CN104504370A - Finger vein recognition method combining bionic texture feature and linear texture feature - Google Patents

Abstract

The invention discloses a finger vein recognition method combining a bionic texture feature and a linear texture feature. By adopting the method, the bionic texture feature and the linear texture feature of finger veins can be utilized effectively, the adverse effect of noise on the recognition accuracy is avoided, and the recognition performance and robustness of a fingerprint vein recognition system are enhanced. The bionic texture feature and the linear texture feature of a finger vein image are extracted respectively. The bionic texture feature is extracted by using a Gabor filter with good similarity of a visual cortex simple cell receptive field model of a mammal. When the method is used for performing feature recognition, the effect is similar to object recognition through human eyes, and a good effect is achieved.

Description

A kind of finger vein identification method in conjunction with bionical textural characteristics and wire textural characteristics

Technical field

The present invention relates to finger vena identification field, specifically a kind of finger vein identification method in conjunction with bionical textural characteristics and wire textural characteristics.

Background technology

Finger vena identification is a kind of emerging biometrics identification technology with better development prospect, and the key of finger vena identification is how accurately to extract vein network, carries out feature extraction and matching on this basis.In order to overcome the impact of low-quality finger vein image on recognition result, the finger vein identification method in conjunction with bionical textural characteristics and wire textural characteristics is suggested.Its ultimate principle first does certain pre-service to the finger venous image gathered, and comprises image enhaucament, size normalization etc., then to the feature of pretreated image zooming-out in conjunction with bionical texture and wire texture, and merges coding and produce proper vector.Finally utilize the Hamming distances between proper vector to calculate the characteristic similarity of two width finger venous images, mate according to the feature power set and threshold value, whether both checkings are from same piece of finger.

Traditional finger vein identification method, when feature extraction, often only extracts single feature, and incomplete, the characteristic of finger venous image well can not be described out, and like this, incomplete feature can cause the reduction of discrimination usually.

Summary of the invention

The object of the invention is for solving above-mentioned the deficiencies in the prior art, and provide a kind of improve finger vein recognition system recognition performance and the finger vein identification method in conjunction with bionical texture and wire texture of robustness.

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

First pre-service is carried out to training image, then, extract bionical texture and the wire textural characteristics of image after pre-service respectively, and it is fused into the feature of complete finger venous image, be encoded into proper vector, finally utilize the hamming between proper vector apart from the characteristic similarity of calculating two width finger venous images, mate according to the feature power set and threshold value, whether both checkings are from same piece of finger.

Described bionical textural characteristics, the Gabor filter using mammalian visual cortex simple cell to accept field model good approximation is extracted, and the functional form of two-dimensional Gabor filter can be expressed as formula (1):

$G\left(\stackrel{\→}{x}\right)=\frac{{\left|\right|\stackrel{\→}{k}\left|\right|}^2}{{\mathrm{\σ}}^2}{e}^{-\frac{{\left|\right|\stackrel{\→}{k}\left|\right|}^2\left|\right|{\stackrel{\→}{x}\left|\right|}^2}{2{\mathrm{\σ}}^2}}(e^{i(\stackrel{\→}{k}\·\stackrel{\→}{x})}-e^{-\frac{{\mathrm{\σ}}^2}{2}})---\left(1\right)$

Wherein for the image coordinate of given position; $\stackrel{\→}{k}=\left(\begin{array}{c}{k}_1\\ k_2\end{array}\right)=\left(\begin{array}{c}{k}_v\cos \mathrm{\α}\\ k_v\sin \mathrm{\α}\end{array}\right)$ For the centre frequency of wave filter; α is the direction of wave filter texture feature extraction; σ ²for the variance of wave filter Gaussian envelope.In order to allow the bionical texture extracted not by the impact of gradation of image absolute figure, and insensitive to the illumination variation of image.Deduct at the real part of two-dimensional Gabor filter shown in (1).

