CN105787451A - Fingerprint matching method based on multi-judgment point mode - Google Patents

Abstract

The invention provides a fingerprint matching method based on multi-judgment point mode. The method uses the central point of the fingerprint image to construct a local detail structure, uses the principle of similar triangles to obtain the reference point on the structure, and uses the variable bounding box The method performs two-level matching, and finally uses the multi-condition judgment method to select the threshold for final judgment. Using the method of the present invention can basically accurately locate the matching reference point, accurately obtain the transformation parameters, solve the problem of rotation and translation of the fingerprint image to be recognized relative to the template fingerprint image, and effectively perform image matching. Compared with the general point pattern In the matching method, the matching speed of the present invention is faster, FAR and FRR are reduced, and the correct rate of fingerprint identification is improved.

Description

A Fingerprint Matching Method Based on Multiple Decision Point Patterns

technical field

The invention relates to a fingerprint matching method based on a multi-judgment point mode, and belongs to the technical field of image processing.

Background technique

The ultimate goal of fingerprint identification is to determine whether two fingerprints come from the same finger. To complete fingerprint identification, it is necessary to match the features extracted from the current input fingerprint with the template features saved in advance, so as to achieve the ultimate goal of fingerprint identification, so the matching algorithm is related to the success of fingerprint identification, which is a key step in fingerprint identification. Fingerprint matching is the last step in the fingerprint identification system, and it is also the most important basis for evaluating the performance of the entire fingerprint identification system. Fingerprint matching is based on the extracted fingerprint features to determine whether two fingerprints come from the same finger. When the quality of the fingerprint image is good, the result of the matching algorithm is usually better. However, when the fingerprint image in the prior art is read, there will be translation, rotation and nonlinear deformation, the dryness and wetness of the finger surface, and the difference in fingerprint collection equipment, all of which affect the effect of fingerprint feature extraction, and then affect the fingerprint matching. result. Therefore, the key issue of fingerprint matching is how to correctly match and identify fingerprints when the quality of fingerprint images is not good. In order to accurately and quickly determine whether two fingerprints come from the same finger, the fingerprint matching algorithm must also have a certain degree of fault tolerance, and the computational complexity should not be too high, and the time cost should be small and the accuracy should be high.

According to different fingerprint details, fingerprint matching algorithms mainly include point-based pattern matching algorithms, texture-based pattern matching algorithms and graph-based matching algorithms. Texture-based matching can overcome the shortcomings of minutiae-based methods, and as a new matching idea is being paid attention to and applied. The method of texture matching makes full use of the rich ridges, which can overcome the difficulty of extracting the details of poor quality areas to a certain extent. In some application fields, it can make up for the defects of minutiae point matching. However, since this method needs to perform multiple convolutions on the image, the amount of calculation is very large, and it is difficult to deal with the matching of large deformed fingerprint images, so it is not suitable for the recognition system of 1:N mode. Graph matching is a method for structural pattern recognition that can be applied to classification, minutiae indexing and matching of fingerprints. Many scholars have proposed fingerprint feature matching based on structural information, using the topological structure information in the fingerprint image to overcome the interference of fingerprint image noise, rotation and deformation on recognition.

The above classification methods are not absolute, and various methods are interrelated. At the same time, there are many algorithms that cross each other. Each algorithm has its own characteristics and is aimed at special applications. For example, the graph matching method has better anti-interference ability to noise of poor-quality fingerprints, but the existing methods have not been confirmed by large-scale experiments; minutiae matching is accurate for good-quality fingerprint images, but the distinguishability of texture features Not strong; the mixed matching method based on texture information and string matching can improve the recognition rate to a certain extent, but the calculation cost is very high; the mixed matching method based on triangle matching and dynamic programming can solve the problem of nonlinear transformation, but the extracted The feature is too large to meet the requirements of online use.

