CN107341485A - Face identification method and device - Google Patents
Face identification method and device Download PDFInfo
- Publication number
- CN107341485A CN107341485A CN201710631337.6A CN201710631337A CN107341485A CN 107341485 A CN107341485 A CN 107341485A CN 201710631337 A CN201710631337 A CN 201710631337A CN 107341485 A CN107341485 A CN 107341485A
- Authority
- CN
- China
- Prior art keywords
- mrow
- matrix
- msup
- coefficient matrix
- sample set
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Granted
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V40/00—Recognition of biometric, human-related or animal-related patterns in image or video data
- G06V40/10—Human or animal bodies, e.g. vehicle occupants or pedestrians; Body parts, e.g. hands
- G06V40/16—Human faces, e.g. facial parts, sketches or expressions
- G06V40/172—Classification, e.g. identification
-
- 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
Landscapes
- Engineering & Computer Science (AREA)
- Health & Medical Sciences (AREA)
- General Health & Medical Sciences (AREA)
- Oral & Maxillofacial Surgery (AREA)
- Human Computer Interaction (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Multimedia (AREA)
- Theoretical Computer Science (AREA)
- Other Investigation Or Analysis Of Materials By Electrical Means (AREA)
- Image Analysis (AREA)
Abstract
The invention discloses a kind of face identification method and device, belong to field of biological recognition.Methods described includes:Collection is pre-processed to the sample identified, obtains sample set matrix to be identified;The sample set matrix to be identified is handled using the sparse convex Non-negative Matrix Factorization method of new iteration, obtains the coefficient matrix and optimal coefficient matrix of the optimal basic matrix of the sample set to be identified;The optimal coefficient matrix of the sample set to be identified is classified using the grader trained, completes recognition of face.The present invention is iterated optimization using new rule of iteration and carries out rarefaction to the coefficient matrix of basic matrix during iteration, so as to which iterative goes out the coefficient matrix and optimal coefficient matrix of the optimal basic matrix of sample set to be identified, improve discrimination, reduce operand so that the final face identification method has higher discrimination and shorter operation time.
Description
Technical field
The present invention relates to field of biological recognition, more particularly to a kind of face identification method and device.
Background technology
Recognition of face is a kind of emerging biological identification technology, due to having untouchable, friendly, use in application aspect
The advantages that convenient, directly perceived, make its had a wide range of applications in fields such as criminal's identification, certificate verification and medical science with it is huge
Market potential.
Face recognition technology common at present can be divided into several classes:Identification technology based on geometric properties, based on mathematical modeling
Identification technology, identification technology based on subspace analysis etc..Identification technology based on subspace analysis is current face's identification
In one of main stream approach, its basic thought be by a mapping by the facial image in higher dimensional space project to one it is low
In n-dimensional subspace n, Classification and Identification is carried out to characteristic coefficient in this lower-dimensional subspace.Traditional sub-space analysis method is typically adopted
With principal component analysis (Principal Components Analysis, PCA), sparse Non-negative Matrix Factorization (Sparse Non-
Negative Matrix Factorization, SNMF), the Feature Dimension Reduction such as convex Non-negative Matrix Factorization (Convex NMF, CNMF)
Method.
Non-negative Matrix Factorization is the decomposition that matrix is realized under conditions of all elements of matrix are non-negative.Image intensity value
Nonnegativity cause the more unconfined principal component analysis of Non-negative Matrix Factorization to have more interpretation.Directly use Non-negative Matrix Factorization
(NMF) method carries out face characteristic extraction, because the coefficient matrix of basic matrix does not have optimised and sparse, causes face identification rate
It is not high;CNMF is NMF popularization, and it is explanatory to may be such that data more have, can improve face identification rate to a certain extent.CNMF be by
Half Non-negative Matrix Factorization (Semi-NMF) develops, in Semi-NMF, X=FGT, F and X are abandoned, only require G
Non-negative, Ding et al. is substituted for matrix F original matrix X non-negative convex combination, i.e. F=XW, and then obtains a kind of new decomposition
Form X=XWGT, wherein F and G be constrained for nonnegative matrix, and X is not constrained, and then proposes CNMF mathematical modelings.Clearly
This decomposed form has expanded NMF application, data is more had explanatory.In CNMF, G is coefficient matrix, and F is group moment
Battle array, because F=XW, W are basic matrix F coefficient matrixes.
During the present invention is realized, inventor has found that prior art at least has problems with:
Existing CNMF decomposition methods using traditional multiplying property rule of iteration, cause the coefficient matrix of basic matrix inadequate
Optimization, discrimination be not high;Meanwhile existing CNMF does not carry out threshold value rarefaction to the coefficient matrix of basic matrix so that feature system
Number is excessively scattered, causes discrimination not high, and does not use the calculating of the coefficient matrix of the basic matrix of threshold value rarefaction complex,
Operand is excessive, and speed is excessively slow, and with K increase, CNMF methods discrimination is to reduce on the contrary, is unfavorable for later stage face figure
The reconstruction of picture.
The content of the invention
The face identification rate of CNMF in the prior art is not high, operand is big, is unfavorable for the problems such as image reconstruction in order to solve,
The embodiments of the invention provide a kind of face identification method and device.The technical scheme is as follows:
In a first aspect, the embodiments of the invention provide a kind of face identification method, methods described includes:To the sample identified
Collection is pre-processed, and obtains sample set matrix to be identified;Using the sparse convex Non-negative Matrix Factorization method of new iteration to described to be identified
Sample set matrix is handled, and obtains the coefficient matrix and optimal coefficient matrix of the optimal basic matrix of the sample set to be identified,
The coefficient matrix and optimal coefficient matrix of the optimal basic matrix of the sample set to be identified are produced using following iterative formula iteration:
Wherein, S is the sample set matrix to be identified, and size is J × I, and J, I are positive integer, and J is the sample to be identified
The characteristics of low-frequency dimension of each sample of this concentration, I be the sample set to be identified sample size, STFor S transposed matrix;U
The coefficient matrix of the basic matrix obtained for nth iteration, Q are the coefficient matrix that nth iteration obtains, and U' is (n-1)th iteration
The coefficient matrix of obtained basic matrix, Q' are the coefficient matrix that (n-1)th iteration obtains, and U, Q, U' and Q' size are I × K, K
S characteristic dimension is represented, K is positive integer and K≤J, n are the positive integer more than 1;U'TFor U' transposed matrix, Q'TFor turning for Q'
Put matrix, QTFor Q transposed matrix;UikThe element arranged for U the i-th row kth, QikThe element arranged for Q the i-th row kth, i, k are equal
For positive integer, and i≤I, k≤K;U'ikThe element arranged for U' the i-th row kth, Q'ikThe element arranged for Q' the i-th row kth;Work as J
(V) during value minimum, U is the coefficient matrix of optimal basic matrix, and Q is optimal coefficient matrix, J (V)=Tr (- 2VTF+-VTE-VD),
Wherein F+=(STS)+Q, E-=(STS)-, D=QTQ, V=U, VTFor V transposed matrix;
The optimal coefficient matrix of the sample set to be identified is classified using the grader trained, face is completed and knows
Not.
It is described to use the sparse convex Non-negative Matrix Factorization method of new iteration to institute in a kind of implementation of the embodiment of the present invention
State sample set matrix to be identified to be handled, obtain the coefficient matrix and most major clique of the optimal basic matrix of the sample set to be identified
Matrix number, including:K value is determined in setting range;For the K determined value, using the sparse convex nonnegative matrix of new iteration
Decomposition method is decomposed to the sample set matrix to be identified, obtains the optimal basic matrix of the sample set to be identified corresponding to K
Coefficient matrix and optimal coefficient matrix.
It is described for the K determined value in another implementation of the embodiment of the present invention, it is sparse using new iteration
Convex Non-negative Matrix Factorization method is decomposed to the sample set matrix to be identified, obtains the sample set to be identified corresponding to K
The coefficient matrix and optimal coefficient matrix of optimal basic matrix, including:The coefficient matrix of initial basic matrix and initial system are determined according to K
Matrix number;Meter is iterated according to the coefficient matrix of the initial basic matrix and initial coefficients matrix and the iterative formula
Calculate;The coefficient matrix and coefficient matrix for the basic matrix that each step is iterated to calculate out substitute into object function:
J (V)=Tr (- 2VTF+-VTE-VD), wherein F+=(STS)+Q, E-=(STS)-, D=QTQ, V=U;
When the value of the object function reaches stable state, terminate iterative calculation, and will iterate to calculate out for the last time
Coefficient matrix and most major clique of the coefficient matrix and coefficient matrix of basic matrix as the optimal basic matrix of the sample set to be identified
Matrix number, wherein, the stable state refers to that the value of the object function keeps constant or amplitude of fluctuation to be less than predetermined amplitude;
Or when iterations reaches iterations threshold value, choose the coefficient matrix of basic matrix that last time iterates to calculate out and
Coefficient matrix and optimal coefficient matrix of the coefficient matrix as optimal basic matrix.
It is described for the K determined value in another implementation of the embodiment of the present invention, it is sparse using new iteration
Convex Non-negative Matrix Factorization method is decomposed to the sample set matrix to be identified, obtains the sample set to be identified corresponding to K
The coefficient matrix and optimal coefficient matrix of optimal basic matrix, in addition to:Determine rarefaction threshold value;After the iterative calculation of each step,
Each numerical value and the size of rarefaction threshold value in the coefficient matrix for the basic matrix that judgement iterates to calculate out;By what is iterated to calculate out
Numerical value in the coefficient matrix of basic matrix more than rarefaction threshold value is arranged to 1, by the coefficient matrix of the basic matrix iterated to calculate out
In be less than or equal to rarefaction threshold value numerical value be arranged to 0.
In another implementation of the embodiment of the present invention, methods described also includes:Training sample set is located in advance
Reason, obtains training sample set matrix;The training sample set matrix is carried out using new iteration sparse convex Non-negative Matrix Factorization method
Processing, the coefficient matrix and optimal coefficient matrix of the optimal basic matrix of the training sample set are obtained, the optimal basic matrix
Coefficient matrix and optimal coefficient matrix use identical iterative formula iteration when being handled with the sample set to be identified to produce;Using
The optimal coefficient matrix training grader of the training sample set.
