CN1818927A - Fingerprint identifying method and system - Google Patents

Abstract

A fingerprint recognition methods and system. It comprises of two steps that are the character distill and the character matching of the fingerprint in order to solve problems of the existing method that are low identify rate and slow identify speed. The character distill step is: collecting fingerprint image, preprocess and standardization the fingerprint image; distilling the bizarre-point by calculating the dis-block directive-image, calculating the directive-image, divide the background area and particularity the bizarre-point; the filtering and raising the image; calculate the ridge-line density; binary and particularity the image, distilling the detail point, validating the detail point, deleting the pseudo detail point; memorizing the fingerprint character module that compressed of the detail point of the fingerprint, bizarre-point, the average ridge-line density and the character of the block directive-image; the character matching step is: collecting the local fingerprint image, distilling the detail point of the fingerprint, bizarre-point, the average ridge-line density and the character of the block directive-image according to the upper step; contrasting the detail point of the fingerprint, bizarre-point, the average ridge-line density and the character of the block directive-image between the fingerprint character module and the local fingerprint image, estimating whether for not is a same finger by the comparability. The invention has high distinguishability, quick distinguish-speed, more reliability and high maneuverability.

Description

Fingerprint identification method and system

Technical field

The present invention relates to a kind of biometric discrimination method, particularly relate to a kind of fingerprint identification method and system.

Background technology

At present, the fingerprint identification method that is used for person identification is different, and still, all the ubiquity discrimination is low for existing fingerprint identification method, the problem that recognition speed is slow.

Summary of the invention

The objective of the invention is to overcome the above-mentioned defective of prior art, a kind of discrimination height is provided, the fingerprint identification method that recognition speed is fast the present invention also aims to provide the recognition system of this method of enforcement.

For achieving the above object, the special feature of fingerprint identification method of the present invention is to be made up of fingerprint characteristic extraction and two steps of characteristic matching:

Characteristic extraction step is: gather fingerprint image, fingerprint image is carried out pre-service and normalization; Calculate to divide block directed graph to extract a singular point, calculated direction figure, cut apart background area and refinement singular point; The filtering of image and enhancing; Calculate ridge density; Minutiae point is extracted in binary image and refinement, minutiae point checking, deletion fake minutiae; Fingerprint minutiae, singular point, average ridge density and block directed graph feature finally are compacted into the fingerprint characteristic template stores;

Character matching step is: the collection site fingerprint image, by fingerprint minutiae, singular point, average ridge density and the block directed graph feature of above-mentioned steps extraction fingerprint on site image; Fingerprint minutiae, singular point, average ridge density and the block directed graph feature of contrast fingerprint characteristic template and fingerprint on site image, the similarity by both features judges whether it is same finger.This fingerprint identification method has the discrimination height, the advantage that recognition speed is fast.

As optimization, character matching step is:

Respectively in computational data library template and the fingerprint on site template minutiae point to line distance, minutiae point to the angle of line and minutiae point direction and minutiae point angle to line; Provided details point to line apart from higher limit and lower limit, the deletion minutiae point to line distance greater than this higher limit and less than the minutiae point of this lower limit to data, obtain a minutiae point more among a small circle to data U;

Adopt the histogram calculation anglec of rotation;

All angles parameter from the fingerprint template of database, comprise minutiae point angle, singular point angle, divide block directed graph and coupling minutiae point to the line direction among the U etc., the angle of calculating according to previous step is rotated, and makes it have consistent direction with the fingerprint template of collection in worksite:

It is right greater than the match point of a designated value to delete corresponding minutiae point differential seat angle from U, makes the coupling minutiae point among the U right to only comprising the most reliable coupling minutiae point;

Same method, calculate the histogram of ranks direction, calculate the statistic histogram of the ranks coordinate difference of the right corresponding minutiae point of all coupling minutiae point, find out the maximum of points in these two arrays, exactly the translational movement of two fingerprint templates after being rotated angular alignment;

Each location parameter from the fingerprint template of database comprises minutiae point coordinate, singular point coordinate, piece direction position etc., carries out translation, two fingerprint template complete matchings;

Delete from U that to comprise ranks coordinate difference right greater than the right coupling of the minutiae point of a designated value, the right similarity of these couplings adds up and obtains the final similarity of two fingerprint template details point sets;

Calculate the similarity of global characteristics:

The singular point similarity is to compare the position of singular point, direction and type in twos, the similarity addition that obtains;

Average ridge density similarity is differing from of two fingerprint template ridge density and gets inverse;

The similarity of block directed graph is the public part two fingerprint template effective coverages, the difference of calculated direction, and inverse is got in the back mean deviation that adds up;

The similarity of two last fingerprint templates is formed by top part and the fusion of global characteristics similarity;

When carrying out the identification of one-to-many, the average ridge density with fingerprint template in the database sorts earlier, and when fingerprint on site was discerned, the immediate fingerprint template of average ridge density in elder generation and the database mated, to accelerate recognition speed; Average ridge density is the average ridge density of whole fingerprint image.

As optimization, when fingerprint characteristic extracted: fingerprint image was expressed as a two-dimensional matrix, and each pixel is exactly an element of matrix, and value is 0～255, and the dimension of matrix is exactly the wide and high of image;

The minutiae point of fingerprint is meant end points or the bifurcation on the fingerprint ridge line, and fingerprint minutiae comprises following feature: coordinate xy-is illustrated in the position in the fingerprint image; Type t-represents the end points or the bifurcation of crestal line; Direction d-represents the direction of minutiae point, if the minutiae point of end points type, then this direction point to crestal line from the minutiae point position, if the branch type minutiae point, then this direction is pointed to the centre of two crestal lines behind the bifurcated from the minutiae point position; Ridge density g-is illustrated near the average density of the crestal line this minutiae point; Ridge curvature c-represents crestal line direction intensity of variation herein;

Divide block directed graph: be the mutually disjoint fritter that fingerprint image is divided into BLOCK_SIZE * BLOCK_SIZE size,, calculate the mean direction of crestal line, be thereby obtain size to the little image of each piece

(HEIGHT/BLOCK_SIZE) * (WIDTH/BLOCK_SIZE) branch block directed graph; The overall crestal line trend of dividing block directed graph portrayal fingerprint image; In addition, on minute block directed graph with one the background area after illegal direction value representation is cut apart fingerprint image;

Singular point: have some local crestal line directions discontinuous on the fingerprint image, these places are called the singular point of fingerprint, and its feature has: coordinate x y is illustrated in the position in the fingerprint image; Type t, singular point are divided into three kinds of core point, double-core point and trigpoints; Direction d, during away from singular point, the fingerprint ridge line direction changes minimum along this direction in expression.Ridge density c is illustrated near the average distance of the crestal line this singular point.

