CN101414350A - Fingerprint image matching method syncretizing multi-reference node integral justification and direction field - Google Patents

Abstract

The invention discloses a fingerprint image matching method which integrates a plurality of reference nodes which are integrally aligned and direction fields. The method comprises the following steps: 1) extracting the information of local relative directions of the relative nodes surrounding a node; 2) carrying out rotation and translational rough aligning on the node set M(T) and M(Q) of a template fingerprint T and an input fingerprint Q; 3) building a process which leads (Rn#-(i)<Q>, Phi n#-(i)<Q>, Psi n#-(i)<Q>> i is equal to or more than 1, and is equal to or less than C) and (Rm n#-(i)<T>, Phi n#-(i)<T>, Psi m#-(i)<T>> i is equal to or more than 1, and is equal to or less than C) to be integrally aligned; 4) converting Mr#-(1)<P>(T) and Mr#-(2)P<Q> into orthogonal coordinates and working out the finally matched node pairs Np of T and Q; 5) calculating the matched fraction STQ of T and Q; 6)giving out the direction fields O(T) and O(Q) of the fingerprint images T and Q to the fingerprint images which contain less nodes or do not contain nodes; 7) aligning the O(T) and O(Q) to obtain O<#>(Q) and calculating the matching degree c of O(T) and O<#>(Q); 8) applying a BP neuronic network to carry out similarity calculation. The fingerprint image matching method is suitable for the fingerprint images which do not contain nodes or are not abundant in nodes and has the advantages of quicker matching speed and higher identifying accuracy rate.

Description

A kind of fingerprint image matching method that merges the whole alignment of many reference modes and the field of direction

Technical field

The present invention relates to the matching process in a kind of fingerprint image identification.

Background technology

Fingerprint matching is a key issue in the fingerprint image recognition technology, and whether fingerprint matching is accurate, is directly connected to the correctness of the final identification of fingerprint image.Present fingerprint recognition system mainly adopts the matching process based on node, and a group node commonly used is represented a fingerprint.

The method of fingerprint matching is a lot, brainstrust has proposed many fingerprint matching algorithms, comprise method based on the sentence structure coupling, method based on images match, based on the method for optical correlation, the method for mating based on the WFMT feature of gray level image is based on the method for textural characteristics, extract the direction of regional area and the method that frequency is carried out fingerprint matching, the matching process of binding site pattern and texture pattern etc. based on GBF.In these methods, node mode is the most widely used fingerprint representation method.Like this, the fingerprint matching problem has just become the node mode matching problem, just seeks the right problem of matched node from the set of two group nodes.On the method for two fingerprints of alignment, Ratha uses GHT to carry out the node alignment, the common practice of Jain and Tico is to seek a reference mode from template fingerprint node set and input fingerprint node set respectively, two group nodes align according to reference mode then, and we are referred to as the method based on single reference point alignment.This method can substantially roughly be alignd two group nodes, but after the alignment of two group nodes, position difference and the direction difference right from reference point matched node far away just might become very big.Therefore, we can expect naturally: matched node to the time, the constraint window size should adjust size with variable in distance.Jain adjusts window size adaptively when node matches, but size how reasonably to adjust window is a problem.Another kind of alternative method is: construct a kind of alignment schemes, make that the position difference of each matched node of two group nodes and direction difference all compare evenly, so just can use the constraint window of fixed size after the alignment of two group nodes.

The shortcoming that existing finger print matching method exists has: (1), the position between pairing node node far away and direction difference become big; (2), the size based on the right alignment schemes self-adaptation adjustment constraint window of single node is comparatively difficult; (3) be only applicable to the more rich fingerprint image of node diagnostic, the not abundant fingerprint image of node or node is inapplicable for not containing; (4), fingerprint matching speed is slower.

Summary of the invention

In order to overcome the node far away that exists in the existing finger print matching method position and direction difference are become big, abundant or do not contain the shortcoming that the node fingerprint image can't be suitable for, matching speed is slower to node, the invention provides a kind of be applicable to do not contain the abundant fingerprint image of node or node, make position difference that each matched node is right and direction difference evenly, matching speed merges the fingerprint image matching method of the whole alignment of many reference modes and the field of direction faster.

The technical solution adopted for the present invention to solve the technical problems is:

A kind of fingerprint image matching method that merges the whole alignment of many reference modes and the field of direction, this matching process may further comprise the steps:

1), extract around the node the local relative direction information of node relatively, comprising:: 1., the coordinate of node (x, y); 2., the direction γ of node, 0≤γ＜2 π; 3., local relative direction D=＜D _α, D _β; 4., ordering number Δ, the computing formula of the number Δ that wherein sorts is:

$\mathrm{\&Delta;}=\underset{-N\&le;i\&le;N,i\&NotEqual;0}{\mathrm{\&Sigma;}}\frac{i}{\left|i\right|}{\mathrm{\&alpha;}}_i/2N---\left(1\right)$

α wherein _iBe an A _iThe accumulative total direction of locating relative node is poor, and N is the sampled point quantity of node one side; With node in the fingerprint image

Be expressed as polar form

If Δ 〉=0 this node of title is a right hand node, otherwise is the left hand node; All nodes that will extract from fingerprint image I increase preface according to Δ to be arranged, and obtains $M\left(I\right)=\left\{\&lang;x_1^I,y_1^I,{\mathrm{\&gamma;}}_1^I,D_1^I,{\mathrm{\&Delta;}}_1^I\&rang;,\&lang;x_2^I,y_2^I,{\mathrm{\&gamma;}}_2^I,D_2^I,{\mathrm{\&Delta;}}_2^I\&rang;,\&CenterDot;\&CenterDot;\&CenterDot;,\&lang;x_p^I,y_p^I,{\mathrm{\&gamma;}}_p^I,D_p^I,{\mathrm{\&Delta;}}_p^I\&rang;\right\}$ Wherein p is the node number, Δ ₁≤ Δ ₂≤ ... ≤ Δ _p, $D_i^I=\&lang;D_{\mathrm{\&alpha;}}^{I,i},D_{\mathrm{\&beta;}}^{I,i}\&rang;,$ $D_{\mathrm{\&alpha;}}^{I,i}=\left\{{\mathrm{\&alpha;}}_k^{I,i}{|}_{k=-N}^N\right\},$ $D_{\mathrm{\&beta;}}^{I,i}=\left\{{\mathrm{\&beta;}}_k^{I,i}{|}_{k=-N}^N\right\};$

2), for the node set of template fingerprint image T and input fingerprint image Q $M\left(T\right)=\{\&lang;x_i^T,y_i^T,{\mathrm{\&gamma;}}_i^T,D_i^T,{\mathrm{\&Delta;}}_i^T\&rang;|1\&le;i\&le;p\}$ With $M\left(Q\right)=\{\&lang;x_i^Q,y_i^Q,{\mathrm{\&gamma;}}_i^Q,D_i^Q,{\mathrm{\&Delta;}}_i^Q\&rang;|1\&le;i\&le;p\}$ Be rotated with translation and slightly align, with

Be reference mode, obtain the polar form of M (T) and Q (T) respectively With

Note C=Count[u] [v] be

With

The matched node logarithm, when it gets maximal value, the set of the matched node that obtains $\{\&lang;x_{m_i}^T,y_{m_i}^T,{\mathrm{\&gamma;}}_{m_i}^T,D_{m_i}^T,{\mathrm{\&Delta;}}_{m_i}^T\&rang;|1\&le;i\&le;C\}$ With $\{\&lang;x_{n_j}^Q,y_{n_j}^Q,{\mathrm{\&gamma;}}_{n_j}^Q,D_{n_j}^Q,{\mathrm{\&Delta;}}_{n_j}^Q\&rang;|1\&le;j\&le;C\}$ Be exactly require many, corresponding to polar coordinates to reference mode