According to finger venous image size and noise situations, select and fix suitable Gabor filter variance parameter σ ², regulate filter frequency parameter k _v, and direction parameter α, obtain one group of two-dimensional Gabor filter and with them to pretreated finger venous image carry out convolution, shown in (2):

$J_{i,j}\left(\stackrel{\→}{x}\right)=\∫I\left({\stackrel{\→}{x}}^{\′}\right)G_{k_i,{\mathrm{\α}}_j}(\stackrel{\→}{x}-{\stackrel{\→}{x}}^{\′})d^2{\stackrel{\→}{x}}^{\′}---\left(2\right)$

Each point in image the response of multiple Gabor filter can be obtained get and wherein respond the strongest frequency parameter k _vi, and direction parameter α _jas the bionical textural characteristics of this point, shown in (3):

$D_g\left({\stackrel{\→}{x}}_0\right)={\arg \max }_{i,j}\left(J_{i,j}\left({\stackrel{\→}{x}}_0\right)\right)---\left(3\right)$

Then whole finger venous image bionical textural characteristics be

Described wire textural characteristics, adopts wire texture filter to extract, and regulates wave filter live width parameter d, and angle parameter theta, obtains one group of wire texture filter and with to pretreated finger venous image carry out convolution, shown in (4):

$L_{i,j}=\∫I\left({\stackrel{\→}{x}}^{\′}\right)H_{d_i,{\mathrm{\θ}}_j}(\stackrel{\→}{x}-{\stackrel{\→}{x}}^{\′})d^2{\stackrel{\→}{x}}^{\′}---\left(4\right)$

Each point in image the response of multiple Gabor filter can be obtained get and wherein respond the strongest live width parameter d _i, and angle parameter theta _jas the wire textural characteristics of this point, shown in (5):

$D_n\left({\stackrel{\→}{x}}_0\right)={\arg \max }_{i,j}\left(L_{i,j}\left({\stackrel{\→}{x}}_0\right)\right)---\left(5\right)$

Then whole finger venous image bionical textural characteristics be

Described Fusion Features and coding, single bionical textural characteristics or wire textural characteristics complete all not, merge them and can obtain complete feature set

$D\left(\stackrel{\→}{x}\right)=[D_g\left(\stackrel{\→}{x}\right),D_n\left(\stackrel{\→}{x}\right)]---\left(6\right)$

Feature set comprise frequency parameter and the direction parameter of bionical texture, and the live width parameter of wire texture and angle parameter, be encoded into binary code $D{\left(\stackrel{\→}{x}\right)}_{\left(2\right)}=[D_g{\left(\stackrel{\→}{x}\right)}_{\left(2\right)},D_n{\left(\stackrel{\→}{x}\right)}_{\left(2\right)}].$

The coupling of described feature based power and Hamming distances, the characteristic similarity of two width finger venous images is obtained by the Hamming distances calculated between proper vector, but due to the difference of image collecting device, bionical textural characteristics is different to the significance level identified with wire textural characteristics under various circumstances, vein as image is clear, wire texture is obvious, wire textural characteristics in proper vector even more important, otherwise, then bionical textural characteristics important, so in order to improve discrimination, adopt and calculate Hamming distances respectively, then the image similarity Dist (a, b) of weighting mating, shown in (7):

$\mathrm{Dist}(a,b)={\mathrm{\ω}}_g\frac{{D_g}^{\left(a\right)}{\left(\stackrel{\→}{x}\right)}_{\left(2\right)}\⊕{D_g}^{\left(b\right)}{\left(\stackrel{\→}{x}\right)}_{\left(2\right)}}{\left|\right|{D_g}^{\left(a\right)}{\left(\stackrel{\→}{x}\right)}_{\left(2\right)}\left|\right|}+{\mathrm{\ω}}_n\frac{{D_n}^{\left(a\right)}{\left(\stackrel{\→}{x}\right)}_{\left(2\right)}\⊕{D_n}^{\left(b\right)}{\left(\stackrel{\→}{x}\right)}_{\left(2\right)}}{\left|\right|{D_n}^{\left(a\right)}{\left(\stackrel{\→}{x}\right)}_{\left(2\right)}\left|\right|}---\left(7\right)$

Wherein, ⊕ is xor operation, and Dist (a, b) is the similarity of finger venous image a and image b, for the bionical textural characteristics of finger venous image a, for the wire textural characteristics of finger venous image b.

