CN106980833A - Method based on multiple linear regression associative memory recognition of face - Google Patents

Abstract

The invention discloses a kind of method based on multiple linear regression associative memory recognition of face, comprise the following steps：S1：Face picture is gathered, face picture is converted into two-value picture, and obtains associative memory input matrix and output matrix；S2：Multiple linear regression human face recognition model is built, and the multiple linear regression human face recognition model is simplified, the multiple linear regression with unknown regression parameter is obtained and simplifies human face recognition model；S3：The unknown regression parameter that multiple linear regression simplifies human face recognition model is calculated, it is final to determine multiple linear regression human face recognition model；S4：Face picture is identified.Beneficial effect：Associative memory and multiple linear regression model are combined, face picture is changed into parameter, safety coefficient is high, and good reliability, recognition effect is good, good to face picture protecting effect, and crypticity is high.

Description

Method based on multiple linear regression associative memory recognition of face

Technical field

It is specifically a kind of to be based on multiple linear regression associative memory the present invention relates to view data Techniques of preserving field The method of recognition of face.

Background technology

With the development in big data epoch, people generally preserve daily photo even identity document photo to database In, personal information is easily so stolen by hacker, to be peddled or carried out illegal activity, the private letter of such people Breath is compromised, and daily life is easily disturbed or even is involved among crime dramas, easily causes inconvenience.

In order to be hidden preservation to face information, people are preserved frequently with face characteristic matrix, but with people Use from the point of view of, such protected mode is also insecure, and the face picture of preservation is easily cracked, and information will also result in It is bad to propagate, the life of people is much bothered.

The content of the invention

In view of the above-mentioned problems, the invention provides a kind of method based on multiple linear regression associative memory recognition of face, Combining association is remembered and multiple linear regression model, is converted into series of parameters to face picture and preserves, safe, can It is good by property.

To reach above-mentioned purpose, the concrete technical scheme that the present invention is used is as follows：

A kind of method based on multiple linear regression associative memory recognition of face, its key is to comprise the following steps：

S1：Face picture is gathered, by setting binary map luminance threshold, face picture two-value picture is processed as, and obtain To associative memory input matrix and output matrix；

S2：Using cell neural network structure, multiple linear regression human face recognition model is built, and to the multiple linear Return human face recognition model to be simplified, obtain the multiple linear regression with unknown regression parameter and simplify human face recognition model；

S3：The multiple linear regression that input matrix, output matrix and the step S2 obtained according to step S1 is obtained simplifies people The unknown regression parameter in face identification model, calculation procedure S2, it is final to determine multiple linear regression human face recognition model；

S4：Based on associative memory criterion, face picture is identified.

Further, the binary map luminance threshold K=(0,1,2,3...255)；

Include m width face pictures in step sl, binary map described in face picture described in each width and each width includes N The pixel composition of row M row, wherein pixel number n=N × M；

If face picture matrix data is input matrix Γ=(X of associative memory¹,X²,…,X^m), wherein,The one-dimensional vector of pixel composition all in a width face picture is represented,Table Show the value of j-th of pixel in the i-th width picture；

If two-value picture matrix data is the output matrix of associative memoryWherein,The one-dimensional vector of pixel composition all in a width two-value picture is represented,Represent The value of j-th of pixel in i-th width picture.

Further describe, step S2 particular content is：

Multiple linear regression human face recognition model is built based on cell neural network structure, is specially：

Wherein, i=1,2 ..., N；J=1,2 ..., M；k₁(i, r)=max { 1-i ,-r }；k₂(i, r)=min { N-i, r }； l₁(j, r)=max { 1-j ,-r }；l₂(i, r)=min { M-i, r }；R is to represent template radius,For input parameter, y_ijIt is defeated Go out,It is input,It is offset,It is input template；

Input templateExpression formula it is as follows：

It is 1 to make r, then in equation (2)

Formula (1) is rewritten as：

Y=AX+V (3)

Wherein, input vector：

Output vector

Offset

Dot-blur pattern A and offset V is the unknown regression parameter；

Wherein, dot-blur pattern A=(a_ij)_n×nForm can be written as：

Wherein,

Further describe, unknown regression parameter in step S3 in calculation procedure S2 is concretely comprised the following steps：

