CN100514352C - Vena characteristic extracting method of finger vena identification system - Google Patents

Abstract

The invention relates to an image vein feature extraction method for finger vein verification system, including the following steps: firstly, a vale-shaped area of a finger vein image is picked up; as for each pixel of a finger vein image, the trend of veins is divided into eight directions within a 9x9 window taking the pixel as the center; then, according to the eight directions, eight vale-shaped detection operators are designed; the points within the 9x9 neighborhood of each pixel point are multiplied and then progression and summation of the product are completed; finally, the maximal progression sum is adopted as the grey value of the point pixel to obtain an image; however, false vein feature and noise exist in the image feature; therefore, threshold subdivision of the matrix is carried out to obtain extraction vein feature finally. The invention realizes accurate extraction of interesting vein feature in image and reduces extraction error feature and missing true feature; meanwhile, the invention can increase the recognition precision of the whole system and image processing rate.

Description

The vena characteristic extracting method of finger vena identification system

Technical field

The biological characteristic that the invention belongs in the human-body biological field of authentication extracts part, is specifically related to the extraction of finger venous image vein pattern.

Background technology

In modern society, along with the high speed development of computing machine and network technology, information security demonstrates unprecedented importance.Identity authentication is the prerequisite that guarantees security of system, in applications such as finance, national security, the administration of justice, ecommerce, E-Government, all needs identity authentication accurately.Whether, be national resident, protect etc. for safety is provided with password or key separately at company computer's network with on the Internet if whether having the right to enter safety zone (security system) as someone, whether having the right to carry out particular transaction.Current, be used for personal identification and differentiate the main means such as ID card (as I.D., employee's card, smart card, computing machine sign card and deposit card etc.) and password that rely on, carry inconvenience, lose easily yet these means exist, perhaps since use too much or improperly damage, problems such as not readable and password easily is cracked.Therefore, present widely used dependence certificate, Personal Identification Number (Personal identification number, PIN), classic method such as password confirms that the technology of personal identification is faced with stern challenge, and the more and more needs of incompatibility modern development in science and technology and social progress that seem.

People wish to have a kind of convenient reliable way to carry out identity authentication.Biometrics identification technology brings possibility to everything.People may forget or lose their card or forget Password, but people but can not forget or lose oneself biological characteristic such as people's face, fingerprint, iris, palmmprint, vein etc.In numerous identifying schemes, fingerprint recognition is present most convenient, reliable, non-infringement and technical relative ripe biological identification technology.Yet fingerprint recognition also has the deficiency of himself: some people or colony's fingerprint characteristic seldom do not reach the requirement of filing; Part user can't use finger print identifying owing to finger wear; Can stay user's finger mark when using fingerprint on fingerprint capturer, there is the possibility that is replicated utilization in these fingerprint traces at every turn.Authentication based on finger vena has its special advantages, has solved traditional many difficult problems that biometrics identification technology faced effectively, mainly shows: speed height, precision height, safe class height, vivo identification, internal feature, contactless.

As seen, finger vena is comparatively desirable a kind of biological characteristic that is used for authentication in the numerous biological characteristic.Simultaneously, be to have high precision, high-speed most advanced in the world authentication techniques based on the individual identity identification system of finger vein identification technology.In various biological identification technology, because of it is to utilize the outside biology interior feature that can't see to carry out authentication technology, so receive much concern as second generation biological identification technology with high antifalsification.

Finger vena base of recognition and key are the finger venous image feature extractions, and can vein pattern accurately extract the accuracy of identification that is directly connected to total system.Yet present existing finger vein features extracting method is less, and not accurate enough to some low-quality vein image feature extractions.

Summary of the invention

The objective of the invention is to deficiency at existing feature technology existence, a kind of feature extracting method of vein image medium sized vein of finger vena identification system is proposed, the nucleus module that belongs to finger vena identification system, be intended to extract interested vein pattern in the image accurately, reduce and extract error characteristic and miss real features, can improve the accuracy of identification of total system simultaneously and improve image processing speed.

