CN101329723B

CN101329723B - A Fast and Robust Location Method for Fingerprint Core Points

Info

Publication number: CN101329723B
Application number: CN200810020824XA
Authority: CN
Inventors: 曹国; 孙权森; 夏德深
Original assignee: Nanjing University of Science and Technology
Current assignee: Nanjing University of Science and Technology
Priority date: 2008-08-04
Filing date: 2008-08-04
Publication date: 2010-06-02
Anticipated expiration: 2028-08-04
Also published as: CN101329723A

Abstract

The invention discloses a fast and robust positioning method for fingerprint core points. The steps of the method are as follows: input image segmentation; calculation of direction field; smoothing of direction field; edge detection; within the threshold range [T _l , T _h ], delete the pixel from the edge; core point positioning: use the outermost pixels of the 3×3 neighborhood window to calculate the Con, dx, and dy values of the remaining edge points respectively , the condition for positioning the core point is: when dx>α, dy<β, select the point with the smallest direction consistency as the Lower core point, and when dx<β, dy>α, select the point with the smallest direction consistency as It is the core point of Upper. The invention can not only effectively suppress the influence of noise, but also quickly and accurately locate the position of the core point of the fingerprint, and can reliably detect the core point for all types of fingerprints.

Description

A Fast and Robust Location Method for Fingerprint Core Points

技术领域technical field

本发明属于指纹识别方法，尤其涉及一种指纹核心点的快速鲁棒定位方法。The invention belongs to a fingerprint recognition method, in particular to a fast and robust positioning method of a fingerprint core point.

背景技术Background technique

在众多的生物识别系统中，指纹识别系统由于其体积小、成本低、易操作、可靠性高等优点越来越受到人们的青睐，相应地，基于指纹识别技术的产品市场需求正在日益扩大，应用也越来越广泛。Among many biometric identification systems, fingerprint identification systems are more and more popular due to their advantages of small size, low cost, easy operation, and high reliability. Correspondingly, the market demand for products based on fingerprint identification technology is expanding day by day. is also becoming more widespread.

指纹图像中核心点被定义为最大曲率方向的点。在指纹识别系统中核心点常被用作参考点来提高指纹匹配的速度和性能，而指纹分类也大多根据指纹中核心点的类型、数目和相对位置等信息来实现。所以，准确、可靠的定位核心点及其方向是指纹识别系统中的一项关键技术。The core point in the fingerprint image is defined as the point in the direction of maximum curvature. In the fingerprint identification system, core points are often used as reference points to improve the speed and performance of fingerprint matching, and fingerprint classification is mostly realized according to the type, number and relative position of core points in fingerprints. Therefore, accurate and reliable positioning of the core point and its direction is a key technology in the fingerprint identification system.