Second aspect, the embodiment of the present invention additionally provide a kind of face identification device, and described device includes:Pretreatment is single
Member, pre-processed for collecting to the sample identified, obtain sample set matrix to be identified;Resolving cell, for using new iteration
Sparse convex Non-negative Matrix Factorization method is handled the sample set matrix to be identified, obtains the optimal of the sample set to be identified
The coefficient matrix and optimal coefficient matrix of basic matrix, the coefficient matrix and most major clique of the optimal basic matrix of the sample set to be identified
Matrix number is produced using following iterative formula iteration:
Wherein, S is the sample set matrix to be identified, and size is J × I, and J, I are positive integer, and J is the sample to be identified
The characteristics of low-frequency dimension of each sample of this concentration, I be the sample set to be identified sample size, STFor S transposed matrix;U
The coefficient matrix of the basic matrix obtained for nth iteration, Q are the coefficient matrix that nth iteration obtains, and U' is (n-1)th iteration
The coefficient matrix of obtained basic matrix, Q' are the coefficient matrix that (n-1)th iteration obtains, and U, Q, U' and Q' size are I × K, K
S characteristic dimension is represented, K is positive integer and K≤J, n are the positive integer more than 1;U'TFor U' transposed matrix, Q'TFor turning for Q'
Put matrix, QTFor Q transposed matrix;UikThe element arranged for U the i-th row kth, QikThe element arranged for Q the i-th row kth, i, k are equal
For positive integer, and i≤I, k≤K;U'ikThe element arranged for U' the i-th row kth, Q'ikThe element arranged for Q' the i-th row kth;Work as J
(V) during value minimum, U is the coefficient matrix of optimal basic matrix, and Q is optimal coefficient matrix, J (V)=Tr (- 2VTF+-VTE-VD),
Wherein F+=(STS)+Q, E-=(STS)-, D=QTQ, V=U, VTFor V transposed matrix;
Taxon, for being divided using the grader trained the optimal coefficient matrix of the sample set to be identified
Class, complete recognition of face.
In a kind of implementation of the embodiment of the present invention, the resolving cell, for determining K's in setting range
Value;For the K determined value, the sample set matrix to be identified is carried out using the sparse convex Non-negative Matrix Factorization method of new iteration
Decompose, obtain the coefficient matrix and optimal coefficient matrix of the optimal basic matrix of the sample set to be identified corresponding to K.
In another implementation of the embodiment of the present invention, the resolving cell, for determining initial basic matrix according to K
Coefficient matrix and initial coefficients matrix;According to the coefficient matrix of the initial basic matrix and initial coefficients matrix and it is described repeatedly
Calculating is iterated for formula;The coefficient matrix and coefficient matrix for the basic matrix that each step is iterated to calculate out substitute into target letter
Number:
J (V)=Tr (- 2VTF+-VTE-VD), wherein F+=(STS)+Q, E-=(STS)-, D=QTQ, V=U;
When the value of the object function reaches stable state, terminate iterative calculation, and will iterate to calculate out for the last time
Coefficient matrix and most major clique of the coefficient matrix and coefficient matrix of basic matrix as the optimal basic matrix of the sample set to be identified
Matrix number, wherein, the stable state refers to that the value of the object function keeps constant or amplitude of fluctuation to be less than predetermined amplitude
(predetermined amplitude can be set according to being actually needed);Or when iterations reaches iterations threshold value, choose last
Coefficient matrix and optimal coefficient square of the coefficient matrix and coefficient matrix of the secondary basic matrix iterated to calculate out as optimal basic matrix
Battle array.
In another implementation of the embodiment of the present invention, the resolving cell, it is additionally operable to determine rarefaction threshold value;
After the iterative calculation of each step, the big of each numerical value in the coefficient matrix of basic matrix that iterates to calculate out and rarefaction threshold value is judged
It is small;The numerical value for being more than rarefaction threshold value in the coefficient matrix of the basic matrix iterated to calculate out is arranged to 1, by what is iterated to calculate out
Numerical value in the coefficient matrix of basic matrix less than or equal to rarefaction threshold value is arranged to 0.
In another implementation of the embodiment of the present invention, described device also includes training unit;The pretreatment is single
Member, it is additionally operable to pre-process training sample set, obtains training sample set matrix;The resolving cell, it is additionally operable to using new
The sparse convex Non-negative Matrix Factorization method of iteration is handled the training sample set matrix, obtains the optimal of the training sample set
The coefficient matrix and optimal coefficient matrix of basic matrix, the coefficient matrix and optimal coefficient matrix of the optimal basic matrix use and institute
Identical iterative formula iteration produces when stating sample set processing to be identified;The training unit, for using the training sample
The optimal coefficient matrix training grader of collection.
The third aspect, the embodiments of the invention provide a kind of face identification device, described device includes:Memory, with depositing
The processor of reservoir connection, the memory are used to store software program and module, when the processor is used to run or hold
When row is stored in the software program and module in the memory, the method described in first aspect can be performed.
Fourth aspect, the embodiment of the present invention additionally provide a kind of computer-readable medium, are filled for storing for recognition of face
The program code of execution is put, described program code includes performing the instruction of the method described in first aspect.
The beneficial effect that technical scheme provided in an embodiment of the present invention is brought is:
The present invention is iterated the coefficient matrix and optimal coefficient square of the optimal basic matrix of Optimization Solution using new rule of iteration
Battle array, and rarefaction has been carried out to the coefficient matrix of basic matrix in an iterative process, new rule of iteration and LS-SVM sparseness are than tradition
Multiplying property rule of iteration it is more excellent, the characteristic of the coefficient matrix of obtained optimal basic matrix is more concentrated, so that optimal
The weight distribution of coefficient matrix is more concentrated, is easier to classify, and effectively increases face identification rate, and reduces operand, finally
So that the face identification method has higher discrimination and shorter run time, discrimination is up to 100%.And the present invention
Method gradually increases with K increase, discrimination, is advantageous to the reconstruction of later stage facial image.
Brief description of the drawings
Technical scheme in order to illustrate the embodiments of the present invention more clearly, make required in being described below to embodiment
Accompanying drawing is briefly described, it should be apparent that, drawings in the following description are only some embodiments of the present invention, for
For those of ordinary skill in the art, on the premise of not paying creative work, other can also be obtained according to these accompanying drawings
Accompanying drawing.
Fig. 1 is face identification method flow chart provided in an embodiment of the present invention;
The basic matrix image that Fig. 2 a- Fig. 2 d are PCA and various NMF methods obtain;
Fig. 3 is the face identification rate of PCA and various NMF methods with K change schematic diagram
Fig. 4 is the run time of PCA and various NMF methods with K change schematic diagram
Fig. 5 is face identification device structural representation provided in an embodiment of the present invention.
Embodiment
To make the object, technical solutions and advantages of the present invention clearer, below in conjunction with accompanying drawing to embodiment party of the present invention
Formula is described in further detail.
Fig. 1 is a kind of flow chart of face identification method provided in an embodiment of the present invention, and referring to Fig. 1, this method includes:
Step S11:Training sample set is obtained, and is expressed as initial matrix.
In embodiments of the present invention, training sample set is obtained to include but is not limited to obtain training from existing Network Picture Database
Sample set, or training sample set is made according to image, image here can shoot acquisition in advance.
It is for instance possible to use in ORL (Olivetti Research Laboratory) picture library that Cambridge University provides
Face is as training sample set.The facial image of the different expressions of 40 people is have collected in ORL picture libraries altogether, everyone each 10 images
Totally 400, every image has 256 gray levels, and size is 112 × 92.Wherein everyone countenance and facial detail
There is different degrees of change, such as laugh at not laughing at, eyes are opened and are closed, worn with not wearing glasses;Human face posture also has certain degree
Change, depth rotation and plane are rotated up to up to 20 degree;The yardstick of face also has up to 10% change.To everyone with
Machine chooses 5 image composition training sample sets.So training sample, which is concentrated, 200 facial images.Certainly, training sample here
The acquisition of this collection is only for example, and can also be realized in practice by the image that other databases or user itself prepare.
Training sample set generally includes multiple sample images, this one matrix of multiple image constructions.With above-mentioned training sample
Exemplified by collection, 200 sample images, each sample image includes 10304 (112 × 92) individual pixels, then this training sample set structure
Into initial matrix X' size be 10304 × 200.
Step S12:Training sample set is pre-processed, obtains training sample set matrix, the training sample set matrix is
The matrix of the characteristics of low-frequency information of every image is concentrated comprising training sample.
Because training sample set has been expressed as initial matrix by step S11, thus step S12 really using initial matrix as
The pretreatment that object is carried out.
Due in recognition of face, the influence of the external environment condition such as illumination condition, picture pick-up device, and expression, attitudes vibration,
The face Self-variation such as age, covering, all make the defects of acquisition image often there are noise, contrast is inadequate, data are with showing
Differ greatly, the discrimination of algorithm is influenceed very big in fact.In order to ensure that the feature of extraction has preferable robust to face change
Property to facial image, it is necessary to pre-process.By being pre-processed to facial image, certain noise and illumination can be removed
Influence, reduce interference of the high-frequency information to discrimination.In embodiments of the present invention, carrying out pretreatment to facial image can wrap
Include following steps:
The first step, gray scale normalization is carried out to facial image.
Gray scale normalization is used for compensating the uneven illumination of original image, so as to overcome illumination variation to bring identification
Influence, there is certain robustness.Main process is as follows:The gray average and variance of given image, will with linear method
Set-point assigns gray average and variance, so can be with the brightness and contrast of unified image so that face images are all abided by
Follow same or analogous intensity profile.By carrying out gray scale normalization to image, illumination variation can be overcome to recognition effect
Influence.
Second step, facial image low-frequency information is extracted using wavelet transformation.
Wavelet transformation is the partial transformation of time and frequency, information can be more efficiently extracted from signal and analysis is local
Signal, there is the ability of very strong sign signal local feature in time domain, frequency domain.
