CN103955676A - Human face identification method and system - Google Patents

Abstract

The invention discloses a human face identification method and system. The method comprises the steps that a PCA method is used for carrying out initial dimensionality reduction on a training sample set, the category label information of a training sample is used for establishing a matrix with classified information, then the optimum secondary projection matrix is determined, secondary dimensionality reduction is carried out on the initial dimensionality reduction training sample set, then secondary dimensionality reduction is also carried out on a tested sample, and after secondary dimensionality reduction, classifying is carried out in a low-dimensional space. By secondary dimensionality reduction processing, human face identification accuracy and efficiency are improved.

Description

A kind of face identification method and system

Technical field

The application relates to mode identification technology, more particularly, relates to a kind of face identification method and system.

Background technology

Face is the pattern of a kind of complexity, changeable, higher-dimension, in recognition of face, face data need to be mapped to low n-dimensional subspace n from higher dimensional space.Because recognition of face has extensive use at the aspect such as authentication, security system, therefore recognition of face has become an important field of research in computer vision and pattern-recognition.

Traditional face identification method adopts pivot analysis PCA (Principal Component Analysis) dimension reduction method conventionally, but the method is only applicable to the data of linear expression.Therefore, someone has proposed neighbour and has kept embedding algorithm again, and the method is applicable to popular data.But neighbour keeps embedding calculation method and do not obtain global structure information and feature in the time carrying out linear reconstruction, and neighborhood is not done to classification judgement, the classification information of sample itself has been ignored, cause recognition of face rate not high.

Summary of the invention

In view of this, the application provides a kind of face identification method and system, for solving the not high problem of discrimination of existing face identification method.

To achieve these goals, the existing scheme proposing is as follows:

A kind of face identification method, comprising:

Utilize principle component analysis PCA to carry out initial dimensionality reduction to training sample set, obtain initial dimensionality reduction training sample set, and preserve a projection matrix in initial reduction process;

Utilize the class label information of training sample, construct a matrix with classified information;

Determine optimum reprojection's matrix, described initial dimensionality reduction training sample set is carried out to secondary dimensionality reduction, obtain secondary dimensionality reduction training sample set;

Utilize a described projection matrix and described reprojection matrix, test sample book is carried out to dimensionality reduction twice, obtain secondary dimensionality reduction test sample book;

Utilize nearest neighbor classifier, described secondary dimensionality reduction test sample book is classified.

Preferably, the described principle component analysis PCA that utilizes carries out initial dimensionality reduction to training sample set, obtains initial dimensionality reduction training sample set, and preserves a projection matrix in initial reduction process, is specially:

Definition training sample set is x _i∈ R ^d, y _i=1,2 ..., c} is training sample x _iclass label information, wherein D is the dimension of training sample, l is the number of training sample data, c is the classification number of training sample data;

Utilize pivot analysis method of descent PCA to carry out initial dimensionality reduction to training sample, obtain initial dimensionality reduction training sample set $\{{\stackrel{\‾}{x}}_i,y_i{\}}_{i=1}^l,{\stackrel{\‾}{x}}_i\∈R^d,$ y _i＝{1,2,…,c}；

Definition training sample matrix is X=[x ₁, x ₂..., x _l] ∈ R ^{d × l}, initial dimensionality reduction training sample matrix X ₁=Α ₁x, wherein A ₁∈ R ^{d × D}it is the projection matrix that PCA method of descent obtains.

Preferably, the described class label information of utilizing training sample, constructs a matrix with classified information, is specially:

Utilize the class label information y of training sample _i, construct the matrix with classified information a: H=[h ₁, h ₂..., h _l] ∈ R ^{c × l}, wherein work as class label y _iwhen=c, h _ic component be 1, all the other are 0.

Preferably, reprojection's matrix of described definite optimum, carries out secondary dimensionality reduction by described initial dimensionality reduction training sample set, obtains secondary dimensionality reduction training sample set, is specially:

Determine optimum reprojection's matrix A, by initial dimensionality reduction training sample set carry out secondary dimensionality reduction, obtain secondary dimensionality reduction training sample set wherein x' _i∈ R ^c, wherein reprojection's matrix A deterministic process is as follows:

