CN101877065A - Extraction and identification method of non-linear authentication characteristic of facial image under small sample condition - Google Patents
Extraction and identification method of non-linear authentication characteristic of facial image under small sample condition Download PDFInfo
- Publication number
- CN101877065A CN101877065A CN2009102346624A CN200910234662A CN101877065A CN 101877065 A CN101877065 A CN 101877065A CN 2009102346624 A CN2009102346624 A CN 2009102346624A CN 200910234662 A CN200910234662 A CN 200910234662A CN 101877065 A CN101877065 A CN 101877065A
- Authority
- CN
- China
- Prior art keywords
- matrix
- facial image
- test
- calculate
- sample
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Granted
Links
Images
Landscapes
- Image Analysis (AREA)
- Image Processing (AREA)
Abstract
The invention discloses an extraction and identification method of non-linear authentication characteristics of facial images under the small sample condition, comprising the following stages: (1) training stage: firstly, utilizing a facial image sample to calculate a kernel matrix, constructing two new matrixes, calculating two new kernel matrixes on the basis, and carrying out Cholesky decomposition on one kernel matrix to obtain an upper triangular matrix so as to calculate a non-liner characteristic vector to each type of facial images; and (2) identification stage: firstly, calculating the non-liner characteristic vector of a facial image to be identified, then calculating the distance between the non-liner characteristic vector and each type of facial image characteristic vector, and classifying the facial image to be identified into the facial classification corresponding to the minimum distance. The non-linear authentication vector and the characteristics obtained by the invention are irreverent so as to eliminate redundancy among non-linear authentication characteristics and improve the authentication capability of the authentication characteristics. In the training stage and the identification stage, the method has high calculation efficiency and better numerical value calculation stability.
Description
Technical field
Invention relates to small sample Optimal Nonlinear diagnostic characteristics to be extracted and the identification field, specifically is that the non-linear diagnostic characteristics of facial image under a kind of condition of small sample extracts and recognition methods.The present invention can be used for machine learning and area of pattern recognition, except the feature extraction of facial image and identification, also can be used for other images under the condition of small sample and the feature extraction and the identification of data.
Background technology
The feature extraction technology of facial image can be divided into based on geometric properties with based on statistical nature two big classes.Early stage method is mainly based on geometrical feature extraction, and its basic thought is the position of at first locating people's unique point on the face, calculates the relative position and the distance of these unique points then, with its tolerance as face characteristic.Because the extraction of geometric properties is responsive to the variation of illumination, attitude and expression, and is difficult to accurate location, has influenced the stability and the discrimination of these class methods.The more in recent years abstracting method that is based on statistical nature.
In the abstracting method based on statistical nature, the face characteristic abstracting method that is based on linear and nonlinear transformation commonly used, these class methods obtain a conversion battle array by criterion function of optimization, with the extremely low n-dimensional subspace n of original higher-dimension facial image dimensionality reduction, make the face characteristic in low n-dimensional subspace n compacter, better separability is arranged.Nonlinear Feature Extraction Methods based on kernel method also has been applied in the face characteristic extraction, it is with the feature space of people's face sample Nonlinear Mapping to higher-dimension, make sample better separability be arranged at feature space, be more conducive to handle the various nonlinearities change (as illumination, expression and attitude etc.) of facial image, therefore can obtain having more the non-linear diagnostic characteristics of resolving ability than the linear feature extraction method.These class methods are aerial to son with higher-dimension facial image feature linearity or Nonlinear Mapping, therefore also are referred to as subspace method.
The recognition of face stage is divided into each zone at the sorting technique that the characteristic Design that is drawn into is fit to the sample characteristics space, according to the zone at sample characteristics to be identified place it is included in the corresponding classification then.The feature extraction stage is used the nearest neighbor classifier classification always after obtaining the diagnostic characteristics of facial image.
In people's face statistical nature abstracting method, commonly based on the linear discriminant analysis of Fisher criterion (being called for short FLDA) method based on the subspace.The FLDA method is by optimization Fisher criterion, make the discriminant vector that obtains to the sample dimensionality reduction after, divergence minimum in the between class scatter maximum of lower dimensional space sample characteristics and class, thus the gained diagnostic characteristics has best class separability after the dimensionality reduction conversion.But when the dimension of sample greater than class in during the divergence rank of matrix, then find the solution best diagnostic characteristics and have ill singular problem, this problem is also referred to as the ill singular problem under the condition of small sample.
