CN101303729A

CN101303729A - A New Fingerprint Singularity Detection Method

Info

Publication number: CN101303729A
Application number: CNA200810138119XA
Authority: CN
Inventors: 杨公平; 翁大伟; 尹义龙; 任春晓; 詹小四
Original assignee: Shandong University
Current assignee: Shandong University
Priority date: 2008-07-01
Filing date: 2008-07-01
Publication date: 2008-11-12
Anticipated expiration: 2028-07-01
Also published as: CN101303729B

Abstract

The invention discloses a novel fingerprint singular point detection method. It solves the problems that the current singular point extraction method relies heavily on the fingerprint direction field, cannot effectively deal with low-quality fingerprint images, and is difficult to be suitable for engineering applications. The method is divided into two stages. The Lay index method is effectively improved, and it is used to extract the candidate singular points in the fingerprint image. In the latter stage, the Gaussian-Hermite Gauss-Hermitian moments are effectively used to remove the spurious processing of the candidate singular points. This method effectively combines the The ridge direction change information in the neighborhood around the singular point and the ridge consistency change trend information in the local area around the singular point can accurately and reliably extract the singular point in the fingerprint image, which has strong noise resistance, high accuracy and reliability, and engineering Features such as great application value.

Description

A New Fingerprint Singularity Detection Method

技术领域 technical field

本发明涉及一种指纹图像检测方法，具体地说一种实用自动指纹识别系统(AFIS)中的低质量指纹图像的新的指纹奇异点检测方法。The invention relates to a fingerprint image detection method, in particular to a novel fingerprint singular point detection method for low-quality fingerprint images in an automatic fingerprint identification system (AFIS).

背景技术 Background technique

目前在实用自动指纹识别系统中，指纹分类技术是加快系统识别速度的关键技术之一，而现今主流的分类技术多是依据奇异点的数目、类型和位置等信息来实现的，而且在处理低质量指纹图像时所采用的纹理匹配算法也需要准确可靠的奇异点信息。现今自动指纹识别系统中采用的主流的奇异点提取方法，绝大多数依赖于指纹方向场的准确提取，但在处理低质量指纹图像时，由于可靠的计算纹线方向本身就是一个难题，因而这些方法提取的奇异点不仅定位不够准确，在纹线方向计算有误的地方以及一些噪声污染的地方，还往往容易检测到许多虚假的奇异点。这使得这些方法难以有效满足工程应用。在实际的应用中需要一种准确可靠的奇异点提取算法。At present, in the practical automatic fingerprint identification system, the fingerprint classification technology is one of the key technologies to speed up the identification speed of the system, but the current mainstream classification technology is mostly realized based on the number, type and position of singular points, and is low in processing. The texture matching algorithm used in quality fingerprint images also requires accurate and reliable singular point information. Most of the mainstream singular point extraction methods used in today's automatic fingerprint identification systems rely on the accurate extraction of the fingerprint direction field. The singular points extracted by the method are not only inaccurately positioned, but also often detect many false singular points easily in places where the calculation of the ridge direction is wrong and where some noise is polluted. This makes these methods difficult to effectively meet engineering applications. In practical applications, an accurate and reliable singular point extraction algorithm is needed.

发明内容 Contents of the invention

本发明的目的就是为了解决现有的奇异点提取方法严重依赖指纹方向场、不能有效处理低质量指纹图像、难以适于工程应用等问题，提供一种准确可靠的新的指纹奇异点检测方法，该方法分为前后两个阶段，前一阶段对现有的Poincare index庞加莱指数方法进行了有效的改进，并利用其提取指纹图像中的候选奇异点，后一阶段效利用Gaussian-Hermite高斯-埃尔米特矩矩对候选奇异点进行去伪处理，该方法有效结合了奇异点周围邻域的纹线方向变化信息和奇异点周围局部区域的纹线一致性变化趋势信息，能准确可靠地提取指纹图像中的奇异点，具有抗噪性强、准确可靠性高、工程应用价值大等特点。The purpose of the present invention is to provide a new accurate and reliable fingerprint singular point detection method to solve the problems that the existing singular point extraction method relies heavily on the fingerprint direction field, cannot effectively process low-quality fingerprint images, and is difficult to be suitable for engineering applications. The method is divided into two stages. In the former stage, the existing Poincare index method is effectively improved and used to extract candidate singular points in the fingerprint image. In the latter stage, the Gaussian-Hermite Gaussian -The Hermitian moment de-aliases the candidate singular points. This method effectively combines the ridge direction change information of the neighborhood around the singular point and the ridge consistency change trend information of the local area around the singular point, which can be accurate and reliable. It has the characteristics of strong anti-noise, high accuracy and reliability, and great engineering application value.

为实现上述目的，本发明采用如下技术方案：To achieve the above object, the present invention adopts the following technical solutions:

一种准确可靠的奇异点提取方法，它的方法为，An accurate and reliable singular point extraction method, its method is,

(1)预备阶段：包括指纹图像的背景分离和方向场信息计算。(1) Preparatory stage: including the background separation of the fingerprint image and the calculation of the direction field information.