The technical scheme that the present invention takes is:

A kind of image vena characteristic extracting method of finger vena identification system is the key component of finger vena identification system, and the quality of extracting method directly influences the accuracy of identification of total system.By analyzing finger venous image as can be known, wherein vein pattern is in zone dark on the image, and background is brighter relatively, and then in conjunction with the knowledge of digital picture, obtaining vein pattern is exactly the zone that is in paddy shape in image, as shown in Figure 3.Therefore, the extraction vein pattern just is converted into the paddy shape zone in the detected image.For this reason, design a kind of operator and detected this zone.Because all there is the field of direction more clearly in finger venous image paddy shape zone (veinprint), the accuracy that the field of direction is estimated has directly determined the effect of figure image intensifying and partitioning algorithm.For estimating the field of direction, the trend of vein is divided into 8 directions, as shown in Figure 2, and i=1,2 ..., 8 represent the position of these 8 directions respectively.Design a kind of operator then and detect this paddy shape zone, promptly utilize 8 template operator corresponding elements (as shown in Figure 3) on the direction that the gray-scale value in this neighborhood of a point is done product respectively, then accumulation is added summation (size of template is big or small identical with neighborhood).Add up and, obtain an image that contains vein pattern maximum on 8 directions at last as the gray-scale value of this pixel.But have pseudo-vein pattern and a noise in this image.Next utilize threshold value to mark off most probable not have the characteristic area of vein (background area), both contained the zone of noise and pseudo-vein pattern

(fuzzy region), most probable have only three zones, zone (target area) of vein pattern.Respectively each zone is processed then, promptly removing does not have characteristic area, the reservation most probable of vein to have only the zone of vein pattern, segmentation to cut the zone of both containing noise and pseudo-vein pattern to most probable.Because the zone of both having contained noise and pseudo-vein pattern belonged to fuzzy region, to this part cutting apart of cutting also be the key that whole vein pattern extracts.To the processing of this part, at first to carry out sharpening to fuzzy part, can reduce the noise of this part like this and reject pseudo-vein and help Threshold Segmentation.Carry out Threshold Segmentation then, but can not utilize single threshold value to come split image, will produce like this and lose some vein pattern information when removing noise and pseudo-vein pattern.The way that solves is to adopt many threshold segmentation methods.

Processing concrete grammar to each zone is as follows:

(1) removes the characteristic area that most probable does not have vein, promptly remove the picture element that is evident as the non-vein feature.At first, because through making product, then accumulation is added grey scale pixel value that summation makes paddy shape zone greater than 0, smooth zone is 0, and the zone that forms convex is less than 0.So, the pixel one of the gray-scale value correspondence greater than zero is decided to be vein pattern, and its value is constant, and is the non-vein feature smaller or equal to zero point.Getting threshold value is zero, carries out cutting apart the first time.

(2) keep the zone that most probable has only vein pattern, promptly keep the pixel that is evident as vein pattern.After first step processing, the feature major part in the image is a vein pattern, but also has pseudo-vein pattern and noise.Just can keep tangible vein pattern by getting appropriate threshold.Gray-scale value greater than this threshold value is changed to threshold value, and is constant less than the gray-scale value of threshold value.

(3) fuzzy part sharpening.Because through after the processing of first two steps, pseudo-vein pattern and noise mainly are present in this fuzzy part, therefore before cutting apart, even the at first essential blur level that reduces fuzzy part is fuzzy part sharpening.Can further reduce noise like this and reject pseudo-vein, and help next step and accurately cut apart.Its way is: the individual layer fuzzy enhancement algorithm is improved, has been proposed a kind of regional fuzzy enhancement algorithm, (1) to original image by formula $u_{\mathrm{mn}}=\frac{\mathrm{Gray}(m,n)}{K-1}$ ((m, n) (K-1 represents maximum gray scale to remarked pixel to Gray, u for m, the gray-scale value of n) locating _MnThe expression expression transforms to vague plane mid point (m, the value of n) locating.) carry out the fuzzy characteristics extraction, obtain the fuzzy characteristics plane of image; (2) on the fuzzy characteristics plane, fuzzy characteristics is pressed the enhancing formula ${u\&prime;}_{\mathrm{mn}}=\left\{\begin{array}{cc}{u}_{\mathrm{mn}}{\left(\frac{u_{\mathrm{mn}}}{T_{\mathrm{mn}}}\right)}^{2^r}& u_{\mathrm{mn}}\&le;T_{\mathrm{mn}}\\ 1-(1-u_{\mathrm{mn}}){\left(\frac{{1-u}_{\mathrm{mn}}}{1-T_{\mathrm{mn}}}\right)}^{2^r}& T_{\mathrm{mn}}<u_{\mathrm{mn}}\end{array}\right.$ (

Expression is through strengthening (m, the value of n) locating, T on the fuzzy characteristics plane, back _MnThe average of all elements in the expression window, r represents to strengthen parameter,) degree of comparing enhancing conversion, fuzzy characteristics plane after being enhanced, promptly use (2p+1) * (2p+1) window (size can be selected) to travel through point in the fuzzy characteristics plane successively, calculate the threshold value of the interior average of this point (2p+1) * (2p+1) window as central point, if the gray-scale value of central point is greater than this threshold value, then utilize and strengthen formula its value increase, otherwise, then its value is reduced, thereby reduce blur level, reach the purpose of enhancing; (3) to new fuzzy characteristics plane press transformation for mula Gray (m, n) '=(K-1) u ' _Mn(u ' _Mn(Gray (m, n) ' gray-scale value after being enhanced) carries out inverse transformation for m, the value of n) locating, and draws the output image after the corresponding enhancing through strengthening on the fuzzy characteristics plane, back in expression.