With the feature of small echo extraction facial image, the main low-frequency information using small echo extraction facial image, conduct is reduced
The image high-frequency information of noise is identifying interference when classifying.The facial image of input passes through (one layer of two-dimensional discrete wavelet conversion
Wavelet decomposition) after can produce 4 subgraphs (LL, LH, HL, HH).Wherein, LL is the low-frequency component (low frequency containing facial image
Information), the most information of original image is contained, can be as the approximation of original image, while suppressed and made an uproar at random significantly
The high-frequency informations such as sound.Continue to obtain including a large amount of low-frequency informations of sample after carrying out two-dimensional discrete wavelet conversion to LL low frequency subgraph pictures
Matrix X.
Training sample set matrix X comprising low-frequency information (characteristics of low-frequency), the matrix are finally given by above-mentioned pretreatment
Size be J × I, J is the characteristics of low-frequency dimension that training sample concentrates each sample, and I is the sample size that training sample is concentrated,
J, I is positive integer.So that size above is 10304 × 200 initial matrix X' as an example, matrix X' is after above-mentioned pretreatment
Obtained training sample set matrix X size is 2784 × 200.
The embodiment of the present invention this, preprocess method is not limited to the first step included by above-mentioned steps S12 to second step, also
Other modes can be used to realize, such as mean filter, medium filtering.The present invention is without limitation.
Step S13:Training sample set matrix is handled using new iteration sparse convex Non-negative Matrix Factorization method, obtains instruction
Practice the coefficient matrix and optimal coefficient matrix of the optimal basic matrix of sample set, the coefficient matrix of the optimal basic matrix of training sample set
Produced with optimal coefficient matrix using following iterative formula iteration:
Wherein, X is the training sample set matrix, and size is J × I, and J, I are positive integer, and J is the training sample set
In each sample characteristics of low-frequency dimension, I is the sample size of the training sample set, XTFor X transposed matrix;W is n-th
The coefficient matrix for the basic matrix that iteration obtains, G are the coefficient matrix that nth iteration obtains, and W' is the base that (n-1)th iteration obtains
The coefficient matrix of matrix, G' are the coefficient matrix that (n-1)th iteration obtains, and W, G, W' and G' size are I × K, and K represents X's
Characteristic dimension, K is positive integer and K≤J, n are the positive integer more than 1;W'TFor W' transposed matrix, G'TFor G' transposed matrix,
GTFor G transposed matrix;WikThe element arranged for W the i-th row kth, GikThe element arranged for G the i-th row kth, i, k are just whole
Number, and i≤I, k≤K;W'ikThe element arranged for W' the i-th row kth, G'ikThe element arranged for G' the i-th row kth;When J's (H)
When being worth minimum, W is the coefficient matrix of optimal basic matrix, and G is optimal coefficient matrix, J (H)=Tr (- 2HTB+-HTA-HC), wherein B+
=(XTX)+G, A-=(XTX)-, C=GTG, H=W, HTFor H transposed matrix.
Wherein, training sample set matrix namely the matrix that training sample set obtains after pre-processing above, W is specially n-th
The coefficient matrix of the basic matrix for the training sample set matrix that iteration obtains, G are the training sample set matrix that nth iteration obtains
Coefficient matrix.
Each element in the coefficient matrix and coefficient matrix of basic matrix can be iterated using above-mentioned formula, these
Element forms the coefficient matrix and coefficient matrix of basic matrix, so by the iteration of element in matrix, that is, completes to basic matrix
Coefficient matrix and coefficient matrix interative computation.
In above-mentioned formula, [(XTX)+G']ikRepresenting matrix (XTX)+It is multiplied by G'GainedThe member of the i-th row kth row of the matrix arrived
Element;Other bracket operations are similar therewith, repeat no more here.
That is (XTX)+Representing matrix (XTX element) takes absolute value again and matrix
(XTX corresponding element) is added resulting matrix;That is (XTX)-Representing matrix (XTX)
Element takes absolute value subtracts matrix (X againTX the matrix obtained by corresponding element).
For example, X size is 2784 × 200, as K=30, i.e., characteristic dimension be reduced to 30, W size for 200 ×
30, G size is 200 × 30;Or as K=25, the size that now characteristic dimension is reduced to 25, W is 200 × 25, G's
Size is 200 × 25.
In step s 13, iterative formula determines to obtain according to the sparse convex Non-negative Matrix Factorization method of new iteration, its principle
It is as follows:
First, basic CNMF object function is:
J ' (H)=Tr (- 2HTB+-HTA-HC+2HTB-+HTA+HC);
A new object function J (H) is invented herein as shown in formula (1), and new J (H) is simpler than traditional J ' (H) expression
Singly it is easy to calculate, convergence rate can be effectively improved, and it is more preferable to W, G effect of optimization, face identification rate is higher.
J (H)=Tr (- 2HTB+-HTA-HC) (1)
(1) in formula, B+=(XTX)+G,A-=(XTX)-, C=GTG, H=W, and Wherein, | (XTX)ik| represent to matrix XTEach element in X takes absolute value.
Define a binary auxiliary function Z (H, H ') on H and H', it is desirable to which it meets formula (2).
Z (H, H ') >=J (H), Z (H, H)=J (H) (2)
Define HminH value when binary function Z (H, H') takes minimum value during for for giving H', as shown in formula (3):
Understand:
J (H ')=Z (H ', H ') >=Z (Hmin,H′)≥J(Hmin)
Therefore, as long as finding the Z (H, H ') of eligible (2), it is ensured that object function J (H) is a nonincreasing function, i.e. mesh
Scalar functions are restrained.The object function as shown in formula (4) of equivalence is obtained as formula (1).
J (H)=Tr (- 2HTB--HTA+HC-2HTB++2HTB-+HTA+HC-HTA-HC) (4)
Wherein:
Obtain:
Wherein, HikAnd HilThe element of representing matrix the i-th rows of H kth row and l row respectively, k and l can with it is equal can not also
It is equal, H'jkThe element of representing matrix H' jth rows kth row, j and i can with it is equal can also be unequal.
Therefore, the Z (H, H ') of eligible (2) is defined by formula (5).To meet formula (3), demand goes out Z (H, H ') and taken most
H during small value, then the derivative that can make Z (H, H ') is 0, obtains formula (6).
Z (H, H ') minimum value is calculated by formula (6), you can obtains making Z take H values during minimum value (i.e. in formula (3)
Hmin), as shown in formula (7).
Substitution formula (8) is arrived in formula (7),
B+=(XTX)+G,B-=(XTX)-G, A=XTX, C=GTG, H=W (8)
Shown in the rule of iteration such as formula (9) that can obtain W.
Shown in the iterative formula such as formula (10) that similarly can obtain G:
Therefore, the rule of iteration as defined in formula (9), (10) can ensure J (H ')=Z (H ', H ') >=Z (H, H ') >=J
(H) so that object function J (H) is a nonincreasing function, i.e. formula (9), (10) can ensure object function stable convergence.
In above-mentioned formula, the coefficient matrix W and optimal coefficient matrix G of optimal basic matrix disclosure satisfy that target function value J
(H) it is minimum so that object function stable convergence.
In embodiments of the present invention, step S13 can include:
K value is determined in setting range;For the K determined value, using the sparse convex Non-negative Matrix Factorization of new iteration
Method is decomposed to matrix, obtains the coefficient matrix and optimal coefficient matrix of the optimal basic matrix of training sample set corresponding to K.
In embodiments of the present invention, setting range is preferably 180 >=K >=20.When K values are too small, dimension is down to too low lead
Cause Character losing serious, and as K increase, facial contour each several part major organs can gradually appear, detail section is increasingly
Clearly, i.e. K value is bigger, and the effect of reconstruction image is better, so K value is unsuitable too small, when K is less than 20, basic matrix
Coefficient matrix dimension it is too low, then feature is excessively lost, can cause later image rebuild distortion it is serious;When K is close to J, weight
The facial image built is clear as original image, without visual any difference.But K, which crosses conference, to be caused to calculate the time
It is long, and discrimination has arrived at the upper limit, will not infinitely increase with K increase, so K value is unsuitable excessive, when K surpasses
When 180, run time is long and discrimination holding 100% is constant.Therefore in the embodiment of the present invention, set K scopes as 180 >=
K≥20。
Specifically, for determination K value, using the sparse convex Non-negative Matrix Factorization method of new iteration to the non-of training sample set
Negative matrix is decomposed, including:
The first step, the coefficient matrix and initial coefficients matrix of initial basic matrix are determined according to K.In embodiments of the present invention,
The first step is completed in the following way:According to the coefficient matrix of initial basic matrix and the dimension of initial coefficients matrix produce respectively with
Machine matrix.Random matrix W and I × K that dimension is I × K random matrix G are produced, and random number is between 0-1.
Second step, meter is iterated according to the coefficient matrix of initial basic matrix and initial coefficients matrix and iterative formula
Calculate.
3rd step, the coefficient matrix and coefficient matrix of the basic matrix that each step is iterated to calculate out substitute into object function:
J (H)=Tr (- 2HTB+-HTA-HC), wherein B+=(XTX)+G, A-=(XTX)-, C=GTG, H=W;
When the value of object function reaches stable state, end iterative calculation, and the group moment that will be iterated to calculate out for the last time
Coefficient matrix and optimal coefficient matrix of the coefficient matrix and coefficient matrix of battle array as the optimal basic matrix of training sample set;Or
Person, when iterations reaches iterations threshold value, choose the coefficient matrix for the basic matrix that last time iterates to calculate out and be
Coefficient matrix and optimal coefficient matrix of the matrix number as the optimal basic matrix of training sample set.
Calculating is iterated using formula (9), (10), each iteration goes out new WikAnd Gik, will be respectively by WikAnd GikForm
Matrix W and G substitute into formula (1) calculate J (H) value;As iterations increases, J (H) value constantly reduces, when J's (H)
When value reaches stable state, terminate iteration, and the W that last time iteration is obtainedikAnd GikThe W formed and G matrix conduct
The coefficient matrix and optimal coefficient matrix of optimal basic matrix.Here stable state refers to, J (H) value keeps constant or become
Dynamic amplitude is less than predetermined amplitude, such as when J (H) value change is less than one thousandth, it is believed that J (H) value reaches stably shape
State.Further, since formula (9), (10) can be such that object function J (H) does not increase and tend to convergence stabilization, so directly being chosen after stable
Coefficient matrix and optimal coefficient matrix of the iteration result of last time as optimal basic matrix.
