WO2007107050A1

WO2007107050A1 - Fingerprint identification method and system

Info

Publication number: WO2007107050A1
Application number: PCT/CN2006/000677
Authority: WO
Inventors: Quanhong Che; Shukai Chen; Zhinong Li
Original assignee: Zksoftware Beijing Inc.
Priority date: 2006-03-23
Filing date: 2006-04-14
Publication date: 2007-09-27
Also published as: CN1818927A; CN100412883C

Abstract

The present invention relates to a fingerprint identification method. In order to overcoming the low identification rate and slow identification speed of known identifying method, it comprises the steps of fingerprint characteristic extracting and characteristic matching. The fingerprint characteristic extracting step is: acquiring the fingerprint image, preprocessing and normalizing the fingerprint image; calculating blocking directional diagram and extracting singular point, calculating directional diagram, partitioning background area and thinning the singular point; filtering and enhancement of the image; calculating the density of ridge line, binarization and thinning of the image, extracting and verifying the detailed point, then deleting the counterfeit detailed point; compressing the fingerprint detailed point, singular point, average density of the ridge line and the characteristic of blocking directional diagram into fingerprint characteristic template and saving. The fingerprint characteristic matching step is: acquiring the fingerprint image, extracting the fingerprint detailed point, singular point, average density of the ridge line and the characteristic of blocking directional diagram of the fingerprint image; comparing the fingerprint detailed point, singular point, average density of the ridge line and the characteristic of blocking directional diagram of the fingerprint characteristic template and the fingerprint image, determining whether the fingerprints are from same finger by the similarity of their characteristic. The present invention has high identification rate, fast identification speed, high reliability and good operability.

Description

Fingerprint identification method and system

The invention relates to a biometric identification method, in particular to a fingerprint identification method and system. BACKGROUND OF THE INVENTION At present, fingerprint recognition methods for personal identification are different. However, existing fingerprint recognition methods generally have problems of low recognition rate and slow recognition speed. Summary of the invention

SUMMARY OF THE INVENTION An object of the present invention is to provide a fingerprint recognition method with high recognition rate and fast recognition speed, and to provide an identification system for implementing the method.

In order to achieve the above object, the fingerprint identification method of the present invention is special in that it consists of two steps: fingerprint feature extraction and feature matching:

The feature extraction steps are: collecting fingerprint images, pre-processing and normalizing the fingerprint images; calculating the block direction, extracting the singular points, calculating the direction map, segmenting the background regions, and refining the singular points; filtering and enhancing the image; calculating the ridge Line density; binarized image and refined, extract minutiae points, detail point verification, delete pseudo-detail points; fingerprint minutiae, singular points, average ridge density and block directional features are finally compressed into fingerprint feature template storage;

The feature matching step is: collecting the fingerprint image of the scene, and extracting the fingerprint detail point, the singular point, the average ridge density and the block pattern feature of the fingerprint image according to the above steps; comparing the fingerprint feature template with the fingerprint detail point of the fingerprint image of the scene, the singular point, The average ridge density and block pattern characteristics are judged by the similarity of the two features to determine whether they are the same finger. This fingerprint identification method has the advantages of high recognition rate and fast recognition speed.

As an optimization, the feature matching steps are:

Calculate the angle of the detail point pair connection, the angle between the detail point pair connection and the detail point direction, and the angle of the detail point pair connection in the database template and the on-site fingerprint template respectively; Specify the upper limit and lower limit of the connection point to the connection distance Value, delete the detail point pair connection point is greater than the upper limit value and the detail point pair data less than the lower limit value, to obtain a smaller range of detail point pair data U;

Calculate the rotation angle using a histogram;

The angle parameters of the fingerprint template from the database, including the point of detail point, the singular point angle, the block side The direction of the connection and the matching detail point pair U, etc., are rotated according to the angle calculated in the previous step, so that it has the same direction as the fingerprint template collected in the field:

The matching point pair corresponding to the detail point angle difference greater than a specified value is deleted from U, so that the matching detail point pair in U only contains the most reliable matching detail point pair;

In the same way, the histogram of the row and column directions is calculated, and the statistical histogram of the row and column coordinate differences of the corresponding minutiae points of all matching minutiae pairs is calculated, and the maximum point in the two arrays is found, that is, two fingerprint templates are in progress. The amount of translation after the rotation angle is aligned; „ The position parameters of the fingerprint template from the database, including the coordinates of the detail point, the coordinates of the singular point, the position of the block direction, etc., are translated, and the two fingerprint templates are completely aligned;

The matching pairs containing the detail point pairs whose row and column coordinates are larger than a specified value are deleted from [/, and the similarities of the matching pairs are added to obtain the final similarity of the two fingerprint template minutiae points;

Calculate the similarity of the global features - the singularity similarity is the position, direction and type of the singular points of the pairwise comparison, and the obtained similarities are added; the average ridge density similarity is the difference between the ridge density of the two fingerprint templates and the reciprocal ;

The similarity of the block pattern is the difference between the calculated directions in the common part of the effective area of the two fingerprint templates, and the average is taken and the reciprocal is taken;

The similarity of the last two fingerprint templates is formed by the above partial and full-circumference feature similarity;

When performing one-to-many recognition, first sort the average ridge density of the fingerprint template in the database, and when identifying the on-site fingerprint, first match the fingerprint template closest to the average ridge density in the database to speed up the recognition; Ridge density is the average ridge density of the entire fingerprint image.