S31：Order vectori∈{1,2,…,n},

Make Y'=((Y¹)^T,(Y²)^T,…,(Yi)^T,…,(Y^m)^T)^T；Wherein YⁱRepresent picture all in a width two-value picture One dimensional vector of vegetarian refreshments composition；

L ∈ { 1,2 ..., m } and q ∈ { 1,2 ..., N } are made,

It can then be obtained by formula (3)：

X ' L=Y'(6)

Then step S2 is obtained

Wherein, L is constant.

S32：In autoassociative memories criterion, exist：

By the associative memory input matrix Γ=(X obtained in step S1¹,X²,…,X^m) and output matrixBring into formula (8), and Γ is converted into X ', willY' is converted into, X ' L are obtained₁= Y', so as to draw L=L₁=pinv (X ') Y'；

S33：Obtained according to formula (7)：

Wherein, i ∈ { 1,2 ..., n },

Try to achieve v_iAnd input template

S34：The input template that step S33 is obtainedAnd v_iBring formula (4) and (5) into, obtain dot-blur pattern A and skew V is measured, multiple linear regression human face recognition model is finally drawn.

Further describe, face picture is identified in step S4 concretely comprises the following steps：

S41：The face picture for needing to recognize is obtained, face picture will be changed and be divided into the pixel of N rows M row, and obtained The input matrix of the face picture；

S42：The input matrix is input in multiple linear regression human face recognition model and obtains corresponding model output square Battle array；

S43：The input matrix that step S41 is obtained is matched with the model output matrix that step S42 is obtained；

S44：If the match is successful that rate is then regarded as matching into more than match settings value H, wherein H=(0~1) by step S43 Work(, otherwise it fails to match.

Beneficial effects of the present invention：Associative memory and multiple linear regression model are combined, face picture is changed into Series of parameters is preserved, and crypticity is high, and safety coefficient is high, and good reliability effectively prevents that people's personal information is compromised；Adopt With by picture through model conversation into the form of parameter, simple and convenient, practicality is good, and picture recognition effect is good, to face picture protect Protect effect good.

Brief description of the drawings

Fig. 1 is the face identification method flow chart of the present invention；

Fig. 2 is multiple linear regression model location parameter resolution principle figure.

Embodiment

The embodiment and operation principle to the present invention are described in further detail below in conjunction with the accompanying drawings.

It will be seen from figure 1 that a kind of method based on multiple linear regression associative memory recognition of face, including following step Suddenly：

S1：Face picture is gathered, by setting binary map luminance threshold, face picture two-value picture is processed as, and obtain To associative memory input matrix and output matrix；

The binary map luminance threshold K=(0,1,2,3...255)；In the present embodiment, K=100 is set.

Include m width face pictures in step sl, binary map described in face picture described in each width and each width includes N The pixel of row M row is constituted, then number n=N × M of pixel；

If face picture matrix data is input matrix Γ=(X of associative memory¹,X²,…,X^m), wherein,The one-dimensional vector of pixel composition all in a width face picture is represented,Table Show the value of j-th of pixel in the i-th width picture；

If two-value picture matrix data is the output matrix of associative memoryWherein,The one-dimensional vector of pixel composition all in a width two-value picture is represented,Represent The value of j-th of pixel in i-th width picture.

In the present embodiment, black pixel point (0) is mapped as -1 in two-value picture, by white pixel point in two-value picture (255) it is mapped as 1.

