CN101281598A - Method for recognizing human face based on amalgamation of multicomponent and multiple characteristics - Google Patents

Abstract

The invention relates to a face identification method based on the multi-component and multi-feature fusion, belonging to the technical field of the image processing, the computer vision, the pattern recognition. The method includes the following steps: processing Gabor filter on the face images of all persons for training, obtaining the Gabor characteristic image of five parts, respectively extracting the projecting characteristic value of the face grey level image of five persons on the Gabor characteristic image by the same method of the known person and the person to be recognized; obtaining the mixed projecting characteristic value of the face parts of five persons; respectively computing the europeanism distance between the mixed projecting characteristic value of the person to be recognized and the known person as the degree of similarity R, to obtain the integrate similarity R0, if R0 is larger than or equal to T, the person to be recognized is the same person as the known person; if R0 is smaller than T, the person to be recognized is not the same person as the known person. The invention has higher face recognition rate.

Description

Face identification method based on the multicomponent and multiple characteristics fusion

Technical field

The invention belongs to Flame Image Process, computer vision, mode identification technology, particularly face identification method.

Background technology

Biometrics identification technology is the effective technology of identification, recently with fastest developing speed is face recognition technology and the biometrics identification technology that merges mutually with face recognition technology.Therefore, the present invention relates to man face image acquiring and recognizer, have important use and be worth.

Present existing face identification method is mainly discerned whole people's face, and in many recognition methodss, mainly adopt methods such as principal component analysis (PCA-Principal Component Analysis), Elastic Matching, neural network, geometric properties.

Simultaneously, the difficult point of recognition of face also is:

(1) people's face plastic yield of expressing one's feelings and causing

(2) people's face diversity of causing of attitude

(3) people's face of causing of age changes

(4) multiplicity of people's face pattern of causing of factors such as hair style, beard, glasses, makeup

(5) otherness of the facial image that causes of factors such as the angle of illumination, intensity and sensor characteristic

Therefore many factors make recognition of face become a thorny and challenging problem, are also becoming the focus of scientific research in recent years.

Prior art related to the present invention is described as follows:

Multi-mode face identification method based on the face component principal component analysis: its principal feature is people's face to be carried out parts extract, and again face component is carried out principal component analysis and multi-mode identification, to reach high discrimination.This method comprises:

1) at first adopt the method for template matches and projection histogram that facial image is located, determine the home position on people's face coarse positioning district, left and right sides eyeball, nose, mouth, lower jaw summit, from whole people's face, extract naked face, eyebrow+eyes, eyes, nose, five kinds of face components of mouth then;

2) five kinds of face components of naked face, eyebrow+eyes, eyes, nose, mouth to from training set people face, extracting, utilize the eigenface method in the principal component method, form the naked face of feature, feature (eyes+eyebrow), feature eyes, feature nose, feature face respectively;

3) five kinds of face components of naked face, eyebrow+eyes, eyes, nose, mouth that the facial image of known person is extracted, utilize the projection properties value analytical approach in the principal component method, the projection properties value of the naked face of extraction known person face, eyes+eyebrow, eyes, nose, five kinds of face components of mouth;

4) five kinds of face components of naked face, eyebrow+eyes, eyes, nose, mouth that people's to be identified facial image is extracted, utilize the feature projection value analytical approach in the principal component method, extract naked face, eyes+eyebrow, eyes, the nose of people's face to be identified, the projection properties value of face;

5) calculate similarity between the projection properties value of the projection properties value of known face component image and people's face corresponding component image to be identified respectively; Similarity fusion to naked face, eyes+eyebrow, eyes, nose, face obtains multimodal overall face identification method, the identification of the similarity of single naked face, eyes+eyebrow, eyes, nose, mouth, or naked face, eyes+eyebrow, eyes, nose, mouth combination identification each other are exactly multimodal local face identification method.

The Gabor method: its principal feature is one group of similar Gabor function that the Gabor basis function is obtained later on through displacement, rotation and transformation of scale, can describe the frequency structure in the picture of publishing picture when keeping spatial relationship, and can provide the result in spatial domain.In the application of recognition of face, owing to Gabor wavelet multiresolution rate, multi-direction reflect the image local characteristic, thus insensitive and use overall gray feature to discern and compare for the reaction of illumination, illumination is had better adaptability.

