CN104363981B - Face verification method and system - Google Patents
Face verification method and system Download PDFInfo
- Publication number
- CN104363981B CN104363981B CN201480000558.8A CN201480000558A CN104363981B CN 104363981 B CN104363981 B CN 104363981B CN 201480000558 A CN201480000558 A CN 201480000558A CN 104363981 B CN104363981 B CN 104363981B
- Authority
- CN
- China
- Prior art keywords
- boltzmann machine
- data
- face
- layer
- node
- 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
- 238000012795 verification Methods 0.000 title claims abstract description 40
- 238000000034 method Methods 0.000 title claims abstract description 31
- 238000000513 principal component analysis Methods 0.000 claims abstract description 11
- 238000004422 calculation algorithm Methods 0.000 claims abstract description 9
- 238000004458 analytical method Methods 0.000 claims abstract description 7
- 230000009467 reduction Effects 0.000 claims abstract description 6
- 239000011159 matrix material Substances 0.000 claims description 8
- 230000008569 process Effects 0.000 claims description 6
- 238000012549 training Methods 0.000 claims description 4
- 238000012360 testing method Methods 0.000 claims description 3
- 238000012545 processing Methods 0.000 claims description 2
- 230000000717 retained effect Effects 0.000 claims description 2
- 238000012512 characterization method Methods 0.000 claims 1
- 230000004069 differentiation Effects 0.000 claims 1
- 230000006870 function Effects 0.000 description 14
- 230000001815 facial effect Effects 0.000 description 8
- 238000010586 diagram Methods 0.000 description 6
- 238000005457 optimization Methods 0.000 description 6
- 238000013524 data verification Methods 0.000 description 4
- 230000000694 effects Effects 0.000 description 4
- 238000002474 experimental method Methods 0.000 description 3
- 238000005286 illumination Methods 0.000 description 3
- 241001269238 Data Species 0.000 description 2
- 230000008859 change Effects 0.000 description 2
- 238000005516 engineering process Methods 0.000 description 2
- 238000011160 research Methods 0.000 description 2
- 238000007476 Maximum Likelihood Methods 0.000 description 1
- 238000004364 calculation method Methods 0.000 description 1
- 230000007547 defect Effects 0.000 description 1
- 238000013461 design Methods 0.000 description 1
- 230000002708 enhancing effect Effects 0.000 description 1
- 238000011478 gradient descent method Methods 0.000 description 1
- 230000006872 improvement Effects 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 238000012544 monitoring process Methods 0.000 description 1
- 238000003909 pattern recognition Methods 0.000 description 1
- 230000006641 stabilisation Effects 0.000 description 1
- 238000011105 stabilization Methods 0.000 description 1
- 238000006467 substitution reaction Methods 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/20—Image preprocessing
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Multimedia (AREA)
- Theoretical Computer Science (AREA)
- Image Analysis (AREA)
Abstract
The present invention relates to a kind of face verification method and system, method includes：Higher-dimension face characteristic data are pre-processed respectively using principal component analysis and linear discriminant analysis, including the data dimension after setting principal component analysis dimensionality reduction；Discriminate high-order Boltzmann machine is established, the number of nodes of hidden layer is set；The model parameter of the discriminate high-order Boltzmann machine is reduced using the strategy of tensor diagonalization；Paired human face data is input in discriminate high-order Boltzmann machine, the other conditional probability of relation object is maximized using stochastic gradient descent algorithm, so as to iteratively optimize the weight of the Boltzmann machine, so as to obtain final discriminate high-order Boltzmann machine；The paired human face data to be verified to discriminate high-order Boltzmann machine mode input, obtains corresponding verification result data.The present invention enhances Model checking power, is more suitable for the face verification with required precision by introducing the other information of data relation object in unsupervised Boltzmann machine model.
Description
Technical field
The present invention relates to a kind of face verification method and systems more particularly to one kind to be based on discriminate high-order Boltzmann machine
Face verification method and system.
Background technology
Face verification technology is widely applied the hot research problem of always area of pattern recognition, such as conduct due to it
Authentication is applied in customs and bank etc..
Given a pair of facial image, the target of face verification are that judge that this two facial images indicate whether same
People.Face verification under the mainly research constraint environment of work before, that is, the facial image in testing all is in controlled ring
It is gathered under the factor of border, including fixed illumination, posture and expression.However in practical applications, these factors are all uncontrollable
, such as facial image in video monitoring may include arbitrary illumination.So many researchers start then grind recently
Study carefully the face verification problem under unconstrained condition.Facial image under unconstrained condition factor has very big variance within clusters, place
Reason is got up will be challenging.