The invention has the beneficial effects as follows: by extracting bionical textural characteristics and the wire textural characteristics of finger venous image: the Gabor filter that bionical textural characteristics uses mammalian visual cortex simple cell to accept field model good approximation is extracted, feature identification is carried out with it, the similar eye recognition object of effect, has good completeness; Meanwhile, in finger venous image, vein texture is generally the feature of wire texture, also proposed the wire textural characteristics based on Radon conversion, it is fused in the proper vector of finger venous image, further increases the completeness of proper vector; Meanwhile, in the characteristic matching stage, have employed the coupling of feature based power and Hamming distances, can under different running environment, adjustment feature power, reaches best matching effect.Thus, improve recognition performance and the robustness of finger vein recognition system.

Accompanying drawing explanation

Figure l is wire texture filter (width d be 2, angle θ be respectively 0 °, 30 °, 60 °, 90 °, 120 °, 150 °);

Fig. 2 is identifying process flow diagram of the present invention.

Embodiment

Below in conjunction with accompanying drawing and embodiment, the present invention will be further described.

The present invention is divided into two processes: registration process and identifying.First registration process carries out pre-service to image, then extracts bionical textural characteristics and wire textural characteristics separately, and merges coding generation proper vector, and be stored in template database, registration process is shown on the left of Fig. 2.In identifying, first pre-service is carried out to image, then carry out bionical textural characteristics and wire texture feature extraction, finally, the image similarity Dist (a, b) between database template is calculated, according to the threshold value determination recognition result of setting by formula (7).Identify that detailed process is shown on the right side of Fig. 2.

1. pre-service

Owing to there is some useless backgrounds and more noise in the image of original vein that collects so first pre-service will be carried out to original image.Pre-service of the present invention comprises region of interesting extraction, size normalization, image enhaucament.Region of interesting extraction, by being rich in the finger vena extracted region of a large amount of useful information out, detects the profile of finger vena by rim detection.Size normalization is rectangular area area-of-interest linear stretch being become L*K pixel size, M=of the present invention, 96, N=48.Image enhaucament adopts gaussian filtering method.

2. feature extraction

Existing finger vein identification method, when feature extraction, often only extracts single feature, and incomplete, the characteristic of finger venous image well can not be described out.For addressing this problem, a kind of finger vein identification method in conjunction with bionical texture and wire texture is provided, the method extracts bionical textural characteristics and the wire textural characteristics of finger venous image: the Gabor filter that bionical textural characteristics uses mammalian visual cortex simple cell to accept field model good approximation is extracted, feature identification is carried out with it, the similar eye recognition object of effect, has good completeness; Meanwhile, in finger venous image, vein texture is generally the feature of wire texture, also proposed the wire textural characteristics based on Radon conversion, it is fused in the proper vector of finger venous image, further increases the completeness of proper vector.

The Gabor filter using mammalian visual cortex simple cell to accept field model good approximation extracts bionical textural characteristics, and the functional form of two-dimensional Gabor filter can be expressed as formula (1):

$G\left(\stackrel{\→}{x}\right)=\frac{{\left|\right|\stackrel{\→}{k}\left|\right|}^2}{{\mathrm{\σ}}^2}{e}^{-\frac{{\left|\right|\stackrel{\→}{k}\left|\right|}^2\left|\right|{\stackrel{\→}{x}\left|\right|}^2}{2{\mathrm{\σ}}^2}}(e^{i(\stackrel{\→}{k}\·\stackrel{\→}{x})}-e^{-\frac{{\mathrm{\σ}}^2}{2}})---\left(1\right)$

Wherein for the image coordinate of given position; $\stackrel{\→}{k}=\left(\begin{array}{c}{k}_1\\ k_2\end{array}\right)=\left(\begin{array}{c}{k}_v\cos \mathrm{\α}\\ k_v\sin \mathrm{\α}\end{array}\right)$ For the centre frequency of wave filter; α is the direction of wave filter texture feature extraction; σ ²for the variance of wave filter Gaussian envelope.In order to allow the bionical texture extracted not by the impact of gradation of image absolute figure, and insensitive to the illumination variation of image.Deduct at the real part of two-dimensional Gabor filter shown in (1).