$\underset{A}{\min }\underset{i=1}{\overset{l}{\mathrm{\Σ}}}({\left|\right|{x^{\′}}_i-\underset{j=1}{\overset{k}{\mathrm{\Σ}}}({\mathrm{\αs}}_{\mathrm{ij}}+(1-\mathrm{\α})w_{\mathrm{ij}})x_i^{\′i}\left|\right|}^2+\mathrm{\β}{\left|\right|{x^{\′}}_i-h_i\left|\right|}^2)$

Wherein represent x' _ij Neighbor Points; β ∈ (0 ,+∞), α ∈ (0,1), α and β are all predefined values; W=(w _ij) _{c × l}for reconstruct weights matrix of coefficients, it obtains by the optimization problem solving below:

$\min \underset{i=1}{\overset{l}{\mathrm{\Σ}}}\left(\right||{\stackrel{\‾}{x}}_i-\underset{j=1}{\overset{k}{\mathrm{\Σ}}}{w}_{\mathrm{ij}}{\stackrel{\‾}{x}}_i^j|\left|\right)$

And, meet $\underset{j=1}{\overset{k}{\mathrm{\Σ}}}{w}_{\mathrm{ij}}=1,$

Wherein, represent j neighbour's sample,

S _ijthe sparse reconstruct weight matrix S=[s of composition ₁, s ₂..., s _l] ∈ R ^{c × l}, s _i, i=1,2 ..., l obtains by the optimization problem solving below:

$\underset{s_i}{\min }\left|\right|s_i\left|\right|$

And, meet $\left|\right|{\stackrel{\&OverBar;}{x}}_i-\underset{j=1}{\overset{k}{\mathrm{\&Sigma;}}}{s}_{\mathrm{ij}}{\stackrel{\&OverBar;}{x}}_i^j\left|\right|<\mathrm{\&epsiv;},1=I^T{s}_i,$

Wherein, represent j neighbour's sample, ε is greater than 0 default constant,

Solving reprojection's matrix A obtains:

A＝β(X ₁MX ₁ ^T+βX ₁X ₁ ^T) ^-TX ₁H ^T

Wherein M=(I-(α S+ (1-α) W)) ^t(I-(α S+ (1-α) W)), I is unit matrix;

Secondary dimensionality reduction training sample matrix is X ₂=AX ₁.

Preferably, a described described projection matrix and the described reprojection matrix of utilizing, carries out dimensionality reduction twice by test sample book, obtains secondary dimensionality reduction test sample book, is specially:

Utilize projection matrix A one time ₁with reprojection matrix A, test sample book x is carried out to dimensionality reduction, obtain secondary dimensionality reduction test sample book: x'=AA ₁x.

Preferably, the described nearest neighbor classifier that utilizes, classifies to described secondary dimensionality reduction test sample book, is specially:

Calculate described secondary dimensionality reduction test sample book x' and multiple secondary dimensionality reduction training sample x' _ibetween distance;

Determine the secondary dimensionality reduction training sample x' with the distance minimum of described secondary dimensionality reduction test sample book x' _i, and by this secondary dimensionality reduction training sample x' _icorresponding class label y _ibe given to described secondary dimensionality reduction test sample book x'.

Preferably, the described secondary dimensionality reduction of described calculating test sample book x' and multiple secondary dimensionality reduction training sample x' _ibetween distance, be specially:

Calculate described secondary dimensionality reduction test sample book x' and multiple secondary dimensionality reduction training sample x' _ibetween Euclidean distance.

A kind of face identification system, comprising:

Initial dimensionality reduction unit, for utilizing principle component analysis PCA to carry out initial dimensionality reduction to training sample set, obtains initial dimensionality reduction training sample set, and preserves a projection matrix in initial reduction process;

Analytical information matrix construction unit, for utilizing the class label information of training sample, constructs a matrix with classified information;

Secondary dimensionality reduction unit, for determining optimum reprojection's matrix, carries out secondary dimensionality reduction by described initial dimensionality reduction training sample set, obtains secondary dimensionality reduction training sample set;

Test sample book converter unit, for utilizing a described projection matrix and described reprojection matrix, carries out dimensionality reduction twice by test sample book, obtains secondary dimensionality reduction test sample book;

Taxon, for utilizing nearest neighbor classifier, classifies to described secondary dimensionality reduction test sample book.

Preferably, described secondary dimensionality reduction unit comprises:

Reprojection's matrix determining unit, for determining optimum reprojection's matrix;

Processing unit, for described initial dimensionality reduction training sample set is carried out to secondary dimensionality reduction, obtains secondary dimensionality reduction training sample set.

