CN102819754A - Fingerprint score fusion system and method based on Sigmoid expansion - Google Patents

Abstract

The invention discloses a fingerprint score fusion system and a fingerprint score fusion method based on Sigmoid expansion. The system comprises a training process and a testing process; the training process is realized via a training image preprocessing unit, a training feature extracting unit, a training feature aligning unit and a Sigmoid parameter training unit; and the testing process is realized via a testing image preprocessing unit, a testing feature extracting unit, a testing feature aligning unit and a score fusion unit. According to the score fusion unit, Sigmoid expansion is conducted on each score by using parameters output from the Sigmoid parameter training unit respectively; the scores after the Sigmoid expansion are fused; and a result after the fusion is taken as a final matching score. The final matching score is used for deciding whether the input fingerprint feature and the template fingerprint feature are originated from the same finger, thereby finishing fingerprint matching.

Description

A kind of print scores emerging system and method based on the Sigmoid expansion

Technical field

The invention belongs to the biometrics identification technology field; Relate to forward position knowledge such as Flame Image Process, pattern-recognition, computer technology; Be particularly related to a kind of print scores emerging system and method based on the Sigmoid expansion, expansion realizes fingerprint matching through the print scores fusion with based on Sigmoid.

Background technology

Fingerprint identification technology is to study and use one of proven technique the most in the present living things feature recognition field.Because fingerprint has the advantage of uniqueness and stability, and gatherer process is convenient, with low cost, has been widely used in a lot of aspects such as identity authentication, information security, access control at present.

The research of spending decades, fingerprint identification technology have obtained development at full speed, but the performance of algorithm for recognizing fingerprint does not also reach the precision of estimated in theory far away, and the time performance of algorithm also has much room for improvement.

Fingerprint minutiae is acknowledged as tool distinguishing ability of fingerprint and local feature the most reliably, also is the main flow algorithm of present fingerprint identification technology based on the matching algorithm of fingerprint minutiae feature.Wherein, Using more widely, minutiae feature comprises the description of minutiae point local direction, minutiae point local triangle structure, minutiae point topological structure etc.; Also there is other finger print information of associating to carry out Matching Algorithm, like the continuous crestal line information of associating minutiae point, local direction field information, local grain information etc.These methods finally all are to confirm the matching fractional of fingerprint image through the similarity of fingerprint minutiae and supplementary structure thereof.But; Because the translation between two width of cloth fingerprint images and the relative conversion of rotation and fingerprint deformation degree are all unknown in advance; So the corresponding relation between two groups of fingerprint minutiaes also is uncertain, this has only just determined by the matching fractional of the fingerprint image of the similarity decision of its supplementary structure of minutiae point its unreliability to be arranged equally.

In order to overcome the above problems, Many researchers begins to consider to merge the coupling that various features or multiple mark carry out fingerprint image.Feng has proposed 17 dimensional feature fusion methods of SVMs, but the method for SVMs is one " black box " for system, has no way of learning its internal processes, and quite consuming time.The fingerprint characteristic few in number that other has been delivered merges or also there is defective in the mark fusion method at time performance or others.

In view of improving constantly to the fingerprint recognition system performance requirement; Finger print matching method based on fingerprint minutiae can not meet the demands and not have improved space gradually merely, and just more and more receives researcher's attention based on the fingerprint identification method of Feature Fusion or mark fusion.Additional resource and time loss that mark fusion ratio Feature Fusion needs are all wanted much less, so practicality is better, but often improve limited on the performance.This is because different marks have different distributions and different attribute.If can different marks be expanded to identical space and unified as far as possible attribute through certain strategy, must improve the performance that mark merges.

Summary of the invention

The technical matters that (one) will solve

In existing print scores fusion process; Different marks have different distributions and different attribute, thus the problem of the performance boost of restriction print scores emerging system, and fundamental purpose of the present invention is to propose a kind of print scores emerging system and method based on the Sigmoid expansion; Through different marks being carried out the Sigmoid expansion; Make different marks have similar distribution and attribute, thereby improved the efficient that mark merges, improved the performance of print scores emerging system.

(2) technical scheme

For reaching said purpose, the invention provides a kind of print scores emerging system based on the Sigmoid expansion, comprising:

Training image pretreatment unit 1, each the training fingerprint image that is used for the training finger print data is concentrated obtains right refinement fingerprint image and the direction of fingerprint field picture of training fingerprint image respectively to carrying out pre-service respectively;

Training characteristics extraction unit 2; Be connected with training image pretreatment unit 1; Be used for from the right refinement fingerprint image of training fingerprint image of the training image pretreatment unit 1 output minutiae point information that takes the fingerprint, and from the right direction of fingerprint field picture of the training fingerprint image of training image pretreatment unit 1 output, extract the description of minutiae point local direction for each fingerprint minutiae;

Training characteristics aligned units 3; Be connected with training characteristics extraction unit 2; Be used for the right details in fingerprint dot information of the training fingerprint image of training characteristics extraction unit 2 outputs, the description of minutiae point local direction and direction of fingerprint field picture are calculated, obtain training average similarity score of the right minutiae point of fingerprint image and field of direction mean distance mark;

Sigmoid parameter training unit 4; Be connected with training characteristics aligned units 3; Be used for the right average similarity score of minutiae point and the field of direction mean distance mark of training fingerprint image of training characteristics aligned units 3 outputs calculated, obtain the Sigmoid spreading parameter of average similarity score of minutiae point and field of direction mean distance mark respectively;

Test pattern pretreatment unit 5 is used for the test fingerprint image obtaining right refinement fingerprint image of test fingerprint image and direction of fingerprint field picture respectively to doing pre-service respectively;

Test feature extraction unit 6; Be connected with test pattern pretreatment unit 5; Be used for from the right refinement fingerprint image of test fingerprint image of the test pattern pretreatment unit 5 output minutiae point information that takes the fingerprint, and from the right direction of fingerprint field picture of the test fingerprint image of test pattern pretreatment unit 5 outputs, extract the description of minutiae point local direction for each fingerprint minutiae;

Test feature aligned units 7; Be connected with test feature extraction unit 6; Be used for the right details in fingerprint dot information of the test fingerprint image of test feature extraction unit 6 outputs, the description of minutiae point local direction and direction of fingerprint field picture are calculated, obtain average similarity score of the right minutiae point of test fingerprint image and field of direction mean distance mark; And

Mark integrated unit 8; Be connected with test feature aligned units 7 and Sigmoid parameter training unit 4; Be used to utilize the Sigmoid spreading parameter of the average similarity score of minutiae point of Sigmoid parameter training unit 4 outputs that the right average similarity score of minutiae point of test fingerprint image of test feature aligned units 7 outputs is carried out the Sigmoid expansion, and utilize the Sigmoid spreading parameter of the field of direction mean distance mark of Sigmoid parameter training unit 4 outputs that the right field of direction mean distance mark of test fingerprint image of test feature aligned units 7 outputs is carried out the Sigmoid expansion; Two marks after will expanding then merge, and obtain final matching fraction, thereby judge that whether the test fingerprint image to deriving from same finger.

For reaching said purpose; The present invention also provides a kind of print scores fusion method based on the Sigmoid expansion; This method at first trains the Sigmoid spreading parameter of average similarity score of minutiae point and field of direction mean distance mark on training fingerprint image image set; Then the test fingerprint image is also aimed at extracting characteristic, on the information behind the aligning, extracted average similarity score of minutiae point and field of direction mean distance mark, and respectively each mark is carried out the Sigmoid expansion; Average similarity score of minutiae point and field of direction mean distance mark to after the Sigmoid expansion merge; The result who obtains after the fusion is as final matching fractional; Judge according to the size of final matching fraction whether input fingerprint characteristic and template fingerprint characteristic derive from same finger, thereby accomplish the coupling of fingerprint.

