CN106127151A - Based on the finger vein identification method and the device that improve local binary model - Google Patents

Abstract

The present invention discloses a kind of based on the finger vein identification method and the device that improve local binary model, it is possible to overcome effect of noise, improves the accuracy identified and is prone to parallelization realization.S1, obtaining finger vein image to be identified, location refers to vein pattern point position, and calculates the LmTP feature of each characteristic point based on the local binary model improved；The finger vein pattern point of described finger vein pattern point and default template image is mated by S2, similarity by means of fusion distance and LmTP feature；S3, according to coupling result, calculate matching double points overall similarity；S4, carry out described finger vein image referring to hand vein recognition according to described overall situation similarity.

Description

Based on the finger vein identification method and the device that improve local binary model

Technical field

The present invention relates to Image Processing and Pattern Recognition field, be specifically related to a kind of based on the finger improving local binary model Vein identification method and device.

Background technology

The oxygen loss hemoglobin that vein identification technology is found that in vein blood vessel when being and stand in medical field research day can be inhaled Receive the near infrared ray of certain wavelength, and it was found that the vein blood vessel of people has good uniqueness.Refer to that hand vein recognition is as having The second filial generation biological identification technology of high antifalsification gets most of the attention.

The most common refer to that hand vein recognition algorithm is the method using local message, as LBP and derivative LDP thereof, LTP, The methods such as EQP.First the method using local shape information positions the characteristic point referring to vein image, afterwards by the shape of characteristic point Shape mates, identifies new finger vein image.Owing to only using characteristic point, thus the template that such algorithm obtains is the least, it is easy to Storage, and when identifying, amount of calculation is little.But the method for LBP class is very sensitive to noise, the impact on recognition accuracy is bigger.Another Aspect, when mating characteristic point, the method frequently with SVD and hungarian is mated.SVD method has ratio not The advantage such as degeneration, rotational invariance.But, its shortcoming is, SVD decomposes in the singular vectors obtained with the presence of negative, it is impossible to very Explain well its physical significance；On the other hand, the method that SVD decomposes is difficult to parallelization, causes its speed the slowest.Huangrian Method also known as Hungary Algorithm, this algorithm by matching problem as assignment problem, but the most difficult convergence of this algorithm, it is impossible to Find optimal solution.In addition, during both approaches matching characteristic point, occur that the probability of error hiding is bigger.

Summary of the invention

The deficiency existed for prior art and defect, the present invention provides a kind of quiet based on the finger improving local binary model Arteries and veins recognition methods and device.

On the one hand, the embodiment of the present invention proposes a kind of finger vein identification method based on improvement local binary model, including:

S1, obtaining finger vein image to be identified, location refers to vein pattern point position, and based on the local binary mould improved Type calculates the LmTP feature of each characteristic point；

S2, by means of the similarity of fusion distance and LmTP feature to described finger vein pattern point and default template image Finger vein pattern point mate；

S3, according to coupling result, calculate matching double points overall similarity；

S4, carry out described finger vein image referring to hand vein recognition according to described overall situation similarity.

On the other hand, the embodiment of the present invention proposes a kind of finger vein identification device based on improvement local binary model, bag Include:

LmTP feature calculation unit, for obtaining finger vein image to be identified, location refers to vein pattern point position, and base The LmTP feature of each characteristic point is calculated in the local binary model improved；

Matching unit, for and presetting described finger vein pattern point by means of the similarity of fusion distance and LmTP feature The finger vein pattern point of template image mate；

Overall situation similarity calculated, for the result according to coupling, calculates the overall similarity of matching double points；

Recognition unit, for carrying out referring to hand vein recognition to described finger vein image according to described overall situation similarity.