在指纹图像核心点检测的各种不同方法中，Poincare索引方法是较为经典的方法，许多学者针对该方法进行了很多改进，但是Poincare索引方法对于拱型指纹无法检测到核心点，易受噪声影响等问题依然存在(Asker M.Bazen and Sabih H.Gerez，SystematicMethods for the Computation of the Directional Fields and Singular Points of Fingerprints，IEEE Trans.Pattern Analysis and Machine Intelligence，2002，24(7)：905-919.)。Jain等人提出了Sine-Map方法，该方法根据核心点的属性建立模型通过多分辨率分析进行检测，这种方法需要计算多次方向场，计算复杂度很高，并且也不适用于指纹旋转的情况(Jain，A.K.，S.Prabhakar，Hong，L.and Pankanti，S.，Filterbank-Based Fingerprint Matching，IEEETrans.Image Processing，2000，9(5)：846-859.)。其他基于数学模型(Yi Wang，Jiankun Hu，and Damien Phillips，A Fingerprint Orientation Model Based on 2D Fourier Expansion(FOMFE)and Its Application to Singular-Point Detection and Fingerprint Indexing，IEEETrans.Pattern Analysis and Machine Intelligence，2007，29(4)：573-585.)、指纹结构(XuchuWang，Jianwei Li and Yanmin Niu，Definition and extraction of stable points fromfingerprint images，Pattern Recognition，2006，40：1804-1815.)以及多尺度分析(ManhuaLiu，Xudong Jiang，and Kot，A.C.，Fingerprint Reference-Point Detection，EURASIP Journalon Applied Signal Processing，2005，4：498-509.)的方法，都需要较长的处理时间，并且当图像质量较差时核心点定位误差较大，这些缺点对嵌入式指纹识别系统的性能带来了较大影响。Among the different methods for core point detection of fingerprint images, the Poincare index method is a more classic method. Many scholars have made many improvements to this method, but the Poincare index method cannot detect core points for arched fingerprints and is easily affected by noise. Such problems still exist (Asker M.Bazen and Sabih H.Gerez, Systematic Methods for the Computation of the Directional Fields and Singular Points of Fingerprints, IEEE Trans. Pattern Analysis and Machine Intelligence, 2002, 24(7): 905-919.) . Jain et al. proposed the Sine-Map method, which builds a model based on the attributes of the core points for detection through multi-resolution analysis. This method needs to calculate the direction field multiple times, and the computational complexity is very high, and it is not suitable for fingerprint rotation. (Jain, A.K., S. Prabhakar, Hong, L. and Pankanti, S., Filterbank-Based Fingerprint Matching, IEEE Trans. Image Processing, 2000, 9(5): 846-859.). Others based on mathematical models (Yi Wang, Jiankun Hu, and Damien Phillips, A Fingerprint Orientation Model Based on 2D Fourier Expansion (FOMFE) and Its Application to Singular-Point Detection and Fingerprint Indexing, IEEETrans.Pattern Analysis and Machine Intelligence, 2007, 29 (4):573-585.), fingerprint structure (XuchuWang, Jianwei Li and Yanmin Niu, Definition and extraction of stable points from fingerprint images, Pattern Recognition, 2006, 40:1804-1815.) and multi-scale analysis (ManhuaLiu, Xudong Jiang, and Kot, A.C., Fingerprint Reference-Point Detection, EURASIP Journalon Applied Signal Processing, 2005, 4: 498-509.) method, all require a long processing time, and when the image quality is poor, the core point positioning error Larger, these shortcomings have a greater impact on the performance of the embedded fingerprint recognition system.

发明内容Contents of the invention

本发明的目的在于提供一种实现指纹核心点定位的快速鲁棒方法。The purpose of the present invention is to provide a fast and robust method for locating fingerprint core points.

实现本发明目的的技术方案为：一种指纹核心点快速鲁棒定位方法，步骤如下：The technical solution for realizing the purpose of the present invention is: a fast and robust positioning method for fingerprint core points, the steps are as follows:

第一步，输入图像分割：将指纹图像分成不重叠的12×12象素大小的块，运用灰度方差和均值信息将图像划分为前景区域F与背景区域B，并将前景区域F内缩12个象素；The first step is to segment the input image: the fingerprint image is divided into non-overlapping 12×12 pixel blocks, and the image is divided into the foreground area F and the background area B by using the gray variance and mean value information, and the foreground area F is indented 12 pixels;

第二步，计算方向场：将指纹图像分成不重叠的大小为w×w的象素块，利用Sobel算子计算每个块(i，j)中每个象素(u，v)的梯度G_x和G_y，然后计算每一小块图像的方向θ(i，j)，即The second step is to calculate the direction field: divide the fingerprint image into non-overlapping pixel blocks of size w×w, and use the Sobel operator to calculate the gradient of each pixel (u, v) in each block (i, j) G _x and G _y , and then calculate the direction θ(i, j) of each small image, namely

$θ θ ((i i,, j j)) = = \frac{11}{22} {tan the tan}^{- - 11} [[\frac{{Σ Σ}_{u u = = i i - - w w / / 22}^{i i + + w w / / 22} {Σ Σ}_{v v = = j j - - w w / / 22}^{j j + + w w / / 22} 22 Gx Gx ((u u,, v v)) Gy Gy ((u u,, v v))}{{Σ Σ}_{u u = = i i - - w w / / 22}^{i i + + w w / / 22} {Σ Σ}_{v v = = j j - - w w / / 22}^{j j + + w w / / 22} ((G G {x x}^{22} ((u u,, v v)) - - G G {y the y}^{22} ((u u,, v v))))}]],,$

当图像中划分的每一象素块的方向都计算完后，便得到了方向场O；When the direction of each pixel block divided in the image is calculated, the direction field O is obtained;

第三步，方向场平滑：定义方向一致性，即The third step, direction field smoothing: define direction consistency, namely

$((s the s)) = = \frac{\sqrt{{(({Σ Σ}_{((i i,, j j)) &Element; &Element; Ω Ω ((s the s))} cos cos ((22 θ θ ((i i,, j j))))))}^{22} + + {(({Σ Σ}_{((i i,, j j)) &Element; &Element; Ω Ω ((s the s))} sin sin ((22 θ θ ((i i,, j j))))))}^{22}}}{M m}$