Existing face verification problem includes two steps：Data characteristics represents and data relationship metric.It is special for data
Sign represents that people represent human face data, such as LBP, SIFT and Gabor usually using the Feature Descriptor that some manually set
Deng.After obtaining paired human face data and representing, the similarity size of each pair image is usually measured using COS distance, and then
It may determine that this indicates whether same person to image.
The content of the invention
The defects of the purpose of the present invention is being directed to the prior art, provides a kind of face verification method and system, to realize more
Suitable for having the face verification of required precision.
To achieve the above object, the present invention provides a kind of face verification method, the described method includes：
Step 1 pre-processes higher-dimension face characteristic data using principal component analysis and linear discriminant analysis respectively,
Include setting the data dimension after principal component analysis dimensionality reduction；
Step 2 establishes discriminate high-order Boltzmann machine, sets the number of nodes of hidden layer；
Step 3, the model parameter that the discriminate high-order Boltzmann machine is reduced using the strategy of tensor diagonalization；
Step 4 is input to paired human face data in discriminate high-order Boltzmann machine, is calculated using stochastic gradient descent
Method maximizes the other conditional probability of relation object, final so as to obtain so as to iteratively optimize the weight of the Boltzmann machine
Discriminate high-order Boltzmann machine；
Step 5, to the discriminate high-order Boltzmann machine mode input paired human face data to be verified, corresponded to
Verification result data.
The present invention also provides a kind of face verification system, the system comprises：Network establishes module, network weight optimization
Module and Data Verification module, wherein：
The network establishes module, for establishing discriminate high-order Boltzmann machine, and sets the node of network hidden layer
Number；
The network weight optimization module minimizes the target letter of the Boltzmann machine using stochastic gradient descent algorithm
The Boltzmann machine weight after being optimized is counted, so as to obtain final discriminate high-order Boltzmann machine model；
The Data Verification module, for inputting paired face number to be verified to the discriminate high-order Boltzmann machine
According to obtaining two nodal value of classification layer, face verification result data can be drawn by comparing the relative size of two values.
Face verification method of the present invention and system, it is other by introducing data relation object in unsupervised Boltzmann machine model
Information enhances Model checking power, is more suitable for the face verification with required precision.
Description of the drawings
Fig. 1 is the flow chart of face verification method of the present invention；
Fig. 2 shows discriminate high-order Boltzmann machine structure diagram used in the present invention；
Fig. 3 is the schematic diagram of tensor diagonalization；
Fig. 4 is the schematic diagram of face verification system of the present invention；
Experiment effect variation diagram when Fig. 5 is different input data dimensions and different implicit node quantity.
Specific embodiment
Below by drawings and examples, technical scheme is described in further detail.
Invention face verification method and system mainly propose a more effective data relationship measure, Ke Yi
Largely promote the precision of face verification.
Fig. 1 is the flow chart of face verification method of the present invention, as shown in the figure, the present invention specifically comprises the following steps：
Step 101 pre-processes higher-dimension face characteristic data using principal component analysis and linear discriminant analysis respectively,
It wherein needs to set the data dimension after principal component analysis dimensionality reduction；
Wherein, principal component analysis (Principal Component Analysis, PCA) is a kind of common Data Dimensionality Reduction
Method, this method are needed to set the data dimension after dimensionality reduction, usually differed from a few hundred to several thousand.Linear discriminant analysis
(Linear Discriminant Analysis, LDA) is a kind of common sub-space learning method for having supervision, can after processing
So that the identification enhancing of data.
Step 102 establishes discriminate high-order Boltzmann machine, sets the number of nodes of hidden layer；
Wherein, the discriminate high-order Boltzmann machine be multitiered network structure, including two data input layers, one it is hidden
Containing layer and a classification output layer.Two input layers of the discriminate high-order Boltzmann machine train number for paired face
According to, such as the character representation of facial image.In an embodiment of the present invention, it is desirable that all human face datas keep identical size,
Such as the vector of similary length；Classification output layer is used to be expressed as the verification result to face training data, due to face verification
Result there was only two classes, that is, match or mismatch, output layer corresponds to both classifications respectively only comprising two nodes here；
The discriminate high-order Boltzmann machine has network weight, for obtaining next node layer value according to current layer nodal value.It is described
The input layer of discriminate high-order Boltzmann machine and the number of nodes of output layer are fixed, but the number of nodes of its hidden layer needs
Manual adjustment is so that the effect of the model is optimal.