(2)第一阶段：候选奇异点的提取，包括以下几个步骤：(2) The first stage: the extraction of candidate singular points, including the following steps:

(a)方向域O′中的每一个点(i，j)按公式(1)计算其相应的Poincare index庞加莱指数P_G，C(i，j)。本算法中采用了两条长度分别为5×5和7×7的封闭曲线来计算每个点的Poincare index庞加莱指数值，只要其中一条的计算结果符合奇异点条件，则认定该点为候选奇异点，若两条曲线检测出的奇异点类型不一致，则直接认定该点为伪点。(a) Each point (i, j) in the direction domain O' calculates its corresponding Poincare index _{PG, C} (i, j) according to the formula (1). In this algorithm, two closed curves with lengths of 5×5 and 7×7 are used to calculate the Poincare index value of each point. As long as the calculation result of one of them meets the singular point condition, the point is considered to be Candidate singular points, if the types of singular points detected by the two curves are inconsistent, the point is directly identified as a false point.

(b)对上一步检测到的core核心点点和delta三角点运用聚类算法按欧氏距离分别进行聚类，并统计各个聚类包含的奇异点的个数N_m。N_m＜N，(N表示指纹模式区应该能够检测到的奇异点的个数，这里N取为25)的聚类被删除。(b) Use the clustering algorithm to cluster the core points and delta triangle points detected in the previous step according to the Euclidean distance, and count the number N _m of singular points contained in each cluster. N _m <N, (N represents the number of singular points that should be detectable in the fingerprint pattern area, where N is taken as 25) clusters are deleted.

(c)求取剩余的各个聚类的平均突变程度，并按奇异点类型、突变程度大小分别对这些聚类进行排序，选取前M(本方法M＝3，core核心点点聚类和delta三角点点聚类都取3个，当实际聚类个数少于3个时取实际聚类个数)个突变程度较大的聚类作为候选奇异点聚类，候选奇异点最终就定位在这些聚类的质心上。(c) Find the average mutation degree of each remaining cluster, and sort these clusters according to the type of singular point and the degree of mutation, and select the top M (this method M=3, core core point clustering and delta triangle Take 3 point clusters, and when the actual number of clusters is less than 3, take the actual number of clusters) clusters with a greater degree of mutation as candidate singular point clusters, and the candidate singular points are finally located in these clusters on the centroid of the class.

(3)第二阶段：候选奇异点的去伪，包含以下几个步骤：(3) The second stage: de-falsification of candidate singular points, including the following steps:

(a)对每个候选奇异点，按公式(2)计算其周围半径为6τ的圆形邻域内每个像素的分布一致性coherence值(τ为平均纹线距离)。(a) For each candidate singular point, calculate the distribution consistency coherence value of each pixel in a circular neighborhood with a radius of 6τ (τ is the average ridge distance) according to formula (2).

(b)然后将该圆形区域划分为32个扇形和一个中心小圆形区域并计算各个部分的平均coherence值，圆形区域的半径和两个圆带的宽度均为2τ。(b) Then divide the circular area into 32 sectors and a central small circular area and calculate the average coherence value of each part. The radius of the circular area and the width of the two circular bands are both 2τ.

(c)最后，比较16个方向每个方向上的三个区域的平均分布一致性coherence值的大小(中心圆形区域为16个方向公共的区域)，若自里向外分布一致性coherence值越来越大，则标记该方向为有效方向。对某一候选奇异点而言，若有效方向数大于等于10，则提取该点为奇异点，否则，认为该点为伪点。若奇异点的某些方向出了指纹边界则将剩余的方向作为参考方向，在这些方向上计算有效方向。此外，经实验分析，计算每个象素分布一致性coherence值的窗口大小取为(4τ+1)×(4τ+1)较为合适。(c) Finally, compare the average coherence value of the three areas in each direction of the 16 directions (the central circular area is the area common to the 16 directions), if the coherence value is distributed from the inside to the outside If it becomes larger and larger, this direction is marked as a valid direction. For a candidate singular point, if the number of effective directions is greater than or equal to 10, the point is extracted as a singular point, otherwise, the point is considered as a false point. If some directions of the singular point are out of the fingerprint boundary, the remaining directions are taken as reference directions, and effective directions are calculated on these directions. In addition, through experimental analysis, it is more appropriate to take (4τ+1)×(4τ+1) as the window size for calculating the coherence value of each pixel distribution consistency.

