CN109145986B - Large-scale face recognition method - Google Patents
Large-scale face recognition method Download PDFInfo
- Publication number
- CN109145986B CN109145986B CN201810956517.6A CN201810956517A CN109145986B CN 109145986 B CN109145986 B CN 109145986B CN 201810956517 A CN201810956517 A CN 201810956517A CN 109145986 B CN109145986 B CN 109145986B
- Authority
- CN
- China
- Prior art keywords
- loss function
- face recognition
- optimization
- steps
- softmax loss
- 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.)
- Active
Links
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- 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
- G06V40/172—Classification, e.g. identification
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- General Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- Life Sciences & Earth Sciences (AREA)
- Evolutionary Biology (AREA)
- Evolutionary Computation (AREA)
- Bioinformatics & Computational Biology (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Artificial Intelligence (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Health & Medical Sciences (AREA)
- General Health & Medical Sciences (AREA)
- Oral & Maxillofacial Surgery (AREA)
- Human Computer Interaction (AREA)
- Multimedia (AREA)
- Image Analysis (AREA)
- Collating Specific Patterns (AREA)
Abstract
A large-scale face recognition method comprises the following steps: the method comprises the following steps: training data and cleaning the trained data; step two: setting a network; step three: inputting settings, adopting 5 key points to calibrate the human face and normalizing the human face to a 112 × 112 color image; step four: outputting the setting; step five: and optimizing the Softmax loss function to obtain a new Softmax loss function, and improving the accuracy of face recognition according to the new Softmax loss function. The invention provides a large-scale face recognition method, which improves the face recognition precision, thereby realizing large-scale face recognition.
Description
Technical Field
The invention relates to the technical field of face recognition, in particular to a large-scale face recognition method.
Background
The face recognition identity authentication technology is widely applied to the fields of security, entrance guard, attendance checking and the like due to the characteristics of convenience, rapidness, non-contact and the like. The convolutional neural network has the characteristic of large capacity in the aspect of characteristic, and is suitable for large-scale face recognition. However, in the face recognition process, intra-class changes caused by factors such as facial expressions, postures, ages, positions and coverings of a person and inter-class changes caused by different identities from outside light, backgrounds and the like need to be considered, the two changes are highly complex and nonlinear, and are generally analyzed by using a Softmax loss function, but the Softmax loss function only separates inter-class features and does not gather features belonging to the same class, and the features are not effective enough for face recognition.
Disclosure of Invention
The invention aims to provide a large-scale face recognition method, which improves the face recognition precision and further realizes large-scale face recognition.
In order to achieve the purpose, the invention adopts the following technical scheme:
a large-scale face recognition method comprises the following steps:
the method comprises the following steps: training data and cleaning the trained data;
step two: setting a network;
step three: inputting settings, adopting 5 key points to calibrate the human face and normalizing the human face to a 112 × 112 color image;
step four: outputting the setting;
step five: optimizing the Softmax loss function to obtain a new Softmax loss function, and improving the accuracy of face recognition according to the new Softmax loss function, wherein the Softmax loss function is defined as follows:
where m represents the number of samples, n represents the number of sample classifications, s represents the scale coefficient,m+θyrepresents an angle increment manner, wherein m is 0.5.
Preferably, in step five, the optimization of the Softmax loss function includes a primary optimization and a secondary optimization, and the primary optimization includes the following steps:
setting bias as 0, and adding a margin to the included angle of projection to obtain a primary optimization formula as follows:
where m represents the number of samples, n represents the number of sample classifications, s represents the scale coefficient,m+θyrepresents an angle increment manner, wherein m is 0.5.
Preferably, the second optimization comprises adding a fixed value to the angle.
Preferably, in step one, the training data comprises training using a VGG network.
Preferably, in step four, the output setting includes a method of using the optimal structure (contribution- > BN- > Dropout- > Fullyconnected- > BN) to realize the connection of the last Convolution layer to the feature vector.
Has the advantages that:
by optimizing a traditional softmax loss function, theta is expressed as an included angle formed by normalizing w and x, so that only one straight line exists between classes, the boundary is enlarged, and classification is facilitated. Meanwhile, training data are optimized and cleaned, and an optimal network structure is selected, so that the recognition precision is greatly improved, and along with the improvement of the recognition precision, the recognition efficiency and accuracy of large-scale human faces are improved.