The kernel method that develops out from Statistical Learning Theory is mapped to feature space with original sample from the input space by a nonlinear transformation, makes the sample after the mapping at feature space better class separability be arranged.Carry out linear analysis at feature space and be equivalent in the input space and carry out nonlinear analysis, thereby obtain nonlinear learning algorithm in the input space.
There is small sample morbid state singular problem equally in FLDA and kernel method non-linear Fisher discriminatory analysis (the being called for short KFDA) method that obtains based on nuclear that combines.Early stage KFDA method only can be handled the classification problem of two classes, and finds the solution and have ill singular problem, and its method with scrambling kinetic moment battle array solves; Further the KFDA of two classes is generalized to the situation of multicategory classification based on the broad sense discriminatory analysis method of nuclear, but, has replaced divergence battle array in the class, can not obtain best non-linear diagnostic characteristics in essence with total divergence battle array for solving ill singular problem; There is ill-conditioning problem equally in multiclass discriminatory analysis based on nuclear, and it solves with method of perturbation equally, and its shortcoming is to be difficult to select the disturbance factor that is fit to.
Summary of the invention
The present invention seeks to provides the non-linear diagnostic characteristics of facial image under a kind of condition of small sample to extract and recognition methods at the defective of existing facial image feature extraction and recognition technology existence.
The present invention adopts following technical scheme for achieving the above object:
The present invention utilizes the facial image training sample to find the solution the nonlinear characteristic vector of every class people's face, and facial image to be identified is found the solution the nonlinear characteristic vector, and relatively both distances are classified to facial image to be identified with the arest neighbors criterion, and the specific implementation step is as follows:
(1) training stage:
1. calculate nuclear matrix K, structural matrix A calculates nuclear matrix K
A:
C people's face classification arranged, and everyone face classification has N facial image training sample, and total number of training is M=NC; With the facial image sample matrix that collects stretching be a vector, with vector representation facial image sample;
Calculate the nuclear matrix K of M * M, the element kernel function k (x of the capable j row of its m
m, x
j) calculate kernel function wherein
Be 2 rank polynomial kernel functions, vector x
mAnd x
jRepresent m and j facial image training sample in total training sample respectively,
Expression x
mThe transposition computing, m=1,2 ..., M, j=1,2 ..., M, down with; Structural matrix A=[A
1, A
2..., A
C], A wherein
iBe the matrix of M * N-1, i=1,2 ..., C is defined as follows:
Matrix A
iIn: capable 0, the (i-1) N+1 capable being-1, the (i-1) N+1 that is is the unit matrix of a N-1 * N-1 below capable from the 1st row to (i-1) N, and all the other elements are 0;
Calculate nuclear matrix K with matrix K and A
A=A
TKA, the computing of subscript T representing matrix transposition;
2. structural matrix B calculates nuclear matrix K
B:
Structural matrix B is the matrix of M * C-1, and building method is as follows:
In the matrix B: the 1st the row be-1, the q * N+1 is capable, and q classifies 1 as, wherein the value of q is q=1,2 ..., C-1, all the other elements are 0;
Calculate nuclear matrix K
B:
K
B=B
T[K-2KDK+KDKDK]B
B wherein
TThe transposed matrix of expression B, matrix D=A (K
A)
-1A
T, matrix (K
A)
-1Expression K
AInverse matrix, A
TThe transposed matrix of expression A;
3. to nuclear matrix K
BCarry out the Cholesky decomposition and obtain upper triangular matrix R
B,
4. calculate the nonlinear characteristic vector Y of i class people face
i:
Wherein
N people's face image pattern representing i class people face, i=1,2 ..., C, n=1,2,3 ..., N; I representation unit battle array,
Expression R
BThe inverse matrix of transposed matrix;
(2) cognitive phase:
A) calculate facial image sample x to be identified
TestThe nonlinear characteristic vector Y
Test:
H=[k (x wherein
1, x
Test), k (x
2, x
Test) ..., k (x
M, x
Test)]
T
B) calculate Y
TestAnd Y
iMinor increment:
I=1,2 ..., C, wherein || Y
i-Y
Test|| expression Y
TestAnd Y
iEuclidean distance, min represents to ask minor increment; The criterion of identification is with facial image sample x to be identified
TestBe included in people's face classification of minor increment correspondence.