其中，Ω(s)为块方向场中的点(i，j)的邻域，M为邻域内点的数目，窗口大小为s×s，s从3开始，计算不同大小窗口的Con(s)值，根据Con(s)值分为两种情况处理：(1)如果计算的Con(s)值小于给定的阈值T_con或小于Con(s-2)，则s增加2；若s达到s_max值，则令s＝3；Among them, Ω(s) is the neighborhood of point (i, j) in the block direction field, M is the number of points in the neighborhood, the window size is s×s, s starts from 3, and the Con(s ) value, according to Con(s) value is divided into two cases: (1) If the calculated Con(s) value is less than the given threshold T _con or less than Con(s-2), then s increases by 2; if s To reach the s _max value, let s=3;

(2)若在计算过程中Con(s)大于给定阈值T_con或是大于Con(s-2)，则令s取当前值；(2) If Con(s) is greater than a given threshold T _con or greater than Con(s-2) during the calculation process, then let s take the current value;

根据上述两种情况设置的s值重新计算块方向场中的点(i，j)方向，即Recalculate the direction of point (i,j) in the block direction field according to the s value set in the above two cases, i.e.

${θ θ}^{' '} ((i i,, j j)) = = \frac{11}{22} arctan arctan ((\frac{\underset{((i i,, j j)) &Element; &Element; Ω Ω ((s the s))}{Σ Σ} sin sin ((22 θ θ ((i i,, j j))))}{\underset{((i i,, j j)) &Element; &Element; Ω Ω ((s the s))}{Σ Σ} cos cos ((22 θ θ ((i i,, j j))))})),,$

当对方向场O中的所有点遍历并重新计算方向后，得到平滑后的方向场O′；After traversing and recalculating the direction for all points in the direction field O, the smoothed direction field O' is obtained;

第四步，边缘检测：在指纹的前景区域F内，对于平滑后的方向场O′，利用Max-Min算子在3×3邻域窗口内进行边缘检测，如果方向场O′内所进行边缘检测的点的梯度高于阈值T_o，则该点标记为边缘象素；The fourth step, edge detection: in the foreground area F of the fingerprint, for the smoothed direction field O', use the Max-Min operator to perform edge detection in a 3×3 neighborhood window, if the direction field O' is If the gradient of the point detected by the edge is higher than the threshold T _o , then the point is marked as an edge pixel;

第五步，边缘象素删除：对于边缘象素点，运用其4邻域计算梯度信息，如果计算的4个梯度值均不在给定阈值范围[T_l，T_h内，则从边缘中删除该象素；The fifth step, edge pixel deletion: For the edge pixel point, use its 4 neighbors to calculate the gradient information, if the calculated 4 gradient values are not within the given threshold range [T _l , T _h , then delete it from the edge the pixel;

第六步，核心点定位：利用3×3邻域窗口的最外围象素，分别计算剩余边缘点(i，j)的Con，dx，dy值，其中Con为方向一致性度量，即The sixth step, core point positioning: use the outermost pixels of the 3×3 neighborhood window to calculate the Con, dx, dy values of the remaining edge points (i, j), where Con is the direction consistency measure, that is

$Con Con ((i i,, j j)) = = ((\sqrt{{(({Σ Σ}_{((u u,, v v)) &Element; &Element; Ω Ω} cos cos ((22 θ θ ((u u,, v v))))))}^{22} + + {(({Σ Σ}_{((u u,, v v)) &Element; &Element; Ω Ω} sin sin ((22 θ θ ((u u,, v v))))))}^{22}})) / / 88,,$

$dx dx ((i i,, j j)) = = {Σ Σ}_{v v = = - - 11}^{11} cos cos ((22 θ θ ((i i - - 11,, v v)))) - - {Σ Σ}_{v v = = - - 11}^{11} cos cos ((22 θ θ ((i i + + 11,, v v)))),,$

$dy dy ((i i,, j j)) = = {Σ Σ}_{u u = = - - 11}^{11} sin sin ((22 θ θ ((u u,, i i - - 11)))) - - {Σ Σ}_{u u = = - - 11}^{11} sin sin ((22 θ θ ((u u,, i i + + 11)))),, ((u u,, v v)) &Element; &Element; Ω Ω,,$

给出核心点定位的条件为：当dx＞α，dy＜β时，选择方向一致性最小的点即为下核心点，当dx＜β，dy＞α时选择方向一致性最小的点即为上核心点，其中α、β的值是利用FVC2002指纹数据库进行核心点定位实验，最终根据实验结果选取具体数值。The conditions for positioning the core point are: when dx>α, dy<β, the point with the smallest direction consistency is selected as the lower core point, and when dx<β, dy>α, the point with the smallest direction consistency is selected as On the core point, the values of α and β are the core point positioning experiments using the FVC2002 fingerprint database, and finally select specific values according to the experimental results.