Preferably, described taxon comprises:

Metrics calculation unit, for calculating described secondary dimensionality reduction test sample book x' and multiple secondary dimensionality reduction training sample x' _ibetween distance;

Apart from comparing unit, for more described secondary dimensionality reduction test sample book x' and multiple secondary dimensionality reduction training sample x' respectively _ibetween distance, and by the minimum secondary dimensionality reduction training sample x' of distance _ichoose out;

Class label determining unit, for by the secondary dimensionality reduction training sample x' that chooses out _icorresponding class label y _ibe given to described secondary dimensionality reduction test sample book x'.

Can find out from above-mentioned technical scheme, the disclosed face identification method of the application, by PCA method, training sample set is carried out to initial dimensionality reduction, and utilize the class label information structuring of training sample to there is the matrix of classified information, then determine optimum reprojection's matrix, initial dimensionality reduction training sample set is carried out to secondary dimensionality reduction, then test sample book is carried out to secondary dimensionality reduction equally, in the lower dimensional space after secondary dimensionality reduction, classify.By secondary dimension-reduction treatment, more improve accuracy and the efficiency of recognition of face.

Brief description of the drawings

In order to be illustrated more clearly in the embodiment of the present application or technical scheme of the prior art, to the accompanying drawing of required use in embodiment or description of the Prior Art be briefly described below, apparently, accompanying drawing in the following describes is only some embodiment of the application, for those of ordinary skill in the art, do not paying under the prerequisite of creative work, can also obtain according to these accompanying drawings other accompanying drawing.

Fig. 1 is the disclosed a kind of face identification method process flow diagram of the embodiment of the present application;

Fig. 2 is the disclosed a kind of face identification system structural representation of the embodiment of the present application;

Fig. 3 is the disclosed secondary dimensionality reduction of the embodiment of the present application cellular construction figure;

Fig. 4 is the disclosed taxon structural drawing of the embodiment of the present application.

Embodiment

Below in conjunction with the accompanying drawing in the embodiment of the present application, the technical scheme in the embodiment of the present application is clearly and completely described, obviously, described embodiment is only some embodiments of the present application, instead of whole embodiment.Based on the embodiment in the application, those of ordinary skill in the art are not paying all other embodiment that obtain under creative work prerequisite, all belong to the scope of the application's protection.

Embodiment mono-

Referring to Fig. 1, Fig. 1 is the disclosed a kind of face identification method process flow diagram of the embodiment of the present application.

As shown in Figure 1, the method comprises:

Step 101: utilize principle component analysis PCA to carry out initial dimensionality reduction to training sample set, obtain initial dimensionality reduction training sample set, and preserve a projection matrix in initial reduction process;

Particularly, principle component analysis PCA is a kind of existing dimension reduction method, and a projection matrix of using in reduction process can be provided, and by a projection matrix, training sample is carried out to dimension-reduction treatment.

Step 102: utilize the class label information of training sample, construct a matrix with classified information;

Particularly, what each training sample was corresponding has a class label, and this class label is indicated the classification of this training sample.

Step 103: determine optimum reprojection's matrix, described initial dimensionality reduction training sample set is carried out to secondary dimensionality reduction, obtain secondary dimensionality reduction training sample set;

Particularly, find optimum reprojection's matrix, the training sample set through initial dimensionality reduction is carried out to secondary dimensionality reduction, project to the space that dimension is lower.

Step 104: utilize a described projection matrix and described reprojection matrix, test sample book is carried out to dimensionality reduction twice, obtain secondary dimensionality reduction test sample book;

Particularly, in order better sample to be tested, test sample book is projected to the space at training sample place.

Step 105: utilize nearest neighbor classifier, described secondary dimensionality reduction test sample book is classified.

The disclosed face identification method of the embodiment of the present application, by PCA method, training sample set is carried out to initial dimensionality reduction, and utilize the class label information structuring of training sample to there is the matrix of classified information, then determine optimum reprojection's matrix, initial dimensionality reduction training sample set is carried out to secondary dimensionality reduction, then test sample book is carried out to secondary dimensionality reduction equally, in the lower dimensional space after secondary dimensionality reduction, classify.By secondary dimension-reduction treatment, more improve accuracy and the efficiency of recognition of face.

Embodiment bis-

In the present embodiment, will be described in detail above-mentioned each step.