Fig. 2 shows the discriminate high-order Boltzmann machine structure diagram used in one embodiment of the invention.Such as Fig. 2
Shown, this is one four layers of network, and the circular dot in every layer represents network node.Bottom Far Left represents two two layers respectively
Human face data input layer, the column vector that it is all I dimensions that input layer inputted, which is,One node of every dimension of vector
It represents, it is 1 that value, which is all normalized to 0 average and variance,.Top is classification output layer, and the classification of face verification is expressed as
The vector of one 2 dimensionThe value of vector is [1,0] or [0,1], two kinds of situation Corresponding matchings or mismatch.It is intermediate
Hidden layer be K dimension vectorVector is 0 or 1 per one-dimensional value.The value of h and z can be by them
The vector value of front layer is calculated, wherein h and x and y layers of interconnection：
Wherein, σ (x)=1/ (1+e-x), x=(xi)i∈I, y=(yj)j∈J, h=(hk)k∈K, z=(zt)t∈{1,2}.C=
(ck)k∈KWith d=(dt)t∈{1,2}The respectively offset parameter of hidden layer and classification layer.W=(Wijk)i∈I,j∈J,k∈KIt is two inputs
Connection weight between layer and hidden layer, and U=(Ukt)k∈K,t∈{1,2}It is the connection weight between hidden layer and classification layer.
Step 103, the model parameter that the discriminate high-order Boltzmann machine is reduced using the strategy of tensor diagonalization；Sentence
Other formula high-order Boltzmann machine contains substantial amounts of model parameter and is learnt, when the data dimension of input is hundreds of, three
The weight tensor W of dimension may can contain weight parameter up to a million.In this way, it is not very more situations in amount of training data
Under, it is easy to cause over-fitting.Therefore parameter is reduced using the method for tensor diagonalization here, as shown in Figure 3.On h axis,
Each hkCorrespond to a weight matrix section:
W..k=(Wijk)i∈I,j∈J
Each weight W wherein in matrixijkAll correspond to node xiAnd yjConnection relation.Implicit hypothesis is in x
Any one node all in y thus node there are correlativities.Such hypothesis is for human face data and improper.Specifically
For, we usually extract the feature of face key point attachment using local description, for example, face eyes, nose and
LBP features are extracted at the positions such as mouth, then connect these features to obtain final feature.So each node only table in x and y
The regional area of face is levied, as shown in the R1-R5 in Fig. 3.When carrying out face verification, that people are typically concerned about is two width people
Whether the same position such as eyes in face image are matched, without concern for the eyes of a people and the nose of another people
Whether match.So x only closed on positioniNode and yjNode represent human face region it is the same when, they just exist compared with
Strong correlation, such as the block diagonalization matrix in Fig. 3, wherein black representative has correlation, and white is no correlation.Therefore it is right
In arbitrary node xi, the node that we define relative ε-neighborhood is { yj, j ∈ [i- ε, i+ ε] }, then by each weight
Matrix section W..kIt is reduced to the weight only retained on its+1 diagonal of 2 ε：
Parameter ranges in so entire weight tensor W are i.e. from O (n3) it is reduced to O (n2)。
Step 104 is input to paired human face data in discriminate high-order Boltzmann machine, utilizes stochastic gradient descent
Algorithm maximizes the other conditional probability of relation object, final so as to obtain so as to iteratively optimize the weight of the Boltzmann machine
Discriminate high-order Boltzmann machine；
It is illustrated by taking the network in Fig. 2 as an example, two input layer datas, intermediate hidden layer and classification output layer table
Show a discriminate high-order Boltzmann machine, energy function E (x, y, h, z) is defined as：
Wherein, a=(ai)i∈IWith b=(b)j∈JThe bias term parameter of two input layers is represented respectively.