Or when iterations reaches predetermined iterations, the W and G that selection last time iteration obtains are as most
The coefficient matrix and optimal coefficient matrix of excellent basic matrix.For example, predetermined iterations is 500 times, and terminate in 500 iteration
When, J (H) value does not keep constant yet, coefficient matrixes of the W and G that selection last time iteration obtains as optimal basic matrix
With optimal coefficient matrix.
Need to carry out rarefaction to the coefficient matrix of basic matrix while being iterated.The coefficient square of rarefaction basic matrix
The process of battle array includes:Determine rarefaction threshold value;After the iterative calculation of each step, the coefficient square of basic matrix iterated to calculate out is judged
Each numerical value and the size of rarefaction threshold value in battle array;Rarefaction threshold will be more than in the coefficient matrix of the basic matrix iterated to calculate out
The numerical value of value is arranged to 1, and the numerical value that rarefaction threshold value is less than or equal in the coefficient matrix of the basic matrix iterated to calculate out is set
It is set to 0.During next step iteration, using the coefficient matrix of the basic matrix after the rarefaction.The rarefaction threshold value pre-sets for one,
It can be given in advance according to being actually needed.
0,1 matrix is turned to by the coefficient matrix of basic matrix is sparse using thresholding method in iteration, may be such that image base
Characteristic coefficient data in the coefficient matrix of matrix are more concentrated, are more sparse, more prominent face characteristic, can effectively extract energy
The characteristic coefficient of outstanding behaviours face characteristic, and the weight coefficient in corresponding coefficient matrix G is more concentrated, it is easier to point
Class, so as to improve discrimination;Meanwhile the matrix rarefaction after processing, matrix operation amount is reduced, accelerates the calculating of the present invention
Speed.
In embodiments of the present invention, W', G' initial value are all the random number matrix between 0-1, therefore, rarefaction threshold value
Selected scope between 0-1.Preferably, the rarefaction threshold value could be arranged to 0.05.
Step S14:Grader is trained using the optimal coefficient matrix of training sample set.
In embodiments of the present invention, grader is classified for SVMs (Support Vector Machine, SVM)
Device.SVM is substantially a two classification device, and it is a typical more classification that the face of multiple classifications is trained into classification
Problem.The more classification problems of SVM processing can use " one-to-one " and " one-to-many " two kinds of strategies, and " one-to-one " tactful classification
As a result it is more accurate.Therefore, the N classes of sample are carried out pairwise classification, construct N by the present invention using the strategy of " one-to-one "
(N-1)/2 grader.For example, when the classification sum N of face sample is 40, using 780 classification of " one-to-one " method construct
Device.
The dimension that different K value corresponds to the coefficient matrix of the optimal basic matrix of training sample set is different, and dimension is bigger, special
Sign coefficient data reservation is more, and Classification and Identification rate is higher.In step S14, it is preferable that using K=180 training sample set most
Major clique matrix number trains grader, ensures higher discrimination.
Trained the optimal coefficient matrix G of training sample set and class label matrix Y as the input of above-mentioned SVM classifier
Collection, grader is trained with the training set.Class label matrix is the square for marker samples classification used when doing two classification
Battle array, its data only have two values such as 0,1, and each value represents the classification belonging to face sample respectively, and such as 1 represents one kind, and 0 represents
Another kind of (two classification).
Specifically training process is:The size for decomposing the optimal coefficient matrix G of obtained training sample set is 200 × K.Because
The face training sample of every class people is 5 images in 40 class people, so to distinguish pth class (39 >=p >=1) and q classes (40 >=q
>=p+1) sample (two classification) when, pth class is belonged in G shares 5 samples, and belong to q classes shares 5 samples.Pth class sample
This composition size be 5 × K (180 >=K >=20) matrix V 1, its class label matrix be 5 × 1 complete 1 column vector Y1, q classes
Sample forms the matrix V 2 that size is 5 × K (180 >=K >=20), and its class label matrix is the column vector Y2 of full 0.By matrix V 1
With V2 be combined into size be 10 × K sample matrix V, Y1 and Y2 be combined into 10 × 1 class label matrix Y, by V, Y matrix make
For the input training set of SVM classifier, the two graders ginseng that can correctly divide pth class and q class samples is calculated by SVM algorithm
Number information.P is from 1 continuous value to 39, while q needs 40 (40-1)/2 time SVM to calculate altogether from the continuous values of p+1 to 40, will be every
The continuous parameters of secondary correct two classification calculated are stored in a file, you can are obtained multi-categorizer Parameter File, classified
When call this document, obtain the parameter information of multi-categorizer.
Step S15:Collection is pre-processed to the sample identified, obtains sample set matrix to be identified.
In embodiments of the present invention, sample set to be identified both can be the collection for being actually needed the facial image classified
Close or the set of the facial image for being tested, the mode that the present invention obtains to it are not limited.Sample to be identified
This collection is also to be made up of multiple facial images, for example, forming 200 samples by multiple facial images of multiclass people (such as 40 class people)
Image, each sample image include 10304 (112 × 92) individual pixels, then the initial matrix S' that this sample set to be identified is formed
Size be 10304 × 200.
To training in process and step S12 that the initial matrix collected to the sample identified in step S15 is pre-processed
The process that the initial matrix of sample set is pre-processed is identical, repeats no more here.By S12 preprocessing process, finally give
Sample set matrix S to be identified comprising low-frequency information, the size of the matrix is J × I, and J is each sample in sample set to be identified
Characteristics of low-frequency dimension, I be sample set to be identified in sample size, J, I are positive integer.Using size above as 10304 ×
Exemplified by 200 initial matrix S', the sample set matrix S to be identified obtained after pretreatment size is 2784 × 200.
Step S16:Collect matrix to the sample identified using the sparse convex Non-negative Matrix Factorization method of new iteration to be handled, obtain
The coefficient matrix and optimal coefficient matrix of the optimal basic matrix of sample set to be identified, the optimal basic matrix of sample set to be identified are
Matrix number and optimal coefficient matrix are produced using following iterative formula iteration:
Wherein, S is the sample set matrix to be identified, and size is J × I, and J, I are positive integer, and J is the sample to be identified
The characteristics of low-frequency dimension of each sample of this concentration, I be the sample set to be identified sample size, STFor S transposed matrix;U
The coefficient matrix of the basic matrix obtained for nth iteration, Q are the coefficient matrix that nth iteration obtains, and U' is (n-1)th iteration
The coefficient matrix of obtained basic matrix, Q' are the coefficient matrix that (n-1)th iteration obtains, and U, Q, U' and Q' size are I × K, K
S characteristic dimension is represented, K is positive integer and K≤J, n are the positive integer more than 1;U'TFor U' transposed matrix, Q'TFor turning for Q'
Put matrix, QTFor Q transposed matrix;UikThe element arranged for U the i-th row kth, QikThe element arranged for Q the i-th row kth, i, k are equal
For positive integer, and i≤I, k≤K;U'ikThe element arranged for U' the i-th row kth, Q'ikThe element arranged for Q' the i-th row kth;Work as J
(V) during value minimum, U is the coefficient matrix of optimal basic matrix, and Q is optimal coefficient matrix, J (V)=Tr (- 2VTF+-VTE-VD),
Wherein F+=(STS)+Q, E-=(STS)-, D=QTQ, V=U, VTFor V transposed matrix.
Wherein, sample set matrix to be identified namely the matrix obtained above after sample set pretreatment to be identified, U is n-th
The coefficient matrix of the basic matrix for the sample set matrix to be identified that iteration obtains, Q are the sample set square to be identified that nth iteration obtains
The coefficient matrix of battle array.
In step S16 to collecting what matrix was handled to the sample identified using the sparse convex Non-negative Matrix Factorization method of new iteration
Process, with the mistake handled in step s 13 using the sparse convex Non-negative Matrix Factorization method of new iteration training sample set matrix
Cheng Xiangtong, repeat no more here.
Different K value corresponds to different decomposition dimensions, and decomposition dimension is bigger, and Character losing is fewer, and Classification and Identification is more accurate.
In step s 16, it is preferable that collect matrix to the sample identified using K=180 and decomposed, so as to ensure higher discrimination.
Step S17:The optimal coefficient matrix collected to the sample identified using the grader trained is classified, and completes people
Face identifies.
The optimal coefficient matrix collected to the sample identified is classified, than directly being divided with sample set matrix to be identified
Class, feature are more concentrated, and operand is smaller.In embodiments of the present invention, the optimal coefficient collected to the sample identified using grader
Matrix, which carries out classification, to be included:The optimal coefficient matrix collected to the sample identified is classified using grader.
Classification and Identification process is:The optimal coefficient matrix Q that size is 200 × K is inputted to SVM classifier and classified,
And define a category vote matrix and be used to vote to each sample generic, category vote matrix size is m × s,
Wherein m is sample total number 200, and s is classification sum 40.The multi-categorizer Parameter File for training to obtain in S14 is called, to 200
Each sample carries out two classification judgements successively in individual sample, judges that it belongs to pth class or q classes, and p is from 1 continuous value
To 39, while q is from the continuous values of p+1 to 40.If the sample is judged as belonging to pth class, then the pth of category vote matrix
Row plus 1, i.e., pth is arranged and throw 1 ticket;, whereas if the sample, which is classified device, is judged as q classes, then the q of category vote matrix is arranged
Add 1, i.e., q is arranged and throw 1 ticket.The column number for counting that row of who gets the most votes in all row of category vote matrix is exactly the sample
This class number.The classification situation of 200 samples is counted, class number's matrix that size is 200 × 1 can be obtained.Classification
The corresponding number of samples of row value of numbering matrix, sample generic corresponding to train value are numbered.Experiment shows, using " a pair of SVM
One " grader correctly Classification and Identification difference expression can belong to same face.