(1), utilize principle component analysis PCA to carry out initial dimensionality reduction to training sample set, obtain initial dimensionality reduction training sample set, and preserve a projection matrix in initial reduction process.

Particularly, definition training sample set is x _i∈ R ^d, y _i=1,2 ..., c} is training sample x _iclass label information, wherein D is the dimension of training sample, l is the number of training sample data, c is the classification number of training sample data;

Utilize pivot analysis method of descent PCA to carry out initial dimensionality reduction to training sample, obtain initial dimensionality reduction training sample set $\{{\stackrel{\&OverBar;}{x}}_i,y_i{\}}_{i=1}^l,{\stackrel{\&OverBar;}{x}}_i\&Element;R^d,$ y _i＝{1,2,…,c}；

Definition training sample matrix is X=[x ₁, x ₂..., x _l] ∈ R ^{d × l}, initial dimensionality reduction training sample matrix X ₁=Α ₁x, wherein A ₁∈ R ^{d × D}it is the projection matrix that PCA method of descent obtains.

(2), utilize the class label information of training sample, construct a matrix with classified information.

Be specially: the class label information y that utilizes training sample _i, construct the matrix with classified information a: H=[h ₁, h ₂..., h _l] ∈ R ^{c × l}, wherein work as classification y _iwhen=c, h _ic component be 1, all the other are 0.

(3), determine and optimum reprojection's matrix described initial dimensionality reduction training sample set carried out to secondary dimensionality reduction, obtain secondary dimensionality reduction training sample set.

Be specially:

Determine optimum reprojection's matrix A, by initial dimensionality reduction training sample set carry out secondary dimensionality reduction, obtain secondary dimensionality reduction training sample set wherein x' _i∈ R ^c, wherein reprojection's matrix A deterministic process is as follows:

$\underset{A}{\min }\underset{i=1}{\overset{l}{\mathrm{\&Sigma;}}}({\left|\right|{x^{\&prime;}}_i-\underset{j=1}{\overset{k}{\mathrm{\&Sigma;}}}({\mathrm{\&alpha;s}}_{\mathrm{ij}}+(1-\mathrm{\&alpha;})w_{\mathrm{ij}})x_i^{\&prime;i}\left|\right|}^2+\mathrm{\&beta;}{\left|\right|{x^{\&prime;}}_i-h_i\left|\right|}^2)$

Wherein represent x' _ij Neighbor Points; β ∈ (0 ,+∞), α ∈ (0,1), α and β are all predefined values; W=(w _ij) _{c × l}for reconstruct weights matrix of coefficients, it obtains by the optimization problem solving below:

$\min \underset{i=1}{\overset{l}{\mathrm{\&Sigma;}}}\left(\right||{\stackrel{\&OverBar;}{x}}_i-\underset{j=1}{\overset{k}{\mathrm{\&Sigma;}}}{w}_{\mathrm{ij}}{\stackrel{\&OverBar;}{x}}_i^j|\left|\right)$

And, meet $\underset{j=1}{\overset{k}{\mathrm{\&Sigma;}}}{w}_{\mathrm{ij}}=1,$

Wherein, represent j neighbour's sample,

S _ijthe sparse reconstruct weight matrix S=[s of composition ₁, s ₂..., s _l] ∈ R ^{c × l}, s _i, i=1,2 ..., l obtains by the optimization problem solving below:

$\underset{s_i}{\min }\left|\right|s_i\left|\right|$

And, meet $\left|\right|{\stackrel{\&OverBar;}{x}}_i-\underset{j=1}{\overset{k}{\mathrm{\&Sigma;}}}{s}_{\mathrm{ij}}{\stackrel{\&OverBar;}{x}}_i^j\left|\right|<\mathrm{\&epsiv;},1=I^T{s}_i,$

Wherein, represent j neighbour's sample, ε is greater than 0 default constant,

Finally, solving reprojection's matrix A obtains:

A＝β(X ₁MX ₁ ^T+βX ₁X ₁ ^T) ^-TX ₁H ^T

Wherein M=(I-(α S+ (1-α) W)) ^t(I-(α S+ (1-α) W)), I is unit matrix;

Secondary dimensionality reduction training sample matrix is X ₂=AX ₁.

(4), utilize a described projection matrix and described reprojection matrix, test sample book is carried out to dimensionality reduction twice, obtain secondary dimensionality reduction test sample book, and utilize nearest neighbor classifier, described secondary dimensionality reduction test sample book is classified.