第一阶段，所述步骤(a)中，计算每一个奇异点(i，j)的Poincare index庞加莱指数值时按如下的公式进行：In the first stage, in the step (a), the Poincare index Poincare index value of each singular point (i, j) is calculated according to the following formula:

$Poincare Poincare ((i i . . j j)) = = \frac{11}{22 π π} {Σ Σ}_{k k = = 00}^{Nψ Nψ - - 11} Δ Δ ((k k)),,$

$Δ Δ ((k k)) = = \{\begin{matrix} δ δ ((k k)),, & if if | | δ δ ((k k)) | | < < \frac{π π}{22} \\ π π + + δ δ ((k k)),, & ifδ ifδ ((k k)) \leq \leq - - \frac{π π}{22} \\ π π - - δ δ ((k k)),, & otherwise otherwise \end{matrix} - - - - - - ((11))$

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

i′(i+1)mod N_ψ，i′(i+1)mod N _ψ ,

其中ψ_x(i)和ψ_y(i)分别是以给定点为中心的具有Nψ个像素的封闭曲线上第k个点的x和y坐标。δ(k)表示两相邻方向角的差值，Δ(k)表示对差值调整以后的结果，i′表示第i个点之后的下一个点。where ψ _x (i) and ψ _y (i) are the x and y coordinates, respectively, of the kth point on a closed curve with Nψ pixels centered at a given point. δ(k) represents the difference between two adjacent direction angles, Δ(k) represents the result after adjusting the difference, and i' represents the next point after the i-th point.

本算法中采用了5×5和7×7的正方形封闭曲线，该公式的形成基于对原有Poincare Index庞加莱指数方法的改进，用于计算某点对应封闭曲线上方向角变化的累积。In this algorithm, 5×5 and 7×7 square closed curves are used. The formation of this formula is based on the improvement of the original Poincare Index method, which is used to calculate the accumulation of direction angle changes on the closed curve corresponding to a certain point.

所述步骤(a)中，奇异点条件是指本方法提出的奇异点应满足的条件，该条件是对公式(1)的计算过程进行约束，具体是指，在沿封闭曲线做方向角变化的累加时附加以下限制条件：In the step (a), the singular point condition refers to the condition that the singular point proposed by this method should satisfy, and this condition is to constrain the calculation process of formula (1), specifically refers to the direction angle change along the closed curve The following restrictions are added to the accumulation of :

(1)统计这Nψ个方向的符号变化次数，若方向由正到负和由负到正各发生一次，且仅有一次，则继续计算其Poincare index庞加莱指数值，否则，认为该点是普通点。(1) Count the number of sign changes in these Nψ directions. If the direction changes from positive to negative and from negative to positive, and only once, continue to calculate its Poincare index value, otherwise, consider this point It's ordinary.

(2)统计这Nψ个差中绝对值

的差的个数，若个数多于一个则认为该点是普通点。(2) Statistics of the absolute value of the Nψ differences

The number of differences, if the number is more than one, the point is considered to be an ordinary point.

(3)如果最终Poincare index庞加莱指数的值为1/2，那么该给定点(i，j)就被确定为core核心点，如果Poincare index庞加莱指数值为-1/2，那么该给定点(i，j)就被确定为delta三角点。(3) If the final Poincare index Poincare index value is 1/2, then the given point (i, j) is determined as the core core point, if the Poincare index Poincare index value is -1/2, then The given point (i, j) is determined as a delta triangle point.

第一阶段，所述步骤(b)中，N取值25是基于对试验结果的分析，在指纹的奇异点区能够检测到的奇异点数目是相对稳定的且集中在奇异点区的最内部，对于低质量指纹图像，只要奇异点区不受大噪声的污染，那么在奇异点区内能够检测到的奇异点的数目也是稳定的，本方法N取25。In the first stage, in the step (b), the N value of 25 is based on the analysis of the test results, and the number of singular points that can be detected in the singular point area of the fingerprint is relatively stable and concentrated in the innermost part of the singular point area , for low-quality fingerprint images, as long as the singular point area is not polluted by large noise, the number of singular points that can be detected in the singular point area is also stable, and N is set to 25 in this method.

第一阶段，所述步骤(c)中，突变程度是指，封闭曲线上相邻两方向角作差过程中，绝对值

的那个差值，平均突变程度是指每一聚类内所有像素的突变程度的平均。In the first stage, in the step (c), the degree of mutation refers to the absolute value of the difference between two adjacent direction angles on the closed curve.

The difference, the average mutation degree refers to the average of the mutation degree of all pixels in each cluster.

第二阶段，所述步骤(a)中，本方法利用分布一致性coherence值来表达纹线的一致性信息，而分布一致性coherence值的计算采用了Gaussian-Hermite高斯-埃尔米特矩。本方法用四个矩来描述指纹的纹线一致性信息：M_0，1，M_1，0，M_0，3，M_3，0。并对这四个不同阶的Gaussian-Hermite矩做如下定义：In the second stage, in the step (a), the method uses the coherence value of the distribution consistency to express the consistency information of the ridges, and the calculation of the coherence value of the distribution consistency adopts the Gaussian-Hermite Gaussian-Hermite moment. This method uses four moments to describe the ridge consistency information of the fingerprint: M _0,1 , M _1,0 , M _0,3 , M _3,0 . And the Gaussian-Hermite moments of these four different orders are defined as follows:

$\{\begin{matrix} {M m}_{u u} ((x x,, y the y)) = = λ λ {M m}_{1,0 1,0} ((x x,, y the y,, I I ((x x,, y the y)))) + + ((11 - - λ λ)) {M m}_{3,0 3,0} ((x x,, y the y,, I I ((x x,, y the y)))) \\ {M m}_{v v} ((x x,, y the y)) = = λ λ {M m}_{0,1 0,1} ((x x,, y the y,, I I ((x x,, y the y)))) + + ((11 - - λ λ)) {M m}_{0,3 0,3} ((x x,, y the y,, I I ((x x,, y the y)))) \end{matrix} - - - - - - ((22))$