In the such scheme, described print scores fusion method based on the Sigmoid expansion specifically may further comprise the steps:

Step S1: fingerprint image is carried out pre-service, obtain refinement fingerprint image and direction of fingerprint field picture;

Step S2: the minutiae point that takes the fingerprint information and minutiae point local direction are described;

Step S3: describe the similarity between the calculated fingerprint minutiae point according to the minutiae point local direction;

Step S4: in the parameter training stage, choose one group of maximum fingerprint minutiae of minutiae point similarity, minutiae point information and direction of fingerprint field picture are aimed to right as initial point;

Step S5: in the parameter training stage, according to alignment result, the computational details is put average similarity score and field of direction mean distance mark;

Step S6: in the parameter training stage, average similarity score of minutiae point and field of direction mean distance mark are trained respectively, the computational details is put the Sigmoid spreading parameter of average similarity score and the Sigmoid spreading parameter of field of direction mean distance mark;

Step S7: at the fingerprint test phase,, choose N maximum fingerprint minutiae of minutiae point similarity, minutiae point information and direction of fingerprint field picture are aimed to right as aiming at initial point according to the fingerprint minutiae similarity that step S3 obtains;

Step S8: at the fingerprint test phase, according to each group alignment result, the computational details is put average similarity score and field of direction mean distance mark respectively;

Step S9: N group average similarity score of minutiae point and field of direction mean distance mark are carried out the Sigmoid expansion respectively, obtain average similarity score of minutiae point and field of direction mean distance mark after the N group is expanded;

Step S10: average similarity score of minutiae point and field of direction mean distance mark to after each group expansion merge;

Step S11: choose in N the matching fractional maximum matching fractional as final matching fraction, judge that according to the size of final matching fraction whether the test fingerprint image to deriving from same finger.

(3) beneficial effect

Print scores emerging system and method based on the Sigmoid expansion provided by the invention adopt the Sigmoid expanding policy to assimilate different print scores, it are distributed and the approximate unification of attribute.Because the different marks after the Sigmoid expansion have had similar distribution and attribute, therefore can more effectively merge, and have improved the precision of fingerprint matching; Simultaneously; The Sigmoid expansion is a simple computation process, and added space resource and time loss are all very little, therefore in the matching precision performance that has improved the mark emerging system; Its real-time performance does not obviously descend, and has finally improved the performance of fingerprint recognition system.

Description of drawings

Fig. 1 is the synoptic diagram based on the print scores emerging system of Sigmoid expansion according to the embodiment of the invention;

Fig. 2 is the structural representation based on training characteristics alignment modules in the print scores emerging system of Sigmoid expansion according to the embodiment of the invention;

Fig. 3 is the structural representation based on test feature alignment modules in the print scores emerging system of Sigmoid expansion according to the embodiment of the invention;

Fig. 4 is the structural representation based on the print scores emerging system mid-score Fusion Module of Sigmoid expansion according to the embodiment of the invention;

Fig. 5 is the method flow diagram that the print scores based on Sigmoid expansion according to the embodiment of the invention merges;

Fig. 6 shows the contrast and experiment of method on FVC2002DB1 that merges based on the print scores of Sigmoid expansion provided by the invention;

Fig. 7 shows the contrast and experiment of method on FVC2004DB1 that merges based on the print scores of Sigmoid expansion provided by the invention.

Embodiment

For making the object of the invention, technical scheme and advantage clearer, below in conjunction with specific embodiment, and with reference to accompanying drawing, to further explain of the present invention.To combine accompanying drawing that the present invention is specified below, and be to be noted that described embodiment only is intended to be convenient to understanding of the present invention, and it is not played any qualification effect.

The print scores that core concept of the present invention is based on the Sigmoid expansion merges; This fusion process at first trains the Sigmoid spreading parameter of average similarity score of minutiae point and field of direction mean distance mark on training fingerprint image image set; Then the test fingerprint image is also aimed at extracting characteristic; On the information behind the aligning, extract average similarity score of minutiae point and field of direction mean distance mark, and respectively each mark is carried out the Sigmoid expansion; Average similarity score of minutiae point and field of direction mean distance mark to after the Sigmoid expansion merge; The result who obtains after the fusion is as final matching fractional; Judge according to the size of final matching fraction whether input fingerprint characteristic and template fingerprint characteristic derive from same finger, thereby accomplish the coupling of fingerprint.

Based on above-mentioned thinking and purpose; Identification processing procedure with fingerprint image is divided into several steps below; Briefly introduce the key issue that should be noted that when carrying out each step; Design and improve our system, set up theoretical frame and system prototype that the final print scores based on the Sigmoid expansion merges.

Core of the present invention is the print scores that is used to merge is carried out the Sigmoid expansion, and then the print scores after the expansion is merged the coupling of accomplishing fingerprint.Prior art improved realized system architecture of the present invention; As shown in Figure 1; Print scores emerging system based on the Sigmoid expansion provided by the invention comprises training image pretreatment unit 1, training characteristics extraction unit 2, training characteristics aligned units 3, Sigmoid parameter training unit 4, test pattern pretreatment unit 5, test feature extraction unit 6, test feature aligned units 7 and mark integrated unit 8.Wherein, training image pretreatment unit 1, training characteristics extraction unit 2, training characteristics aligned units 3 and Sigmoid parameter training unit 4 belong to training process, are used for the required Sigmoid spreading parameter of training test process on training finger print data collection; Test pattern pretreatment unit 5, test feature extraction unit 6, test feature aligned units 7 and mark integrated unit 8 belong to test process, and the Sigmoid spreading parameter that utilizes training process to obtain is expanded and coupling carrying out Sigmoid the test fingerprint image of test process.

Training image pretreatment unit 1, each the training fingerprint image that is used for the training finger print data is concentrated obtains right refinement fingerprint image and the direction of fingerprint field picture of training fingerprint image respectively to carrying out pre-service respectively.Each training fingerprint image that 1 pair of training of training image pretreatment unit finger print data is concentrated is to carrying out pre-service respectively, and this pre-service comprises image segmentation, figure image intensifying, image binaryzation, micronization processes and field of direction extraction.

Training characteristics extraction unit 2; Be connected with training image pretreatment unit 1; Be used for from the right refinement fingerprint image of training fingerprint image of the training image pretreatment unit 1 output minutiae point information that takes the fingerprint, and from the right direction of fingerprint field picture of the training fingerprint image of training image pretreatment unit 1 output, extract the description of minutiae point local direction for each fingerprint minutiae.The training characteristics extraction unit 2 minutiae point information that from the right refinement fingerprint image of the training fingerprint image of training image pretreatment unit 1 output, takes the fingerprint, this fingerprint minutiae information comprises position, direction and the type of minutiae point.

Training characteristics aligned units 3; Be connected with training characteristics extraction unit 2; Be used for the right details in fingerprint dot information of the training fingerprint image of training characteristics extraction unit 2 outputs, the description of minutiae point local direction and direction of fingerprint field picture are calculated, obtain training average similarity score of the right minutiae point of fingerprint image and field of direction mean distance mark.