S2：Using cell neural network structure, multiple linear regression human face recognition model is built, and to the multiple linear Return human face recognition model to be simplified, obtain the multiple linear regression with unknown regression parameter and simplify human face recognition model；

In the present embodiment, it is the multiple linear regression human face recognition model built based on cell neural network structure, tool Body is：

Wherein, i=1,2 ..., N；J=1,2 ..., M；k₁(i, r)=max { 1-i ,-r }；k₂(i, r)=min { N-i, r }； l₁(j, r)=max { 1-j ,-r }；l₂(i, r)=min { M-i, r }；R is to represent template radius,For input parameter, y_ijIt is defeated Go out,It is input,It is offset,It is input template；

Input templateExpression formula it is as follows：

It is 1 to make r, then in equation (2)

To concretely comprising the following steps that the multiple linear regression human face recognition model is simplified：

Formula (1) is rewritten as：

Y=AX+V (3)

Wherein, input vector：

Output vector

Offset

Dot-blur pattern A and offset V is the unknown regression parameter；

Wherein, dot-blur pattern A=(a_ij)_n×nForm can be written as：

Wherein,

S3：The multiple linear regression that input matrix, output matrix and the step S2 obtained according to step S1 is obtained simplifies people The unknown regression parameter in face identification model, calculation procedure S2, it is final to determine multiple linear regression human face recognition model；

S31：Order vectori∈{1,2,…,n}；

Make Y'=((Y¹)^T,(Y²)^T,…,(Yⁱ)^T,…,(Y^m)^T)^T；Wherein YⁱRepresent picture all in a width two-value picture One dimensional vector of vegetarian refreshments composition；

L ∈ { 1,2 ..., m } and q ∈ { 1,2 ..., N } are made,

It can then be obtained by formula (3)：

X ' L=Y'(6)

Then step S2 is obtained

Wherein, L is constant；

S32：In autoassociative memories criterion, exist：

By the associative memory input matrix Γ=(X obtained in step S1¹,X²,…,X^m) and output matrixBring into formula (8), and Γ is converted into X ', willY' is converted into, X ' L are obtained₁= Y', so as to draw L=L₁=pinv (X ') Y'；

S33：Obtained according to formula (7)

Try to achieve v_iAnd input template

S34：The input template that step S33 is obtainedAnd v_iBring formula (4) and (5) into, obtain dot-blur pattern A and skew V is measured, multiple linear regression human face recognition model is finally drawn.

S4：Based on associative memory criterion, the face picture for needing to recognize is identified.

S41：The face picture for needing to recognize is obtained, face picture will be changed and be divided into the pixel of N rows M row, and obtained The input matrix of the face picture；

S42：The input matrix is input in multiple linear regression human face recognition model and obtains corresponding model output square Battle array；

S43：The input matrix that step S41 is obtained is matched with the model output matrix that step S42 is obtained；

S44：If the match is successful that rate is then regarded as matching into more than match settings value H, wherein H=(0~1) by step S43 Work(, otherwise it fails to match.

In the present embodiment, H=0.98.

It should be pointed out that described above is not limitation of the present invention, the present invention is also not limited to the example above, What those skilled in the art were made in the essential scope of the present invention changes, is modified, adds or replaces, and also should Belong to protection scope of the present invention.

Claims (5)

1. a kind of method based on multiple linear regression associative memory recognition of face, it is characterised in that comprise the following steps：

S1：Face picture is gathered, by setting binary map luminance threshold, face picture two-value picture is processed as, and joined Think memory input matrix and output matrix；

S2：Using cell neural network structure, multiple linear regression human face recognition model is built, and to the multiple linear regression Human face recognition model is simplified, and is obtained the multiple linear regression with unknown regression parameter and is simplified human face recognition model；

S3：The multiple linear regression letter that associative memory input matrix, output matrix and the step S2 obtained according to step S1 is obtained Change the unknown regression parameter in human face recognition model, calculation procedure S2, it is final to determine multiple linear regression recognition of face mould Type；

S4：Based on associative memory criterion, face picture is identified.

2. the method according to claim 1 based on multiple linear regression associative memory recognition of face, it is characterised in that：Institute State binary map luminance threshold K=(0,1,2,3...255)；

Include m width face pictures in step sl, binary map described in face picture described in each width and each width includes N rows M The pixel composition of row, wherein pixel number n=N × M；

If face picture matrix data is input matrix Γ=(X of associative memory¹,X²,…,X^m), wherein,The one-dimensional vector of pixel composition all in a width face picture is represented,Table Show the value of j-th of pixel in the i-th width picture；

If two-value picture matrix data is the output matrix of associative memoryWherein,The one-dimensional vector of pixel composition all in a width two-value picture is represented,Represent The value of j-th of pixel in i-th width picture.

