CN101329723B - Method for rapidly positioning robust of finger print core point - Google Patents

Abstract

The invention discloses a fingerprint core point fast robust positioning method, comprising the following steps: images are input and divided; a direction field is calculated; the direction field is smoothed; edge detection is carried out; edge pixel is deleted; for edge pixel points, 4 neighborhoods thereof are applied to calculate gradient information and if the 4 gradient magnitudes calculatedare not in the range of a given threshold value (Tl, Th), the pixel is deleted from the edge; the core point is positioned: the outermost pixel of a 3 multiplied by 3 neighborhood window is utilized,the Con, dx and dy values of the rest edge points are calculated respectively and the conditions for positioning the core point are that: when dx is more than Alpha and dy is less than Beta, a point with minimum selected direction consistency is the Lower core point, and when dx is less than Beta and dy is more than Alpha, a point with minimum selected direction consistency is the Upper core point. The method of the invention can not only effectively restrain effect brought by noise, but also can quickly and accurately position the core point of the fingerprint, and reliably detect the core point for all types of fingerprints.

Description

Method for rapidly positioning robust of finger print core point

Technical field

The invention belongs to fingerprint identification method, relate in particular to a kind of method for rapidly positioning robust of finger print core point.

Background technology

In numerous biological recognition systems, fingerprint recognition system is because its volume is little, cost is low, easy to operate, favor that the reliability advantages of higher more and more is subjected to people, correspondingly, based on the produce market demand expansion day by day of fingerprint identification technology, use also more and more widely.

Core point is defined as the point of maximum curvature direction in the fingerprint image.Core point is often used as speed and the performance that reference point improves fingerprint matching in fingerprint recognition system, and fingerprint classification also realizes according to the information such as type, number and relative position of core point in the fingerprint mostly.So, locate core point and direction thereof accurately, reliably and be a gordian technique in the fingerprint recognition system.

In the various distinct methods that the fingerprint image core point detects, the Poincare indexing means is a classic methods comparatively, many scholars have carried out a lot of improvement at this method, but the Poincare indexing means can't detect core point for the arch form fingerprint, still there is (Asker M.Bazen and Sabih H.Gerez in problem such as easily affected by noise, 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.).People such as Jain have proposed the Sine-Map method, and this method is set up model according to the attribute of core point and detected by multiresolution analysis, and this method need be calculated the repeatedly field of direction, computation complexity is very high, and also is not suitable for situation (Jain, the A.K. of fingerprint rotation, S.Prabhakar, Hong, L.and Pankanti, S., Filterbank-Based Fingerprint Matching, IEEETrans.Image Processing, 2000,9 (5): 846-859.).Other are based on mathematical model (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.), dactylotype (XuchuWang, Jianwei Li and Yanmin Niu, Definition and extraction of stable points fromfingerprint images, Pattern Recognition, 2006,40:1804-1815.) and multiscale analysis (ManhuaLiu, Xudong Jiang, and Kot, A.C., Fingerprint Reference-Point Detection, EURASIP Journalon Applied Signal Processing, 2005,4:498-509.) method, all need the long processing time, and the core point positioning error is bigger when picture quality is relatively poor, these shortcomings have been brought considerable influence to the performance of embedded fingerprint recognition system.

Summary of the invention

The object of the present invention is to provide a kind of fast robust method that realizes the finger print core point location.

The technical scheme that realizes the object of the invention is: a kind of method for rapidly positioning robust of finger print core point, and step is as follows:

The first step, input picture is cut apart: fingerprint image is divided into the piece of nonoverlapping 12 * 12 pixel sizes, the utilization gray variance and all value information be foreground area F and background area B with image division, and with 12 pixels that contract in the foreground area F;

Second the step, the calculated direction field: fingerprint image is divided into the pixel block that nonoverlapping size is w * w, utilize the Sobel operator calculate each piece (i, j) in each pixel (u, gradient G v) _xAnd G _y, calculate then each small images direction θ (i, j), promptly