As optimization, image pre-service and normalization are at first image to be carried out even value filtering, make image more level and smooth, then, image are formatd; Calculating branch block directed graph extraction singular point is on block directed graph, calculates the Poincare Index of every bit earlier:

${\mathrm{pindex}}^n=\frac{1}{\mathrm{\π}}\underset{i=0}{\overset{n}{\mathrm{\Σ}}}\mathrm{\τ}(O_{(i+1)\mathrm{mod}n}-O_i)$

$\mathrm{\&tau;}\left(k\right)=\left\{\begin{array}{c}k,\mathrm{if}\left|k\right|<\frac{\mathrm{\&pi;}}{2},\\ \mathrm{\&pi;}+k,\mathrm{if}\left|k\right|<\frac{\mathrm{\&pi;}}{2},\\ \mathrm{\&pi;}-k,\mathrm{otherwise}\end{array}\right.$

Wherein, n is the number of surrounding pixel point, O _iThe direction of representing i point; Getting radius earlier is 1, promptly Zhou Bian 8 points calculate Poincare Index, get p1, if its Poincare Index non-zero, again with radius 2, promptly Zhou Bian outer one deck calculates Poincare Index, it is identical with p2 to get p2:p1, illustrates that this point is a singular point, is core point type singular point if p1 is 1, if p1 is a trigpoint for-1, is the dikaryon singular point if p1 is 2; If p2 is different with p1, but p2＞0, p1＞0 then is the dikaryon singular point; Other situations then are not singular points.

As optimization, calculated direction figure, cut apart the background area and the refinement singular point is: the image after the normalization, calculate first crestal line direction, and calculate the consistance of crestal line direction simultaneously, and obtain directional diagram, redefine the singular point position, original position from these singular points, find the exact position of singular point,, calculate the singular point direction that makes new advances in new position.

As optimization, the filtering and the enhancing of image are: after handling by anisotropic filter, and the fingerprint image that is enhanced; The calculating ridge density is: first calculated fingerprint ridge density figure, carry out 33 * 33 mean filter again to ridge density figure.

As optimization, binary image and refinement are: the image after coming binaryzation to strengthen with the image behind 33 * 33 mean filters as adaptive threshold values; Then the image thinning of binaryzation is become the crestal line figure of single-point width; Image thinning is that each black picture element in the image has 8 consecutive point, judges according to them whether current point should be changed to white.Through repeatedly multiple scanning, made into white like this, just obtained the fingerprint ridge line chart of refinement up to the neither one black color dots.

As optimization, extract minutiae point and be: eliminate burr and noise earlier, promptly, follow the tracks of crestal line, if, just it is erased from refinement figure from the pixel distance of crestal line origin-to-destination threshold values less than a setting by the crestal line figure of scanning refinement; Then, extract minutiae point: promptly to any one black color dots on the image, if in its 8 adjacent points, optional starting point, run-down is got back to starting point in the direction of the clock, and its change in color illustrates that this point is a termination type minutiae point if 2 times; If more than 4 times, this point is the branch type minutiae point, and other situations then can be ignored, and by scanning effective fingerprint image zone, has obtained all minutiae point;

Follow the tracks of crestal line at the minutiae point place, obtain the direction of crestal line; The crestal line curvature of minutiae point is represented with the variation of direction, on the directional diagram of fingerprint image, calculates curvature with near the direction this point and the direction difference of this point.

As optimization, minutiae point checking and deletion fake minutiae are: any one minutiae point, if exist to come a minutiae point, then delete this minutiae point with it apart from less than a setting value D1; As an end points type minutiae point with come end points type minutiae point distance less than a setting value D2, and their directions are opposite, then delete this two minutiae point simultaneously; If an end points type minutiae point and a branch type minutiae point distance are less than a setting value D3, and their directions are opposite, then delete this two minutiae point simultaneously; If less than a setting value D4, and direction outwardly, then deletes this minutiae point from the inactive area of fingerprint image for minutiae point; Obtain final minutiae point by above-mentioned deletion.

A kind of recognition system that is used to implement fingerprint identification method of the present invention, its special feature be to comprise fingerprint capturer, fingerprint recognition system, identification or and the control signal output mechanism; Comprising fingerprint image storer, fingerprint image processor and fingerprint characteristic data storer; Fingerprint image processor is to utilize to require the described method of one of 1-9 that fingerprint image is handled and discerned.It has the discrimination height, and recognition speed is fast, good reliability, workable advantage.

Wherein: the character representation of fingerprint minutiae (x, y, t, d, g c) comprises more information, helps improving the discrimination of system; The character representation of fingerprint singularity (x, y, t, d g) comprises more information, helps improving the discrimination of system; Average ridge line density G can carry out index with this as a global characteristics, and aid identification is with pick up speed.The block directed graph of fingerprint is kept in the fingerprint template as a global characteristics, carries out the block directed graph comparison in comparison process, and its similarity is fused among the last result; The extracting method of singular point can calculate singular point position and feature accurately fast; Anisotropic filter is used to strengthen fingerprint image, and effect is fine; After wave filter is modulated by the each point direction on the fingerprint image, adopt the method for convolution, this point is carried out filtering.Because the filter kernel of every bit all is subjected to the modulation of this direction, so the filtering of the effect of filtering comparison image block is far better; By preserving the anisotropic filter coefficient of all directions, make and when convolution, to use look-up table.Improved the speed of filtering greatly; The flow process of fingerprint comparison, the last similarity of fingerprint template coupling obtains by the similarity that merges various features, and this makes the result more reliable; Minutiae point alignment schemes, this method be by estimating the right transformation parameter of minutiae point line of preliminary coupling, valuation added up in the histogram of generation and found final transformation parameter.

After adopting technique scheme, fingerprint identification method of the present invention has the discrimination height, and recognition speed is fast, good reliability, workable advantage.

Description of drawings:

Fig. 1 is the synoptic diagram of three kinds of singular points in the fingerprint identification method of the present invention;

Fig. 2 is the process flow diagram of fingerprint identification method of the present invention;

Fig. 3 is the synoptic diagram of 8 points of periphery of p1 in the fingerprint identification method of the present invention;

Fig. 4 is the synoptic diagram of 12 points of periphery of p2 in the fingerprint identification method of the present invention;

Fig. 5 is that direction is the synoptic diagram that zero anisotropic filter is examined in the fingerprint identification method of the present invention;

Fig. 6 is the organigram that 8 consecutive point are converted to table index numbers 22 in the fingerprint identification method of the present invention;

Fig. 7 is three kinds of main refinement crestal line noise patterns in the fingerprint identification method of the present invention;

Fig. 8 is adjacent 8 the change color figure of termination type minutiae point in the fingerprint identification method of the present invention;

Fig. 9 is adjacent 8 the change color figure of branch type minutiae point in the fingerprint identification method of the present invention;

Figure 10 be in the fingerprint identification method of the present invention minutiae point between line graph;

Figure 11 is the former fingerprint image in the fingerprint identification method of the present invention;

Figure 12 is the fingerprint image after the normalization in the fingerprint identification method of the present invention;

Figure 13 is the directional diagram of the fingerprint in the fingerprint identification method of the present invention;

Figure 14 is the enhancing image of the fingerprint in the fingerprint identification method of the present invention;

Figure 15 is the binary image of the fingerprint in the fingerprint identification method of the present invention;

Figure 16 is the refinement crestal line figure of the fingerprint in the fingerprint identification method of the present invention.