本发明与现有技术相比，其显著优点：(1)为开发一种快速的基于嵌入式系统的指纹识别系统提供基础。在嵌入式指纹识别系统中，利用检测的核心点作为参考点建立细节点模板，从而消除指纹间的平移影响，使得指纹图像能够快速配准，然后进行细节点匹配识别。(2)由于采用了平滑的方向场信息，显著提高了指纹识别的抗噪声能力。(3)核心点定位时，运用了边缘信息，对边缘点进行了有效的删除，极大了缩短了核心点定位的时间。(4)提出的方法对于指纹旋转具有较好的鲁棒性，同时对于所有类型的指纹均能够检测到核心点，显著提高了指纹的识别性能。Compared with the prior art, the present invention has the following remarkable advantages: (1) It provides a basis for developing a fast fingerprint identification system based on an embedded system. In the embedded fingerprint identification system, the detected core points are used as reference points to establish minutiae templates, so as to eliminate the impact of translation between fingerprints, so that fingerprint images can be quickly registered, and then minutiae matching and identification are performed. (2) Due to the use of smooth direction field information, the anti-noise ability of fingerprint recognition is significantly improved. (3) When locating the core point, the edge information is used to effectively delete the edge point, which greatly shortens the time for locating the core point. (4) The proposed method is robust to fingerprint rotation, and can detect core points for all types of fingerprints, which significantly improves the fingerprint recognition performance.

下面结合附图对本发明作进一步详细描述。The present invention will be described in further detail below in conjunction with the accompanying drawings.

附图说明Description of drawings

图1是本发明指纹核心点快速鲁棒定位方法的流程图。Fig. 1 is a flow chart of the fast and robust positioning method for fingerprint core points of the present invention.

图2是原始指纹图像。Figure 2 is the original fingerprint image.

图3是指纹分割结果图。Figure 3 is a graph of fingerprint segmentation results.

图4是计算的方向场图。Figure 4 is the calculated direction field diagram.

图5是平滑后的方向场图。Figure 5 is the smoothed direction field diagram.

图6是边缘检测图。Figure 6 is an edge detection diagram.

图7象素点四邻域示意图。Figure 7 is a schematic diagram of the four-neighborhood of a pixel point.

图8是边缘象素删除后的图。Fig. 8 is a picture after edge pixels are deleted.

图9是核心点定位结果图。Figure 9 is a diagram of the core point positioning results.

图10是指纹旋转前后核心点定位结果图。Figure 10 is a diagram of the core point positioning results before and after fingerprint rotation.

图11是拱形指纹核心点定位结果图。Fig. 11 is a diagram of the positioning result of the core point of the arched fingerprint.

具体实施方式Detailed ways

结合图1，本发明指纹核心点快速鲁棒定位方法包含下列步骤：In conjunction with Fig. 1, the fingerprint core point fast and robust positioning method of the present invention comprises the following steps:

第一步，指纹图像分割：将图像分成不重叠的12×12大小的象素块，计算每一块的均值mean和方差var，当var＜T_var且mean＜M₁时或者(mean-M₁)×8+var＜0时，设置该图像块为背景，否则设置为图像前景区域F，即指纹有效区域，其中T_var为给定阈值，根据经验设置T_var＝80，M₁是整幅指纹图像的均值，图像分割后再将有效区域边界内缩12个象素，以避免背景区域的影响，图2是输入的指纹图像，图3给出了图像分割的结果，其中白色区域为指纹的有效区域。The first step, fingerprint image segmentation: divide the image into non-overlapping 12×12 pixel blocks, calculate the mean mean and variance var of each block, when var<T _var and mean<M ₁ or (mean-M ₁ )×8+var<0, set the image block as the background, otherwise set it as the foreground area F of the image, that is, the effective area of the fingerprint, where T _var is a given threshold, set T _var =80 according to experience, and M ₁ is the entire image The average value of the fingerprint image, after image segmentation, shrink the effective area boundary by 12 pixels to avoid the influence of the background area. Figure 2 is the input fingerprint image, and Figure 3 shows the result of image segmentation, in which the white area is the fingerprint effective area.