In fact, it is very difficult to determine whether two fingerprints come from the same finger. First of all, when collecting fingerprints, it is impossible to make strict restrictions on the contact position and direction of the finger and the surface of the collection device, the pressure of the finger and the direction of the pressing force, etc., which will not only make the image area collected by the same finger at different times incomplete. The same, and there are inevitably translation deformation, rotation deformation, scale deformation and nonlinear deformation between images. Secondly, when the image quality is poor, the minutiae extraction process will produce many errors, including the generation of false minutiae, the omission of real minutiae, and the deviation of minutiae position and direction. These factors cause even the fingerprint images representing the same finger, in which the number, position, direction, etc. of the minutiae points are not exactly the same, making the fingerprint minutiae matching problem very difficult. And the present invention can well solve the above problems.

Contents of the invention

The purpose of the present invention is to solve the deficiencies of the above-mentioned prior art, and propose a fingerprint matching method based on a multi-decision point mode, which uses the central point of the fingerprint image to construct a local detail structure, and uses the similar triangle principle to obtain Take the reference point, and use the variable bounding box method to perform secondary matching, and finally use the multi-condition judgment method to select the threshold for the final judgment.

The technical solution adopted by the present invention to solve the technical problem is: a fingerprint matching method based on a multi-decision point mode, which can accurately locate the matching reference point, accurately obtain the transformation parameters, and solve the problem of relative fingerprint images to be recognized. Due to the problem of rotation and translation of the template fingerprint image, image matching is carried out effectively. Compared with the general point pattern matching method, the matching speed of the present invention is faster, FAR and FRR are reduced, and the correct rate of fingerprint recognition is improved.

Method flow:

Step 1: use the principle of similar triangles to obtain the reference point, and obtain the position deviation and rotation transformation factor in the X and Y coordinate directions according to the reference point pair in the input fingerprint image and the template fingerprint image, so as to determine the transformation factor set;

Step 2: Perform first-level matching on the input fingerprint image and the template fingerprint image, respectively calculate the number of bifurcation points and endpoints of the interest area of the input fingerprint image and the template fingerprint image, and determine whether to enter the next level of matching;

Step 3: Use the variable bounding box method for secondary matching, select the rectangular area around the feature point as the bounding box, as long as the transformed feature points of the fingerprint to be recognized fall in this area, the direction is the same and the type is the same, then it is considered is the matching point;

Step 4: Set thresholds such as the number of successfully matched point pairs, the ratio of paired points to the total number of corresponding fingerprint features, and the sum of difference scores of each matching point pair to finally determine whether the input fingerprint image matches the template fingerprint image.

The present invention uses the basic properties of triangles to determine the reference point and transformation factor; performs secondary matching and introduces the concept of a variable-sized bounding box. The size of the bounding box is determined by the distance between the current feature point and the center point, as long as the transformation If the feature points of the final fingerprint to be identified fall in this area, and the type is the same, and the direction is basically the same, then the two feature point pairs can be considered as a pair of matching feature points.

Beneficial effect:

1. The present invention can accurately locate matching reference points, accurately obtain transformation parameters, and effectively perform image matching, compared with general point pattern matching methods.

2. The matching speed of the present invention is fast, FAR and FRR are reduced, and the correct rate of fingerprint identification is well improved.

3. The positioning of the reference point determined by the present invention is relatively accurate, and the time consumption is shortened, which greatly improves the recognition rate and execution efficiency.

Description of drawings

Fig. 1 is a schematic diagram of a variable bounding box in the second-level matching of the present invention.

Fig. 2 is a flow chart of the method of the present invention.

detailed description

The invention will be described in further detail below in conjunction with the accompanying drawings.

