CN101303728A - Method for identifying fingerprint facing image quality - Google Patents

Abstract

The invention discloses a fingerprint identifying method facing to the quality of an image, which solves the defect that the compatibility of the existing fingerprint identifying method to an ideal fingerprint image and a non ideal fingerprint image is not good. The method of the invention includes: (1) reading the collected fingerprint image g(x, y), wherein, g (x, y) is the gray value of the pixel points (x, y); (2) carrying out quality characteristic extracting on the fingerprint images and respectively extracting three characteristics of grads consistency QT, the frequency spectrum characteristic QF and the gray standard difference Qs; (3) adopting an SVM sorter to carry out learning and sorting on the qualities of the fingerprint images and confirming the qualities of the fingerprint images as two defined quality types; (4) respectively adopting a matching arithmetic based on detail points and a matching arithmetic based on textures to the two fingerprints with better and poorer qualities to accomplish the identification.

Description

Fingerprint identification method towards picture quality

Technical field

Invention relates to a kind of fingerprint identification method, relates in particular to a kind of fingerprint identification method towards picture quality.

Background technology

In automatic fingerprint identification technology, the quality of picture quality is a key factor that influences recognition performance at present.In existing fingerprint recognition disposal route, normally mate, but this method to be for second-rate image by the minutiae point of obtaining fingerprint image, performance descends obviously.And some other recognition methods, though as based on streakline and based on the recognition methods of texture second-rate image is had certain effect, for the quality better image, not only performance improves not quite, and the resource that takies and time complexity are relatively large.Therefore, press for a kind of fingerprint identification method of differentiate between images quality, both guaranteed the recognition accuracy that reaches certain, saving resource that again can big limit.

Summary of the invention

The objective of the invention is in order to solve a kind of fingerprint identification method that existing fingerprint identification method proposes the desirable and bad shortcoming of the imperfect fingerprint image property taken into account towards picture quality.It is a kind of disposal route of judging based on picture quality, fingerprint image is divided into better, second-rate two types of quality, and then adopts different recognizers to carry out fingerprint recognition.

For achieving the above object, the present invention adopts following technical scheme:

The fingerprint image g of the collection of (1) reading (x, y), wherein (x y) is pixel (x, gray-scale value y) to g;

(2) fingerprint image is carried out qualitative character and extract, extract gradient consistance Q respectively _T, spectrum signature Q _F, gray standard deviation Q _sTotally three features;

(3) employing SVM support vector machine classifier is learnt the quality of fingerprint image and is classified, and it is defined as defined two kinds of quality types;

(4) to better, the second-rate two kinds of fingerprints of quality, adopt respectively based on the matching algorithm of minutiae point with based on the matching algorithm of texture and finish identification.

In the described step (2), three features have reflected the quality of quality respectively from different aspects, specifically be calculated as follows,

$Q_T=\frac{1}{r}\underset{i=1}{\overset{r}{\mathrm{\Σ}}}{\tilde{k}}_i,$ R is the sum of foreground blocks,

Be the gradient consistance of a piece in the block image, its computing formula is: $\tilde{k}=\frac{{(j_{11}-j_{22})}^2+4j_{12}^2}{{(j_{11}+j_{22})}^2},$ Wherein, j ₁₁, j ₁₂, j ₂₁, j ₂₂Be the element among the gradient vector covariance matrix J.As if the image block size is b*b, then all b in the piece ²The gradient vector covariance matrix of individual point $J=\frac{1}{b^2}\underset{s\∈B}{\mathrm{\Σ}}{g}_s{g}_s^T\≡\left[\begin{array}{cc}{j}_{11}& j_{12}\\ j_{21}& j_{22}\end{array}\right],$ b ²Be the size of block image, s is the point in this piece, and B is the set of all pixels in this piece, g _sBe the gradient vector of a s, g _s ^TTransposition for gradient vector; The qualitative character of final entire image is the conforming average of gradient of all pieces.