Sigmoid parameter training unit 4; Be connected with training characteristics aligned units 3; Be used for the right average similarity score of minutiae point and the field of direction mean distance mark of training fingerprint image of training characteristics aligned units 3 outputs calculated, obtain the Sigmoid spreading parameter of average similarity score of minutiae point and field of direction mean distance mark respectively.

Test pattern pretreatment unit 5 is used for the test fingerprint image obtaining right refinement fingerprint image of test fingerprint image and direction of fingerprint field picture respectively to doing pre-service respectively.5 pairs of test fingerprint images of test pattern pretreatment unit are to doing pre-service respectively, and this pre-service comprises image segmentation, figure image intensifying, image binaryzation, micronization processes and field of direction extraction.

Test feature extraction unit 6; Be connected with test pattern pretreatment unit 5; Be used for from the right refinement fingerprint image of test fingerprint image of the test pattern pretreatment unit 5 output minutiae point information that takes the fingerprint, and from the right direction of fingerprint field picture of the test fingerprint image of test pattern pretreatment unit 5 outputs, extract the description of minutiae point local direction for each fingerprint minutiae.The test feature extraction unit 6 minutiae point information that from the right refinement fingerprint image of the test fingerprint image of test pattern pretreatment unit 5 output, takes the fingerprint, this fingerprint minutiae information comprises position, direction and the type of minutiae point.

Test feature aligned units 7; Be connected with test feature extraction unit 6; Be used for the right details in fingerprint dot information of the test fingerprint image of test feature extraction unit 6 outputs, the description of minutiae point local direction and direction of fingerprint field picture are calculated, obtain average similarity score of the right minutiae point of test fingerprint image and field of direction mean distance mark.

Mark integrated unit 8; Be connected with test feature aligned units 7 and Sigmoid parameter training unit 4; Be used to utilize the Sigmoid spreading parameter of the average similarity score of minutiae point of Sigmoid parameter training unit 4 outputs that the right average similarity score of minutiae point of test fingerprint image of test feature aligned units 7 outputs is carried out the Sigmoid expansion, and utilize the Sigmoid spreading parameter of the field of direction mean distance mark of Sigmoid parameter training unit 4 outputs that the right field of direction mean distance mark of test fingerprint image of test feature aligned units 7 outputs is carried out the Sigmoid expansion; Two marks after will expanding then merge, and obtain final matching fraction, thereby judge that whether the test fingerprint image to deriving from same finger.

Formation synoptic diagram based on training characteristics aligned units 3 in the print scores emerging system of Sigmoid expansion provided by the invention is as shown in Figure 2, comprising:

Training minutiae point similarity calculated 31 is used for describing the similarity between all right fingerprint minutiaes of calculation training fingerprint image according to the right details in fingerprint dot information and the minutiae point local direction of training fingerprint image of training characteristics extraction unit 2 outputs;

Training is aimed at initial point to choosing unit 32; Be connected with training minutiae point similarity calculated 31; Be used for from all minutiae point similarities of training minutiae point similarity calculated 31 outputs, it is right as aiming at initial point to choose the maximum a pair of fingerprint minutiae of similarity;

Training aligned units 33 is aimed at initial point with training and is connected choosing unit 32, be used for based on training aim at initial point to the aligning initial point choosing unit 32 and choose to details in fingerprint dot information and direction of fingerprint field picture are carried out translation and rotation, with the completion aligning;

The average similarity score computing unit 34 of training minutiae point is connected with training aligned units 33, and the details in fingerprint dot information and the minutiae point local direction that are used for after training aligned units 33 is aimed at are described the basis, and the computational details is put average similarity score; And

Training field of direction mean distance score calculating unit 35 is connected with training aligned units 33, is used for the direction of fingerprint field picture basis after training aligned units 33 is aimed at, calculated direction field mean distance mark.

Formation synoptic diagram based on test feature aligned units 7 in the print scores emerging system of Sigmoid expansion provided by the invention is as shown in Figure 3, comprising:

Test detail point similarity calculated 71 is used for describing the similarity of calculating between all right minutiae point of test fingerprint image according to the test fingerprint image of test feature extraction unit 6 outputs right details in fingerprint dot information and minutiae point local direction;

Test is aimed at initial point to choosing unit 72; Be connected with test detail point similarity calculated 71; Be used for from all minutiae point similarities of test detail point similarity calculated 71 outputs; It is right as aiming at initial point to fingerprint minutiae to choose the maximum N of similarity, and wherein N is the integer more than or equal to 1;

Test aligned units 73; Aiming at initial point with test is connected choosing unit 72; Be used for based on test aim at initial point to the N that chooses unit 72 and choose to aiming at initial point to respectively details in fingerprint dot information and direction of fingerprint field picture being carried out translation and rotation, accomplish N group aligning altogether;

Test detail is put average similarity score computing unit 74; Be connected with test aligned units 73; Be used for aiming at according to the N group of test aligned units 73 outputs; Describe on the basis at details in fingerprint dot information behind the aligning and minutiae point local direction, calculate N the average similarity score of minutiae point respectively; And

Measurement direction field average distance score calculating unit 75 is connected with test aligned units 73, is used for aiming at based on the N group of test aligned units 73 outputs, on the direction of fingerprint field picture basis behind the aligning, calculates N field of direction average distance mark respectively.

The formation synoptic diagram of the print scores emerging system mid-score integrated unit 8 based on Sigmoid expansion provided by the invention is as shown in Figure 4, comprising:

The average similarity score expanding element 81 of minutiae point; Be used to utilize the Sigmoid spreading parameter of the average similarity score of minutiae point of Sigmoid parameter training unit 4 outputs that N the average similarity score of minutiae point of characteristic aligned units 7 outputs carried out the Sigmoid expansion, obtain N the average similarity score of minutiae point after expanding;

Field of direction mean distance mark expanding element 82; Be used to utilize the Sigmoid spreading parameter of the field of direction mean distance mark of Sigmoid parameter training unit 4 outputs that N field of direction mean distance mark of characteristic aligned units 7 outputs carried out the Sigmoid expansion, obtain N the field of direction mean distance mark after expanding;

Sigmoid expansion mark integrated unit 83; Be connected with field of direction mean distance mark expanding element 82 with the average similarity score expanding element 81 of minutiae point; Be used for the field of direction mean distance mark after N the expansion of average similarity score of minutiae point after N the expansion of average similarity score expanding element 81 outputs of minutiae point and 82 outputs of field of direction mean distance mark expanding element is merged respectively, obtain N and merge mark; And

Final matching fraction output unit 84; Be connected with Sigmoid expansion mark integrated unit 83; Be used for N fusion mark of Sigmoid expansion mark integrated unit 83 outputs compared; Choose maximum fusion mark as final matching fraction, thereby judge that whether the test fingerprint image to deriving from same finger.