$\mathrm{\θ}(i,j)=\frac{1}{2}{\tan }^{-1}\left[\frac{{\mathrm{\Σ}}_{u=i-w/2}^{i+w/2}{\mathrm{\Σ}}_{v=j-w/2}^{j+w/2}2\mathrm{Gx}(u,v)\mathrm{Gy}(u,v)}{{\mathrm{\Σ}}_{u=i-w/2}^{i+w/2}{\mathrm{\Σ}}_{v=j-w/2}^{j+w/2}(G{x}^2(u,v)-G{y}^2(u,v))}\right],$

After the direction of each pixel block of dividing in the image has all been calculated, just obtained field of direction O;

In the 3rd step, the field of direction is level and smooth: definition direction consistance, promptly

$\left(s\right)=\frac{\sqrt{{\left({\mathrm{\Σ}}_{(i,j)\∈\mathrm{\Ω}\left(s\right)}\cos \left(2\mathrm{\θ}(i,j)\right)\right)}^2+{\left({\mathrm{\Σ}}_{(i,j)\∈\mathrm{\Ω}\left(s\right)}\sin \left(2\mathrm{\θ}(i,j)\right)\right)}^2}}{M}$

Wherein, Ω (s) is point (i, neighborhood j) in the piece field of direction, M is the number of point in the neighborhood, and window size is s * s, and s is since 3, calculate Con (s) value of different size windows, be divided into two kinds of situations according to Con (s) value and handle: (1) if Con (s) value of calculating less than given threshold value T _ConOr less than Con (s-2), then s increases by 2; If s reaches s _MaxValue then makes s=3;

(2) if Con (s) is greater than given threshold value T in computation process _ConOr, then make s get currency greater than Con (s-2);

According to the s value of above-mentioned two kinds of situation settings recomputate point in the piece field of direction (i, j) direction, promptly

${\mathrm{\θ}}^{\′}(i,j)=\frac{1}{2}\arctan \left(\frac{\underset{(i,j)\∈\mathrm{\Ω}\left(s\right)}{\mathrm{\Σ}}\sin \left(2\mathrm{\θ}(i,j)\right)}{\underset{(i,j)\∈\mathrm{\Ω}\left(s\right)}{\mathrm{\Σ}}\cos \left(2\mathrm{\θ}(i,j)\right)}\right),$

When among the field of direction O have a few traversal and again after the calculated direction, obtain the field of direction O ' after level and smooth;

In the 4th step, rim detection: in the foreground area F of fingerprint, the field of direction O ' for after level and smooth utilizes the Max-Min operator to carry out rim detection in 3 * 3 neighborhood windows, if the interior gradient of carrying out the point of rim detection of field of direction O ' is higher than threshold value T _o, then this point is labeled as edge pixel;

The 5th step, edge pixel deletion: for the edge pixel point, use its 4 neighborhood compute gradient information, if 4 Grad that calculate are not all at given threshold range [T _l, T _hIn, then from the edge, delete this pixel;

The 6th step, the core point location: utilize the outermost pixel of 3 * 3 neighborhood windows, the marginal point of calculating residue respectively (i, Con j), dx, the dy value, wherein Con is the direction consistency metric, promptly

$\mathrm{Con}(i,j)=\left(\sqrt{{\left({\mathrm{\Σ}}_{(u,v)\∈\mathrm{\Ω}}\cos \left(2\mathrm{\θ}(u,v)\right)\right)}^2+{\left({\mathrm{\Σ}}_{(u,v)\∈\mathrm{\Ω}}\sin \left(2\mathrm{\θ}(u,v)\right)\right)}^2}\right)/8,$

$\mathrm{dx}(i,j)={\mathrm{\Σ}}_{v=-1}^1\cos \left(2\mathrm{\θ}(i-1,v)\right)-{\mathrm{\Σ}}_{v=-1}^1\cos \left(2\mathrm{\θ}(i+1,v)\right),$

$\mathrm{dy}(i,j)={\mathrm{\Σ}}_{u=-1}^1\sin \left(2\mathrm{\θ}(u,i-1)\right)-{\mathrm{\Σ}}_{u=-1}^1\sin \left(2\mathrm{\θ}(u,i+1)\right),(u,v)\∈\mathrm{\Ω},$

The condition that provides the core point location is: as dx＞α, during dy＜β, the point of choice direction consistance minimum is down core point, as dx＜β, the point of choice direction consistance minimum is core point during dy＞α, wherein the value of α, β is to utilize the FVC2002 fingerprint database to carry out the core point positioning experiment, finally chooses concrete numerical value according to experimental result.

