CN101303728A

CN101303728A - Image Quality Oriented Fingerprint Recognition Method

Info

Publication number: CN101303728A
Application number: CNA2008101381170A
Authority: CN
Inventors: 尹义龙; 杨公平; 骆功庆; 张宇; 詹小四; 任春晓
Original assignee: Shandong University
Current assignee: Shandong University
Priority date: 2008-07-01
Filing date: 2008-07-01
Publication date: 2008-11-12
Anticipated expiration: 2028-07-01
Also published as: CN100592323C

Abstract

The invention discloses an image quality-oriented fingerprint identification method. It solves the shortcomings of the existing fingerprint identification methods that are not good at both ideal and non-ideal fingerprint images. The method is: (1) read the collected fingerprint image g(x, y), where g(x, y) is the gray value of the pixel point (x, y); (2) Extract the quality features of the fingerprint image, and extract three features: gradient consistency Q _T , spectral feature Q _F , and gray standard deviation Q _s ; (3 ) use SVM support vector machine classifier to learn and classify the quality of fingerprint images, and determine them as two defined quality types; The matching algorithm and texture-based matching algorithm complete the recognition.

Description

Image Quality Oriented Fingerprint Recognition Method

技术领域 technical field

发明涉及一种指纹识别方法，尤其涉及一种面向图像质量的指纹识别方法。The invention relates to a fingerprint identification method, in particular to an image quality-oriented fingerprint identification method.

背景技术 Background technique

目前在自动指纹识别技术中，图像质量的好坏是影响识别性能的一个重要因素。在现有的指纹识别处理方法中，通常是通过获取指纹图像的细节点来进行匹配，但这种方法对于质量较差的图像，性能下降明显。而其它一些识别方法，如基于纹线的和基于纹理的识别方法虽然对质量较差的图像有一定的效果，但对于质量较好的图像，不仅性能提高不大，而且占用的资源和时间复杂度相对较大。因此，迫切需要一种区分图像质量的指纹识别方法，既保证达到一定的识别准确率，又能够较大限度的节省资源。At present, in the automatic fingerprint identification technology, the image quality is an important factor affecting the identification performance. In the existing fingerprint identification processing methods, the minutiae points of the fingerprint image are usually obtained for matching, but the performance of this method drops significantly for images with poor quality. Other recognition methods, such as line-based and texture-based recognition methods, have certain effects on poor-quality images, but for better-quality images, not only the performance is not improved much, but the resources and time taken up are complicated. degree is relatively large. Therefore, there is an urgent need for a fingerprint recognition method that distinguishes image quality, which not only ensures a certain recognition accuracy rate, but also saves resources to the greatest extent.

发明内容 Contents of the invention

本发明的目的是为了解决现有指纹识别方法对理想和非理想指纹图像兼顾性不好的缺点而提出的一种面向图像质量的指纹识别方法。它是一种基于图像质量判断的处理方法，将指纹图像分为质量较好、质量较差两种类型，进而采用不同的识别算法进行指纹识别。The object of the present invention is to propose an image quality-oriented fingerprint identification method in order to solve the disadvantage that the existing fingerprint identification method has poor compatibility with ideal and non-ideal fingerprint images. It is a processing method based on image quality judgment, which divides fingerprint images into two types with good quality and poor quality, and then uses different recognition algorithms for fingerprint recognition.

为实现上述目的，本发明采用如下技术方案：To achieve the above object, the present invention adopts the following technical solutions:

(1)读取的采集的指纹图像g(x，y)，其中g(x，y)为像素点(x，y)的灰度值；(1) The fingerprint image g(x, y) of the collection of reading, wherein g(x, y) is the gray value of pixel point (x, y);

(2)对指纹图像进行质量特征提取，分别提取梯度一致性Q_T、频谱特征Q_F、灰度标准差Q_s共三个特征；(2) Perform quality feature extraction on the fingerprint image, and extract three features: gradient consistency Q _T , spectral feature Q _F , and gray standard deviation Q _s ;

(3)采用SVM支持向量机分类器对指纹图像的质量进行学习和分类，将其确定为已定义的两种质量类型；(3) adopt SVM support vector machine classifier to learn and classify the quality of the fingerprint image, and determine it as two defined quality types;

(4)对质量较好、质量较差两种指纹，分别采用基于细节点的匹配算法和基于纹理的匹配算法完成识别。(4) For two kinds of fingerprints with good quality and poor quality, the matching algorithm based on minutiae and the matching algorithm based on texture are used to complete the identification respectively.