To print scores emerging system based on the Sigmoid expansion shown in Figure 4, Fig. 5 shows the method flow diagram that the print scores based on the Sigmoid expansion according to the embodiment of the invention merges based on Fig. 1, and this method comprises that parameter training and fingerprint test two stages; Solid line representes that fingerprint test phase, dotted line represent the parameter training stage among Fig. 5; When the enterprising line parameter of training fingerprint image image set is trained, train fingerprint images to handling to all according to the flow process shown in the dotted line; When to the test fingerprint image when testing, according to the flow performing shown in the solid line; The parameter training stage carried out before the fingerprint test phase; Comprise that specifically step is following:

Step S1: fingerprint image is carried out pre-service, and this pre-service comprises respectively extracts image segmentation, figure image intensifying, image binaryzation, micronization processes and the field of direction of fingerprint image, obtains refinement fingerprint image and direction of fingerprint field picture; The concrete detailed step of this pre-service has:

Step S11: the equalization of gray scale, to eliminate the difference of contrast between the different images;

Step S12: use the LPF algorithm to eliminate speckle noise and Gaussian noise;

Step S13: the estimation of the field of direction calculates the direction of each pixel of fingerprint image;

Step S14: utilize the field of direction consistance fingerprint image to be divided into foreground area and background area with the average and the variance of image;

Step S15: binaryzation, come fingerprint image is treated to the image that has only two kinds of pixels of black and white according to the direction of each pixel;

Step S16: refinement according to binary image, to having only a pixel, generates fingerprint thinning figure with the crestal line width reduction of fingerprint;

Step S17: the refinement aftertreatment, remove the bad crestal line structure in the refined image, said bad crestal line structure comprises the burr on tangible bridge, the crestal line, too short crestal line and single spot between tangible broken string, crestal line at least.

Step S2: the minutiae point that takes the fingerprint information and minutiae point local direction are described.The minutiae point that in the refinement fingerprint image, takes the fingerprint information comprises (x, the y) coordinate on the direction, the direction and the type of each fingerprint minutiae; In the direction of fingerprint field picture, each fingerprint minutiae being extracted the minutiae point local direction respectively then describes.The concrete grammar that said extraction minutiae point local direction is described is: around minutiae point, getting L radius is rl, and (concentric circles of 0≤l≤L) is got K then on each is justified _lIndividual sampled point α _{K, l}, (0≤k≤K _l) with the circumference five equilibrium; So just describing, the minutiae point local direction can enough sampling point set close a={ α _{K, l}Form represent α wherein _{K, l}The direction of representing l round last k sample point; Here, l, k, L, K _lBe nonnegative integer.

Step S3: describe the similarity between the calculated fingerprint minutiae point according to the minutiae point local direction.

The concrete grammar of the similarity between the said calculated fingerprint minutiae point is: for come respectively the self-training fingerprint image to or two fingerprint images of test fingerprint image pair on two minutiae point, use a={ α _{K, l}And b={ β _{K, l}Represent that respectively the local direction of these two minutiae point describes, wherein, α _{K, l}And β _{K, l}The direction of k sample point during the local direction of representing two minutiae point is respectively described on l concentric circles, then the similarity S of a and b _AbComputing method following:

S _ab＝mean(s(Λ(α _k，l，β _k，l))) (2)

Wherein, Λ (α _{K, l}, β _{K, l}) be α _{K, l}And β _{K, l}Between the direction difference, s (Λ (α _{K, l}, β _{K, l})) be about Λ (α _{K, l}, β _{K, l}) similarity function, represent the variable of similarity function s (x) with x, then similarity function s (x) can be expressed as s (x)=e ^{-x/ σ}, σ is taken as π/16.

Step S4: in the parameter training stage, choose one group of maximum fingerprint minutiae of minutiae point similarity, minutiae point information and direction of fingerprint field picture are aimed to right as initial point.At first calculate this group fingerprint minutiae between the translation rotation parameter, confirm the corresponding relation between the fingerprint minutiae according to the translation rotation parameter then, confirm the aligned relationship between the direction of fingerprint field according to the translation rotation parameter at last;

Step S5: in the parameter training stage, according to alignment result, the computational details is put average similarity score and field of direction mean distance mark.After aiming at, the right minutiae point similarity averaged of all minutiae point of corresponding relation is arranged, thereby obtain the average similarity score of minutiae point.For the direction of fingerprint field after aiming at, calculate the absolute value of the field of direction difference of all corresponding point, then all absolute difference are asked on average, thereby obtained field of direction mean distance mark.

Step S6: in the parameter training stage; Average similarity score of minutiae point and field of direction mean distance mark are trained respectively; The computational details is put the Sigmoid spreading parameter of average similarity score and the Sigmoid spreading parameter of field of direction mean distance mark, so far accomplishes the parameter training stage.True matching sequence and the false matching sequence of representing mark j with

and

respectively; Wherein G and I represent true matching times and the false matching times that the parameter training process is carried out altogether respectively, and g and i represent the g time true coupling and the i time false coupling respectively; J=1,2, during j=1, the average similarity score of expression minutiae point; During j=2, expression field of direction mean distance mark; Score (j, g) and Score (j i) is non-negative real number, and G and I are the integers greater than 0; For mark j, the step of calculating the Sigmoid spreading parameter is following:

Step S61: calculate Sigmoid spreading parameter Max (j) and Min (j), and mark j is carried out normalization; Seek maximal value Max (j) and minimum M in (j) in

and

, the method below utilizing is then carried out normalization:

$\stackrel{\‾}{\mathrm{Score}{(j,g)}_{g=1}^G}=(\mathrm{Score}{(j,g)}_{g=1}^G-\mathrm{Min}\left(j\right))/(\mathrm{Max}\left(j\right)-\mathrm{Min}\left(j\right))$ (2)

$\stackrel{\‾}{\mathrm{Score}{(j,i)}_{i=1}^I}=(\mathrm{Score}{(j,i)}_{i=1}^I-\mathrm{Min}\left(j\right))/(\mathrm{Max}\left(j\right)-\mathrm{Min}\left(j\right))$

Wherein,

With

Represent true matching sequence and false matching sequence after the normalization respectively; If j=2 also will carry out negate and calculate: $\stackrel{\‾}{\mathrm{Score}{(j,g)}_{g=1}^G}=1-\stackrel{\‾}{\mathrm{Score}{(j,g)}_{g=1}^G},$ $\stackrel{\‾}{\mathrm{Score}{(j,i)}_{i=1}^I}=1-\stackrel{\‾}{\mathrm{Score}{(j,i)}_{i=1}^I}.$

Step S62: erroneous matching rate FMR sequence and the mistake of the calculating mark j rate FNMR sequence that do not match; Get threshold value Thr={0,1,2 ..., 1000}/1000, for each Thr (s), s=1,2,3 ..., 1001, calculate

Times N um _M(j, s) with

Times N um _N(j s), thereby obtains the erroneous matching rate FMR sequence of mark j ${\{\mathrm{Fmr}(j,s)={\mathrm{Num}}_M(j,s)/I\}}_{s=1}^{1001}$ With the mistake rate FNMR sequence that do not match ${\{\mathrm{Fnmr}(j,s)={\mathrm{Num}}_N(j,s)/G\}}_{s=1}^{1001}.$

Step S63: seek unique point s _{μ 1}(j) and s _{μ 2}(j); Comparison error matching rate FMR sequence With the mistake rate FNMR sequence that do not match

Searching Fmr (j, s)>=Fnmr (j, the s value s of maximum s) _{μ 1}(j) and Fmr (j, s)≤Fnmr (j, the s value s of minimum s) _{μ 2}(j); If seek failure, then s _{μ 1}(j)=0, s _{μ 2}(j)=1001.

Step S64: seek unique point s _l(j) and s _h(j); In erroneous matching rate FMR sequence

In, seek Fmr (j, the s value s of minimum s)＜0.01 _h(j), if do not have Fmr (j, s)＜0.01, s then _h(j)=1001; In mistake does not match rate FNMR sequence, seek Fnmr (j, the s value s of maximum s)＜0.01 _l(j), if do not have Fnmr (j, s)＜0.01, s then _l(j)=0.