With

With

3), many reference modes are alignd, being exactly will $\{\&lang;x_{m_i}^T,y_{m_i}^T,{\mathrm{\&gamma;}}_{m_i}^T\&rang;|1\&le;i\&le;C\}$ With $\{\&lang;x_{n_j}^Q,y_{n_j}^Q,{\mathrm{\&gamma;}}_{n_j}^Q\&rang;|1\&le;j\&le;C\}$ Carry out integral body alignment, promptly will construct and make

With

The process of whole alignment; Right

Be rotated and translation the rotation parameter-Δ of calculating _ψAnd translation parameters (Δ _x,-Δ _y);

4), will

Be converted to rectangular coordinate $M_{r_1}^c\left(T\right)=\{\&lang;X_i^T,Y_i^T,{\mathrm{\&psi;}}_i^T,D_i^T,{\mathrm{\&Delta;}}_i^T\&rang;|1\&le;i\&le;p\},$ Equally will

Be converted to $M_{r_2}^c\left(Q\right)=\{\&lang;X_i^Q,Y_i^Q,{\mathrm{\&psi;}}_i^Q,D_i^Q,{\mathrm{\&Delta;}}_i^Q\&rang;|1\&le;i\&le;q\},$ Ask the final matched node logarithm N of T and Q _p, computing formula is:

$N_p=\underset{1\&le;i\&le;p,1\&le;j\&le;p}{\mathrm{\&Sigma;}}p(i,j)---\left(2\right)$

Wherein,

5), the coupling mark S of calculated fingerprint image T and Q _TQ, S _TQBe defined as:

$S_{\mathrm{TQ}}=\frac{\underset{p(i,j)=1}{\mathrm{\&Sigma;}}s(m_i^T,m_j^Q)}{N_T+N_Q-N_P}---\left(4\right)$

After two fingerprint alignment, an overlapping public domain F is arranged, establish N _TFor fingerprint image T is arranged in the number of nodes of F, N _QFor being arranged in the number of nodes of F among the fingerprint image Q; S _TQCan be regarded as the similarity of fingerprint image T and Q, S _TQBe worth greatly more, then T is similar more with Q;

Wherein Be $m_i^T=\&lang;x_i^T,y_i^T,{\mathrm{\&gamma;}}_i^T,D_i^T,{\mathrm{\&Delta;}}_i^T\&rang;$ With $m_j^Q=\&lang;x_i^Q,y_i^Q,{\mathrm{\&gamma;}}_i^Q,D_i^Q,{\mathrm{\&Delta;}}_i^Q\&rang;$ Similarity, computing formula is:

$s(m_i^T,m_j^Q)=\exp [-\frac{\mathrm{\&rho;}\&CenterDot;{\mathrm{\&Sigma;}}_{k=-N}^N(|{\mathrm{\&alpha;}}_k^{T,i}-{\mathrm{\&alpha;}}_k^{Q,j}|+|{\mathrm{\&beta;}}_k^{T,i}-{\mathrm{\&beta;}}_k^{Q,j}|)}{2\mathrm{\&pi;N}}]---\left(5\right)$

Wherein ρ is a constant, gets 10 in the experiment, and N is the quantity parameter of sampled point.

6), less or do not contain the fingerprint image of node for containing number of nodes, the field of direction that provides fingerprint image T and Q is O (T) and O (Q), O (T)={ O (W ^T(i, j)) | 0≤i＜m/w, 0≤j＜n/w}, O (Q)={ O (W ^Q(i, j)) | 0≤i＜m/w, 0≤j＜n/w}, with the ridge orientation stipulations to [0, π), for background piece W (i, j), its direction is meaningless, puts O (W (i, j))=-1; If W (i j) is foreground blocks, then O (W (i, j)) ∈ [and 0, π);

7), field of direction O (T) and the O (Q) of fingerprint image T and Q alignd, obtain: $O^{\#}\left(Q\right)=\{O\left(W^{Q^{\#}}(i,j)\right)|0\&le;i<m/w,0\&le;j<n/w\},$ Calculate O (T) and O ^#(Q) matching degree c, computing formula is:

$c=\frac{\underset{0\&le;i<m/w,0\&le;j<n/w}{\mathrm{\&Sigma;}}d(i,j)}{\underset{0\&le;i<m/w,0\&le;j<n/w}{\mathrm{\&Sigma;}}\mathrm{\&eta;}(i,j)}\&times;100---\left(6\right)$

C is the proportion of the shared overlapping region of direction that differs greatly after the alignment of both direction field;

8), utilization BP neural network is carried out similarity and is calculated containing number of nodes fingerprint image less or that do not contain node.The input of neural network is 5) node matching mark S _TQWith field of direction matching degree c, be output as the similarity of two fingerprints that participate in coupling; Training stage, for two identical fingerprints, be output as 1, otherwise be output as 0; At cognitive phase, the network output valve is between 0 and 1, and similarity is big more, and then two fingerprints are similar more;

As preferred a kind of scheme: described 1), come node diagnostic is described, two nodes are mated with local relative direction information.By around node, choosing some sampled points, calculate a LRO value at each sampled point, LRO is calculated as from node along straight line to the accumulative total direction difference SOD of sampled point; If (i, j) ((i j)＜π) is that (the local relative direction that centers on node is expressed as D=＜D for i, the ridge orientation of j) locating at point to 0≤O to O _α, D _β, wherein $D_{\mathrm{\&alpha;}}=\left\{{\mathrm{\&alpha;}}_i{|}_{i=-N}^N\right\},$ $D_{\mathrm{\&beta;}}=\left\{{\mathrm{\&beta;}}_i{|}_{i=-N}^N\right\},$ α ₀＝0，β ₀＝0。α _iAnd β _iBe respectively at an A _iAnd B _iLocate the SOD value of relative node.Suppose that node coordinate is for (x, y), direction is γ (0≤γ＜2 π), so

For from (x is y) to A _iSOD, A _i=(x+ilcos (γ), y+ilsin (γ)).Use starting condition α ₀=0, i is from 1 to N, α _i| i〉0 calculate by step (1.1) and (1.2) iteration:

(1.1), utilize known quantity α _I-1Calculate σ _i, computing formula is:

O(A _i)-(γ+α _i-i/|i|)＝2kπ+σ _i，k∈Z，0≤σ _i<2π (7)

(1.2), utilize known quantity α _I-1And σ _iCalculate α _i, computing formula is:

${\mathrm{\&alpha;}}_i=\left\{\begin{array}{cc}{\mathrm{\&alpha;}}_{i-i/\left|i\right|}+{\mathrm{\&sigma;}}_i& 0\&le;{\mathrm{\&sigma;}}_i\&le;2\mathrm{\&pi;}\\ {\mathrm{\&alpha;}}_{i-i/\left|i\right|}+{\mathrm{\&sigma;}}_i-\mathrm{\&pi;}& \mathrm{\&pi;}/2<{\mathrm{\&sigma;}}_i\&le;3\mathrm{\&pi;}/2\\ {\mathrm{\&alpha;}}_{i-i/\left|i\right|}+{\mathrm{\&sigma;}}_i-2\mathrm{\&pi;}& 3\mathrm{\&pi;}/2<{\mathrm{\&sigma;}}_i\&le;2\mathrm{\&pi;}\end{array}\right.---\left(8\right)$

I is from-1 to-N, α _i| i＜0 can be calculated by step (1.3) and (1.4) iteration:

(1.3), by formula (7), utilize known quantity α _I+1Calculate σ _i

(1.4), by formula (8), utilize known quantity α _I+1And σ _iCalculate α _i

(1.5), be similar to Computation process,

For from (x is y) to B _iSOD, B _i=(x+ilcos (γ-2 π), y+ilsin (γ-2 π)).Use starting condition β ₀=0, Calculate by formula (9) and (10) iteration:

O(B _i)-(γ+β _i-i/|i|)＝2kπ+σ _i，k∈Z，0≤σ _i<2π (9)

${\mathrm{\&beta;}}_i=\left\{\begin{array}{cc}{\mathrm{\&beta;}}_{i-i/\left|i\right|}+{\mathrm{\&sigma;}}_i& 0\&le;{\mathrm{\&sigma;}}_i\&le;2\mathrm{\&pi;}\\ {\mathrm{\&beta;}}_{i-i/\left|i\right|}+{\mathrm{\&sigma;}}_i-\mathrm{\&pi;}& \mathrm{\&pi;}/2<{\mathrm{\&sigma;}}_i\&le;3\mathrm{\&pi;}/2\\ {\mathrm{\&beta;}}_{i-i/\left|i\right|}+{\mathrm{\&sigma;}}_i-2\mathrm{\&pi;}& 3\mathrm{\&pi;}/2<{\mathrm{\&sigma;}}_i\&le;2\mathrm{\&pi;}\end{array}\right.---\left(10\right)$

As preferred another scheme: described 3) in, the whole alignment of many reference modes is defined as follows: establish $\{\&lang;x_i^A,y_i^A,{\mathrm{\&gamma;}}_i^A\&rang;|1\&le;i\&le;C\}$ With $\{\&lang;x_i^B,y_i^B,{\mathrm{\&gamma;}}_i^B\&rang;|1\&le;i\&le;C\}$ Be respectively the subclass of the node set of fingerprint image A and B, if ${\mathrm{\&Sigma;}}_{i=1}^C(x_i^A-x_i^B)=0,$ ${\mathrm{\&Sigma;}}_{i=1}^C(y_i^A-y_i^B)=0,$ ${\mathrm{\&Sigma;}}_{i=1}^C({\mathrm{\&gamma;}}_i^A-{\mathrm{\&gamma;}}_i^B)=0,$ Then claim $\{\&lang;x_i^A,y_i^A,{\mathrm{\&gamma;}}_i^A\&rang;|1\&le;i\&le;C\}$ With $\{\&lang;x_i^B,y_i^B,{\mathrm{\&gamma;}}_i^B\&rang;|1\&le;i\&le;C\}$ Be whole the alignment; If function f makes $f(\&lang;x_i^B,y_i^B,{\mathrm{\&gamma;}}_i^B\&rang;)=\&lang;x_i^{\&prime;B},y_i^{\&prime;B},{\mathrm{\&gamma;}}_i^{\&prime;B}\&rang;,$ And $\{\&lang;x_i^A,y_i^A,{\mathrm{\&gamma;}}_i^A\&rang;|1\&le;i\&le;C\}$ With $\{\&lang;x_i^{\&prime;B},y_i^{\&prime;B},{\mathrm{\&gamma;}}_i^{\&prime;B}\&rang;|1\&le;i\&le;C\}$ Be whole the alignment, then claim f to make $\{\&lang;x_i^A,y_i^A,{\mathrm{\&gamma;}}_i^A\&rang;|1\&le;i\&le;C\}$ With $\{\&lang;x_i^B,y_i^B,{\mathrm{\&gamma;}}_i^B\&rang;|1\&le;i\&le;C\}$ The function of whole alignment;

Whole alignment can similarly be defined as follows for many reference modes under the polar form: establish

With

Be respectively the subclass of node set of the polar form of fingerprint image A and B, if

${\mathrm{\&Sigma;}}_{i=1}^C({\mathrm{\&psi;}}_i^A-{\mathrm{\&psi;}}_i^B)=0,$ Then claim

With

Be whole the alignment.If function f makes

And

With

Be whole the alignment, then claim f to make With

The function of whole alignment;

Described 2) in calculated many to reference mode $\{\&lang;x_{m_i}^T,y_{m_i}^T,{\mathrm{\&gamma;}}_{m_i}^T\&rang;|1\&le;i\&le;C\}$ With $\{\&lang;x_{n_j}^Q,y_{n_j}^Q,{\mathrm{\&gamma;}}_{n_j}^Q\&rang;|1\&le;j\&le;C\}$ Wherein

With Be that matched node is right, corresponding to polar coordinates

With In

With

Basic thought based on many reference mode alignment is exactly will $\{\&lang;x_{m_i}^T,y_{m_i}^T,{\mathrm{\&gamma;}}_{m_i}^T\&rang;|1\&le;i\&le;C\}$ With $\{\&lang;x_{n_j}^Q,y_{n_j}^Q,{\mathrm{\&gamma;}}_{n_j}^Q\&rang;|1\&le;j\&le;C\}$ Carry out integral body alignment, promptly will construct and make With

The process of whole alignment; It is right that this process can be decomposed into Be rotated and translation the rotation parameter-Δ of calculating _ψAnd translation parameters (Δ _x,-Δ _y).Concrete steps are as follows:

(3.1), calculate

With

Direction difference, computing formula is as follows:

${\mathrm{\&Delta;}}_{\mathrm{\&psi;}}=\frac{\underset{i=1}{\overset{C}{\mathrm{\&Sigma;}}}({\mathrm{\&psi;}}_{n_i}^Q-{\mathrm{\&psi;}}_{m_i}^T)}{C}---\left(11\right)$

(3.2), will

Rotation-Δ _ψObtain

Calculate again its with

Position difference, computing formula is:

(3.3) ,-Δ _ψ(Δ _x,-Δ _y) be rotation parameter and translation parameters.Will Rotation-Δ _ψ, translation (Δ then _x,-Δ _y), obtain

Then With

Be whole the alignment;

(3.4), will

Rotation-Δ _ψ, translation (Δ then _x,-Δ _y), obtain

Then fingerprint image T and Q have obtained alignment based on multiple reference points;

As preferred another kind of again scheme: 4) in the fingerprint image T that addresses and the final matched node logarithm N of Q _pFind the solution because

With According to the Δ sort ascending, the left hand node comes right hand node front.If j _ROFor

First right hand node location, j _LSatisfy ${\mathrm{\&Delta;}}_{j_R}^Q-{\mathrm{\&Delta;}}_{j_{\mathrm{RO}}}^Q\&GreaterEqual;-e$ And ${\mathrm{\&Delta;}}_{j_R-1}^Q-{\mathrm{\&Delta;}}_{j_{\mathrm{RO}}}^Q<-e,$ E is a threshold value.N then _pCalculating according to following algorithm:

Step1. initialization N _p=0, i=0, j=0;

If Step2.i=i+1 is i〉p then returns N _p

Step3. if

In i node be right hand node, then

Step4.j＝j+1；

Step5. if ${\mathrm{\&Delta;}}_j^Q-{\mathrm{\&Delta;}}_i^T<-e,$ Then forward Step4 to;

Step6. if ${\mathrm{\&Delta;}}_j^Q-{\mathrm{\&Delta;}}_i^T>-e,$ Then j=0 forwards Step2 to;

Step7. if p (i, j)=1, N then _p=N _p+ 1, j=0 forwards Step2 to;

If Step8.j=j+1 is j〉q, then j=0 forwards Step2 to, otherwise forwards Step6 to.