第二步，计算方向场：指纹图像具有很强的方向性，计算方向场时，通常将图像分块，然后计算每一块的方向作为指纹脊线的方向。这里将图像分成不重叠的大小为w×w的象素块，如w可以为5、7、9、11等，当w＝5，利用Sobel算子计算每个块(i，j)中每个象素(u，v)的梯度G_x和G_y，G_x和G_y分别是x和y方向上的梯度，然后计算每一小块图像的方向θ(i，j)，即The second step is to calculate the direction field: the fingerprint image has strong directionality. When calculating the direction field, the image is usually divided into blocks, and then the direction of each block is calculated as the direction of the fingerprint ridge. Here, the image is divided into non-overlapping pixel blocks with a size of w×w, such as w can be 5, 7, 9, 11, etc., when w=5, use the Sobel operator to calculate each block (i, j) The gradients G _x and G _y of pixels (u, v), G _x and G _y are the gradients in the x and y directions respectively, and then calculate the direction θ(i, j) of each small block image, that is

$θ θ ((i i,, j j)) = = \frac{11}{22} {tan the tan}^{- - 11} [[\frac{{Σ Σ}_{u u = = i i - - w w / / 22}^{i i + + w w / / 22} {Σ Σ}_{v v = = j j - - w w / / 22}^{j j + + w w / / 22} 22 Gx Gx ((u u,, v v)) Gy Gy ((u u,, v v))}{{Σ Σ}_{u u = = i i - - w w / / 22}^{i i + + w w / / 22} {Σ Σ}_{v v = = j j - - w w / / 22}^{j j + + w w / / 22} ((G G {x x}^{22} ((u u,, v v)) - - G G {y the y}^{22} ((u u,, v v))))}]] - - - - - - ((11))$

当图像中划分的每一象素块的方向都计算完后，便得到了方向场O，计算得到的方向场规范化为灰度图像后如图4所示。When the direction of each pixel block divided in the image is calculated, the direction field O is obtained, and the calculated direction field is normalized into a grayscale image as shown in Figure 4.

第三步，方向场平滑：计算出的方向场对噪声敏感，需要加以平滑，采用低通滤波方法会导致核心点定位不准确，这里采用自适应的多窗口方法，定义方向一致性，即The third step is to smooth the direction field: the calculated direction field is sensitive to noise and needs to be smoothed. Using the low-pass filtering method will lead to inaccurate positioning of the core points. Here, an adaptive multi-window method is used to define the direction consistency, namely

$((s the s)) = = \frac{\sqrt{{(({Σ Σ}_{((i i,, j j)) &Element; &Element; Ω Ω ((s the s))} cos cos ((22 θ θ ((i i,, j j))))))}^{22} + + {(({Σ Σ}_{((i i,, j j)) &Element; &Element; Ω Ω ((s the s))} sin sin ((22 θ θ ((i i,, j j))))))}^{22}}}{M m} - - - - - - ((22))$

其中，Ω(s)为块方向场中点(i，j)的邻域，窗口大小为s×s，M为邻域内点的数目，s从3开始，计算不同大小窗口的Con(s)值，根据Con(s)值分为两种情况处理：(1)如果计算的Con(s)值小于给定的阈值T_com或小于Con(s-2)，则s增加2；若s达到s_max＝9的，则令s＝3。Among them, Ω(s) is the neighborhood of the point (i, j) in the block direction field, the window size is s×s, M is the number of points in the neighborhood, s starts from 3, and the Con(s) of different size windows is calculated value, according to the Con(s) value is divided into two cases: (1) If the calculated Con(s) value is less than the given threshold T _com or less than Con(s-2), then s increases by 2; if s reaches If s _max =9, then set s=3.

(2)若在计算过程中Con(s)大于给定阈值T_com或是大于Con(s-2)，则令s取当前值。(2) If Con(s) is greater than a given threshold T _com or greater than Con(s-2) during the calculation process, let s take the current value.

根据上述设置的s值通过公式(3)重新计算块方向场中点(i，j)方向，根据经验设置T_com＝0.5，当对方向场O中的所有点遍历并重新计算方向后，得到平滑后的方向场O′。图5为平滑方向场后以灰度图显示的结果，与图4所示结果进行比较，从图5中可以看出，指纹噪声得到了有效的抑制，各个区域变得比较平滑，使得后续核心点定位受噪声影响较小，定位结果更加稳定。Recalculate the direction of the center point (i, j) of the block direction field through formula (3) according to the s value set above, and set T _com =0.5 according to experience. After traversing and recalculating the direction for all points in the direction field O, we get The smoothed direction field O'. Figure 5 shows the results displayed in grayscale after smoothing the direction field. Compared with the results shown in Figure 4, it can be seen from Figure 5 that the fingerprint noise has been effectively suppressed, and each area has become smoother, making the subsequent core Point positioning is less affected by noise, and the positioning result is more stable.