Can be obtained on the basis of energy function input layer data x, y and its relation classification z probability distribution P (x, y,
z)：
For learning model parameter, it is common practice to minimize object function using stochastic gradient descent algorithm, i.e.,
Minimize the maximum likelihood function L of negative logarithmgen：
Wherein α represents the index of training data.Object function LgenGradient on parameter W can be calculated as below：
Wherein ＜ ＞qRepresent the expectation on variable q.But the Section 2 calculated in above formula is extremely difficult, because meter
Calculation process needs to travel through all value states of variable x, y, h and z.Use LgenModel learning process during as object function
As shown in Figure 6.In addition, calculating process for simplicity, the object function of a discriminate has also been attempted in we, i.e., under negative logarithm
The other conditional probability of relation object：
Condition distribution P (z | x, y) wherein on z can be expressed as:
Different from object function Lgen, object function L in this casedisGradient on parameter W can be counted accurately
It calculates:
After gradient is obtained, we in an iterative manner parameter to being adjusted：
Wherein, ε represents a constant learning rate.
Step 105, to the trained discriminate high-order Boltzmann machine mode input paired face number to be verified
According to, such as x and y, output layer nodal value z is obtained, face verification is can determine whether by comparing the relative size of two of which numerical value
Result.If z0＞ z1Then mismatch, it is on the contrary then match.
Fig. 4 is the schematic diagram of face verification system of the present invention, as shown in the figure, the present invention specifically includes：Network establishes module
1st, network weight optimization module 2 and Data Verification module 3, wherein：
Specifically, network establishes module 1 for establishing discriminate high-order Boltzmann machine, and the section of network hidden layer is set
Points；Discriminate high-order Boltzmann machine is multitiered network structure, including two input data layers, a hidden layer and a class
Other output layer.Input layer be trained to human face data, output layer expression is trained to verification result to human face data；Differentiate
The discriminate high-order Boltzmann machine of formula has network weight, for obtaining next node layer value according to current layer nodal value.
Network weight optimization module 2 using stochastic gradient descent algorithm come minimize the object function of the Boltzmann machine come
Boltzmann machine weight after being optimized, so as to obtain final discriminate high-order Boltzmann machine model.
Data Verification module 3 is used to input paired human face data to be verified to discriminate high-order Boltzmann machine, obtains
Two nodal value of classification layer, face verification result can be drawn by comparing the relative size of two values.
For the specific embodiment that the present invention will be described in detail, here by taking two image data sets (LFW and WDRef) as an example
Explanation.LFW data sets include 13233 facial images for coming from 5749 people.Wherein 1680 people possess 1 image, and
The image for having two or more of remaining people.WDRef data sets include 99773 facial images thrown the net on network, wherein WDRef with
Human face data in LFW is not overlapped.Face images are all directly collected from Internet, in illumination, posture etc.
Aspect has very big variation.
We carry out face verification experiment under being set at two kinds：
1) the data of LFW are divided into 10 parts by the supervised learning under initial data, take wherein 9 points to be trained, remaining
1 part is tested, and is so repeated 10 times.
2) supervised learning under external data.It is trained using WDRef data, is tested on LFW databases.
It is as follows：
For all data, 2000 dimensions are dropped to using principal component analysis by step S1 for original higher-dimension face LBP features,
Linear discriminant analysis is recycled to be handled.
Step S2, using one four layers of discriminate high-order Boltzmann machine model, two input layer, hidden layer and defeated
Go out layer and include 2000,2000,1000 and 2 nodes respectively.
Step S3 eliminates the network weight in addition to diagonal using the strategy of tensor diagonalization, weight quantity is reduced
To O (n2)。
Step S4 utilizes the object function L of gradient descent algorithm optimization networkdis, wherein gradient can be calculated accurately
It is and non-approximated.Optimization is to carry out in an iterative manner, sets maximum iteration that can ensure to restrain for 30 here.
Step S6 is input to the paired human face data of test in trained model, and after output layer output verification
Result.Output the result is that the vector of a bidimensional, the relative size for comparing two values can determine whether to match or mismatch.
We compared some highest certain methods of face verification precision now.It can be sent out by the comparison of ROC curve
Under present two kinds of experimental setups, our method achieves best experimental result.In addition, we are tested using different
When input data dimension and different implicit node quantity, experiment effect situation of change of the invention.As shown in figure 5, transverse axis table
Show the quantity (number of hidden units) of implicit node, the longitudinal axis represents recognition of face precision (accuracy), in figure
Each curve represents a kind of face verification precision situation of change of data dimension in different implicit nodes.Input data dimension
1000 dimensions are changed to from 200 dimensions, implicit node quantity changes to 1200 from 200.It can be seen from the figure that fixed implicit knot
The quantity of point, with the dimension of input data is used to increase, precision is to rise.For the dimension of any input data, when hidden
When being more than 400 containing node, precision keeps stabilization no longer to rise.