Further, described 7) in to O (T) and O ^#(Q) matching degree c calculates, and the detailed calculated method is:

Step1. calculate W ^T(i, j) with

Direction poor, computing formula is:

Step2. direction difference stipulations are arrived [0, pi/2]:

$\mathrm{\&sigma;}(i,j)=\left\{\begin{array}{cc}\mathrm{\&pi;}-\mathrm{\&sigma;}\&prime;(i,j)& \mathrm{\&sigma;}\&prime;(i,j)>\mathrm{\&pi;}/2\\ \mathrm{\&sigma;}\&prime;(i,j)& 0\&le;\mathrm{\&sigma;}\&prime;(i,j)\&le;\mathrm{\&pi;}/2\\ -1& \mathrm{\&sigma;}\&prime;(i,j)<0\end{array}\right.---\left(15\right)$

Step3. the stipulations result to the direction difference classifies according to threshold value t:

Wherein the t in the formula (16) is a threshold value; In the formula (14)

Computing method as follows:

u＝i×ω+ω/2，v＝j×ω+ω/2 (18)

$\mathrm{\&gamma;}=\sqrt{{(u-x_{r_1}^T)}^2+{(v-y_{r_1}^T)}^2}---\left(19\right)$

$\mathrm{\&alpha;}\&prime;=\arctan ((v-y_{r_1}^T)/(u-x_{r_1}^T))---\left(20\right)$

$\mathrm{\&alpha;}=\left\{\begin{array}{cc}\mathrm{\&pi;}/2& u=x_{r_1}^T,v>y_{r_1}^T\\ 0& u=x_{r_1}^T,v=y_{r_1}^T\\ -\mathrm{\&pi;}/2& u=x_{r_1}^T,v<y_{r_1}^T\\ \mathrm{\&alpha;}\&prime;& u>x_{r_1}^T\\ \mathrm{\&alpha;}\&prime;+\mathrm{\&pi;}& u<x_{r_1}^T\end{array}\right.---\left(21\right)$

$x=x_{r_2}^Q+\mathrm{\&gamma;}\&times;\cos (\mathrm{\&alpha;}-{\mathrm{\&theta;}}_{r_1}^T+{\mathrm{\&theta;}}_{r_2}^Q)---\left(22\right)$

$y=y_{r_2}^Q+\mathrm{\&gamma;}\&times;\sin (\mathrm{\&alpha;}-{\mathrm{\&theta;}}_{r_1}^T+{\mathrm{\&theta;}}_{r_2}^Q)---\left(23\right)$

Wherein,

Expression

Be prospect.

Further again, described 8) in neural network judge whether identical differentiation feature specifically describes as follows two fingerprint images:

(a) based on the coupling mark S under many reference mode alignment _TQ

(b) alignment back matched node logarithm N _p

(c) overlapping region, alignment back area;

(d) the image block number of direction difference between [0 °, 5 °] in the overlapping region;

(e) in the overlapping region direction difference (5 °, 10 °] between the image block number;

(f) in the overlapping region direction difference (10 °, 20 °] between the image block number.

Beneficial effect of the present invention mainly shows: 1, can be applicable to node that contains lesser amt or the automatic identification of the fingerprint image that does not contain node; 2, fingerprint recognition accuracy height; 3, algorithm complex is low, and is higher than existing algorithm efficient.

Description of drawings

Fig. 1 is the sampled point of local direction around the node.

Fig. 2 is the simplified example of node alignment, wherein, (a) is the every template node set; (b) be the input node set; (c) be based on a pair of node＜a, a '〉align, obtain two couples of matched node＜a, a '〉and＜b, b ' 〉; (d) be based on two couples of node＜a, a '〉and＜b, b '〉alignment, obtain three pairs of matched node.

Fig. 3 is based on single reference mode alignment and the alignment of many reference modes, wherein, (a) is based on single reference mode alignment, totally 18 pairs of matched node; (b) be based on many reference mode alignment, totally 23 pairs of matched node.

Fig. 4 compares the ROC curve that obtains for this method and other method, and its horizontal ordinate is a false acceptance rate, and ordinate is a false rejection rate, and the ROC curve of recognizer illustrates that more near transverse axis the accuracy of this algorithm is high more.Algorithm A is this method among the figure.

Fig. 5 is the matching process in conjunction with the multinode and the field of direction.

Fig. 6 is that wherein algorithm C is this method in conjunction with the finger print matching method of the node and the field of direction and other method ROC curve relatively.

Embodiment

Below in conjunction with accompanying drawing the present invention is further described.

With reference to 1～Fig. 6, a kind of fingerprint image matching method that merges the whole alignment of many reference modes and the field of direction, this matching process may further comprise the steps:

1), extract around the node the local relative direction information of node relatively, comprising: 1., the coordinate of node (x, y); 2., the direction γ of node, 0≤γ＜2 π; 3., local relative direction D=＜D _α, D _β; 4., ordering number Δ, the computing formula of the number Δ that wherein sorts is:

$\mathrm{\&Delta;}=\underset{-N\&le;i\&le;N,i\&NotEqual;0}{\mathrm{\&Sigma;}}\frac{i}{\left|i\right|}{\mathrm{\&alpha;}}_i/2N---\left(1\right)$

α wherein _iBe an A _iThe accumulative total direction of locating relative node is poor, and N is the sampled point quantity of node one side;

With node in the fingerprint image

Be expressed as polar form

If Δ 〉=0 this node of title is a right hand node, otherwise is the left hand node.All nodes that will extract from fingerprint image I increase preface according to Δ to be arranged, and obtains $M\left(I\right)=\left\{\&lang;x_1^I,y_1^I,{\mathrm{\&gamma;}}_1^I,D_1^I,{\mathrm{\&Delta;}}_1^I\&rang;,\&lang;x_2^I,y_2^I,{\mathrm{\&gamma;}}_2^I,D_2^I,{\mathrm{\&Delta;}}_2^I\&rang;,\&CenterDot;\&CenterDot;\&CenterDot;,\&lang;x_p^I,y_p^I,{\mathrm{\&gamma;}}_p^I,D_p^I,{\mathrm{\&Delta;}}_p^I\&rang;\right\}$ Wherein p is the node number, Δ ₁≤ Δ ₂≤ ... ≤ Δ _p, $D_i^I=\&lang;D_{\mathrm{\&alpha;}}^{I,i},D_{\mathrm{\&beta;}}^{I,i}\&rang;,$ $D_{\mathrm{\&alpha;}}^{I,i}=\left\{{\mathrm{\&alpha;}}_k^{I,i}{|}_{k=-N}^N\right\},$ $D_{\mathrm{\&beta;}}^{I,i}=\left\{{\mathrm{\&beta;}}_k^{I,i}{|}_{k=-N}^N\right\}.$

2), for the node set of template fingerprint image T and input fingerprint image Q $M\left(T\right)=\{\&lang;x_i^T,y_i^T,{\mathrm{\&gamma;}}_i^T,D_i^T,{\mathrm{\&Delta;}}_i^T\&rang;|1\&le;i\&le;p\}$ With $M\left(Q\right)=\{\&lang;x_i^Q,y_i^Q,{\mathrm{\&gamma;}}_i^Q,D_i^Q,{\mathrm{\&Delta;}}_i^Q\&rang;|1\&le;i\&le;p\}$ Be rotated with translation and slightly align, with

Be reference mode, obtain the polar form of M (T) and Q (T) respectively

With Note C=Count[u] [v] be

With

The matched node logarithm, when it gets maximal value, the set of the matched node that obtains $\{\&lang;x_{m_i}^T,y_{m_i}^T,{\mathrm{\&gamma;}}_{m_i}^T,D_{m_i}^T,{\mathrm{\&Delta;}}_{m_i}^T\&rang;|1\&le;i\&le;C\}$ With $\{\&lang;x_{n_j}^Q,y_{n_j}^Q,{\mathrm{\&gamma;}}_{n_j}^Q,D_{n_j}^Q,{\mathrm{\&Delta;}}_{n_j}^Q\&rang;|1\&le;j\&le;C\}$ Be exactly require many, corresponding to polar coordinates to reference mode

With

With

3), many reference modes are alignd, being exactly will $\{\&lang;x_{m_i}^T,y_{m_i}^T,{\mathrm{\&gamma;}}_{m_i}^T\&rang;|1\&le;i\&le;C\}$ With $\{\&lang;x_{n_j}^Q,y_{n_j}^Q,{\mathrm{\&gamma;}}_{n_j}^Q,D_{n_j}^Q,{\mathrm{\&Delta;}}_{n_j}^Q\&rang;|1\&le;j\&le;C\}$ Carry out integral body alignment, promptly will construct and make With

The process of whole alignment.Right

Be rotated and translation the rotation parameter-Δ of calculating _ψAnd translation parameters (Δ _x,-Δ _y).