所述步骤(2)中，三个特征分别从不同的方面反映了质量的好坏，具体计算如下，In the step (2), the three characteristics reflect the quality of the quality from different aspects, and the specific calculation is as follows,

$Q_{T} = \frac{1}{r} Σ_{i = 1}^{r} {\tilde{k}}_{i},$ r是前景块的总数，

为分块图像中一个块的梯度一致性，其计算公式为：

\tilde{k} = \frac{{(j_{11} - j_{22})}^{2} + 4 j_{12}^{2}}{{(j_{11} + j_{22})}^{2}},

其中，j₁₁，j₁₂，j₂₁，j₂₂为梯度向量协方差矩阵J中的元素。若图像分块大小为b*b，则块中所有b²个点的梯度向量协方差矩阵

J = \frac{1}{b^{2}} \underset{s &Element; B}{Σ} g_{s} g_{s}^{T} &equiv; [\begin{matrix} j_{11} & j_{12} \\ j_{21} & j_{22} \end{matrix}],

b²为分块图像的大小，s为该块中的点，B为该块中所有像素点的集合，g_s为点s的梯度向量，g_s ^T为梯度向量的转置；最终整幅图像的质量特征为所有块的梯度一致性的均值。

Q_{T} = \frac{1}{r} Σ_{i = 1}^{r} {\tilde{k}}_{i},

r is the total number of foreground blocks,

is the gradient consistency of a block in the block image, and its calculation formula is:

\tilde{k} = \frac{{(j_{11} - j_{twenty two})}^{2} + 4 j_{12}^{2}}{{(j_{11} + j_{twenty two})}^{2}},

Wherein, j ₁₁ , j ₁₂ , j ₂₁ , j ₂₂ are elements in the gradient vector covariance matrix J. If the image block size is b*b, the gradient vector covariance matrix of all b ² points in the block

J = \frac{1}{b^{2}} \underset{the s &Element; B}{Σ} g_{the s} g_{the s}^{T} &equiv; [\begin{matrix} j_{11} & j_{12} \\ j_{twenty one} & j_{twenty two} \end{matrix}],

b ² is the size of the block image, s is the point in the block, B is the set of all pixels in the block, g _s is the gradient vector of point s, g _s ^T is the transposition of the gradient vector; the final whole image The quality feature of an image is the mean of the gradient consistency of all blocks.

Q_F的计算公式为： $Q_{F} = \frac{1}{9} Σ_{r = r_{0} - 4}^{r_{0} + 4} Q (r),$ 其中， $Q (r) = \frac{1}{# C_{r}} \underset{(u, v) &Element; C_{r}}{Σ} | G_{(u, v)} |$ 为能量强度函数，

为环(r₀-4＜＝r＜＝r₀+4)内像素点的数目，而The formula for calculating _QF is:

Q_{f} = \frac{1}{9} Σ_{r = r_{0} - 4}^{r_{0} + 4} Q (r),

in,

Q (r) = \frac{1}{# C_{r}} \underset{(u, v) &Element; C_{r}}{Σ} | G_{(u, v)} |

is the energy intensity function,

is the number of pixels in the ring (r ₀ -4<=r<=r ₀ +4), and

$| | {G G}_{((u u,, v v))} | | = = {Σ Σ}_{x x = = 00}^{N N - - 11} {Σ Σ}_{y the y = = 00}^{N N - - 11} \sqrt{{(({g g}_{((x x,, y the y))} cos cos {((- - 22 π π < < ((x x,, y the y)) ((u u,, v v)) > > / / N N))}^{22} + + (({g g}_{((x x,, y the y))} sin sin ((- - 22 π π < < ((x x,, y the y)) ((u u,, v v)) > > / / N N))))}^{22}}$

|G_(u，v)|反映了傅里叶变换后的频域图像中点(u，v)处的能量强度，|G_(u，v)|构成了频域的强度谱。设g(x，y)表示大小为N×N的数字图像中坐标为(x，y)像素点的灰度值，则g(x，y)的离散傅立叶变换(DFT)G(u，v)定义为，|G _{(u, v)} | reflects the energy intensity at the point (u, v) in the frequency domain image after Fourier transform, and |G _{(u, v)} | constitutes the intensity spectrum in the frequency domain. Let g(x, y) represent the gray value of a pixel whose coordinates are (x, y) in a digital image of size N×N, then the discrete Fourier transform (DFT) G(u, v) of g(x, y) )defined as,

