CN101303729B - Novel method for detecting fingerprint singularity - Google Patents

Abstract

The invention discloses a novel detecting method for fingerprint singular point, which solves the problems that the existing singular point extracting method seriously depends on the direction field of the fingerprint, can not effectively treat the fingerprint images of low quality and is hard to be suitable for engineering application. The method is divided into a front stage and a back stage; the front stage carries out effective improvement on the existing Poincare index method and utilizes the Poincare index to extract the candidacy singular points in the fingerprint images. The back stageeffectively utilizes a Gaussian-Hermite moment to carry out false-removing treatment on the candidacy singular points. The method effectively combines the changing information of the dermal ridge directions of the neighboring areas surrounding the singular points and the change trend information of the dermal ridge consistency of the partial areas surrounding the singular points, can accurately and reliably extract the singular points in the fingerprint images and has the characteristics of high noise immunity, high accuracy and reliability, large engineering application value, etc.

Description

A kind of new method for detecting fingerprint singularity

Technical field

The present invention relates to a kind of fingerprint image detection method, the new method for detecting fingerprint singularity of the inferior quality fingerprint image in specifically a kind of practical Automated Fingerprint Identification System (AFIS).

Background technology

At present in practical Automated Fingerprint Identification System, the fingerprint classification technology is one of gordian technique of accelerating system identification speed, and mostly the sorting technique of main flow is that information such as number, type and position according to singular point realize now, and the texture matching algorithm that is adopted when handling the inferior quality fingerprint image also needs singular point information accurately and reliably.The singular point extracting method of the main flow that adopts in the Automated Fingerprint Identification System now, the overwhelming majority depends on the accurate extraction of direction of fingerprint field, but when handling the inferior quality fingerprint image, because calculating streakline direction itself reliably is exactly a difficult problem, thereby the singular point that these methods are extracted is not only located not accurate enough, in the place of wrong local and some noise pollutions of streakline direction calculating, also often detect the singular point of many falsenesses easily.This makes these methods be difficult to effectively satisfy engineering and uses.In the application of reality, need a kind of extraction algorithm of singular point accurately and reliably.

Summary of the invention

Purpose of the present invention seriously relies on the direction of fingerprint field in order to solve existing singular point extracting method exactly, can not effectively handle the inferior quality fingerprint image, be difficult to be suitable for problems such as engineering application, a kind of new accurately and reliably method for detecting fingerprint singularity is provided, this method is divided into former and later two stages, carried out effective improvement to existing P oincare index Poincare index method previous stage, and utilize candidate's singular point in its image that takes the fingerprint, the latter half imitates and utilizes Gaussian-Hermite Gauss-Hermite square square that candidate's singular point is gone pseudo-the processing, this method effectively combines the streakline consistance variation tendency information of regional area around the streakline direction change information of neighborhood around the singular point and the singular point, can take the fingerprint accurately and reliably singular point in the image, it is strong to have a noise immunity, the accurate reliability height, characteristics such as the engineering using value is big.

For achieving the above object, the present invention adopts following technical scheme:

A kind of extracting method of singular point accurately and reliably, its method be,

(1) preproduction phase: the background separation and the field of direction information calculations that comprise fingerprint image.

(2) phase one: the extraction of candidate's singular point comprises following step:

(a) (i, j) by formula (1) calculates its corresponding Poincare index Poincare indices P to each point among the O ' of direction territory _{G, C}(i, j).Adopted two length to be respectively the Poincare index Poincare exponential quantity that 5 * 5 and 7 * 7 closed curve calculates each point in this algorithm, as long as wherein one result of calculation meets the singular point condition, assert that then this point is candidate's singular point, if two singular point Type-Inconsistencies that curve detection goes out assert directly that then this point is pseudo-point.

(b) the detected core core point of previous step and delta trigpoint utilization clustering algorithm is carried out cluster respectively by Euclidean distance, and add up the number N of the singular point that each cluster comprises _mN _m＜N, the cluster of (N represents the number that the fingerprint pattern district should be able to detected singular point, and N is taken as 25 here) is deleted.