4), will

Be converted to rectangular coordinate $M_{r_1}^c\left(T\right)=\{\&lang;X_i^T,Y_i^T,{\mathrm{\&psi;}}_i^T,D_i^T,{\mathrm{\&Delta;}}_i^T\&rang;|1\&le;i\&le;p\},$ Equally will Be converted to $M_{r_2}^c\left(Q\right)=\{\&lang;X_i^Q,Y_i^Q,{\mathrm{\&psi;}}_i^Q,D_i^Q,{\mathrm{\&Delta;}}_i^Q\&rang;|1\&le;i\&le;q\},$ Ask the final matched node logarithm N of T and Q _p, computing formula is:

$N_p=\underset{1\&le;i\&le;p,1\&le;j\&le;p}{\mathrm{\&Sigma;}}p(i,j)---\left(2\right)$

Wherein,

5), the coupling mark S of calculated fingerprint image T and Q _TQ, S _TQBe defined as:

$S_{\mathrm{TQ}}=\frac{\underset{p(i,j)=1}{\mathrm{\&Sigma;}}s(m_i^T,m_j^Q)}{N_T+N_Q-N_P}---\left(4\right)$

After two fingerprint alignment, an overlapping public domain F is arranged, establish N _TFor fingerprint image T is arranged in the number of nodes of F, N _QFor being arranged in the number of nodes of F among the fingerprint image Q.S _TQCan be regarded as the similarity of fingerprint image T and Q, S _TQBe worth greatly more, then T is similar more with Q.

Wherein

Be $m_i^T=\&lang;x_i^T,y_i^T,{\mathrm{\&gamma;}}_i^T,D_i^T,{\mathrm{\&Delta;}}_i^T\&rang;$ With $m_j^Q=\&lang;x_i^Q,y_i^Q,{\mathrm{\&gamma;}}_i^Q,D_i^Q,{\mathrm{\&Delta;}}_i^Q\&rang;$ Similarity, computing formula is:

$s(m_i^T,m_j^Q)=\exp [-\frac{\mathrm{\&rho;}\&CenterDot;{\mathrm{\&Sigma;}}_{k=-N}^N(|{\mathrm{\&alpha;}}_k^{T,i}-{\mathrm{\&alpha;}}_k^{Q,j}|+|{\mathrm{\&beta;}}_k^{T,i}-{\mathrm{\&beta;}}_k^{Q,j}|)}{2\mathrm{\&pi;N}}]---\left(5\right)$

Wherein ρ is a constant, gets 10 in the experiment, and N is the quantity parameter of sampled point.

6), less or do not contain the fingerprint image of node for containing number of nodes described method also comprises:, utilize the field of direction to mate.The field of direction of fingerprint image T and Q is O (T) and O (Q), O (T)={ O (W ^T(i, j)) | 0≤i＜m/w, 0≤j＜n/w}, O (Q)={ O (W ^Q(i, j)) | 0≤i＜m/w, 0≤j＜n/w}.This method with the ridge orientation stipulations to [0, π), for background piece W (i, j), its direction is meaningless, puts O (W (i, j))=-1; If W (i j) is foreground blocks, then O (W (i, j)) ∈ [and 0, π).

7), field of direction O (T) and the O (Q) of fingerprint image T and Q alignd, obtain:

$O^{\#}\left(Q\right)=\{O\left(W^{Q^{\#}}(i,j)\right)|0\&le;i<m/w,0\&le;j<n/w\},$ Calculate O (T) and O ^#(Q) matching degree c, computing formula is:

$c=\frac{\underset{0\&le;i<m/w,0\&le;j<n/w}{\mathrm{\&Sigma;}}d(i,j)}{\underset{0\&le;i<m/w,0\&le;j<n/w}{\mathrm{\&Sigma;}}\mathrm{\&eta;}(i,j)}\&times;100---\left(6\right)$

C is the proportion of the shared overlapping region of direction that differs greatly after the alignment of both direction field.

8), utilization BP neural network is carried out similarity and is calculated containing number of nodes fingerprint image less or that do not contain node.The input of neural network is 5) node matching mark S _TQWith field of direction matching degree c, be output as the similarity of two fingerprints that participate in coupling.Training stage, for two identical fingerprints, be output as 1, otherwise be output as 0; At cognitive phase, the network output valve is between 0 and 1, and similarity is big more, and then two fingerprints are similar more.

Choosing as shown in Figure 1 of sampled point, sampling point distributions is on two mutually perpendicular straight lines that intersect at node, and wherein straight line is parallel with node direction.On every straight line, be the center with the node, choose equally spacedly and comprise node at an interior 2N+1 sampled point, promptly respectively get N point on the node both sides, the spacing between two sampled points is l, is called step-length.Fig. 1 has provided node m and its sampled point A _iAnd B _i(1≤| i|≤N).

Node mode coupling is right in order to seek corresponding node from every template node set and input node set, and Fig. 2 has provided a simple example to comparing based on single reference point with based on the match pattern of multiple reference points.Fig. 2 (a) is the node set of template fingerprint, comprises three node a, b and c; Fig. 2 (b) is the node set of input fingerprint, comprises three node a ', b ' and c ', respectively with a, b and the c correspondence of template fingerprint.Suppose method based on single reference mode, with＜a, a '〉be that reference mode is right, be about to a and a ' complete matching, obtain two couples of matched node＜a, a '〉and＜b, b ' 〉, however that c and c ' do not become matched node is right, because based on after single reference mode alignment, the distance difference of c and c ' is too big, shown in Fig. 2 (c).Node will be rotated and translation if employing based on the thought of many reference modes, will be imported, and make＜a, a '〉and＜b, b '〉obtain whole alignment, although a and a ' and b and b ' are not overlapping fully, thereby but the distance of c and c ' dwindle and become matched node, shown in Fig. 2 (d).In the example of Fig. 2,＜a, a '〉and＜b, b '〉be that many reference modes are right, in the matching process based on many reference modes, at first need to determine many to reference mode, then it being carried out integral body alignment, is alignment of the overall situation based on the method for many reference modes, rather than local alignment.

Fig. 3 has provided the strength based on many reference mode alignment, under situation based on single reference mode alignment, shown in Fig. 3 (a), be labeled as 1,2,3,4,5 node is right to become the pairing node too greatly and not owing to alternate position spike XOR direction difference, and under the situation based on many reference mode alignment, shown in Fig. 3 (b), their difference is dwindled and to become matched node right, be labeled as 6 node to not become matched node too greatly right because of difference based on many reference modes alignment the time, thus Fig. 3 (b) to have increased by 5 pairs of matched node than Fig. 3 (a) right.From Fig. 3 (a), near the regional area alignment effect the reference mode is fine, but relatively poor from reference mode region alignment effect far away; Among Fig. 3 (b), near the region alignment effect the former reference mode is not as Fig. 3 (a), but whole alignment effect is better than Fig. 3 (a), so consider whole best alignment based on the alignment schemes of many reference modes.

ROC curve shown in Figure 4 is based on the comparison of method and other method effect of many reference modes, is algorithm A based on the method for many reference modes.Wherein horizontal ordinate is represented false acceptance rate, and ordinate is represented false rejection rate, and the ROC curve of algorithm shows that more near horizontal ordinate this algorithm accuracy is high more.Can see that by comparing the ROC curve matching process of examining the node alignment based on misincorporation has improved the accuracy of identification really.