Do further explanation below in conjunction with accompanying drawing and instantiation:

Algorithm for recognizing fingerprint relates to two topmost steps: feature extraction and characteristic matching.

Feature extraction: the Flame Image Process of fingerprint and the take the fingerprint overall situation and local feature, and save as fingerprint template;

Characteristic matching: two fingerprint characteristic templates are compared, obtain a coupling mark, determine according to this mark whether two fingerprints are same then.

One, feature extraction

1, notion and agreement

1) expression of fingerprint image

Fingerprint image is expressed as a two-dimensional matrix, and each pixel is exactly an element of a matrix, and value is (0～255), and the dimension of matrix is exactly the wide WIDTH and the high HEIGHT of image.The gray-scale value of the capable j row of the i on the fingerprint image is expressed as I _Ij

2) expression of local feature

The local feature of fingerprint is meant end points or the bifurcation on the fingerprint ridge line, is called the minutiae point of fingerprint.Fingerprint minutiae comprise following feature (x, y, t, d, g, c):

Coordinate xy: be illustrated in the position in the fingerprint image;

Type t: expression is the end points or the bifurcation of crestal line;

Direction d: the direction of expression minutiae point.If the minutiae point of end points type, then this direction points to crestal line from the minutiae point position; If the branch type minutiae point, then this direction is pointed to the centre of two crestal lines behind the bifurcated from the minutiae point position.

Ridge density g: the average density that is illustrated near the crestal line of this minutiae point.The spacing distance of crestal line is big more, and density is just more little;

Ridge curvature c: expression crestal line direction intensity of variation herein

3) expression of global characteristics

Divide block directed graph

Fingerprint image is divided into the mutually disjoint fritter of BLOCK_SIZE * BLOCK_SIZE size,, calculates the mean direction of crestal line, be thereby obtain size to the little image of each piece

(HEIGHT/BLOCK_SIZE) * (WIDTH/BLOCK_SIZE) branch block directed graph.Block directed graph has been portrayed the overall crestal line trend of fingerprint image, and the global characteristics as fingerprint image is stored, and is used for later comparison.In addition, on minute block directed graph with one the background area (do not have fingerprint image herein, or the fingerprint image quality being too poor) after illegal direction value representation is cut apart fingerprint image.

Singular point

The crestal line direction of fingerprint has successional feature, and promptly in general the crestal line direction of adjacent position is consistent or changes little.Yet, also there are some local crestal line directions discontinuous on the fingerprint image, these places are called the singular point of fingerprint.

The feature of singular point have (x, y, t, d, c): x y coordinate: be illustrated in the position in the fingerprint image.Type t: as shown in Figure 1, singular point is divided into three kinds of core point 1-1, double-core point 1-2 and trigpoint 1-3.Direction d: during away from singular point, the fingerprint ridge line direction changes minimum along this direction in expression.Ridge density c: the average distance that is illustrated near the crestal line of this singular point.The average ridge line density is the average ridge line density of whole fingerprint image.

2, algorithm flow

2.1 process flow diagram is asked for an interview accompanying drawing 2.

2.2 image pre-service and normalization

At first image is carried out 3 * 3 mean filter, make image more level and smooth.

$R_{i,j}=\frac{\underset{y=i=w}{\overset{i+w}{\mathrm{\&Sigma;}}}\underset{x=j=w}{\overset{j+w}{\mathrm{\&Sigma;}}}{I}_{y,x}}{{(w+1)}^2}$

Wherein, I _{Y, x}Be original image, R _{Y, x}Be the image after level and smooth, get w=1 here.

Then, image is standardized:

$R_{i,j}=\frac{255(I_{i,j}-{\mathrm{Min}}_{i,j})}{{\mathrm{\&Delta;}}_{i,j}}$

Min _i，j＝I _i，j-Var _i，j

Max _i，j＝I _i，j+Var _i，j

Δ _i，j＝Max _i，j-Min _i，j

${\mathrm{Var}}_{i,j}=\frac{\underset{y=i-w}{\overset{i+w}{\mathrm{\&Sigma;}}}\stackrel{j+w}{\underset{x=j-w}{\mathrm{\&Sigma;}}}|I_{y,x}-S_{y,x}|}{{(w+1)}^2}$

S wherein _{Y, x}Be original image through the image of 5 * 5 mean filter, in the calculating of Var, get a bigger neighborhood w=80.

Divide block directed graph to extract singular point 2.3 calculate

Divide the calculating of block directed graph the same with the calculating of complete computation directional diagram, the direction of center of only only calculating piecemeal is just passable, and the calculating of calculated direction figure can be introduced below specially.

On block directed graph, calculate the Poincare Index of every bit:

${\mathrm{pindex}}^n=\frac{1}{\mathrm{\&pi;}}\underset{i=0}{\overset{n}{\mathrm{\&Sigma;}}}\mathrm{\&tau;}(O_{(i+1)\mathrm{mod}n}-O_i)$

$\mathrm{\&tau;}\left(k\right)=\left\{\begin{array}{c}k,\mathrm{if}\left|k\right|<\frac{\mathrm{\&pi;}}{2},\\ \mathrm{\&pi;}+k,\mathrm{if}\left|k\right|<\frac{\mathrm{\&pi;}}{2},\\ \mathrm{\&pi;}-k,\mathrm{otherwise}\end{array}\right.$

Wherein, n is the number of surrounding pixel point, O _iThe direction of representing i point.In order to guarantee computation's reliability, getting radius earlier is 1, and promptly Zhou Bian 8 points calculate Poincare Index, gets p1, if its Poincare Index non-zero, again with radius 2, promptly Zhou Bian outer one deck calculates Poincare Index, gets p2.Wherein the synoptic diagram of 8 points of periphery of p1 is seen accompanying drawing 3, and the synoptic diagram of 12 points of periphery of p2 is seen accompanying drawing 4.

There is following situation:

P1 is identical with p2, illustrates that this point is a singular point, is core point type singular point if p1 is 1, if p1 is a trigpoint for-1, is the dikaryon singular point if p1 is 2; If p2 is different with p1, but p2＞0, p1＞0 then is the dikaryon singular point; Other situations then are not singular points.