(c) ask for the average sudden change degree of remaining each cluster, and respectively these clusters are sorted by singular point type, sudden change degree size, choose preceding M (this method M=3, core core point point cluster and delta trigpoint point cluster are all got 3, when actual cluster number is less than 3, get actual cluster number) the more cluster of individual sudden change is as candidate's singular point cluster, and candidate's singular point finally just is positioned on the barycenter of these clusters.

(3) subordinate phase: candidate's singular point go puppet, comprise following step:

(a) to each candidate's singular point, by formula (2) calculate the distribution consistance coherence value (τ is average ridge distance) of each pixel in the circular neighborhood that radius around it is 6 τ.

(b) then this border circular areas is divided into 32 fan-shaped and center small circular zones and calculate the average coherence value of various piece, the width of the radius of border circular areas and two round belts is 2 τ.

(c) last, compare the trizonal size (central circular is 16 zones that direction is public) that is evenly distributed consistance coherence value on 16 each directions of direction, if the consistance coherence value that outwards distributes from the lining is increasing, then this direction of mark is a useful direction.For a certain candidate's singular point, more than or equal to 10, then extracting this point is singular point as if the useful direction number, otherwise, think that this point is pseudo-point.If some direction of singular point gone out the fingerprint border then with remaining direction as the reference direction, on these directions, calculate useful direction.In addition, through experimental analysis, it is comparatively suitable that the window size that calculates each pixel distribution consistance coherence value is taken as (4 τ+1) * (4 τ+1).

Phase one, in the described step (a), calculate each singular point (i, undertaken by following formula during j) Poincare index Poincare exponential quantity:

$\mathrm{Poincare}(i.j)=\frac{1}{2\mathrm{\π}}\underset{k=0}{\overset{\mathrm{N\ψ}-1}{\mathrm{\Σ}}}\mathrm{\Δ}\left(k\right),$

$\mathrm{\&Delta;}\left(k\right)=\left\{\begin{array}{cc}\mathrm{\&delta;}\left(k\right),& \mathrm{if}\left|\mathrm{\&delta;}\left(k\right)\right|<\frac{\mathrm{\&pi;}}{2}\\ \mathrm{\&pi;}+\mathrm{\&delta;}\left(k\right),& \mathrm{if\&delta;}\left(k\right)\&le;-\frac{\mathrm{\&pi;}}{2}\\ \mathrm{\&pi;}-\mathrm{\&delta;}\left(k\right),& \mathrm{otherwise}\end{array}\right.---\left(1\right)$

δ(k)＝O′(ψ _x(i′)，ψ _y(i′))-O′(ψ _x(i)，ψ _y(i))，

i′＝(i+1)mod?N _ψ，

ψ wherein _x(i) and ψ _y(i) being respectively is the x and the y coordinate of k point on the closed curve with N ψ pixel at center with the set point.The difference of δ (k) expression two adjacent deflections, Δ (k) expression is adjusted later result to difference, the next point after i point of i ' expression.

Adopted 5 * 5 and 7 * 7 square closed curve in this algorithm, the formation of this formula is used to calculate certain and puts the accumulation that deflection changes on the corresponding closed curve based on the improvement to original Poincare Index Poincare index method.

In the described step (a), the singular point condition is meant the condition that singular point that this method proposes should satisfy, and this condition is that the computation process to formula (1) retrains, and specifically is meant additional following restrictive condition when doing adding up that deflection changes along closed curve:

(1) add up the sign change number of times of this N ψ direction, if direction by just to negative and by negative to just each takes place once, and only once, its Poincare index Poincare exponential quantity of continuation calculating then, otherwise, think that this point is a general point.

(2) add up absolute value in this N ψ the difference

The number of difference, if number thinks then that more than one this point is a general point.

(3) if the value of final Poincare index Poincare index is 1/2, (i j) just is confirmed as the core core point to this set point, if Poincare index Poincare exponential quantity is-1/2 so, (i j) just is confirmed as the delta trigpoint to this set point so.