Advantage of the present invention is:
(1) finds the solution non-linear discriminant vector in the kernel of the present invention's divergence battle array in facial image sample total divergence rank of matrix space and class, can obtain the Optimal Nonlinear feature of optimization Fisher criterion.(2) the non-linear discriminant vector that the present invention calculated has orthonormal character.By constructing new matrix, calculate new nuclear matrix, the Cholesky that the orthogonalization procedure in the feature space is converted into new nuclear matrix decomposes, and has improved numerical stability and counting yield.(3) the present invention's matrix that only need utilize the training stage to calculate at cognitive phase just can calculate the nonlinear characteristic vector of sample to be identified.Criterion of identification is to calculate the eigenvector of sample to be identified and the distance of every class facial image eigenvector, and facial image to be identified is included in the minor increment corresponding class.Owing to only need calculate a nonlinear characteristic vector to every class facial image, the computation complexity of cognitive phase only and the facial image number of samples in the relevant and every class people's face of people's face classification number have nothing to do, have higher counting yield at cognitive phase.(4) for the less recognition of face problem of number of training, because therefore the general divergence rank of matrix in the class of the dimension behind the stretching one-tenth vector of facial image matrix exists the problem of the ill generalized character equation under the condition of small sample of finding the solution.This invention is head it off preferably.
Embodiment
Accompanying drawing 1 is the calculation flow chart of performing step of the present invention, is elaborated below in conjunction with 1 pair of technical scheme of the present invention of accompanying drawing:
(1), cognitive phase
1. calculate nuclear matrix K, structural matrix A calculates nuclear matrix K
A
2. structural matrix B calculates nuclear matrix K
B
3. to nuclear matrix K
BCarry out the Cholesky decomposition and obtain upper triangular matrix R
B
4. calculate the nonlinear characteristic vector Y of every class people's face
i
(2), the training stage
A) the nonlinear characteristic vector Y of calculating facial image to be identified
Test
B) calculate Y
iAnd Y
TestDistance, sample to be identified is included in people's face classification of minor increment correspondence.
Embodiments of the invention technical scheme according to the present invention is implemented, and provided concrete embodiment and calculating operation process, but protection scope of the present invention is not limited to following embodiment.
Embodiment one:
Adopt public AT﹠amp; T standard faces image data base.AT﹠amp; The T storehouse comprises 40 people's face classifications, and everyone face classification has the facial image of 10 different human face postures, expression and face detail, and the image size is 112 * 92.
(1), the training stage:
The data pre-service: the image array with 112 * 92 carries out down-sampling, and size becomes 28 * 23.Stretching by row is the column vector of 644 dimensions, and the pixel value of image is normalized between the 0-1.With every class people's face sample separated into two parts at random, a part is as training sample, and a part is as test sample book.The number of training of every class people's face is N, and total number of training is M=40N.The span of N is N=2,3,4,5,6,7,8,9.
With everyone face image pattern column vector x
mExpression, m=1,2 ..., M, the category order is arranged in training sample matrix [x with training sample
1, x
2..., x
M].
Structural matrix A=[A at first
1, A
2..., A
40], A
iBe the matrix of M * N-1, i=1,2 ..., 40, building method
Personnel selection face image pattern and 2 rank polynomial kernel function calculation obtain nuclear matrix K, and the element of the capable j row of the m of nuclear matrix K is
M=1,2 ..., M, j=1,2 ..., M, x
mAnd x
jRepresent m and j training sample in total training sample respectively.
Calculate new nuclear matrix K
A=A
TKA
Compute matrix D=A (K
A)
-1A
T
The matrix B of structure M * 39
Calculate new nuclear matrix K
B=B
T[K-2KDK+KDKDK] B, wherein B
TThe transposed matrix of expression B, matrix D=A (K
A)
-1A
T, matrix (K
A)
-1Expression K
AInverse matrix, A
TThe transposed matrix of expression A.