${G G}_{((u u,, v v))} = = {Σ Σ}_{x x = = 00}^{N N - - 11} {Σ Σ}_{y the y = = 00}^{N N - - 11} {g g}_{((x x,, y the y))} {e e}^{- - 22 πj πj < < ((x x,, y the y)) ((u u,, v v)) > > / / N N}$

$= = {Σ Σ}_{x x = = 00}^{N N - - 11} {Σ Σ}_{y the y = = 00}^{N N - - 11} {g g}_{((x x,, y the y))} ((cos cos ((- - 22 π π < < ((x x,, y the y)) ((u u,, v v)) > > / / N N)) + + j j sin sin ((- - 22 π π < < ((x x,, y the y)) ((u u,, v v)) > > / / N N))))$

Q_F通过对频谱图像中亮环带的能量进行计算，并用其大小来表示质量的好坏。 _QF calculates the energy of the bright ring in the spectrum image, and uses its size to represent the quality.

Q_s的计算公式为： $Q_{S} = \frac{1}{N} Σ_{k = 1}^{N} S_{k},$ The calculation formula of Q _s is: $Q_{S} = \frac{1}{N} Σ_{k = 1}^{N} S_{k},$

其中， $S_{k} = \sqrt{\frac{1}{{\overset{&OverBar;}{w}}^{2}} Σ_{x = 1}^{\overset{&OverBar;}{w}} Σ_{y = 1}^{\overset{&OverBar;}{w}} {(g (x, y) - g (k))}^{2}}$ 为分块图像中第个k块的标准差，g(x，y)为像素点(x，y)的灰度值，g(k)是第个k块灰度均值，w为分块图像的块边长。Q_s通过对所有的分块图像的标准差求均值来表示整幅图像的质量。in, $S_{k} = \sqrt{\frac{1}{{\overset{&OverBar;}{w}}^{2}} Σ_{x = 1}^{\overset{&OverBar;}{w}} Σ_{the y = 1}^{\overset{&OverBar;}{w}} {(g (x, the y) - g (k))}^{2}}$ is the standard deviation of the kth block in the block image, g(x, y) is the gray value of the pixel point (x, y), g(k) is the gray value of the kth block, and w is the block image The side length of the block. Q _s represents the quality of the entire image by averaging the standard deviations of all block images.

所述步骤(4)中，所采用的基于细节点的匹配算法和基于纹理的匹配算法都为经典的算法。In the step (4), the minutiae-based matching algorithm and the texture-based matching algorithm adopted are both classical algorithms.

本发明处理方法中分块图像的块大小为8×8。The block size of the block image in the processing method of the present invention is 8×8.

本发明的有益效果：由于上述的处理方法综合指纹图像空间域、频率域等多方面的特征，可以很好地区分指纹图像质量，从而使得基于细节点的匹配算法和基于纹理的匹配算法具有更好的适应性。Beneficial effects of the present invention: because the above-mentioned processing method integrates the characteristics of fingerprint image space domain, frequency domain and other aspects, it can distinguish the quality of fingerprint image well, so that the matching algorithm based on minutiae and the matching algorithm based on texture have more advantages. good adaptability.

附图说明 Description of drawings

图1为本发明的识别方法流程图。Fig. 1 is a flowchart of the identification method of the present invention.

具体实施方式 Detailed ways

下面结合附图与实施例对本发明做进一步说明。The present invention will be further described below in conjunction with the accompanying drawings and embodiments.

图1中，一种面向图像质量的指纹识别方法，它的步骤为：In Figure 1, an image quality-oriented fingerprint recognition method, its steps are:

(3)步骤(2)中提取出的三个特征指标形成三维特征向量，作为支持向量机的输入向量；采用SVM支持向量机分类器对指纹图像的质量进行学习和分类，将其确定为已定义的两种质量类型；(3) The three feature indicators extracted in step (2) form a three-dimensional feature vector, which is used as the input vector of the support vector machine; the SVM support vector machine classifier is used to learn and classify the quality of the fingerprint image, and it is determined as the Two quality types are defined;