Phase one, in the described step (b), N value 25 is based on the analysis to test findings, can detected singular point number in the singular point district of fingerprint be metastable and concentrate on the penetralia in singular point district, for the inferior quality fingerprint image, as long as the singular point district is not subjected to the pollution of big noise, number that can detected singular point in the singular point district also is stable so, and this method N gets 25.

Phase one, in the described step (c), the sudden change degree is meant that adjacent two deflections are done in the difference process on the closed curve, absolute value

That difference, on average sudden change degree is meant sudden change degree average of all pixels in each cluster.

Subordinate phase, in the described step (a), this method utilization distribution consistance coherence value is expressed the consensus information of streakline, and Gaussian-Hermite Gauss-Hermite square has been adopted in the calculating of distribution consistance coherence value.This method is described the streakline consensus information of fingerprint: M with four squares _0,1, M _1,0, M _0,3, M _3,0And to these four not the Gaussian-Hermite square of same order do as giving a definition:

$\left\{\begin{array}{c}{M}_u(x,y)={\mathrm{\&lambda;M}}_{\mathrm{1,0}}(x,y,I(x,y))+(1-\mathrm{\&lambda;})M_{\mathrm{3,0}}(x,y,I(x,y))\\ M_v(x,y)={\mathrm{\&lambda;M}}_{\mathrm{0,1}}(x,y,I(x,y))+(1-\mathrm{\&lambda;})M_{0.3}(x,y,I(x,y))\end{array}\right.---\left(2\right)$

Here λ (0＜λ＜1) is the associating weight coefficient (λ value 0.5 in this method) of the Gaussian-Hermite Gauss-Hermite square of not same order.Utilize definition (2), (i j) can calculate a proper vector [M to each pixel in the fingerprint image _u, M _v] ^TAt fingerprint singularity district and non-singular point district [M _u, M _v] ^TDistribution take on a different character, at non-singular point district [M _u, M _v] ^TDistribute along direction, at singular point district [M perpendicular to crestal line _u, M _v] ^TThen be evenly distributed on all directions, this method adopts principal component method to extract [M _u, M _v] ^TDistribution character, [M _u, M _v] ^TCovariance matrix C _mDefine by following formula:

$C_M=\left[\begin{array}{cc}\underset{w}{\mathrm{\&Sigma;}}{(M_u-m_u)}^2& \underset{w}{\mathrm{\&Sigma;}}(M_u-m_u)(M_v-m_v)\\ \underset{w}{\mathrm{\&Sigma;}}(M_u-m_u)(M_v-m_v)& \underset{w}{\mathrm{\&Sigma;}}{(M_v-m_v)}^2\end{array}\right]---\left(3\right)$

Wherein,

m _uAverage M in the expression window W _uValue, m _vAverage M in the expression window W _vValue, n * n is the size of window W, and this method n value is 4 τ+1, and τ is average ridge distance.

If λ 1 and λ 2 are covariance matrix C _MTwo eigenwerts, then when λ 1＞＞during λ 2, [M _u, M _v] ^TDistribution mainly be to distribute along long axis direction, also promptly distribute along direction perpendicular to crestal line, and in the noise range or the value of singular point district λ 1 and λ 2 be very approaching, therefore, definition [M _u, M _v] ^TDistribution consistance coherence feature as follows:

$\mathrm{cohdrence}=\frac{{{\mathrm{\&lambda;}}_1}^2-{{\mathrm{\&lambda;}}_2}^2}{{{\mathrm{\&lambda;}}_1}^2+{{\mathrm{\&lambda;}}_2}^2}---\left(4\right)$

$=\frac{(\underset{w}{\mathrm{\&Sigma;}}{(M_u-m_u)}^2+\underset{w}{\mathrm{\&Sigma;}}{(M_v-M_v)}^2)\sqrt{{(\underset{w}{\mathrm{\&Sigma;}}{(M_u-m_u)}^2-\underset{w}{\mathrm{\&Sigma;}}{(M_v-m_v)}^2)}^2+4{\left(\underset{w}{\mathrm{\&Sigma;}}(M_u-m_u)(M_v-m_v)\right)}^2}}{{\left(\underset{w}{\mathrm{\&Sigma;}}{(M_u-m_u)}^2\right)}^2+{\left(\underset{w}{\mathrm{\&Sigma;}}{(M_v-m_v)}^2\right)}^2+2{\left(\underset{w}{\mathrm{\&Sigma;}}(M_u-m_u)(M_v-m_v)\right)}^2}$

Thus, the streakline consistance is good more, and distribution consistance coherence is big more.