To K
BCarry out the Cholesky decomposition and obtain upper triangular matrix R
B
Facial image training sample with the i class
The nonlinear characteristic vector that calculates i class facial image sample is:
Wherein I is a unit matrix,
M training sample representing the i class, i=1,2 ..., 40.x
mRepresent m training sample in total training sample.K represents 2 rank polynomial kernel functions.
(2) cognitive phase: calculate facial image sample x to be identified
TestThe nonlinear characteristic vector Y
Test
H=[k (x wherein
1, x
Test), k (x
2, x
Test) ..., k (x
M, x
Test)]
T, k represents 2 rank polynomial kernel functions.
Cognitive phase is included into facial image sample to be identified in people's face classification of minor increment correspondence
Table 1 is to use AT﹠amp; The result of T face database test carries out 8 experiments altogether.To every class people's face, test 2 to 9 samples of picked at random as training sample at every turn, remaining sample is as test sample book, and each experiment repeats 20 times, calculates average recognition rate, and the kernel function of the inventive method is with 2 rank polynomial kernel functions.The result of method of the present invention and Fisherface method is compared, and under identical training sample and test sample book condition, method of the present invention all is better than the Fisherface method.
Table 1 average recognition rate (%)
Embodiment two:
Adopt public UMIST standard faces image data base.Comprise 20 people's face classifications in the UMIST storehouse, everyone face classification is selected the facial image of 20 different human face postures, and the image size is 220 * 220.
(1), the training stage:
The data pre-service: the image array with 220 * 220 carries out down-sampling, and size becomes 28 * 23, and stretching is the column vector of 644 dimensions, and the pixel value of image is normalized between the 0-1.With every class people's face sample separated into two parts at random, a part is as training sample, and a part is as test sample book.The number of training of every class people's face is N, and total number of training is M=20N.The value of N is N=3 in the present embodiment, 5,7,9,11,13,15,17.
With everyone face image pattern column vector x
mExpression, m=1,2 ..., M, the category order is arranged in training sample matrix [x with training sample
1, x
2..., x
M].
Structural matrix A=[A at first
1, A
2..., A
20], A wherein
iBe the matrix of M * N-1, i=1,2 ..., 20,
Personnel selection face image pattern and 2 rank polynomial kernel function calculation obtain nuclear matrix K, and the element of the capable j row of the m of nuclear matrix K is
x
mAnd x
jRepresent m and j training sample in total training sample respectively.
Calculate new nuclear matrix K
A=A
TKA
Compute matrix D=A (K
A)
-1A
T
The matrix B of structure M * 19
Calculate new nuclear matrix K
B=B
T[K-2KDK+KDKDK] B, wherein B
TThe transposed matrix of expression B, matrix D=A (K
A)
-1A
T, matrix (K
A)
-1Expression K
AInverse matrix, A
TThe transposed matrix of expression A.
To K
BCarry out the Cholesky decomposition and obtain upper triangular matrix R
B
Facial image training sample with the i class
The nonlinear characteristic vector that calculates i class facial image sample is
Wherein I is a unit matrix,
N training sample representing the i class.x
mRepresent m training sample in total training sample.K represents 2 rank polynomial kernel functions.
(2) cognitive phase: calculate facial image sample x to be identified
TestThe nonlinear characteristic vector
H=[k (x wherein
1, x
Test), k (x
2, x
Test) ..., k (x
M, x
Test)]
T, k represents 2 rank polynomial kernel functions.
Cognitive phase is included into facial image sample to be identified in people's face classification of minor increment correspondence
Table 2 is the results with the test of UMIST face database, carries out 8 experiments altogether.To every class people's face, each experiment is picked at random N=3 respectively, and 5,7,9,11,13,15,17 samples are as training sample, remaining sample is as test sample book, and each experiment repeats 20 times, calculates average recognition rate, and the kernel function of the inventive method is with 2 rank polynomial kernel functions.The result of method of the present invention and Fisherface method is compared, and under identical training sample and test sample book condition, method of the present invention all is better than the Fisherface method.