${θ θ}^{' '} ((i i,, j j)) = = \frac{11}{22} arctan arctan ((\frac{\underset{((i i,, j j)) &Element; &Element; Ω Ω ((s the s))}{Σ Σ} sin sin ((22 θ θ ((i i,, j j))))}{\underset{((i i,, j j)) &Element; &Element; Ω Ω ((s the s))}{Σ Σ} cos cos ((22 θ θ ((i i,, j j))))})) - - - - - - ((33))$

第四步，边缘检测：在指纹的前景区域F内，对于平滑的方向场O′，利用Max-Min算子以块方向为计算单元在3×3邻域窗口内进行边缘检测，如果梯度高于阈值T_o(根据经验设置T_o＝2.15)，则该点标记为边缘象素。边缘检测后规范化为灰度图如图6所示，图中存在两条黑色的边缘，一条边缘起始于图像中上部，另一边缘位于图像中下部。The fourth step, edge detection: in the foreground area F of the fingerprint, for the smooth direction field O′, use the Max-Min operator to perform edge detection in the 3×3 neighborhood window with the block direction as the calculation unit, if the gradient is high At the threshold T _o (set T _o = 2.15 based on experience), the point is marked as an edge pixel. After edge detection, it is normalized into a grayscale image as shown in Figure 6. There are two black edges in the image, one edge starts at the upper middle of the image, and the other edge is located at the middle and lower part of the image.

第五步，边缘象素删除：对于检测的边缘象素点，运用其4邻域计算梯度信息，4邻域关系如附图7所示，如果计算的4个梯度值均不在给定阈值范围[T_l，T_h]内，其中T_l＝0.55，T_h＝1.75，则从边缘中删除该象素。见图8所示，边缘象素删除后用白色显示，可以看出边缘象素删除后只有少数黑色的边缘象素点得到保留，从而大大减少了核心点定位的搜索空间，继而极大缩短了核心点定位时间。The fifth step, edge pixel deletion: For the detected edge pixel point, use its 4 neighborhoods to calculate the gradient information, the relationship between the 4 neighborhoods is shown in Figure 7, if the calculated 4 gradient values are not within the given threshold range Within [T _l , T _h ], where T _l = 0.55, _Th = 1.75, the pixel is removed from the edge. As shown in Figure 8, after the edge pixels are deleted, they are displayed in white. It can be seen that only a few black edge pixels are retained after the edge pixels are deleted, which greatly reduces the search space for core point positioning, and then greatly shortens the Core point positioning time.

第六步，核心点定位：对于保留下的边缘象素点，依据下述公式(4-6)，利用3×3邻域窗口的最外围象素，分别计算剩余边缘点(i，j)的Con dx，dy值，其中Con为方向一致性度量，即The sixth step, core point positioning: for the remaining edge pixels, according to the following formula (4-6), use the outermost pixels of the 3×3 neighborhood window to calculate the remaining edge points (i, j) respectively Con dx, dy value, where Con is the direction consistency measure, that is

$Con Con ((i i,, j j)) = = ((\sqrt{{(({Σ Σ}_{((u u,, v v)) &Element; &Element; Ω Ω} cos cos ((22 θ θ ((u u,, v v))))))}^{22} + + {(({Σ Σ}_{((u u,, v v)) &Element; &Element; Ω Ω} sin sin ((22 θ θ ((u u,, v v))))))}^{22}})) / / 88 - - - - - - ((44))$

$dx dx ((i i,, j j)) = = {Σ Σ}_{v v = = - - 11}^{11} cos cos ((22 θ θ ((i i - - 11,, v v)))) - - {Σ Σ}_{v v = = - - 11}^{11} cos cos ((22 θ θ ((i i + + 11,, v v)))) - - - - - - ((55))$

$dy dy ((i i,, j j)) = = {Σ Σ}_{u u = = - - 11}^{11} sin sin ((22 θ θ ((u u,, i i - - 11)))) - - {Σ Σ}_{u u = = - - 11}^{11} sin sin ((22 θ θ ((u u,, i i + + 11)))),, ((u u,, v v)) &Element; &Element; Ω Ω - - - - - - ((66))$

