2018-03-06 | 天津大学 | The image, semantic dividing method that a kind of Fusion Features coefficient can learn |
CN107784284A (en) * | 2017-10-24 | 2018-03-09 | 哈尔滨工业大学深圳研究生院 | Face identification method and system |
CN107784284B (en) * | 2017-10-24 | 2021-07-06 | 哈尔滨工业大学深圳研究生院 | Face recognition method and system |
CN107833328A (en) * | 2017-11-20 | 2018-03-23 | 北京奇虎科技有限公司 | Gate inhibition's verification method and device, computing device based on recognition of face |
CN107833328B (en) * | 2017-11-20 | 2020-07-07 | 北京奇虎科技有限公司 | Access control verification method and device based on face recognition and computing equipment |
CN110033420B (en) * | 2018-01-12 | 2023-11-07 | 京东科技控股股份有限公司 | Image fusion method and device |
CN110033420A (en) * | 2018-01-12 | 2019-07-19 | 北京京东金融科技控股有限公司 | A kind of method and apparatus of image co-registration |
CN108062545A (en) * | 2018-01-30 | 2018-05-22 | 北京搜狐新媒体信息技术有限公司 | A kind of method and device of face alignment |
CN108062545B (en) * | 2018-01-30 | 2020-08-28 | 北京搜狐新媒体信息技术有限公司 | Face alignment method and device |
CN108364000A (en) * | 2018-03-26 | 2018-08-03 | 南京大学 | A kind of similarity preparation method based on the extraction of neural network face characteristic |
CN108416327A (en) * | 2018-03-28 | 2018-08-17 | 京东方科技集团股份有限公司 | A kind of object detection method, device, computer equipment and readable storage medium storing program for executing |
CN109308681B (en) * | 2018-09-29 | 2023-11-24 | 北京字节跳动网络技术有限公司 | Image processing method and device |
CN109308681A (en) * | 2018-09-29 | 2019-02-05 | 北京字节跳动网络技术有限公司 | Image processing method and device |
US10956778B2 (en) | 2019-03-05 | 2021-03-23 | International Business Machines Corporation | Multi-level deep feature and multi-matcher fusion for improved image recognition |
CN111079514A (en) * | 2019-10-28 | 2020-04-28 | 湖北工业大学 | Face recognition method based on CLBP and convolutional neural network |
CN111241998B (en) * | 2020-01-09 | 2023-04-28 | 中移(杭州)信息技术有限公司 | Face recognition method, device, electronic equipment and storage medium |
CN111241998A (en) * | 2020-01-09 | 2020-06-05 | 中移(杭州)信息技术有限公司 | Face recognition method and device, electronic equipment and storage medium |
CN115641637A (en) * | 2022-11-11 | 2023-01-24 | 杭州海量信息技术有限公司 | Face recognition method and system for mask |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN107066941A (en) | A kind of face identification method and system | |
CN109543606B (en) | Human face recognition method with attention mechanism | |
CN107609497B (en) | Real-time video face recognition method and system based on visual tracking technology | |
CN105975931B (en) | A kind of convolutional neural networks face identification method based on multiple dimensioned pond | |
CN104978580B (en) | A kind of insulator recognition methods for unmanned plane inspection transmission line of electricity | |
CN106407986B (en) | A kind of identification method of image target of synthetic aperture radar based on depth model | |
CN105512680B (en) | A kind of more view SAR image target recognition methods based on deep neural network | |
CN104850845B (en) | A kind of traffic sign recognition method based on asymmetric convolutional neural networks | |
CN103093250B (en) | A kind of Adaboost method for detecting human face based on new Haar-like feature | |
CN105849747B (en) | Method and system for facial image identification | |
CN107766850A (en) | Based on the face identification method for combining face character information | |
CN107239736A (en) | Method for detecting human face and detection means based on multitask concatenated convolutional neutral net | |
CN107871100A (en) | The training method and device of faceform, face authentication method and device | |
CN108681707A (en) | Wide-angle model recognizing method and system based on global and local Fusion Features | |
CN107545243A (en) | Yellow race's face identification method based on depth convolution model | |
CN110516595A (en) | Finger multi-modal fusion recognition methods based on convolutional neural networks | |
CN106372581A (en) | Method for constructing and training human face identification feature extraction network | |
CN112801015B (en) | Multi-mode face recognition method based on attention mechanism | |
CN109190643A (en) | Based on the recognition methods of convolutional neural networks Chinese medicine and electronic equipment | |
CN104484658A (en) | Face gender recognition method and device based on multi-channel convolution neural network | |
CN107633511A (en) | A kind of blower fan vision detection system based on own coding neutral net | |
CN104850857B (en) | Across the video camera pedestrian target matching process of view-based access control model spatial saliency constraint | |
CN113516005B (en) | Dance action evaluation system based on deep learning and gesture estimation | |
CN107578091A (en) | A kind of Pedestrians and vehicles real-time detection method based on lightweight depth network | |
CN110175615A (en) | The adaptive visual position recognition methods in model training method, domain and device |
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: 20170818 |