Finger vein identification method based on improvement local binary model described in the embodiment of the present invention and device, on the one hand, Improve local binary model so that it is be susceptible to effect of noise, on the other hand, it is proposed that LmTP feature, this LmTP feature energy Enough overcome effect of noise, and when referring to vein pattern Point matching, the similarity by means of fusion distance and LmTP feature is entered Row coupling, it is possible to obtain Optimum Matching result so that it is when match point is less, be difficult to the situation that error hiding occurs, the standard of coupling Exactness is higher and is prone to parallelization realization, thus compared to prior art, the present invention can overcome effect of noise, improves and identifies Accuracy and be prone to parallelization realize.

Accompanying drawing explanation

Fig. 1 is that the present invention is based on the schematic flow sheet referring to vein identification method one embodiment improving local binary model；

Fig. 2 is the schematic flow sheet of S2 mono-embodiment in Fig. 1；

Fig. 3 is that the present invention is based on the structural representation referring to vein identification device one embodiment improving local binary model.

Detailed description of the invention

For making the purpose of the embodiment of the present invention, technical scheme and advantage clearer, below in conjunction with the embodiment of the present invention In accompanying drawing, the technical scheme in the embodiment of the present invention is explicitly described, it is clear that described embodiment is the present invention A part of embodiment rather than whole embodiments.Based on the embodiment in the present invention, those of ordinary skill in the art are not having Make the every other embodiment obtained under creative work premise, broadly fall into the scope of protection of the invention.

Referring to Fig. 1, the present embodiment discloses a kind of finger vein identification method based on improvement local binary model, including:

S1, obtaining finger vein image to be identified, location refers to vein pattern point position, and based on the local binary mould improved Type calculates the LmTP feature of each characteristic point；

S2, by means of the similarity of fusion distance and LmTP feature to described finger vein pattern point and default template image Finger vein pattern point mate；

S3, according to coupling result, calculate matching double points overall similarity；

S4, carry out described finger vein image referring to hand vein recognition according to described overall situation similarity.

Finger vein identification method based on improvement local binary model described in the embodiment of the present invention, on the one hand, improve Local binary model so that it is be susceptible to effect of noise, on the other hand, it is proposed that LmTP feature, this LmTP feature can overcome Effect of noise, and when referring to vein pattern Point matching, the similarity by means of fusion distance and LmTP feature is mated, Can obtain Optimum Matching result so that it is when match point is less, be difficult to situation error hiding occur, the accuracy of coupling is higher And be prone to parallelization realization, thus compared to prior art, the present invention can overcome effect of noise, improves the accuracy identified And it is prone to parallelization realization.

Alternatively, in the present invention is based on another embodiment referring to vein identification method improving local binary model, institute State location and refer to vein pattern point position, including:

Described finger vein image is carried out pretreatment；

Finger venosomes in the image that pretreatment described in use maximum curvature algorithm identification obtains and background area, to institute State the image that pretreatment obtains and carry out binaryzation, obtain bianry image, and described bianry image is refined, wherein, described The pixel value of finger venosomes is 1, and the pixel value of described background area is 0；

Take each position of the bianry image after the window described refinement of traversal of 3*3, it is judged that the central spot of this window Whether the pixel value of pixel is 1, if 1, then calculates N_trans, it is judged that N_transWhether it is 2 or not less than 6, if N_transFor 2, then the pixel of this position is end points end points (End Point, EP), if or N_transNot less than 6, then the picture of this position Vegetarian refreshments is branch point (Bifurcation Point, BP), and wherein, the central point of this window is positioned at this position, N_transCalculating public Formula is

$N_{trans}=\underset{i=1}{\overset{8}{\Σ}}{p}_{i+1}-p_i,$

In formula, p₁To p₈It is followed successively by this window from the beginning of upper left position, the pixel being distributed clockwise along window periphery The pixel value of point, p₉=p₁。

In the embodiment of the present invention, by described finger vein image is carried out pretreatment, it is possible to while smooth noise, strengthen Refer to the difference of venosomes and background area.