When amount of images if necessary to carry out Classification and Identification exceedes this quantity, training sample set can be expanded simultaneously
The dimension of the initial matrix of initial matrix and sample set to be identified.1600 faces of such as 80 people, everyone 20 faces,
If everyone randomly selects 10 image composition training sample sets, everyone remaining 10 images form sample to be identified
Collection, then respectively have 800 facial images in training sample set and sample set to be identified, if every image size is 130 × 100,
The dimension for expanding training sample set initial matrix X' and sample set initial matrix S' to be identified is J' × I', wherein J'=
13000, I'=800, successively using step S12-S17, you can complete Classification and Identification.If necessary to carry out the image of Classification and Identification
When quantity is less than this quantity, can reduce training sample set initial matrix and sample set initial matrix to be identified dimension i.e.
Can.Such as 20 people, everyone 10 facial images, 200 images altogether, if everyone, which randomly selects 5, is used as training sample set,
It is left 5 conduct sample sets to be identified, then respectively there are 100 facial images in training sample set and sample set to be identified.If every
Image size is 90 × 60, then the dimension for specifying training sample set initial matrix X' and sample set initial matrix S' to be identified is
J'×I':J'=5400, I'=100, perform step S12-S17, you can complete Classification and Identification.
, now can be by the figure to be identified if images to be recognized is not present in training set when carrying out Classification and Identification
As being added in training set, it is using what the sparse convex Non-negative Matrix Factorization method of new iteration recalculated to obtain new optimal basic matrix
Matrix number and optimal coefficient matrix;And be used as and inputted by the use of new optimal coefficient matrix, re -training grader.
Below by contrast test, the effect of face identification method provided in an embodiment of the present invention is illustrated:
The three kinds of control methods used in contrast test are respectively:A, PCA methods;B, SNMF methods;C, basic CNMF side
Method.Method provided in an embodiment of the present invention is:D, the sparse CNMF methods of new iteration.
Each 10 images of everyone in ORL picture libraries totally 400 are taken, every image there are 256 gray levels, and size is 112 × 92.
Wherein everyone countenance and facial detail suffer from different degrees of change, such as laugh at do not laugh at, eyes are opened and close, wear with
Do not wear glasses;Human face posture also has considerable degree of change, and depth rotation and plane are rotated up to up to 20 degree;The chi of face
Degree also has up to 10% change.Preceding 5 images are randomly selected to everyone as training image, composing training sample set, are remained
Under 5 be used as images to be recognized, form sample set to be identified.So respectively have 200 in training sample set and sample set to be identified
.
Fig. 2 a- Fig. 2 d are tetra- kinds of methods of a-d provided in an embodiment of the present invention respectively concentrate to obtain from training sample it is optimal
Basic matrix image, wherein figure d is to concentrate the coefficient matrix for training obtained optimal basic matrix to pass through F=XW formula from training sample
Optimal base matrix image after being reduced.As seen from Figure 2, the optimal base matrix image that d methods obtain can accurately reflect
The position feature information of face eyes, nose so that face characteristic data are more concentrated, be sparse;And a, b, c method obtain most
Excellent basic matrix image is as shown in Fig. 2 a, 2b, 2c, characteristic information excessively Decentralized Fuzzy, not enough concentrates.Accordingly, with respect to a, b, c side
Method, the new sparse CNMF methods of iteration proposed by the present invention, the characteristic that the optimal basic matrix calculated is included is more sparse, feature
Information is more accurately concentrated, and the classification of the optimal coefficient matrix of corresponding face characteristic can be more accurate.Face when K takes different value
Spent by discrimination, algorithm in terms of the time, the comparing result of four kinds of methods is as shown in table 1 below, table 2, four kinds during the continuous values of K
The Contrast on effect curve of method, as shown in Figure 3, Figure 4.
Table 1-K takes the face identification rate of the various methods of different value to contrast
K | 20 | 35 | 55 | 75 | 180 | 220 |
Method a | 21% | 16.5% | 18.5% | 15.5% | 13.5% | 12.5% |
Method b | 85% | 88% | 89% | 87% | 75% | 70% |
Method c | 67% | 54% | 22% | 16% | 3% | 2.5% |
Method d | 98.5% | 99% | 99% | 99% | 100% | 100% |
Table 2-K takes the run time of the various methods of different value to contrast (unit:Second)
K | 20 | 35 | 55 | 75 | 180 | 220 |
Method a | 13.8 | 17.3 | 20.9 | 33.7 | 51.3 | 57.7 |
Method b | 12.5 | 14.5 | 16.2 | 20.3 | 35.1 | 38.2 |
Method c | 20.0 | 21.5 | 24.1 | 25.8 | 37.3 | 40.3 |
Method d | 18.5 | 18.8 | 19.2 | 20.3 | 24.8 | 25.3 |
From table 1 and Fig. 3, using the new sparse CNMF methods of iteration proposed by the present invention, face knowledge can be increased substantially
Not rate, discrimination will be apparently higher than tri- kinds of methods of a-c;This is due to the new rule of iteration ratio invented in technical scheme
Traditional multiplying property rule of iteration is more excellent, and resulting optimal coefficient matrix data more concentrates more sparse, concentration.From the figure 3, it may be seen that this
The sparse CNMF methods of new iteration proposed are invented as K increase, discrimination constantly increase, as K=180, discrimination is up to
When 100%, K continue increase, discrimination holding 100% is constant;And the SNMF methods that method b is provided, when K increases to 55, identification
Rate highest, only 89%, and K continues to increase, discrimination reduces on the contrary;Method a and method c discrimination are relatively low, and as K increases
Big discrimination constantly reduces.And K is bigger, image reconstruction is more accurate, therefore from the figure 3, it may be seen that only proposed by the invention newly changes
It can ensure under higher discrimination for CNMF methods, can also obtain higher K value, so as to guarantee preferably reconstruction figure
Picture.
As the operation time required for the various methods provided in table 2 and Fig. 4, it is known that when K constantly increases, the present invention carries
The recognition speed of the new iteration CNMF methods supplied is substantially compared with other three kinds of methods faster.Because in technical scheme
The new rule of iteration of invention simultaneously can obtain the coefficient matrix of optimal, rarefaction basic matrix and optimal with reference to threshold value Sparse methods
Coefficient matrix, reduce amount of calculation.
From Fig. 3 and Fig. 4, existing tri- kinds of methods of a, b, c, with K increase, its discrimination is constantly reduced, consumed
It is time-consuming to be but continuously increased;With K increase, the increase of technical scheme run time is slowly and discrimination constantly increases
It is high;Method provided by the invention K values when discrimination is up to 100% are 180, and during the discrimination highest of other method
K values are smaller, and K is bigger, and it is more accurate that later image is rebuild.Therefore, algorithm time, the bulking property of discrimination are considered
The accuracy rate that energy and later image are rebuild, institute extracting method d of the present invention will be substantially better than existing related tri- kinds of methods of a, b, c.
Based on the new sparse CNMF methods of iteration provided in an embodiment of the present invention, discrimination, operation time, later stage are considered
The factors such as image reconstruction, K=180 is selected, the face recognition software based on MATLAB has been write, with 100% from 200 faces
Probability choose the facial image to be identified of different people at random, in the situation whether different expressions, eyes open and close, wear glasses
Under, the equal Classification and Identification of software is correct.Software can be chosen a people under with 100% probability and carry out correct identification and output identification
The frontal faces of this person's identity, while the face generic and sample set to be identified can be correctly provided in software interface text box
Discrimination.
In summary, the face recognition technology provided by the invention based on the sparse CNMF methods of new iteration, there is higher reason
Value;Technical scheme can obtain high face identification rate simultaneously, and the calculating time is short, and can ensure the later stage
The higher accuracy rate of human face rebuilding, and this technology realizes that engineering application value is big by software.
Fig. 5 is the embodiments of the invention provide a kind of structural representation of face identification device, referring to Fig. 5, the device bag
Include:
Pretreatment unit 201, is pre-processed for collecting to the sample identified, obtains sample set matrix to be identified;
Resolving cell 202, carried out for collecting matrix to the sample identified using the sparse convex Non-negative Matrix Factorization method of new iteration
Processing, obtains the coefficient matrix and optimal coefficient matrix of the optimal basic matrix of sample set to be identified, sample set to be identified it is optimal
The coefficient matrix and optimal coefficient matrix of basic matrix are produced using following iterative formula iteration:
Wherein, S is the sample set matrix to be identified, and size is J × I, and J, I are positive integer, and J is the sample to be identified
The characteristics of low-frequency dimension of each sample of this concentration, I be the sample set to be identified sample size, STFor S transposed matrix;U
The coefficient matrix of the basic matrix obtained for nth iteration, Q are the coefficient matrix that nth iteration obtains, and U' is (n-1)th iteration
The coefficient matrix of obtained basic matrix, Q' are the coefficient matrix that (n-1)th iteration obtains, and U, Q, U' and Q' size are I × K, K
S characteristic dimension is represented, K is positive integer and K≤J, n are the positive integer more than 1;U'TFor U' transposed matrix, Q'TFor turning for Q'
Put matrix, QTFor Q transposed matrix;UikThe element arranged for U the i-th row kth, QikThe element arranged for Q the i-th row kth, i, k are equal
For positive integer, and i≤I, k≤K;U'ikThe element arranged for U' the i-th row kth, Q'ikThe element arranged for Q' the i-th row kth;Work as J
(V) during value minimum, U is the coefficient matrix of optimal basic matrix, and Q is optimal coefficient matrix, J (V)=Tr (- 2VTF+-VTE-VD),
Wherein F+=(STS)+Q, E-=(STS)-, D=QTQ, V=U, VTFor V transposed matrix;
Taxon 203, the optimal coefficient matrix for being collected to the sample identified using the grader trained are divided
Class, complete recognition of face.
In embodiments of the present invention, resolving cell 202, for determining K value in setting range;For the K determined
Value, matrix is collected using the sparse convex Non-negative Matrix Factorization method of new iteration to the sample identified and decomposed, is obtained corresponding to K described
The coefficient matrix and optimal coefficient matrix of the optimal basic matrix of sample set to be identified.
In embodiments of the present invention, setting range preferably can be 180 >=K >=20.