Drawings
Fig. 1 is a flow chart of the present invention for improving the accuracy of large-scale face recognition.
Detailed Description
The technical scheme of the invention is further explained by the specific implementation mode in combination with the attached drawings.
As shown in fig. 1, the large-scale face recognition method of this embodiment includes the following steps:
the method comprises the following steps: training data and cleaning the trained data; the training set used VGG network training 8000 people, contained 3000000 images, the test set contained 160000 images of 500 people, image quality spanning pose, age, lighting, race, occupation, etc. The training data is cleaned.
Step two: setting a network; various network settings were verified using the MxNet architecture, using VGG2 as training data, softmax as a loss function. The batch size is set to 512, the learning rate starts from 0.1, the learning rate is reduced by 10 times at 100k, 140k and 160k respectively, the impulse unit takes a value of 0.9, and the weight attenuation is set to 5 e-4.
Step three: inputting settings, adopting 5 key points to calibrate the human face and normalizing the human face to a 112 × 112 color image; the face was aligned using 5 keypoints and normalized to a 112 × 112 color image, and in order to ensure the resolution of the feature map, this example replaced conv7 × 7stride2 with conv3 × 3stride 1. The output of the last convolution is 7 x 7.
Step four: outputting the setting; including the implementation of the last Convolution layer connected to the feature vector by the optimal structure (constraint- > BN- > Dropout- > Fullyconnected- > BN).
Step five: optimizing the Softmax loss function to obtain a new Softmax loss function, and improving the accuracy of face recognition according to the new Softmax loss function, wherein the definition of the common Softmax loss function is as follows:
the traditional Softmax can not optimize the inter-class distance and the intra-class distance of the feature points, and bias is set to be 0 for calculation convenience, so that the included angle after normalization is represented. In such Softmax, class-to-class boundaries are simply one line. Such points that fall near the boundary may make the generalization capability of the entire model poor. To make this limit larger, the points between the different classes are made as far apart as possible. This can be achieved by adding a margin to the included angle of the projection. The function after the initial optimization is therefore as follows:
after the primary optimization, the Softmax function stops converging in the initial stage, so that secondary optimization is performed, a fixed value is added to the angle, and the function after the secondary optimization is as follows:
after optimization, the angular distance is more directly affected on the angle than the cosine distance.
The technical principle of the present invention is described above in connection with specific embodiments. The description is made for the purpose of illustrating the principles of the invention and should not be construed in any way as limiting the scope of the invention. Based on the explanations herein, those skilled in the art will be able to conceive of other embodiments of the present invention without inventive effort, which would fall within the scope of the present invention.
Claims (1)
1. A large-scale face recognition method is characterized in that: the method comprises the following steps:
the method comprises the following steps: training data and cleaning the trained data;
step two: setting a network, using an MxNet architecture, using a VGG2 network as training data, and using softmax as a loss function to verify various network settings; the batch size is set to 512, the learning rate is reduced by 10 times from 0.1 at 100k, 140k and 160k respectively, the impulse unit takes a value of 0.9, and the weight attenuation is set to 5 e-4;
step three: inputting settings, adopting 5 key points to calibrate the human face and normalizing the human face to a 112 × 112 color image;
step four: output settings including methods for implementing the last Convolution layer connected to the feature vector using the optimal structure Convolume- > BN- > Dropout- > Fullyconnected- > BN;
step five: optimizing the Softmax loss function to obtain a new Softmax loss function, and improving the accuracy of face recognition according to the new Softmax loss function, wherein the method comprises the following steps:
optimizing the Softmax loss function comprises a primary optimization and a secondary optimization, wherein the primary optimization comprises the following steps:
setting bias as 0, and adding a margin to the included angle of projection to obtain a primary optimization formula as follows:
m represents the number of samples, i represents the number of samples, represents the ith sample, n represents the number of sample classifications, j represents the jth number of sample classifications, s represents the scale coefficient,
representing a fixed value, representing the angle of projection of the feature points,
representing the included angle of the feature point after projection normalization;