Abundant or do not contain the fingerprint image of node for node, the present invention proposes the fingerprint image matching method that merges the whole alignment of many reference modes and the field of direction, as shown in Figure 5.For the fingerprint image that contains node, at first carry out based on the right coupling of many reference modes, obtain the coupling mark S of two fingerprint images _TQIf the fingerprint image nodal information is not abundant or do not comprise nodal information, then the field of direction of fingerprint image is mated, obtain the field of direction coupling mark c of two fingerprint images, in conjunction with S _TQWith c fingerprint is carried out accurate match.The matching result that utilizes this method is shown in the ROC curve of Fig. 6, and wherein algorithm A is the method based on single reference mode, and algorithm B is the method based on many reference modes, and algorithm C is this method.ROC curve by comparison algorithm A and algorithm B can be seen, improves the node alignment schemes and can improve the accuracy of identification; The ROC curve of comparison algorithm B and algorithm C can see that the bonding position matching result also can improve the accuracy of identification.

The present invention can be to fingerprint image based on many reference modes to carrying out the coupling of precise and high efficiency, bonding position field coupling is applicable to the abundant coupling demand that does not even comprise the fingerprint image of node of nodal information simultaneously.The experiment proved that this method has effectively improved the accuracy of fingerprint matching.

Claims (6)

1, a kind of fingerprint image matching method that merges the whole alignment of many reference modes and the field of direction, it is characterized in that: this matching process may further comprise the steps:

1), extract around the node the local relative direction information of node relatively, comprising: 1., the coordinate of node (x, y); 2., the direction γ of node, 0≤γ＜2 π; 3., local relative direction D=＜D _α, D _β; 4., ordering number Δ, the computing formula of the number Δ that wherein sorts is:

$\mathrm{\&Delta;}=\underset{-N\&le;i\&le;N,i\&NotEqual;0}{\mathrm{\&Sigma;}}\frac{i}{\left|i\right|}{\mathrm{\&alpha;}}_i/2N---\left(1\right)$

Wherein, α wherein _iBe an A _iThe accumulative total direction of locating relative node is poor, and N is the sampled point quantity of node one side; With node in the fingerprint image

Be expressed as polar form

If Δ 〉=0 this node of title is a right hand node, otherwise is the left hand node; All nodes that will extract from fingerprint image I increase preface according to Δ to be arranged, and obtains $M\left(I\right)=\{\&lang;x_1^I,y_1^I,{\mathrm{\&gamma;}}_1^I,D_1^I,{\mathrm{\&Delta;}}_1^I\&rang;,\&lang;x_2^I,y_2^I,{\mathrm{\&gamma;}}_2^I,D_2^I,{\mathrm{\&Delta;}}_2^I\&rang;,\&CenterDot;\&CenterDot;\&CenterDot;,\&lang;x_p^I,y_p^I,{\mathrm{\&gamma;}}_p^I,D_p^I,{\mathrm{\&Delta;}}_p^I\&rang;\},$ Wherein p is the node number, ${\mathrm{\&Delta;}}_1\&le;{\mathrm{\&Delta;}}_2\&le;\&CenterDot;\&CenterDot;\&CenterDot;\&le;{\mathrm{\&Delta;}}_p,$ $D_i^I=\&lang;D_{\mathrm{\&alpha;}}^{I,i},D_{\mathrm{\&beta;}}^{I,i}\&rang;,$ $D_{\mathrm{\&alpha;}}^{I,i}=\left\{{\mathrm{\&alpha;}}_k^{I,i}{|}_{k=-N}^N\right\},$ $D_{\mathrm{\&beta;}}^{I,i}=\left\{{\mathrm{\&beta;}}_k^{I,i}{|}_{k=-N}^N\right\};$

2), for the node set of template fingerprint image T and input fingerprint image Q $M\left(T\right)=\{\&lang;x_i^T,y_i^T,{\mathrm{\&gamma;}}_i^T,D_i^T,{\mathrm{\&Delta;}}_i^T\&rang;|1\&le;i\&le;p\}$ With $M\left(Q\right)=\{\&lang;x_i^Q,y_i^Q,{\mathrm{\&gamma;}}_i^Q,D_i^Q,{\mathrm{\&Delta;}}_i^Q\&rang;|1\&le;i\&le;q\}$ Be rotated with translation and slightly align, with Be reference mode, obtain the polar form of M (T) and Q (T) respectively

With

Note C=Count[u] [v] be With

The matched node logarithm, when it gets maximal value, the set of the matched node that obtains $\{\&lang;x_{m_i}^T,y_{m_i}^T,{\mathrm{\&gamma;}}_{m_i}^T,D_{m_i}^T,{\mathrm{\&Delta;}}_{m_i}^T\&rang;|1\&le;i\&le;C\}$ With $\{\&lang;x_{n_j}^Q,y_{n_j}^Q,{\mathrm{\&gamma;}}_{n_j}^Q,D_{n_j}^Q,{\mathrm{\&Delta;}}_{n_j}^Q\&rang;|1\&le;j\&le;C\}$ Be many, corresponding to polar coordinates to reference mode

(T) and

With

3), many reference modes are alignd: will $\{\&lang;x_{m_i}^T,y_{m_i}^T,{\mathrm{\&gamma;}}_{m_i}^T\&rang;|1\&le;i\&le;C\}$ With $\{\&lang;x_{n_j}^Q,y_{n_j}^Q,{\mathrm{\&gamma;}}_{n_j}^Q\&rang;|1\&le;j\&le;C\}$ Carry out integral body alignment, promptly will construct and make

With

Whole alignment; Right Be rotated and translation the rotation parameter-Δ of calculating _ψAnd translation parameters (Δ _x,-Δ _y);

4), will Be converted to rectangular coordinate $M_{r_1}^c\left(T\right)=\{\&lang;X_i^T,Y_i^T,{\mathrm{\&psi;}}_i^T,D_i^T,{\mathrm{\&Delta;}}_i^T\&rang;|1\&le;i\&le;p\},$ Equally will

Be converted to $M_{r_2}^c\left(Q\right)=\{\&lang;X_i^Q,Y_i^Q,{\mathrm{\&psi;}}_i^Q,D_i^Q,{\mathrm{\&Delta;}}_i^Q\&rang;|1\&le;i\&le;q\},$ Ask the final matched node logarithm N of T and Q _p, computing formula is:

$N_p=\underset{1\&le;i\&le;p,1\&le;j\&le;p}{\mathrm{\&Sigma;}}p(i,j)---\left(2\right)$

Wherein,

5), the coupling mark S of calculated fingerprint image T and Q _TQ, S _TQBe defined as:

$S_{\mathrm{TQ}}=\frac{\underset{p(i,j)=1}{\mathrm{\&Sigma;}}s(m_i^T,m_j^Q)}{N_T+N_Q-N_P}---\left(4\right)$

After two fingerprint alignment, an overlapping public domain F is arranged, establish N _TFor fingerprint image T is arranged in the number of nodes of F, N _QFor being arranged in the number of nodes of F among the fingerprint image Q; S _TQBe the similarity of fingerprint image T and Q, S _TQBe worth greatly more, T is similar more with Q;

Wherein

Be $m_i^T=\&lang;x_i^T,y_i^T,{\mathrm{\&gamma;}}_i^T,D_i^T,{\mathrm{\&Delta;}}_i^T\&rang;$ With $m_j^Q=\&lang;x_i^Q,y_i^Q,{\mathrm{\&gamma;}}_i^Q,D_i^Q,{\mathrm{\&Delta;}}_i^Q\&rang;$ Similarity be that computing formula is:

$s(m_i^T,m_j^Q)=\exp [-\frac{\mathrm{\&rho;g}{\mathrm{\&Sigma;}}_{k=-N}^N(|{\mathrm{\&alpha;}}_k^{T,i}-{\mathrm{\&alpha;}}_k^{Q,j}|+|{\mathrm{\&beta;}}_k^{T,i}-{\mathrm{\&beta;}}_k^{Q,j}|)}{2\mathrm{\&pi;N}}]---\left(5\right)$

Wherein ρ is a constant, and N is the quantity parameter of sampled point;

6), less or do not contain the fingerprint image of node for containing number of nodes, provide field of direction O (T) and the O (Q) of fingerprint image T and Q, O (T)={ O (W ^T(i, j)) | 0≤i＜m/w, 0≤j＜n/w}, O (Q)={ O (W ^Q(i, j)) | 0≤i＜m/w, 0≤j＜n/w}; With the direction stipulations to [0, π), for background piece W (i, j), its direction is meaningless, puts O (W (i, j))=-1; If W (i j) is foreground blocks, then O (W (i, j)) ∈ [and 0, π);

7), field of direction O (T) and the O (Q) of fingerprint image T and Q alignd, obtain:

$O^{\#}\left(Q\right)=\{O\left(W^{Q^{\#}}(i,j)\right)|0\&le;i<m/w,0\&le;j<n/w\},$ Calculate O (T) and O ^#(Q) matching degree c, computing formula is:

$c=\frac{\underset{0\&le;i<m/w,0\&le;j<n/w}{\mathrm{\&Sigma;}}d(i,j)}{\underset{0\&le;i<m/w,0\&le;j<n/w}{\mathrm{\&Sigma;}}\mathrm{\&eta;}(i,j)}\&times;100---\left(6\right)$

C is the proportion of the shared overlapping region of direction that differs greatly after the alignment of both direction field;

8), utilization BP neural network is carried out similarity and is calculated the input of neural network above-mentioned 5) node matching mark S _TQWith field of direction matching degree c, be output as the similarity of two fingerprints that participate in coupling; Training stage, for two identical fingerprints, be output as 1, otherwise be output as 0; At cognitive phase, the network output valve is between 0 and 1, and similarity is big more, and then two fingerprints are similar more.

2, a kind of fingerprint image matching method that merges the whole alignment of many reference modes and the field of direction as claimed in claim 1, it is characterized in that: described 1), come node diagnostic is described with local relative direction information, two nodes are mated, by around node, choosing sampled point, calculate the LRO value of each sampled point, LRO is calculated as from node along straight line to the accumulative total direction difference SOD of sampled point; If (i, j) ((i j)＜π) is that (the local relative direction that centers on node is expressed as for i, the ridge orientation of j) locating at point to 0≤O to O $D=\&lang;D_{\mathrm{\&alpha;}},D_{\mathrm{\&beta;}}\&rang;,$ Wherein $D_{\mathrm{\&alpha;}}=\left\{{\mathrm{\&alpha;}}_i{|}_{i=-N}^N\right\},$ $D_{\mathrm{\&beta;}}=\left\{{\mathrm{\&beta;}}_i{|}_{i=-N}^N\right\},$ α ₀=0, β ₀=0; α _iAnd β _iBe respectively at an A _iAnd B _iLocate the SOD value of relative node; Suppose that node coordinate is for (x, y), direction is γ (0≤γ＜2 π), so

For from (x is y) to A _iSOD, A _i=(x+ilcos (γ), y+ilsin (γ)); Use starting condition α ₀=0, i is from 1 to N, α _i| i〉0 calculate by step (1.1) and (1.2) iteration:

(1.1), utilize known quantity α _I-1Calculate σ _i, computing formula is:

O(A _i)-(γ+α _i-i/|i|)＝2kπ+σ _i，k∈Z，0≤σ _i<2π (7)

(1.2), utilize known quantity α _I-1And σ _iCalculate α _i, computing formula is:

${\mathrm{\&alpha;}}_i=\left\{\begin{array}{cc}{\mathrm{\&alpha;}}_{i-i/\left|i\right|}+{\mathrm{\&sigma;}}_i& 0\&le;{\mathrm{\&sigma;}}_i\&le;2\mathrm{\&pi;}\\ {\mathrm{\&alpha;}}_{i-i/\left|i\right|}+{\mathrm{\&sigma;}}_i-\mathrm{\&pi;}& \mathrm{\&pi;}/2<{\mathrm{\&sigma;}}_i\&le;3\mathrm{\&pi;}/2\\ {\mathrm{\&alpha;}}_{i-i/\left|i\right|}+{\mathrm{\&sigma;}}_i-2\mathrm{\&pi;}& 3\mathrm{\&pi;}/2<{\mathrm{\&sigma;}}_i\&le;2\mathrm{\&pi;}\end{array}\right..---\left(8\right)$

I is from-1 to-N, α _i| step (1.3) and (1.4) iterative computation are passed through in i＜0:

(1.3), by formula (7), utilize known quantity α _I+1Calculate σ _i

(1.4), by formula (8), utilize known quantity α _I+1And σ _iCalculate α _i

(1.5), be similar to

Computation process,

For from (x is y) to B _iSOD, B _i=(x+ilcos (γ-2 π), y+il sin (γ-2 π)); Use starting condition β ₀=0,

Calculate by formula (9) and (10) iteration:

O(B _i)-(γ+β _i-i/|i|)＝2kπ+σ _i，k∈Z，0≤σ _i<2π (9)

${\mathrm{\&beta;}}_i=\left\{\begin{array}{cc}{\mathrm{\&beta;}}_{i-i/\left|i\right|}+{\mathrm{\&sigma;}}_i& 0\&le;{\mathrm{\&sigma;}}_i\&le;2\mathrm{\&pi;}\\ {\mathrm{\&beta;}}_{i-i/\left|i\right|}+{\mathrm{\&sigma;}}_i-\mathrm{\&pi;}& \mathrm{\&pi;}/2<{\mathrm{\&sigma;}}_i\&le;3\mathrm{\&pi;}/2\\ {\mathrm{\&beta;}}_{i-i/\left|i\right|}+{\mathrm{\&sigma;}}_i-2\mathrm{\&pi;}& 3\mathrm{\&pi;}/2<{\mathrm{\&sigma;}}_i\&le;2\mathrm{\&pi;}\end{array}\right..---\left(10\right)$

3, a kind of fingerprint image matching method that merges the whole alignment of many reference modes and the field of direction as claimed in claim 1 or 2 is characterized in that: described 3) in, the whole alignment of many reference modes is defined as follows: establish $\{\&lang;x_i^A,y_i^A,{\mathrm{\&gamma;}}_i^A\&rang;|1\&le;i\&le;C\}$ With $\{\&lang;x_i^B,y_i^B,{\mathrm{\&gamma;}}_i^B\&rang;|1\&le;i\&le;C\}$ Be respectively the subclass of the node set of fingerprint image A and B, if ${\mathrm{\&Sigma;}}_{i=1}^C(x_i^A-x_i^B)=0,$ ${\mathrm{\&Sigma;}}_{i=1}^C(y_i^A-y_i^B)=0,$ ${\mathrm{\&Sigma;}}_{i=1}^C({\mathrm{\&gamma;}}_i^A-{\mathrm{\&gamma;}}_i^B)=0,$ Then claim $\{\&lang;x_i^A,y_i^A,{\mathrm{\&gamma;}}_i^A\&rang;|1\&le;i\&le;C\}$ With $\{\&lang;x_i^B,y_i^B,{\mathrm{\&gamma;}}_i^B\&rang;|1\&le;i\&le;C\}$ Be whole the alignment; If function f makes $f(\&lang;x_i^B,y_i^B,{\mathrm{\&gamma;}}_i^B\&rang;)=\&lang;x_i^{\&prime;B},y_i^{\&prime;B},{\mathrm{\&gamma;}}_i^{\&prime;B}\&rang;,$ And $\{\&lang;x_i^A,y_i^A,{\mathrm{\&gamma;}}_i^A\&rang;|1\&le;i\&le;C\}$ With $\{\&lang;x_i^{\&prime;B},y_i^{\&prime;B},{\mathrm{\&gamma;}}_i^{\&prime;B}\&rang;|1\&le;i\&le;C\}$ Be whole the alignment, then claim f to make $\{\&lang;x_i^A,y_i^A,{\mathrm{\&gamma;}}_i^A\&rang;|1\&le;i\&le;C\}$ With $\{\&lang;x_i^B,y_i^B,{\mathrm{\&gamma;}}_i^B\&rang;|1\&le;i\&le;C\}$ The function of whole alignment;