According to finger venous image size and noise situations, select and fix suitable Gabor filter variance parameter σ ², regulate filter frequency parameter k _v, and direction parameter α, obtain one group of two-dimensional Gabor filter and with them to pretreated finger venous image carry out convolution, shown in (2):

$J_{i,j}\left(\stackrel{\→}{x}\right)=\∫I\left({\stackrel{\→}{x}}^{\′}\right)G_{k_i,{\mathrm{\α}}_j}(\stackrel{\→}{x}-{\stackrel{\→}{x}}^{\′})d^2{\stackrel{\→}{x}}^{\′}---\left(2\right)$

Each point in image the response of multiple Gabor filter can be obtained get and wherein respond the strongest frequency parameter k _vi, and direction parameter α _jas the bionical textural characteristics of this point, shown in (3):

$D_g\left({\stackrel{\→}{x}}_0\right)={\arg \max }_{i,j}\left(J_{i,j}\left({\stackrel{\→}{x}}_0\right)\right)---\left(3\right)$

Then whole finger venous image bionical textural characteristics be frequency parameter k _viget 8 kinds of possibilities (1,1/2,1/4,1/6,1/8,1/10,1/12,1/14; Unit: 1/ pixel), direction parameter α _jgetting 6 kinds may (0 °, 60 °, 120 °, 180 °, 240 °, 300 °).

Adopt wire texture filter to extract wire textural characteristics, wire texture filter as shown in Figure 1.Regulate wave filter live width parameter d, and angle parameter theta, obtain one group of wire texture filter and with to pretreated finger venous image carry out convolution, shown in (4):

$L_{i,j}=\∫I\left({\stackrel{\→}{x}}^{\′}\right)H_{d_i,{\mathrm{\θ}}_j}(\stackrel{\→}{x}-{\stackrel{\→}{x}}^{\′})d^2{\stackrel{\→}{x}}^{\′}---\left(4\right)$

Each point in image the response of multiple Gabor filter can be obtained get and wherein respond the strongest live width parameter d _i, and angle parameter theta _jas the wire textural characteristics of this point, shown in (5):

$D_n\left({\stackrel{\→}{x}}_0\right)={\arg \max }_{i,j}\left(L_{i,j}\left({\stackrel{\→}{x}}_0\right)\right)---\left(5\right)$

Then whole finger venous image bionical textural characteristics be live width parameter d _iget 4 kinds of possibilities (1,2,4,8; Unit: pixel), and angle parameter theta _jgetting 6 kinds may (0 °, 30 °, 60 °, 90 °, 120 °, 150 °).

3. Fusion Features and coding

Single bionical textural characteristics or wire textural characteristics complete all not, merge them and can obtain complete feature set

$D\left(\stackrel{\→}{x}\right)=[D_g\left(\stackrel{\→}{x}\right),D_n\left(\stackrel{\→}{x}\right)]---\left(6\right)$

Feature set comprise frequency parameter and the direction parameter of bionical texture, and the live width parameter of wire texture and angle parameter, be encoded into binary code totally 96 × 48 × (3+3+2+3)=50688bit.