这里λ(0＜λ＜1)是不同阶的Gaussian-Hermite高斯-埃尔米特矩的联合权重系数(本方法中λ取值0.5)。利用定义(2)，指纹图像中的每一个象素(i，j)都可以计算得到一个特征向量[M_u，M_v]^T。在指纹奇异点区和非奇异点区[M_u，M_v]^T的分布具有不同的特征，在非奇异点区{M_u，M_v]^T沿着垂直于脊线的方向分布，在奇异点区[M_u，M_v]^T则均匀地分布在各个方向上，本方法采用主分量分析方法提取[M_u，M_v]^T的分布特性，[M_u，M_v]^T协方差矩阵C_M由下式定义：Here λ (0<λ<1) is the joint weight coefficient of Gaussian-Hermite moments of different orders (λ takes the value of 0.5 in this method). Using definition (2), each pixel (i, j) in the fingerprint image can be calculated to obtain a feature vector [M _u , M _v ] ^T . The distribution of [M _u , M _v ] ^T in the fingerprint singular point area and the non-singular point area has different characteristics, in the non-singular point area {M _u , M _v ] ^T distributes along the direction perpendicular to the ridge line, The point area [M _u , M _v ] ^T is evenly distributed in all directions. This method uses the principal component analysis method to extract the distribution characteristics of [M _u , M _v ] ^T , and the covariance matrix of [M _u , M _v ] ^T C _M is defined by:

${C C}_{M m} = = [\begin{matrix} \underset{w w}{Σ Σ} {(({M m}_{u u} {- - m m}_{u u}))}^{22} & \underset{w w}{Σ Σ} (({M m}_{u u} - - {m m}_{u u})) (({M m}_{v v} - - {m m}_{v v})) \\ \underset{w w}{Σ Σ} (({M m}_{u u} - - {m m}_{u u})) (({M m}_{v v} - - {m m}_{v v})) & \underset{w w}{Σ Σ} {(({M m}_{v v} - - {m m}_{v v}))}^{22} \end{matrix}] - - - - - - ((33))$

其中， $m_{u} = \frac{1}{n \times n} \underset{w}{Σ} M_{u},$ $m_{v} = \frac{1}{n \times n} \underset{w}{Σ} M_{v},$ m_u表示窗口W内平均的M_u值，m_v表示窗口W内平均的M_v值，n×n是窗口W的大小，，n×n是窗口W的大小，本方法n取值为4τ+1，τ为平均纹线距离。in, $m_{u} = \frac{1}{no \times no} \underset{w}{Σ} m_{u},$ $m_{v} = \frac{1}{no \times no} \underset{w}{Σ} m_{v},$ m _u _represents the average value of Mu in the window W, m _v represents the average value of M _v in the window W, n×n is the size of the window W, n×n is the size of the window W, and the value of n in this method is 4τ +1, τ is the average ridge distance.

设λ1和λ2是协方差矩阵C_M的两个特征值，则当λ1＞＞λ2时，[M_u，M_v]^T的分布主要是沿着长轴方向分布，也即沿着垂直于脊线的方向分布，而在噪声区或者奇异点区λ1和λ2的值是很接近的，因此，定义[M_u，M_v]^T的分布一致性coherence特征如下：Let λ1 and λ2 be the two eigenvalues of the covariance matrix C _M , then when λ1>>λ2, the distribution of [M _u , M _v ] ^T is mainly distributed along the long axis, that is, along the direction perpendicular to the ridge The direction distribution of the line, and the values of λ1 and λ2 in the noise area or singular point area are very close. Therefore, the distribution consistency coherence characteristics of [M _u , M _v ] ^T are defined as follows:

$coherence coherence = = \frac{{λ λ}_{11}^{22} - - {λ λ}_{22}^{22}}{{λ λ}_{11}^{22} + + {λ λ}_{22}^{22}} - - - - - - ((44))$

$= = \frac{((\underset{w w}{Σ Σ} {(({M m}_{u u} - - {m m}_{u u}))}^{22} + + \underset{w w}{Σ Σ} {(({M m}_{v v} - - {m m}_{v v}))}^{22})) \sqrt{{((\underset{w w}{Σ Σ} {(({M m}_{u u} - - {m m}_{u u}))}^{22} - - \underset{w w}{Σ Σ} {(({M m}_{v v} - - {m m}_{v v}))}^{22}))}^{22} + + 44 {((\underset{w w}{Σ Σ} (({M m}_{u u} - - {m m}_{u u})) (({M m}_{v v} - - {m m}_{v v}))))}^{22}}}{{((\underset{w w}{Σ Σ} {(({M m}_{u u} - - {m m}_{u u}))}^{22}))}^{22} + + {((\underset{w w}{Σ Σ} {(({M m}_{v v} - - {m m}_{v v}))}^{22}))}^{22} + + 22 {((\underset{w w}{Σ Σ} (({M m}_{u u} - - {m m}_{u u})) (({M m}_{v v} - - {m m}_{v v}))))}^{22}}$

由此，纹线一致性越好，分布一致性coherence越大。Therefore, the better the consistency of the ridges, the greater the coherence of the distribution consistency.