Professional should further appreciate that, be described with reference to the embodiments described herein each exemplary
Unit and algorithm steps can be realized with the combination of electronic hardware, computer software or the two, hard in order to clearly demonstrate
The interchangeability of part and software generally describes each exemplary composition and step according to function in the above description.
These functions are performed actually with hardware or software mode, specific application and design constraint depending on technical solution.
Professional technician can realize described function to each specific application using distinct methods, but this realization
It is it is not considered that beyond the scope of this invention.
The step of method or algorithm for being described with reference to the embodiments described herein, can use hardware, processor to perform
The combination of software module or the two is implemented.Software module can be placed in random access memory (RAM), memory, read-only memory
(ROM), electrically programmable ROM, electrically erasable ROM, register, hard disk, moveable magnetic disc, CD-ROM or technical field
In any other form of storage medium well known to interior.
Above-described specific embodiment has carried out the purpose of the present invention, technical solution and advantageous effect further
It is described in detail, it should be understood that the foregoing is merely the specific embodiments of the present invention, is not intended to limit the present invention
Protection domain, within the spirit and principles of the invention, any modification, equivalent substitution, improvement and etc. done should all include
Within protection scope of the present invention.
Claims (3)
- A kind of 1. face verification method, which is characterized in that the described method includes：Step 1 pre-processes higher-dimension face characteristic data using principal component analysis and linear discriminant analysis respectively, wherein wrapping Include the data dimension after principal component analysis dimensionality reduction is set；Step 2 establishes discriminate high-order Boltzmann machine, sets the number of nodes of hidden layer；The model of the discriminate high-order Boltzmann machine in the step 2 is to include two data input layers, and one implicit The four-layer network network structure of layer and a classification output layer；Described two data input layers are paired training human face data, and the classification output layer represents face verification result；And And the discriminate high-order Boltzmann machine has network weight, to obtain next node layer value according to current layer nodal value；That input layer inputs is all column vector x, the y ∈ R of I dimensionsI×1, vectorial every dimension represents that value is all with a node It is 1 to be normalized to 0 average and variance；The top vector for being classification output layer, the classification of face verification being expressed as to one 2 dimension Z∈R2×1, vectorial value is [1,0] or [0,1], two kinds of situation Corresponding matchings or mismatch；Intermediate hidden layer is one The vectorial h ∈ R of K dimensionsK×1, vectorial is 0 or 1 per one-dimensional value；Step 3, the model parameter that the discriminate high-order Boltzmann machine is reduced using the strategy of tensor diagonalization；Specifically, on h axis, each hkCorrespond to a weight matrix section:W..k=(Wijk)I ∈ I, j ∈ JEach weight W wherein in matrixijkAll correspond to node xiAnd yjConnection relation, so each node characterization in x and y The regional area of face, the x closed on positioniNode and yjIt is right there are correlation when the human face region that node represents is the same In arbitrary node xi, the node for defining relative ε-neighborhood is { yj, j ∈ [i- ε, i+ ε] }, then by each weight matrix Cut into slices W..kIt is reduced to the weight only retained on+1 diagonal of 2 ε：Parameter ranges in so entire weight tensor W are from O (n3) it is reduced to O (n2)；Step 4 is input to paired human face data in discriminate high-order Boltzmann machine, using stochastic gradient descent algorithm come The other conditional probability of relation object is maximized, so as to iteratively optimize the weight of the Boltzmann machine, so as to obtain final differentiation Formula high-order Boltzmann machine；Step 5, to the discriminate high-order Boltzmann machine mode input paired human face data to be verified, obtain corresponding test Demonstrate,prove result data.
- 2. according to the method described in claim 1, it is characterized in that, the strategy of the tensor diagonalization in the step 3 is by institute State weight matrix of the three dimensional weight cubes processing between two data input layers and hidden layer for multiple two dimensions.
- 3. according to the method described in claim 1, it is characterized in that, the object function of the discriminate high-order Boltzmann machine is The other conditional probability of relation object of negative logarithm.