The present invention compared with prior art, its remarkable advantage: (1) a kind ofly provides the basis based on the fingerprint recognition system of embedded system fast for developing.In the embedded fingerprint recognition system, the core point that utilize to detect is as setting up the minutiae point template with reference to point, thereby eliminates the translation influence between fingerprint, makes that fingerprint image can rapid registering, carries out the identification of minutiae point coupling then.(2) owing to adopted level and smooth field of direction information, significantly improved the noise resisting ability of fingerprint recognition.(3) core point when location, used marginal information, marginal point has been carried out effective deletion, the very big time that has shortened the core point location.(4) rotation has robustness preferably to the method for Ti Chuing for fingerprint, all can detect core point for all types of fingerprints simultaneously, has significantly improved the recognition performance of fingerprint.

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

Description of drawings

Fig. 1 is the process flow diagram of method for rapidly positioning robust of finger print core point of the present invention.

Fig. 2 is the original fingerprint image.

Fig. 3 is fingerprint segmentation result figure.

Fig. 4 is the direction field pattern that calculates.

Fig. 5 is the direction field pattern after level and smooth.

Fig. 6 is an edge detection graph.

Fig. 7 picture element neighbours territory synoptic diagram.

Fig. 8 is the figure after the edge pixel deletion.

Fig. 9 is core point positioning result figure.

Figure 10 is core point positioning result figure before and after the fingerprint rotation.

Figure 11 is arch finger print core point positioning result figure.

Embodiment

In conjunction with Fig. 1, method for rapidly positioning robust of finger print core point of the present invention comprises the following step:

The first step, fingerprint image is cut apart: image is divided into the pixel block of nonoverlapping 12 * 12 sizes, calculates the average mean and the variance var of each piece, as var＜T _VarAnd mean＜M ₁The time or (mean-M ₁) * 8+var＜0 o'clock, it is background that this image block is set, otherwise is set to display foreground zone F, i.e. fingerprint effective coverage, wherein T _VarBe given threshold value, T rule of thumb is set _Var=80, M ₁Be the average of view picture fingerprint image, again with 12 pixels that contract in the border, effective coverage, to avoid the influence of background area, Fig. 2 is the fingerprint image of input after the image segmentation, and Fig. 3 has provided the result of image segmentation, and wherein white portion is the effective coverage of fingerprint.

In second step, the calculated direction field: fingerprint image has very strong directivity, during the calculated direction field, usually with image block, calculates the direction of the direction of each piece as fingerprint ridge line then.Here image is divided into the pixel block that nonoverlapping size is w * w, can be 5,7,9,11 etc., work as w=5 as w, utilize the Sobel operator calculate each piece (i, j) in each pixel (u, gradient G v) _xAnd G _y, G _xAnd G _yBe respectively the gradient on x and the y direction, calculate then each small images direction θ (i, j), promptly

$\mathrm{\θ}(i,j)=\frac{1}{2}{\tan }^{-1}\left[\frac{{\mathrm{\Σ}}_{u=i-w/2}^{i+w/2}{\mathrm{\Σ}}_{v=j-w/2}^{j+w/2}2\mathrm{Gx}(u,v)\mathrm{Gy}(u,v)}{{\mathrm{\Σ}}_{u=i-w/2}^{i+w/2}{\mathrm{\Σ}}_{v=j-w/2}^{j+w/2}(G{x}^2(u,v)-G{y}^2(u,v))}\right]---\left(1\right)$

After the direction of each pixel block of dividing in the image has all been calculated, just obtained field of direction O, the field of direction standard that calculates turns to behind the gray level image as shown in Figure 4.