Be specially:

Utilize projection matrix A one time ₁with reprojection matrix A, test sample book x is carried out to dimensionality reduction, obtain secondary dimensionality reduction test sample book: x'=AA ₁x;

Calculate described secondary dimensionality reduction test sample book x' and multiple secondary dimensionality reduction training sample x' _ibetween distance;

Determine the secondary dimensionality reduction training sample x' with the distance minimum of described secondary dimensionality reduction test sample book x' _i, and by this secondary dimensionality reduction training sample x' _icorresponding class label y _ibe given to described secondary dimensionality reduction test sample book x'.

Wherein, calculate described secondary dimensionality reduction test sample book x' and multiple secondary dimensionality reduction training sample x' _ibetween distance can be Euclidean distance.Distance is less, is representing that two goodnesses of fit between sample are higher.

Embodiment tri-

Referring to Fig. 2, Fig. 2 is the disclosed a kind of face identification system structural representation of the embodiment of the present application.

Corresponding with the disclosed face identification method of above-described embodiment, the present embodiment discloses a kind of face identification system, as shown in Figure 2:

Initial dimensionality reduction unit 21, for utilizing principle component analysis PCA to carry out initial dimensionality reduction to training sample set, obtains initial dimensionality reduction training sample set, and preserves a projection matrix in initial reduction process;

Analytical information matrix construction unit 22, for utilizing the class label information of training sample, constructs a matrix with classified information;

Secondary dimensionality reduction unit 23, for determining optimum reprojection's matrix, carries out secondary dimensionality reduction by described initial dimensionality reduction training sample set, obtains secondary dimensionality reduction training sample set;

Test sample book converter unit 24, for utilizing a described projection matrix and described reprojection matrix, carries out dimensionality reduction twice by test sample book, obtains secondary dimensionality reduction test sample book;

Taxon 25, for utilizing nearest neighbor classifier, classifies to described secondary dimensionality reduction test sample book.

Wherein, the specific works principle of unit can, referring to the introduction of above-described embodiment, not repeat them here.The disclosed face identification system of the embodiment of the present application, by PCA method, training sample set is carried out to initial dimensionality reduction, and utilize the class label information structuring of training sample to there is the matrix of classified information, then determine optimum reprojection's matrix, initial dimensionality reduction training sample set is carried out to secondary dimensionality reduction, then test sample book is carried out to secondary dimensionality reduction equally, in the lower dimensional space after secondary dimensionality reduction, classify.By secondary dimension-reduction treatment, more improve accuracy and the efficiency of recognition of face.

It should be noted that, referring to Fig. 3, Fig. 3 is the disclosed secondary dimensionality reduction of the embodiment of the present application cellular construction figure.As shown in Figure 3, secondary dimensionality reduction unit 23 can be further divided into:

For determining reprojection's matrix determining unit 231 of optimum reprojection's matrix and for described initial dimensionality reduction training sample set is carried out to secondary dimensionality reduction, obtaining the processing unit 232 of secondary dimensionality reduction training sample set.

It should be noted that, referring to Fig. 4, Fig. 4 is the disclosed taxon structural drawing of the embodiment of the present application.As shown in Figure 3, taxon 25 comprises:

Metrics calculation unit 251, for calculating described secondary dimensionality reduction test sample book x' and multiple secondary dimensionality reduction training sample x' _ibetween distance;

Apart from comparing unit 252, for more described secondary dimensionality reduction test sample book x' and multiple secondary dimensionality reduction training sample x' respectively _ibetween distance, and by the minimum secondary dimensionality reduction training sample x' of distance _ichoose out;

Class label determining unit 253, for by the secondary dimensionality reduction training sample x' that chooses out _icorresponding class label y _ibe given to described secondary dimensionality reduction test sample book x'.

Finally, also it should be noted that, in this article, relational terms such as the first and second grades is only used for an entity or operation to separate with another entity or operational zone, and not necessarily requires or imply and between these entities or operation, have the relation of any this reality or sequentially.And, term " comprises ", " comprising " or its any other variant are intended to contain comprising of nonexcludability, thereby the process, method, article or the equipment that make to comprise a series of key elements not only comprise those key elements, but also comprise other key elements of clearly not listing, or be also included as the intrinsic key element of this process, method, article or equipment.The in the situation that of more restrictions not, the key element being limited by statement " comprising ... ", and be not precluded within process, method, article or the equipment that comprises described key element and also have other identical element.