Table 2 average recognition rate (%)
Claims (1)
1. the non-linear diagnostic characteristics of the facial image under the condition of small sample extracts and recognition methods, it is characterized in that comprising the steps:
(1) training stage:
1. calculate nuclear matrix K, structural matrix A calculates nuclear matrix K
A:
C people's face classification arranged, and everyone face classification has N facial image training sample, and total number of training is M=NC; With the facial image sample matrix that collects stretching be a vector, with vector representation facial image sample;
Calculate the nuclear matrix K of M * M, the element kernel function k (x of the capable j row of its m
m, x
j) calculate kernel function wherein
Be 2 rank polynomial kernel functions, vector x
mAnd x
jRepresent m and j facial image training sample in total training sample respectively,
Expression x
mThe transposition computing, m=1,2 ..., M, j=1,2 ..., M, down with; Structural matrix A=[A
1, A
2..., A
i..., A
C], A wherein
iBe the matrix of M * N-1, i=1,2 ..., C is defined as follows:
Matrix A
iIn: capable 0, the (i-1) N+1 capable being-1, the (i-1) N+1 that is is the unit matrix of a N-1 * N-1 below capable from the 1st row to (i-1) N, and all the other elements are 0;
Calculate nuclear matrix K with matrix K and A
A=A
TKA, the computing of subscript T representing matrix transposition;
2. structural matrix B calculates nuclear matrix K
B:
Structural matrix B is the matrix of M * C-1, and building method is as follows:
In the matrix B: the 1st the row be-1, the q * N+1 is capable, and q classifies 1 as, wherein the value of q is q=1,2 ..., C-1, all the other elements are 0;
Calculate nuclear matrix K
B:
K
B=B
T[K-2KDK+KDKDK]B
B wherein
TThe transposed matrix of expression B, matrix D=A (K
A)
-1A
T, matrix (K
A)
-1Expression K
AInverse matrix, A
TThe transposed matrix of expression A;
3. to nuclear matrix K
BCarry out the Cholesky decomposition and obtain upper triangular matrix R
B,
4. calculate the nonlinear characteristic vector Y of i class people face
i:
Wherein
N people's face image pattern representing i class people face, i=1,2 ..., C, n=1,2,3 ..., N; I representation unit battle array,
Expression R
BThe inverse matrix of transposed matrix;
(2) cognitive phase:
A) calculate facial image sample x to be identified
TestThe nonlinear characteristic vector Y
Test:
H=[k (x wherein
1, x
Test), k (x
2, x
Test) ..., k (x
M, x
Test)]
T
B) calculate Y
TestAnd Y
iMinor increment:
I=1,2 ..., C, wherein || Y
i-Y
Test|| expression Y
TestAnd Y
iEuclidean distance, min represents to ask minor increment; The criterion of identification is with facial image sample x to be identified
TestBe included in people's face classification of minor increment correspondence.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN2009102346624A CN101877065B (en) | 2009-11-26 | 2009-11-26 | Extraction and identification method of non-linear authentication characteristic of facial image under small sample condition |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN2009102346624A CN101877065B (en) | 2009-11-26 | 2009-11-26 | Extraction and identification method of non-linear authentication characteristic of facial image under small sample condition |
Publications (2)
Publication Number | Publication Date |
---|---|
CN101877065A true CN101877065A (en) | 2010-11-03 |
CN101877065B CN101877065B (en) | 2012-07-18 |
Family
ID=43019617
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN2009102346624A Expired - Fee Related CN101877065B (en) | 2009-11-26 | 2009-11-26 | Extraction and identification method of non-linear authentication characteristic of facial image under small sample condition |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN101877065B (en) |
Cited By (6)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN101984455A (en) * | 2010-12-01 | 2011-03-09 | 南京信息工程大学 | Method for solving linear discrimination vector in matrix rank spaces of between-class scatter and total scattering |
CN102013023A (en) * | 2010-12-01 | 2011-04-13 | 南京信息工程大学 | Rapid real-time extraction method of small sample linear identification characters |
CN102142082A (en) * | 2011-04-08 | 2011-08-03 | 南京邮电大学 | Virtual sample based kernel discrimination method for face recognition |
CN104463085A (en) * | 2013-09-23 | 2015-03-25 | 深圳市元轩科技发展有限公司 | Face recognition method based on local binary pattern and KFDA |
CN107392212A (en) * | 2017-07-19 | 2017-11-24 | 上海电机学院 | A kind of image information method for quickly identifying |
CN112232298A (en) * | 2020-11-05 | 2021-01-15 | 北京理工大学 | Face classification method for small sample training data |
