KR100834995B1 - Fingerprints matching Apparatus and Method - Google Patents

Abstract

The fingerprint matching device according to the present invention includes a scanner scanning a user fingerprint, a database storing template fingerprint information including fingerprint feature points, feature points of a template fingerprint read from a database, and feature points extracted from fingerprint information input from a scanner. A local neighborhood structure generation unit for generating a local neighborhood structure containing nearest feature point information for each feature point with respect to a local neighborhood structure of a template fingerprint and an input fingerprint generated by the local neighborhood structure generation unit. neighborhood fingerprint) to align the input fingerprint with the template fingerprint, and the template fingerprint and the input using the clustered feature points in the alignment and clustering processing unit, the alignment and clustering processing unit for clustering the feature points matched from the aligned input fingerprint and the template fingerprint Fingerprint Matching Score Between Fingerprints It comprises parts of the matching score calculator for calculating.

Fingerprint matching, feature point, local neighbor structure, alignment, clustering, verdict

Description

Fingerprint matching device and method {Fingerprints matching Apparatus and Method}

1 is a block diagram showing the configuration of a fingerprint matching device according to the present invention;

2A is a diagram illustrating a local neighbor structure generated by the local neighbor structure generating unit according to the present invention;

2B is a diagram illustrating a feature point encoding variable for a k th nearest feature point generated by a local neighbor structure generator according to the present invention;

Figure 3a is an illustration of the alignment result of the alignment and clustering processing unit according to the present invention,

Figure 3b is an illustration of the feature grouping of the alignment and clustering processing unit according to the present invention,

Figure 4 is an exemplary fingerprint modification of the alignment and clustering processing unit according to the present invention,

5 is a graph comparing fingerprint recognition performance in a fingerprint recognition system using a fingerprint matching device according to the present invention;

6 is a fingerprint matching flowchart of a fingerprint matching device according to the present invention;

7 is a flowchart of generating a local neighborhood structure according to the present invention;

8 is a flow chart of alignment and clustering according to the present invention;

9 is a flowchart illustrating feature clustering according to an embodiment of the present invention;

<Brief Description of Drawings>

21,22: fingerprint boundary box 31: deformation pattern

110: scanner 120: template fingerprint DB

130: control unit

  131: local neighbor structure generation unit 133: alignment and clustering processing unit

  135: matching score calculation unit

The present invention relates to a fingerprint recognition technology, and more particularly, to a fingerprint matching device and a method for solving a nonlinear distortion problem to effectively perform fingerprint recognition.

Fingerprint recognition is attracting the most attention as a technology that shows good characteristics evenly in various judgment items such as discrimination, persistence, and performance among various biometrics, and does not give a user a feeling of rejection. In recent years, the field of Automatic Fingerprint Identification System has made great progress. The most important technique to be dealt with in such an automatic fingerprint identification system is fingerprint matching. Fingerprint matching refers to the expression of similarity for two fingerprint images. The similarity should be large for the fingerprint of the same person, and similarity to the fingerprint with other people. ) Should be small.

Although many fingerprint matching theories exist, the fingerprint matching technique has attracted much attention as a minutiae-based method, which is the most used and accurate method. The fingerprint feature point (minutiae) is a term used to collectively refer to a point where a ridge line ends (ridge ending, hereinafter referred to as a disadvantage) and a ridge bifurcation (hereinafter, referred to as a branch point) in a fingerprint pattern. Fingerprint Identification System supports compatibility with standards of the American National Standards Institute (ANSI) and the National Institute of Standards and Technology (NIST), and uses location and orientation information of feature points to use template fingerprint feature points extracted from existing systems. It is required to use only. Accordingly, fingerprint matching can be thought of as a problem of finding the number of pairs of feature points that match, such as simple dot pattern matching.

However, there are several problems that make this simple dot pattern matching problem difficult. The most influential problem is caused by the elasticity of the skin, which causes non-linear distortion in the fingerprint pattern due to changes in the position and pressure of the finger on the fingerprint reader. In fact, human fingers are three-dimensional curvatures, and fingerprint readers are usually flat two-dimensional shapes, so non-linear distortions, whether large or small, always occur when fingerprints are acquired. All Fingerprint Identification Systems must be designed to cope with non-linear distortion issues during the fingerprint matching stage to ensure good performance.