As an optimization, when the fingerprint feature is extracted: the fingerprint image is represented as a two-dimensional matrix, and each pixel is an element of the matrix, and the value is 0 to 255, and the dimension of the matrix is the width and height of the image;

The detail point of the fingerprint refers to the endpoint or bifurcation point on the fingerprint ridgeline. The fingerprint detail point includes the following features: The coordinate X indicates the position in the fingerprint image; the type indicates whether the ridge line is the end point or the bifurcation point; the direction indicates the detail The direction of the point, if it is the end point type detail point, the direction points from the detail point position to the ridge line. If it is a bifurcated detail point, the direction points from the detail point position to the middle of the bifurcated ridge line; The density g - represents the average density of the ridge lines near the minutiae point; the ridge curvature _C - represents the degree of change of the ridge line direction here;

Blocking pattern: divides the fingerprint image into disjoint small blocks of size BLOCK_SIZEXBLOCK_SIZE. For each small image, the average direction of the ridge line is calculated, so that the size is

(HEIGHT/BLOCK_SIZE) X (WIDTH/BLOCK— SIZE) Blocking pattern; the block pattern depicts the global ridgeline of the fingerprint image; in addition, an illegal direction value is used to represent the background area of the fingerprint image segmentation on the block pattern; - Singularity: Fingerprint image There are some places where the ridges are not continuous. These places are called singular points of fingerprints. They are characterized by: coordinates xy, indicating the position in the fingerprint image; type t, singular points are divided into core points, dual core points, and triangles. Point three; direction d, indicating that the direction of the fingerprint ridge line changes the least when moving away from the singular point in this direction. The ridge density c represents the average separation distance of the ridge lines near the singular point.

As an optimization, image pre-processing and normalization are to first uniformly filter the image to make the image smoother, and then format the image. Calculate the block pattern to extract the singular point on the block pattern, first calculate each point. Poincare Index: pindex"

k, if μ |< -,

2

π + k, if \ k \< ~

2

π - k, otherwise where M is the number of surrounding pixels, Ο, · indicates the direction of the first point; first take the radius of 1, ie the surrounding 8 points to calculate the Poincare Index, get pl, if its Poincare Index is not Zero, then calculate the Poincare Index with a radius of 2, that is, the outer layer of the periphery, and p2 _: pi is the same as p2, indicating that the point is a singular point. If pi is 1, it is a core point type singular point, if pi is one 1 is a triangular point. If pi is 2, it is a dual-core singular point. If p2 is different from pi, but p2>0, pl>0, it is a dual-core singular point; otherwise, it is not a singular point.

As an optimization, calculate the direction map, segment the background area and refine the singular points: the normalized image, calculate the ridge direction of the first point, and simultaneously calculate the consistency of the ridge line direction, obtain the direction map, and re-determine the singularity Point position, starting from the original position of these singular points, find the exact position of the singular point, and calculate the new singular point direction in the new position.

As an optimization, the filtering and enhancement of the image is: after the anisotropic filter is processed, an enhanced fingerprint image is obtained; the calculated ridge density is: first, the fingerprint ridge density map is calculated, and then the ridge density map is 33 X 33 Mean filtering.

As an optimization, the binarized image is refined and refined by: binarizing the enhanced image with the 33×33 mean filtered image as an adaptive threshold; then refining the binarized image into a single-point width ridge Line graph; Image refinement is that each black pixel in the image has 8 adjacent points, according to which it is judged whether the current point should be changed to white. This way After repeated scans, until no black dots are changed to white, a refined fingerprint ridge plot is obtained. As an optimization, the point of extraction is: first eliminate the glitch and noise, that is, by scanning the refined ridge diagram, tracking the ridge line, if the pixel distance from the start point to the end point of the ridge line is less than a set threshold, take it from Wipe off the refinement map; then, extract the minutiae point: that is, any black point on the image, if it is one of the adjacent 8 points, select one starting point, scan clockwise one week to return to the starting point If the change of color is 2 times, it means that the point is a final type of detail point; if it is more than 4 times, the point is a forked type detail point, otherwise it can be ignored, by scanning the effective fingerprint image area , got all the details points;

Track the ridge line at the detail point to get the direction of the ridge line; the ridge line curvature of the detail point is represented by the change of direction. On the pattern of the fingerprint image, the direction difference between the direction near the point and the point is used. Calculate the curvature.