Beneficial effect of the present invention: owing to adopted two stage disposal route, the reliability that singular point extracts has had raising largely, and particularly when handling the inferior quality fingerprint image, the omission of singular point, flase drop phenomenon have had minimizing largely.In addition, original Poincare index Poincare index method is improved, improve its anti-noise ability, avoided repeatedly smoothing processing, the singular point accuracy of final extraction is effectively improved the field of direction.Satisfied the application demand of practical Automated Fingerprint Identification System (AFIS).

Description of drawings

Fig. 1 is for removing pseudo-template image;

Fig. 2 is an algorithm flow chart.

Embodiment

The present invention will be further described below in conjunction with accompanying drawing and embodiment.

A kind of extracting method of singular point accurately and reliably, its method be,

(1) preproduction phase: the background separation and the field of direction information calculations that comprise fingerprint image.

(2) phase one: the extraction of candidate's singular point comprises following step:

(a) (i, j) by formula (1) calculates its corresponding Poincare index Poincare indices P to each point among the O ' of direction territory _{G, C}(i, j).Adopted two length to be respectively the Poincareindex Poincare exponential quantity that 5 * 5 and 7 * 7 closed curve calculates each point in this algorithm, as long as wherein one result of calculation meets the singular point condition, assert that then this point is candidate's singular point, if two singular point Type-Inconsistencies that curve detection goes out assert directly that then this point is pseudo-point.

(b) the detected core core point of previous step and delta trigpoint utilization clustering algorithm is carried out cluster respectively by Euclidean distance, and add up the number N of the singular point that each cluster comprises _m, N _mThe cluster of＜N (N is taken as 25 here) is deleted.

(c) ask for the average sudden change degree of remaining each cluster, and respectively these clusters are sorted by singular point type, sudden change degree size, choose preceding M (this method M=3, core point cluster and delta point cluster are all got 3, when actual cluster number is less than 3, get actual cluster number) the more cluster of individual sudden change is as candidate's singular point cluster, and candidate's singular point finally just is positioned on the barycenter of these clusters.

(3) subordinate phase: candidate's singular point go puppet, as shown in Figure 2, comprise following step:

(a) to each candidate's singular point, by formula (2) calculate the distribution consistance coherence value (τ is average ridge distance) of each pixel in the circular neighborhood that radius around it is 6 τ.

What (b) by template shown in Figure 1 this border circular areas is divided into 32 fan-shaped and center small circular zones then and calculates various piece is evenly distributed consistance coherence value, and the width of the radius of border circular areas and two round belts is 2 τ.

(c) last, compare the trizonal size (central circular is 16 zones that direction is public) that is evenly distributed consistance coherence value on 16 each directions of direction, if the consistance coherence value that outwards distributes from the lining is increasing, then this direction of mark is a useful direction.For a certain candidate's singular point, more than or equal to 10, then extracting this point is singular point as if the useful direction number, otherwise, think that this point is pseudo-point.If some direction of singular point gone out the fingerprint border then with remaining direction as the reference direction, on these directions, calculate useful direction.In addition, through experimental analysis, it is comparatively suitable that the window size that calculates each pixel distribution consistance coherence value is taken as (4 τ+1) * (4 τ+1).Concrete algorithm flow chart as shown in Figure 2.