There are various methods of fingerprint matching such as image based method, texture based method and fingerprint feature based method. Many papers suggest various methods to solve this nonlinear distortion problem. Among them, `` On-line fingerprint verification '', written by Jane and two others, samples the ridge line information near the feature point, adds it to the information of the feature point itself, and changes the size of each feature point. We used a method to find the matching pairs in the However, this method basically only considers rigid transformation, so it is not effective for large distortion errors. In addition, since information other than the feature point itself is used, there is a compatibility problem, and the sampled ridge information has a problem of increasing the template capacity. Meanwhile, in Modeling plastic distortion in fingerprint images by Cappelli and two others, he proposed a mathematical deformation model describing nonlinear distortions. However, this model is not suitable for the purpose of the fingerprint matching method, which basically distinguishes between the warped states because it represents the relationship between the warped state and the warped state. In fact, the relationship between two warped surfaces is not easy to describe mathematically. Fingerprint matching by thin-plate spline modeling plastic deformation, written by Bazen and Gerez, used a method of iterative application of thin-plate spline (TPS) models to model warped fingerprint images. . However, this method does not have appropriate constraints to control the iterative deformation process. Although this method can reliably increase the similarity score of the same person's fingerprint matching, the similarity score of matching with another person's fingerprint also increases to a similar distribution, resulting in an increase in the false matched rate (FMR). Recently, written by Ross and two others, Fingerprint warping using ridge curve correspondences suggests using a TPS model based on ridge curve correspondence information. However, this method requires training a deformation pattern for each person, and the averaged trained deformation model also has a limitation in that it does not give a correct result when a large distortion occurs.

SUMMARY OF THE INVENTION The present invention has been proposed in the above background, and an object of the present invention is to provide a fingerprint matching device and method for effectively performing fingerprint recognition by solving a nonlinear distortion problem.

Fingerprint matching device according to an aspect of the present invention, a scanner for scanning a user fingerprint, a database for storing the template fingerprint information including the fingerprint feature point, extracted from the feature points of the template fingerprint read from the database and the fingerprint information input from the scanner A local neighbor structure generation unit for generating a local neighborhood structure including nearest feature point information based on each feature point with respect to the extracted feature points, and a local neighbor of a template fingerprint and an input fingerprint generated by the local neighbor structure generator A template is used by aligning an input fingerprint with a template fingerprint using a local neighborhood structure, and clustering feature points in an alignment and clustering processor and an alignment and clustering processor that cluster the feature points matched by the aligned input fingerprint and the template fingerprint. Fingerprint between fingerprint and input fingerprint And a matching score calculator for calculating the matching score.

The fingerprint matching device of the present invention is implemented to cluster local feature point matching pairs and aggregate information from each cluster to calculate a fingerprint matching score, thereby effectively performing fingerprint recognition while solving the nonlinear distortion problem. To make it work.

According to another aspect of the present invention, there is provided a fingerprint matching method between a template fingerprint and an input fingerprint, including: local neighborhood structure including nearest feature point information based on each feature point of the feature points of the template fingerprint and the feature points of the input fingerprint; ) Aligning the input fingerprint with the template fingerprint using a local neighborhood structure of the generated template fingerprint and the input fingerprint, and matching feature points in the aligned input fingerprint and the template fingerprint. And a step of sorting and clustering, and calculating a fingerprint matching score between the template fingerprint and the input fingerprint using the clustered feature points.

Hereinafter, the present invention will be described in detail with reference to the accompanying drawings so that those skilled in the art can easily understand and reproduce the present invention.

1 is a block diagram illustrating a configuration of a fingerprint matching device according to an embodiment of the present invention, and FIG. 2A is a diagram illustrating a local neighbor structure generated by a local neighbor structure generating unit according to the present invention, and FIG. FIG. 3A is an exemplary view illustrating alignment result of the alignment and clustering processing unit according to the present invention, and FIG. 3B illustrates the alignment and clustering process according to the present invention. 4 is an exemplary view illustrating clustering of feature points in a clustering processing unit, and FIG. 4 is a diagram illustrating fingerprint modification of an alignment and clustering processing unit according to the present invention.