As an optimization, the detail point verification and deletion pseudo-detail point is: any detail point, if there is a detail point with a distance less than a set value D1, delete the detail point; such as an end point type detail point and an end point If the distance of the plow point is less than a set value D2, and they are opposite directions, the two minutiae points are deleted at the same time; if the distance between an end point type detail point and a bifurcated type detail point is less than a set value D3, and they are opposite directions , at the same time delete the two minutiae points; if a detail point is less than a set value D4 from the invalid area of the fingerprint image, and the direction is outward, the minutiae point is deleted; the final minutiae point is obtained by the above deletion.

An identification system for implementing the fingerprint identification method of the present invention, which is characterized in that it comprises a fingerprint collector, a fingerprint recognition system, an identification or control signal output mechanism, and includes a fingerprint image memory, a fingerprint image processor and fingerprint feature data. The fingerprint image processor processes and recognizes the fingerprint image using the method described in one of claims 1-9. It has the advantages of high recognition rate, fast recognition speed, high reliability and strong operability.

Among them: The characteristic representation of the fingerprint detail point (x, , t, ^ g, _C ) contains more information, which is beneficial to improve the recognition rate of the system; the characteristic representation of the fingerprint singular point (x, y, i, g) contains more Information, which is conducive to the recognition rate of the system; the average ridge density G as a global feature can be indexed to assist in recognition to speed up. The block pattern of the fingerprint is saved as a global feature in the fingerprint template. The block pattern is compared in the comparison process, and the similarity is merged into the final result. The singular point extraction method can quickly calculate the accurate singularity. Point position and feature; Anisotropic filter is used to enhance the fingerprint image, and the effect is very good; after the filter is modulated by the direction of each point on the fingerprint image, the convolution method is used to filter the point. Since the filter kernel of each point is modulated by the direction of the point, the filtering effect is much better than that of image block filtering; by storing the anisotropic filter coefficients in all directions, it is possible to use the look-up table during convolution. law. The speed of filtering is greatly improved; the process of fingerprint matching, the final similarity of fingerprint template matching is obtained by fusing the similarity of various features, which makes the result more reliable; the detail point alignment method, which estimates the preliminary match The details of the point-to-line pair transformation parameters are statistically generated for the valuation. Find the final transformation parameters in the histogram.

After adopting the above technical solution, the fingerprint identification method of the invention has the advantages of high recognition rate, fast recognition speed, high reliability and strong operability. DRAWINGS

1 is a schematic diagram of three singular points in the fingerprint identification method of the present invention;

2 is a flow chart of a fingerprint identification method of the present invention;

3 is a schematic diagram of 8 points around the pi in the fingerprint identification method of the present invention;

4 is a schematic diagram of 12 points around p2 in the fingerprint identification method of the present invention;

5 is a schematic diagram of an anisotropic filter core with a zero direction in the fingerprint identification method of the present invention;

6 is a schematic structural diagram of converting eight adjacent points into a table index number 22 in the fingerprint identification method of the present invention; FIG. 7 is a view showing three main refinement ridge noise patterns in the fingerprint identification method of the present invention;

8 is a color change diagram of adjacent 8 points of a final type detail point in the fingerprint identification method of the present invention;

9 is a color change diagram of adjacent 8 points of a bifurcated detail point in the fingerprint identification method of the present invention;

Figure 10 is a connection diagram between the pairs of minutiae points in the fingerprint identification method of the present invention;

11 is an original fingerprint image in the fingerprint identification method of the present invention;

12 is a normalized fingerprint image in the fingerprint identification method of the present invention;

13 is a direction diagram of a fingerprint in the fingerprint identification method of the present invention;

14 is an enhanced image of a fingerprint in the fingerprint recognition method of the present invention;

15 is a binarized image of a fingerprint in the fingerprint identification method of the present invention;

Figure 16 is a detailed ridge diagram of a fingerprint in the fingerprint recognition method of the present invention. detailed description

The following further describes with reference to the accompanying drawings and specific examples - the fingerprint identification algorithm involves two main steps: feature extraction and feature matching.

Feature extraction: Image processing of fingerprints and extraction of fingerprint global and local features, and saved as fingerprint templates; Feature matching: Compare two fingerprint feature templates to obtain a matching score, and then determine whether the two fingerprints are the same according to this score.

First, feature extraction

1, concepts and conventions 1) Representation of fingerprint image

The fingerprint image is represented as a two-dimensional matrix. Each pixel is an element of a matrix, and the value is (0~255). The dimension of the matrix is the width WIDTH and 髙 HEIGHT of the image. The gray value of the i-line j column on the fingerprint image is expressed as k.

2) Representation of local features

The local feature of the fingerprint refers to the endpoint or bifurcation point on the fingerprint ridgeline, which is called the detail point of the fingerprint. The fingerprint detail points include the following features (x, _y , t, d, g, c:

Coordinate; : indicates the position in the fingerprint image;

• Type t : Indicates whether it is the end point or the bifurcation point of the ridge line;

Direction : Indicates the direction of the detail point. If it is the end point type detail point, the direction points from the detail point position to the ridge line; if it is the bifurcation type detail point, the direction points from the detail point position to the middle of the bifurcated two ridge lines. „ Ridge density g : indicates the average density of the ridges near the minutiae. The greater the separation distance of the ridges, the lower the density;

Ridge curvature C : Indicates how much the ridge line changes here.