3. the method according to claim 2 based on multiple linear regression associative memory recognition of face, it is characterised in that step Suddenly S2 particular content is：

Multiple linear regression human face recognition model is built based on cell neural network structure, is specially：

$y_{ij}=\underset{k=k_1(i,r)}{\overset{k_2(i,r)}{\Σ}}\underset{l=l_1(j,r)}{\overset{l_2(j,r)}{\Σ}}{\stackrel{\‾}{a}}_{kl}{\stackrel{\‾}{x}}_{(k+i)(l+j)}+{\stackrel{\‾}{v}}_{ij}---\left(1\right)$

Wherein, i=1,2 ..., N；J=1,2 ..., M；k₁(i, r)=max { 1-i ,-r }；k₂(i, r)=min { N-i, r }；l₁ (j, r)=max { 1-j ,-r }；l₂(i, r)=min { M-i, r }；R is to represent template radius,For input parameter, y_ijIt is output,It is input,It is offset,It is input template；

Input templateExpression formula it is as follows：

It is 1 to make r, then in equation (2)

$\stackrel{\‾}{a}={\left[\begin{array}{ccc}{\stackrel{\‾}{a}}_{-1,-1}& {\stackrel{\‾}{a}}_{-1,0}& {\stackrel{\‾}{a}}_{-1,1}\\ {\stackrel{\‾}{a}}_{0,-1}& {\stackrel{\‾}{a}}_{0,0}& {\stackrel{\‾}{a}}_{0,1}\\ {\stackrel{\‾}{a}}_{1,-1}& {\stackrel{\‾}{a}}_{1,0}& {\stackrel{\‾}{a}}_{1,1}\end{array}\right]}_{3\×3}.$

Formula (1) is rewritten as：

Y=AX+V (3)

Wherein, input vector：

$X={(x_1,x_2,...,x_n)}^T={({\stackrel{\‾}{x}}_{11},{\stackrel{\‾}{x}}_{12}...{\stackrel{\‾}{x}}_{1M},{\stackrel{\‾}{x}}_{21},...,{\stackrel{\‾}{x}}_{NM})}^T$

Output vector

$Y={(y_1,y_2,...,y_n)}^T={({\stackrel{\‾}{y}}_{11},{\stackrel{\‾}{y}}_{12},...,{\stackrel{\‾}{y}}_{1M},{\stackrel{\‾}{y}}_{21},...,{\stackrel{\‾}{y}}_{NM})}^T$

Offset

$V={(v_1,v_2,...v_i...,v_n)}^T={({\stackrel{\‾}{v}}_{11},{\stackrel{\‾}{v}}_{12},...,{\stackrel{\‾}{v}}_{1M},{\stackrel{\‾}{v}}_{21}...,{\stackrel{\‾}{v}}_{NM})}^T---\left(4\right)$

Dot-blur pattern A and offset V is the unknown regression parameter；

Wherein, dot-blur pattern A=(a_ij)_n×nForm can be written as：

Wherein,

4. the method according to claim 3 based on multiple linear regression associative memory recognition of face, it is characterised in that step Unknown regression parameter in rapid S3 in calculation procedure S2 is concretely comprised the following steps：

S31：Order vector

OrderWherein YⁱRepresent pixel all in a width two-value picture One dimensional vector of composition；

L ∈ { 1,2 ..., m } and q ∈ { 1,2 ..., N } are made,

$X_q^{\′l}={\left(\begin{array}{ccc}0& {\mathrm{\α}}_{(q-1)M+1}^l& {\mathrm{\α}}_{(q-1)M+2}^l\\ {\mathrm{\α}}_{(q-1)M+1}^l& {\mathrm{\α}}_{(q-1)M+2}^l& {\mathrm{\α}}_{(q-1)M+3}^l\\ {\mathrm{\α}}_{(q-1)M+2}^l& {\mathrm{\α}}_{(q-1)M+3}^l& {\mathrm{\α}}_{(q-1)M+4}^l\\ .& .& .\\ .& .& .\\ .& .& .\\ {\mathrm{\α}}_{qM-2}^l& {\mathrm{\α}}_{qM-1}^l& {\mathrm{\α}}_{qM}^l\\ {\mathrm{\α}}_{qM-1}^l& {\mathrm{\α}}_{qM}^l& 0\end{array}\right)}_{M\×3};$