Referring to FIG. 1, the fingerprint matching device according to the present invention includes a scanner 110, a template fingerprint database 120, and a controller 130. The scanner 110 is implemented as a fingerprint scanner that scans a user's fingerprint. The template fingerprint database 120 stores template fingerprint information including fingerprint feature points. The controller 130 is preferably implemented as a microprocessor integrated with ROM, RAM, and peripheral devices, and includes a local neighbor structure generator 131, an alignment and clustering processor 133, and a matching score calculator 135. .

The local neighbor structure generation unit 131 includes nearest feature point information based on each feature point for feature points of the template fingerprint read from the database 120 and feature points extracted from the fingerprint information input from the scanner 110. Create a local neighborhood structure. That is, the present invention proposes a local neighborhood structure containing information on neighboring feature points that are locally adjacent to each fingerprint feature point. The local neighborhood structure is a structure that contains information that is invariant to translation and rotation, and is stable to distortion propagating over a wide area. In an example, the local neighbor structure generator 131 may include a nearest feature point calculator 1312 and a feature point encoding variable calculator 1324.

The nearest feature point calculator 1312 refers to each feature point of the fingerprint as a reference point (hereinafter, referred to as a reference feature point. For example, when generating a local neighboring structure for the i-th fingerprint feature point, the i-th fingerprint feature becomes the reference feature point). By setting a local coordinate system, the plane of the local coordinate system is divided into K equal-angle regions to calculate nearest feature points in each region. Here, the operation of the nearest feature point calculator 1312 will be described with reference to FIG. 2A.

In the local coordinate system, each feature point has a position and direction as attributes, so the direction of each feature point can be set in the x-axis. For example, as shown in FIG. 2A, the nearest feature point calculator 1312 first sets a local coordinate system using the reference feature as the origin and the ridge line of the reference feature as the x axis. The nearest feature point calculator 1312 divides the local coordinate system plane into an area having a 45 ° angle (k = 8), and calculates nearest feature points with respect to the reference feature point among feature points included in each area. In Fig. 2A, the rectangular, circular, and x-shaped line segments respectively represent reference feature points, neighbors selected as reference feature points and nearest feature points in each region, and remaining non-selected feature points.

The feature point encoding variable calculator 1324 calculates encoding variables of the nearest feature points calculated by the nearest feature point calculator 1312. Here, an operation of calculating encoding variables of nearest feature points in the feature point encoding variable calculator 1324 will be described with reference to FIG. 2B. As shown in FIG. 2B, the feature point encoding variable calculating unit 1324 has an angle θ _k , a distance d _k , and a position of a neighbor measured with respect to the k th nearest feature point based on the center feature point of the local structure. Calculate the encoding variables of the nearest feature points, including the angle φ _k for.

The alignment and clustering processor 133 aligns the input fingerprint with the template fingerprint by using the template fingerprint generated by the local neighbor structure generator 131 and the local neighborhood structure of the input fingerprint, and the aligned input fingerprint. Cluster the feature points that match in the template fingerprint. In one embodiment, the alignment and clustering processing unit 133 may be implemented to include a similar local neighbor structure acquisition unit 1332, an alignment parameter setting unit 1334, and a feature point cluster setting unit 1336.

The similar local neighbor structure obtaining unit 1332 compares each local neighborhood structure of the template fingerprint generated by the local neighbor structure generating unit 131 with each local neighborhood structure of the input fingerprint to obtain a similarity degree. And local neighborhood structures of the input fingerprint and the pair of template fingerprints whose calculated similarity is greater than the threshold are obtained. In one example, the similar local neighbor structure obtaining unit 1332 calculates a difference between the nearest feature point encoding variables θ _k , d _k , Φ _k included in the local neighbor structure of the template fingerprint and the input fingerprint, and the difference between the variables. Is compared with the threshold to determine whether the variables are similar to each other. The similar local neighbor structure obtainer 1332 may be implemented to count the number of nearest feature points having similar encoding variables to calculate the similarity. For example, the similar local neighbor structure acquirer 1332 may be implemented to calculate the similarity by calculating (the number of similar nearest feature points) / (the total number of nearest feature points).