本发明的有益效果：由于采用了两阶段的处理方法，奇异点提取的可靠性有了较大程度的提高，特别是在处理低质量指纹图像时，奇异点的漏检、误检现象有了较大程度的减少。此外，对原有的Poincare index庞加莱指数方法进行了改进，提高其抗噪能力，避免了对方向场的多次平滑处理，使最终提取的奇异点准确性得到有效提高。满足了实用自动指纹识别系统(AFIS)的应用需求。Beneficial effects of the present invention: due to the adoption of the two-stage processing method, the reliability of singular point extraction has been greatly improved, especially when processing low-quality fingerprint images, the phenomenon of missed detection and false detection of singular points has been eliminated. a greater degree of reduction. In addition, the original Poincare index method is improved to improve its anti-noise ability, avoid multiple smoothing of the direction field, and effectively improve the accuracy of the final extracted singular points. It meets the application requirements of practical automatic fingerprint identification system (AFIS).

附图说明 Description of drawings

图1为去伪模板图像；Fig. 1 is to remove false template image;

图2为算法流程图。Figure 2 is the flow chart of the algorithm.

具体实施方式 Detailed ways

下面结合附图与实施例对本发明做进一步说明。The present invention will be further described below in conjunction with the accompanying drawings and embodiments.

(4)预备阶段：包括指纹图像的背景分离和方向场信息计算。(4) Preparatory stage: including the background separation of the fingerprint image and the calculation of the direction field information.

(5)第一阶段：候选奇异点的提取，包括以下几个步骤：(5) The first stage: the extraction of candidate singular points, including the following steps:

(d)方向域O′中的每一个点(i，j)按公式(1)计算其相应的Poincare index庞加莱指数P_G，C(i，j)。本算法中采用了两条长度分别为5×5和7×7的封闭曲线来计算每个点的Poincare index庞加莱指数值，只要其中一条的计算结果符合奇异点条件，则认定该点为候选奇异点，若两条曲线检测出的奇异点类型不一致，则直接认定该点为伪点。(d) For each point (i, j) in the direction field O', calculate its corresponding Poincare index _{PG, C} (i, j) according to the formula (1). In this algorithm, two closed curves with lengths of 5×5 and 7×7 are used to calculate the Poincare index value of each point. As long as the calculation result of one of them meets the singular point condition, the point is considered to be Candidate singular points, if the types of singular points detected by the two curves are inconsistent, the point is directly identified as a false point.

(e)对上一步检测到的core核心点和delta三角点运用聚类算法按欧氏距离分别进行聚类，并统计各个聚类包含的奇异点的个数N_m，N_m＜N(这里N取为25)的聚类被删除。(e) Use the clustering algorithm to cluster the core points and delta triangle points detected in the previous step according to the Euclidean distance, and count the number N _m of singular points contained in each cluster, N _m <N (here The clusters where N is taken as 25) are deleted.

(f)求取剩余的各个聚类的平均突变程度，并按奇异点类型、突变程度大小分别对这些聚类进行排序，选取前M(本方法M＝3，core点聚类和delta点聚类都取3个，当实际聚类个数少于3个时取实际聚类个数)个突变程度较大的聚类作为候选奇异点聚类，候选奇异点最终就定位在这些聚类的质心上。(f) Calculate the average degree of mutation of the remaining clusters, and sort these clusters according to the type of singular point and the degree of mutation, and select the top M (this method M=3, core point clustering and delta point clustering If the number of clusters is less than 3, the actual number of clusters is taken as the number of clusters with a large mutation degree as candidate singular point clusters, and the candidate singular points are finally located in the clusters of these clusters. on the centroid.

(6)第二阶段：候选奇异点的去伪，如图2所示，包含以下几个步骤：(6) The second stage: the de-falsification of candidate singular points, as shown in Figure 2, includes the following steps:

(d)对每个候选奇异点，按公式(2)计算其周围半径为6τ的圆形邻域内每个像素的分布一致性coherence值(τ为平均纹线距离)。(d) For each candidate singular point, calculate the distribution consistency coherence value of each pixel in a circular neighborhood with a radius of 6τ (τ is the average ridge distance) according to formula (2).

(e)然后按图1所示的模板将该圆形区域划分为32个扇形和一个中心小圆形区域并计算各个部分的平均分布一致性coherence值，圆形区域的半径和两个圆带的宽度均为2τ。(e) Then divide the circular area into 32 sectors and a central small circular area according to the template shown in Figure 1 and calculate the average distribution consistency coherence value of each part, the radius of the circular area and two circular bands The width of is 2τ.