Alternatively, in the present invention is based on another embodiment referring to vein identification method improving local binary model, institute State and described finger vein image is carried out pretreatment, including:

Use even symmetry Gabor filter that described finger vein image is filtered, wherein, Gabor filter function table It is shown as

In formula, x'=xcos θ+ysin θ, y'=-xsin θ+ycos θ, (x, y) is original coordinates, and (x', y') is through filtering Coordinate after ripple, λ is wavelength；θ is direction, and this state modulator Gabor function stripe direction, value is 0-360 °；For phase place Skew, span is-180 °-180 °；γ is length-width ratio, determines the ellipticity of Gabor function, and σ represents the height of Gabor function The standard deviation of this factor；

Use CLAHE that filtered image is strengthened.

In the embodiment of the present invention, using CLAHE to carry out filtered image strengthening is prior art, the most superfluous State, as the method etc. in the patent of Application No. 201510661549.X can be used.

Alternatively, in the present invention is based on another embodiment referring to vein identification method improving local binary model, institute State the finger venosomes in the image using pretreatment described in maximum curvature algorithm identification to obtain and background area, including:

Calculating the curvature of each pixel in the image that described pretreatment obtains, computing formula isWherein, K_fZ () represents the curvature of the pixel that position is z,(x₀,y₀) it is the coordinate of position z, As i > 0 time, x_iRepresent at the z of position the abscissa of ith pixel point on the right side of pixel, when i < when 0, x_iRepresent pixel at the z of position Left side ith pixel point abscissa, as i > 0 time, y_iThe vertical coordinate of ith pixel point directly over pixel at expression position z, When i < when 0, y_iThe vertical coordinate of ith pixel point immediately below pixel at expression position z, w is parameter, putting down for reference section Average, its empirical value is 8；

Determine the maximum of described curvature, and using pixel corresponding for described maximum as possible finger vein center Point；

To each possible finger vein center point marking, score value computing formula is S'_cr=K_f×W_r, wherein, S'_crIt is first Score value, K_fFor curvature, W_rIt is positive peak width for referring to the local curvature of vein center point；

For each pixel in the image that described pretreatment obtains, to horizontal direction, vertical direction, and ± First score value summation of the possible finger vein center point on 45 ° of directions, and using summed result as second point of this pixel Value；

For each pixel in the image that described pretreatment obtains, calculated level direction, vertical direction respectively, with And the C of this pixel on ± 45 ° of directions_d(x, y), and to horizontal direction, vertical direction, and this pixel on ± 45 ° of directions C_d(x, y) maximizing, obtain this pixel G (x, y), wherein,

C_d(x, y)=min{max (S_cr(x-1,y),S_cr(x-2,y)),max(S_cr(x+1,y),S_cr(x+2, y)) }, (x, Y) coordinate of pixel, S are represented_cr(x-1 is y) that in a direction, coordinate is (x, the of front first pixel of pixel y) Two score values, S_cr(x-2 is y) that in a direction, coordinate is (x, the second score value of front second pixel of pixel y), S_cr(x + 1, it is y) that in a direction, coordinate is (x, the second score value of first pixel after pixel y), S_cr(x+2 is y) a certain On direction, coordinate is (x, the second score value of second pixel after pixel y)；

For each pixel in the image that described pretreatment obtains, (x, y) with the first number to compare the G of this pixel The magnitude relationship of value, if (x, y) less than described first numerical value, it is determined that this pixel belongs to background area, no for the G of this pixel Then, it is determined that this pixel belongs to finger venosomes.

In the embodiment of the present invention, the possible finger vein center point in horizontal direction, vertical direction crosses this pixel respectively Horizontal direction straight line, possible finger vein center point on vertical direction straight line, possible on+45 ° of directions ,-45 ° of directions Refer to that vein center point crosses the possible finger vein center point on this pixel and straight line that slope is 1 or-1 respectively.For first Numerical value, can by G (x, y) in value more than 0 according to being ranked up from big to small, and take the value conduct of the neutral element after sequence First numerical value, it is also possible to take other value as required, other value near the value of the most described neutral element, the present invention is to this not It is construed as limiting.