3) Representation of global features '

Blocking pattern

Divide the fingerprint image into disjoint small blocks of size BLOCK_SIZE X BLOCK-SIZE. For each small image, calculate the average direction of the ridge line, and get the size

(HEIGHT/BLOCK—SIZE) Blocking pattern of X (WIDTH/BLOCK—SIZE). The block pattern plots the global ridgeline of the fingerprint image and stores it as a global feature of the fingerprint image for later comparison. In addition, an illegal direction value is used on the block pattern to indicate the background area after the fingerprint image is segmented (there is no fingerprint image here, or the fingerprint image quality is too poor).

Singularity

The ridge direction of the fingerprint has the characteristic of continuity, that is, the ridge direction of the adjacent position is generally uniform or does not change much. However, there are also some places on the fingerprint image where the ridge directions are not continuous. These places are called singular points of the fingerprint <=

The singular points are characterized by (x, y, t, d, c) _: X y coordinates: indicates the position in the fingerprint image. Type t: As shown in Figure 1, the singular points are divided into core point 1-1, dual core point 1-2, and triangle point 1-3. Direction d: Indicates that the direction of the fingerprint ridge line changes the least when moving away from the singular point in this direction. Ridge density c : indicates the ridge line near the singular point Average separation distance. The average density is average ridge ridge entire fingerprint image density _t

2, the algorithm flow

2.1 Flow chart, please see Figure 2.

2.2 Image Preprocessing and Normalization

First, the image is subjected to 3X3 mean filtering to make the image smoother.

Among them, is the original image, 3⁄4 _x is the smoothed image, here take w=l.

Then, normalize the image:

Δ

^Mn i,j ^{= I} i - ^Var i

Max _i . - /. j + Var _iy

u = Max _J. One Min _u

∑ ∑

Var.. = ° '

'J (w+iy

Where ^ is the image of the original image after 5X5 mean filtering, and in the calculation of r, take a larger neighborhood w=80.

2.3 Calculate the block direction map to extract singular points.

The calculation of the block pattern is the same as the calculation of the complete calculation pattern, except that only the direction of the center position of the block is calculated. The calculation of the calculation pattern is specifically described below.

On the block pattern, calculate the Poincare Index for each point:

Pindex" =丄τ(Ο _{Μ)η0άη - O _j )

^ /=0

k, if | t|< -,

+ if I A: |<—

2

π- k, otherwise where "is the number of surrounding pixel points, indicates the direction of _Ζ · points. In order to ensure the reliability of the calculation, first Take a radius of 1, that is, 8 points around to calculate the Poincare Index, and get pi. If the Poincare Index is non-zero, calculate the Poincare Index with a radius of 2, that is, the outer layer of the periphery, and obtain p2. See Fig. 3 for a schematic diagram of 8 points around the pi, and Fig. 4 for a schematic view of 12 points around p2.

There are the following situations:

Pi is the same as p2, indicating that the point is a singular point. If pi is 1, it is a core point type singular point. If pi is a 1, it is a triangle point. If pi is 2, it is a dual-core type singular point; if p2 and pi Different, but p2>0, pl>0, is a dual-core singular point; in other cases it is not a singular point.

2.4 Calculate the direction map, segment the background area and refine the singular points

For the normalized image, calculate the ridge direction of each point by the following formula O _v 3⁄4 = ( ) ² , (3⁄4;= ( ) ² , 3⁄4 =

And calculate the consistency of the ridge direction at the same time j

i _{imij≥Threshold}

73⁄4^/wW is a set threshold, Q; =0 means that this is the background area of the fingerprint image.

The position of the singular point of the fingerprint obtained from the above block pattern is inaccurate, and the singular point position can be re-determined using the currently obtained pattern. Starting from the original position of these singular points, the minimum point can be found in the vicinity, which is the exact position of the singular point. In the new position, the new singular point direction is calculated.

2.5 Image filtering and enhancement

Design an anisotropic filter: , ,

, y,

, , \α, if χ ² + y ² < r ²

ρ(χ, y) = <

[0, otherwise where r is the effective radius, usually taken as 6, the amplitude coefficient, usually 1024, and is the shape control parameter of the filter, usually taken as 8 and 1. Is the modulation direction of the filter. See Figure 5 for an anisotropic filter kernel with zero direction. Thus, an enhanced fingerprint image (convolution) can be calculated using the following formula:

In order to quickly calculate the above formula, the filter coefficients of all angles and the denominator term in the above formula are pre-calculated and saved, and the table is specifically calculated to directly convolve with the image.