2.4 calculated direction figure, cut apart background area and refinement singular point

To the image after the normalization, calculate the crestal line direction O of every bit by following formula _{I, j}

$G_{i,j}^{\mathrm{xx}}={\left(G_{i,j}^x\right)}^2,G_{i,j}^{\mathrm{yy}}={\left(G_{i,j}^y\right)}^2,G_{i,j}^{\mathrm{xy}}=G_{i,j}^x{G}_{i,j}^y$

$G_{i,j}^x=(I_{i-1,j+1}-I_{i-1,j-1})+4(I_{i,j+1}-I_{i,j-1})+(I_{i+1,j+1}-I_{i+1,j-1})$

$G_{i,j}^y=(I_{i+1,j-1}-I_{i-1,j-1})+4(I_{i+1,j}-I_{i-1,j})+(I_{i+1,j+1}-I_{i-1,j+1})$

$g_{i,j}^{\mathrm{xx}}=\frac{\underset{y=i-w}{\overset{i+w}{\mathrm{\&Sigma;}}}\underset{x=j-w}{\overset{j+w}{\mathrm{\&Sigma;}}}{G}_{y,x}^{\mathrm{xx}}}{{(2w+1)}^2},g_{i,j}^{\mathrm{yy}}\frac{\underset{y=i-w}{\overset{i+w}{\mathrm{\&Sigma;}}}\underset{x=j-w}{\overset{j+w}{\mathrm{\&Sigma;}}}{G}_{y,x}^{\mathrm{yy}}}{{(2w+1)}^2},g_{i,j}^{\mathrm{xy}}=\frac{\underset{y=i-w}{\overset{i+w}{\mathrm{\&Sigma;}}}\underset{x=j-w}{\overset{j+w}{\mathrm{\&Sigma;}}}{G}_{y,x}^{\mathrm{xy}}}{{(2w+1)}^2}$

$O_{i,j}={\tan }^{-1}\frac{2g_{i,j}^{\mathrm{xy}}}{g_{i,j}^{\mathrm{xx}}-g_{i,j}^{\mathrm{yy}}}$

And calculate the consistance C of crestal line direction simultaneously _{I, j}

$m_{i,j}=\sqrt{g_{i,j}^{\mathrm{xx}}+g_{i,j}^{\mathrm{yy}}}$

$C_{i,j}=\left\{\begin{array}{c}\frac{\sqrt{{(g_{i,j}^{\mathrm{xx}}-g_{i,j}^{\mathrm{yy}})}^2+4g_{i,j}^{\mathrm{xy}2}}}{g_{i,j}^{\mathrm{xx}}+g_{i,j}^{\mathrm{yy}}},\mathrm{if}{m}_{i,j}\&GreaterEqual;\mathrm{<figure-callout\; id="0"\; label="Threshold"\; filenames="A200610065297.500211.png,A200610065297.500221.png"\; state="\{\{state\}\}">Threshold</figure-callout>}\\ 0,\mathrm{otherwise}\end{array}\right.$

Threshold is the threshold values of a setting, C _{I, j}=0 expression is the fingerprint image background area herein.

The position of the singular point of the fingerprint that is obtained by top block directed graph is coarse, can use the directional diagram of obtaining now, redefines these singular point positions.From the original position of these singular points, can find C in its vicinity _{I, j}Minimum point has been exactly the exact position of singular point, in new position, calculates the singular point direction that makes new advances.

2.5 the filtering of image and enhancing

Design an anisotropic filter:

$h(x,y,\mathrm{\&theta;})=\mathrm{\&rho;}(x,y)\exp (-\frac{{(x\cos \mathrm{\&theta;}+y\sin \mathrm{\&theta;})}^2}{{\mathrm{\&delta;}}_1^2}-\frac{{(x\sin \mathrm{\&theta;}-y\cos \mathrm{\&theta;})}^2}{{\mathrm{\&delta;}}_2^2})$

$\mathrm{\&rho;}(x,y)=\left\{\begin{array}{c}a,\mathrm{if}{\mathrm{<figure-callout\; id="2"\; label="x"\; filenames="A200610065297.500211.png,A200610065297.500221.png"\; state="\{\{state\}\}">x</figure-callout>}}^2+{\mathrm{<figure-callout\; id="2"\; label="y"\; filenames="A200610065297.500211.png,A200610065297.500221.png"\; state="\{\{state\}\}">y</figure-callout>}}^2\&le;{\mathrm{<figure-callout\; id="2"\; label="r"\; filenames="A200610065297.500211.png,A200610065297.500221.png"\; state="\{\{state\}\}">r</figure-callout>}}^2\\ 0,\mathrm{otherwise}\end{array}\right.$

Wherein, r is an effective radius, gets 6 usually, and a is an amplitude coefficient, gets 1024 usually, δ ₁ ²And δ ₂ ²Be the shape controlled variable of wave filter, be taken as 8 and 1 usually.θ is the modulation direction of this wave filter.Direction is that zero anisotropic filter nuclear is asked for an interview accompanying drawing 5.Thereby, can use following formula to calculate the fingerprint image (convolution) that strengthens:

$R_{i,j}=\frac{\underset{y=i-r}{\overset{i+r}{\mathrm{\&Sigma;}}}\underset{x=j-r}{\overset{j+r}{\mathrm{\&Sigma;}}}h(x-i,y-j,O_{y,x})I_{y,x}}{\underset{y=i-r}{\overset{i+r}{\mathrm{\&Sigma;}}}\underset{x=j-r}{\overset{j+r}{\mathrm{\&Sigma;}}}h(x-i,y-j,O_{y,x})}$

In order to calculate following formula apace, calculate and preserve the filter factor h of all angles and the denominator term in the following formula in advance, table look-up when specifically calculating with image and directly carry out convolution.

2.6 calculate ridge density figure

Density map D as shown in the formula the calculated fingerprint crestal line:

$D_{i,j}=\frac{\underset{y=i-w}{\overset{i+w}{\mathrm{\&Sigma;}}}\underset{x=j-w}{\overset{j+w}{\mathrm{\&Sigma;}}}{P}_{y,x}}{\underset{y=i-w}{\overset{i+w}{\mathrm{\&Sigma;}}}\underset{x=j-w}{\overset{j+w}{\mathrm{\&Sigma;}}}{C}_{y,x}}g\frac{255\mathrm{Koef}\left(O_{i,j}\right)}{\mathrm{KoefP}}$

$P_{i,j}=\left\{\begin{array}{c}1,\mathrm{if}\underset{y=i-w}{\overset{i+w}{\mathrm{\&Sigma;}}}\underset{x=j-w}{\overset{j+w}{\mathrm{\&Sigma;}}}{I}_{y,x}\&Element;[{\mathrm{Threshold}}_{\mathrm{bottom}},{\mathrm{Threshold}}_{\mathrm{top}}]\\ \mathrm{and}[i,j]\mathrm{is\; not\; in\; the\; <figure-callout\; id="0"\; label="bad\; area"\; filenames="A200610065297.500211.png,A200610065297.500221.png"\; state="\{\{state\}\}">bad\; area</figure-callout>}\\ 0,\mathrm{otherwise}\end{array}\right.$

$C_{i,j}=\left\{\begin{array}{c}1,[i,j]\mathrm{is\; not\; in\; the\; <figure-callout\; id="0"\; label="bad\; area"\; filenames="A200610065297.500211.png,A200610065297.500221.png"\; state="\{\{state\}\}">bad\; area</figure-callout>}\\ 0,\mathrm{otherwise}\end{array}\right.$

The ridge density of fingerprint is very continuous, therefore to ridge density figure D _{I, j}Carry out 33 * 33 mean filter again to eliminate noise.