1. Calculation of datum point and conversion factor

As shown in Fig. 1 and Fig. 2, the present invention determines the reference point and calculates the transformation parameters according to the relationship among the three neighboring feature points. The matching of two fingerprint images is mainly to solve the problems of rotation, translation and deformation. It is assumed that the input of fingerprint matching is a set of two feature points P and Q, P is extracted from the input fingerprint image, another click on Q will From the fingerprint feature database, that is, extracted from the template fingerprint image. These two point sets are represented as:

P={p ₁ ,p ₂ ,...,p _m }={(x _p1 ,y _p1 ,θ _p1 ,T _p1 ),(x _p2 ,y _p2 ,θ _p2 ,T _p2 ),..., (x _pm ,y _pm ,θ _pm ,T _pm )}

Q＝{q ₁ ,q ₂ ,...,q _n }＝{(x _q1 ,y _q1 ,θ _q1 ,T _q1 ),(x _q2 ,y _q2 ,θ _q2 ,T _q2 ),..., (x _qn ,y _qn ,θ _qn ,T _qn )}

Among them (x _pi , y _pi , θ _pi , T _pi ) and (x _qj , y _qj , θ _qj , T _qj ) respectively record the i-th feature point in point set P and the j-th feature point in point set Q 4 pieces of information: X coordinate, Y coordinate, direction and feature point type. If the two fingerprint images are completely matched, the template feature point set can be obtained by performing some transformation (rotation, translation, and stretching) on the input fingerprint feature point set. Therefore, the point set P can be approximated as Point set Q. However, in practical applications, the fingerprint areas collected twice may not be exactly the same, and due to deformation, noise and other reasons, the positions of some detailed feature points have a certain offset, and some feature points are added or deleted, so it is necessary to find A transformation that matches as many points in two point sets as possible. If under a certain transformation condition, two feature points are close in position, basically in the same direction, and of the same type, it is considered that the two points form a match under the transformation condition.

In order to convert a feature point in the input fingerprint image into a corresponding position in the template fingerprint image according to a certain transformation method, it is necessary to know the corresponding transformation factor. Since all fingerprint images are recorded by the same fingerprint collector, this paper does not consider the deformation of fingerprint images for the time being, and considers that they are basically unchanged. In addition, the types of feature points should not change before and after transformation, that is,

Assume that the characteristic information of a point p _i in the input point set P is (x _pi , y _pi , θ _pi , T _pi ), after the formula transformation, it is The feature information of the corresponding point q _j in the template point set Q is (x _qj , y _qj , θ _qj , T _qj ). if It is considered that the conversion factor is (Δx, Δy, Δθ), and p _i is similar to q _j .

$\left[\begin{array}{c}{x}_{pi}^T\\ {\mathrm{the\; y}}_{pi}^T\\ {\mathrm{\θ}}_{pi}^T\\ T_{pi}^T\end{array}\right]=\left[\begin{array}{cccc}\cos \mathrm{\Δ}\mathrm{\θ}& -\sin \mathrm{\Δ}\mathrm{\θ}& 0& \mathrm{\Δ}x\\ \sin \mathrm{\Δ}\mathrm{\θ}& \cos \mathrm{\Δ}\mathrm{\θ}& 0& \mathrm{\Δ}\mathrm{the\; y}\\ 0& 0& 1& \mathrm{\Δ}\mathrm{\θ}\\ 0& 0& 0& 1\end{array}\right]\left[\begin{array}{c}{x}_{pi}\\ {\mathrm{the\; y}}_{pi}\\ {\mathrm{\θ}}_{pi}\\ T_{pi}\end{array}\right]$ Formula 1

In the above formula, Δx and Δy are the parallel factors in the X and Y directions respectively, and Δθ is the rotation factor. In order to match two fingerprints accurately, these three transformation factors need to be determined.

Any feature point p _i in the input point set and any feature point q _j in the template point set form a point pair. Find the feature point closest to p _i in the input fingerprint image, denoted as p ₁ , find the feature point closest to p _i , denoted as p ₂ , similarly, find q ₁ , q ₂ . In this way, two triangles (p _i , p ₁ , p ₂ ) and (q _j , q ₁ , q ₂ ) are formed in the input fingerprint image and the template fingerprint image. Judging the degree of similarity between two triangles, if the degree of similarity is high, it is a possible matching pair, and the transformation parameters obtained according to the two sets of feature point subsets are the transformation parameters of the two images. Transform (rotate and translate) the two fingerprint images according to the obtained transformation parameters, and further judge the similarity of the two images.