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| PCT/CN2014/082149 WO2016008071A1 (en) | 2014-07-14 | 2014-07-14 | Face verification method and system |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN104363981A CN104363981A (en) | 2015-02-18 |
| CN104363981B true CN104363981B (en) | 2018-06-05 |
Family
ID=52530957
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201480000558.8A Active CN104363981B (en) | 2014-07-14 | 2014-07-14 | Face verification method and system |
Country Status (2)
| Country | Link |
|---|---|
| CN (1) | CN104363981B (en) |
| WO (1) | WO2016008071A1 (en) |
Families Citing this family (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109978164B (en) * | 2019-03-18 | 2022-12-06 | 西安电子科技大学 | Method for identifying high-resolution range profile of morphing aircraft based on deep confidence network |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101021900A (en) * | 2007-03-15 | 2007-08-22 | 上海交通大学 | Method for making human face posture estimation utilizing dimension reduction method |
| CN103605972A (en) * | 2013-12-10 | 2014-02-26 | 康江科技(北京)有限责任公司 | Non-restricted environment face verification method based on block depth neural network |
| CN103793718A (en) * | 2013-12-11 | 2014-05-14 | 台州学院 | Deep study-based facial expression recognition method |
Family Cites Families (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US8239336B2 (en) * | 2009-03-09 | 2012-08-07 | Microsoft Corporation | Data processing using restricted boltzmann machines |
-
2014
- 2014-07-14 WO PCT/CN2014/082149 patent/WO2016008071A1/en active Application Filing
- 2014-07-14 CN CN201480000558.8A patent/CN104363981B/en active Active
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101021900A (en) * | 2007-03-15 | 2007-08-22 | 上海交通大学 | Method for making human face posture estimation utilizing dimension reduction method |
| CN103605972A (en) * | 2013-12-10 | 2014-02-26 | 康江科技(北京)有限责任公司 | Non-restricted environment face verification method based on block depth neural network |
| CN103793718A (en) * | 2013-12-11 | 2014-05-14 | 台州学院 | Deep study-based facial expression recognition method |
Also Published As
| Publication number | Publication date |
|---|---|
| CN104363981A (en) | 2015-02-18 |
| WO2016008071A1 (en) | 2016-01-21 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Guo et al. | Human age estimation using bio-inspired features | |
| Ghosh et al. | A multi-label convolutional neural network approach to cross-domain action unit detection | |
| Ma et al. | Linear dependency modeling for classifier fusion and feature combination | |
| Keinert et al. | A robust group-sparse representation variational method with applications to face recognition | |
| Qin et al. | Finger-vein quality assessment by representation learning from binary images | |
| CN103927522B (en) | A kind of face identification method based on manifold self-adaptive kernel | |
| Khan et al. | Human gait analysis: A sequential framework of lightweight deep learning and improved moth-flame optimization algorithm | |
| Han et al. | High-order statistics of microtexton for hep-2 staining pattern classification | |
| Salhi et al. | Fast and efficient face recognition system using random forest and histograms of oriented gradients | |
| Puthenputhussery et al. | A sparse representation model using the complete marginal fisher analysis framework and its applications to visual recognition | |
| Xie et al. | Deep nonlinear metric learning for 3-D shape retrieval | |
| Balaji et al. | Multi-level feature fusion for group-level emotion recognition | |
| CN112101087A (en) | Facial image identity de-identification method and device and electronic equipment | |
| de Oliveira et al. | A novel 2D shape signature method based on complex network spectrum | |
| CN110490028A (en) | Recognition of face network training method, equipment and storage medium based on deep learning | |
| Zhang et al. | Discriminative tensor sparse coding for image classification. | |
| Alom et al. | Digit recognition in sign language based on convolutional neural network and support vector machine | |
| Yesu et al. | Hybrid features based face recognition method using Artificial Neural Network | |
| Rodrigues et al. | HEp-2 cell image classification based on convolutional neural networks | |
| CN104363981B (en) | Face verification method and system | |
| Zhong et al. | A diversified deep belief network for hyperspectral image classification | |
| Rohmatillah et al. | Automatic cervical cell classification using features extracted by convolutional neural network | |
| Sangari et al. | Paper texture classification via multi-scale restricted Boltzman machines | |
| Demontis et al. | Super-sparse regression for fast age estimation from faces at test time | |
| Wu et al. | Collaborative representation for classification, sparse or non-sparse? |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| C06 | Publication | ||
| PB01 | Publication | ||
| C10 | Entry into substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |