CN100452082C - Signature identifying method - Google Patents

Abstract

A method for identifying signature includes obtaining envelope curves of original signature and verification signature, separately picking up envelope curve curvity of original and verification signatures as identifications of signature, carrying out regularization-corresponding of envelope curve curvity on original signature to envelope curve curvity on verification signature and utilizing weighted Eustachian distance grader to identify envelope curve curvities of original and verification signatures.

Description

A kind of signature identifying method

Technical field

The invention belongs to a kind of method of personal identification identification, specifically, is that a kind of signature identifying method is verified in individual's notes.

Background technology

At present, in the personal identification recognition technology, there are the signature notes that utilize the individual to discern, but this identification can only rest on the basis of eye-observation, be that the litigant writes original signature in advance as checking authority, when needs signature identification litigant identity, write out certifying signature, whether whether other people rely on the personal experience to observe the comparison original signature consistent with certifying signature, true to judge litigant's identity.

Its shortcoming is: do not have the Authentication devices of specialty and verification method that original signature and certifying signature are carried out objective identification relatively, only observe differentiation with the personal experience, be subjected to the interference of subjective factor easily, verification the verifying results is relatively poor.

Summary of the invention

The purpose of this invention is to provide a kind of signature identifying method, original signature and certifying signature are carried out objective identification relatively, improve verification the verifying results.

For achieving the above object, a kind of signature identifying method of the present invention, its key is:

Include following steps:

Step 1, obtain the envelope curve of original signature and certifying signature respectively:

Read original signature and certifying signature pattern respectively with scanner, scanner sends to computing machine with pattern-information, computing machine adopts the stroke overstriking of expansion operator to the signature writing, the stroke of signature writing is linked to each other, become as a whole, utilize the Contour tracing operator to obtain the envelope curve of original signature and certifying signature respectively then;

Step 2, extract original signature and certifying signature respectively envelope curve curvature as the recognition feature k (t) of signature ₁And k (t) ₂:

The discrete curvature computing method of envelope curve are the computing formula of curvature of curve:

$\mathrm{\κ}\left(t\right)=\frac{x^{\′}\left(t\right)y^{\′\′}\left(t\right)-y^{\′}\left(t\right)x^{\′\′}\left(t\right)}{\sqrt{{(x^{\′}{\left(t\right)}^2+y^{\′}{\left(t\right)}^2)}^3}}$

Wherein, x, y are the coordinate figure of any point that obtains on the envelope curve, x ' (t), y ' (t), x " (t), y " is respectively single order, the second derivative of envelope coordinate x (t) and y (t) (t), recognition feature k (t) is the discrete curvature value of envelope curve,

Wherein said original signature envelope curve curvature k (t) ₁As criterion of identification, be kept in the computer system described certifying signature envelope curve curvature k (t) ₂Foundation as signature identification afterwards supplies the computer system examination;

Step 3, to original signature envelope curve curvature k (t) ₁With certifying signature envelope curve curvature k (t) ₂Carry out the regularization correspondence:

Utilize the polygon matching principle to seek original signature envelope curve curvature k (t) ₁With certifying signature envelope curve curvature k (t) ₂Feature mode corresponding calculated starting point, to the writing feature of twice signature carry out angle rotation, amplify or dwindle, position translation carries out normalization, make twice signature carry out the regularization correspondence, make its position, starting point and angle unified, make it reach the unchangeability of rotation, yardstick, position, be convenient to comparison, the computing formula of regularization is:

$\mathrm{xnew}\left(t\right)=\frac{1}{\mathrm{end}-\mathrm{start}+1}\underset{i=\mathrm{start}}{\overset{\mathrm{end}}{\mathrm{\Σ}}}x\left(i\right)$

$\mathrm{ynew}\left(t\right)=\frac{1}{\mathrm{end}-\mathrm{start}+1}\underset{i=\mathrm{start}}{\overset{\mathrm{end}}{\mathrm{\Σ}}}y\left(i\right)$ (t∈[0，T]).