Alternatively, in the present invention is based on another embodiment referring to vein identification method improving local binary model, institute State the LmTP feature calculating each characteristic point based on the local binary model improved, including:

For each characteristic point, calculateWithBy described WithEven Pick up the LmTP feature obtaining this feature point, wherein,

a₀To a₇It is followed successively by the image that described pretreatment obtains in the 3*3 window centered by this feature point from the upper left corner Position starts, along the gray value of the pixel that window periphery is distributed clockwise, b₀To b₁₅It is followed successively by what described pretreatment obtained In image in 5*5 window centered by this feature point from the beginning of upper left position, the picture being distributed clockwise along window periphery The gray value of vegetarian refreshments, c₀To c₂₃Be followed successively by the image that described pretreatment obtains in the 7*7 window centered by this feature point from Upper left position starts, along the gray value of the pixel that window periphery is distributed clockwise, i ∈ (1,2 ..., 7), m is this feature The gray value of point,

T is default numerical value, and general value is 5.

Alternatively, referring to Fig. 2, in the present invention based on another enforcement referring to vein identification method improving local binary model In example, described S2, including:

S20, judge whether current cycle time reaches second value；

If S21 reaches described second value, then preserve optimal matching pointsAnd Best Affine conversion ginseng Number S, T, perform step S32, otherwise, then perform step S22；

S22, determine from described finger vein figure according to the end points of described finger vein image and the end points of described template image As to described template image affine transformation parameter；

S23, judge that described affine transformation parameter, whether in default scope, if in the range of described, then performs step S24, otherwise, then performs step S20；

S24, by affine transformation parameter, the characteristic point of described finger vein image is carried out affine transformation, the spy after conversion Levy position a little and be designated as (f_xi',f_yi'), i=1 ..., n₁, n₁Characteristic point quantity for described finger vein image；

S25, use range information calculate (f_xi',f_yi') and (g_xj,g_yj) optimal matching points, wherein, (g_xj,g_yj) it is The characteristic point position of described template image, j=1 ..., n₂, n₂Characteristic point quantity for described template image；

S26, the local similarity S1 based on distance of calculating current matching point pair；

Whether S27, the described local similarity S1 based on distance of judgement be less than third value；

If S28 is less than described third value, then updating S is described affine transformation parameter, updates optimal matching points For described optimal matching points, updating described third value is S1, and performs step S29, otherwise, directly performs step S29,d_lBeing the Euclidean distance of l optimal matching points, M is the quantity of optimal matching points；

S29, the local similarity Similarity based on LmTP feature of calculating current matching point pair_L；

S30, the described local similarity Similarity based on LmTP feature of judgement_LWhether more than the 4th numerical value；

If S31 is more than described 4th numerical value, then updating T is described affine transformation parameter, updates optimal matching points For described optimal matching points, updating described 4th numerical value is Similarity_L, and perform step S20, otherwise, directly perform Step S20；

The branch point of described finger vein image is imitated by Best Affine transformation parameter S, T that S32, utilization calculate respectively Penetrate conversion, and the branch point of the branch point after conversion each time and described template image is mated, obtain matching double points

In the embodiment of the present invention, second value can be arranged as required to, and such as could be arranged to 50,60 etc..For referring to Vein image translates, rotates, scales situation about all existing, it is possible to use the formula of following homogeneous coordinates form calculates affine change Change parameter p₁-p₆:

$\left[\begin{array}{ccc}{p}_1& p_2& p_3\\ p_4& p_5& p_6\\ 0& 0& 1\end{array}\right]\left[\begin{array}{ccc}{x}_1& x_2& x_3\\ y_1& y_2& y_3\\ 1& 1& 1\end{array}\right]=\left[\begin{array}{ccc}{x}_1^{\&prime;}& x_2^{\&prime;}& x_3^{\&prime;}\\ y_1^{\&prime;}& y_2^{\&prime;}& y_3^{\&prime;}\\ 1& 1& 1\end{array}\right],$