Q _FComputing formula be: $Q_F=\frac{1}{9}\underset{r=r_0-4}{\overset{r_0+4}{\mathrm{\Σ}}}Q\left(r\right),$ Wherein, $Q\left(r\right)=\frac{1}{\#C_r}\underset{(u,v)\∈C_r}{\mathrm{\Σ}}\left|G_{(u,v)}\right|$ Be the energy intensity function,

Be ring (r ₀-4＜=r＜=r ₀+ 4) number of interior pixel point, and

$\left|G_{(u,v)}\right|=\underset{x=0}{\overset{N-1}{\mathrm{\&Sigma;}}}\underset{y=0}{\overset{N-1}{\mathrm{\&Sigma;}}}\sqrt{{(g_{(x,y)}\cos {(-2\mathrm{\&pi;}<(x,y)(u,v)>/N)}^2+(g_{(x,y)}\sin (-2\mathrm{\&pi;}<(x,y)(u,v)>/N))}^2}$

| G _{(u, v)}| reflected the frequency domain figure after the Fourier transform as mid point (u, the energy intensity of v) locating, | G _{(u, v)}| constituted the intensity spectrum of frequency domain.If g (x, y) expression size be in the digital picture of N * N coordinate be (x, y) gray values of pixel points, then g (x, discrete Fourier transform (DFT) y) (DFT) G (u v) is defined as,

$G_{(u,v)}=\underset{x=0}{\overset{N-1}{\mathrm{\&Sigma;}}}\underset{y=0}{\overset{N-1}{\mathrm{\&Sigma;}}}{g}_{(x,y)}{e}^{-2\mathrm{\&pi;j}<(x,y)(u,v)>/N}$

$=\underset{x=0}{\overset{N-1}{\mathrm{\&Sigma;}}}\underset{y=0}{\overset{N-1}{\mathrm{\&Sigma;}}}{g}_{(x,y)}(\cos (-2\mathrm{\&pi;}<(x,y)(u,v)>/N)+j\sin (-2\mathrm{\&pi;}<(x,y)(u,v)>/N))$

Q _FCalculate by energy, and represent the quality of quality with its size bright ring band in the spectral image.

Q _sComputing formula be: $Q_S=\frac{1}{N}\underset{k=1}{\overset{N}{\mathrm{\&Sigma;}}}{S}_k,$

Wherein, $S_k=\sqrt{\frac{1}{{\stackrel{\&OverBar;}{w}}^2}\underset{x=1}{\overset{\stackrel{\&OverBar;}{w}}{\mathrm{\&Sigma;}}}\underset{y=1}{\overset{\stackrel{\&OverBar;}{w}}{\mathrm{\&Sigma;}}}{(g(x,y)-g\left(k\right))}^2}$ Be the standard deviation of the k piece in the block image, g (x, y) be pixel (x, gray-scale value y), g (k) they are the k piece gray averages, w is the piece length of side of block image.Q _sBy being averaged, the standard deviation of all block images represents the quality of entire image.

In the described step (4), what adopted all is classical algorithm based on the matching algorithm of minutiae point with based on the matching algorithm of texture.

The block size of block image is 8 * 8 in the disposal route of the present invention.

Beneficial effect of the present invention: because many-sided features such as the above-mentioned comprehensive fingerprint image spatial domain of disposal route, frequency fields, can distinguish the fingerprint image quality well, thereby make to have better adaptability based on the matching algorithm of minutiae point with based on the matching algorithm of texture.

Description of drawings

Fig. 1 is a recognition methods process flow diagram of the present invention.

Embodiment

The present invention will be further described below in conjunction with accompanying drawing and embodiment.

Among Fig. 1, a kind of fingerprint identification method towards picture quality, its step is:

The fingerprint image g of the collection of (1) reading (x, y), wherein (x y) is pixel (x, gray-scale value y) to g;

(2) fingerprint image is carried out qualitative character and extract, extract gradient consistance Q respectively _T, spectrum signature Q _F, gray standard deviation Q _sTotally three features;

(3) three characteristic indexs that extract in the step (2) form the three-dimensional feature vector, as the input vector of support vector machine; Employing SVM support vector machine classifier is learnt the quality of fingerprint image and is classified, and it is defined as defined two kinds of quality types;

(4) to better, the second-rate two kinds of fingerprints of quality, adopt respectively based on the matching algorithm of minutiae point with based on the matching algorithm of texture and finish identification.

In the described step (2), three features have reflected the quality of quality respectively from different aspects, specifically be calculated as follows,

$Q_T=\frac{1}{r}\underset{i=1}{\overset{r}{\mathrm{\&Sigma;}}}{\tilde{k}}_i,$ R is the sum (r=40 here) of foreground blocks,

Be the gradient consistance of a piece in the block image, its computing formula is: $\tilde{k}=\frac{{(j_{11}-j_{22})}^2+4j_{12}^2}{{(j_{11}+j_{22})}^2},$ Wherein, j ₁₁, j ₁₂, j ₂₁, j ₂₂Be the element among the gradient vector covariance matrix J.The gradient vector covariance matrix of being had a few in the piece $J=\frac{1}{b^2}\underset{s\&Element;B}{\mathrm{\&Sigma;}}{g}_s{g}_s^T\&equiv;\left[\begin{array}{cc}{j}_{11}& j_{12}\\ j_{21}& j_{22}\end{array}\right],$ Here b ²=64, i.e. the size of block image, s is the point in this piece, B is the set of all pixels in this piece, g _sBe the gradient vector of a s, g _s ^TTransposition for gradient vector; The qualitative character of final entire image is the conforming average of gradient of all pieces.