(f)最后，比较16个方向每个方向上的三个区域的平均分布一致性coherence值的大小(中心圆形区域为16个方向公共的区域)，若自里向外分布一致性coherence值越来越大，则标记该方向为有效方向。对某一候选奇异点而言，若有效方向数大于等于10，则提取该点为奇异点，否则，认为该点为伪点。若奇异点的某些方向出了指纹边界则将剩余的方向作为参考方向，在这些方向上计算有效方向。此外，经实验分析，计算每个象素分布一致性coherence值的窗口大小取为(4τ+1)×(4τ+1)较为合适。具体的算法流程图如图2所示。(f) Finally, compare the average coherence value of the three areas in each direction of the 16 directions (the central circular area is the common area of the 16 directions), if the coherence value is distributed from the inside to the outside If it becomes larger and larger, this direction is marked as a valid direction. For a candidate singular point, if the number of effective directions is greater than or equal to 10, the point is extracted as a singular point, otherwise, the point is considered as a false point. If some directions of the singular point are out of the fingerprint boundary, the remaining directions are taken as reference directions, and effective directions are calculated on these directions. In addition, through experimental analysis, it is more appropriate to take (4τ+1)×(4τ+1) as the window size for calculating the coherence value of each pixel distribution consistency. The specific algorithm flow chart is shown in Figure 2.

第一阶段，所述步骤(a)中，计算每一个点(i，j)的Poincare index值时按如下的公式进行：In the first stage, in the step (a), the Poincare index value of each point (i, j) is calculated according to the following formula:

i′＝(i+1)mod N_ψ，i'=(i+1)mod N _ψ ,

所述步骤(a)中，奇异点条件是指本方法提出的奇异点应满足的条件，该条件是对公式(1)的计算过程进行约束，具体是指，在沿封闭曲线做方向角变化的累加时附加以下限制条件：In the step (a), the singular point condition refers to the condition that the singular point proposed by this method should satisfy, and this condition is to constrain the calculation process of the formula (1), specifically refers to the direction angle change along the closed curve The following restrictions are added to the accumulation of :

(2)统计这Nψ个差中绝对值

(3)如果最终Poincare index庞加莱指数的值为1/2，那么该给定点(i，j)就被确定为core核心点，如果Poincare index庞加莱指数值为-1/2，那么该给定点(i，j)就被确定为delta三角点点。(3) If the value of the final Poincare index Poincare index is 1/2, then the given point (i, j) is determined as the core core point, if the value of the Poincare index Poincare index is -1/2, then The given point (i, j) is determined as a delta triangle point.

第二阶段，所述步骤(a)中，本方法利用分布一致性coherence值来表达纹线的一致性信息，而分布一致性coherence值的计算采用了Gaussian-Hermite矩。本方法用四个矩来描述指纹的纹线一致性信息：M_0，1，M_1，0，M_0，3，M_3，0。并对这四个不同阶的Gaussian-Hermite高斯-埃尔米特矩做如下定义：In the second stage, in the step (a), the method uses the coherence value of the distribution consistency to express the consistency information of the ridges, and the calculation of the coherence value of the distribution consistency adopts Gaussian-Hermite moments. This method uses four moments to describe the ridge consistency information of the fingerprint: M _0,1 , M _1,0 , M _0,3 , M _3,0 . And the Gaussian-Hermite Gaussian-Hermite moments of these four different orders are defined as follows:

这里λ(0＜λ＜1)是不同阶的Gaussian-Hermite矩的联合权重系数(本方法中λ取值0.5)。利用定义(2)，指纹图像中的每一个象素(i，j)都可以计算得到一个特征向量[M_u，M_v]^T。在指纹奇异点区和非奇异点区[M_u，M_v]^T的分布具有不同的特征，在非奇异点区[M_u，M_v]^T沿着垂直于脊线的方向分布，在奇异点区[M_u，M_v]^T则均匀地分布在各个方向上，本方法采用主分量分析方法提取[M_u，M_v]^T的分布特性，[M_u，M_v]^T协方差矩阵C_M由下式定义：Here λ (0<λ<1) is the joint weight coefficient of Gaussian-Hermite moments of different orders (in this method, λ takes a value of 0.5). Using definition (2), each pixel (i, j) in the fingerprint image can be calculated to obtain a feature vector [M _u , M _v ] ^T . The distribution of [M _u , M _v ] ^T in the fingerprint singular point area and non-singular point area has different characteristics. In the non-singular point area [M _u , M _v ] ^T distributes along the direction perpendicular to the ridge line. The point area [M _u , M _v ] ^T is evenly distributed in all directions. This method uses the principal component analysis method to extract the distribution characteristics of [M _u , M _v ] ^T , and the covariance matrix of [M _u , M _v ] ^T C _M is defined by:

其中， $m_{u} = \frac{1}{n \times n} \underset{w}{Σ} M_{u},$ $m_{v} = \frac{1}{n \times n} \underset{w}{Σ} M_{v},$ δ(k)表示两相邻方向角的差值，Δ(k)表示对差值调整以后的结果，i′表示第i个点之后的下一个点，n×n是窗口W的大小，本方法n取值为4τ+1，τ为平均纹线距离。in, $m_{u} = \frac{1}{no \times no} \underset{w}{Σ} m_{u},$ $m_{v} = \frac{1}{no \times no} \underset{w}{Σ} m_{v},$ δ(k) represents the difference between two adjacent direction angles, Δ(k) represents the result after adjusting the difference, i' represents the next point after the i-th point, n×n is the size of the window W, this The value of method n is 4τ+1, and τ is the average ridge distance.