2.7 binary image and refinement

According to following formula the image after strengthening is carried out binaryzation:

$R_{i,j}=\left\{\begin{array}{c}0,\mathrm{if}{I}_{i,j}<S_{i,j}\\ 255,\mathrm{otherwise}\end{array}\right.$

Wherein, S _{I, j}Be to the image of the fingerprint image after strengthening after, come the binaryzation fingerprint image as adaptive threshold values with this image with one 33 * 33 mean filter.

Then the image thinning of binaryzation is become the crestal line figure of single-point width.Algorithm is: consider 8 consecutive point of each black picture element in the image, judge according to them whether current point should be changed to white.Through repeatedly multiple scanning, made into white like this, just obtained the fingerprint ridge line chart of refinement up to the neither one black color dots.

In the binary image, 8 consecutive point of a black picture element can have 256 kinds of situations altogether, can be by the quick judgement of having tabled look-up in the actual calculation.The table of setting up 256 elements is 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，

0,0,0,0,0,0,0,0,0,0,0,0,1,0,0,0,0,0,0,0,1,0,0,0,1,0,1, its value of 0} is whether 1 expression will should be changed to white to current pixel, and the index of table is constructed with following mode: 8 consecutive point are converted to table index number 22 (binary zeros 0010110), ask for an interview accompanying drawing 6.

2.8 extraction minutiae point

Because the noise of image, the fingerprint ridge line after the refinement is strivied for survival at jagged phenomenon and noise, so when extracting minutiae point, must eliminate them earlier, otherwise will extract a lot of false minutiae point.Three kinds of main refinement crestal line noise patterns are asked for an interview accompanying drawing 7.

By the crestal line figure of scanning refinement, follow the tracks of crestal line, if, just can erase it from refinement figure from the pixel distance of crestal line origin-to-destination threshold values less than a setting.

Then, can extract minutiae point easily:

To any one black color dots on the image, if in its 8 adjacent points, optional starting point, run-down is got back to starting point in the direction of the clock, and its change in color illustrates that this point is a termination type minutiae point if 2 times; If more than 4 times, this point is the branch type minutiae point, and other situations then can be ignored.Adjacent 8 change color of termination type minutiae point is 2 times: 6-＞7, and accompanying drawing 8 is asked for an interview in 7-＞0; Adjacent 8 change color of branch type minutiae point is more than 4 times: 1-＞2, and 2-＞3,3-＞4,4-＞5,6-＞7, accompanying drawing 9 is asked for an interview in 7-＞0.

Like this, by scanning effective fingerprint image zone, all minutiae point have been obtained.Follow the tracks of crestal line at the minutiae point place, can obtain the direction of crestal line.

The crestal line curvature of minutiae point can be represented with the variation of direction.On the directional diagram of fingerprint image, calculate curvature with near the direction this point and the direction difference of this point:

$\underset{x^2+{\mathrm{<figure-callout\; id="2"\; label="y"\; filenames="A200610065297.500211.png,A200610065297.500221.png"\; state="\{\{state\}\}">y</figure-callout>}}^2\&NotEqual;0}{\underset{x^2+{\mathrm{<figure-callout\; id="2"\; label="y"\; filenames="A200610065297.500211.png,A200610065297.500221.png"\; state="\{\{state\}\}">y</figure-callout>}}^2<r^2}{\mathrm{\&Sigma;}}}|O_{i+x,j+y}-O_{i,j}|$

Wherein, r is a radius constant, gets 10. usually

2.9 minutiae point checking

So far the minutiae point that obtains because the cause of picture noise still has a lot of fake minutiaes inside, needs further to reject.Consider following situation:

Consider any one minutiae point,, then delete this minutiae point if exist to come a minutiae point with it apart from less than a setting value D1;

As an end points type minutiae point with come end points type minutiae point distance less than a setting value D2, and their directions are opposite, then delete this two minutiae point simultaneously;

If an end points type minutiae point and a branch type minutiae point distance are less than a setting value D3, and their directions are opposite, then delete this two minutiae point simultaneously;

If less than a setting value D4, and direction outwardly, then deletes this minutiae point from the inactive area of fingerprint image for minutiae point.

Obtain final minutiae point like this.

Fingerprint minutiae and other global characteristics finally are compacted into the fingerprint characteristic template stores.

Two, finger print matching method

The minutiae point matching process is based on the minutiae point line.

Consider two minutiae point m on the fingerprint image _i, m _jLine, the definition:

d _IjBe the length of line segment, i.e. distance between two minutiae point;

a _iAnd b _jBe respectively the angle of line and minutiae point direction;

u _IjAngle for line.

Minutiae point between line, as shown in Figure 10.

The base unit of such line segment, come the minutiae point on two fingerprint images of comparison right as the fingerprint minutiae coupling.Right for a minutiae point, d _Ij, a ₁, a ₂With type t, curvature c, the ridge density g of two minutiae point all be translation invariant and invariable rotary.Thus, can compare this tittle, determine two similaritys that minutiae point is right.

For the minutiae point on the fingerprint on site template to (m _I1, m _J1) and database in minutiae point on the fingerprint template to (m _I2, m _J2), " similarity " that the definition minutiae point is right is as follows:

$D={\mathrm{<figure-callout\; id="1"\; label="cof"\; filenames="A200610065297.500211.png,A200610065297.500221.png"\; state="\{\{state\}\}">cof</figure-callout>}}_1|d_{i_1{j}_1}-d_{i_2{\mathrm{<figure-callout\; id="2"\; label="j"\; filenames="A200610065297.500211.png,A200610065297.500221.png"\; state="\{\{state\}\}">j</figure-callout>}}_2}|+{\mathrm{cof}}_2\left(\right|a_{i_1}-a_{{\mathrm{<figure-callout\; id="2"\; label="i"\; filenames="A200610065297.500211.png,A200610065297.500221.png"\; state="\{\{state\}\}">i</figure-callout>}}_2}|+|b_{j_1}-{\mathrm{<figure-callout\; id="2"\; label="b\; j"\; filenames="A200610065297.500211.png,A200610065297.500221.png"\; state="\{\{state\}\}">b</figure-callout>}}_{j_{}}\left|\right)$

$+{\mathrm{cof}}_3\left(\right|c_{i_1}-{\mathrm{<figure-callout\; id="2"\; label="c\; i"\; filenames="A200610065297.500211.png,A200610065297.500221.png"\; state="\{\{state\}\}">c</figure-callout>}}_{i_{}}|+|c_{j_1}-{\mathrm{<figure-callout\; id="2"\; label="c\; j"\; filenames="A200610065297.500211.png,A200610065297.500221.png"\; state="\{\{state\}\}">c</figure-callout>}}_{j_{}}\left|\right)+{\mathrm{cof}}_4\left(\right|g_{i_1}-{\mathrm{<figure-callout\; id="2"\; label="g\; i"\; filenames="A200610065297.500211.png,A200610065297.500221.png"\; state="\{\{state\}\}">g</figure-callout>}}_{i_{}}|+|g_{j_1}-{\mathrm{<figure-callout\; id="2"\; label="g\; j"\; filenames="A200610065297.500211.png,A200610065297.500221.png"\; state="\{\{state\}\}">g</figure-callout>}}_{j_{}}\left|\right)$