The two-dimensional Gabor function is equivalent to the two-dimensional Gaussian function of the multiple sine function modulation of a quilt, then being that the result after the translation has all taken place on two frequency axiss two-dimensional Gaussian function on the frequency domain, is a two-dimentional bandpass filter.Because each Gabor wave filter is equivalent to a bandpass filter,, can adopt a plurality of Gabor wave filters on the different scale different directions to form bank of filters usually in order to extract the feature of facial image on a plurality of yardsticks of a plurality of directions.When carrying out filtering, with input picture successively with each wave filter convolution of bank of filters, and get its amplitude, i.e. the Gabor image of input picture as output.

The weighted sum rule: for different features, the identification sex all is not quite similar, and the weighted sum rule is exactly to adopt different weights to merge to different features.The weights of each feature are determined by the characteristic of this feature itself (separability, discrimination etc.), the fusion weights that different fusion features is corresponding different.Give bigger weights to the good feature of recognition performance, and the feature of recognition performance difference is given less weights.

Summary of the invention

The objective of the invention is in order to improve the adaptability of face recognition algorithms, propose a kind of face identification method that merges based on multicomponent and multiple characteristics, this method has higher recognition of face rate.

The face identification method based on the multicomponent and multiple characteristics fusion that the present invention proposes is characterized in that, may further comprise the steps:

1) facial image is carried out five yardsticks from all directions to Gabor filtering, obtain corresponding people's face Gabor image;

2) from people's face Gabor image, extract naked face Gabor image, eyebrow+eyes Gabor image, eyes Gabor image, nose Gabor image, five kinds of parts Gabor of face Gabor image image;

3) respectively to described naked face Gabor image, eyebrow+eyes Gabor image, eyes Gabor image, nose Gabor image, the even piecemeal of face Gabor image, the mean value of all picture elements is as the feature picture element of this piece in getting every, all feature picture elements are combined into the Gabor characteristic image of five kinds of parts, realize the dimensionality reduction of five kinds of parts Gabor images;

4) all images in the training set is adopted described step 1)-3) the Gabor characteristic image of five kinds of parts obtaining, utilization forms the naked face of Gabor feature, Gabor feature eyebrow+eyes, Gabor feature eyes, Gabor feature nose, Gabor feature face based on the eigenface method in the principal component method;

5) facial image to known person adopts described step 1)-3) obtain the Gabor characteristic image of five kinds of parts, utilization is based on the projection properties method of vector analysis in the principal component method, extracts naked face Gabor image, eyebrow+eyes Gabor image, eyes Gabor image, the nose Gabor image of described known person face, the projection properties vector of face Gabor image respectively;

6) utilize multi-mode face identification method, extract naked face, eyes+eyebrow, eyes, the nose of known facial image, the projection properties value of five kinds of face component gray level images of face based on the parts principal component analysis;

7) the gray level image projection properties value of the same parts of known facial image and corresponding Gabor image projection eigenwert are merged according to the weighted sum rule, obtain naked face, eyes+eyebrow, eyes, the nose of known facial image, the mixing projection properties value of five kinds of face components of face;

8) people's to be identified facial image is adopted step 1)-3) obtain the Gabor characteristic image of five kinds of parts, utilization is based on the projection properties method of vector analysis in the principal component method, extracts naked face Gabor image, eyebrow+eyes Gabor image, eyes Gabor image, the nose Gabor image of described people's face to be identified, the projection properties vector of face Gabor image respectively;

9) utilize multi-mode face identification method, extract naked face, eyes+eyebrow, eyes, the nose of facial image to be identified, the projection properties value of five kinds of face component gray level images of face based on the parts principal component analysis;

10) the gray level image projection properties value of the same parts of facial image to be identified and corresponding Gabor image projection eigenwert are merged according to the weighted sum rule, obtain naked face, eyes+eyebrow, eyes, the nose of facial image to be identified, the mixing projection properties value of five kinds of face components of face;

11) calculate Euclidean distance between the mixing projection properties value of the mixing projection properties value of each parts of people's face to be identified and each parts of known person face respectively as the similarity R of the image of component of people's face to be identified and known person face, each image of component similarity is respectively naked face image R1, eyes+eyebrow image R2, eye image R3, nose image R4, face image R5;

12) similarity R1, R2, R3, R4, R5 are merged according to the weighted sum rule, obtain the people's face to be identified and the comprehensive similarity R0 of known person face, this R0 is as the human face similarity degree of recognition of face;

13) comparison step 12) similarity R0 that obtains and the size of pre-set threshold T, T chooses the similarity value that wrong acceptance rate is 0.1% o'clock correspondence, if R0 〉=T judges that then people to be identified and known person are same individuals; If R0＜T judges that then people to be identified and known person are not same individuals.