Step S65: calculate Sigmoid spreading parameter μ (j), δ _l(j) and δ _h(j).μ (j), δ _l(j) and δ _h(j) computing method are following:

μ(j)＝(s _μ1(j)+s _μ2(j))/1001/2；

δ _l(j)＝(μ(j)-s _l(j)/1001)/2； (3)

δ _h(j)＝(s _h(j)/1001-μ(j))/2.

Step S7: at the fingerprint test phase,, choose N maximum fingerprint minutiae of minutiae point similarity, minutiae point information and direction of fingerprint field picture are aimed to right as aiming at initial point according to the fingerprint minutiae similarity that step S3 obtains.Wherein N is the integer more than or equal to 1, and N is taken as 10 in the experiment.Calculate each group aim at initial point between the translation rotation parameter, respectively details in fingerprint dot information and direction of fingerprint field picture are aimed at according to each group translation rotation parameter then, obtain N altogether and organize alignment result;

Step S8: at the fingerprint test phase, according to each group alignment result, the computational details is put average similarity score and field of direction mean distance mark respectively.After aiming at, the right minutiae point similarity averaged of all minutiae point of corresponding relation is arranged, thereby obtain the average similarity score of minutiae point; For the direction of fingerprint field after aiming at, calculate the absolute value of the field of direction difference of all corresponding point, then all absolute difference are asked on average, thereby obtained field of direction mean distance mark; N organizes alignment result, obtains N group average similarity score of minutiae point and field of direction mean distance mark altogether.

Step S9: N group average similarity score of minutiae point and field of direction mean distance mark are carried out the Sigmoid expansion respectively, obtain average similarity score of minutiae point and field of direction mean distance mark after the N group is expanded.The expression N number of components with

; Wherein, Scr (j; N) be non-negative real number, represent the n number of components; J=1,2, represent the average similarity score of minutiae point during j=1, represent the field of direction mean distance mark during j=2, at first mark j is carried out normalization through following method:

$\stackrel{\‾}{\mathrm{Scr}(j,n)}=(\mathrm{Scr}(j,n)-\mathrm{Min}\left(j\right))/(\mathrm{Max}\left(j\right)-\mathrm{Min}\left(j\right))---\left(4\right)$

Wherein,

expression is to Scr (j; N) carry out result after the normalization; Be after the normalization Scr (j, n), Min (j) and Max (j) are that the Sigmoid of mark j of the step S6 output in parameter training stage expands. open up parameter.If j 2, also will carry out negate and calculate:

carries out the Sigmoid expansion through following method to mark j then:

$E_{\mathrm{Sig}}=\left\{\begin{array}{cc}\frac{\mathrm{\μ}\left(j\right)-\stackrel{\‾}{\mathrm{Scr}(j,n)}}{{\mathrm{\δ}}_l\left(j\right)},& \stackrel{\‾}{\mathrm{Scr}(j,n)}\≤\mathrm{\μ}\left(j\right);\\ \frac{\mathrm{\μ}\left(j\right)-\stackrel{\‾}{\mathrm{Scr}(j,n)}}{{\mathrm{\δ}}_h\left(j\right)},& \stackrel{\‾}{\mathrm{Scr}(j,n)}>\mathrm{\μ}\left(j\right).\end{array}\right.---\left(5\right)$

$\mathrm{Sig}\stackrel{\‾}{\left(\mathrm{Scr}(j,n)\right)}=\frac{1}{1+e^{E_{\mathrm{Sig}}}}$

Wherein, It is right to represent

Carry out the result after Sigmoid expands, i.e. after the Sigmoid expansion

E _SigIt is intermediate variable; μ (j), δ _l(j) and δ _h(j) be the Sigmoid spreading parameter of the mark j of parameter training stage step S6 output.

Step S10: average similarity score of minutiae point and field of direction mean distance mark to after each group expansion merge.Fusion method adopts addition, and the field of direction mean distance mark addition after the average similarity score of minutiae point after N expansion is expanded with N respectively obtains N matching fractional.

Step S11: choose in N the matching fractional maximum matching fractional as final matching fraction, judge the test fingerprint image to whether deriving from same finger according to the size of final matching fraction, thus completion fingerprint test phase.

In specific embodiment, we are applied to us with the method and design voluntarily in the fingerprint image processing system of realization.The fingerprint image processing system that we develop is based on Window XP, adopts Object Oriented method and soft project standard, Flame Image Process and analytic system that realize with C Plus Plus, object fingerprint identification field.

In order to verify the overall performance that incorporates among the present invention based on the print scores blending algorithm (to call algorithm Sigmoid in the following text) of Sigmoid expansion; We have realized three contrast algorithms: the independent respectively details of use of algorithm priSmi and algorithm oriDiff are put average similarity score and field of direction mean distance mark; Not doing mark merges; Algorithm Mean merges two marks without the Sigmoid expansion, and other processing is identical with algorithm Sigmoid.

Four algorithms are applied to respectively in the aforementioned fingerprint image processing system, and on FVC2002DB1 and FVC2004DB1 fingerprint base, test respectively.Fig. 6 and Fig. 7 show the ROC curve contrast of four algorithms experimental result on FVC2002DB1 and FVC2004DB1 fingerprint base respectively.Contrast and experiment can find out that the performance of algorithm Sigmoid obviously is superior to other algorithm, and this makes the overall performance of fingerprint recognition system be improved.

The Sigmoid expansion is simple calculating just, only needs seldom additional calculations resource and time loss, and therefore real-time and the practicality to algorithm can not have a significant effect.

Above-described specific embodiment; The object of the invention, technical scheme and beneficial effect have been carried out further explain, and institute it should be understood that the above is merely specific embodiment of the present invention; Be not limited to the present invention; All within spirit of the present invention and principle, any modification of being made, be equal to replacement, improvement etc., all should be included within protection scope of the present invention.

Claims (24)

1. the print scores emerging system based on the Sigmoid expansion is characterized in that, comprising:

Training image pretreatment unit (1), each the training fingerprint image that is used for the training finger print data is concentrated obtains right refinement fingerprint image and the direction of fingerprint field picture of training fingerprint image respectively to carrying out pre-service respectively;

Training characteristics extraction unit (2); Be connected with training image pretreatment unit (1); Be used for from the right refinement fingerprint image of training fingerprint image of training image pretreatment unit (1) the output minutiae point information that takes the fingerprint, and from the right direction of fingerprint field picture of the training fingerprint image of training image pretreatment unit (1) output, extract the description of minutiae point local direction for each fingerprint minutiae;

Training characteristics aligned units (3); Be connected with training characteristics extraction unit (2); Be used for the right details in fingerprint dot information of the training fingerprint image of training characteristics extraction unit (2) output, the description of minutiae point local direction and direction of fingerprint field picture are calculated, obtain training average similarity score of the right minutiae point of fingerprint image and field of direction mean distance mark;

Sigmoid parameter training unit (4); Be connected with training characteristics aligned units (3); Be used for the right average similarity score of minutiae point and the field of direction mean distance mark of training fingerprint image of training characteristics aligned units (3) output calculated, obtain the Sigmoid spreading parameter of average similarity score of minutiae point and field of direction mean distance mark respectively;

Test pattern pretreatment unit (5) is used for the test fingerprint image obtaining right refinement fingerprint image of test fingerprint image and direction of fingerprint field picture respectively to doing pre-service respectively;