(4) to fuzzy Threshold Segmentation partly.This be whole vein pattern extract key component, adopt regional NiBlack, it is the selection of threshold method of gray level image, in putting a certain size neighborhood, each utilize the NiBlack method to calculate the threshold value of this point, make in the image each point that a threshold value is all arranged, utilize each threshold value to cut apart then respectively, finally obtain vein pattern.

The present invention can extract the feature of finger vena accurately, and is special very effective for the feature extraction of some inferior quality finger venous images.Can improve simultaneously the accuracy of identification of total system and improve image processing speed.

Description of drawings

Fig. 1. schematic flow sheet of the present invention.

Fig. 2 .8 vein crestal line direction synoptic diagram.

Fig. 3 .8 vein crestal line direction synoptic diagram.

Wherein Fig. 3 a is the gray scale coordinate diagram of cross section part; Fig. 3 b is the position view of xsect in referring to vein image;

Fig. 4 is the operator of all directions;

Wherein Fig. 4 a is 0 ⁰(level) directional operator; Fig. 4 b is 157.5 ⁰Directional operator; Fig. 4 c is 135 ⁰Directional operator; Fig. 4 d is 112.5 ⁰Directional operator; Fig. 4 e is 90 ⁰Directional operator; Fig. 4 f is 67.5 ⁰Directional operator; Fig. 4 g is 45 ⁰Directional operator; Fig. 4 h is 22.5 ⁰Directional operator

Embodiment

General finger venous image all has the field of direction more clearly, and the accuracy that the field of direction is estimated has directly determined the effect of figure image intensifying and partitioning algorithm.

For estimating the field of direction, the trend of vein is divided into 8 directions, as shown in Figure 2.

In conjunction with Fig. 1, finger venous image characteristic extraction method is as follows:

Step1: extract paddy shape zone

For each pixel of finger-image,, be in 9 * 9 windows at center with this pixel in order to determine direction at this pixel place crestal line, calculate the product accumulation and Fgray (the i) (i=1 of operator (as shown in Figure 4) on corresponding with it 8 directions respectively, 2 ..., 8).The maximum that obtains then on these 8 directions adds up and Gmax.

$G\max =\underset{i}{\mathrm{Max}}\left(\mathrm{Fgray}\left(i\right)\right)$ (1)

Use the gray-scale value of maximal value Gmax then as this point.

Gray(m，n)＝Gmax； (2)

In the formula: Gray (m, n) remarked pixel (m, the gray-scale value of n) locating.

Step2: Threshold Segmentation

Step2.1: carry out the Threshold Segmentation first time

Step2.2: carry out the Threshold Segmentation second time

Gmean＝sum(Gray)/Num

In the formula: the mean value of nonzero element in the Gmean presentation video; Sum (Gray) and Num respectively in the presentation video nonzero element and and number.

Step2.3: the fuzzy enhancing

After preceding cut apart for twice, at this moment the gray-scale value scope is [0, Gmean] in the image, and this regional value exists the possibility of pseudo-vein pattern very big, belongs to fuzzy region, can strengthen with fuzzy operator.Its algorithm is as follows:

(1) calculates degree of membership, promptly

$u_{\mathrm{mn}}=G\left(\mathrm{Gray}(m,n)\right)=\frac{\mathrm{Gray}(m,n)}{K-1}---\left(5\right)$

(2) the average T of all elements in calculating (2p+1) * (2p+1) window _Mn

(3) in (2p+1) * (2p+1) window, calculate adjusted pixel grey scale degree of membership u _MnAnd gray-scale value Gray (m, n) ', its mathematic(al) representation is

${u\&prime;}_{\mathrm{mn}}=\left\{\begin{array}{cc}{u}_{\mathrm{mn}}{\left(\frac{u_{\mathrm{mn}}}{T_{\mathrm{mn}}}\right)}^{2^r}& u_{\mathrm{mn}}\&le;T_{\mathrm{mn}}\\ 1-(1-u_{\mathrm{mn}}){\left(\frac{{1-u}_{\mathrm{mn}}}{1-T_{\mathrm{mn}}}\right)}^{2^r}& T_{\mathrm{mn}}<u_{\mathrm{mn}}\end{array}\right.---\left(6\right)$