Characteristics of the present invention and effect

The present invention carries out parts to gray level image and Gabor image to be cut apart, and gray level image and the Gabor image to parts carries out principal component analysis and fusion again, has higher recognition of face rate.

Embodiment

The face identification method based on the multicomponent and multiple characteristics fusion that the present invention proposes reaches embodiment in conjunction with the accompanying drawings and is described in detail as follows, and may further comprise the steps:

1) facial image is carried out five yardsticks from all directions to Gabor filtering, obtain corresponding people's face Gabor image;

The definition of Gabor wave filter is suc as formula (1):

${\mathrm{\ψ}}_{\mathrm{\μ},v}\left(z\right)=\frac{{\left|\right|k_{\mathrm{\μ},v}\left|\right|}^2}{{\mathrm{\σ}}^2}{e}^{-\frac{{\left|\right|k_{\mathrm{\μ},v}\left|\right|}^2{\left|\right|z\left|\right|}^2}{2{\mathrm{\σ}}^2}}[e^{{\mathrm{ik}}_{\mathrm{\μ},v}z}-e^{-\frac{{\mathrm{\σ}}^2}{2}}]---\left(1\right)$

Wherein, z=(x y) is the corresponding point coordinate, and wave vector is defined as: $k_{\mathrm{\μ},v}=k_v{e}^{i{\mathrm{\θ}}_{\mathrm{\μ}}},$ K wherein _v=k _Max/ λ ^vAnd θ _μ=π μ/n.V and μ have defined the yardstick and the direction of Gabor wave filter respectively.Get n=8, v ∈ 0,1,2,3,4}, μ ∈ 0,1 ..., 7}, σ=2 π, k _Max=pi/2 and $\mathrm{\λ}=\sqrt{2}$ Obtain five yardsticks from all directions to the Gabor bank of filters.

Facial image I (z) is carried out convolution with the Gabor wave filter, get the amplitude part of convolution results, obtain corresponding people's face Gabor image A _{μ, v}(z).

O _μ，v(z)＝I(z)*ψ _μ，v(z)(2)

$A_{\mathrm{\μ},v}\left(z\right)=\sqrt{\mathrm{Re}{\left(O_{\mathrm{\μ},v}\left(z\right)\right)}^2+\mathrm{Im}{\left(O_{\mathrm{\μ},v}\left(z\right)\right)}^2}---\left(3\right)$

2) from people's face Gabor image A _{μ, v}(z) extract naked face Gabor image, eyebrow+eyes Gabor image, eyes Gabor image, nose Gabor image, five kinds of parts Gabor of face Gabor image image in, naked face Gabor size of images is 90 * 120, eyebrow+eyes Gabor size of images is 45 * 35, eyes Gabor size of images is 40 * 20, nose Gabor size of images is 20 * 30, and face Gabor size of images is 24 * 32.

3) respectively to naked face Gabor image, eyebrow+eyes Gabor image, eyes Gabor image, nose Gabor image, the even piecemeal of face Gabor image, every block size is 3 * 3, the mean value of all picture elements is as the feature picture element of this piece in getting every, all feature picture elements are combined into the Gabor characteristic image, realize the dimensionality reduction of five kinds of parts Gabor images, the Gabor characteristic image size of five parts is respectively 30 * 40,15 * 11,13 * 6,6 * 10,8 * 10 behind the dimensionality reduction;

4) all images in the training set is adopted described step 1)-3) the Gabor characteristic image of five kinds of parts obtaining, utilization forms the naked face of Gabor feature, Gabor feature eyebrow+eyes, Gabor feature eyes, Gabor feature nose, Gabor feature face based on the eigenface method in the principal component method;

Concrete way is:

The vector set of representing in the training set everyone naked face of face Gabor feature or Gabor feature eyebrow+eyes or Gabor feature eyes or Gabor feature nose or Gabor feature face with n * N matrix X respectively, n is the pixel count of the naked face of people's face Gabor feature or Gabor feature eyebrow+eyes or Gabor feature eyes or Gabor feature nose or Gabor feature face, N (N＞1000) is a training set people face sum, then:

$C=\frac{1}{N}{\mathrm{XX}}^T,X=(X_1,\·\·\·,X_k,\·\·\·,X_N)$

(4) X in the formula _k=(x _1k, x _2k..., x _Nk) ^T, k=(1,2 ..., N) can represent a naked face vector of people's face Gabor feature, Gabor feature eyebrow+eyes vector, Gabor feature eyes vector, Gabor feature nose vector, Gabor feature face vector, X ^TThe transposition of representing matrix X.