给出核心点定位的条件为：当dx＞α，dy＜β时，选择方向一致性最小的点即为Lower核心点，当dx＜β，dy＞α时选择方向一致性最小的点即为Upper核心点。α、β的值是利用FVC2002指纹数据库进行核心点定位实验，最终根据实验结果选取具体数值，如α＝0.1，β＝-0.1，即核心点定位条件是：The conditions for positioning the core point are: when dx>α, dy<β, select the point with the smallest direction consistency as the Lower core point, and when dx<β, dy>α, select the point with the smallest direction consistency as Upper core point. The values of α and β are core point positioning experiments using the FVC2002 fingerprint database, and finally select specific values according to the experimental results, such as α=0.1, β=-0.1, that is, the core point positioning conditions are:

当dx＞0.1，dy＜-0.1时，选择方向一致性最小的点即为Lower核心点；When dx>0.1, dy<-0.1, select the point with the smallest direction consistency as the Lower core point;

当dx＜-0.1，dy＞0.1时，选择方向一致性最小的点即为Upper核心点。When dx<-0.1, dy>0.1, select the point with the smallest direction consistency as the Upper core point.

图9给出了一幅指纹核心点检测的结果。Figure 9 shows a result of fingerprint core point detection.

图10给出了指纹旋转前后核心点定位结果对照图。图10.a为一幅指纹图像，图10.f为图10.a顺时针旋转90度的结果图像，图10.b和图10.g分别为相应平滑后的方向场转化为灰度图显示的结果，从这两个图中可以看出方向场差异较大；图10.c和图10.h为对应的边缘检测图，两个图中的边缘位置各不相同，图10.d和图10.i为边缘象素删除后的结果，图10.e为图10.a的核心点定位结果图，图10.j为图10.f的核心点定位结果图。从图10中可以看出，虽然图像进行了旋转，得到的边缘也不一样，但是最终定位的核心点位置却基本一致。利用FVC2002指纹数据库进行实验，结果表明核心点定位对于指纹旋转具有较好的鲁棒性。Figure 10 shows the comparison diagram of the core point positioning results before and after the fingerprint rotation. Figure 10.a is a fingerprint image, Figure 10.f is the result image rotated 90 degrees clockwise in Figure 10.a, and Figure 10.b and Figure 10.g are the corresponding smoothed direction fields converted into grayscale images From the displayed results, it can be seen that the direction field is quite different from these two figures; Figure 10.c and Figure 10.h are the corresponding edge detection images, and the edge positions in the two figures are different, and Figure 10.d And Fig. 10.i is the result after the edge pixels are deleted, Fig. 10.e is the core point positioning result map of Fig. 10.a, and Fig. 10.j is the core point positioning result map of Fig. 10.f. It can be seen from Figure 10 that although the image is rotated, the resulting edges are different, but the final core point positions are basically the same. Using the FVC2002 fingerprint database for experiments, the results show that the core point location has better robustness to fingerprint rotation.

对于拱形指纹，如图11.a，经过边缘检测后，结果如图11.b。边缘象素删除过程会将所有边缘象素删除，如图11.c，此时运用删除之前的边缘象素按照核心点定位条件进行核心点定位就可检测到正确的核心点位置，见图11.d。For arched fingerprints, as shown in Figure 11.a, after edge detection, the result is shown in Figure 11.b. The edge pixel deletion process will delete all edge pixels, as shown in Figure 11.c. At this time, the correct core point position can be detected by using the edge pixels before deletion according to the core point positioning conditions to locate the core point, as shown in Figure 11. .d.

Claims

1. A fast and robust location method for fingerprint core points, characterized in that the steps are as follows:

The first step is to segment the input image: the fingerprint image is divided into non-overlapping 12×12 pixel blocks, and the image is divided into the foreground area F and the background area B by using the gray variance and mean value information, and the foreground area F is indented 12 pixels;

The second step is to calculate the direction field: divide the fingerprint image into non-overlapping pixel blocks of size w×w, and use the Sobel operator to calculate the gradient of each pixel (u, v) in each block (i, j) G _x and G _y , and then calculate the direction θ(i, j) of each small image, namely