2.6 Calculate the ridge density map

Calculate the density map of the fingerprint ridge by the following formula:

c Koef ~~

i+w j+w

i, ^if ∑ ∑ ^J y,x ^E - ^Threshold bottom > Threshold^ ]

y=i-iv x= J'- '

And [i, j] is not in the bad area

0, otherwise

1, [ , j] is not in the bad area

0, otherwise The ridge density of the fingerprint is very continuous, so the ridge density map Α·· is then subjected to a mean filtering of 33 X 33 to eliminate noise.

2.7 Binarized image and refined The enhanced image is binarized according to the following formula: = j0, if ,; < S _u

l [255, otherwise where 3⁄4;· is an image filtered by the average of 33 X 33 on the enhanced fingerprint image, and this image is used as an adaptive threshold to binarize the fingerprint image.

The binarized image is then refined into a single point width ridge plot. The algorithm is to consider the eight adjacent points of each black pixel in the image and use them to determine if the current point should be changed to white. This is followed by repeated iterations until a black dot is changed to white, resulting in a refined fingerprint ridge plot. '

In the binarized image, there are a total of 256 cases of 8 adjacent points of a black pixel, and the actual calculation can be quickly judged by looking up the table. Create a table of 256 elements as follows:

{0,0,0,0,0,0,0, 1 ,0,0,0,0,0,0, 1 , 1 ,0,0,0,0,0,0,0,0,0 ,0,0,0, 1,0,1,1 ,0,0,0,0,0,0,0,0,0,0,0,0,

0,0,0,0,0,0,0,0,0,0,0,0,1,0,0,0,1,0,1,1,0,0,0,1,0, 0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,1,0,

0,0,0,0,0,0,0,1,0,1,0,0,0,0,0,0,0,0,0,0,0,0,1,0,0, 0,0,0,0,0,1,0,0,0,1,0,1,0,0,0,0,1,0,0,0,1,

0,0,0,0,0,0,0,1,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,1,0, 0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,

0,0,0,0,0,0,0,0,0, 1,0,1,0,1 ,0,0,0,0,0,0,0,0,0,0,0, 0, 1 ,0,0,0,0,0,0,0, 1,0,0,0,0,0,0,0,1,1,0,1,

0,0,0,0,0,0,0,0,0,0,0 ₅ 0,1,0,0,0,0,0,0,0,1,0,0,0,1 , 0,1,0}

A value of 1 indicates whether the current pixel should be changed to white. The index of the table is constructed as follows: 8 adjacent points are converted to table index number 22 (binary 00010110), see Figure 6.

2.8 Extracting minutiae

Due to the noise of the image, the refined fingerprint ridge diagram has burrs and noise, so when extracting the minutiae, they must be eliminated first, otherwise many false minutiae will be extracted. See Figure 7 for the three main refinement ridge noise maps.

By scanning the refined ridge plot, the ridgeline is tracked, and if the pixel distance from the start to the end of the ridge is less than a set threshold, it can be erased from the refinement.

Then, it is very convenient to extract the minutiae point - any black point on the image, if it is adjacent to the 8 points, choose a starting point, scan clockwise one week back to the starting point, its color If the change is 2 times, it means that the point is a final detail point; if it is more than 4 times, the point is a forked detail point, otherwise it can be ignored. The color change of the adjacent 8 points of the final detail point is 2 times: 6->7, 7->0, see Figure 8; the color of the adjacent 8 points of the bifurcated detail point changes more than 4 times: 1->2, 2->3, 3->4, 4->5, 6->7, 7->0, see Figure 9.

Thus, by scanning the valid fingerprint image area, all the minutiae points are obtained. Track the ridgeline at the point of detail, The direction of the ridge line can be obtained.

The ridge curvature of the detail point can be represented by a change in direction. On the pattern of the fingerprint image, calculate the curvature using the difference between the direction near the point and the direction of the point -

Where _r is the radius constant, usually taken as 10.

2.9 Detail point verification

The details obtained so far, due to image noise, there are still many pseudo-details in it, which need to be further removed. Consider the following:

Consider any detail point, if there is a detail point with a distance less than a set value Dl, delete the detail point;

If an end point type detail point is less than a set value D2 from an end point type detail point, and they are opposite in direction, the two minutiae points are simultaneously deleted;

If an end point type detail point is less than a set value D3 from a fork type detail point, and they are opposite in direction, the two minutiae points are simultaneously deleted;

If a detail point is less than a set value D4 from the invalid area of the fingerprint image, and the direction is outward, the detail point is deleted.

This gives the final details.

Fingerprint minutiae points and other global features are eventually compressed into fingerprint feature template stores.

Second, the fingerprint matching method

The minutiae point matching method is based on the detail point connection.

Consider the connection of two minutiae points on a fingerprint image, defined as the length of the line segment, ie the distance between the two minutiae points;

And bj are the angles between the connection and the direction of the detail point;

Uy is the angle of the connection.

The connection between the point pairs of detail is shown in Figure 10.