Wherein { (x_i,y_i) | i=1,2,3} is the coordinate of three end points referring to vein image, { (x_i',y_i') | i=1,2, 3} is the position of three end points of template image, wherein, (x_i,y_i) corresponding (x_i',y_i').By three groups of corresponding coordinate informations, Just can be in the hope of one group of affine transformation parameter p₁-p₆.Refer in the scope now preset:

0.5≤p₁≤ 1.5,

-0.5≤p₂≤ 0.5,

-40≤p₃≤ 40,

-0.5≤p₄≤ 0.5,

0.5≤p₅≤ 1.5,

-40≤p₆≤40。

When translating, rotate, scale part existence, affine transformation parameter is p₁-p₆In partial parameters, herein Repeat no more.When characteristic point is calculated optimum affine transformation parameter, use end points, because end points is more more than branch point, meter The affine transformation parameter arrived is more accurate.

The initial value of third value can be a bigger numerical value, is such as 10⁴、1.7×10³⁰⁸Deng.At the beginning of 4th numerical value Initial value is generally 0.

Alternatively, in the present invention is based on another embodiment referring to vein identification method improving local binary model, institute State the local similarity Similarity based on LmTP feature calculating current matching point pair_L, including:

Calculate the Euclidean distance d of the jth characteristic point of the ith feature point after conversion and described template image_ij, calculate public affairs Formula is

d_ij=((f_xi'-g_xj)²+(f_yi'-g_yj)²)^0.5；

Computed range matrix D, wherein,

Wherein, D_ij=exp (-d_ij)；

By the element value characteristic of correspondence point the most maximum in affiliated row and affiliated row in the D spy to being defined as coupling It is a little right to levy, and the feature point pairs letter now mated is calculated as M1={ (f_n',g_n), n=1 ..., N, N are the sum of matching double points；

Calculate Similarity_L, computing formula is

Wherein,L_n1(k) and L_n2K () is respectively the n-th matching double points In the kth position of LmTP feature of two characteristic points, Len is 64.

For a pair LmTP feature L_i1And L_i2, length is 64, owing to LmTP feature comprises-1, it is impossible to use Hamming distance From calculating similarity, thus the present invention calculates this similarity Sim to feature by following formula_G(i):

${\mathrm{Sim}}_G\left(i\right)=\frac{1}{2Len}\underset{j=1}{\overset{Len}{\&Sigma;}}|L_{i1}\left(j\right)-L_{i2}\left(j\right)|,$

If matching N number of characteristic point, similarity Similarity of matching double points_GCan be calculated by following formula:

${\mathrm{Similarity}}_G=\frac{1}{N}\underset{i=1}{\overset{N}{\&Sigma;}}{\mathrm{Sim}}_G\left(i\right),$

Similarity Similarity_GThe least, represent the two template the most similar.But this similarity can cause error hiding, As when the characteristic point matched is little, overall situation Similarity value may be the least, can cause matching error.Thus, above-mentioned similarity It is referred to as overall situation similarity.Meanwhile, overall situation similarity is modified by the present invention, L_i1And L_i2This local similarity to feature Calculated by following formula:

${\mathrm{Sim}}_L\left(i\right)=\frac{1}{2Len}\underset{j=1}{\overset{Len}{\&Sigma;}}(1-|L_{i1}\left(j\right)-L_{i2}\left(j\right)\left|\right),$

The local similarity Similarity of matching double points_LJust can be calculated by following formula:

${\mathrm{Similarity}}_L=\underset{i=1}{\overset{N}{\&Sigma;}}{\mathrm{Sim}}_L\left(i\right),$

Local similarity is the biggest, and the two template is the most similar.Use after local similarity, when the feature point pairs matched relatively Time few, local similarity is not too large, thus avoids the occurrence of error hiding.