In this instructions, each embodiment adopts the mode of going forward one by one to describe, and what each embodiment stressed is and the difference of other embodiment, between each embodiment identical similar part mutually referring to.

To the above-mentioned explanation of the disclosed embodiments, make professional and technical personnel in the field can realize or use the application.To be apparent for those skilled in the art to the multiple amendment of these embodiment, General Principle as defined herein can, in the case of not departing from the application's spirit or scope, realize in other embodiments.Therefore, the application will can not be restricted to these embodiment shown in this article, but will meet the widest scope consistent with principle disclosed herein and features of novelty.

Claims (10)

1. a face identification method, is characterized in that, comprising:

Utilize principle component analysis PCA to carry out initial dimensionality reduction to training sample set, obtain initial dimensionality reduction training sample set, and preserve a projection matrix in initial reduction process;

Utilize the class label information of training sample, construct a matrix with classified information;

Determine optimum reprojection's matrix, described initial dimensionality reduction training sample set is carried out to secondary dimensionality reduction, obtain secondary dimensionality reduction training sample set;

Utilize a described projection matrix and described reprojection matrix, test sample book is carried out to dimensionality reduction twice, obtain secondary dimensionality reduction test sample book;

Utilize nearest neighbor classifier, described secondary dimensionality reduction test sample book is classified.

2. face identification method according to claim 1, is characterized in that, the described principle component analysis PCA that utilizes carries out initial dimensionality reduction to training sample set, obtains initial dimensionality reduction training sample set, and preserves a projection matrix in initial reduction process, is specially:

Definition training sample set is x _i∈ R ^d, y _i=1,2 ..., c} is training sample x _iclass label information, wherein D is the dimension of training sample, l is the number of training sample data, c is the classification number of training sample data;

Utilize pivot analysis method of descent PCA to carry out initial dimensionality reduction to training sample, obtain initial dimensionality reduction training sample set $\{{\stackrel{\&OverBar;}{x}}_i,y_i{\}}_{i=1}^l,{\stackrel{\&OverBar;}{x}}_i\&Element;R^d,$ y _i＝{1,2,…,c}；

Definition training sample matrix is X=[x ₁, x ₂..., x _l] ∈ R ^{d × l}, initial dimensionality reduction training sample matrix X ₁=Α ₁x, wherein A ₁∈ R ^{d × D}it is the projection matrix that PCA method of descent obtains.

3. face identification method according to claim 2, is characterized in that, the described class label information of utilizing training sample is constructed a matrix with classified information, is specially:

Utilize the class label information y of training sample _i, construct the matrix with classified information a: H=[h ₁, h ₂..., h _l] ∈ R ^{c × l}, wherein work as class label y _iwhen=c, h _ic component be 1, all the other are 0.

4. face identification method according to claim 3, is characterized in that, described initial dimensionality reduction training sample set is carried out secondary dimensionality reduction by reprojection's matrix of described definite optimum, obtains secondary dimensionality reduction training sample set, is specially:

Determine optimum reprojection's matrix A, by initial dimensionality reduction training sample set carry out secondary dimensionality reduction, obtain secondary dimensionality reduction training sample set wherein x' _i∈ R ^c, wherein reprojection's matrix A deterministic process is as follows:

$\underset{A}{\min }\underset{i=1}{\overset{l}{\mathrm{\&Sigma;}}}({\left|\right|{x^{\&prime;}}_i-\underset{j=1}{\overset{k}{\mathrm{\&Sigma;}}}({\mathrm{\&alpha;s}}_{\mathrm{ij}}+(1-\mathrm{\&alpha;})w_{\mathrm{ij}})x_i^{\&prime;i}\left|\right|}^2+\mathrm{\&beta;}{\left|\right|{x^{\&prime;}}_i-h_i\left|\right|}^2)$

Wherein represent x' _ij Neighbor Points; β ∈ (0 ,+∞), α ∈ (0,1), α and β are all predefined values; W=(w _ij) _{c × l}for reconstruct weights matrix of coefficients, it obtains by the optimization problem solving below:

$\min \underset{i=1}{\overset{l}{\mathrm{\&Sigma;}}}\left(\right||{\stackrel{\&OverBar;}{x}}_i-\underset{j=1}{\overset{k}{\mathrm{\&Sigma;}}}{w}_{\mathrm{ij}}{\stackrel{\&OverBar;}{x}}_i^j|\left|\right)$