(4)Gray(m，n)′＝(K-1)u′ _mn (7)

Wherein, K-1 represents maximum gray scale

Step2.4: carry out the 3rd subthreshold and cut apart

Adopt (2p+1) * (2p+1) window on original image, to slide, this window center pixel gray-scale value be Gray (m, n) ', then all pixel values are constructed as follows set in this window:

S(i，j)＝{G _ray(m+k，n+L)′|k，L＝-p，…，-1，0，1，…，p}

Obtain all pixels in this window mean value Average (S (m, n)) and variances sigma (and m, n), computing formula is as follows:

$\mathrm{Average}\left(S(m,n)\right)=\frac{1}{(2p+1)(2p+1)}\underset{m=-p}{\overset{p}{\mathrm{\&Sigma;}}}\underset{n=-p}{\overset{p}{\mathrm{\&Sigma;}}}\mathrm{Gray}(m,n)\&prime;---\left(8\right)$

$\mathrm{\&sigma;}(m,n)=\sqrt{\frac{1}{(2p+1)(2p+1)}\underset{m=-p}{\overset{p}{\mathrm{\&Sigma;}}}\underset{n=-p}{\overset{p}{\mathrm{\&Sigma;}}}{(\mathrm{Gray}(m,n)\&prime;-\mathrm{Average}\left(S(m,n)\right))}^2}---\left(9\right)$

Utilize variance and average obtain segmentation threshold T (m, n)

T(m，n)＝Average(S(m，n))+α×σ(m，n)　　　　　　(10)

Each pixel all has a threshold value in this sampled images.Utilizing each threshold value to carry out two-value then respectively turns to

Obtain finally only containing vein pattern image Gray (m, n) '.

Claims (3)

1, a kind of vena characteristic extracting method of finger vena identification system is characterized in that it may further comprise the steps:

(1) the paddy shape zone of detection finger venous image

For each pixel of finger venous image, being in 9 * 9 windows at center with this pixel, the trend of vein is divided into 8 directions, detect operator according to 8 paddy shapes of 8 direction designs then; Utilize the paddy shape detection operator of 8 directions respectively the point in the neighborhood of each pixel 9 * 9 to be multiplied each other, then accumulation is added summation, at last with maximum add up and as the gray-scale value of this pixel, obtain an image, wherein be in paddy shape zone greater than zero value, the value smaller or equal to zero is in non-paddy shape zone;

(2) Threshold Segmentation

With the image division that obtains is three zones, and promptly not have the characteristic area of vein, the zone of containing noise and pseudo-vein pattern be the zone that fuzzy region and most probable have only vein pattern to most probable; Utilize threshold value to mark off this three zones then, respectively each zone is processed at last, concrete steps are as follows:

The first step is removed the characteristic area that most probable is not had vein, promptly removes the pixel that is evident as the non-vein feature; At first, getting threshold value is zero, carries out cutting apart the first time, promptly remains unchanged more than or equal to zero gray-scale value, and minus gray-scale value is changed to zero;

Second step kept the zone that most probable has only vein pattern, promptly kept the pixel that is evident as vein pattern; The mean value of getting all non-zero pixels point gray-scale values in the image carries out cutting apart the second time as segmentation threshold, even its gray values of pixel points greater than threshold value equals threshold value, remains unchanged less than the gray-scale value of threshold value;

In the 3rd step, blur the part sharpening: after cutting apart through preceding twice, at this moment the gray-scale value scope is [0, Gmean] in the image, and this regional value exists the possibility of pseudo-vein pattern very big, belongs to fuzzy region, strengthens with regional fuzzy enhancement algorithm;

The 4th step, Threshold Segmentation to fuzzy part: the selection of threshold method that adopts the area grayscale image, in putting a certain size neighborhood, each utilize the selection of threshold method of gray level image to calculate the threshold value of this point, make in the image each point that a threshold value is all arranged, utilize each threshold value to cut apart then respectively, finally obtain vein pattern.

2, the vena characteristic extracting method of finger vena identification system according to claim 1 is characterized in that the specific practice in the paddy shape zone of described detection finger venous image is:

For each pixel of image, be in 9 * 9 windows at center with this pixel, calculate adding up and Fgray (i) (i=1,2 of operator on corresponding with it 8 directions respectively ..., 8), obtain adding up and G max on these 8 directions;

$G\max =\underset{i}{\mathrm{Max}}\left(\mathrm{Fgray}\left(i\right)\right)---\left(1\right)$

Use the gray-scale value of maximal value Gmax then as this point

Gray(m，n)＝Gmax (2)

In the formula: Gray (m, n) remarked pixel (m, the gray-scale value of n) locating.