$+{\mathrm{cof}}_5\left(\right|t_{i_1}-{\mathrm{<figure-callout\; id="2"\; label="t\; i"\; filenames="A200610065297.500211.png,A200610065297.500221.png"\; state="\{\{state\}\}">t</figure-callout>}}_{i_{}}|+|t_{j_1}-{\mathrm{<figure-callout\; id="2"\; label="t\; j"\; filenames="A200610065297.500211.png,A200610065297.500221.png"\; state="\{\{state\}\}">t</figure-callout>}}_{j_{}}\left|\right)$

Wherein, cof ₁, cof ₂, cof ₃, cof ₄, cof ₅Be positive constant coefficient.As the front definition, c, g, t are respectively curvature, ridge density and the type codes (the branch type minutiae point is 1, and terminal type minutiae point is 0) of each minutiae point respectively for d, a, b.As D during, think that just these two minutiae point are to being complementary less than a specified value ThresholdD.

If a fingerprint image has N minutiae point, it is right then can to produce the individual minutiae point of C (N, 2), will be they C (M with M minutiae point generation of another fingerprint image, 2) individual minutiae point just need be carried out C (M, 2) * C (N to comparison one by one, 2) inferior comparison, if M=N=80, then the number of times that will compare is 39942400, and this can be very slow, so, be necessary before comparison, to do restriction.Stipulate two value ThresholdD1, ThresholdD2 only considers that the minutiae point of wire length between these two values is right.This can reduce comparison institute's time spent greatly.

From the computing formula of D, be easy to find out: if first result of calculation is greater than ThresholdD, need not calculate the item of back and just know that two minutiae point are not to being complementary, therefore can be earlier the minutiae point in two fingerprints to sorting according to its wire length d, just can in the small neighbourhood of a wire length d, compare.This has accelerated computing velocity greatly.

It is right to neglect the minutiae point that is not complementary, and can obtain a minutiae point that is complementary to tabulating:

$U=\{{\mathrm{\&mu;}}_{i_1{\mathrm{<figure-callout\; id="1"\; label="j"\; filenames="A200610065297.500211.png,A200610065297.500221.png"\; state="\{\{state\}\}">j</figure-callout>}}_1,i_2{\mathrm{<figure-callout\; id="2"\; label="j"\; filenames="A200610065297.500211.png,A200610065297.500221.png"\; state="\{\{state\}\}">j</figure-callout>}}_2}=(l_{i_1{\mathrm{<figure-callout\; id="1"\; label="j"\; filenames="A200610065297.500211.png,A200610065297.500221.png"\; state="\{\{state\}\}">j</figure-callout>}}_1},l_{i_2{\mathrm{<figure-callout\; id="2"\; label="j"\; filenames="A200610065297.500211.png,A200610065297.500221.png"\; state="\{\{state\}\}">j</figure-callout>}}_2},S_{i_1{\mathrm{<figure-callout\; id="1"\; label="j"\; filenames="A200610065297.500211.png,A200610065297.500221.png"\; state="\{\{state\}\}">j</figure-callout>}}_1,i_2{j}_2})|D_{i_1{\mathrm{<figure-callout\; id="1"\; label="j"\; filenames="A200610065297.500211.png,A200610065297.500221.png"\; state="\{\{state\}\}">j</figure-callout>}}_1,i_2{\mathrm{<figure-callout\; id="2"\; label="j"\; filenames="A200610065297.500211.png,A200610065297.500221.png"\; state="\{\{state\}\}">j</figure-callout>}}_2}<\mathrm{ThresholdD}\}$

$S_{i_1{j}_1{i}_2{\mathrm{<figure-callout\; id="2"\; label="j"\; filenames="A200610065297.500211.png,A200610065297.500221.png"\; state="\{\{state\}\}">j</figure-callout>}}_2}=1-\frac{D_{i_1{\mathrm{<figure-callout\; id="1"\; label="j"\; filenames="A200610065297.500211.png,A200610065297.500221.png"\; state="\{\{state\}\}">j</figure-callout>}}_1,i_2{\mathrm{<figure-callout\; id="2"\; label="j"\; filenames="A200610065297.500211.png,A200610065297.500221.png"\; state="\{\{state\}\}">j</figure-callout>}}_2}}{\mathrm{ThresholdD}}$

L wherein _I1j1Be that minutiae point from the fingerprint on site template is to (m _I1, m _J1) and line, l _I12j2Be that minutiae point from the fingerprint template of database is to (m _I2, m _J2) and line, D be these two pairs of minutiae point to " distance ", S be two pairs of minutiae point to similarity.

Adopt following method, tabulate from this and calculate the similarity of two fingerprint templates:

1, adopt histogram method to calculate the anglec of rotation

Set an one-dimension array { H _d| 0≤d＜360}, its subscript is represented the angle from 0～359, each element is as shown in the formula calculating:

$H_d=\underset{{\mathrm{\&mu;}}_{i_1{j}_1{i}_2{\mathrm{<figure-callout\; id="2"\; label="j"\; filenames="A200610065297.500211.png,A200610065297.500221.png"\; state="\{\{state\}\}">j</figure-callout>}}_2}\&Element;U}{\mathrm{\&Sigma;}}{S}_{i_1{j}_1{i}_2{j}_2}g({\mathrm{\&sigma;}}_d(d_{i_1}-d_{i_2})+{\mathrm{\&sigma;}}_d(d_{j_1}-d_{j_2})$

Wherein, σ _dBe to be 1 in the d value, other all value be 0 uni-impulse function.As seen, { H _dBe actually the statistic histogram of the differential seat angle of the right corresponding minutiae point of all coupling minutiae point.Find out the maximum of points in this array, the anglec of rotation θ of two fingerprint templates that need exactly.

2, all angles parameter from the fingerprint template of database, comprise minutiae point angle, singular point angle, divide block directed graph and coupling minutiae point to the line direction among the U etc., the angle of calculating according to previous step is rotated, and makes it have consistent direction with the fingerprint template of collection in worksite:

(d _i+θ)mod360→d _i

3, it is right greater than the match point of a designated value to delete corresponding minutiae point differential seat angle from U, and like this, the coupling minutiae point among the U is right to just only having comprised the most reliable coupling minutiae point.