And, meet $\underset{j=1}{\overset{k}{\mathrm{\&Sigma;}}}{w}_{\mathrm{ij}}=1,$

Wherein, represent j neighbour's sample,

S _ijthe sparse reconstruct weight matrix S=[s of composition ₁, s ₂..., s _l] ∈ R ^{c × l}, s _i, i=1,2 ..., l obtains by the optimization problem solving below:

$\underset{s_i}{\min }\left|\right|s_i\left|\right|$

And, meet $\left|\right|{\stackrel{\&OverBar;}{x}}_i-\underset{j=1}{\overset{k}{\mathrm{\&Sigma;}}}{s}_{\mathrm{ij}}{\stackrel{\&OverBar;}{x}}_i^j\left|\right|<\mathrm{\&epsiv;},1=I^T{s}_i,$

Wherein, represent j neighbour's sample, ε is greater than 0 default constant,

Solving reprojection's matrix A obtains:

A＝β(X ₁MX ₁ ^T+βX ₁X ₁ ^T) ^-TX ₁H ^T

Wherein M=(I-(α S+ (1-α) W)) ^t(I-(α S+ (1-α) W)), I is unit matrix;

Secondary dimensionality reduction training sample matrix is X ₂=AX ₁.

5. face identification method according to claim 4, is characterized in that, a described described projection matrix and the described reprojection matrix of utilizing, carries out dimensionality reduction twice by test sample book, obtains secondary dimensionality reduction test sample book, is specially:

Utilize projection matrix A one time ₁with reprojection matrix A, test sample book x is carried out to dimensionality reduction, obtain secondary dimensionality reduction test sample book: x'=AA ₁x.

6. face identification method according to claim 5, is characterized in that, the described nearest neighbor classifier that utilizes is classified to described secondary dimensionality reduction test sample book, is specially:

Calculate described secondary dimensionality reduction test sample book x' and multiple secondary dimensionality reduction training sample x' _ibetween distance;

Determine the secondary dimensionality reduction training sample x' with the distance minimum of described secondary dimensionality reduction test sample book x' _i, and by this secondary dimensionality reduction training sample x' _icorresponding class label y _ibe given to described secondary dimensionality reduction test sample book x'.

7. face identification method according to claim 6, is characterized in that, the described secondary dimensionality reduction of described calculating test sample book x' and multiple secondary dimensionality reduction training sample x' _ibetween distance, be specially:

Calculate described secondary dimensionality reduction test sample book x' and multiple secondary dimensionality reduction training sample x' _ibetween Euclidean distance.

8. a face identification system, is characterized in that, comprising:

Initial dimensionality reduction unit, for utilizing principle component analysis PCA to carry out initial dimensionality reduction to training sample set, obtains initial dimensionality reduction training sample set, and preserves a projection matrix in initial reduction process;

Analytical information matrix construction unit, for utilizing the class label information of training sample, constructs a matrix with classified information;

Secondary dimensionality reduction unit, for determining optimum reprojection's matrix, carries out secondary dimensionality reduction by described initial dimensionality reduction training sample set, obtains secondary dimensionality reduction training sample set;

Test sample book converter unit, for utilizing a described projection matrix and described reprojection matrix, carries out dimensionality reduction twice by test sample book, obtains secondary dimensionality reduction test sample book;

Taxon, for utilizing nearest neighbor classifier, classifies to described secondary dimensionality reduction test sample book.

9. face identification system according to claim 8, is characterized in that, described secondary dimensionality reduction unit comprises:

Reprojection's matrix determining unit, for determining optimum reprojection's matrix;

Processing unit, for described initial dimensionality reduction training sample set is carried out to secondary dimensionality reduction, obtains secondary dimensionality reduction training sample set.

10. face identification system according to claim 9, is characterized in that, described taxon comprises:

Metrics calculation unit, for calculating described secondary dimensionality reduction test sample book x' and multiple secondary dimensionality reduction training sample x' _ibetween distance;

Apart from comparing unit, for more described secondary dimensionality reduction test sample book x' and multiple secondary dimensionality reduction training sample x' respectively _ibetween distance, and by the minimum secondary dimensionality reduction training sample x' of distance _ichoose out;

Class label determining unit, for by the secondary dimensionality reduction training sample x' that chooses out _icorresponding class label y _ibe given to described secondary dimensionality reduction test sample book x'.