Alternatively, in the present invention is based on another embodiment referring to vein identification method improving local binary model, institute State S3, including:

Calculate optimal matching pointsOverall similarityAnd Overall phase Like degreeAnd choose the smaller's overall similarity as described matching double points.

Alternatively, in the present invention is based on another embodiment referring to vein identification method improving local binary model, institute State S4, including:

Described overall situation similarity is compared with the 5th numerical value, if described overall situation similarity is more than the 5th numerical value, the most really It is set to different finger, otherwise, it is determined that for identical finger.

In the embodiment of the present invention, the 5th general threshold value of numerical value takes 0.25.

Referring to Fig. 3, the present embodiment discloses a kind of finger vein identification device based on improvement local binary model, including:

LmTP feature calculation unit 1, for obtaining finger vein image to be identified, location refers to vein pattern point position, and The LmTP feature of each characteristic point is calculated based on the local binary model improved；

Matching unit 2, for the similarity by means of fusion distance and LmTP feature to described finger vein pattern point with pre- If the finger vein pattern point of template image mate；

Overall situation similarity calculated 3, for the result according to coupling, calculates the overall similarity of matching double points；

Recognition unit 4, for carrying out referring to hand vein recognition to described finger vein image according to described overall situation similarity.

Finger vein identification device based on improvement local binary model described in the embodiment of the present invention, on the one hand, improve Local binary model so that it is be susceptible to effect of noise, on the other hand, it is proposed that LmTP feature, this LmTP feature can overcome Effect of noise, and when referring to vein pattern Point matching, the similarity by means of fusion distance and LmTP feature is mated, Can obtain Optimum Matching result so that it is when match point is less, be difficult to situation error hiding occur, the accuracy of coupling is higher And be prone to parallelization realization, thus compared to prior art, the present invention can overcome effect of noise, improves the accuracy identified And it is prone to parallelization realization.

Although being described in conjunction with the accompanying embodiments of the present invention, but those skilled in the art can be without departing from this Making various modifications and variations in the case of bright spirit and scope, such amendment and modification each fall within by claims Within limited range.

Claims (10)

1. a finger vein identification method based on improvement local binary model, it is characterised in that including:

S1, obtaining finger vein image to be identified, location refers to vein pattern point position, and based on the local binary model meter improved Calculate the LmTP feature of each characteristic point；

S2, by means of the similarity of fusion distance and LmTP feature to described finger vein pattern point and the finger of default template image Vein pattern point mates；

S3, according to coupling result, calculate matching double points overall similarity；

S4, carry out described finger vein image referring to hand vein recognition according to described overall situation similarity.

Method the most according to claim 1, it is characterised in that described location refers to vein pattern point position, including:

Described finger vein image is carried out pretreatment；

Finger venosomes in the image that pretreatment described in use maximum curvature algorithm identification obtains and background area, to described pre- Processing the image obtained and carry out binaryzation, obtain bianry image, and refine described bianry image, wherein, described finger is quiet The pixel value in arteries and veins region is 1, and the pixel value of described background area is 0；

Take each position of the bianry image after the window described refinement of traversal of 3*3, it is judged that the pixel of the central spot of this window Whether the pixel value of point is 1, if 1, then calculates N_trans, it is judged that N_transWhether it is 2 or not less than 6, if N_transIt is 2, then The pixel of this position is end points, if or N_transNot less than 6, then the pixel of this position is branch point, wherein, and this window The central point of mouth is positioned at this position, N_transComputing formula be

$N_{trans}=\underset{i=1}{\overset{8}{\&Sigma;}}{p}_{i+1}-p_i,$

In formula, p₁To p₈It is followed successively by this window from the beginning of upper left position, the pixel being distributed clockwise along window periphery Pixel value, p₉=p₁。

Method the most according to claim 2, it is characterised in that described described finger vein image is carried out pretreatment, including:

Using even symmetry Gabor filter to be filtered described finger vein image, wherein, Gabor filter function representation is

In formula, x'=xcos θ+ysin θ, y'=-xsin θ+ycos θ, (x, y) is original coordinates, (x', y') be after filtering after Coordinate, λ is wavelength；θ is direction, and this state modulator Gabor function stripe direction, value is 0-360 °；For phase offset, Span is-180 °-180 °；γ is length-width ratio, determines the ellipticity of Gabor function, σ represent the Gauss of Gabor function because of The standard deviation of son；

Use CLAHE that filtered image is strengthened.