Q _FComputing formula be: $Q_F=\frac{1}{9}\underset{r=r_0-4}{\overset{r_0+4}{\mathrm{\&Sigma;}}}Q\left(r\right),$ Wherein, $Q\left(r\right)=\frac{1}{\#C_r}\underset{(u,v)\&Element;C_r}{\mathrm{\&Sigma;}}\left|G_{(u,v)}\right|$ Be the energy intensity function, Be ring (r0-4＜=r＜=r0+4) number of interior pixel point), and

$\left|G_{(u,v)}\right|=\underset{x=0}{\overset{N-1}{\mathrm{\&Sigma;}}}\underset{y=0}{\overset{N-1}{\mathrm{\&Sigma;}}}\sqrt{{(g_{(x,y)}\cos {(-2\mathrm{\&pi;}<(x,y)(u,v)>/N)}^2+(g_{(x,y)}\sin (-2\mathrm{\&pi;}<(x,y)(u,v)>/N))}^2}$

| G _{(u, v)}| reflected the frequency domain figure after the Fourier transform as mid point (u, the energy intensity of v) locating, | G _{(u, v)}| constituted the intensity spectrum of frequency domain.If g (x, y) expression size be in the digital picture of N * N coordinate be (x, y) gray values of pixel points, then g (x, discrete Fourier transform (DFT) y) (DFT) G (u v) is defined as,

$G_{(u,v)}=\underset{x=0}{\overset{N-1}{\mathrm{\&Sigma;}}}\underset{y=0}{\overset{N-1}{\mathrm{\&Sigma;}}}{g}_{(x,y)}{e}^{-2\mathrm{\&pi;j}<(x,y)(u,v)>/N}$

$=\underset{x=0}{\overset{N-1}{\mathrm{\&Sigma;}}}\underset{y=0}{\overset{N-1}{\mathrm{\&Sigma;}}}{g}_{(x,y)}(\cos (-2\mathrm{\&pi;}<(x,y)(u,v)>/N)+j\sin (-2\mathrm{\&pi;}<(x,y)(u,v)>/N))$

Q _FCalculate by energy, and represent the quality of quality with its size bright ring band in the spectral image.

Q _sComputing formula be: $Q_S=\frac{1}{N}\underset{k=1}{\overset{N}{\mathrm{\&Sigma;}}}{S}_k,$

Wherein, $S_k=\sqrt{\frac{1}{{\stackrel{\&OverBar;}{w}}^2}\underset{x=1}{\overset{\stackrel{\&OverBar;}{w}}{\mathrm{\&Sigma;}}}\underset{y=1}{\overset{\stackrel{\&OverBar;}{w}}{\mathrm{\&Sigma;}}}{(g(x,y)-g\left(k\right))}^2}$ Be the standard deviation of the k piece in the block image, g (x, y) be pixel (x, gray-scale value y), g (k) they are the k piece gray averages, w is the piece length of side (w=8 here) of block image.Q _sBy being averaged, the standard deviation of all block images represents the quality of entire image.

In the described step (4), what adopted all is classical algorithm based on the matching algorithm of minutiae point with based on the matching algorithm of texture.

Claims (3)

1. the fingerprint identification method towards picture quality is characterized in that, its step is:

(1) read collection fingerprint image g (x, y), wherein (x y) is pixel (x, gray-scale value y) to g;

(2) fingerprint image is carried out qualitative character and extract, extract gradient consistance Q respectively _T, spectrum signature Q _F, gray standard deviation Q _sTotally three features;

(3) employing SVM support vector machine classifier is learnt the quality of fingerprint image and is classified, and it is defined as the better or second-rate two kinds of quality types of quality;

(4) to better, the second-rate two kinds of fingerprints of quality, adopt respectively based on the matching algorithm of minutiae point with based on the matching algorithm of texture and finish identification.