Test feature extraction unit (6); Be connected with test pattern pretreatment unit (5); Be used for from the right refinement fingerprint image of test fingerprint image of test pattern pretreatment unit (5) the output minutiae point information that takes the fingerprint, and from the right direction of fingerprint field picture of the test fingerprint image of test pattern pretreatment unit (5) output, extract the description of minutiae point local direction for each fingerprint minutiae;

Test feature aligned units (7); Be connected with test feature extraction unit (6); Be used for the right details in fingerprint dot information of the test fingerprint image of test feature extraction unit (6) output, the description of minutiae point local direction and direction of fingerprint field picture are calculated, obtain average similarity score of the right minutiae point of test fingerprint image and field of direction mean distance mark; And

Mark integrated unit (8); Be connected with test feature aligned units (7) and Sigmoid parameter training unit (4); Be used to utilize the Sigmoid spreading parameter of the average similarity score of minutiae point of Sigmoid parameter training unit (4) output that the right average similarity score of minutiae point of test fingerprint image of test feature aligned units (7) output is carried out the Sigmoid expansion, and utilize the Sigmoid spreading parameter of the field of direction mean distance mark of Sigmoid parameter training unit (4) output that the right field of direction mean distance mark of test fingerprint image of test feature aligned units (7) output is carried out the Sigmoid expansion; Two marks after will expanding then merge, and obtain final matching fraction, thereby judge that whether the test fingerprint image to deriving from same finger.

2. the print scores emerging system based on the Sigmoid expansion according to claim 1 is characterized in that said training characteristics aligned units (3) comprising:

Training minutiae point similarity calculated (31) is used for describing the similarity between all right fingerprint minutiaes of calculation training fingerprint image according to the right details in fingerprint dot information and the minutiae point local direction of training fingerprint image of training characteristics extraction unit (2) output;

Training is aimed at initial point to choosing unit (32); Be connected with training minutiae point similarity calculated (31); Be used for from all minutiae point similarities of training minutiae point similarity calculated (31) output, it is right as aiming at initial point to choose the maximum a pair of fingerprint minutiae of similarity;

Training aligned units (33); Aiming at initial point with training is connected choosing unit (32); Be used for based on training aim at initial point to the aligning initial point choosing unit (32) and choose to details in fingerprint dot information and direction of fingerprint field picture are carried out translation and rotation, to accomplish aligning;

The training average similarity score computing unit of minutiae point (34) is connected with training aligned units (33), and the details in fingerprint dot information and the minutiae point local direction that are used for after training aligned units (33) is aimed at are described the basis, and the computational details is put average similarity score; And

Training field of direction mean distance score calculating unit (35) is connected with training aligned units (33), is used for the direction of fingerprint field picture basis after training aligned units (33) is aimed at, calculated direction field mean distance mark.

3. the print scores emerging system based on the Sigmoid expansion according to claim 1 is characterized in that said test feature aligned units (7) comprising:

Test detail point similarity calculated (71) is used for describing the similarity of calculating between all right minutiae point of test fingerprint image according to the test fingerprint image of test feature extraction unit (6) output right details in fingerprint dot information and minutiae point local direction;

Test is aimed at initial point to choosing unit (72); Be connected with test detail point similarity calculated (71); Be used for from all minutiae point similarities of test detail point similarity calculated (71) output; It is right as aiming at initial point to fingerprint minutiae to choose the maximum N of similarity, and wherein N is the integer more than or equal to 1;

Test aligned units (73); Aiming at initial point with test is connected choosing unit (72); Be used for based on test aim at initial point to the N that chooses unit (72) and choose to aiming at initial point to respectively details in fingerprint dot information and direction of fingerprint field picture being carried out translation and rotation, accomplish N group aligning altogether;

Test detail is put average similarity score computing unit (74); Be connected with test aligned units (73); Be used for aiming at according to the N group of test aligned units (73) output; Describe on the basis at details in fingerprint dot information behind the aligning and minutiae point local direction, calculate N the average similarity score of minutiae point respectively; And

Measurement direction field mean distance score calculating unit (75); Be connected with test aligned units (73); Be used for aiming at, on the direction of fingerprint field picture basis behind the aligning, calculate N field of direction mean distance mark respectively according to the N group of test aligned units (73) output.

4. the print scores emerging system based on the Sigmoid expansion according to claim 1 is characterized in that said mark integrated unit (8) comprising:

The average similarity score expanding element of minutiae point (81); Be used to utilize the Sigmoid spreading parameter of the average similarity score of minutiae point of Sigmoid parameter training unit (4) output that N the average similarity score of minutiae point of characteristic aligned units (7) output carried out the Sigmoid expansion, obtain N the average similarity score of minutiae point after expanding;

Field of direction mean distance mark expanding element (82); Be used to utilize the Sigmoid spreading parameter of the field of direction mean distance mark of Sigmoid parameter training unit (4) output that N field of direction mean distance mark of characteristic aligned units (7) output carried out the Sigmoid expansion, obtain N the field of direction mean distance mark after expanding;

Sigmoid expands mark integrated unit (83); Be connected with field of direction mean distance mark expanding element (82) with the average similarity score expanding element of minutiae point (81); Be used for the field of direction mean distance mark after N the expansion of average similarity score of minutiae point after N the expansion of the average similarity score expanding element of minutiae point (81) output and field of direction mean distance mark expanding element (82) output is merged respectively, obtain N and merge mark; And

Final matching fraction output unit (84); Be connected with Sigmoid expansion mark integrated unit (83); Be used for N fusion mark of Sigmoid expansion mark integrated unit (83) output compared; Choose maximum fusion mark as final matching fraction, thereby judge that whether the test fingerprint image to deriving from same finger.

5. the print scores emerging system based on the Sigmoid expansion according to claim 1 is characterized in that,

Said training image pretreatment unit (1) trains fingerprint image to carrying out pre-service respectively to concentrated each of training finger print data, and this pre-service comprises image segmentation, figure image intensifying, image binaryzation, micronization processes and field of direction extraction;

Said training characteristics extraction unit (2) the minutiae point information that from the right refinement fingerprint image of the training fingerprint image of training image pretreatment unit (1) output, takes the fingerprint, this fingerprint minutiae information comprises position, direction and the type of minutiae point;

To doing pre-service respectively, this pre-service comprises image segmentation, figure image intensifying, image binaryzation, micronization processes and field of direction extraction to said test pattern pretreatment unit (5) to the test fingerprint image;

Said test feature extraction unit (6) the minutiae point information that from the right refinement fingerprint image of the test fingerprint image of test pattern pretreatment unit (5) output, takes the fingerprint, this fingerprint minutiae information comprises position, direction and the type of minutiae point.

6. print scores fusion method based on Sigmoid expansion; It is characterized in that; This method at first trains the Sigmoid spreading parameter of average similarity score of minutiae point and field of direction mean distance mark on training fingerprint image image set; Then the test fingerprint image is also aimed at extracting characteristic, on the information behind the aligning, extracted average similarity score of minutiae point and field of direction mean distance mark, and respectively each mark is carried out the Sigmoid expansion; Average similarity score of minutiae point and field of direction mean distance mark to after the Sigmoid expansion merge; The result who obtains after the fusion is as final matching fractional; Judge according to the size of final matching fraction whether input fingerprint characteristic and template fingerprint characteristic derive from same finger, thereby accomplish the coupling of fingerprint.