$X^{\′l}={\left(\begin{array}{ccc}0& X_1^{\′l}& X_2^{\′l}\\ X_1^{\′l}& X_2^{\′l}& X_3^{\′l}\\ X_2^{\′l}& X_3^{\′l}& X_4^{\′l}\\ .& .& .\\ .& .& .\\ .& .& .\\ X_q^{\′l}& X_{q+1}^{\′l}& X_{q+2}^{\′l}\\ .& .& .\\ .& .& .\\ .& .& .\\ X_{N-2}^{\′l}& X_{N-1}^{\′l}& X_N^{\′l}\\ X_{N-1}^{\′l}& X_N^{\′l}& 0\end{array}\right)}_{n\×9};$

$X^{\′}={\left[\begin{array}{cc}1& X^{\′1}\\ 1& X^{\′2}\\ .& .\\ .& .\\ .& .\\ 1& X^{\′l}\\ .& .\\ .& .\\ .& .\\ 1& X^{\′m}\end{array}\right]}_{(n\×m)\×10}.$

It can then be obtained by formula (3)：

X ' L=Y'(6)

Then step S2 is obtained

$L=pinv\left(X^{\′}\right)\·Y^{\′}={(v_i,{\stackrel{\‾}{a}}_{-1,-1},{\stackrel{\‾}{a}}_{-1,0},{\stackrel{\‾}{a}}_{-1,1},{\stackrel{\‾}{a}}_{0,-1},{\stackrel{\‾}{a}}_{0,0},{\stackrel{\‾}{a}}_{0,1},{\stackrel{\‾}{a}}_{1,-1},{\stackrel{\‾}{a}}_{1,0},{\stackrel{\‾}{a}}_{1,1})}_{10\×1}^T---\left(7\right)$

Wherein, L is constant；

S32：In autoassociative memories criterion, exist：

$\mathrm{\Γ}=\hat{Y}---\left(8\right)$

By the associative memory input matrix Γ=(X obtained in step S1¹,X²,…,X^m) and output matrixBring into formula (8), and Γ is converted into X ', willY' is converted into, X ' L are obtained₁= Y', so as to draw L=L₁=pinv (X ') Y'；

S33：Obtained according to formula (7)

$L={(v_i,{\stackrel{\‾}{a}}_{-1,-1},{\stackrel{\‾}{a}}_{-1,0},{\stackrel{\‾}{a}}_{-1,1},{\stackrel{\‾}{a}}_{0,-1},{\stackrel{\‾}{a}}_{0,0},{\stackrel{\‾}{a}}_{0,1},{\stackrel{\‾}{a}}_{1,-1},{\stackrel{\‾}{a}}_{1,0},{\stackrel{\‾}{a}}_{1,1})}_{10\×1}^T=L_1;$

Try to achieve v_iAnd input template

S34：The input template that step S33 is obtainedAnd v_iBring formula (4) and (5) into, obtain dot-blur pattern A and offset V, most Multiple linear regression human face recognition model is drawn eventually.

5. the method based on multiple linear regression associative memory recognition of face according to claim 1 or 2 or 3 or 4, it is special Levy and be that face picture is identified in step S4 concretely comprises the following steps：

S41：The face picture for needing to recognize is obtained, face picture will be changed and be divided into the pixel of N rows M row, and obtain the people The input matrix of face picture；

S42：The input matrix is input in multiple linear regression human face recognition model and obtains corresponding model output matrix；

S43：The input matrix that step S41 is obtained is matched with the model output matrix that step S42 is obtained；

S44：If the match is successful that rate then regards as more than match settings value H, wherein H=(0~1) that the match is successful by step S43, no Then it fails to match.