4, same method is calculated the histogram of ranks direction

Set two one-dimension array { HX _DxAnd { HY _Dy}:

{HX _dx|-MaxDim≤dx≤MaxDim}

${\mathrm{HX}}_{\mathrm{dx}}=\underset{{\mathrm{\&mu;}}_{i_1{j}_1{i}_2{\mathrm{<figure-callout\; id="2"\; label="j"\; filenames="A200610065297.500211.png,A200610065297.500221.png"\; state="\{\{state\}\}">j</figure-callout>}}_2}\&Element;U}{\mathrm{\&Sigma;}}{S}_{i_1{j}_1{i}_2{j}_2}g\frac{{\mathrm{\&sigma;}}_{\mathrm{dx}}(x_{i_1}-x_{i_2})+{\mathrm{\&sigma;}}_{\mathrm{dx}}(x_{j_1}-x_{j_2})}{2}$

{HY _dy|-MaxDim≤dy≤MaxDim}

${\mathrm{HY}}_{\mathrm{dy}}=\underset{{\mathrm{\&mu;}}_{i_1{j}_1{i}_2{\mathrm{<figure-callout\; id="2"\; label="j"\; filenames="A200610065297.500211.png,A200610065297.500221.png"\; state="\{\{state\}\}">j</figure-callout>}}_2}\&Element;U}{\mathrm{\&Sigma;}}{S}_{i_1{j}_1{i}_2{j}_2}g\frac{{\mathrm{\&sigma;}}_{\mathrm{dy}}(y_{i_1}-y_{i_2})+{\mathrm{\&sigma;}}_{\mathrm{dy}}(y_{j_1}-y_{j_2})}{2}$

As seen, { HY _Dy, { HX _DxBe actually the statistic histogram of the ranks coordinate difference of the right corresponding minutiae point of all coupling minutiae point.Find out the maximum of points in these two arrays, the translational movement (x after being rotated angular alignment of two fingerprint templates that need exactly ₀, y ₀).

5, each location parameter, comprise minutiae point coordinate, singular point coordinate, piece direction position etc., carry out translation from the fingerprint template of database

$\left(\begin{array}{c}{x}_i\\ y_i\end{array}\right)+\left(\begin{array}{c}{x}_0\\ {\mathrm{<figure-callout\; id="0"\; label="y"\; filenames="A200610065297.500211.png,A200610065297.500221.png"\; state="\{\{state\}\}">y</figure-callout>}}_0\end{array}\right)\&RightArrow;\left(\begin{array}{c}{x}_i\\ y_i\end{array}\right)$

Now, two fingerprint templates are with regard to complete matching.

6, delete from U that to comprise ranks coordinate difference right greater than the right coupling of the minutiae point of a designated value, like this, the right coupling of the minutiae point among the U is to mating fully.Adding up from the right similarity of these couplings is exactly the final similarity S of two fingerprint template details point sets _m

7, in the top calculating, also, global characteristics is alignd by translation and rotation.At this moment, can calculate the similarity of global characteristics very simply.

Singular point similarity S _s: compare position, direction and the type of singular point in twos, the similarity addition that obtains;

Average ridge density similarity S _g: the differing from and get inverse of two fingerprint template ridge density;

The similarity S of block directed graph _d: the public part two fingerprint template effective coverages, the difference of calculated direction, inverse is got in the back mean deviation that adds up.

8, the similarity of two last fingerprint templates is formed by top part and the fusion of global characteristics similarity:

S＝k _mS _m+k _sS _s+k _gS _g+k _dS _d

Wherein, k _m, k _s, k _g, k _dIt is the weight coefficient of various characteristic matching similarities.

Note average ridge density similarity S _gCalculating be exactly the poor of two average ridge density, therefore, if carry out the identification of one-to-many, then can sort the database of fingerprint template according to average ridge density G earlier, when fingerprint on site is discerned, just can be preferentially and the fingerprint template that taps into most of the average ridge density in the database mate, like this, because the database of fingerprint template according to the G index, therefore can be accelerated identifying greatly.

Fingerprint image in the processing procedure is asked for an interview accompanying drawing:

Wherein: accompanying drawing 11 is former fingerprint images, and accompanying drawing 12 is the images after the normalization, and accompanying drawing 13 is directional diagrams, and accompanying drawing 14 is to strengthen image, and accompanying drawing 15 is binary images, and accompanying drawing 16 is refinement crestal line figure.

The fingerprint recognition system that is used to implement fingerprint identification method of the present invention comprise fingerprint capturer, fingerprint recognition system, identification or and the control signal output mechanism; Comprising fingerprint image storer, fingerprint image processor and fingerprint characteristic data storer; Fingerprint image processor is to utilize to require the described method of one of 1-9 that fingerprint image is handled and discerned.

Claims (10)

1, a kind of fingerprint identification method is characterized in that being made up of fingerprint characteristic extraction and two steps of characteristic matching:

Characteristic extraction step is: gather fingerprint image, fingerprint image is carried out pre-service and normalization; Calculate to divide block directed graph to extract a singular point, calculated direction figure, cut apart background area and refinement singular point; The filtering of image and enhancing; Calculate ridge density; Minutiae point is extracted in binary image and refinement, minutiae point checking, deletion fake minutiae; Fingerprint minutiae, singular point, average ridge density and block directed graph feature finally are compacted into the fingerprint characteristic template stores;

Character matching step is: the collection site fingerprint image, by fingerprint minutiae, singular point, average ridge density and the block directed graph feature of above-mentioned steps extraction fingerprint on site image; Fingerprint minutiae, singular point, average ridge density and the block directed graph feature of contrast fingerprint characteristic template and fingerprint on site image, the similarity by both features judges whether it is same finger.

2,, it is characterized in that character matching step is according to the described fingerprint identification method of claim 1:

Respectively in computational data library template and the fingerprint on site template minutiae point to line distance, minutiae point to the angle of line and minutiae point direction and minutiae point angle to line; Provided details point to line apart from higher limit and lower limit, the deletion minutiae point to line distance greater than this higher limit and less than the minutiae point of this lower limit to data, obtain a minutiae point more among a small circle to data U;

Adopt the histogram calculation anglec of rotation;

All angles parameter from the fingerprint template of database, comprise minutiae point angle, singular point angle, divide block directed graph and coupling minutiae point to the line direction among the U etc., the angle of calculating according to previous step is rotated, and makes it have consistent direction with the fingerprint template of collection in worksite:

It is right greater than the match point of a designated value to delete corresponding minutiae point differential seat angle from U, makes the coupling minutiae point among the U right to only comprising the most reliable coupling minutiae point;

Same method, calculate the histogram of ranks direction, calculate the statistic histogram of the ranks coordinate difference of the right corresponding minutiae point of all coupling minutiae point, find out the maximum of points in these two arrays, exactly the translational movement of two fingerprint templates after being rotated angular alignment;

Each location parameter from the fingerprint template of database comprises minutiae point coordinate, singular point coordinate, piece direction position etc., carries out translation, two fingerprint template complete matchings;

Delete from U that to comprise ranks coordinate difference right greater than the right coupling of the minutiae point of a designated value, the right similarity of these couplings adds up and obtains the final similarity of two fingerprint template details point sets;

Calculate the similarity of global characteristics:

The singular point similarity is to compare the position of singular point, direction and type in twos, the similarity addition that obtains;

Average ridge density similarity is differing from of two fingerprint template ridge density and gets inverse;

The similarity of block directed graph is the public part two fingerprint template effective coverages, the difference of calculated direction, and inverse is got in the back mean deviation that adds up;

The similarity of two last fingerprint templates is formed by top part and the fusion of global characteristics similarity;

When carrying out the identification of one-to-many, the average ridge density with fingerprint template in the database sorts earlier, and when fingerprint on site was discerned, the immediate fingerprint template of average ridge density in elder generation and the database mated, to accelerate recognition speed; Average ridge density is the average ridge density of whole fingerprint image.