θ θ ((i i,, j j)) = = \frac{11}{22} {tan the tan}^{- - 11} [[\frac{{Σ Σ}_{u u = = i i - - w w / / 22}^{i i + + w w / / 22} {Σ Σ}_{v v = = j j - - w w / / 22}^{j j + + w w / / 22} 22 Gx Gx ((u u,, v v)) Gy Gy ((u u,, v v))}{{Σ Σ}_{u u = = i i - - w w / / 22}^{i i + + w w / / 22} {Σ Σ}_{v v = = j j - - w w / / 22}^{j j + + w w / / 22} (({Gx Gx}^{22} ((u u,, v v)) - - {Gy Gy}^{22} ((u u,, v v))))}]],,

When the direction of each pixel block divided in the image is calculated, the direction field O is obtained;

The third step, direction field smoothing: define direction consistency, namely

Con Con ((s the s)) = = \frac{\sqrt{{(({Σ Σ}_{((i i,, j j)) &Element; &Element; Ω Ω ((s the s))} cos cos ((22 θ θ ((i i,, j j))))))}^{22} + + {(({Σ Σ}_{((i i,, j j)) &Element; &Element; Ω Ω ((s the s))} sin sin ((22 θ θ ((i i,, j j))))))}^{22}}}{M m},,

Among them, Ω(s) is the neighborhood of point (i, j) in the block direction field, M is the number of points in the neighborhood, the window size is s×s, s starts from 3, and the Con(s ) value, according to Con(s) value is divided into two cases: (1) If the calculated Con(s) value is less than the given threshold T _con or less than Con(s-2), then s increases by 2; if s To reach the s _max value, let s=3;

(2) If Con(s) is greater than a given threshold T _con or greater than Con(s-2) during the calculation process, then let s take the current value;

Recalculate the direction of point (i,j) in the block direction field according to the s value set in the above two cases, i.e.

{θ θ}^{' '} ((i i,, j j)) = = \frac{11}{22} arctan arctan ((\frac{\underset{((i i,, j j)) &Element; &Element; Ω Ω ((s the s))}{Σ Σ} sin sin ((22 θ θ ((i i,, j j))))}{\underset{((i i,, j j)) &Element; &Element; Ω Ω ((s the s))}{Σ Σ} cos cos ((22 θ θ ((i i,, j j))))})),,

After traversing and recalculating the direction for all points in the direction field O, the smoothed direction field O' is obtained;

The fourth step, edge detection: in the foreground area F of the fingerprint, for the smoothed direction field O', use the Max-Min operator to perform edge detection in a 3×3 neighborhood window, if the direction field O' is If the gradient of the point detected by the edge is higher than the threshold T _o , the point is marked as an edge pixel;

The fifth step, edge pixel deletion: For the edge pixel point, use its 4 neighbors to calculate the gradient information, if the calculated 4 gradient values are not within the given threshold range [T _l , T _h ], then from the edge delete the pixel;

The sixth step, core point positioning: use the outermost pixels of the 3×3 neighborhood window to calculate the Con dx and dy values of the remaining edge points (i, j) respectively, where Con is the direction consistency measure, namely

Con Con ((i i,, j j)) = = ((\sqrt{{(({Σ Σ}_{((u u,, v v)) &Element; &Element; Ω Ω} cos cos ((22 θ θ ((u u,, v v))))))}^{22} + + {(({Σ Σ}_{((u u,, v v)) &Element; &Element; Ω Ω} sin sin ((22 θ θ ((u u,, v v))))))}^{22}})) / / 88,,

dx dx ((i i,, j j)) = = {Σ Σ}_{v v = = - - 11}^{11} cos cos ((22 θ θ ((i i - - 11,, v v)))) - - {Σ Σ}_{v v = = - - 11}^{11} cos cos ((22 θ θ ((i i + + 11,, v v)))),,

dy dy ((i i,, j j)) = = {Σ Σ}_{u u = = - - 11}^{11} sin sin ((22 θ θ ((u u,, j j - - 11)))) - - {Σ Σ}_{u u = = - - 11}^{11} sin sin ((22 θ θ ((u u,, j j + + 11)))),, ((u u,, v v)) &Element; &Element; Ω Ω,,

The conditions for positioning the core point are: when dx>α, dy<β, select the point with the smallest direction consistency as the lower core point, and when dx<β, dy>α, select the point with the smallest direction consistency as On the core point, the values of α and β are the core point positioning experiments using the FVC2002 fingerprint database, and finally select specific values according to the experimental results.

2. The fingerprint core point fast and robust positioning method according to claim 1 is characterized in that: for part of the arched fingerprint image, the edge pixel deletion process will delete all edge pixels, this time using the edge image before deletion Elements perform core point positioning according to the core point positioning conditions.