Compare such line segments as the basic unit of fingerprint detail point matching to compare the point pairs on the two fingerprints. For a detail point pair, , _ai , and two minutiae types t, curvature c, and ridge density g are both translationally invariant and rotationally invariant. Thus, the similarity of the two minutiae pairs can be determined by comparing these quantities. For the point-in-point pair (m^mji) on the fingerprint template and the minutiae pair (mi m _j2 ) on the fingerprint template in the database, define the "similarity" of the point-to-point pair as follows: - )

Where cqi,co/ ₂ , c ₀ / ₃ , c ₀ / ₄ , co/ ₅ are positive constant coefficients. d, a, 6 are respectively defined as before, c, g, t are the curvature, ridge density and type code of each minutiae (the bifurcation minutiae is 1, and the terminal type minutiae is 0). When D is less than a given value ThresholdD, the two minutiae pairs are considered to match.

' If a fingerprint image has N minutiae points, then C(N, 2) minutiae pairs can be generated, and C(M, 2) minutires pairs that are to be generated with M minutiae points of another fingerprint map To compare one by one, you need to perform C(M,2)XC(N,2) comparisons. If M=N=80, the number of comparisons is 39942400, which is very slow, so it is necessary to compare Make some restrictions on the front. Specify two values r^^to A, ThresholdDz, and only consider the point-to-point pairs whose connection length is between these two values. This can greatly reduce the time spent on the alignment.

From the calculation formula of D, it is easy to see: If the calculation result of the first item is larger than J3⁄4r 1⁄2, it is not necessary to calculate the following items to know that the two detail points do not match, so the details of the two fingerprints can be first By sorting according to the length d. of the connection, it is possible to perform comparison in a small neighborhood of a connection length d. This greatly speeds up the ignoring of mismatched minutiae pairs and will result in a matching list of minutiae points:

^U = J _M , , U ^D W < ThresholdD]

D ...

S ~ γ J\ , hJ2 .

^Wlh - ThresholdD

Among them is a detail point pair (η^, π^) from the fingerprint template of the site and its connection, /, 7 is a detail point pair (m _i2 , mj ₂ ) and its connection from the fingerprint template of the database, D is this The "distance" between two pairs of points of detail, S is the similarity between two pairs of point pairs.

Use the following method to calculate the similarity of two fingerprint templates from this list:

1, using the histogram method to calculate the rotation angle

Set a one-dimensional array { | 0<i<360 }, the subscripts represent the angle from 0 to 359, each element is as follows

Among them, it is a unit impact function with a value of 1 for d and a value of 0 for others. It can be seen that it is actually a statistical histogram of the angular differences of all corresponding minutiae points that match the minutiae points. Find the maximum point in this array, which is the rotation angle θ of the two fingerprint templates you need.

2. Rotate the angle parameters of the fingerprint template from the database, including the point of detail point, the angle of the singular point, the direction of the block, and the direction of the matching point to the line in t/, etc., according to the angle calculated in the previous step, so that It has the same direction as the fingerprint template collected on site -

+ ^) mod 360 - d _t

3. Remove the matching point pair corresponding to the detail point angle difference greater than a specified value from U. Thus, the matching 'detail point pair' in U contains only the most reliable matching detail point pairs.

4, the same method, calculate the histogram of the direction of the row and column

Set two one-dimensional arrays { } and {H3⁄4}:

{HX _dx I-MaxD im≤ dx < MaxD im }

{HY _dy I -MaxD im < dy < MaxD im }

It can be seen that {Hi }, {H } is actually a statistical histogram of the row and column coordinate differences of all corresponding minutiae points matching the minutiae point pairs. Finding the maximum point in the two arrays is the amount of translation (xo, y ₀ ) of the two fingerprint templates needed after the rotation angle is aligned.

5. Translate the position parameters of the fingerprint template from the database, including the coordinates of the detail points, the coordinates of the singular points, the position of the block direction, etc.

Now, the two fingerprint templates are fully aligned.

6. Remove the matching pair containing the detail point pairs whose row and column coordinates are larger than a specified value from /7, so that the matching pairs of the detail point pairs in the pair are perfectly matched. Adding the similarities from these matching pairs is two fingerprint patterns The final similarity S of the plate detail point set, „.

7. In the above calculations, the global features have also been aligned by translation and rotation. At this point, the similarity of the global features can be easily calculated.

Singular point similarity & : The similarity of the position, direction and type of the singular points of the pairwise comparison; the average ridge density similarity: the difference between the ridge density of the two fingerprint templates and the reciprocal;

The similarity of the block pattern & : In the common part of the effective area of the two fingerprint templates, the difference of the direction is calculated, and the average is taken and the reciprocal is taken.

8. The similarity of the last two fingerprint templates is formed by the above partial and global feature similarity:

Where k _m , k _s , k _s , k _d are the weighting coefficients of the matching similarities of various features.

Note that the calculation of the average ridge density similarity is the difference between the two average ridge densities. Therefore, if one-to-many identification is performed, the fingerprint templates in the database can be sorted according to the average ridge density G, for the on-site fingerprint. When identifying, it is possible to match the fingerprint template with the highest average ridge density in the database. Therefore, since the fingerprint template in the database is indexed according to G, the recognition process can be greatly accelerated.