The three sides can determine a unique triangle, so the similarity of the two triangles can be judged according to the distance between the three feature points and their mutual positional relationship.

The procedure for judging the similarity of triangles and obtaining transformation parameters is as follows:

(1) Calculate the side lengths |p ₁ p ₂ | and |q ₁ q ₂ | corresponding to vertices p _i and q _j respectively;

(2) If ||p ₁ p ₂ |-|q ₁ q ₂ ||>D ₁ , it means that the triangles cannot be congruent. This judgment is over, and the vertices p _i and q _j to be judged and the distance between them are selected again The nearest two feature points p ₁ , p ₂ and q ₁ , q ₂ return to step (1);

(3) Otherwise, calculate the distances from p _i and q _j to p ₁ , p ₂ and q ₁ , q ₂ respectively |p _i p ₁ |, |p _i p ₂ | and |q _j q ₁ |, |q _j q ₂ |. If there is ||p _i p ₁ |-|p _i p ₂ ||<=D ₂ and ||q _j q ₁ |-|q _j q ₂ ||<=D ₂ , then the three sides of the two triangles are respectively Approximately equal means that the two triangles are approximately congruent. Otherwise, reselect the vertices p _i and q _j that need to be judged and the two nearest feature points p ₁ , p ₂ and q ₁ , q ₂ , and return to step (1);

(4) According to the vertices corresponding to the two triangles, respectively calculate the direction difference between the possible matching feature points Direction difference calculation formula:

Formula 2

If the angle differences between corresponding vertices are approximately equal, that is Then it is considered that the angle between the two feature point subsets (p _i , p ₁ , p ₂ ) and (q _j , q ₁ , q ₂ ) satisfies a rotation transformation relationship, and the calculation formula of the rotation transformation factor at this time is as follows:

$\mathrm{\&Delta;}\mathrm{\&theta;}=\frac{1}{3}({\mathrm{\&Delta;\&theta;}}_{p_i{q}_j}+{\mathrm{\&Delta;\&theta;}}_{p_1{q}_1}+{\mathrm{\&Delta;\&theta;}}_{p_2{q}_2})$ Formula 3

Otherwise, it is determined that a matching relationship cannot be formed between the two feature subsets, reselect the vertices p _i and q _j that need to be judged and the two feature points p ₁ , p ₂ and q ₁ , q ₂ closest to them, and return step 1);

(5) Select (p _i , q _j ) as the transformation origin of the rotation transformation, perform rotation transformation on (q _j ,q ₁ ,q ₂ ), and the transformed value is (q _j ,q′ ₁ ,q′ ₂ ) , and calculate the position deviation in the X and Y directions after the rotation transformation respectively Its calculation formula is as follows:

Δx _pq ＝x _p -x _q Formula 4

Δy _pq = y _p -y _q Formula 5

At this time, if there is and Then it is considered that the two feature point subsets (p _i , p ₁ , p ₂ ) and (q _j , q ₁ , q ₂ ) also satisfy a transformation in the X and Y directions. It is considered that the two feature point subsets (p _i , p ₁ , p ₂ ) and (q _j , q ₁ , q ₂ ) form a matching relationship, and the transformation factors of translation and rotation at this time are ( Δx, Δy, Δθ). Among them, Δx, Δy, Δθ are respectively:

$\mathrm{\&Delta;}x=\frac{1}{3}({\mathrm{\&Delta;x}}_{p_i{q}_j}+{\mathrm{\&Delta;x}}_{p_1{q}_1^{\&prime;}}+{\mathrm{\&Delta;x}}_{p_2{q}_2^{\&prime;}})$ Formula 6