Phase one, in the described step (a), calculate each point (i, undertaken by following formula during j) Poincare index value:

$\mathrm{Poincare}(i.j)=\frac{1}{2\mathrm{\&pi;}}\underset{k=0}{\overset{\mathrm{N\&psi;}-1}{\mathrm{\&Sigma;}}}\mathrm{\&Delta;}\left(k\right),$

$\mathrm{\&Delta;}\left(k\right)=\left\{\begin{array}{cc}\mathrm{\&delta;}\left(k\right),& \mathrm{if}\left|\mathrm{\&delta;}\left(k\right)\right|<\frac{\mathrm{\&pi;}}{2}\\ \mathrm{\&pi;}+\mathrm{\&delta;}\left(k\right),& \mathrm{if\&delta;}\left(k\right)\&le;-\frac{\mathrm{\&pi;}}{2}\\ \mathrm{\&pi;}-\mathrm{\&delta;}\left(k\right),& \mathrm{otherwise}\end{array}\right.---\left(1\right)$

δ(k)＝O′(ψ _x(i′)，ψ _y(i′))-O′(ψ _x(i)，ψ _y(i))，

i′＝(i+1)mod?N _ψ，

ψ wherein _x(i) and ψ _y(i) being respectively is the x and the y coordinate of k point on the closed curve with N ψ pixel at center with the set point.The difference of δ (k) expression two adjacent deflections, Δ (k) expression is adjusted later result to difference, the next point after i point of i ' expression.

Adopted 5 * 5 and 7 * 7 square closed curve in this algorithm, the formation of this formula is used to calculate certain and puts the accumulation that deflection changes on the corresponding closed curve based on the improvement to original Poincare Index Poincare index method.

In the described step (a), the singular point condition is meant the condition that singular point that this method proposes should satisfy, and this condition is that the computation process to formula (1) retrains, and specifically is meant additional following restrictive condition when doing adding up that deflection changes along closed curve:

(1) add up the sign change number of times of this N ψ direction, if direction by just to negative and by negative to just each takes place once, and only once, its Poincare index Poincare exponential quantity of continuation calculating then, otherwise, think that this point is a general point.

(2) add up absolute value in this N ψ the difference

The number of difference, if number thinks then that more than one this point is a general point.

(3) if the value of final Poincare index Poincare index is 1/2, (i j) just is confirmed as the core core point to this set point, if Poincare index Poincare exponential quantity is-1/2 so, (i j) just is confirmed as the delta trigpoint to this set point so.

Phase one, in the described step (b), N value 25 is based on the analysis to test findings, can detected singular point number in the singular point district of fingerprint be metastable and concentrate on the penetralia in singular point district, for the inferior quality fingerprint image, as long as the singular point district is not subjected to the pollution of big noise, number that can detected singular point in the singular point district also is stable so, and this method N gets 25.

Phase one, in the described step (c), the sudden change degree is meant that adjacent two deflections are done in the difference process on the closed curve, absolute value That difference, on average sudden change degree is meant sudden change degree average of all pixels in each cluster.

Subordinate phase, in the described step (a), this method utilization distribution consistance coherence value is expressed the consensus information of streakline, and the Gaussian-Hermite square has been adopted in the calculating of distribution consistance coherence value.This method is described the streakline consensus information of fingerprint: M with four squares _0,1, M _1,0, M _0,3, M _3,0And to these four not the Gaussian-Hermite Gauss-Hermite square of same order do as giving a definition:

$\left\{\begin{array}{c}{M}_u(x,y)={\mathrm{\&lambda;M}}_{\mathrm{1,0}}(x,y,I(x,y))+(1-\mathrm{\&lambda;})M_{\mathrm{3,0}}(x,y,I(x,y))\\ M_v(x,y)={\mathrm{\&lambda;M}}_{\mathrm{0,1}}(x,y,I(x,y))+(1-\mathrm{\&lambda;})M_{0.3}(x,y,I(x,y))\end{array}\right.---\left(2\right)$

Here λ (0＜λ＜1) is the associating weight coefficient (λ value 0.5 in this method) of the Gaussian-Hermite square of not same order.Utilize definition (2), (i j) can calculate a proper vector [M to each pixel in the fingerprint image _u, M _v] ^TAt fingerprint singularity district and non-singular point district [M _u, M _v] ^TDistribution take on a different character, at non-singular point district [M _u, M _v] ^TDistribute along direction, at singular point district [M perpendicular to crestal line _u, M _v] ^TThen be evenly distributed on all directions, this method adopts principal component method to extract [M _u, M _v] ^TDistribution character, [M _u, M _v] ^TCovariance matrix C _MDefine by following formula:

$C_M=\left[\begin{array}{cc}\underset{w}{\mathrm{\&Sigma;}}{(M_u-m_u)}^2& \underset{w}{\mathrm{\&Sigma;}}(M_u-m_u)(M_v-m_v)\\ \underset{w}{\mathrm{\&Sigma;}}(M_u-m_u)(M_v-m_v)& \underset{w}{\mathrm{\&Sigma;}}{(M_v-m_v)}^2\end{array}\right]---\left(3\right)$

Wherein,

N * n is the size of window W, and this method n value is 4 τ+1, and τ is average ridge distance.

If λ 1 and λ 2 are covariance matrix C _MTwo eigenwerts, then when λ 1＞＞during λ 2, [M _u, M _v] ^TDistribution mainly be to distribute along long axis direction, also promptly distribute along direction perpendicular to crestal line, and in the noise range or the value of singular point district λ 1 and λ 2 be very approaching, therefore, definition [M _u, M _v] ^TDistribution consistance coherence feature as follows:

$\mathrm{cohdrence}=\frac{{{\mathrm{\&lambda;}}_1}^2-{{\mathrm{\&lambda;}}_2}^2}{{{\mathrm{\&lambda;}}_1}^2+{{\mathrm{\&lambda;}}_2}^2}---\left(4\right)$

$=\frac{(\underset{w}{\mathrm{\&Sigma;}}{(M_u-m_u)}^2+\underset{w}{\mathrm{\&Sigma;}}{(M_v-M_v)}^2)\sqrt{{(\underset{w}{\mathrm{\&Sigma;}}{(M_u-m_u)}^2-\underset{w}{\mathrm{\&Sigma;}}{(M_v-m_v)}^2)}^2+4{\left(\underset{w}{\mathrm{\&Sigma;}}(M_u-m_u)(M_v-m_v)\right)}^2}}{{\left(\underset{w}{\mathrm{\&Sigma;}}{(M_u-m_u)}^2\right)}^2+{\left(\underset{w}{\mathrm{\&Sigma;}}{(M_v-m_v)}^2\right)}^2+2{\left(\underset{w}{\mathrm{\&Sigma;}}(M_u-m_u)(M_v-m_v)\right)}^2}$

Thus, the streakline consistance is good more, and distribution consistance coherence is big more.

Claims (3)

1. a new method for detecting fingerprint singularity is characterized in that, its method is,

(1) preproduction phase: the background separation and the field of direction information calculations of carrying out fingerprint image;

(2) phase one: the extraction of candidate's singular point comprises following step:

(a) (i j) calculates its corresponding Poincare index Poincare exponential quantity P to each point among the O ' of direction territory _{G, C}(i, j), adopt two length to be respectively the Poincare index Poincare exponential quantity that 5 * 5 and 7 * 7 closed curve calculates each some during calculating, as long as wherein one result of calculation meets the singular point condition, assert that then this point is candidate's singular point, if two singular point Type-Inconsistencies that curve detection goes out assert directly that then this point is pseudo-point; Calculate each singular point (i, undertaken by following formula during j) Poincare index Poincare exponential quantity:

$\mathrm{Poincare}(i.j)=\frac{1}{2\mathrm{\&pi;}}\underset{k=0}{\overset{N_{\mathrm{\&psi;}}-1}{\mathrm{\&Sigma;}}}\mathrm{\&Delta;}\left(k\right),$

$\mathrm{\&Delta;}\left(k\right)=\left\{\begin{array}{cc}\mathrm{\&delta;}\left(k\right)& \mathrm{if}\left|\mathrm{\&delta;}\left(k\right)\right|<\frac{\mathrm{\&pi;}}{2}\\ \mathrm{\&pi;}+\mathrm{\&delta;}\left(k\right)& \mathrm{if\&delta;}\left(k\right)\&le;-\frac{\mathrm{\&pi;}}{2}\\ \mathrm{\&pi;}-\mathrm{\&delta;}\left(k\right)& \mathrm{otherwise}\end{array}\right.---\left(1\right)$