3, the vena characteristic extracting method of finger vena identification system according to claim 1 is characterized in that described Threshold Segmentation specific practice is:

The first step: carry out the Threshold Segmentation first time

Second step: carry out the Threshold Segmentation second time

Gmean＝sum(Gray)/Num

In the formula: the mean value of nonzero element in the Gmean presentation video, sum (Gray) and Num respectively in the presentation video nonzero element and and number;

The 3rd step: the fuzzy enhancing

Fuzzy region is strengthened with fuzzy operator, and algorithm is as follows:

(1) by formula to original image $u_{\mathrm{mn}}=\frac{\mathrm{Gray}(m,n)}{K-1}$ Carry out fuzzy characteristics and extract, obtain the fuzzy characteristics plane of image; Wherein (m, n) (K-1 represents maximum gray scale to remarked pixel to Gray, u for m, the gray-scale value of n) locating _MnExpression transforms to vague plane mid point (m, the value of n) locating;

(2) on the fuzzy characteristics plane, fuzzy characteristics is pressed the enhancing formula ${u\&prime;}_{\mathrm{mn}}=\left\{\begin{array}{cc}{u}_{\mathrm{mn}}{\left(\frac{u_{\mathrm{mn}}}{T_{\mathrm{mn}}}\right)}^{2^r}& u_{\mathrm{mn}}\&le;T_{\mathrm{mn}}\\ 1-(1-u_{\mathrm{mn}}){\left(\frac{1-u_{\mathrm{mn}}}{1-T_{\mathrm{mn}}}\right)}^{2^r}& T_{\mathrm{mn}}<u_{\mathrm{mn}}\end{array}\right.$ Degree of comparing strengthens conversion, fuzzy characteristics plane after being enhanced, promptly use (2p+1) * (2p+1) window, travel through the point in the fuzzy characteristics plane successively, calculate the threshold value of the interior average of this point (2p+1) * (2p+1) window as central point, if the gray-scale value of central point is greater than this threshold value, then utilize to strengthen formula with its value increase, otherwise, then its value is reduced, thereby the reduction blur level reaches the purpose of enhancing; U ' wherein _MnExpression is through strengthening (m, the value of n) locating, T on the fuzzy characteristics plane, back _MnThe average of all elements in the expression window, r represents to strengthen parameter;

(3) transformation for mula Gray (m, n) the u ＇ of ＇=(K-1) are pressed in new fuzzy characteristics plane _MnCarry out inverse transformation, draw the output image after the corresponding enhancing; U ＇ wherein _MnExpression is through strengthening (m, the value of n) locating, Gray (m, the gray-scale value after n) ＇ is enhanced on the fuzzy characteristics plane, back;

The 4th step: to bluring the Threshold Segmentation of part:

Adopt (2p+1) * (2p+1) window on original image, to slide, this window center pixel gray-scale value be Gray (m, n) ', then all pixel values are constructed as follows set in this window:

S(i，j)＝{G _ray(m+k，n+L)＇|k，L＝-p，…，-1，0，1，…，p}

Obtain all pixels in this window mean value Average (S (m, n)) and variances sigma (and m, n), computing formula is as follows:

$\mathrm{Average}\left(S(m,n)\right)=\frac{1}{(2p+1)(2p+1)}\underset{m=-p}{\overset{p}{\mathrm{\&Sigma;}}}\underset{n=-p}{\overset{p}{\mathrm{\&Sigma;}}}\mathrm{Gray}(m,n)\&prime;---\left(8\right)$

$\text{\&sigma;}(m,n)=\sqrt{\frac{1}{(2p+1)(2p+1)}\underset{m=-p}{\overset{p}{\mathrm{\&Sigma;}}}\underset{n=-p}{\overset{p}{\mathrm{\&Sigma;}}}{(\mathrm{Gray}(m,n)\&prime;-\mathrm{Average}\left(S(m,n)\right))}^2}---\left(9\right)$

Utilize variance and average obtain segmentation threshold T (m, n)

T(m，n)＝Average(S(m，n))+α×σ(m，n)　　　　　　　　　　　(10)

Each pixel all has a threshold value in this sampled images, utilizes each threshold value to carry out two-value then respectively and turns to

Obtain finally only containing vein pattern image Gray (m, n) '.

Images