7. the print scores fusion method based on the Sigmoid expansion according to claim 6 is characterized in that, specifically may further comprise the steps:

Step S1: fingerprint image is carried out pre-service, obtain refinement fingerprint image and direction of fingerprint field picture;

Step S2: the minutiae point that takes the fingerprint information and minutiae point local direction are described;

Step S3: describe the similarity between the calculated fingerprint minutiae point according to the minutiae point local direction;

Step S4: in the parameter training stage, choose one group of maximum fingerprint minutiae of minutiae point similarity, minutiae point information and direction of fingerprint field picture are aimed to right as initial point;

Step S5: in the parameter training stage, according to alignment result, the computational details is put average similarity score and field of direction mean distance mark;

Step S6: in the parameter training stage, average similarity score of minutiae point and field of direction mean distance mark are trained respectively, the computational details is put the Sigmoid spreading parameter of average similarity score and the Sigmoid spreading parameter of field of direction mean distance mark;

Step S7: at the fingerprint test phase,, choose N maximum fingerprint minutiae of minutiae point similarity, minutiae point information and direction of fingerprint field picture are aimed to right as aiming at initial point according to the fingerprint minutiae similarity that step S3 obtains;

Step S8: at the fingerprint test phase, according to each group alignment result, the computational details is put average similarity score and field of direction mean distance mark respectively;

Step S9: N group average similarity score of minutiae point and field of direction mean distance mark are carried out the Sigmoid expansion respectively, obtain average similarity score of minutiae point and field of direction mean distance mark after the N group is expanded;

Step S10: average similarity score of minutiae point and field of direction mean distance mark to after each group expansion merge;

Step S11: choose in N the matching fractional maximum matching fractional as final matching fraction, judge that according to the size of final matching fraction whether the test fingerprint image to deriving from same finger.

8. the print scores fusion method based on the Sigmoid expansion according to claim 7; It is characterized in that; Described in the step S1 fingerprint image is carried out pre-service; Be respectively image segmentation, figure image intensifying, image binaryzation, micronization processes and the field of direction of fingerprint image to be extracted, specifically comprise:

Step S11: the equalization of gray scale, to eliminate the difference of contrast between the different images;

Step S12: use the LPF algorithm to eliminate speckle noise and Gaussian noise;

Step S13: the estimation of the field of direction calculates the direction of each pixel of fingerprint image;

Step S14: utilize the field of direction consistance fingerprint image to be divided into foreground area and background area with the average and the variance of image;

Step S15: binaryzation, come fingerprint image is treated to the image that has only two kinds of pixels of black and white according to the direction of each pixel;

Step S16: refinement according to binary image, to having only a pixel, generates fingerprint thinning figure with the crestal line width reduction of fingerprint;

Step S17: the refinement aftertreatment, remove the bad crestal line structure in the refined image.

9. the print scores fusion method based on the Sigmoid expansion according to claim 8; It is characterized in that bad crestal line structure described in the step S17 comprises the burr on tangible bridge, the crestal line, too short crestal line and single spot between tangible broken string, crestal line at least.

10. the print scores fusion method based on the Sigmoid expansion according to claim 7; It is characterized in that; The minutiae point that takes the fingerprint described in step S2 information and minutiae point local direction are described, and are the minutiae point information that in the refinement fingerprint image, takes the fingerprint, and comprise (the x of each fingerprint minutiae; Y) coordinate on the direction, direction and type; In the direction of fingerprint field picture, each fingerprint minutiae being extracted the minutiae point local direction respectively then describes.

11. the print scores fusion method based on the Sigmoid expansion according to claim 10; It is characterized in that; The concrete grammar that said extraction minutiae point local direction is described is: around minutiae point, getting L radius is rl, and (concentric circles of 0≤l≤L) is got K then on each is justified _lIndividual sampled point α _{K, l}, (0≤k≤K _l) with the circumference five equilibrium; So just describing, the minutiae point local direction can enough sampling point set close a={ α _{K, l}Form represent α wherein _{K, l}The direction of representing l round last k sample point; Here, l, k, L, K _lBe nonnegative integer.

12. the print scores fusion method based on the Sigmoid expansion according to claim 7 is characterized in that, describes the similarity between the calculated fingerprint minutiae point according to the minutiae point local direction described in the step S3, comprising:

For come respectively the self-training fingerprint image to or two fingerprint images of test fingerprint image pair on two minutiae point, use a={ α _{K, l}And b=(β _{K, l}Represent that respectively the local direction of these two minutiae point describes, wherein, α _{K, l}And β _{K, l}The direction of k sample point during the local direction of representing two minutiae point is respectively described on l concentric circles, then the similarity S of a and b _AbComputing method following:

S _ab＝mean(s(Λ(α _k，l，β _k，l))) (1)

Wherein, Λ (α _{K, l}, β _{K, l}) be α _{K, l}And β _{K, l}Between the direction difference, s (Λ (α _{K, l}, β _{K, l})) be about Λ (α _{K, l}, β _{K, l}) similarity function, represent the variable of similarity function s (x) with x, then similarity function s (x) can be expressed as s (x)=e ^{-x/ σ}, σ is taken as π/16.

13. the print scores fusion method based on the Sigmoid expansion according to claim 7 is characterized in that, described in the step S4 minutiae point information and direction of fingerprint field picture is aimed at, and comprising:

At first computational details dot information and direction of fingerprint field picture this group fingerprint minutiae between the translation rotation parameter; Confirm the corresponding relation between the fingerprint minutiae according to the translation rotation parameter then, confirm the aligned relationship between the direction of fingerprint field according to the translation rotation parameter at last.

14. the print scores fusion method based on the Sigmoid expansion according to claim 7 is characterized in that, puts average similarity score and field of direction mean distance mark according to the alignment result computational details described in the step S5, comprising:

After aiming at, the right minutiae point similarity averaged of all minutiae point of corresponding relation is arranged, thereby obtain the average similarity score of minutiae point;

For the direction of fingerprint field after aiming at, calculate the absolute value of the field of direction difference of all corresponding point, then all absolute difference are asked on average, thereby obtained field of direction mean distance mark.

15. the print scores fusion method based on the Sigmoid expansion according to claim 7; It is characterized in that; Described in the step S6 average similarity score of minutiae point and field of direction mean distance mark are trained respectively; The computational details is put the Sigmoid spreading parameter of average similarity score and the Sigmoid spreading parameter of field of direction mean distance mark; Be true matching sequence and the false matching sequence of representing mark j with

and

respectively; Wherein G and I represent true matching times and the false matching times that the parameter training process is carried out altogether respectively, and g and i represent the g time true coupling and the i time false coupling respectively; J=1,2, during j=1, the average similarity score of expression minutiae point; During j=2, expression field of direction mean distance mark; Score (j, g) and Score (j i) is non-negative real number, and G and I are the integers greater than 0; For mark j, the step of calculating the Sigmoid spreading parameter is following:

Step S61: calculate Sigmoid spreading parameter Max (j) and Min (j), and mark j is carried out normalization;

Step S62: erroneous matching rate FMR sequence and the mistake of the calculating mark j rate FNMR sequence that do not match;

Step S63: seek unique point s _{μ 1}(j) and s _{μ 2}(j);

Step S64: seek unique point s _l(j) and s _h(j);

Step S65: calculate Sigmoid spreading parameter μ (j), δ _l(j) and δ _h(j).