Wherein T is the length of envelope curve after the regularization, and start is the maximum integer that is less than or equal to t*N/T, and end is less than or equal to the maximum integer of (t+1) * N/T; Xnew that is drawn (t) and ynew (t) determine described original signature envelope curve curvature k (t) ₁With certifying signature envelope curve curvature k (t) ₂The span of following;

Start is the starting point of envelope curve, and end is the terminal point of envelope curve.

Step 4, identification signature

Adopt the weighted euclidean distance sorter to carry out described original signature envelope curve curvature k (t) ₁With certifying signature envelope curve curvature k (t) ₂Identification, the pattern character vector of the two is compared,

$\mathrm{Wed}(F_t,\mathrm{\μ},\widehat{\mathrm{\Σ}})=\sqrt{{(F_t-\mathrm{\μ})}^t{\widehat{\mathrm{\Σ}}}^{-1}(F_t-\mathrm{\μ})}$ $\widehat{\mathrm{\Σ}}=\mathrm{diag}\left(\mathrm{\Σ}\right)$

Wherein ∑ is the covariance matrix of training sample signature packets trace curve proper vector, F _tBe exactly certifying signature envelope curve curvature k (t) to be identified ₂The proper vector of forming, μ is original signature envelope curve curvature k (t) ₁Proper vector (F _t) mean vector.When the certifying signature envelope curve is compared resulting Euclidean distance with the original signature envelope curve

Less than pre-set threshold, during just less than the permissible error scope, signature then to be verified is true signature, otherwise thinks that signature to be verified is for forging a signature.

This signature identifying method is the technology of a kind of off-line signature identification, be separated from each other between the stroke of individual's signature, the lines, the present invention is by extracting signature, the lines of separation are closed up, form envelope curve, because the discrete packets trace curve can be thought a polygon, therefore according to the principle of polygon coupling envelope curve is handled, make that the curvature of envelope curve is the invariant of a rotation, size, translation, thereby use the weighted euclidean distance sorter that the curvature feature of envelope curve is carried out the identity that the signer is judged in pattern classification identification at last.

People are owing to through repeatedly practice, formed relatively-stationary style at signature, all are being difficult to forge aspect the smoothness of the connectedness of the profile of signature and stroke, and this method utilizes this characteristic of signature to reach authentication purposes just.

The each signature of litigant all has the subtle change of rotation, size, translation, is discerned at the extraction signature character, need carry out pre-service, and pretreated result directly influences the precision of identification; And the present invention reaches the unchangeability of rotation, size, translation in feature extraction phases by feasible signature of the principle of polygon coupling, reaches high recognition.

Before execution in step two,, then before the computing formula of carrying out curvature of curve, need utilize Gauss's smooth kernel if uncontinuity appears in the discrete curve derivative

$g(t,\mathrm{\σ})=\frac{1}{\mathrm{\σ}\sqrt{2\mathrm{\π}}}{e}^{\frac{-t^2}{2{\mathrm{\σ}}^2}}$

Come smooth curve, make it can obtain stable original signature envelope curve curvature k (t) ₁With certifying signature envelope curve curvature k (t) ₂Wherein σ represents variance, that is:

$X^{\′}(t,\mathrm{\σ})=\frac{\∂X(t,\mathrm{\σ})}{\∂t}=x^{\′}\left(t\right)*g(t,\mathrm{\σ})=x\left(t\right)*g^{\′}(t,\mathrm{\σ})=\underset{-\∞}{\overset{+\∞}{\∫}}x\left(s\right)\frac{\∂g(t-s,\mathrm{\σ})}{\∂s}\mathrm{ds}$

With X ' (t, σ) x ' in the computing formula of replacement curvature of curve (t), in like manner to y (t), x " (t), y " (t) does similar operation, so just x (t) is transferred to g (t to the derivative of t, σ) to coming on the derivative of t, (t σ) can stability Calculation to the t derivative, draws calculated curve curvature stably because g.

Described computing machine adopt read image in the figure instrument performing step one read and read expansion operator in the figure instrument to the stroke overstriking of signature writing, and obtain envelope curve; Described computing machine adopts the recognition feature k (t) of signature in the operational tool performing step two ₁And k (t) ₂Obtain; Operational tool is regularization calculating in the performing step three and the signature of the identification in the step 4 also.

Remarkable result of the present invention is: original signature and certifying signature are carried out objective identification relatively, improve verification the verifying results, unified the identification comparative standard, to the identification more convenient and quicker of personal identification.

Description of drawings

Fig. 1 is a workflow diagram of the present invention;

Fig. 2 is the envelope curve that obtains original input signature image, pretreated signature image and this signature through step 1, two, wherein:

Fig. 2. (1) is two different original input signature images;

Fig. 2. (2) are to two signature images pretreated image of operator that expands;

Fig. 2. (3) are that two signature images are obtained different envelope curves;

Fig. 3 is the envelope curve of four similar signature images, four the anglec of rotation, varies in size, and initial point start choice of location difference, the displacement of formation is also different;

Fig. 4 is to the curvature characteristic curve diagram of four envelope curves among Fig. 3; Obtain four different curvature characteristic curvees, as Fig. 4 .1, as Fig. 4 .2, as Fig. 4 .3, shown in Fig. 4 .4: owing to do not pass through the step 3 regularization, can see that its initial point start is inconsistent, relatively signature is very difficult;

The synoptic diagram that Fig. 5 step 3 is carried out regularization to envelope curve among Fig. 3, wherein:

Fig. 5 .1 is signature packets trace curve figure;

Fig. 5 .2 is the polygon match map;

Fig. 5 .3 is envelope curve figure after the polygon coupling, can find the starting point start of its signature packets trace curve unanimity;

Fig. 6 be earlier through step 3 to the envelope curve regularization among Fig. 3 after, four curvature characteristic curvees reentrying, the corresponding starting point of signature packets trace curve after polygon match map and the polygon coupling, can see four curvature characteristic curvees initial point start unanimity just, help relatively signing;

Fig. 7 is the structure principle chart of weighted euclidean distance sorter, is prior art;

Fig. 8 system chart of the present invention.

Embodiment

Below in conjunction with the drawings and specific embodiments the present invention is described in further detail.

As Fig. 1, shown in Figure 8: a kind of signature identifying method includes following steps:

Step 1, obtain the envelope curve of original signature and certifying signature respectively:

Read original signature and certifying signature pattern respectively with scanner, shown in Fig. 2 .1, original input signature image, scanner sends to computing machine with pattern-information, computing machine adopts the stroke overstriking of expansion operator to the signature writing, the stroke of signature writing is linked to each other, shown in Fig. 2 .2, become as a whole, utilize the Contour tracing operator to obtain the original signature shown in Fig. 2 .3 and the envelope curve of certifying signature respectively then;

Step 2, extract original signature and certifying signature respectively envelope curve curvature as the recognition feature k (t) of signature ₁And k (t) ₂:

Before execution in step two, differentiate the discrete curve derivative, if uncontinuity appears in the discrete curve derivative, then utilize Gauss's smooth kernel earlier

$g(t,\mathrm{\σ})=\frac{1}{\mathrm{\σ}\sqrt{2\mathrm{\π}}}{e}^{\frac{-t^2}{2{\mathrm{\σ}}^2}}$

Come smooth curve, make it can obtain stable original signature envelope curve curvature k (t) ₁With certifying signature envelope curve curvature k (t) ₂Wherein σ represents variance, that is:

$X^{\′}(t,\mathrm{\σ})=\frac{\∂X(t,\mathrm{\σ})}{\∂t}=x^{\′}\left(t\right)*g(t,\mathrm{\σ})=x\left(t\right)*g^{\′}(t,\mathrm{\σ})=\underset{-\∞}{\overset{+\∞}{\∫}}x\left(s\right)\frac{\∂g(t-s,\mathrm{\σ})}{\∂s}\mathrm{ds}$

With X ' (t, σ) x ' in the computing formula of replacement curvature of curve (t), in like manner to y ' (t), " (t), y " (t) does similar operation to x, so just x (t) is transferred to g (t to the derivative of t, σ) to coming on the derivative of t, (t σ) can stability Calculation to the t derivative because g, after drawing stably calculated curve curvature, carry out the calculating of curvature of curve again.

Obtain as shown in Figure 3 original signature image and certifying signature image;

As shown in Figure 4,, then directly carry out the calculating of curvature of curve, obtain the visual curvature characteristic curve shown in Fig. 4 .1,4.2,4.3,4.4 respectively if the discrete curve derivative itself possesses continuity;

The discrete curvature computing method of envelope curve are the computing formula of curvature of curve:

$\mathrm{\κ}\left(t\right)=\frac{x^{\′}\left(t\right)y^{\′\′}\left(t\right)-y^{\′}\left(t\right)x^{\′\′}\left(t\right)}{\sqrt{{(x^{\′}{\left(t\right)}^2+y^{\′}{\left(t\right)}^2)}^3}}$

Wherein, x, y are the coordinate figure of any point that obtains on the envelope curve, x ' (t), y ' (t), x " (t), y " is respectively single order, the second derivative of envelope coordinate x (t) and y (t) (t), recognition feature k (t) is the discrete curvature value of envelope curve,

Wherein said original signature envelope curve curvature k (t) ₁As criterion of identification, be kept in the computer system described certifying signature envelope curve curvature k (t) ₂Foundation as signature identification afterwards supplies the computer system examination;

As shown in Figure 5, step 3, to original signature envelope curve curvature k (t) ₁With certifying signature envelope curve curvature k (t) ₂Carry out the regularization correspondence:

Signature packets trace curve figure shown in Fig. 5 .1 is carried out the polygon coupling, obtain the polygon match map shown in Fig. 5 .2; After the polygon coupling, obtain the starting point start of the envelope curve unanimity shown in Fig. 5 .3;

Utilize the polygon matching principle to seek original signature envelope curve curvature k (t) ₁With certifying signature envelope curve curvature k (t) ₂Feature mode corresponding calculated starting point, to the writing feature of twice signature carry out angle rotation, amplify or dwindle, position translation carries out normalization, make twice signature carry out the regularization correspondence, make its position, starting point and angle unified, make it reach the unchangeability of rotation, yardstick, position, be convenient to comparison, the computing formula of regularization is:

$\mathrm{xnew}\left(t\right)=\frac{1}{\mathrm{end}-\mathrm{start}+1}\underset{i=\mathrm{start}}{\overset{\mathrm{end}}{\mathrm{\Σ}}}x\left(i\right)$

$\mathrm{ynew}\left(t\right)=\frac{1}{\mathrm{end}-\mathrm{start}+1}\underset{i=\mathrm{start}}{\overset{\mathrm{end}}{\mathrm{\Σ}}}y\left(i\right)$ (t∈[0，T]).

Wherein T is the length of envelope curve after the regularization, and start is the maximum integer that is less than or equal to t*N/T, and end is less than or equal to the maximum integer of (t+1) * N/T; Xnew that is drawn (t) and ynew (t) determine described original signature envelope curve curvature k (t) ₁With certifying signature envelope curve curvature k (t) ₂The span of following.

Step 4, identification signature

The most finally make as shown in Figure 3 different sizes and the signature of angle respectively by unified be as shown in Figure 6 regularization curve, the proper vector of being convenient to curve is compared

Adopt the weighted euclidean distance sorter to carry out described original signature envelope curve curvature k (t) ₁With certifying signature envelope curve curvature k (t) ₂Identification, the pattern character vector of the two is compared,

$\mathrm{Wed}(F_t,\mathrm{\μ},\widehat{\mathrm{\Σ}})=\sqrt{{(F_t-\mathrm{\μ})}^t{\widehat{\mathrm{\Σ}}}^{-1}(F_t-\mathrm{\μ})}$ $\widehat{\mathrm{\Σ}}=\mathrm{diag}\left(\mathrm{\Σ}\right)$

Wherein ∑ is the covariance matrix of training sample signature packets trace curve proper vector, F _tBe exactly certifying signature envelope curve curvature k (t) to be identified ₂The proper vector of forming, μ is original signature envelope curve curvature k (t) ₁Proper vector (F _t) mean vector.When the certifying signature envelope curve is compared resulting Euclidean distance with the original signature envelope curve

Less than pre-set threshold, during just less than the permissible error scope, signature then to be verified is true signature, otherwise thinks that signature to be verified is for forging a signature.