In the 3rd step, the field of direction is level and smooth: the field of direction that calculates is to noise-sensitive, needs in addition smoothly, adopts the low-pass filtering method can cause the core point location inaccurate, adopts adaptive multiwindow method here, definition direction consistance, promptly

$\left(s\right)=\frac{\sqrt{{\left({\mathrm{\Σ}}_{(i,j)\∈\mathrm{\Ω}\left(s\right)}\cos \left(2\mathrm{\θ}(i,j)\right)\right)}^2+{\left({\mathrm{\Σ}}_{(i,j)\∈\mathrm{\Ω}\left(s\right)}\sin \left(2\mathrm{\θ}(i,j)\right)\right)}^2}}{M}---\left(2\right)$

Wherein, Ω (s) is piece field of direction mid point (i, neighborhood j), window size is s * s, and M is the number of point in the neighborhood, and s is since 3, calculate Con (s) value of different size windows, be divided into two kinds of situations according to Con (s) value and handle: (1) if Con (s) value of calculating less than given threshold value T _ComOr less than Con (s-2), then s increases by 2; If s reaches s _Max=9, then make s=3.

(2) if Con (s) is greater than given threshold value T in computation process _ComOr, then make s get currency greater than Con (s-2).

(i, j) direction rule of thumb are provided with T to recomputate piece field of direction mid point according to the s value of above-mentioned setting by formula (3) _Com=0.5, when among the field of direction O have a few traversal and again after the calculated direction, obtain the field of direction O ' after level and smooth.Fig. 5 be after the level and smooth field of direction with the gray-scale map result displayed, compare with result shown in Figure 4, as can be seen from Figure 5, the fingerprint noise has obtained effective inhibition, each zone becomes smoother, and it is affected by noise less to make follow-up core point locate, and positioning result is more stable.

${\mathrm{\θ}}^{\′}(i,j)=\frac{1}{2}\arctan \left(\frac{\underset{(i,j)\∈\mathrm{\Ω}\left(s\right)}{\mathrm{\Σ}}\sin \left(2\mathrm{\θ}(i,j)\right)}{\underset{(i,j)\∈\mathrm{\Ω}\left(s\right)}{\mathrm{\Σ}}\cos \left(2\mathrm{\θ}(i,j)\right)}\right)---\left(3\right)$

In the 4th step, rim detection: in the foreground area F of fingerprint, for level and smooth field of direction O ', utilizing the Max-Min operator is that computing unit carries out rim detection in 3 * 3 neighborhood windows with the piece direction, if gradient is higher than threshold value T _o(T is set rule of thumb _o=2.15), then this point is labeled as edge pixel.Standard turns to gray-scale map as shown in Figure 6 after the rim detection, has the edge of two black among the figure, and an edge originates in the image middle and upper part, and another edge is positioned at the image middle and lower part.

The 5th step, edge pixel deletion: the edge pixel point for detecting, use its 4 neighborhood compute gradient information, 4 neighborhood relationships as shown in Figure 7, if 4 Grad that calculate are not all at given threshold range [T _l, T _h] in, T wherein _l=0.55, T _h=1.75, then from the edge, delete this pixel.See shown in Figure 8ly, edge pixel deletion back shows to have only the edge pixel point of minority black to obtain keeping as can be seen after the edge pixel deletion with white, thereby has significantly reduced the search volume of core point location, has greatly shortened core point positioning time then.

The 6th step, the core point location: for the edge pixel point that retains,, utilize the outermost pixel of 3 * 3 neighborhood windows according to following formula (4-6), the marginal point of calculating residue respectively (i, Con dx j), the dy value, wherein Con is the direction consistency metric, promptly

$\mathrm{Con}(i,j)=\left(\sqrt{{\left({\mathrm{\Σ}}_{(u,v)\∈\mathrm{\Ω}}\cos \left(2\mathrm{\θ}(u,v)\right)\right)}^2+{\left({\mathrm{\Σ}}_{(u,v)\∈\mathrm{\Ω}}\sin \left(2\mathrm{\θ}(u,v)\right)\right)}^2}\right)/8---\left(4\right)$