The alignment parameter setting unit 1334 converts using the position and angle of the reference feature points in the local neighborhood structures of the pair of template fingerprints and the input fingerprints acquired by the similar local neighbor structure acquisition unit 1332. Calculate the parameters. Here, the positions and angles of the reference feature points mean the positions and angles of the reference feature points in the coordinate system for the entire fingerprint image data. The position and angle of these reference feature points can be easily set programmatically. For example, the alignment parameter setting unit 1334 may determine an angle when the positions of the reference feature 1 of the template fingerprint and the reference feature 2 of the input fingerprint are (x1, y1) and (x2, y2), respectively, and the angles are r1, r2, respectively. Calculate the transformation parameters r1-r2, displacements (x1-x2, y1-y2).

In addition, the alignment parameter setting unit 1334 sets the transformation parameter having the maximum similarity among the calculated transformation parameters as the alignment parameter. In one example, the alignment parameter setting unit 1334 may be implemented to select a transformation parameter that maximizes the sum of the similarities for a predetermined space size in the transformation parameter space. Here, the transform parameter space is a three-dimensional space obtained by using a pair of template fingerprints having a high similarity and angles and displacements of feature points belonging to a local neighboring structure of the input fingerprint. It means a sphere with a certain size of space around each feature point. The sorting parameter setting unit 1334 obtains feature points included in a predetermined space size sphere among the displayed feature points on the 3D space, and adds the similarity corresponding to the feature points to obtain a transform parameter having the maximum similarity. Set to.

The feature point cluster setting unit 1336 aligns the input fingerprints according to the alignment parameters set by the alignment parameter setting unit 1334, and features the maximum feature point matching pairs gathered in a predetermined area from the aligned input fingerprints and the template fingerprints. Set to cluster. Referring to FIG. 3B, for example, when the maximum feature point matching pairs gathered in a certain area are separated from the other regions except for the feature point cluster (Cluster 1), it can be seen that the similarity of the local neighbor structure is also maintained locally. Therefore, if the above clustering process is performed after reordering only the remaining feature points except for the pairs of corresponding pairs corresponding to the existing clusters, the clusters may have a smaller number of matching pairs than the previous clusters, but may be meaningful. do. Preferably, the feature point group setting unit 1336 is configured to set the feature group including at least one feature point matching pair by repeating the alignment and clustering process with only the remaining feature points except the feature points of the set feature point cluster.

Here, referring to FIG. 3A, it can be seen that the feature points included in the input fingerprint boundary box 22 are aligned with the feature points included in the template fingerprint boundary box 21. In addition, referring to FIG. 3B, feature clusters (Cluster 1, Cluster 2, Cluster 3, Cluster) within overlapped areas of the template fingerprint boundary box 21 and the input fingerprint boundary box 22 are described. 4) You can check.

The matching score calculator 135 calculates a fingerprint matching score between the template fingerprint and the input fingerprint using the clustered feature points in the alignment and clustering processor 133. In one embodiment, the matching score calculator 135 may be implemented to calculate a fingerprint matching score through an algorithm corresponding to Equation 1 below.

Where T is a template fingerprint, I is an input fingerprint, S (T, I) is a fingerprint matching score, N ₁ and N ₂ are the number of feature points of the template fingerprint and the input fingerprint, and N _P is the number of matched feature points. .

In another embodiment, preferably, the matching score calculator 135 may be implemented to calculate a fingerprint matching score through an algorithm corresponding to Equation 2 below.

), N_O 는 중첩된 영역 안에서 매칭된 특징점들의 개수를 의미한다. 여기서, 특징점 정합 비용(w)는 (유사 최근접 특징점들의 개수) / (총 최근접 특징점들의 개수)이다.Where T is the template fingerprint, I is the input fingerprint, S (T, I) is the fingerprint matching score, and A _O is the template fingerprint boundary box 21 and the input fingerprint boundary box 22. The overlapped area, A ₁ is the area of the template fingerprint bounding box 21, S _P is the sum of the feature point matching costs (w) corresponding to the corresponding pairs of each cluster in the clustering process (

), N _O denotes the number of feature points matched in the overlapped region. Here, the feature point matching cost w is (the number of similar nearest feature points) / (the total number of nearest feature points).

In one embodiment, the alignment and clustering processing unit 133 of the present invention further comprises a fingerprint deformation processing unit 1338 for transforming the input fingerprint with respect to the template fingerprint by using all the feature point information clustered by the feature point cluster setting unit 1336. It can be implemented to include. According to such an embodiment, the feature point cluster setting unit 1336 may cluster feature points that are additionally matched with the input fingerprint and the template fingerprint modified by the fingerprint deformation processing unit 1338.