$\mathrm{\&Delta;}\mathrm{the\; y}=\frac{1}{3}({\mathrm{\&Delta;y}}_{p_i{q}_j}+{\mathrm{\&Delta;y}}_{p_1{q}_1^{\&prime;}}+{\mathrm{\&Delta;y}}_{p_2{q}_2^{\&prime;}})$ Formula 7

$\mathrm{\&Delta;}\mathrm{\&theta;}=\frac{1}{3}({\mathrm{\&Delta;\&theta;}}_{p_i{q}_j}+{\mathrm{\&Delta;\&theta;}}_{p_1{q}_1}+{\mathrm{\&Delta;\&theta;}}_{p_2{q}_2})$ Formula 8

Second, the first level of matching

With the center of the reference point and the fixed length as the radius, an equal region of interest (ROI) is drawn on the template fingerprint image and the input fingerprint image as the first-level matching interval. Select the role of ROI: first, it can resist fingerprint deformation; second, it can facilitate matching.

(1) Select the ROI radius as 96, and the unit is pixel.

(2) Search for the number of endpoints and bifurcation points in the ROI, record the number of bifurcation points and the number of endpoints of the template fingerprint image as M _b , M _e , and the number of feature points of the input fingerprint image as N _b , _Ne .

(3) Calculate |M _b -N _b |, |M _e -N _e |:

If |M _b -N _b |<ε ₁ and |M _e -N _e |<ε ₂ , enter the next level of matching. Otherwise, it is directly considered as a mismatch. Wherein ε ₁ =0.5 min(M _b ,N _b ), ε ₂ =0.5 min(M _e ,N _e ).

Third, the second level of matching

In order to overcome the influence of nonlinear deformation, the concept of variable-sized bounding box is introduced. For each feature point in the template fingerprint feature point set, a rectangular area around it is selected as its bounding box. As shown in Figure 1, it can be seen from the figure that the size of the bounding box is determined by the distance between the current feature point and the center point. The polar radius should be smaller and the polar angle should be larger near the center point; on the contrary, at a distance from the center point Where the center point is far away, the polar radius should be larger and the polar angle should be smaller. As long as the transformed feature points of the fingerprint to be recognized fall in this area, and have the same type and basically the same direction, the two feature point pairs can be considered as a pair of matching feature points.

4. Judgment method based on composite mode and multiple conditions

When the present invention matches and recognizes, the following three aspects can be expanded as judgment conditions, including: 1) the number of successfully matched point pairs; 2) the ratio of the number of paired points to the total number of corresponding fingerprint features; 3) the sum of the difference scores of each matched point pair , that is (the weighted sum of direction difference and distance difference), shows that the lower the score, the higher the matching degree.

The present invention improves the general point pattern fingerprint matching method, constructs the local detail structure of the fingerprint image, and uses the triangular similarity principle to obtain the reference point and transformation factor on the structure, introduces the method of bounding box for matching, and finally uses the multi-condition The decision defines the recognition threshold. Compared with the general point pattern fingerprint matching method, the reference point positioning determined by the present invention is more accurate, and the time consumption is shortened, and the recognition rate and execution efficiency can be improved.

Claims (5)

1. the finger print matching method based on many determination points pattern, it is characterised in that described method comprises the steps:

Step 1: utilize similar triangle theory to ask for datum mark, according to datum mark in input fingerprint image and template fingerprint image to asking for X, the position deviation in Y coordinate direction and the rotation transformation factor, so that it is determined that transformation factor collection；

Step 2: input fingerprint image and template fingerprint image are carried out first order coupling, calculates bifurcation and the end points number in input fingerprint image and template fingerprint interesting image district respectively, it is determined whether enter next stage coupling；

Step 3: utilize the method for variable gauge box to carry out second degree matches, the rectangular area around selected characteristic point is as gauge box, as long as the characteristic point of the fingerprint to be identified after conversion drops in this region, direction is consistent and type is identical, then it is assumed that be match point；

Step 4: set the some logarithm of successful match, pairing is counted and the threshold value such as discrepancy score summation of the ratio of corresponding fingerprint characteristic sum, each matching double points finally judges whether input fingerprint image mates with template fingerprint image.