When the proper vector of compute matrix C and eigenwert, owing to calculate XX ^TThe very big (n of dimension ²Dimension), and adopts svd, change into and calculate X ^TX can obtain proper vector and the eigenwert of C so indirectly, and X ^TThe dimension of X is reduced to N ²Dimension, XX ^TWith X ^TThe eigenwert of X is the same, and the relation of the proper vector between them satisfies following formula:

$u_k=\frac{1}{{\mathrm{\λ}}_k}\×{\mathrm{\φ}}_k---\left(5\right)$

(5) u in the formula _kBe XX ^TProper vector, φ _kBe X ^TThe proper vector of X; λ _kBe X ^TThe eigenwert of X also is XX simultaneously ^TEigenwert.By calculating X ^TThe eigenvalue of X _kWith proper vector φ _k, and obtain XX ^TProper vector u _kAccording to λ _kNumerical value is by sorting from big to small, D before taking out (D＜＜N) individual maximum eigenwert and keep corresponding with it D proper vector u _kJust form the naked face of Gabor feature, Gabor feature eyebrow+eyes, Gabor feature eyes, Gabor feature nose, Gabor feature face, for described five kinds of parts, the value of D is respectively 120,60,80,50,40.

5) facial image to known person adopts described step 1)-3) obtain the Gabor characteristic image of five kinds of parts, the computing of (6) formula is promptly based on the projection properties method of vector analysis in the principal component method below adopting, and extracts naked face Gabor image, eyebrow+eyes Gabor image, eyes Gabor image, the nose Gabor image of described known person face, the projection properties vector of face Gabor image respectively;

$B=u_k^T\×q,k=\mathrm{1,2},...,D---\left(6\right)$

(6) q is respectively naked face, eyes+eyebrow, eyes, the nose of known person face, the Gabor characteristic image vector of face, u in the formula _kBe respectively the Gabor eigenface, Gabor feature eyes+eyebrow, Gabor feature eyes, Gabor feature nose, the Gabor feature face that from training set people face, obtain.

6) utilize multi-mode face identification method, extract naked face, eyes+eyebrow, eyes, the nose of known facial image, the projection properties value of five kinds of face component gray level images of face based on the parts principal component analysis;

7) the gray level image projection properties value of the same parts of known facial image and corresponding Gabor image projection eigenwert are merged according to the weighted sum rule, integration percentage is 1: 0.75, obtains naked face, eyes+eyebrow, eyes, the nose of known facial image, the mixing projection properties value of five kinds of face components of face;

8) people's to be identified facial image is adopted step 1)-3) obtain the Gabor characteristic image of five kinds of parts, the computing of (7) formula is promptly based on the projection properties method of vector analysis in the principal component method below adopting, and extracts naked face Gabor image, eyebrow+eyes Gabor image, eyes Gabor image, the nose Gabor image of described people's face to be identified, the projection properties vector of face Gabor image respectively;

$A=u_k^T\×p,k=\mathrm{1,2},...,D---\left(7\right)$

(7) p is respectively naked face, eyes+eyebrow, eyes, the nose of people's face to be identified, the Gabor characteristic image vector of face, u in the formula _kBe respectively the Gabor eigenface, Gabor feature eyes+eyebrow, Gabor feature eyes, Gabor feature nose, the Gabor feature face that from training set people face, obtain.

9) utilize multi-mode face identification method, extract naked face, eyes+eyebrow, eyes, the nose of facial image to be identified, the projection properties value of five kinds of face component gray level images of face based on the parts principal component analysis;

10) the gray level image projection properties value of the same parts of facial image to be identified and corresponding Gabor image projection eigenwert are merged according to the weighted sum rule, integration percentage is 1: 0.75, obtains naked face, eyes+eyebrow, eyes, the nose of facial image to be identified, the mixing projection properties value of five kinds of face components of face;

11) calculate the similarity R of the image of component of people's face to be identified and known person face respectively according to (8) formula, each image of component similarity is respectively naked face image R1, eyes+eyebrow image R2, eye image R3, nose image R4, face image R5;

$R=\sqrt{1-\frac{\left|\right|A-B\left|\right|}{\left|\right|A\left|\right|+\left|\right|B\left|\right|}}---\left(8\right)$

Wherein, B is respectively 7) in naked face, eyes+eyebrow, eyes, nose, the mixing projection properties value of five kinds of face components of face of known facial image, A is respectively 10) in naked face, eyes+eyebrow, eyes, nose, the mixing projection properties value of five kinds of face components of face of facial image to be identified.