4. coupling and identification,

The characteristic similarity of two width finger venous images is obtained by the Hamming distances calculated between proper vector, but due to the difference of image collecting device, bionical textural characteristics is different to the significance level identified with wire textural characteristics under various circumstances, vein as image is clear, wire texture is obvious, wire textural characteristics in proper vector even more important, otherwise, then bionical textural characteristics important, so in order to improve discrimination, adopt and calculate Hamming distances respectively, then the image similarity Dist (a, b) of weighting mating, shown in (7):

$\mathrm{Dist}(a,b)={\mathrm{\ω}}_g\frac{{D_g}^{\left(a\right)}{\left(\stackrel{\→}{x}\right)}_{\left(2\right)}\⊕{D_g}^{\left(b\right)}{\left(\stackrel{\→}{x}\right)}_{\left(2\right)}}{\left|\right|{D_g}^{\left(a\right)}{\left(\stackrel{\→}{x}\right)}_{\left(2\right)}\left|\right|}+{\mathrm{\ω}}_n\frac{{D_n}^{\left(a\right)}{\left(\stackrel{\→}{x}\right)}_{\left(2\right)}\⊕{D_n}^{\left(b\right)}{\left(\stackrel{\→}{x}\right)}_{\left(2\right)}}{\left|\right|{D_n}^{\left(a\right)}{\left(\stackrel{\→}{x}\right)}_{\left(2\right)}\left|\right|}---\left(7\right)$

Wherein, ⊕ is xor operation, and Dist (a, b) is the similarity of finger venous image a and image b, for the bionical textural characteristics of finger venous image a, for the wire textural characteristics of finger venous image b.Normal conditions ω _g=ω _n=0.5, as finger vena clean mark, then suitably tune up ω _n; Otherwise then tune up ω _g.

Claims (4)

1., in conjunction with a finger vein identification method for bionical textural characteristics and wire textural characteristics, it is characterized in that, comprise the following steps:

S1. pre-service: remove useless background and noise; S2. feature extraction: extract the bionical textural characteristics of finger vena and wire textural characteristics; S3. Fusion Features and coding: merge bionical textural characteristics and wire textural characteristics and encode; S4. mate and identify: feature based power and Hamming distances carry out match cognization.

2. a kind of finger vein identification method in conjunction with bionical textural characteristics and wire textural characteristics according to claim 1, it is characterized in that, above-mentioned steps S2 comprises step by step following:

S21. bionical texture feature extraction

The Gabor filter using mammalian visual cortex simple cell to accept field model good approximation is extracted, and the functional form of two-dimensional Gabor filter can be expressed as formula (1):

$G\left(\stackrel{\→}{x}\right)=\frac{{\left|\right|\stackrel{\→}{k}\left|\right|}^2}{{\mathrm{\σ}}^2}{e}^{-\frac{{\left|\right|\stackrel{\→}{k}\left|\right|}^2{\left|\right|\stackrel{\→}{x}\left|\right|}^2}{2{\mathrm{\σ}}^2}}(e^{i(\stackrel{\→}{k}\·\stackrel{\→}{x})}-e^{-\frac{{\mathrm{\σ}}^2}{2}})---\left(1\right)$

Wherein for the image coordinate of given position; $\stackrel{\→}{k}=\left(\begin{array}{c}{k}_1\\ k_2\end{array}\right)=\left(\begin{array}{c}{k}_v\cos \mathrm{\α}\\ k_v\sin \mathrm{\α}\end{array}\right)$ For the centre frequency of wave filter; α is the direction of wave filter texture feature extraction; σ ²for the variance of wave filter Gaussian envelope; According to finger venous image size and noise situations, select and fix suitable Gabor filter variance parameter σ ², regulate filter frequency parameter k _v, and direction parameter α, obtain one group of two-dimensional Gabor filter and with them to pretreated finger venous image carry out convolution, shown in (2):

$J_{i,j}\left(\stackrel{\→}{x}\right)=\∫I\left({\stackrel{\→}{x}}^{\′}\right)G_{k_i,{\mathrm{\α}}_j}(\stackrel{\→}{x}-{\stackrel{\→}{x}}^{\′})d^2{\stackrel{\→}{x}}^{\′}---\left(2\right)$