The fingerprint deformation processing unit 1338 transforms the input fingerprint with respect to the template fingerprint by using all the feature point information grouped by the feature point cluster setting unit 1336. Referring to FIG. 4, a fingerprint deformation result performed by the fingerprint deformation processing unit 1338 may be confirmed. In one example, the fingerprint deformation processing unit 1338 uses a thin-plate spline (TPS) fingerprint deformation method. The TPS fingerprint deformation method is an analytical solution that minimizes the bending energy of the thin plate by interpolating the deformation pattern of the entire surface based on a given point pair.

In one embodiment, the fingerprint modification processing unit 1338 may be implemented to modify the input fingerprint with respect to the template fingerprint through a basic TPS algorithm as shown in Equation 3 below.

Where k Denotes a basic function matrix, w denotes a nonlinear transformation matrix, a denotes an affine transformation matrix, and P and v denote input and template feature point position matrices.

However, the basic TPS algorithm, as shown in Equation 3, determines the deformation pattern by precise interpolation between the spaces between the pairs, so that if the pairs obtained incorrectly exist, deformation of the fingerprint surface occurs in the wrong direction. Therefore, there is a need for a method of obtaining good deformation results even in the presence of unstable matching pairs due to low feature point matching costs. In this specification, a weight normalization method having a feature point matching cost as a function recognition is proposed.

In another embodiment, the fingerprint modification processing unit 1338 of the present invention preferably transforms the input fingerprint to the template fingerprint through an algorithm such as Equation 4 to which a weight normalization method having a feature point matching cost as a function recognition is applied. It can be implemented to.

는 정규화 매개 변수 행렬이고, 각

는 k번째 정합쌍에서 대응쌍 주변의 지문 표면의 평탄화 정도를 정의한다. 도 4를 참조하면 지문 변형 처리부(1338)에서 수행된 지문 변형 결과를 확인할 수 있다. 신뢰도가 낮은 대응쌍 근처의 변형 패턴을 다른 주변의 대응쌍의 정보로 근사시키는 방법으로 전체적인 변형 패턴을 얻게 되어 결과적으로 안정적인 변형 패턴을 얻게 된다. 도 4 에서, 격자 모양의 변형 패턴(31)은 입력 지문의 전체적인 변형 패턴을 나타낸다. 여기서 기존 군집에 해당하지 않는 특징점에 대해서도 적절하게 대응쌍이 맞도록 변형 패턴을 구하였음을 알 수 있다.here,

Is a normalization parameter matrix, and each

Defines the degree of planarization of the fingerprint surface around the corresponding pair in the k-th matching pair. Referring to FIG. 4, a fingerprint deformation result performed by the fingerprint deformation processing unit 1338 may be confirmed. The overall deformation pattern is obtained by approximating the deformation pattern near the pair with low reliability to the information of the corresponding pair of other peripherals, thereby obtaining a stable deformation pattern. In Fig. 4, the lattice-shaped deformation pattern 31 represents the overall deformation pattern of the input fingerprint. Here, it can be seen that the deformation pattern was obtained to match the corresponding pairs appropriately even for the feature points that do not correspond to the existing clusters.

5 is a graph comparing fingerprint recognition performance in a fingerprint recognition system using a fingerprint matching device according to the present invention.

In this experiment, the proposed algorithm, the BOZORTH3 algorithm included in NIST Fingerprint Image Software 2, and the rigid matching 1,2 algorithm were used. On the other hand, in this experiment, we used the FVC2002 DB1_A fingerprint DB including 800 fingerprint images acquired in various situations for 100 different fingers. Following the procedure applied in the process of FVC2002, a total of 2,800 fingerprints of the same person and a total of 4,950 fingerprints of different people were tested.

In the proposed algorithm, the feature extraction method based on the fingerprint image enhancement method and the binarization method is used, and the same feature extraction results are used in all tested matching algorithms to compare the performance of the matching method alone. In the proposed method, the rigid matching 2 algorithm uses the same process except the feature clustering and deformation process. The rigid matching 1 algorithm uses the same process except the process of using the fingerprint matching score in the rigid matching 2. The BOZORTH3 algorithm implicitly allows nonlinear distortion while examining a table that calculates the correspondence between fingerprints.