For the fingerprint image during processing, please refer to the accompanying drawings - wherein: Figure 1 is the original fingerprint image, Figure 12 is the normalized image, Figure 13 is the orientation image, Figure 14 is the enhanced image, Figure 15 is the second image The valued image, Figure 16 is a refined ridge diagram.

A fingerprint identification system for implementing the fingerprint identification method of the present invention includes a fingerprint collector, a fingerprint recognition system, an identification or control signal output mechanism, and includes a fingerprint image memory, a fingerprint image processor, and a fingerprint feature data memory; the fingerprint image processor is The fingerprint image is processed and identified using the method described in one of claims 1-9.

Claims

Claim

A fingerprint identification method, which is characterized by two steps of fingerprint feature extraction and feature matching: The feature extraction step is: collecting a fingerprint image, preprocessing and normalizing the fingerprint image; calculating a segmentation pattern extraction Singular points, calculate the direction map, segment the background area and refine the singular points; filter and enhance the image; calculate the ridge line density; binarize the image and refine it, extract the minutiae points, detail point verification, delete the pseudo-detail points; fingerprint details Points, singular points, average ridge density, and block pattern features are eventually compressed into fingerprint feature template storage;

The feature matching step is: collecting the on-site fingerprint image, and extracting the fingerprint detail point, 'singular point, average ridge density and block pattern feature of the on-the-spot fingerprint image according to the above steps; comparing the fingerprint feature template with the fingerprint detail point of the fingerprint image of the scene, the singular point The average ridge density and block pattern characteristics are judged by the similarity of the two features to determine whether they are the same finger.

2. The fingerprint identification method according to claim 1, wherein the feature matching step is:

Calculate the detail point pair connection distance, the detail point pair connection and the detail point, the angle of the direction and the angle of the detail point pair connection in the database template and the on-site fingerprint template respectively; specify the detail point pair connection distance upper limit value and the lower limit Limit value, delete the detail point to the data of the detail point pair whose connection distance is greater than the upper limit value and less than the lower limit value, and obtain a smaller range of details. Point to the data U; . Calculate the rotation angle by using the histogram;

The angle parameters of the fingerprint template from the database, including the point angle of the detail point, the angle of the singular point, the direction of the block, and the direction of the line connecting the point to the U, are rotated according to the angle calculated in the previous step, so that it is on the same scene. The collected fingerprint templates have a consistent orientation:

The matching point pair corresponding to the detail point angle difference greater than a specified value is deleted from the U, so that the matching detail point pair in U only contains the most reliable matching detail point pair; - the same method, calculating the histogram of the row and column direction, and calculating All statistical histograms of the row and column coordinate differences of the corresponding minutiae points of the matching detail point pairs, and finding the maximum point in the two arrays is the translation amount of the two fingerprint templates after the rotation angle alignment;

Each position parameter of the fingerprint template from the database, including minutiae coordinates, singular point coordinates, block direction position, etc., is translated, and the two fingerprint templates are completely aligned;

The matching pairs of the detail point pairs including the row and column coordinate deviations greater than a specified value are deleted from C7, and the similarity of the matching pairs is added to obtain the final similarity of the two fingerprint template detail point sets;

Calculate the similarity of global features: The singular point similarity is the position, direction and type of the singular points of the pairwise comparison, and the obtained similarities are added; the average ridge density similarity is the difference between the ridge density of the two fingerprint templates and the reciprocal;

The similarity of the block pattern is the common part of the effective area of the two fingerprint templates, the difference of the calculated direction, the average after the accumulation and the reciprocal; The similarity of the last two fingerprint templates is the local and global feature similarity of the above !3⁄4 combined;

3. The fingerprint identification method according to claim 2, wherein when the one-to-many identification is performed, the average ridge density of the fingerprint template in the database is first sorted, and when the on-site fingerprint is recognized, the average ridge in the database is first used. The fingerprint templates with the closest density are matched to speed up the recognition; the average ridge density is the average ridge density of the entire fingerprint image.

4. The fingerprint identification method according to claim 3, wherein when the fingerprint feature is extracted: the fingerprint image is represented as a two-dimensional matrix, and each pixel is an element of the matrix, and the value is 0 to 255, and the dimension of the matrix is The width and width of the image;

5. The fingerprint identification method according to claim 4, wherein the detail point of the fingerprint refers to an end point or a bifurcation point on the fingerprint ridge line, and the fingerprint detail point includes the following features: coordinate x - indicates a position in the fingerprint image ; type t . - indicates whether the end point or the bifurcation point of the ridge line; the direction indicates the direction of the detail point, if it is the detail point of the end point type, the direction points from the detail point position to the ridge line, and if it is a bifurcated detail point, The direction is from the detail point position to the middle of the bifurcated ridge lines; the ridge density represents the average density of the ridge lines near the minutiae point; the ridge curvature c - represents the degree of change of the ridge line direction here;