$Q_{T} = \frac{1}{r} Σ_{i = 1}^{r} {\tilde{k}}_{i},$ r是前景块的总数(这里r＝40)，

为分块图像中一个块的梯度一致性，其计算公式为：

\tilde{k} = \frac{{(j_{11} - j_{22})}^{2} + 4 j_{12}^{2}}{{(j_{11} + j_{22})}^{2}},

其中，j₁₁，j₁₂，j₂₁，j₂₂为梯度向量协方差矩阵J中的元素。块中所有点的梯度向量协方差矩阵

J = \frac{1}{b^{2}} \underset{s &Element; B}{Σ} g_{s} g_{s}^{T} &equiv; [\begin{matrix} j_{11} & j_{12} \\ j_{21} & j_{22} \end{matrix}],

这里b²＝64，即分块图像的大小，s为该块中的点，B为该块中所有像素点的集合，g_s为点s的梯度向量，g_s ^T为梯度向量的转置；最终整幅图像的质量特征为所有块的梯度一致性的均值。

Q_{T} = \frac{1}{r} Σ_{i = 1}^{r} {\tilde{k}}_{i},

r is the total number of foreground blocks (here r=40),

\tilde{k} = \frac{{(j_{11} - j_{twenty two})}^{2} + 4 j_{12}^{2}}{{(j_{11} + j_{twenty two})}^{2}},

Wherein, j ₁₁ , j ₁₂ , j ₂₁ , j ₂₂ are elements in the gradient vector covariance matrix J. Gradient vector covariance matrix for all points in the block

J = \frac{1}{b^{2}} \underset{the s &Element; B}{Σ} g_{the s} g_{the s}^{T} &equiv; [\begin{matrix} j_{11} & j_{12} \\ j_{twenty one} & j_{twenty two} \end{matrix}],

Here b ² =64, which is the size of the block image, s is the point in the block, B is the set of all pixels in the block, g _s is the gradient vector of point s, g _s ^T is the transposition of the gradient vector ; the quality feature of the final whole image is the mean of the gradient consistency of all blocks.

Q_F的计算公式为： $Q_{F} = \frac{1}{9} Σ_{r = r_{0} - 4}^{r_{0} + 4} Q (r),$ 其中， $Q (r) = \frac{1}{# C_{r}} \underset{(u, v) &Element; C_{r}}{Σ} | G_{(u, v)} |$ 为能量强度函数，为环(r0-4＜＝r＜＝r0+4)内像素点的数目)，而The formula for calculating _QF is: $Q_{f} = \frac{1}{9} Σ_{r = r_{0} - 4}^{r_{0} + 4} Q (r),$ in, $Q (r) = \frac{1}{# C_{r}} \underset{(u, v) &Element; C_{r}}{Σ} | G_{(u, v)} |$ is the energy intensity function, is the number of pixels in the ring (r0-4<=r<=r0+4), and

其中， $S_{k} = \sqrt{\frac{1}{{\overset{&OverBar;}{w}}^{2}} Σ_{x = 1}^{\overset{&OverBar;}{w}} Σ_{y = 1}^{\overset{&OverBar;}{w}} {(g (x, y) - g (k))}^{2}}$ 为分块图像中第个k块的标准差，g(x，y)为像素点(x，y)的灰度值，g(k)是第个k块灰度均值，w为分块图像的块边长(这里w＝8)。Q_s通过对所有的分块图像的标准差求均值来表示整幅图像的质量。in, $S_{k} = \sqrt{\frac{1}{{\overset{&OverBar;}{w}}^{2}} Σ_{x = 1}^{\overset{&OverBar;}{w}} Σ_{the y = 1}^{\overset{&OverBar;}{w}} {(g (x, the y) - g (k))}^{2}}$ is the standard deviation of the kth block in the block image, g(x, y) is the gray value of the pixel point (x, y), g(k) is the gray value of the kth block, and w is the block image The side length of the block (here w=8). Q _s represents the quality of the entire image by averaging the standard deviations of all block images.

Claims

1. a kind of fingerprint recognition method facing image quality, it is characterized in that, its step is:

(1) Read the collected fingerprint image g(x, y), where g(x, y) is the gray value of the pixel point (x, y);

(2) Perform quality feature extraction on the fingerprint image, and extract three features: gradient consistency Q _T , spectral feature Q _F , and gray standard deviation Q _s ;

(3) Adopt SVM support vector machine classifier to learn and classify the quality of fingerprint image, and determine it as two quality types with good quality or poor quality;

(4) For two kinds of fingerprints with good quality and poor quality, the matching algorithm based on minutiae and the matching algorithm based on texture are used to complete the identification respectively.