In embodiments of the present invention, resolving cell 202, for determining the coefficient matrix of initial basic matrix and initial according to K
Coefficient matrix;Calculating is iterated according to the coefficient matrix of initial basic matrix and initial coefficients matrix and iterative formula;Will be every
The coefficient matrix and coefficient matrix for the basic matrix that single-step iteration calculates substitute into object function:
J (V)=Tr (- 2VTF+-VTE-VD), wherein F+=(STS)+Q, E-=(STS)-, D=QTQ, V=U;
When the value of object function reaches stable state, end iterative calculation, and the group moment that will be iterated to calculate out for the last time
Coefficient matrix and optimal coefficient matrix of the coefficient matrix and coefficient matrix of battle array as the optimal basic matrix of sample set to be identified, its
In, stable state refers to that the value of object function keeps constant or amplitude of fluctuation to be less than predetermined amplitude;Or when iterations reaches
During to iterations threshold value, the coefficient matrix and coefficient matrix for the basic matrix that selection last time iterates to calculate out are as to be identified
The coefficient matrix and optimal coefficient matrix of the optimal basic matrix of sample set.
In embodiments of the present invention, resolving cell 202, it is additionally operable to determine rarefaction threshold value;After the iterative calculation of each step,
Each numerical value and the size of rarefaction threshold value in the coefficient matrix for the basic matrix that judgement iterates to calculate out;By what is iterated to calculate out
Numerical value in the coefficient matrix of basic matrix more than rarefaction threshold value is arranged to 1, by the coefficient matrix of the basic matrix iterated to calculate out
In be less than or equal to rarefaction threshold value numerical value be arranged to 0.
Further, the device also includes training unit 204;
Pretreatment unit 201, it is additionally operable to pre-process training sample set, obtains training sample set matrix;
Resolving cell 202, it is additionally operable to carry out training sample set matrix using the sparse convex Non-negative Matrix Factorization method of new iteration
Processing, obtain the coefficient matrix and optimal coefficient matrix of the optimal basic matrix of training sample set, the coefficient matrix of optimal basic matrix
Produced with optimal coefficient matrix using identical iterative formula iteration when being handled with sample set to be identified;
Training unit 204, for training grader using the optimal coefficient matrix of the training sample set.
It should be noted that:The face identification device that above-described embodiment provides is in recognition of face, only with above-mentioned each function
The division progress of module, can be as needed and by above-mentioned function distribution by different function moulds for example, in practical application
Block is completed, i.e., the internal structure of equipment is divided into different functional modules, to complete all or part of work(described above
Energy.In addition, the face identification device that above-described embodiment provides belongs to same design with face identification method embodiment, it is specific real
Existing process refers to embodiment of the method, repeats no more here.
The embodiments of the present invention are for illustration only, do not represent the quality of embodiment.
One of ordinary skill in the art will appreciate that hardware can be passed through by realizing all or part of step of above-described embodiment
To complete, by program the hardware of correlation can also be instructed to complete, described program can be stored in a kind of computer-readable
In storage medium, storage medium mentioned above can be read-only storage, disk or CD etc..
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., it should be included in the scope of the protection.
Claims (10)
1. a kind of face identification method, it is characterised in that methods described includes:
Collection is pre-processed to the sample identified, obtains sample set matrix to be identified;
The sample set matrix to be identified is handled using the sparse convex Non-negative Matrix Factorization method of new iteration, obtains and described waits to know
The coefficient matrix and optimal coefficient matrix of the optimal basic matrix of other sample set, the optimal basic matrix of the sample set to be identified are
Matrix number and optimal coefficient matrix are produced using following iterative formula iteration:
<mrow>
<msub>
<mi>U</mi>
<mrow>
<mi>i</mi>
<mi>k</mi>
</mrow>
</msub>
<mo>=</mo>
<msqrt>
<mrow>
<msubsup>
<mi>U</mi>
<mrow>
<mi>i</mi>
<mi>k</mi>
</mrow>
<mrow>
<mo>&prime;</mo>
<mn>2</mn>
</mrow>
</msubsup>
<mo>+</mo>
<msubsup>
<mi>U</mi>
<mrow>
<mi>i</mi>
<mi>k</mi>
</mrow>
<mrow>
<mo>&prime;</mo>
<mn>2</mn>
</mrow>
</msubsup>
<mfrac>
<mrow>
<msub>
<mrow>
<mo>&lsqb;</mo>
<msup>
<mrow>
<mo>(</mo>
<msup>
<mi>S</mi>
<mi>T</mi>
</msup>
<mi>S</mi>
<mo>)</mo>
</mrow>
<mo>+</mo>
</msup>
<msup>
<mi>Q</mi>
<mo>&prime;</mo>
</msup>
<mo>&rsqb;</mo>
</mrow>
<mrow>
<mi>i</mi>
<mi>k</mi>
</mrow>
</msub>
<mo>+</mo>
<msub>
<mrow>
<mo>&lsqb;</mo>
<msup>
<mrow>
<mo>(</mo>
<msup>
<mi>S</mi>
<mi>T</mi>
</msup>
<mi>S</mi>
<mo>)</mo>
</mrow>
<mo>-</mo>
</msup>
<msup>
<mi>U</mi>
<mo>&prime;</mo>
</msup>
<msup>
<mi>Q</mi>
<mrow>
<mo>&prime;</mo>
<mi>T</mi>
</mrow>
</msup>
<msup>
<mi>Q</mi>
<mo>&prime;</mo>
</msup>
<mo>&rsqb;</mo>
</mrow>
<mrow>
<mi>i</mi>
<mi>k</mi>
</mrow>
</msub>
</mrow>
<mrow>
<msub>
<mrow>
<mo>&lsqb;</mo>
<msup>
<mrow>
<mo>(</mo>
<msup>
<mi>S</mi>
<mi>T</mi>
</msup>
<mi>S</mi>
<mo>)</mo>
</mrow>
<mo>-</mo>
</msup>
<msup>
<mi>Q</mi>
<mo>&prime;</mo>
</msup>
<mo>&rsqb;</mo>
</mrow>
<mrow>
<mi>i</mi>
<mi>k</mi>
</mrow>
</msub>
<mo>+</mo>
<msub>
<mrow>
<mo>&lsqb;</mo>
<msup>
<mrow>
<mo>(</mo>
<msup>
<mi>S</mi>
<mi>T</mi>
</msup>
<mi>S</mi>
<mo>)</mo>
</mrow>
<mo>+</mo>
</msup>
<msup>
<mi>U</mi>
<mo>&prime;</mo>
</msup>
<msup>
<mi>Q</mi>
<mrow>
<mo>&prime;</mo>
<mi>T</mi>
</mrow>
</msup>
<msup>
<mi>Q</mi>
<mo>&prime;</mo>
</msup>
<mo>&rsqb;</mo>
</mrow>
<mrow>
<mi>i</mi>
<mi>k</mi>
</mrow>
</msub>
</mrow>
</mfrac>
</mrow>
</msqrt>
<mo>,</mo>
<msub>
<mi>Q</mi>
<mrow>
<mi>i</mi>
<mi>k</mi>
</mrow>
</msub>
<mo>=</mo>
<msqrt>
<mrow>
<msubsup>
<mi>Q</mi>
<mrow>
<mi>i</mi>
<mi>k</mi>
</mrow>
<mrow>
<mo>&prime;</mo>
<mn>2</mn>
</mrow>
</msubsup>
<mo>+</mo>
<msubsup>
<mi>Q</mi>
<mrow>
<mi>i</mi>
<mi>k</mi>
</mrow>
<mrow>
<mo>&prime;</mo>
<mn>2</mn>
</mrow>
</msubsup>
<mfrac>
<mrow>
<msub>
<mrow>
<mo>&lsqb;</mo>
<msup>
<mrow>
<mo>(</mo>
<msup>
<mi>S</mi>
<mi>T</mi>
</msup>
<mi>S</mi>
<mo>)</mo>
</mrow>
<mo>+</mo>
</msup>
<msup>
<mi>U</mi>
<mo>&prime;</mo>
</msup>
<mo>&rsqb;</mo>
</mrow>
<mrow>
<mi>i</mi>
<mi>k</mi>
</mrow>
</msub>
<mo>+</mo>
<msub>
<mrow>
<mo>&lsqb;</mo>
<msup>
<mi>Q</mi>
<mo>&prime;</mo>
</msup>
<msup>
<mi>U</mi>
<mrow>
<mo>&prime;</mo>
<mi>T</mi>
</mrow>
</msup>
<msup>
<mrow>
<mo>(</mo>
<msup>
<mi>S</mi>
<mi>T</mi>
</msup>
<mi>S</mi>
<mo>)</mo>
</mrow>
<mo>-</mo>
</msup>
<msup>
<mi>U</mi>
<mo>&prime;</mo>
</msup>
<mo>&rsqb;</mo>
</mrow>
<mrow>
<mi>i</mi>
<mi>k</mi>
</mrow>
</msub>
</mrow>
<mrow>
<msub>
<mrow>
<mo>&lsqb;</mo>
<msup>
<mrow>
<mo>(</mo>
<msup>
<mi>S</mi>
<mi>T</mi>
</msup>
<mi>S</mi>
<mo>)</mo>
</mrow>
<mo>-</mo>
</msup>
<msup>
<mi>U</mi>
<mo>&prime;</mo>
</msup>
<mo>&rsqb;</mo>
</mrow>
<mrow>
<mi>i</mi>
<mi>k</mi>
</mrow>
</msub>
<mo>+</mo>
<msub>
<mrow>
<mo>&lsqb;</mo>
<msup>
<mi>Q</mi>
<mo>&prime;</mo>
</msup>
<msup>
<mi>U</mi>
<mrow>
<mo>&prime;</mo>
<mi>T</mi>
</mrow>
</msup>
<msup>
<mrow>
<mo>(</mo>
<msup>
<mi>S</mi>
<mi>T</mi>
</msup>
<mi>S</mi>
<mo>)</mo>
</mrow>
<mo>+</mo>
</msup>
<msup>
<mi>U</mi>
<mo>&prime;</mo>
</msup>
<mo>&rsqb;</mo>
</mrow>
<mrow>
<mi>i</mi>
<mi>k</mi>
</mrow>
</msub>
</mrow>
</mfrac>
</mrow>
</msqrt>
<mo>;</mo>
</mrow>
Wherein, S is the sample set matrix to be identified, and size is J × I, and J, I are positive integer, and J is the sample set to be identified
In each sample characteristics of low-frequency dimension, I is the sample size of the sample set to be identified, STFor S transposed matrix;U is n-th
The coefficient matrix for the basic matrix that secondary iteration obtains, Q are the coefficient matrix that nth iteration obtains, and U' is what (n-1)th iteration obtained
The coefficient matrix of basic matrix, Q' are the coefficient matrix that (n-1)th iteration obtains, and U, Q, U' and Q' size are I × K, and K represents S
Characteristic dimension, K is that positive integer and K≤J, n are positive integer more than 1;U'TFor U' transposed matrix, Q'TFor Q' transposition square
Battle array, QTFor Q transposed matrix;UikThe element arranged for U the i-th row kth, QikThe element arranged for Q the i-th row kth, i, k are just
Integer, and i≤I, k≤K;U'ikThe element arranged for U' the i-th row kth, Q'ikThe element arranged for Q' the i-th row kth;As J (V)
Value minimum when, U is the coefficient matrix of optimal basic matrix, and Q is optimal coefficient matrix, J (V)=Tr (- 2VTF+-VTE-VD), its
Middle F+=(STS)+Q, E-=(STS)-, D=QTQ, V=U, VTFor V transposed matrix;
The optimal coefficient matrix of the sample set to be identified is classified using the grader trained, completes recognition of face.