The fingerprint identification method according to claim 5, characterized in that the segmentation pattern is: dividing the fingerprint image into mutually disjoint small blocks of BLOCK-SIZEX BLOCK-SIZE size, and calculating ridge lines for each small image. The average direction, thus getting the size,

(HEIGHT/BLOCK—SIZE) Blocking pattern of X (WIDTH/BLOCK—SIZE); the block pattern depicts the global ridgeline of the fingerprint image; in addition, an illegal direction value is used to represent the pair on the block pattern The background area after the fingerprint image is divided;

7. The fingerprint identification method according to claim 6, characterized in that the singular point: the ridge line direction of the fingerprint image is discontinuous, and the singular points of the fingerprint are called singular points of the fingerprint, and the features are: coordinates xy, indicating The position in the fingerprint image; type t, the singular point is divided into three types: core point, double core point and triangle point; direction d, indicating along When the direction is far from the singular point, the direction of the fingerprint ridge is minimally changed. The ridge density C represents the average separation distance of the ridge lines near the singular point.

8. The fingerprint identification method according to claim 7, wherein the image pre-processing and normalization firstly perform uniform value filtering on the image to make the image smoother, and then format the image; and calculate the block pattern to extract the singularity. The point is on the block pattern, first calculate the Poincare Index for each point:

Pindex" =丄(0 ₍ , ₊₁ „ - O, ) k, if |< - ,

2

π + k. if I A; |<—

2

π - k, otherwise where M is the number of surrounding pixels, O, indicating the direction of the first point; first take the radius of 1, ie the surrounding 8 points to calculate the Poincare Index, get pl, if its Poincare Index is non-zero Then calculate the Poincare Index with a radius of 2, that is, the outer layer of the periphery, and p2: pi is the same as p2, indicating that the point is a singular point. If pi is 1, it is a core point type singular point, if pi is a 1 It is a triangular point. If pi is 2, it is a dual-core singular point. If p2 is different from pi, but p2>0, pl>0, it is a dual-core singular point; otherwise, it is not a singular point.

9. The fingerprint identification method according to claim 1, 2, 3, 4, 5, 6, 7, or 8, wherein the calculating the direction map, dividing the background area, and refining the singular points are: normalized image, calculation The ridge line direction of the first point, and at the same time calculate the consistency of the ridge line direction, obtain the direction map, re-determine the singular point position, and find the exact position of the singular point from the original position of these singular points, in the new position, Calculate the new singular point direction.

10. The fingerprint identification method according to claim 1, 2, 3, 4, 5, 6, 7, or 8, characterized in that the filtering and enhancement of the image are: after the anisotropic filter is processed, an enhanced fingerprint image is obtained. Calculating the ridge density is - first calculate the fingerprint ridge density map, then perform a 33 X 33 mean filter on the ridge density map.

11. A fingerprint identification method according to claim 1, 2, 3, 4, 5, 6, 7, or 8, characterized in that the binarized image is refined and refined: an image that is filtered by 33 X 33 means as an adaptive The threshold is used to binarize the enhanced image; then the binarized image is refined into a single-point width ridge plot; the image refinement is that each black pixel in the image has 8 adjacent points, according to which Determine if the current point should be changed to white. This way, after repeated scans, straight When no black spot is changed to white, a refined fingerprint ridge diagram is obtained.

12. The fingerprint identification method according to claim 1, 2, 3, 4, 5, 6, 7, or 8, wherein the extraction of the minutiae points is: first removing burrs and noise, that is, by scanning the refined ridge diagram, Track the ridge line, if the pixel distance from the start point to the end point of the ridge line is less than a set threshold, erase it from the refinement map; then, extract the minutiae point - that is, any black point on the image If one of the 8 adjacent points is selected, start with a clockwise side. Return to the starting point one week after scanning. If the color change is 2 times, it means that the point is a final detail point. If it is more than 4 times, the point is a forked detail point, otherwise it can be ignored. By scanning the effective fingerprint image area, all the minutiae points are obtained;

13. The fingerprint identification method according to claim 1, 2, 3, 4, 5, 6, 7, or 8, characterized in that the detail point verification and the deletion of the pseudo-detail point are: any one of the minutiae points, if there is a detail point If the distance is less than a set value D1, the detail is deleted; if an end point type detail point is less than a set value D2 from an end point type detail point, and they are in opposite directions, the two minutiae points are simultaneously deleted. If an endpoint type detail point is less than a set value D3 from a fork-type detail point and they are in opposite directions, the two minutiae points are deleted at the same time; if a detail point is smaller than the invalid area of the fingerprint image The value D4 is set, and the direction is outward, the detail point is deleted; the final detail point is obtained by the above deletion. .

14. A fingerprint identification system, comprising: a fingerprint collector, a fingerprint recognition system, an identification or control signal output mechanism; wherein the fingerprint image memory, the fingerprint image processor and the fingerprint feature data memory; the fingerprint image processor is utilized The method described in one of claims 1-9 processes and identifies the fingerprint image.