Method the most according to claim 3, it is characterised in that pretreatment described in described use maximum curvature algorithm identification obtains To image in finger venosomes and background area, including:

Calculating the curvature of each pixel in the image that described pretreatment obtains, computing formula isWherein, K_fZ () represents the curvature of the pixel that position is z, (x₀,y₀) be the coordinate of position z, as i > 0 time, x_iThe abscissa of ith pixel point on the right side of pixel at expression position z, as i < 0 Time, x_iRepresent at the z of position the abscissa of ith pixel point on the left of pixel, as i > 0 time, y_iRepresent that at the z of position, pixel is just gone up Side ith pixel point vertical coordinate, when i < when 0, y_iRepresent at the z of position the vertical coordinate of ith pixel point, w immediately below pixel Being parameter, for the meansigma methods of reference section, its empirical value is 8；

Determine the maximum of described curvature, and using pixel corresponding for described maximum as possible finger vein center point；

To each possible finger vein center point marking, score value computing formula is S'_cr=K_f×W_r, wherein, S'_crIt it is first point Value, K_fFor curvature, W_rIt is positive peak width for referring to the local curvature of vein center point；

For each pixel in the image that described pretreatment obtains, to horizontal direction, vertical direction, and ± 45 ° of sides First score value summation of possible finger vein center point upwards, and using summed result as the second score value of this pixel；

For each pixel in the image that described pretreatment obtains, calculated level direction, vertical direction respectively, and ± The C of this pixel on 45 ° of directions_d(x, y), and to the C of this pixel in horizontal direction, vertical direction, and ± 45 ° of directions_d (x, y) maximizing, obtain this pixel G (x, y), wherein,

C_d(x, y)=min{max (S_cr(x-1,y),S_cr(x-2,y)),max(S_cr(x+1,y),S_cr(x+2, y)) }, (x, y) table Show the coordinate of pixel, S_cr(x-1 is y) that in a direction, coordinate is (x, second point of front first pixel of pixel y) Value, S_cr(x-2 is y) that in a direction, coordinate is (x, the second score value of front second pixel of pixel y), S_cr(x+1, Y) it is that in a direction, coordinate is (x, the second score value of first pixel after pixel y), S_cr(x+2 y) is a direction Upper coordinate is (x, the second score value of second pixel after pixel y)；

For each pixel in the image that described pretreatment obtains, (x, y) with the first numerical value to compare the G of this pixel Magnitude relationship, if the G of this pixel (x, y) less than described first numerical value, it is determined that this pixel belongs to background area, otherwise, Then determine that this pixel belongs to finger venosomes.

Method the most according to claim 1, it is characterised in that it is special that described local binary model based on improvement calculates each Levy LmTP feature a little, including:

For each characteristic point, calculateWithBy described WithConnect Obtain the LmTP feature of this feature point, wherein,

a₀To a₇It is followed successively by the image that described pretreatment obtains in the 3*3 window centered by this feature point from upper left position Start, along the gray value of the pixel that window periphery is distributed clockwise, b₀To b₁₅It is followed successively by the image that described pretreatment obtains In in 5*5 window centered by this feature point from the beginning of upper left position, the pixel being distributed clockwise along window periphery Gray value, c₀To c₂₃It is followed successively by the image that described pretreatment obtains in the 7*7 window centered by this feature point from upper left Angle Position starts, along the gray value of the pixel that window periphery is distributed clockwise, i ∈ (1,2 ..., 7), m is this feature point Gray value,

T is default numerical value, and general value is 5.