$\mathrm{dx}(i,j)={\mathrm{\Σ}}_{v=-1}^1\cos \left(2\mathrm{\θ}(i-1,v)\right)-{\mathrm{\Σ}}_{v=-1}^1\cos \left(2\mathrm{\θ}(i+1,v)\right)---\left(5\right)$

$\mathrm{dy}(i,j)={\mathrm{\Σ}}_{u=-1}^1\sin \left(2\mathrm{\θ}(u,i-1)\right)-{\mathrm{\Σ}}_{u=-1}^1\sin \left(2\mathrm{\θ}(u,i+1)\right),(u,v)\∈\mathrm{\Ω}---\left(6\right)$

The condition that provides core point location is: as dx＞α, during dy＜β, the point of choice direction consistance minimum is the Lower core point, and as dx＜β, the point of choice direction consistance minimum is the Upper core point during dy＞α.The value of α, β is to utilize the FVC2002 fingerprint database to carry out the core point positioning experiment, finally chooses concrete numerical value according to experimental result, as α=0.1, and β=-0.1, promptly the core point location condition is:

When dx＞0.1, dy＜-0.1 o'clock, the point of choice direction consistance minimum is the Lower core point;

When dx＜-0.1, dy＞0.1 o'clock, the point of choice direction consistance minimum is the Upper core point.

Fig. 9 has provided the result that a width of cloth finger print core point detects.

Figure 10 has provided fingerprint rotation front and back core point positioning result contrast figure.Figure 10 .a is a width of cloth fingerprint image, the result images that Figure 10 .f turn 90 degrees for Figure 10 .a dextrorotation, and the field of direction that Figure 10 .b and Figure 10 .g are respectively after the respective smoothed is converted into the gray-scale map result displayed, and the field of direction differs greatly as can be seen from these two figure; Figure 10 .c and Figure 10 .h are corresponding edge detection graph, marginal position among two figure has nothing in common with each other, Figure 10 .d and Figure 10 .i are the result after the edge pixel deletion, and Figure 10 .e is the core point positioning result figure of Figure 10 .a, and Figure 10 .j is the core point positioning result figure of Figure 10 .f.As can be seen from Figure 10, though image rotates, the edge that obtains is also different, and the core point position of final location is basically identical but.Utilize the FVC2002 fingerprint database to experimentize, the result shows that rotation has robustness preferably for fingerprint in the core point location.

For the arch fingerprint, as Figure 11 .a, through after the rim detection, result such as Figure 11 .b.The edge pixel delete procedure can be with all edge pixel deletions, and as Figure 11 .c, the edge pixel before utilization deletion this moment carries out the core point location according to the core point location condition just can detect correct core point position, sees Figure 11 .d.

Claims (2)

1. method for rapidly positioning robust of finger print core point is characterized in that step is as follows:

The first step, input picture is cut apart: fingerprint image is divided into the piece of nonoverlapping 12 * 12 pixel sizes, the utilization gray variance and all value information be foreground area F and background area B with image division, and with 12 pixels that contract in the foreground area F;

Second the step, the calculated direction field: fingerprint image is divided into the pixel block that nonoverlapping size is w * w, utilize the Sobel operator calculate each piece (i, j) in each pixel (u, gradient G v) _xAnd G _y, calculate then each small images direction θ (i, j), promptly

$\mathrm{\θ}(i,j)=\frac{1}{2}{\tan }^{-1}\left[\frac{{\mathrm{\Σ}}_{u=i-w/2}^{i+w/2}{\mathrm{\Σ}}_{v=j-w/2}^{j+w/2}2\mathrm{Gx}(u,v)\mathrm{Gy}(u,v)}{{\mathrm{\Σ}}_{u=i-w/2}^{i+w/2}{\mathrm{\Σ}}_{v=j-w/2}^{j+w/2}({\mathrm{Gx}}^2(u,v)-{\mathrm{Gy}}^2(u,v))}\right],$

After the direction of each pixel block of dividing in the image has all been calculated, just obtained field of direction O;

In the 3rd step, the field of direction is level and smooth: definition direction consistance, promptly