Referring to Table 1 and Figure 5, it can be seen that the fingerprint matching algorithm proposed in this specification shows the fingerprint recognition results that outperform the BOZORTH3 algorithm, rigid matching 1,2.

Method EER FMR100 FMR1000 BOZORTH3 3.35% 5.25% 7.14% Rigid matching 1 2.03% 2.43% 4.18% Rigid matching 2 1.32% 1.39% 2.43% proposed 0.99% 1.00% 1.54%

6 is a flowchart of fingerprint matching in a fingerprint matching device according to the present invention, FIG. 7 is a flowchart of generating a local neighborhood structure according to the present invention, and FIGS. 8 and 9 are a flowchart of alignment and clustering according to the present invention. . Hereinafter, a fingerprint matching method of the fingerprint matching device according to the present invention will be described with reference to FIGS. 6 to 9.

Referring to FIG. 6, the fingerprint matching apparatus according to the present invention first acquires feature points of a template fingerprint and an input fingerprint (S501). The feature data of the template fingerprint can be obtained from a template fingerprint database including the fingerprint feature points. The feature data of the input fingerprint can be exaggerated by preprocessing the fingerprint image input from the fingerprint scanner, finding the direction component of each ridge, ridges and valleys. The process of dividing and binarizing it, determining the thickness of each ridge and thinning it to one point line, and removing the singular point by the direction and coordinate of the thin line may be obtained from the feature point extraction apparatus.

Subsequently, the fingerprint matching apparatus according to the present invention generates a local neighborhood structure including nearest feature point information on the feature points of the template fingerprint and the feature points of the input fingerprint based on each feature point (S503). Referring to FIG. 7, the fingerprint matching apparatus according to the present invention sets a local coordinate system based on each feature point of an input fingerprint (S601), and divides a plane with respect to the local coordinate system into regions of K equal angles in each region. The contact feature points are calculated (S602), and the encoding variables ( θ _k , d _k , Φ _k ) of the calculated nearest feature points are calculated (S603).

Subsequently, the fingerprint matching apparatus according to the present invention sets a local coordinate system based on each feature point of the template fingerprint as a reference point (S604), and divides the plane with respect to the local coordinate system into K equal angle regions to calculate nearest feature points in each region. In operation S605, encoding variables θ _k , d _k , and Φ _k of the calculated nearest feature points are calculated (S606).

Referring back to FIG. 6, the fingerprint matching apparatus according to the present invention generates a local neighborhood structure (S503) and then uses the generated local fingerprint structure and the local neighborhood structure of the input fingerprint. The input fingerprint is aligned with the template fingerprint, and the feature points matched by the aligned input fingerprint and the template fingerprint are clustered (S505). Referring to FIG. 8, the fingerprint matching apparatus according to the present invention calculates a similarity by comparing each local neighborhood structure of the template fingerprint with each local neighborhood structure of the input fingerprint (S701), The calculated similarity is compared with a threshold to obtain local neighborhood structures of a pair of template fingerprints and an input fingerprint in which the similarity is greater than the threshold (S702). For example, the fingerprint matching apparatus according to the present invention calculates a difference between nearest feature point encoding variables θ _k , d _k , Φ _k included in a local neighbor structure of the template fingerprint and the input fingerprint, and thresholds the difference between the variables. Compare with to determine if the variables are similar. Then, the fingerprint matching apparatus according to the present invention may be implemented to count the number of nearest feature points having similar encoding variables and calculate the similarity by calculating (the number of similar nearest feature points) / (the total number of nearest feature points). .

Subsequently, the fingerprint matching apparatus according to the present invention calculates the conversion parameters by using the positions and angles of the reference feature points in the acquired local neighborhood structures of the pair of template fingerprints and the input fingerprints (S703). Here, the positions and angles of the reference feature points mean the positions and angles of the reference feature points in the coordinate system for the entire fingerprint image data. The position and angle of these reference feature points can be easily set programmatically. In the fingerprint matching device according to the present invention, when the position of the reference feature point 1 of the template fingerprint and the reference feature point 2 of the input fingerprint are respectively (x1, y1), (x2, y2) and the angles are respectively r1, r2, the angle r1-r2 Calculate the transformation parameter, which is the displacement (x1-x2, y1-y2).