2. the fingerprint identification method towards picture quality as claimed in claim 1 is characterized in that, in the described step (2), three features have reflected the quality of quality respectively from different aspects, specifically be calculated as follows,

$Q_T=\frac{1}{r}\underset{i=1}{\overset{r}{\mathrm{\&Sigma;}}}{\tilde{k}}_i,$ R is the sum of foreground blocks,

Be the gradient consistance of a piece in the block image, its computing formula is: $\tilde{k}=\frac{{(j_{11}-j_{22})}^2+4j_{12}^2}{{(j_{11}+j_{22})}^2},$ Wherein, j ₁₁, j ₁₂, j ₂₁, j ₂₂Be the element among the gradient vector covariance matrix J;

As if the image block size is b*b, then all b in the piece ²The gradient vector covariance matrix of individual point

$J=\frac{1}{b^2}\underset{s\&Element;B}{\mathrm{\&Sigma;}}{g}_s{g}_s^T\&equiv;\left[\begin{array}{cc}{j}_{11}& j_{12}\\ j_{21}& j_{22}\end{array}\right],$ B wherein ²Be the size of block image, s is the point in this piece, and B is the set of all pixels in this piece, g _sBe the gradient vector of a s, g _s ^TTransposition for gradient vector; The qualitative character of final entire image is the conforming average of gradient of all pieces;

Q _FComputing formula be: $Q_F=\frac{1}{9}\underset{r=r_0-4}{\overset{r_0+4}{\mathrm{\&Sigma;}}}Q\left(r\right),$ Wherein, $Q\left(r\right)=\frac{1}{\#C_r}\underset{(u,v)\&Element;C_r}{\mathrm{\&Sigma;}}\left|G_{(u,v)}\right|$ Be the energy intensity function,

Be ring (r ₀-4＜=r＜=r ₀+ 4) number of interior pixel point, and

$\left|G_{(u,v)}\right|=\underset{x=0}{\overset{N-1}{\mathrm{\&Sigma;}}}\underset{y=0}{\overset{N-1}{\mathrm{\&Sigma;}}}\sqrt{{\left(g_{(x,y)}\cos (-2\mathrm{\&pi;}<(x,y)(u,v)>/N)\right)}^2+{\left(g_{(x,y)}\sin (-2\mathrm{\&pi;}<(x,y)(u,v)>/N)\right)}^2}$

| G _{(u, v)}| reflected the frequency domain figure after the Fourier transform as mid point (u, the energy intensity of v) locating, | G _{(u, v)}| constituted the intensity spectrum of frequency domain; If g (x, y) expression size be in the digital picture of N * N coordinate be (x, y) gray values of pixel points, then g (x, discrete Fourier transform (DFT) G y) (u v) is defined as,

$G_{(u,v)}=\underset{x=0}{\overset{N-1}{\mathrm{\&Sigma;}}}\underset{y=0}{\overset{N-1}{\mathrm{\&Sigma;}}}{g}_{(x,y)}{e}^{-2\mathrm{\&pi;j}<(x,y)(u,v)>/N}$

$=\underset{x=0}{\overset{N-1}{\mathrm{\&Sigma;}}}\underset{y=0}{\overset{N-1}{\mathrm{\&Sigma;}}}{g}_{(x,y)}(\cos (-2\mathrm{\&pi;}<(x,y)(u,v)>/N)+j\sin (-2\mathrm{\&pi;}<(x,y)(u,v)>/N))$

Q _FCalculate by energy, and represent the quality of quality with its size bright ring band in the spectral image;

Q _sComputing formula be: $Q_S=\frac{1}{N}\underset{k=1}{\overset{N}{\mathrm{\&Sigma;}}}{S}_k,$

Wherein, $S_k=\sqrt{\frac{1}{{\stackrel{\&OverBar;}{w}}^2}\underset{x=1}{\overset{\stackrel{\&OverBar;}{w}}{\mathrm{\&Sigma;}}}\underset{y=1}{\overset{\stackrel{\&OverBar;}{w}}{\mathrm{\&Sigma;}}}{(g(x,y)-g\left(k\right))}^2}$ Be the standard deviation of the k piece in the block image, g (x, y) be pixel (x, gray-scale value y), g (k) they are the k piece gray averages, w is the piece length of side of block image; Q _sBy being averaged, the standard deviation of all block images represents the quality of entire image.

3. the fingerprint identification method towards picture quality as claimed in claim 1 is characterized in that, in the described step (4), what adopted all is classical algorithm based on the matching algorithm of minutiae point with based on the matching algorithm of texture.

Images