16. the print scores fusion method based on the Sigmoid expansion according to claim 15 is characterized in that, calculates Sigmoid spreading parameter Max (j) and Min (j) described in the step S61, and mark j is carried out normalization, comprising:

Seek maximal value Max (j) and minimum M in (j) in

and

, the method below utilizing is then carried out normalization:

$\stackrel{\‾}{\mathrm{Score}{(j,g)}_{g=1}^G}=(\mathrm{Score}{(j,g)}_{g=1}^G-\mathrm{Min}\left(j\right))/(\mathrm{Max}\left(j\right)-\mathrm{Min}\left(j\right))$

$\stackrel{\‾}{\mathrm{Score}{(j,i)}_{i=1}^I}=(\mathrm{Score}{(j,i)}_{i=1}^I-\mathrm{Min}\left(j\right))/(\mathrm{Max}\left(j\right)-\mathrm{Min}\left(j\right))$

Wherein,

With Represent true matching sequence and false matching sequence after the normalization respectively; If j=2 also will carry out negate and calculate: $\stackrel{\‾}{\mathrm{Score}{(j,g)}_{g=1}^G}=1-\stackrel{\‾}{\mathrm{Score}{(j,g)}_{g=1}^G},$ $\stackrel{\‾}{\mathrm{Score}{(j,i)}_{i=1}^I}=1-\stackrel{\‾}{\mathrm{Score}{(j,i)}_{i=1}^I}.$

17. the print scores fusion method based on Sigmoid expansion according to claim 15 is characterized in that, erroneous matching rate FMR sequence and the mistake of calculating mark j described in the step S62 rate FNMR sequence that do not match comprises:

Get threshold value Thr={0,1,2 ..., 1000}/1000, for each Thr (s), s=1,2,3 ..., 1001, calculate

Times N um _M(j, s) with

Times N um _N(j s), thereby obtains the erroneous matching rate FMR sequence of mark j ${\{\mathrm{Fmr}(j,s)={\mathrm{Num}}_M(j,s)/I\}}_{s=1}^{1001}$ With the mistake rate FNMR sequence that do not match ${\{\mathrm{Fnmr}(j,s)={\mathrm{Num}}_N(j,s)/G\}}_{s=1}^{1001}.$

18. the print scores fusion method based on the Sigmoid expansion according to claim 15 is characterized in that, seeks unique point s described in the step S63 _{μ 1}(j) and s _{μ 2}(j), comprising:

Comparison error matching rate FMR sequence

With the mistake rate FNMR sequence that do not match Searching Fmr (j, s)>=Fnmr (j, the s value s of maximum s) _{μ 1}(j) and Fmr (j, s)≤Fnmr (j, the s value s of minimum s) _{μ 2}(j); If seek failure, then s _{μ 1}(j)=0, s _{μ 2}(j)=1001.

19. the print scores fusion method based on the Sigmoid expansion according to claim 15 is characterized in that, seeks unique point s described in the step S64 _l(j) and s _h(j), comprising:

In erroneous matching rate FMR sequence

In, seek Fmr (j, the s value s of minimum s)＜0.01 _h(j), if do not have Fmr (j, s)＜0.01, s then _h(j)=1001; In mistake does not match rate FNMR sequence, seek Fnmr (j, the s value s of maximum s)＜0.01 _l(j), if do not have Fnmr (j, s)＜0.01, s then _l(j)=0.

20. the print scores fusion method based on the Sigmoid expansion according to claim 15 is characterized in that, calculates Sigmoid spreading parameter μ (j), δ described in the step S65 _l(j) and δ _h(j), adopt computing formula following: $\begin{array}{c}\mathrm{\μ}\left(j\right)=\left(s_{\mathrm{\μ}1}\left(j\right){+s}_{\mathrm{\μ}2}\left(j\right)\right)/1001/2;\\ {\mathrm{\δ}}_l\left(j\right)=(\mathrm{\μ}\left(j\right)-s_l\left(j\right)/1001)/2;\\ {\mathrm{\δ}}_h\left(j\right)=(s_h\left(j\right)/1001-\mathrm{\μ}\left(j\right))/2.\end{array}.$

21. the print scores fusion method based on the Sigmoid expansion according to claim 7; It is characterized in that; Described in the step S7 minutiae point information and direction of fingerprint field picture are aimed at; Be calculate each group aim at initial point between the translation rotation parameter, respectively details in fingerprint dot information and direction of fingerprint field picture are aimed at according to each group translation rotation parameter then, obtain N altogether and organize alignment result.

22. the print scores fusion method based on the Sigmoid expansion according to claim 7 is characterized in that the computational details described in the step S8 is put average similarity score and field of direction mean distance mark, comprising:

After aiming at, the right minutiae point similarity averaged of all minutiae point of corresponding relation is arranged, thereby obtain the average similarity score of minutiae point; For the direction of fingerprint field after aiming at, calculate the absolute value of the field of direction difference of all corresponding point, then all absolute difference are asked on average, thereby obtained field of direction mean distance mark; N organizes alignment result, obtains N group average similarity score of minutiae point and field of direction mean distance mark altogether.

23. the print scores fusion method based on the Sigmoid expansion according to claim 7; It is characterized in that; Described in the step S9 N group average similarity score of minutiae point and field of direction mean distance mark are carried out the Sigmoid expansion respectively; Obtain average similarity score of minutiae point and field of direction mean distance mark after the N group is expanded, comprising:

The expression N number of components with

; Wherein, Scr (j; N) be non-negative real number, represent the n number of components; J=1,2, represent the average similarity score of minutiae point during j=1, represent the field of direction mean distance mark during j=2, at first mark j is carried out normalization through following method:

$\stackrel{\‾}{\mathrm{Scr}(j,n)}=(\mathrm{Scr}(j,n)-\mathrm{Min}\left(j\right))/(\mathrm{Max}\left(j\right)-\mathrm{Min}\left(j\right))$

Wherein,

expression is to Scr (j; N) carry out result after the normalization; Be after the normalization Scr (j, n), Min (j) and Max (j) they are the Sigmoid spreading parameters of the mark j that exports of the step S6 in parameter training stage;

If j=2 also will carry out negate and calculate:

carries out the Sigmoid expansion through following method to mark j then:

$E_{\mathrm{Sig}}=\left\{\begin{array}{cc}\frac{\mathrm{\μ}\left(j\right)-\stackrel{\‾}{\mathrm{Scr}(j,n)}}{{\mathrm{\δ}}_l\left(j\right)},& \stackrel{\‾}{\mathrm{Scr}(j,n)}\≤\mathrm{\μ}\left(j\right);\\ \frac{\mathrm{\μ}\left(j\right)-\stackrel{\‾}{\mathrm{Scr}(j,n)}}{{\mathrm{\δ}}_h\left(j\right)},& \stackrel{\‾}{\mathrm{Scr}(j,n)}>\mathrm{\μ}\left(j\right).\end{array}\right.$

$\mathrm{Sig}\stackrel{\‾}{\left(\mathrm{Scr}(j,n)\right)}=\frac{1}{1+e^{E_{\mathrm{Sig}}}}$

Wherein,

It is right to represent

Carry out the result after Sigmoid expands, i.e. after the Sigmoid expansion E _SigIt is intermediate variable; μ (j), δ _l(j) and δ _h(j) be the Sigmoid spreading parameter of the mark j of parameter training stage step S6 output.

24. the print scores fusion method based on the Sigmoid expansion according to claim 7; It is characterized in that; Described in the step S10 average similarity score of minutiae point and field of direction mean distance mark after each group expansion are merged; Fusion method adopts addition, and the field of direction mean distance mark addition after the average similarity score of minutiae point after N expansion is expanded with N respectively obtains N matching fractional.

Images