Thereafter, the fingerprint matching apparatus according to the present invention sets the conversion parameter having the maximum similarity among the calculated conversion parameters as the alignment parameter (S704). The fingerprint matching device according to the present invention can be implemented to select a conversion parameter that maximizes the sum of the similarities for a certain space size in the conversion parameter space. Here, the transform parameter space is a three-dimensional space obtained by using a pair of template fingerprints having a high similarity and angles and displacements of feature points belonging to a local neighboring structure of the input fingerprint. It means a sphere with a certain size of space around each feature point. The fingerprint matching device according to the present invention obtains the feature points included in a certain space size sphere among the displayed feature points on the three-dimensional space, sums the similarity corresponding to the feature points, and converts the transform parameter having the maximum similarity as the alignment parameter. Set it.

Then, the fingerprint matching device according to the present invention aligns the input fingerprint according to the set alignment parameter, and sets the maximum feature point matching pairs gathered in a predetermined area in the aligned input fingerprint and the template fingerprint as a feature point cluster (S705). Referring to FIG. 9, in an embodiment, the fingerprint matching apparatus according to the present invention may examine edge connection structures of feature point matching pairs (S901), and the remaining feature points except for feature point matching pairs separated from the cluster as a result of edge connection structure inspection. Matching pairs are set to feature point clusters (S902).

Again, referring to FIG. 8, the fingerprint matching device according to the present invention checks whether there is a feature point matching pair not included in the cluster (S706). When the fingerprint matching device according to the present invention has a feature point matching pair not included in the cluster as a result, the similarity is calculated by comparing local neighboring structures of the remaining feature points except for the feature points of the existing cluster (S707), and realigning and clustering process. S702 to S705 are repeated to obtain a feature point cluster including at least one feature point matching pair.

The fingerprint matching device according to the present invention checks whether there is a feature point matching pair not included in the cluster (S706), and if there is no feature point matching pair not included in the cluster, the input fingerprint is applied to the template fingerprint using all clustered feature point information. In operation S708, the feature points additionally matched with the modified input fingerprint and the template fingerprint are clustered (S709).

Thereafter, the fingerprint matching device according to the present invention calculates a fingerprint matching score between the template fingerprint and the input fingerprint using the clustered feature points (S507). The fingerprint recognition system using the fingerprint matching device according to the present invention may check whether the template fingerprint and the input fingerprint are the same person's fingerprints by using the calculated fingerprint matching score.

As described above, the fingerprint matching algorithm proposed herein first generates a local neighbor structure that is invariant to nonlinear distortion, and then clusters the feature matching pairs in an iterative manner using the local neighbor structure, and finally By modifying the entire fingerprint surface based on the feature point matching pairs, additional feature point matching pairs not included in the existing clusters are obtained and clustered, and the information from each cluster is synthesized to calculate the fingerprint matching score, thereby solving the nonlinear distortion problem. There is a useful effect that can effectively perform fingerprint recognition while solving.

Although the present invention has been described with reference to the accompanying drawings, it will be apparent to those skilled in the art that many various obvious modifications are possible without departing from the scope of the invention from this description. Therefore, it should be interpreted by the claims described to include many such variations.

Claims (9)