2. a kind of finger print matching method based on many determination points pattern according to claim 1, it is characterised in that described method is to adopt the fundamental property of triangle, it is determined that datum mark and transformation factor；Carry out second degree matches, introduce the concept of the gauge box of variable-size, the size of gauge box is determined by the distance between current signature point and central point, as long as the characteristic point of the fingerprint to be identified after conversion drops in this region, and type is identical, direction is basically identical, then it is believed that the two feature point pairs is the characteristic point of a pair coupling.

3. a kind of finger print matching method based on many determination points pattern according to claim 1 and 2, it is characterized in that, described method determines datum mark according to the mutual relation between three neighbour's characteristic points, ask for transformation parameter, the problems such as the coupling of two width fingerprint images mainly solves to rotate, translation and deformation, assume that the input of fingerprint matching is set P and the Q of two characteristic points, P extracts from the fingerprint image of input, another clicks Q then from fingerprint characteristic data storehouse, namely extracting from template fingerprint image, the two point set is expressed as:

P={p₁,p₂,...,p_m}={ (x_p1,y_p1,θ_p1,T_p1),(x_p2,y_p2,θ_p2,T_p2),...,(x_pm,y_pm,θ_pm,T_pm)}

Q={q₁,q₂,...,q_n}={ (x_q1,y_q1,θ_q1,T_q1),(x_q2,y_q2,θ_q2,T_q2),...,(x_qn,y_qn,θ_qn,T_qn)}

Wherein (x_pi,y_pi,θ_pi,T_pi) and (x_qj,y_qj,θ_qj,T_qjnull) have recorded in point set P 4 information of jth characteristic point in ith feature point and point set Q respectively: X-coordinate、Y coordinate、Direction and characteristic point type，If two width fingerprint images mate completely，Then can by the fingerprint characteristic point set of input being done certain conversion，Namely (rotate、Translation is with flexible) obtain template characteristic point set，Therefore，Point set P can pass through to rotate、Translation and stretching are approximated to point set Q，But in practical application，The finger-print region of twice collection can not be completely the same，And due to deformation、The reasons such as noise，There is certain skew the position of some of which details，Also some characteristic point is added or removed，So need to find the some coupling as much as possible that a kind of conversion makes two points concentrate，If when certain conversion，Two characteristic point positions are close、Direction is basically identical，And type is identical，Then think that the two point defines when this conversion once to mate.

4. a kind of finger print matching method based on many determination points pattern according to claim 1 and 2, it is characterised in that the similarity of the triangle of described method judges and the step of asking for of transformation parameter includes:

(1) summit p is calculated respectively_iAnd q_jThe corresponding length of side | p₁p₂| with | q₁q₂|, specification error value D₁And D₂；

(2) if | | p₁p₂|-|q₁q₂||>D₁, then illustrating that triangle can not be congruent, this judgement terminates, and reselects the summit p needing to judge_iAnd q_jAnd apart from they 2 nearest characteristic point p₁, p₂And q₁, q₂, return step (1)；

(3) otherwise, p is calculated respectively_iAnd q_jTo p₁, p₂And q₁, q₂Distance | p_ip₁|, | p_ip₂| with | q_jq₁|, | q_jq₂|, if had | | p_ip₁|-|p_ip₂| |≤D₂And | | q_jq₁|-|q_jq₂| |≤D₂, then 3 limits of two trianglees approximately equal respectively, the approximate congruence of 2 trianglees is described, otherwise reselects the summit p needing to judge_iAnd q_jAnd apart from they 2 nearest characteristic point p₁, p₂And q₁, q₂, return step (1)；