Method the most according to claim 1, it is characterised in that described S2, including:

S20, judge whether current cycle time reaches second value；

If S21 reaches described second value, then preserve optimal matching pointsAnd Best Affine transformation parameter S, T, performs step S32, otherwise, then performs step S22；

S22, according to the end points of described finger vein image and the end points of described template image determine from described finger vein image to Described template image affine transformation parameter；

S23, judge described affine transformation parameter whether in default scope, if in the range of described, then perform step S24, Otherwise, then step S20 is performed；

S24, by affine transformation parameter, the characteristic point of described finger vein image is carried out affine transformation, the characteristic point after conversion Position be designated as (f '_xi,f′_yi), i=1 ..., n₁, n₁Characteristic point quantity for described finger vein image；

S25, use range information calculate (f '_xi,f′_yi) and (g_xj,g_yj) optimal matching points, wherein, (g_xj,g_yj) it is described The characteristic point position of template image, j=1 ..., n₂, n₂Characteristic point quantity for described template image；

S26, the local similarity S1 based on distance of calculating current matching point pair；

Whether S27, the described local similarity S1 based on distance of judgement be less than third value；

If S28 is less than described third value, then updating S is described affine transformation parameter, updates optimal matching pointsFor institute Stating optimal matching points, updating described third value is S1, and performs step S29, otherwise, directly performs step S29,d_lBeing the Euclidean distance of l optimal matching points, M is the quantity of optimal matching points；

S29, the local similarity Similarity based on LmTP feature of calculating current matching point pair_L；

S30, the described local similarity Similarity based on LmTP feature of judgement_LWhether more than the 4th numerical value；

If S31 is more than described 4th numerical value, then updating T is described affine transformation parameter, updates optimal matching pointsFor institute Stating optimal matching points, updating described 4th numerical value is Similarity_L, and perform step S20, otherwise, directly perform step S20；

Best Affine transformation parameter S, T that S32, utilization calculate carry out affine change to the branch point of described finger vein image respectively Change, and the branch point of the branch point after conversion each time and described template image is mated, obtain matching double points

Method the most according to claim 6, it is characterised in that described calculating current matching point pair based on LmTP feature Local similarity Similarity_L, including:

Calculate the Euclidean distance d of the jth characteristic point of the ith feature point after conversion and described template image_ij, computing formula is

d_ij=((f '_xi-g_xj)²+(f′_yi-g_yj)²)^0.5；

Computed range matrix D, wherein,

Wherein, D_ij=exp (-d_ij)；

By the element value characteristic of correspondence point the most maximum in affiliated row and affiliated row in the D characteristic point to being defined as coupling Right, the feature point pairs letter now mated is calculated as M1={ (f '_n,g_n), n=1 ..., N, N are the sum of matching double points；

Calculate Similarity_L, computing formula is

Wherein,L_n1(k) and L_n2K () is respectively in n-th matching double points two The kth position of the LmTP feature of individual characteristic point, Len is 64.

Method the most according to claim 6, it is characterised in that described S3, including:

Calculate optimal matching pointsOverall similarityAnd Overall similarityAnd choose the smaller's overall similarity as described matching double points.

Method the most according to claim 1, it is characterised in that described S4, including:

Described overall situation similarity is compared with the 5th numerical value, if described overall situation similarity is more than the 5th numerical value, it is determined that for Different fingers, otherwise, it is determined that for identical finger.

10. a finger vein identification device based on improvement local binary model, it is characterised in that including:

LmTP feature calculation unit, for obtaining finger vein image to be identified, location refers to vein pattern point position, and based on changing The local binary model entered calculates the LmTP feature of each characteristic point；

Matching unit, for the similarity by means of fusion distance and LmTP feature to described finger vein pattern point and default mould The finger vein pattern point of plate image mates；

Overall situation similarity calculated, for the result according to coupling, calculates the overall similarity of matching double points；

Recognition unit, for carrying out referring to hand vein recognition to described finger vein image according to described overall situation similarity.