δ(k)＝O′(ψ _x(i′)，ψ _y(i′))-O′(ψ _x(i)，ψ _y(i))，

i′＝(i+1)modN _ψ，

ψ wherein _x(i) and ψ _y(i) being respectively is the N that has at center with the set point _ψThe x and the y coordinate of k point on the closed curve of individual pixel; The difference of δ (k) expression two adjacent deflections, Δ (k) expression is adjusted later result to difference, the next point after i point of i ' expression; Adopted 5 * 5 and 7 * 7 square closed curve in this algorithm, the formation of this formula is used to calculate certain and puts the accumulation that deflection changes on the corresponding closed curve based on the improvement to original PoincareIndex Poincare index method;

The singular point condition is meant the condition that singular point should satisfy, and this condition is that the computation process to formula (1) retrains, and specifically is meant additional following restrictive condition when doing adding up that deflection changes along closed curve:

Add up this N _ψThe sign change number of times of individual direction, if direction by just to negative and by negative to just each takes place once, and only once, its Poincare index Poincare exponential quantity of continuation calculating then, otherwise, think that this point is a general point;

Add up this N _ψIn the individual difference

The number of difference, if number thinks then that more than one this point is a general point;

If the value of final Poincare index Poincare index is 1/2, (i j) just is confirmed as the core core point to this tested measuring point, if Poincare index Poincare exponential quantity is-1/2 so, (i j) just is confirmed as delta two angle points to this tested measuring point so;

(b) the detected candidate's singular point of previous step utilization clustering algorithm is carried out cluster respectively by Euclidean distance, and add up the number N of the singular point that each cluster comprises _m, N _mThe cluster of＜N is deleted; Wherein, N represents the number that the fingerprint pattern district should be able to detected singular point, and this number is stable, and N is taken as 25 here;

(c) ask for the average sudden change degree of remaining each cluster, and respectively these clusters are sorted by singular point type, sudden change degree size, M the more cluster of sudden change is as candidate's singular point cluster before choosing, and candidate's singular point finally just is positioned on the barycenter of these clusters; The sudden change degree is meant that adjacent two deflections are done in the difference process on the closed curve,

That difference, on average sudden change degree is meant sudden change degree average of all pixels in each cluster;

(3) subordinate phase: candidate's singular point go puppet, comprise following step:

To each candidate's singular point, calculate radius around it and be the distribution consistance coherence value of each pixel in the circular neighborhood of 6 τ, τ is average ridge distance;

Then this border circular areas is divided into 32 fan-shaped and center small circular zones and calculates the average coherence value of various piece, the width of the radius of border circular areas and two round belts is 2 τ;

At last, compare the trizonal size that is evenly distributed consistance coherence value on 16 each directions of direction, central circular is 16 zones that direction is public, and increasing as if the consistance coherence value that outwards distributes from the lining, then this direction of mark is a useful direction; For a certain candidate's singular point, more than or equal to 10, then extracting this point is singular point as if the useful direction number, otherwise, think that this point is pseudo-point; If some direction of singular point gone out the fingerprint border then with remaining direction as the reference direction, on these directions, calculate useful direction; Detailed process is:

Distribution consistance coherence value is expressed the consensus information of streakline, and Gaussian-Hermite Gauss-Hermite square has been adopted in the calculating of distribution consistance coherence value; The streakline consensus information of fingerprint: M is promptly described with four squares _0,1, M _1,0, M _0,3, M _3,0And to these four not the Gaussian-Hermite Gauss-Hermite square of same order do as giving a definition:

$\left\{\begin{array}{c}{M}_n(x,y)=\mathrm{\&lambda;}{M}_{\mathrm{1,0}}(x,y,I(x,y))+(1-\mathrm{\&lambda;})M_{\mathrm{3,0}}(x,y,I(x,y))\\ M_v(x,y)=\mathrm{\&lambda;}{M}_{\mathrm{0,1}}(x,y,I(x,y))+(1-\mathrm{\&lambda;})M_{\mathrm{0,3}}(x,y,I(x,y))\end{array}\right.---\left(2\right)$

Here λ is the associating weight coefficient of the Gaussian-Hermite Gauss-Hermite square of not same order, 0＜λ＜1 wherein, and the λ value 0.5 herein; Utilize formula (2), (i j) can calculate a proper vector [M to each pixel in the fingerprint image _u, M _v] ^TAt fingerprint singularity district and non-singular point district [M _u, M _v] ^TDistribution take on a different character, at non-singular point district [M _u, M _v] ^TDistribute along direction, at singular point district [M perpendicular to crestal line _u, M _v] ^TThen be evenly distributed on all directions, adopt principal component method to extract [M _u, M _v] ^TDistribution character, [M _u, M _v] ^TCovariance matrix C _MDefine by following formula:

$C_M=\left[\begin{array}{cc}\underset{w}{\mathrm{\&Sigma;}}{(M_u-m_u)}^2& \underset{u}{\mathrm{\&Sigma;}}(M_u-m_u)(M_v-m_v)\\ \underset{w}{\mathrm{\&Sigma;}}(M_u-m_u)(M_v-m_v)& \underset{w}{\mathrm{\&Sigma;}}{(M_v-m_v)}^2\end{array}\right]---\left(3\right)$

Wherein,

m _uAverage M in the expression window W _uValue, m _vAverage M in the expression window W _vValue, n * n is the size of window W, and this method n value is 4 τ+1, and τ is average ridge distance;

If λ 1 and λ 2 are covariance matrix C _MTwo eigenwerts, then when λ 1＞＞during λ 2, [M _u, M _v] ^TDistribution mainly be to distribute along long axis direction, also promptly distribute along direction perpendicular to crestal line, and in the noise range or the value of singular point district λ 1 and λ 2 be very approaching, therefore, definition [M _u, M _v] ^TDistribution consistance coherence feature as follows:

$\mathrm{coherence}=\frac{{{\mathrm{\&lambda;}}_1}^2-{{\mathrm{\&lambda;}}_2}^2}{{{\mathrm{\&lambda;}}_1}^2+{{\mathrm{\&lambda;}}_2}^2}---\left(4\right)$

$=\frac{(\underset{w}{\mathrm{\&Sigma;}}{(M_u-m_u)}^2+\underset{w}{\mathrm{\&Sigma;}}{(M_v-m_v)}^2)\sqrt{{(\underset{u}{\mathrm{\&Sigma;}}{(M_u-m_u)}^2-\underset{u}{\mathrm{\&Sigma;}}{(M_v-m_v)}^2)}^2+4{\left(\underset{u}{\mathrm{\&Sigma;}}(M_u-m_u)(M_v-m_v)\right)}^2}}{{\left(\underset{u}{\mathrm{\&Sigma;}}{(M_u-m_u)}^2\right)}^2+{\left(\underset{u}{\mathrm{\&Sigma;}}{(M_v-m_v)}^2\right)}^2+2{\left(\underset{u}{\mathrm{\&Sigma;}}(M_u-m_u)(M_v-m_v)\right)}^2}$

Thus, the streakline consistance is good more, and distribution consistance coherence is big more.

2. new method for detecting fingerprint singularity as claimed in claim 1, it is characterized in that, in described phase one (b), N for the fingerprint pattern district should be able to detected singular point number, value 25 is based on the analysis to test findings, can detected singular point number in the singular point district of fingerprint be metastable and concentrate on the penetralia in singular point district, for the inferior quality fingerprint image, as long as the singular point district is not subjected to the pollution of big noise, number that can detected singular point in the singular point district also is stable so, and this method N gets 25.

3. new method for detecting fingerprint singularity as claimed in claim 1 is characterized in that, in the described subordinate phase, the window size of each pixel distribution consistance coherence value is taken as (4 τ+1) * (4 τ+1).

Images