(4) according to the summit that 2 trianglees are corresponding, the direction difference between the characteristic point being likely to coupling is calculated respectively Direction mathematic interpolation formula:

Formula 2

If the angle difference approximately equal between corresponding vertex, namelyThen think these 2 characteristic point subset (p_i,p₁,p₂) and (q_j,q₁,q₂) between angle meet a kind of rotation transformation relation, the computing formula of the rotation transformation factor now is as follows:

$\mathrm{\&Delta;}\mathrm{\&theta;}=\frac{1}{3}({\mathrm{\&Delta;\&theta;}}_{p_i{q}_j}+{\mathrm{\&Delta;\&theta;}}_{p_1{q}_1}+{\mathrm{\&Delta;\&theta;}}_{p_2{q}_2})$ Formula 3

Otherwise judge between these 2 character subsets, to form a kind of matching relationship, reselect the summit p needing to judge_iAnd q_jAnd apart from they 2 nearest characteristic point p₁, p₂And q₁, q₂, return step (1)；

(5) (p is chosen_i,q_j) as the conversion initial point of rotation transformation, to (q_j,q₁,q₂) do rotation transformation, the value after conversion is (q_j,q′₁,q'₂), and calculate the position deviation in rotation transformation X, Y-direction later respectively Its computing formula is as follows:

Δx_pq=x_p-x_qFormula 4

Δy_pq=y_p-y_qFormula 5

Now, if had ${\mathrm{\&Delta;x}}_{p_i{q}_j}\&ap;{\mathrm{\&Delta;x}}_{p_1{q}_1^{\&prime;}}\&ap;{\mathrm{\&Delta;x}}_{p_2{q}_2^{\&prime;}}$ And ${\mathrm{\&Delta;y}}_{p_i{q}_j}\&ap;{\mathrm{\&Delta;y}}_{p_1{q}_1^{\&prime;}}\&ap;{\mathrm{\&Delta;y}}_{p_2{q}_2^{\&prime;}},$ Then think these 2 characteristic point subset (p_i,p₁,p₂) and (q_j,q₁,q₂) also meet a kind of conversion in the x, y direction, it is considered as this 2 characteristic point subset (p_i,p₁,p₂) and (q_j,q₁,q₂) between define a kind of matching relationship, translation now and the transformation factor of rotation respectively (Δ x, Δ y, Δ θ), wherein, Δ x, Δ y, Δ θ is respectively as follows:

$\mathrm{\&Delta;}x=\frac{1}{3}({\mathrm{\&Delta;x}}_{p_i{q}_j}+{\mathrm{\&Delta;x}}_{p_1{q}_1^{\&prime;}}+{\mathrm{\&Delta;x}}_{p_2{q}_2^{\&prime;}})$ Formula 6

$\mathrm{\&Delta;}y=\frac{1}{3}({\mathrm{\&Delta;y}}_{p_i{q}_j}+{\mathrm{\&Delta;y}}_{p_1{q}_1^{\&prime;}}+{\mathrm{\&Delta;y}}_{p_2{q}_2^{\&prime;}})$ Formula 7

$\mathrm{\&Delta;}\mathrm{\&theta;}=\frac{1}{3}({\mathrm{\&Delta;\&theta;}}_{p_i{q}_j}+{\mathrm{\&Delta;\&theta;}}_{p_1{q}_1}+{\mathrm{\&Delta;\&theta;}}_{p_2{q}_2})$ Formula 8.

5. a kind of finger print matching method based on many determination points pattern according to claim 1, it is characterised in that described method is based on the judgement of composite mode many condition, and described judgment condition includes: the some logarithm of (1) successful match；(2) ratio that pairing is counted and corresponding fingerprint characteristic is total；(3) the discrepancy score summation of each matching double points, i.e. (weighted sum of direction difference and range difference), finally determine whether input fingerprint image mates with template fingerprint image.