2. according to the method for claim 1, it is characterised in that described to use the sparse convex Non-negative Matrix Factorization method pair of new iteration
The sample set matrix to be identified is handled, and obtains the coefficient matrix of the optimal basic matrix of the sample set to be identified and optimal
Coefficient matrix, including:
K value is determined in setting range;
For the K determined value, the sample set matrix to be identified is entered using the sparse convex Non-negative Matrix Factorization method of new iteration
Row decomposes, and obtains the coefficient matrix and optimal coefficient matrix of the optimal basic matrix of the sample set to be identified corresponding to K.
3. according to the method for claim 2, it is characterised in that it is described for the K determined value, it is sparse using new iteration
Convex Non-negative Matrix Factorization method is decomposed to the sample set matrix to be identified, obtains the sample set to be identified corresponding to K
The coefficient matrix and optimal coefficient matrix of optimal basic matrix, including:
The coefficient matrix and initial coefficients matrix of initial basic matrix are determined according to K;
Calculating is iterated according to the coefficient matrix of the initial basic matrix and initial coefficients matrix and the iterative formula;
The coefficient matrix and coefficient matrix for the basic matrix that each step is iterated to calculate out substitute into object function:
J (V)=Tr (- 2VTF+-VTE-VD), wherein F+=(STS)+Q, E-=(STS)-, D=QTQ, V=U;
When the value of the object function reaches stable state, end iterative calculation, and the group moment that will be iterated to calculate out for the last time
Coefficient matrix and optimal coefficient square of the coefficient matrix and coefficient matrix of battle array as the optimal basic matrix of the sample set to be identified
Battle array, wherein, the stable state refers to that the value of the object function keeps constant or amplitude of fluctuation to be less than predetermined amplitude;Or
Person, when iterations reaches iterations threshold value, choose the coefficient matrix for the basic matrix that last time iterates to calculate out and be
Coefficient matrix and optimal coefficient matrix of the matrix number as the optimal basic matrix of the sample set to be identified.
4. according to the method for claim 3, it is characterised in that it is described for the K determined value, it is sparse using new iteration
Convex Non-negative Matrix Factorization method is decomposed to the sample set matrix to be identified, obtains the sample set to be identified corresponding to K
The coefficient matrix and optimal coefficient matrix of optimal basic matrix, in addition to:
Determine rarefaction threshold value;
After the iterative calculation of each step, each numerical value and rarefaction threshold in the coefficient matrix of basic matrix that iterates to calculate out are judged
The size of value;The numerical value for being more than rarefaction threshold value in the coefficient matrix of the basic matrix iterated to calculate out is arranged to 1, by iteration meter
Numerical value in the coefficient matrix of the basic matrix calculated less than or equal to rarefaction threshold value is arranged to 0.
5. according to the method described in claim any one of 1-4, it is characterised in that methods described also includes:
Training sample set is pre-processed, obtains training sample set matrix;
The training sample set matrix is handled using new iteration sparse convex Non-negative Matrix Factorization method, obtains the training sample
The coefficient matrix and optimal coefficient matrix of the optimal basic matrix of this collection, the coefficient matrix and optimal coefficient square of the optimal basic matrix
Battle array is produced using identical iterative formula iteration when being handled with the sample set to be identified;
Grader is trained using the optimal coefficient matrix of the training sample set.
6. a kind of face identification device, it is characterised in that described device includes:
Pretreatment unit, pre-processed for collecting to the sample identified, obtain sample set matrix to be identified;
Resolving cell, at using the sparse convex Non-negative Matrix Factorization method of new iteration to the sample set matrix to be identified
Reason, obtain the coefficient matrix and optimal coefficient matrix of the optimal basic matrix of the sample set to be identified, the sample set to be identified
Optimal basic matrix coefficient matrix and optimal coefficient matrix using following iterative formula iteration produce:
<mrow>
<msub>
<mi>U</mi>
<mrow>
<mi>i</mi>
<mi>k</mi>
</mrow>
</msub>
<mo>=</mo>
<msqrt>
<mrow>
<msubsup>
<mi>U</mi>
<mrow>
<mi>i</mi>
<mi>k</mi>
</mrow>
<mrow>
<mo>&prime;</mo>
<mn>2</mn>
</mrow>
</msubsup>
<mo>+</mo>
<msubsup>
<mi>U</mi>
<mrow>
<mi>i</mi>
<mi>k</mi>
</mrow>
<mrow>
<mo>&prime;</mo>
<mn>2</mn>
</mrow>
</msubsup>
<mfrac>
<mrow>
<msub>
<mrow>
<mo>&lsqb;</mo>
<msup>
<mrow>
<mo>(</mo>
<msup>
<mi>S</mi>
<mi>T</mi>
</msup>
<mi>S</mi>
<mo>)</mo>
</mrow>
<mo>+</mo>
</msup>
<msup>
<mi>Q</mi>
<mo>&prime;</mo>
</msup>
<mo>&rsqb;</mo>
</mrow>
<mrow>
<mi>i</mi>
<mi>k</mi>
</mrow>
</msub>
<mo>+</mo>
<msub>
<mrow>
<mo>&lsqb;</mo>
<msup>
<mrow>
<mo>(</mo>
<msup>
<mi>S</mi>
<mi>T</mi>
</msup>
<mi>S</mi>
<mo>)</mo>
</mrow>
<mo>-</mo>
</msup>
<msup>
<mi>U</mi>
<mo>&prime;</mo>
</msup>
<msup>
<mi>Q</mi>
<mrow>
<mo>&prime;</mo>
<mi>T</mi>
</mrow>
</msup>
<msup>
<mi>Q</mi>
<mo>&prime;</mo>
</msup>
<mo>&rsqb;</mo>
</mrow>
<mrow>
<mi>i</mi>
<mi>k</mi>
</mrow>
</msub>
</mrow>
<mrow>
<msub>
<mrow>
<mo>&lsqb;</mo>
<msup>
<mrow>
<mo>(</mo>
<msup>
<mi>S</mi>
<mi>T</mi>
</msup>
<mi>S</mi>
<mo>)</mo>
</mrow>
<mo>-</mo>
</msup>
<msup>
<mi>Q</mi>
<mo>&prime;</mo>
</msup>
<mo>&rsqb;</mo>
</mrow>
<mrow>
<mi>i</mi>
<mi>k</mi>
</mrow>
</msub>
<mo>+</mo>
<msub>
<mrow>
<mo>&lsqb;</mo>
<msup>
<mrow>
<mo>(</mo>
<msup>
<mi>S</mi>
<mi>T</mi>
</msup>
<mi>S</mi>
<mo>)</mo>
</mrow>
<mo>+</mo>
</msup>
<msup>
<mi>U</mi>
<mo>&prime;</mo>
</msup>
<msup>
<mi>Q</mi>
<mrow>
<mo>&prime;</mo>
<mi>T</mi>
</mrow>
</msup>
<msup>
<mi>Q</mi>
<mo>&prime;</mo>
</msup>
<mo>&rsqb;</mo>
</mrow>
<mrow>
<mi>i</mi>
<mi>k</mi>
</mrow>
</msub>
</mrow>
</mfrac>
</mrow>
</msqrt>
<mo>,</mo>
<msub>
<mi>Q</mi>
<mrow>
<mi>i</mi>
<mi>k</mi>
</mrow>
</msub>
<mo>=</mo>
<msqrt>
<mrow>
<msubsup>
<mi>Q</mi>
<mrow>
<mi>i</mi>
<mi>k</mi>
</mrow>
<mrow>
<mo>&prime;</mo>
<mn>2</mn>
</mrow>
</msubsup>
<mo>+</mo>
<msubsup>
<mi>Q</mi>
<mrow>
<mi>i</mi>
<mi>k</mi>
</mrow>
<mrow>
<mo>&prime;</mo>
<mn>2</mn>
</mrow>
</msubsup>
<mfrac>
<mrow>
<msub>
<mrow>
<mo>&lsqb;</mo>
<msup>
<mrow>
<mo>(</mo>
<msup>
<mi>S</mi>
<mi>T</mi>
</msup>
<mi>S</mi>
<mo>)</mo>
</mrow>
<mo>+</mo>
</msup>
<msup>
<mi>U</mi>
<mo>&prime;</mo>
</msup>
<mo>&rsqb;</mo>
</mrow>
<mrow>
<mi>i</mi>
<mi>k</mi>
</mrow>
</msub>
<mo>+</mo>
<msub>
<mrow>
<mo>&lsqb;</mo>
<msup>
<mi>Q</mi>
<mo>&prime;</mo>
</msup>
<msup>
<mi>U</mi>
<mrow>
<mo>&prime;</mo>
<mi>T</mi>
</mrow>
</msup>
<msup>
<mrow>
<mo>(</mo>
<msup>
<mi>S</mi>
<mi>T</mi>
</msup>
<mi>S</mi>
<mo>)</mo>
</mrow>
<mo>-</mo>
</msup>
<msup>
<mi>U</mi>
<mo>&prime;</mo>
</msup>
<mo>&rsqb;</mo>
</mrow>
<mrow>
<mi>i</mi>
<mi>k</mi>
</mrow>
</msub>
</mrow>
<mrow>
<msub>
<mrow>
<mo>&lsqb;</mo>
<msup>
<mrow>
<mo>(</mo>
<msup>
<mi>S</mi>
<mi>T</mi>
</msup>
<mi>S</mi>
<mo>)</mo>
</mrow>
<mo>-</mo>
</msup>
<msup>
<mi>U</mi>
<mo>&prime;</mo>
</msup>
<mo>&rsqb;</mo>
</mrow>
<mrow>
<mi>i</mi>
<mi>k</mi>
</mrow>
</msub>
<mo>+</mo>
<msub>
<mrow>
<mo>&lsqb;</mo>
<msup>
<mi>Q</mi>
<mo>&prime;</mo>
</msup>
<msup>
<mi>U</mi>
<mrow>
<mo>&prime;</mo>
<mi>T</mi>
</mrow>
</msup>
<msup>
<mrow>
<mo>(</mo>
<msup>
<mi>S</mi>
<mi>T</mi>
</msup>
<mi>S</mi>
<mo>)</mo>
</mrow>
<mo>+</mo>
</msup>
<msup>
<mi>U</mi>
<mo>&prime;</mo>
</msup>
<mo>&rsqb;</mo>
</mrow>
<mrow>
<mi>i</mi>
<mi>k</mi>
</mrow>
</msub>
</mrow>
</mfrac>
</mrow>
</msqrt>
<mo>;</mo>
</mrow>
Wherein, S is the sample set matrix to be identified, and size is J × I, and J, I are positive integer, and J is the sample set to be identified
In each sample characteristics of low-frequency dimension, I is the sample size of the sample set to be identified, STFor S transposed matrix;U is n-th
The coefficient matrix for the basic matrix that secondary iteration obtains, Q are the coefficient matrix that nth iteration obtains, and U' is what (n-1)th iteration obtained
The coefficient matrix of basic matrix, Q' are the coefficient matrix that (n-1)th iteration obtains, and U, Q, U' and Q' size are I × K, and K represents S
Characteristic dimension, K is that positive integer and K≤J, n are positive integer more than 1;U'TFor U' transposed matrix, Q'TFor Q' transposition square
Battle array, QTFor Q transposed matrix;UikThe element arranged for U the i-th row kth, QikThe element arranged for Q the i-th row kth, i, k are just
Integer, and i≤I, k≤K;U'ikThe element arranged for U' the i-th row kth, Q'ikThe element arranged for Q' the i-th row kth;As J (V)
Value minimum when, U is the coefficient matrix of optimal basic matrix, and Q is optimal coefficient matrix, J (V)=Tr (- 2VTF+-VTE-VD), its
Middle F+=(STS)+Q, E-=(STS)-, D=QTQ, V=U, VTFor V transposed matrix;
Taxon, for being classified using the grader trained to the optimal coefficient matrix of the sample set to be identified,
Complete recognition of face.