A scanner scanning a user fingerprint; A database for storing template fingerprint information including fingerprint feature points; Create a local neighborhood structure including the feature points of the template fingerprint read from the database and the feature points extracted from the fingerprint information input from the scanner, including the nearest feature point information of each feature point based on a local coordinate system. A local neighboring structure generating unit; The input fingerprint is aligned with the template fingerprint using a local neighborhood structure of the template fingerprint and the input fingerprint generated by the local neighbor structure generation unit, and feature points matched by the aligned input fingerprint and the template fingerprint are obtained. Clustering and clustering processing unit; and A matching score calculator configured to calculate a fingerprint matching score between a template fingerprint and an input fingerprint using clustered feature points in the alignment and clustering processor; Fingerprint matching device comprising a. The method of claim 1, wherein the local neighborhood structure generating unit: A nearest feature point calculator configured to set a local coordinate system based on each feature point of the fingerprint, and calculate the nearest feature points in each area by dividing a plane with respect to the local coordinate system of each feature point into K equal angle regions; and A feature point encoding variable calculator for calculating encoding variables of the nearest feature points calculated by the nearest feature point calculator; Fingerprint matching device comprising a. The method of claim 1 or 2, wherein the alignment and clustering processing unit: The similarity is calculated by comparing each local neighborhood structure of the template fingerprint generated by the local neighborhood structure generator with each local neighborhood structure of the input fingerprint, and as long as the similarity is greater than a threshold. A pseudo local neighborhood structure acquisition unit for obtaining local neighborhood structures of the paired template fingerprint and the input fingerprint; The calculated conversion parameters are calculated by calculating the conversion parameters using the position and angle of reference feature points in the pair of template fingerprints and the local neighborhood structures of the input fingerprint obtained by the pseudo local neighborhood structure acquisition unit. A sort parameter setting unit for setting a transform parameter having a maximum similarity as a sort parameter; Align the input fingerprints according to the alignment parameters set by the alignment parameter setting unit, and set the maximum feature point matching pairs gathered in a predetermined area in the aligned input fingerprint and template fingerprint as feature point clusters, A feature point cluster setting unit for setting a feature point cluster including at least one feature point matching pair by repeatedly performing an alignment and clustering process with only the remaining feature points except for the feature points; Fingerprint matching device comprising a. The method of claim 3, wherein the alignment and clustering processing unit: Further comprising a fingerprint deformation processing unit for transforming the input fingerprint with respect to the template fingerprint by using all the feature point information clustered in the feature point group setting unit, And the feature point group setting unit clusters the feature points that are additionally matched by the input fingerprint and the template fingerprint modified by the fingerprint deformation processing unit. In the fingerprint matching method between the template fingerprint and the input fingerprint, Generating a local neighborhood structure including the feature points of the template fingerprint and the feature points of the input fingerprint, including nearest feature point information of each feature point based on a local coordinate system; An alignment and clustering step of aligning the input fingerprint with the template fingerprint using a local neighborhood structure of the generated template fingerprint and the input fingerprint, and clustering the feature points matched from the aligned input fingerprint and the template fingerprint ; And Calculating a fingerprint matching score between a template fingerprint and an input fingerprint using the clustered feature points; Fingerprint matching method comprising a. 6. The method of claim 5, wherein creating the local neighboring structure is: Setting a local coordinate system based on each feature point of the input fingerprint, dividing a plane with respect to the local coordinate system of each feature point into regions of K equal angles, calculating nearest feature points in each region, and calculating the nearest feature points Calculating encoding variables; and Setting a local coordinate system based on each feature point of the template fingerprint, dividing a plane with respect to the local coordinate system of each feature point into regions of K equal angles, calculating nearest feature points in each region, and calculating the nearest feature points Calculating encoding variables; Fingerprint matching method comprising a. The method of claim 5 or 6, wherein the sorting and clustering step is: Comparing each local neighborhood structure of the template fingerprint with each local neighborhood structure of the input fingerprint to calculate a similarity; Comparing the calculated similarity with a threshold to obtain a pair of template fingerprints having a similarity greater than a threshold and local neighborhood structures of the input fingerprint; Calculating transformation parameters using positions and angles of reference feature points in the obtained local neighborhood structures of the pair of template fingerprints and the input fingerprint; Setting a transformation parameter having a maximum similarity among the calculated transformation parameters as an alignment parameter; Arranging input fingerprints according to the set alignment parameter, and setting maximum feature matching pairs gathered in a predetermined area in the aligned input fingerprint and template fingerprint as a feature point cluster; and Setting a feature point cluster including at least one feature point matching pair by repeatedly performing an alignment and clustering process using only the remaining feature points except the feature points of the set feature point cluster; Fingerprint matching method comprising a. 8. The method of claim 7, wherein the setting of the maximum feature point matching pairs gathered in a predetermined area in the aligned input fingerprint and the template fingerprint as a feature point cluster includes: Examining an edge connection structure of the feature point matching pairs; and Setting the other feature point matching pairs except for the feature point matching pairs apart from the cluster as a result of the edge connection structure test; Fingerprint matching method comprising a. 8. The method of claim 7, wherein the sorting- clustering step is: Modifying the input fingerprint with respect to a template fingerprint using all the clustered feature point information; and Clustering feature points that further match in the modified input fingerprint and the template fingerprint; Fingerprint matching method further comprises.

Images