12) similarity R1, R2, R3, R4, R5 are merged according to the weighted sum rule, its fusion coefficients is got respectively and is done 16: 1.5: 2.5: 2: 3, obtain the people's face to be identified and the comprehensive similarity R0 of known person face, with the human face similarity degree of R0 as recognition of face;

13) choosing wrong acceptance rate is that the similarity value of 0.1% o'clock correspondence is predetermined threshold value T, T=87, comparison step 12) similarity R0 that obtains and the size of pre-set threshold T, if R0 〉=T judges that then people to be identified and known person are same individuals; If R0＜T judges that then people to be identified and known person are not same individuals.

Certification of Second Generation real name identity authorization system VC++ Programming with Pascal Language based on recognition of face.The discrimination that is reached is: when false acceptance rate was 0.1%, correct recognition rata was 78.96%.

Claims (1)

1, a kind of face identification method that merges based on multicomponent and multiple characteristics is characterized in that this method may further comprise the steps:

1) facial image is carried out five yardsticks from all directions to Gabor filtering, obtain corresponding people's face Gabor image;

2) from people's face Gabor image, extract naked face Gabor image, eyebrow+eyes Gabor image, eyes Gabor image, nose Gabor image, five kinds of parts Gabor of face Gabor image image;

3) respectively to described naked face Gabor image, eyebrow+eyes Gabor image, eyes Gabor image, nose Gabor image, the even piecemeal of face Gabor image, the mean value of all picture elements is as the feature picture element of this piece in getting every, all feature picture elements are combined into the Gabor characteristic image of five kinds of parts, realize the dimensionality reduction of five kinds of parts Gabor images;

4) all images in the training set is adopted described step 1)-3) the Gabor characteristic image of five kinds of parts obtaining, utilization forms the naked face of Gabor feature, Gabor feature eyebrow+eyes, Gabor feature eyes, Gabor feature nose, Gabor feature face based on the eigenface method in the principal component method;

5) facial image to known person adopts described step 1)-3) obtain the Gabor characteristic image of five kinds of parts, utilization is based on the projection properties method of vector analysis in the principal component method, extracts naked face Gabor image, eyebrow+eyes Gabor image, eyes Gabor image, the nose Gabor image of described known person face, the projection properties vector of face Gabor image respectively;

6) utilize multi-mode face identification method, extract naked face, eyes+eyebrow, eyes, the nose of known facial image, the projection properties value of five kinds of face component gray level images of face based on the parts principal component analysis;

7) the gray level image projection properties value of the same parts of known facial image and corresponding Gabor image projection eigenwert are merged according to the weighted sum rule, obtain naked face, eyes+eyebrow, eyes, the nose of known facial image, the mixing projection properties value of five kinds of face components of face;

8) people's to be identified facial image is adopted step 1)-3) obtain the Gabor characteristic image of five kinds of parts, utilization is based on the projection properties method of vector analysis in the principal component method, extracts naked face Gabor image, eyebrow+eyes Gabor image, eyes Gabor image, the nose Gabor image of described people's face to be identified, the projection properties vector of face Gabor image respectively;

9) utilize multi-mode face identification method, extract naked face, eyes+eyebrow, eyes, the nose of facial image to be identified, the projection properties value of five kinds of face component gray level images of face based on the parts principal component analysis;

10) the gray level image projection properties value of the same parts of facial image to be identified and corresponding Gabor image projection eigenwert are merged according to the weighted sum rule, obtain naked face, eyes+eyebrow, eyes, the nose of facial image to be identified, the mixing projection properties value of five kinds of face components of face;

11) calculate Euclidean distance between the mixing projection properties value of the mixing projection properties value of each parts of people's face to be identified and each parts of known person face respectively as the similarity R of the image of component of people's face to be identified and known person face, each image of component similarity is respectively naked face image R1, eyes+eyebrow image R2, eye image R3, nose image R4, face image R5;

12) similarity R1, R2, R3, R4, R5 are merged according to the weighted sum rule, obtain the people's face to be identified and the comprehensive similarity R0 of known person face, this R0 is as the human face similarity degree of recognition of face;

13) comparison step 12) similarity R0 that obtains and the size of pre-set threshold T, T chooses the similarity value that wrong acceptance rate is 0.1% o'clock correspondence, if R0 〉=T judges that then people to be identified and known person are same individuals; If R0＜T judges that then people to be identified and known person are not same individuals.