7. device according to claim 6, it is characterised in that the resolving cell, for determining K's in setting range
Value;For the K determined value, the sample set matrix to be identified is carried out using the sparse convex Non-negative Matrix Factorization method of new iteration
Decompose, obtain the coefficient matrix and optimal coefficient matrix of the optimal basic matrix of the sample set to be identified corresponding to K.
8. device according to claim 7, it is characterised in that the resolving cell, for determining initial basic matrix according to K
Coefficient matrix and initial coefficients matrix;According to the coefficient matrix of the initial basic matrix and initial coefficients matrix and it is described repeatedly
Calculating is iterated for formula;The coefficient matrix and coefficient matrix for the basic matrix that each step is iterated to calculate out substitute into target letter
Number:
J (V)=Tr (- 2VTF+-VTE-VD), wherein F+=(STS)+Q, E-=(STS)-, D=QTQ, V=U;
When the value of the object function reaches stable state, end iterative calculation, and the group moment that will be iterated to calculate out for the last time
Coefficient matrix and optimal coefficient square of the coefficient matrix and coefficient matrix of battle array as the optimal basic matrix of the sample set to be identified
Battle array, wherein, the stable state refers to that the value of the object function keeps constant or amplitude of fluctuation to be less than predetermined amplitude;Or
Person, when iterations reaches iterations threshold value, choose the coefficient matrix for the basic matrix that last time iterates to calculate out and be
Coefficient matrix and optimal coefficient matrix of the matrix number as the optimal basic matrix of the sample set to be identified.
9. device according to claim 8, it is characterised in that the resolving cell, be additionally operable to determine rarefaction threshold value;
After the iterative calculation of each step, the big of each numerical value in the coefficient matrix of basic matrix that iterates to calculate out and rarefaction threshold value is judged
It is small;The numerical value for being more than rarefaction threshold value in the coefficient matrix of the basic matrix iterated to calculate out is arranged to 1, by what is iterated to calculate out
Numerical value in the coefficient matrix of basic matrix less than or equal to rarefaction threshold value is arranged to 0.
10. according to the device described in claim any one of 6-9, it is characterised in that described device also includes training unit;
The pretreatment unit, it is additionally operable to pre-process training sample set, obtains training sample set matrix;
The resolving cell, it is additionally operable to carry out the training sample set matrix using the sparse convex Non-negative Matrix Factorization method of new iteration
Processing, the coefficient matrix and optimal coefficient matrix of the optimal basic matrix of the training sample set are obtained, the optimal basic matrix
Coefficient matrix and optimal coefficient matrix use identical iterative formula iteration when being handled with the sample set to be identified to produce;
The training unit, for training grader using the optimal coefficient matrix of the training sample set.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201710631337.6A CN107341485B (en) | 2017-07-28 | 2017-07-28 | Face recognition method and device |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201710631337.6A CN107341485B (en) | 2017-07-28 | 2017-07-28 | Face recognition method and device |
Publications (2)
Publication Number | Publication Date |
---|---|
CN107341485A true CN107341485A (en) | 2017-11-10 |
CN107341485B CN107341485B (en) | 2019-12-31 |
Family
ID=60216586
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201710631337.6A Active CN107341485B (en) | 2017-07-28 | 2017-07-28 | Face recognition method and device |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN107341485B (en) |
Cited By (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN109508697A (en) * | 2018-12-14 | 2019-03-22 | 深圳大学 | Face identification method, system and the storage medium of half Non-negative Matrix Factorization based on E auxiliary function |
Citations (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN101719270A (en) * | 2009-12-25 | 2010-06-02 | 武汉大学 | Non-negative matrix factorization-based face super-resolution processing method |
CN102831598A (en) * | 2012-07-04 | 2012-12-19 | 西安电子科技大学 | Remote sensing image change detecting method with combination of multi-resolution NMF (non-negative matrix factorization) and Treelet |
US20170091530A1 (en) * | 2013-03-12 | 2017-03-30 | Yahoo! Inc. | Media content enrichment using an adapted object detector |
-
2017
- 2017-07-28 CN CN201710631337.6A patent/CN107341485B/en active Active
Patent Citations (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN101719270A (en) * | 2009-12-25 | 2010-06-02 | 武汉大学 | Non-negative matrix factorization-based face super-resolution processing method |
CN102831598A (en) * | 2012-07-04 | 2012-12-19 | 西安电子科技大学 | Remote sensing image change detecting method with combination of multi-resolution NMF (non-negative matrix factorization) and Treelet |
US20170091530A1 (en) * | 2013-03-12 | 2017-03-30 | Yahoo! Inc. | Media content enrichment using an adapted object detector |
Cited By (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN109508697A (en) * | 2018-12-14 | 2019-03-22 | 深圳大学 | Face identification method, system and the storage medium of half Non-negative Matrix Factorization based on E auxiliary function |
CN109508697B (en) * | 2018-12-14 | 2022-03-08 | 深圳大学 | Face recognition method, system and storage medium based on semi-nonnegative matrix factorization of E auxiliary function |
Also Published As
Publication number | Publication date |
---|---|
CN107341485B (en) | 2019-12-31 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
Yang et al. | A feature-reduction multi-view k-means clustering algorithm | |
CN108564129B (en) | Trajectory data classification method based on generation countermeasure network | |
CN101968853B (en) | Improved immune algorithm based expression recognition method for optimizing support vector machine parameters | |
CN106599941A (en) | Method for identifying handwritten numbers based on convolutional neural network and support vector machine | |
CN111862093A (en) | Corrosion grade information processing method and system based on image recognition | |
Shrivastava et al. | Learning discriminative dictionaries with partially labeled data | |
CN112464865A (en) | Facial expression recognition method based on pixel and geometric mixed features | |
CN106909946A (en) | A kind of picking system of multi-modal fusion | |
CN104268593A (en) | Multiple-sparse-representation face recognition method for solving small sample size problem | |
Zhou et al. | A method of facial expression recognition based on Gabor and NMF | |
CN107341510A (en) | Image clustering method based on sparse orthogonal digraph Non-negative Matrix Factorization | |
CN105956570B (en) | Smiling face's recognition methods based on lip feature and deep learning | |
CN104751463B (en) | A kind of threedimensional model optimal viewing angle choosing method based on sketch outline feature | |
CN105868796A (en) | Design method for linear discrimination of sparse representation classifier based on nuclear space | |
CN109726725A (en) | The oil painting writer identification method of heterogeneite Multiple Kernel Learning between a kind of class based on large-spacing | |
CN111401156A (en) | Image identification method based on Gabor convolution neural network | |
CN108805054A (en) | A kind of facial image sorting technique, system, equipment and computer storage media | |
CN113628297A (en) | COVID-19 deep learning diagnosis system based on attention mechanism and transfer learning | |
Ma et al. | Joint-label learning by dual augmentation for time series classification | |
CN110070068B (en) | Human body action recognition method | |
CN114913379A (en) | Remote sensing image small sample scene classification method based on multi-task dynamic contrast learning | |
CN114155572A (en) | Facial expression recognition method and system | |
CN114492634A (en) | Fine-grained equipment image classification and identification method and system | |
CN113408418A (en) | Calligraphy font and character content synchronous identification method and system | |
CN107563287A (en) | Face identification method 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 | ||
GR01 | Patent grant | ||
GR01 | Patent grant |