3, according to the described fingerprint identification method of claim 2, when it is characterized in that fingerprint characteristic extracts: fingerprint image is expressed as a two-dimensional matrix, and each pixel is exactly an element of matrix, and value is 0～255, and the dimension of matrix is exactly the wide and high of image;

The minutiae point of fingerprint is meant end points or the bifurcation on the fingerprint ridge line, and fingerprint minutiae comprises following feature: coordinate xy-is illustrated in the position in the fingerprint image; Type t-represents the end points or the bifurcation of crestal line; Direction d-represents the direction of minutiae point, if the minutiae point of end points type, then this direction point to crestal line from the minutiae point position, if the branch type minutiae point, then this direction is pointed to the centre of two crestal lines behind the bifurcated from the minutiae point position; Ridge density g-is illustrated near the average density of the crestal line this minutiae point; Ridge curvature c-represents crestal line direction intensity of variation herein;

Divide block directed graph: be the mutually disjoint fritter that fingerprint image is divided into BLOCK_SIZE * BLOCK_SIZE size,, calculate the mean direction of crestal line, be thereby obtain size to the little image of each piece

(HEIGHT/BLOCK_SIZE) * (WIDTH/BLOCK_SIZE) branch block directed graph; The overall crestal line trend of dividing block directed graph portrayal fingerprint image; In addition, on minute block directed graph with one the background area after illegal direction value representation is cut apart fingerprint image;

Singular point: have some local crestal line directions discontinuous on the fingerprint image, these places are called the singular point of fingerprint, and its feature has: coordinate x y is illustrated in the position in the fingerprint image; Type t, singular point are divided into three kinds of core point, double-core point and trigpoints; Direction d, during away from singular point, the fingerprint ridge line direction changes minimum along this direction in expression.Ridge density c is illustrated near the average distance of the crestal line this singular point.

4, according to the described fingerprint identification method of claim 3, it is characterized in that image pre-service and normalization are at first image to be carried out even value filtering, make image more level and smooth, then, image is formatd; Calculating branch block directed graph extraction singular point is on block directed graph, calculates the Poincare Index of every bit earlier:

${\mathrm{pindex}}^n=\frac{1}{\mathrm{\&pi;}}\underset{i=0}{\overset{n}{\mathrm{\&Sigma;}}}\mathrm{\&tau;}(O_{(i+1)\mathrm{mod}n}-O_i)$

$\mathrm{\&tau;}\left(k\right)=\left\{\begin{array}{c}k,\mathrm{if}\left|k\right|<\frac{\mathrm{\&pi;}}{2},\\ \mathrm{\&pi;}+k,\mathrm{if}\left|k\right|<\frac{\mathrm{\&pi;}}{2}\\ \mathrm{\&pi;}-k,\mathrm{otherwise}\end{array}\right.,$

Wherein, n is the number of surrounding pixel point, O _iThe direction of representing i point; Getting radius earlier is 1, promptly Zhou Bian 8 points calculate Poincare Index, get p1, if its Poincare Index non-zero, again with radius 2, promptly Zhou Bian outer one deck calculates Poincare Index, it is identical with p2 to get p2:p1, illustrates that this point is a singular point, is core point type singular point if p1 is 1, if p1 is a trigpoint for-1, is the dikaryon singular point if p1 is 2; If p2 is different with p1, but p2＞0, p1＞0 then is the dikaryon singular point; Other situations then are not singular points.

5, according to claim 1,2,3 or 4 described fingerprint identification methods, it is characterized in that calculated direction figure, cut apart the background area and the refinement singular point is: the image after the normalization, calculate first crestal line direction, and calculate the consistance of crestal line direction simultaneously, and obtain directional diagram, redefine the singular point position, original position from these singular points, find the exact position of singular point,, calculate the singular point direction that makes new advances in new position.

6,, it is characterized in that the filtering of image and enhancing are according to claim 1,2,3 or 4 described fingerprint identification methods: after handling by anisotropic filter, the fingerprint image that is enhanced; The calculating ridge density is: first calculated fingerprint ridge density figure, carry out 33 * 33 mean filter again to ridge density figure.

7, according to claim 1,2,3 or 4 described fingerprint identification methods, it is characterized in that binary image and refinement are: the image after coming binaryzation to strengthen with the image behind 33 * 33 mean filters as adaptive threshold values; Then the image thinning of binaryzation is become the crestal line figure of single-point width; Image thinning is that each black picture element in the image has 8 consecutive point, judges according to them whether current point should be changed to white.Through repeatedly multiple scanning, made into white like this, just obtained the fingerprint ridge line chart of refinement up to the neither one black color dots.

8, according to claim 1,2,3 or 4 described fingerprint identification methods, it is characterized in that extracting minutiae point is: eliminate burr and noise earlier, promptly by scanning the crestal line figure of refinement, follow the tracks of crestal line, if, just it is erased from refinement figure from the pixel distance of crestal line origin-to-destination threshold values less than a setting; Then, extract minutiae point: promptly to any one black color dots on the image, if in its 8 adjacent points, optional starting point, run-down is got back to starting point in the direction of the clock, and its change in color illustrates that this point is a termination type minutiae point if 2 times; If more than 4 times, this point is the branch type minutiae point, and other situations then can be ignored, and by scanning effective fingerprint image zone, has obtained all minutiae point;

Follow the tracks of crestal line at the minutiae point place, obtain the direction of crestal line; The crestal line curvature of minutiae point is represented with the variation of direction, on the directional diagram of fingerprint image, calculates curvature with near the direction this point and the direction difference of this point.

9,, it is characterized in that minutiae point checking and deletion fake minutiae are according to claim 1,2,3 or 4 described fingerprint identification methods: any one minutiae point, if exist to come a minutiae point, then delete this minutiae point with it apart from less than a setting value D1; As an end points type minutiae point with come end points type minutiae point distance less than a setting value D2, and their directions are opposite, then delete this two minutiae point simultaneously; If an end points type minutiae point and a branch type minutiae point distance are less than a setting value D3, and their directions are opposite, then delete this two minutiae point simultaneously; If less than a setting value D4, and direction outwardly, then deletes this minutiae point from the inactive area of fingerprint image for minutiae point; Obtain final minutiae point by above-mentioned deletion.

10, a kind of fingerprint recognition system, it is characterized in that comprising fingerprint capturer, fingerprint recognition system, identification or and the control signal output mechanism; Comprising fingerprint image storer, fingerprint image processor and fingerprint characteristic data storer; Fingerprint image processor is to utilize to require the described method of one of 1-9 that fingerprint image is handled and discerned.

Images