-
2009
- 2009-11-26 CN CN2009102346624A patent/CN101877065B/en not_active Expired - Fee Related
Cited By (9)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN101984455A (en) * | 2010-12-01 | 2011-03-09 | 南京信息工程大学 | Method for solving linear discrimination vector in matrix rank spaces of between-class scatter and total scattering |
CN102013023A (en) * | 2010-12-01 | 2011-04-13 | 南京信息工程大学 | Rapid real-time extraction method of small sample linear identification characters |
CN102013023B (en) * | 2010-12-01 | 2013-02-27 | 南京信息工程大学 | Rapid real-time extraction method of small sample linear identification characters |
CN101984455B (en) * | 2010-12-01 | 2013-05-08 | 南京信息工程大学 | Method for solving linear discrimination vector in matrix rank spaces of between-class scatter and total scattering |
CN102142082A (en) * | 2011-04-08 | 2011-08-03 | 南京邮电大学 | Virtual sample based kernel discrimination method for face recognition |
CN104463085A (en) * | 2013-09-23 | 2015-03-25 | 深圳市元轩科技发展有限公司 | Face recognition method based on local binary pattern and KFDA |
CN107392212A (en) * | 2017-07-19 | 2017-11-24 | 上海电机学院 | A kind of image information method for quickly identifying |
CN112232298A (en) * | 2020-11-05 | 2021-01-15 | 北京理工大学 | Face classification method for small sample training data |
CN112232298B (en) * | 2020-11-05 | 2022-11-15 | 北京理工大学 | Face classification method for small sample training data |
Also Published As
Publication number | Publication date |
---|---|
CN101877065B (en) | 2012-07-18 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
Harada et al. | Discriminative spatial pyramid | |
Yang et al. | KPCA plus LDA: a complete kernel Fisher discriminant framework for feature extraction and recognition | |
Park et al. | Fingerprint classification using fast Fourier transform and nonlinear discriminant analysis | |
CN101877065B (en) | Extraction and identification method of non-linear authentication characteristic of facial image under small sample condition | |
Mashhoori et al. | Block-wise two-directional 2DPCA with ensemble learning for face recognition | |
CN102930300B (en) | Method and system for identifying airplane target | |
CN103020640A (en) | Facial image dimensionality reduction classification method based on two-dimensional principal component analysis | |
CN103745205A (en) | Gait recognition method based on multi-linear mean component analysis | |
CN106778714B (en) | LDA face identification method based on nonlinear characteristic and model combination | |
Dong et al. | Feature extraction through contourlet subband clustering for texture classification | |
CN107194314A (en) | The fuzzy 2DPCA and fuzzy 2DLDA of fusion face identification method | |
Chergui et al. | Lpq and ldp descriptors with ml representation for kinship verification | |
CN103714340A (en) | Self-adaptation feature extracting method based on image partitioning | |
Dai et al. | Face recognition using support vector machines with the robust feature | |
CN103632145A (en) | Fuzzy two-dimensional uncorrelated discriminant transformation based face recognition method | |
Khalid et al. | Tropical wood species recognition system based on multi-feature extractors and classifiers | |
CN102332087A (en) | Face recognition method based on sparse representation | |
Xu et al. | Spectral-spatial subspace clustering for hyperspectral images via modulated low-rank representation | |
CN102289679B (en) | Method for identifying super-resolution of face in fixed visual angle based on related characteristics and nonlinear mapping | |
CN104361337A (en) | Sparse kernel principal component analysis method based on constrained computation and storage space | |
Li et al. | Face recognition using Gabor-based complete Kernel Fisher Discriminant analysis with fractional power polynomial models | |
Khairuddin et al. | Using two stage classification for improved tropical wood species recognition system | |
Tabejamaat et al. | A coding-guided holistic-based palmprint recognition approach | |
Shao et al. | An NMF-based method for the fingerprint orientation field estimation | |
CN101984455B (en) | Method for solving linear discrimination vector in matrix rank spaces of between-class scatter and total scattering |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
C06 | Publication | ||
PB01 | Publication | ||
C10 | Entry into substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
C14 | Grant of patent or utility model | ||
GR01 | Patent grant | ||
CF01 | Termination of patent right due to non-payment of annual fee |
Granted publication date: 20120718 Termination date: 20141126 |
|
EXPY | Termination of patent right or utility model |