For similar being defined as follows of the whole alignment of many reference modes under the polar form: establish

With

Be respectively the subclass of node set of the polar form of fingerprint image A and B, if

Then claim

With

Be whole the alignment; If function f makes

And

With Be whole the alignment, then claim f to make

With

The function of whole alignment;

Described 2) in calculated many to reference mode $\{\&lang;x_{m_i}^T,y_{m_i}^T,{\mathrm{\&gamma;}}_{m_i}^T\&rang;|1\&le;i\&le;C\}$ With $\{\&lang;x_{n_j}^Q,y_{n_j}^Q,{\mathrm{\&gamma;}}_{n_j}^Q\&rang;|1\&le;j\&le;C\},$ Wherein

With

Be that matched node is right, corresponding to polar coordinates

With

In

With

Based on many reference modes alignment be with $\{\&lang;x_{m_i}^T,y_{m_i}^T,{\mathrm{\&gamma;}}_{m_i}^T\&rang;|1\&le;i\&le;C\}$ With $\{\&lang;x_{n_j}^Q,y_{n_j}^Q,{\mathrm{\&gamma;}}_{n_j}^Q\&rang;|1\&le;j\&le;C\}$ Carry out integral body alignment, promptly will construct and make

With

The process of whole alignment; It is right that this process is decomposed into

Be rotated and translation the rotation parameter-Δ of calculating _ψAnd translation parameters (Δ _x,-Δ _y); Concrete steps are as follows:

(3.1), calculate

With

Direction difference, computing formula is:

${\mathrm{\&Delta;}}_{\mathrm{\&psi;}}=\frac{\underset{i=1}{\overset{C}{\mathrm{\&Sigma;}}}({\mathrm{\&psi;}}_{n_i}^Q-{\mathrm{\&psi;}}_{m_i}^T)}{C}---\left(11\right)$

(3.2), will

Rotation-Δ _ψObtain

Calculate again its with Position difference, computing formula is:

(3.3) ,-Δ _ψ(Δ _x,-Δ _y) be rotation parameter and translation parameters.Will

Rotation-Δ _ψ, translation (Δ then _x,-Δ _y), obtain

Then

With

Be whole the alignment;

(3.4), will

(Q) rotation-Δ _ψ, translation (Δ then _x,-Δ _y), obtain

Then fingerprint image T and Q have obtained alignment based on multiple reference points.

4, a kind of fingerprint image matching method that merges the whole alignment of many reference modes and the field of direction as claimed in claim 3 is characterized in that: described 4) in, the final matched node logarithm N of fingerprint image T and Q _pFind the solution because

With

According to the Δ sort ascending, the left hand node comes right hand node front; If j _ROFor (Q) first right hand node location, j _LSatisfy ${\mathrm{\&Delta;}}_{j_L}^Q-{\mathrm{\&Delta;}}_{j_{\mathrm{RO}}}^Q\&GreaterEqual;-e$ And ${\mathrm{\&Delta;}}_{j_L-1}^Q-{\mathrm{\&Delta;}}_{j_{\mathrm{RO}}}^Q<-e,$ E is a threshold value; N then _pCalculating according to following algorithm:

Step1. initialization N _p=0, i=0, j=0;

If Step2.i=i+1 is i〉p then returns N _p

Step3. if

In i node be right hand node, j=j then _L-1;

Step4.j＝j+1；

Step5. if ${\mathrm{\&Delta;}}_j^Q-{\mathrm{\&Delta;}}_i^T<-e,$ Then forward Step4 to;

Step6. if ${\mathrm{\&Delta;}}_j^Q-{\mathrm{\&Delta;}}_i^T>-e,$ Then j=0 forwards Step2 to;

Step7. if p (i, j)=1, N then _p=N _p+ 1, j=0 forwards Step2 to;

If Step8.j=j+1 is j〉q, then j=0 forwards Step2 to, otherwise forwards Step6 to.

5, a kind of fingerprint image matching method that merges the whole alignment of many reference modes and the field of direction as claimed in claim 4 is characterized in that: described 7) in, to O (T) and O ^#(Q) matching degree c calculates, and computing method are:

Step1. calculate W ^T(i, j) with

Direction poor, computing formula is:

Step2. direction difference stipulations are arrived [0, pi/2]:

$\mathrm{\&sigma;}(i,j)=\left\{\begin{array}{cc}\mathrm{\&pi;}-\mathrm{\&sigma;}\&prime;(i,j)& \mathrm{\&sigma;}\&prime;(i,j)>\mathrm{\&pi;}/2\\ \mathrm{\&sigma;}\&prime;(i,j)& 0\&le;\mathrm{\&sigma;}\&prime;(i,j)\&le;\mathrm{\&pi;}/2\\ -1& \mathrm{\&sigma;}\&prime;(i,j)<0\end{array}\right.---\left(15\right)$

Step3. the stipulations result to the direction difference classifies according to threshold value t:

Wherein the t in the formula (16) is a threshold value; In the formula (14) Computation process as follows:

u＝i×ω+ω/2，v＝j×ω+ω/2 (18)

$\mathrm{\&gamma;}=\sqrt{{(u-x_{r_1}^T)}^2+{(v-y_{r_1}^T)}^2}---\left(19\right)$

$\mathrm{\&alpha;}\&prime;=\arctan ((v-y_{r_1}^T)/(u-x_{r_1}^T))---\left(20\right)$

$\mathrm{\&alpha;}=\left\{\begin{array}{cc}\mathrm{\&pi;}/2& u=x_{r_1}^T,v>y_{r_1}^T\\ 0& u=x_{r_1}^T,v=y_{r_1}^T\\ -\mathrm{\&pi;}/2& u=x_{r_1}^T,v<y_{r_1}^T\\ \mathrm{\&alpha;}\&prime;& u>x_{r_1}^T\\ \mathrm{\&alpha;}\&prime;+\mathrm{\&pi;}& u<x_{r_1}^T\end{array}\right.---\left(21\right)$

$x=x_{r_2}^Q+\mathrm{\&gamma;}\&times;\cos (\mathrm{\&alpha;}-{\mathrm{\&theta;}}_{r_1}^T+{\mathrm{\&theta;}}_{r_2}^Q)---\left(22\right)$

$y=y_{r_2}^Q+\mathrm{\&gamma;}\&times;\sin (\mathrm{\&alpha;}-{\mathrm{\&theta;}}_{r_1}^T+{\mathrm{\&theta;}}_{r_2}^Q)---\left(23\right)$

Wherein, O (

) 〉=0 expression WQ (

) be prospect.

6, a kind of fingerprint image matching method that merges the whole alignment of many reference modes and the field of direction as claimed in claim 5 is characterized in that: described 8) in, neural network judges whether identical differentiation feature specifically describes as follows two fingerprint images:

(a) based on the coupling mark S under many reference mode alignment _TQ

(b) alignment back matched node logarithm N _p

(c) overlapping region, alignment back area;

(d) the image block number of direction difference between [0 °, 5 °] in the overlapping region;

(e) in the overlapping region direction difference (5 °, 10 °] between the image block number;

(f) in the overlapping region direction difference (10 °, 20 °] between the image block number.

Images