Claims

1. A new fingerprint singular point detection method is characterized in that its method is,

(1) Preparatory stage: background separation and direction field information calculation of the fingerprint image;

(2) The first stage: the extraction of candidate singular points, including the following steps:

(a) Calculate the corresponding Poincare index P _{G, C} (i, j) for each point (i, j) in the direction domain O′, using two lengths of 5×5 and a closed curve of 7×7 to calculate the Poincare index value of each point, as long as the calculation result of one of them meets the singular point condition, the point is considered as a candidate singular point, if the singular point detected by the two curves If the types are inconsistent, the point is directly identified as a false point;

(b) Use the clustering algorithm to cluster the candidate singular points detected in the previous step according to the Euclidean distance, and count the number N _m of singular points contained in each cluster, N _m <N (N represents the fingerprint pattern area The number of singular points that should be able to be detected, this number is stable, where N is taken as 25) clusters are deleted;

(c) Find the average mutation degree of the remaining clusters, and sort these clusters according to the type of singularity and the degree of mutation, and select the first M clusters with higher mutation degrees as candidate singularity clusters, Candidate singular points are finally positioned on the centroids of these clusters;

(3) The second stage: de-falsification of candidate singular points, including the following steps:

For each candidate singular point, calculate the distribution consistency coherence value of each pixel in a circular neighborhood with a radius of 6τ, where τ is the average ridge distance;

Then divide the circular area into 32 sectors and a central small circular area and calculate the average coherence value of each part. The radius of the circular area and the width of the two circular bands are both 2τ;

Finally, compare the average distribution consistency coherence value of the three areas in each of the 16 directions. The central circular area is the common area of the 16 directions. If the distribution consistency coherence value is getting larger and larger from the inside to the outside , then mark the direction as an effective direction; for a candidate singular point, if the number of effective directions is greater than or equal to 10, then extract the point as a singular point, otherwise, consider the point as a pseudo point; if some directions of the singular point Out of the fingerprint boundary, the remaining directions are used as reference directions, and effective directions are calculated on these directions.

2. new fingerprint singular point detection method as claimed in claim 1, is characterized in that, in the (a) of described first stage, calculate the Poincare index Poincare index value of each singular point (i, j) according to the following formula:

Poincare Poincare ((i i . . j j)) = = \frac{11}{22 π π} {Σ Σ}_{k k = = 00}^{Nψ Nψ - - 11} Δ Δ ((k k)),,

Δ Δ ((k k)) = = \{\begin{matrix} δ δ ((k k)),, & if if | | δ δ ((k k)) | | < < \frac{π π}{22} \\ π π + + δ δ ((k k)),, & ifδ ifδ ((k k)) \leq \leq - - \frac{π π}{22} \\ π π - - δ δ ((k k)),, & otherwise otherwise \end{matrix} - - - - - - ((11))

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

i'=(i+1)mod N _ψ ,

where ψ _x (i) and ψ _y (i) are the x and y coordinates of the kth point on a closed curve with Nψ pixels centered on a given point, respectively; δ(k) represents the difference between two adjacent orientation angles value, Δ(k) represents the result after adjusting the difference, i′ represents the next point after the i-th point; this algorithm uses a 5×5 and 7×7 square closed curve, and the formation of the formula is based on An improvement to the original PoincareIndex method, which is used to calculate the accumulation of direction angle changes on a closed curve corresponding to a certain point.

3. the new fingerprint singular point detection method as claimed in claim 1 or 2, is characterized in that, in the (a) of described first stage, singular point condition refers to the condition that singular point should satisfy, and this condition is to The calculation process of formula (1) is constrained, specifically, the following restrictions are added when accumulating the direction angle changes along the closed curve:

Count the number of sign changes in these Nψ directions. If the direction changes from positive to negative and from negative to positive, and there is only one time, continue to calculate its Poincare index value, otherwise, consider this point to be an ordinary point ;

Statistics of these Nψ differences

The number of differences, if the number is more than one, the point is considered to be an ordinary point;

If the value of the final Poincare index Poincaré index is 1/2, then the detected point (i, j) is determined as the core core point, if the Poincare index Poincare index value is -1/2, then the detected point (i, j) is determined as the core core point. The detection point (i, j) is determined as a delta triangle point.

4. the novel fingerprint singular point detection method as claimed in claim 1, is characterized in that, in the (b) of described first stage, N is the number of the singular point that fingerprint mode area should be able to detect, takes value 25 is based on the analysis of the test results. The number of singular points that can be detected in the singular point area of the fingerprint is relatively stable and concentrated in the innermost part of the singular point area. For low-quality fingerprint images, as long as the singular point area is not affected by large noise pollution, then the number of singular points that can be detected in the singular point area is also stable, and N is set to 25 in this method.