Each point in image the response of multiple Gabor filter can be obtained get and wherein respond the strongest frequency parameter k _vi, and direction parameter α _jas the bionical textural characteristics of this point, shown in (3):

$D_g\left({\stackrel{\→}{x}}_0\right)=\arg {\max }_{i,j}\left(J_{i,j}\left({\stackrel{\→}{x}}_0\right)\right)---\left(3\right)$

Then whole finger venous image bionical textural characteristics be frequency parameter k _viget 8 kinds of possibilities, direction parameter α _jgetting 6 kinds may;

S22. wire texture feature extraction:

Adopt wire texture filter to extract wire textural characteristics, and regulate wave filter live width parameter d, and angle parameter theta, obtain one group of wire texture filter and with to pretreated finger venous image carry out convolution, shown in (4):

$L_{i,j}\left(\stackrel{\→}{x}\right)=\∫I\left({\stackrel{\→}{x}}^{\′}\right)H_{d_i,{\mathrm{\θ}}_j}(\stackrel{\→}{x}-{\stackrel{\→}{x}}^{\′})d^2{\stackrel{\→}{x}}^{\′}---\left(4\right)$

Each point in image the response of multiple Gabor filter can be obtained get and wherein respond the strongest live width parameter d _i, and angle parameter theta _jas the wire textural characteristics of this point, shown in (5):

$D_n\left({\stackrel{\→}{x}}_0\right)=\arg {\max }_{i,j}\left(L_{i,j}\left({\stackrel{\→}{x}}_0\right)\right)---\left(5\right)$

Then whole finger venous image bionical textural characteristics be live width parameter d _iget 4 kinds of possibilities, and angle parameter theta _jgetting 6 kinds may.

3. a kind of finger vein identification method in conjunction with bionical textural characteristics and wire textural characteristics according to claim 2, it is characterized in that, above-mentioned steps S3 is specific as follows:

Merge bionical textural characteristics and wire textural characteristics, obtain complete feature set

$D\left(\stackrel{\→}{x}\right)=[D_g\left(\stackrel{\→}{x}\right),D_n\left(\stackrel{\→}{x}\right)]---\left(6\right)$

Feature set Dx comprises frequency parameter and the direction parameter of bionical texture, and the live width parameter of wire texture and angle parameter, is encoded into binary code totally 96 × 48 × (3+3+2+3)=50688bit.

4. a kind of finger vein identification method in conjunction with bionical textural characteristics and wire textural characteristics according to claim 3, it is characterized in that, above-mentioned steps S4 is specific as follows:

Obtained the characteristic similarity of two width finger venous images by the Hamming distances calculated between proper vector, adopt and calculate Hamming distances respectively, then the image similarity Dist (a, b) of weighting mate, shown in (7):

$\mathrm{Dist}(a,b)={\mathrm{\ω}}_g\frac{{D_g}^{\left(a\right)}{\left(\stackrel{\→}{x}\right)}_{\left(2\right)}\⊕{D_g}^{\left(b\right)}{\left(\stackrel{\→}{x}\right)}_{\left(2\right)}}{\left|\right|{D_g}^{\left(a\right)}{\left(\stackrel{\→}{x}\right)}_{\left(2\right)}\left|\right|}+{\mathrm{\ω}}_n\frac{{D_n}^{\left(a\right)}{\left(\stackrel{\→}{x}\right)}_{\left(2\right)}\⊕{D_n}^{\left(b\right)}{\left(\stackrel{\→}{x}\right)}_{\left(2\right)}}{\left|\right|{D_n}^{\left(a\right)}{\left(\stackrel{\→}{x}\right)}_{\left(2\right)}\left|\right|}---\left(7\right)$

Wherein, ⊕ is xor operation, and Dist (a, b) is the similarity of finger venous image a and image b, for the bionical textural characteristics of finger venous image a, for the wire textural characteristics of finger venous image b.Normal conditions ω _g=ω _n=0.5, as finger vena clean mark, then suitably tune up ω _n; Otherwise then tune up ω _g.