and performing secondary optimization on the formula after the primary optimization, wherein the secondary optimization comprises adding a fixed value on the angle as follows:
where m represents the number of samples, i represents the ith number of samples, n represents the number of sample classifications, j represents the jth number of sample classifications, s represents the scale coefficient,
indicating an angular incremental manner in which
，
Which represents a fixed value of the time-varying,
and representing the included angle of the projection normalization of the characteristic points.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810956517.6A CN109145986B (en) | 2018-08-21 | 2018-08-21 | Large-scale face recognition method |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810956517.6A CN109145986B (en) | 2018-08-21 | 2018-08-21 | Large-scale face recognition method |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN109145986A CN109145986A (en) | 2019-01-04 |
| CN109145986B true CN109145986B (en) | 2021-12-24 |
Family
ID=64790832
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201810956517.6A Active CN109145986B (en) | 2018-08-21 | 2018-08-21 | Large-scale face recognition method |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN109145986B (en) |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106372581A (en) * | 2016-08-25 | 2017-02-01 | 中国传媒大学 | Method for constructing and training human face identification feature extraction network |
| CN107103281A (en) * | 2017-03-10 | 2017-08-29 | 中山大学 | Face identification method based on aggregation Damage degree metric learning |
| CN108182409A (en) * | 2017-12-29 | 2018-06-19 | 北京智慧眼科技股份有限公司 | Biopsy method, device, equipment and storage medium |
Family Cites Families (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US10509952B2 (en) * | 2016-08-30 | 2019-12-17 | Irida Labs S.A. | Fast, embedded, hybrid video face recognition system |
-
2018
- 2018-08-21 CN CN201810956517.6A patent/CN109145986B/en active Active
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106372581A (en) * | 2016-08-25 | 2017-02-01 | 中国传媒大学 | Method for constructing and training human face identification feature extraction network |
| CN107103281A (en) * | 2017-03-10 | 2017-08-29 | 中山大学 | Face identification method based on aggregation Damage degree metric learning |
| CN108182409A (en) * | 2017-12-29 | 2018-06-19 | 北京智慧眼科技股份有限公司 | Biopsy method, device, equipment and storage medium |
Also Published As
| Publication number | Publication date |
|---|---|
| CN109145986A (en) | 2019-01-04 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN110443143B (en) | Multi-branch convolutional neural network fused remote sensing image scene classification method | |
| CN108898180B (en) | Depth clustering method for single-particle cryoelectron microscope images | |
| CN104966097B (en) | A kind of complex script recognition methods based on deep learning | |
| CN108520202B (en) | Method for extracting image characteristics with robustness resistance based on variational spherical projection | |
| CN105138998B (en) | Pedestrian based on the adaptive sub-space learning algorithm in visual angle recognition methods and system again | |
| CN109190446A (en) | Pedestrian's recognition methods again based on triple focused lost function | |
| CN112580590A (en) | Finger vein identification method based on multi-semantic feature fusion network | |
| CN111126386A (en) | Sequence field adaptation method based on counterstudy in scene text recognition | |
| CN111401156B (en) | Image identification method based on Gabor convolution neural network | |
| CN107203752A (en) | A kind of combined depth study and the face identification method of the norm constraint of feature two | |
| CN107480723B (en) | Texture Recognition based on partial binary threshold learning network | |
| CN108038504A (en) | A kind of method for parsing property ownership certificate photo content | |
| Hu et al. | Toward driver face recognition in the intelligent traffic monitoring systems | |
| CN111160538B (en) | Method and system for updating margin parameter value in loss function | |
| CN115546553A (en) | Zero sample classification method based on dynamic feature extraction and attribute correction | |
| CN108520201A (en) | A kind of robust human face recognition methods returned based on weighted blend norm | |
| CN109508670B (en) | Static gesture recognition method based on infrared camera | |
| CN109145986B (en) | Large-scale face recognition method | |
| CN103226698B (en) | A kind of method for detecting human face | |
| CN111191027B (en) | Generalized zero sample identification method based on Gaussian mixture distribution (VAE) | |
| CN116704585A (en) | Face recognition method based on quality perception | |
| CN111445540A (en) | Automatic registration method for RGB colored three-dimensional point cloud | |
| CN111126173A (en) | High-precision face detection method | |
| Hassan et al. | Intelligent sign language recognition using enhanced fourier descriptor: a case of Hausa sign language | |
| CN110378414A (en) | The personal identification method of multi-modal biological characteristic fusion based on evolution strategy |
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 |