5. The novel fingerprint singular point detection method as claimed in claim 1, characterized in that, in (c) of the first stage, the degree of mutation refers to that in the process of making a difference between two adjacent direction angles on the closed curve,

6. The new fingerprint singular point detection method as claimed in claim 1, characterized in that, in the second stage, the coherence consistency value is used to express the consistency information of the lines, and the calculation of the coherence consistency value adopts The Gaussian-Hermite Gaussian-Hermitian moments; that is, use four moments to describe the consistency information of the fingerprints: M _0,1 , M _1,0 , M _0,3 , M _3,0 ; and for this The Gaussian-Hermite Gaussian-Hermite moments of four different orders are defined as follows:

\{\begin{matrix} {M m}_{u u} ((x x,, y the y)) = = λ λ {M m}_{1,0 1,0} ((x x,, y the y,, I I ((x x,, y the y)))) + + ((11 - - λ λ)) {M m}_{3,0 3,0} ((x x,, y the y,, I I ((x x,, y the y)))) \\ {M m}_{v v} ((x x,, y the y)) = = λ λ {M m}_{0,1 0,1} ((x x,, y the y,, I I ((x x,, y the y)))) + + ((11 - - λ λ)) {M m}_{0,3 0,3} ((x x,, y the y,, I I ((x x,, y the y)))) \end{matrix} - - - - - - ((22))

Here λ (0<λ<1) is the joint weight coefficient of Gaussian-Hermite Gaussian-Hermitian moments of different orders (λ takes a value of 0.5); using formula (2), each pixel in the fingerprint image (i , j) can be calculated to get a eigenvector [M _u , M _v ] ^T ; the distribution of [M u , M v ] T in the fingerprint singular point area and the non-singular point area [M _u , M _v ] ^T has different characteristics, and in the non-singular point area [ Mu _u , M _v ] ^T are distributed along the direction perpendicular to the ridge line, and {M _u , M _v ] ^T are evenly distributed in all directions in the singular point area, and [M _u , M _{v ] T} are extracted by principal component analysis method ] ^T distribution characteristics, [M _u , M _v ] ^T covariance matrix G _M is defined by the following formula:

{C C}_{M m} = = [\begin{matrix} \underset{w w}{Σ Σ} {(({M m}_{u u} - - {m m}_{u u}))}^{22} & \underset{w w}{Σ Σ} (({M m}_{u u} - - {m m}_{u u})) (({M m}_{v v} - - {m m}_{v v})) \\ \underset{w w}{Σ Σ} (({M m}_{u u} - - {m m}_{u u})) (({M m}_{v v} - - {m m}_{v v})) & \underset{w w}{Σ Σ} {(({M m}_{v v} - - {m m}_{v v}))}^{22} \end{matrix}] - - - - - - ((33))

in,

m_{u} = \frac{1}{no \times no} \underset{w}{Σ} m_{u},

m_{v} = \frac{1}{no \times no} \underset{w}{Σ} m_{v},

m _u represents the average M _u value in the window W, m _v represents the average M _v value in the window W, n×n is the size of the window W, the value of n in this method is 4τ+1, and τ is the average ridge distance;

Let λ1 and λ2 be the two eigenvalues of the covariance matrix G _M , then when λ1>>λ2, the distribution of [M _u , M _v ] ^T is mainly distributed along the long axis, that is, along the direction perpendicular to the ridge The direction distribution of the line, and the values of λ1 and λ2 in the noise area or singular point area are very close. Therefore, the distribution consistency coherence characteristics of [M _u , M _v ] ^T are defined as follows:

coherence coherence = = \frac{{λ λ}_{11}^{22} - - {λ λ}_{22}^{22}}{{λ λ}_{11}^{22} + + {λ λ}_{22}^{22}} - - - - - - ((44))

= = \frac{((\underset{w w}{Σ Σ} {(({M m}_{u u} - - {m m}_{u u}))}^{22} + + \underset{w w}{Σ Σ} {(({M m}_{v v} - - {m m}_{v v}))}^{22})) \sqrt{{((\underset{w w}{Σ Σ} {(({M m}_{u u} - - {m m}_{u u}))}^{22} - - \underset{w w}{Σ Σ} {(({M m}_{v v} - - {m m}_{v v}))}^{22}))}^{22} + + 44 {((\underset{w w}{Σ Σ} (({M m}_{u u} - - {m m}_{u u})) (({M m}_{v v} - - {m m}_{v v}))))}^{22}}}{{((\underset{w w}{Σ Σ} {(({M m}_{u u} - - {m m}_{u u}))}^{22}))}^{22} + + {((\underset{w w}{Σ Σ} {(({M m}_{v v} - - {m m}_{v v}))}^{22}))}^{22} + + 22 {((\underset{w w}{Σ Σ} (({M m}_{u u} - - {m m}_{u u})) (({M m}_{v v} - - {m m}_{v v}))))}^{22}}

Therefore, the better the consistency of the ridges, the greater the coherence of the distribution consistency.

7. the novel fingerprint singular point detection method as claimed in claim 1 is characterized in that, in the second stage, the window size of each pixel distribution consistency coherence value is taken as (4τ+1)*(4τ +1).