2. the image quality-oriented fingerprint identification method as claimed in claim 1, is characterized in that, in described step (2), three characteristics reflect the quality of quality from different aspects respectively, concrete calculation is as follows,

Q_{T} = \frac{1}{r} Σ_{i = 1}^{r} {\tilde{k}}_{i},

r is the total number of foreground blocks,

\tilde{k} = \frac{{(j_{11} - j_{twenty two})}^{2} + 4 j_{12}^{2}}{{(j_{11} + j_{twenty two})}^{2}},

Among them, j ₁₁ , j ₁₂ , j ₂₁ , j ₂₂ are elements in the gradient vector covariance matrix J;

If the image block size is b*b, the gradient vector covariance matrix of all b ² points in the block

J = \frac{1}{b^{2}} \underset{the s &Element; B}{Σ} g_{the s} g_{the s}^{T} &equiv; [\begin{matrix} j_{11} & j_{12} \\ j_{twenty one} & j_{twenty two} \end{matrix}],

Where b ² is the size of the block image, s is the point in the block, B is the set of all pixels in the block, g _s is the gradient vector of point s, g _s ^T is the transposition of the gradient vector; the final integer The quality feature of an image is the mean of the gradient consistency of all blocks;

The formula for calculating _QF is:

Q_{f} = \frac{1}{9} Σ_{r = r_{0} - 4}^{r_{0} + 4} Q (r),

in,

Q (r) = \frac{1}{# C_{r}} \underset{(u, v) &Element; C_{r}}{Σ} | G_{(u, v)} |

is the energy intensity function,

is the number of pixels in the ring (r ₀ -4<=r<=r ₀ +4), and

| | {G G}_{((u u,, v v))} | | = = {Σ Σ}_{x x = = 00}^{N N - - 11} {Σ Σ}_{y the y = = 00}^{N N - - 11} \sqrt{{(({g g}_{((x x,, y the y))} cos cos ((- - 22 π π < < ((x x,, y the y)) ((u u,, v v)) > > / / N N))))}^{22} + + {(({g g}_{((x x,, y the y))} sin sin ((- - 22 π π < < ((x x,, y the y)) ((u u,, v v)) > > / / N N))))}^{22}}

|G _{(u, v)} | reflects the energy intensity at the point (u, v) in the frequency domain image after Fourier transform, and |G _{(u, v)} | constitutes the intensity spectrum in the frequency domain; let g( x, y) represents the gray value of a pixel whose coordinates are (x, y) in a digital image whose size is N×N, then the discrete Fourier transform G(u, v) of g(x, y) is defined as,

{G G}_{((u u,, v v))} = = {Σ Σ}_{x x = = 00}^{N N - - 11} {Σ Σ}_{y the y = = 00}^{N N - - 11} {g g}_{((x x,, y the y))} {e e}^{- - 22 πj πj < < ((x x,, y the y)) ((u u,, v v)) > > / / N N}

= = {Σ Σ}_{x x = = 00}^{N N - - 11} {Σ Σ}_{y the y = = 00}^{N N - - 11} {g g}_{((x x,, y the y))} ((cos cos ((- - 22 π π < < ((x x,, y the y)) ((u u,, v v)) > > / / N N)) + + j j sin sin ((- - 22 π π < < ((x x,, y the y)) ((u u,, v v)) > > / / N N))))

_QF calculates the energy of the bright ring in the spectrum image, and uses its size to represent the quality;

The calculation formula of Q _s is:

Q_{S} = \frac{1}{N} Σ_{k = 1}^{N} S_{k},

in,

S_{k} = \sqrt{\frac{1}{{\overset{&OverBar;}{w}}^{2}} Σ_{x = 1}^{\overset{&OverBar;}{w}} Σ_{the y = 1}^{\overset{&OverBar;}{w}} {(g (x, the y) - g (k))}^{2}}

is the standard deviation of the kth block in the block image, g(x, y) is the gray value of the pixel point (x, y), g(k) is the gray value of the kth block, and w is the block image The block side length; Q _s represents the quality of the entire image by averaging the standard deviations of all block images.

3. the image quality-oriented fingerprint recognition method as claimed in claim 1, is characterized in that, in described step (4), the matching algorithm based on minutiae and the matching algorithm based on texture are all classical algorithms.