$\mathrm{Con}\left(s\right)=\frac{\sqrt{{\left({\mathrm{\Σ}}_{(i,j)\∈\mathrm{\Ω}\left(s\right)}\cos \left(2\mathrm{\θ}(i,j)\right)\right)}^2+{\left({\mathrm{\Σ}}_{(i,j)\∈\mathrm{\Ω}\left(s\right)}\sin \left(2\mathrm{\θ}(i,j)\right)\right)}^2}}{M},$

Wherein, Ω (s) is point (i, neighborhood j) in the piece field of direction, M is the number of point in the neighborhood, and window size is s * s, and s is since 3, calculate Con (s) value of different size windows, be divided into two kinds of situations according to Con (s) value and handle: (1) if Con (s) value of calculating less than given threshold value T _ConOr less than Con (s-2), then s increases by 2; If s reaches s _MaxValue then makes s=3;

(2) if Con (s) is greater than given threshold value T in computation process _ConOr, then make s get currency greater than Con (s-2);

According to the s value of above-mentioned two kinds of situation settings recomputate point in the piece field of direction (i, j) direction, promptly

${\mathrm{\θ}}^{\′}(i,j)=\frac{1}{2}\arctan \left(\frac{\underset{(i,j)\∈\mathrm{\Ω}\left(s\right)}{\mathrm{\Σ}}\sin \left(2\mathrm{\θ}(i,j)\right)}{\underset{(i,j)\∈\mathrm{\Ω}\left(s\right)}{\mathrm{\Σ}}\cos \left(2\mathrm{\θ}(i,j)\right)}\right),$

When among the field of direction O have a few traversal and again after the calculated direction, obtain the field of direction O ' after level and smooth;

In the 4th step, rim detection: in the foreground area F of fingerprint, the field of direction O ' for after level and smooth utilizes the Max-Min operator to carry out rim detection in 3 * 3 neighborhood windows, if the interior gradient of carrying out the point of rim detection of field of direction O ' is higher than threshold value T _o, then this point is labeled as edge pixel;

The 5th step, edge pixel deletion: for the edge pixel point, use its 4 neighborhood compute gradient information, if 4 Grad that calculate are not all at given threshold range [T _l, T _h] in, then from the edge, delete this pixel;

The 6th step, the core point location: utilize the outermost pixel of 3 * 3 neighborhood windows, the marginal point of calculating residue respectively (i, Con dx j), the dy value, wherein Con is the direction consistency metric, promptly

$\mathrm{Con}(i,j)=\left(\sqrt{{\left({\mathrm{\Σ}}_{(u,v)\∈\mathrm{\Ω}}\cos \left(2\mathrm{\θ}(u,v)\right)\right)}^2+{\left({\mathrm{\Σ}}_{(u,v)\∈\mathrm{\Ω}}\sin \left(2\mathrm{\θ}(u,v)\right)\right)}^2}\right)/8,$

$\mathrm{dx}(i,j)={\mathrm{\Σ}}_{v=-1}^1\cos \left(2\mathrm{\θ}(i-1,v)\right)-{\mathrm{\Σ}}_{v=-1}^1\cos \left(2\mathrm{\θ}(i+1,v)\right),$

$\mathrm{dy}(i,j)={\mathrm{\Σ}}_{u=-1}^1\sin \left(2\mathrm{\θ}(u,j-1)\right)-{\mathrm{\Σ}}_{u=-1}^1\sin \left(2\mathrm{\θ}(u,j+1)\right),(u,v)\∈\mathrm{\Ω},$

The condition that provides the core point location is: as dx＞α, during dy＜β, the point of choice direction consistance minimum is down core point, as dx＜β, the point of choice direction consistance minimum is core point during dy＞α, wherein the value of α, β is to utilize the FVC2002 fingerprint database to carry out the core point positioning experiment, finally chooses concrete numerical value according to experimental result.

2. method for rapidly positioning robust of finger print core point according to claim 1, it is characterized in that: for part arch fingerprint image, the edge pixel delete procedure can be with all edge pixel deletions, and the edge pixel before utilization deletion this moment carries out the core point location according to the core point location condition.

Images