Can repeatedly sign training relatively by the litigant, draw the best identified threshold value.

As shown in Figure 7, the weighted euclidean distance sorter is a prior art.

Described computing machine adopt read image in the figure instrument performing step one read and read expansion operator in the figure instrument to the stroke overstriking of signature writing, and obtain envelope curve; Described computing machine adopts the recognition feature k (t) of signature in the operational tool performing step two ₁And k (t) ₂Obtain; Operational tool is regularization calculating in the performing step three and the signature of the identification in the step 4 also.

The described figure of readding instrument is photoshop or flash, and described operational tool is Java or Vc or VB.

This signature identifying method is the technology of a kind of off-line signature identification, be separated from each other between the stroke of individual's signature, the lines, the present invention is by extracting signature, the lines of separation are closed up, form envelope curve, because the discrete packets trace curve can be thought a polygon, therefore according to the principle of polygon coupling envelope curve is handled, make that the curvature of envelope curve is the invariant of a rotation, size, translation, thereby use the weighted euclidean distance sorter that the curvature feature of envelope curve is carried out the identity that the signer is judged in pattern classification identification at last.

People are owing to through repeatedly practice, formed relatively-stationary style at signature, all are being difficult to forge aspect the smoothness of the connectedness of the profile of signature and stroke, and this method utilizes this characteristic of signature to reach authentication purposes just.

The each signature of litigant all has the subtle change of rotation, size, translation, is discerned at the extraction signature character, need carry out pre-service, and pretreated result directly influences the precision of identification; And the present invention reaches the unchangeability of rotation, size, translation in feature extraction phases by feasible signature of the principle of polygon coupling, reaches high recognition.

Claims (3)

1, a kind of signature identifying method is characterized in that:

Include following steps:

Step 1, obtain the envelope curve of original signature and certifying signature respectively:

Read original signature and certifying signature pattern respectively with scanner, scanner sends to computing machine with pattern-information, computing machine adopts the stroke overstriking of expansion operator to the signature writing, the stroke of signature writing is linked to each other, become as a whole, utilize the Contour tracing operator to obtain the envelope curve of original signature and certifying signature respectively then;

Step 2, extract original signature and certifying signature respectively envelope curve curvature as the recognition feature k (t) of signature ₁And k (t) ₂:

The discrete curvature computing method of envelope curve are the computing formula of curvature of curve:

$\mathrm{\κ}\left(t\right)=\frac{x^{\′}\left(t\right)y^{\′\′}\left(t\right)-y^{\′}\left(t\right)x^{\′\′}\left(t\right)}{\sqrt{{(x^{\′}{\left(t\right)}^2+y^{\′}{\left(t\right)}^2)}^3}}$

Wherein, x, y are the coordinate figure of any point that obtains on the envelope curve, x ' (t), y ' (t), x " (t), y " is respectively single order, the second derivative of envelope coordinate x (t) and y (t) (t), recognition feature k (t) is the discrete curvature value of envelope curve,

Wherein said original signature envelope curve curvature k (t) ₁As criterion of identification, be kept in the computer system described certifying signature envelope curve curvature k (t) ₂Foundation as signature identification afterwards supplies the computer system examination;

Step 3, to original signature envelope curve curvature k (t) ₁With certifying signature envelope curve curvature k (t) ₂Carry out the regularization correspondence:

Utilize the polygon matching principle to seek original signature envelope curve curvature k (t) ₁With certifying signature envelope curve curvature k (t) ₂Feature mode corresponding calculated starting point, to the writing feature of twice signature carry out angle rotation, amplify or dwindle, position translation carries out normalization, make twice signature carry out the regularization correspondence, make its position, starting point and angle unified, make it reach the unchangeability of rotation, yardstick, position, be convenient to comparison, the computing formula of regularization is:

$\mathrm{xnew}\left(t\right)=\frac{1}{\mathrm{end}-\mathrm{start}+1}\underset{i=\mathrm{start}}{\overset{\mathrm{end}}{\mathrm{\Σ}}}x\left(i\right)$

$\mathrm{ynew}\left(t\right)=\frac{1}{\mathrm{end}-\mathrm{start}+1}\underset{i=\mathrm{start}}{\overset{\mathrm{end}}{\mathrm{\Σ}}}y\left(i\right)$ (t∈[0，T]).

Wherein T is the length of envelope curve after the regularization, and start is the maximum integer that is less than or equal to t*N/T, and end is less than or equal to the maximum integer of (t+1) * N/T; Xnew that is drawn (t) and ynew (t) determine described original signature envelope curve curvature k (t) ₁With certifying signature envelope curve curvature k (t) ₂The span of following;

Step 4, identification signature

Adopt the weighted euclidean distance sorter to carry out described original signature envelope curve curvature k (t) ₁With certifying signature envelope curve curvature k (t) ₂Identification, the pattern character vector of the two is compared,

$\mathrm{Wed}(F_t,\mathrm{\μ},\widehat{\mathrm{\Σ}})=\sqrt{{(F_t-\mathrm{\μ})}^t{\widehat{\mathrm{\Σ}}}^{-1}(F_t-\mathrm{\μ})}$ $\widehat{\mathrm{\Σ}}=\mathrm{diag}\left(\mathrm{\Σ}\right)$

Wherein ∑ is the covariance matrix of training sample signature packets trace curve proper vector, F _tBe exactly certifying signature envelope curve curvature k (t) to be identified ₂The proper vector of forming, μ is original signature envelope curve curvature k (t) ₁Proper vector (F _t) mean vector.When the certifying signature envelope curve is compared resulting Euclidean distance with the original signature envelope curve

Less than pre-set threshold, during just less than the permissible error scope, signature then to be verified is true signature, otherwise thinks that signature to be verified is for forging a signature.

2, a kind of signature identifying method according to claim 1 is characterized in that: before execution in step two, if uncontinuity appears in the discrete curve derivative, then before the computing formula of carrying out curvature of curve, need utilize Gauss's smooth kernel

$g(t,\mathrm{\σ})=\frac{1}{\mathrm{\σ}\sqrt{2\mathrm{\π}}}{e}^{-\frac{-t^2}{{2\mathrm{\σ}}^2}}$

Come smooth curve, make it can obtain stable original signature envelope curve curvature k (t) ₁With certifying signature envelope curve curvature k (t) ₂Wherein σ represents variance, that is:

$X^{\′}(t,\mathrm{\σ})=\frac{\∂X(t,\mathrm{\σ})}{\∂t}=x^{\′}\left(t\right)*g(t,\mathrm{\σ})=x\left(t\right)*g^{\′}(t,\mathrm{\σ})=\underset{-\∞}{\overset{+\∞}{\∫}}x\left(s\right)\frac{\∂g(t-s,\mathrm{\σ})}{\∂s}\mathrm{ds}$

With X ' (t, σ) x ' in the computing formula of replacement curvature of curve (t), in like manner (t) to y ', x " (t), y " (t) does similar operation, so just x (t) is transferred to g (t to the derivative of t, σ) to coming on the derivative of t, (t σ) can stability Calculation to the t derivative, draws calculated curve curvature stably because g.

3, a kind of signature identifying method according to claim 1 is characterized in that: described computing machine adopt read image in the figure instrument performing step one read and read expansion operator in the figure instrument to the stroke overstriking of signature writing, and obtain envelope curve; Described computing machine adopts the recognition feature k (t) of signature in the operational tool performing step two ₁And k (t) ₂Obtain; Operational tool is regularization calculating in the performing step three and the signature of the identification in the step 4 also.

Images