CN102045162A - Personal identification system of permittee with tri-modal biometric characteristic and control method thereof - Google Patents

Abstract

The invention relates to a personal identification system of permittee with tri-modal biometric characteristic and a control method thereof, wherein the system comprises four parts as follows: a terminal equipment hardware device which is used for personal identification, a terminal management software, a terminal service software and a central service software. The terminal equipment is mainly used for collecting personal credit card information, static fingerprint characteristic image, acoustical signal and face image of video; the terminal management software is mainly used for managing the terminal, registering fingerprints, registering faces, registering phonetic features, collecting credit card information and data analyzing and etc.; the terminal service software is mainly used for comparing fingerprints, voices and face characteristics and providing a personal identification result based on fused three modes; and the central service software is mainly used for accessing the data of fingerprints, voices and faces, providing search services to the terminal equipment and controlling the various service requests which are made by the terminal equipment. The personal identification system provided by the invention has the advantage that the high quality and high reliability personal identification service of permittee can be provided to the user.

Description

A kind of three mode biological characteristic holder identity identification system and control methods thereof

Technical field

The present invention relates to the biometrics identification technology field, be specifically related to a kind of three mode holder identity identification system and control methods thereof based on fingerprint, voice and people's face.

Background technology

Along with the accuracy that social safety and identity are differentiated and the raising day by day of reliability requirement, single biometrics identification technology can not satisfy the application demand of society, and then has hindered this field and used widely.Owing to can provide enough accurate and identification reliably without any a single living creature characteristic recognition system, therefore the appearance of multi-biological characteristic recognition system will be an effective optional strategy, such as voice and people's face, voice and fingerprint, perhaps comprehensively discern as the combination of the use voice, people's face and the fingerprint that propose in this project, can greatly improve the accuracy and the reliability of system, and the robustness under the complex environment.With regard to biometrics identification technology, its development trend will be progressively to carry out the transition to the stage that relies on multiple biological characteristic comprehensively to discern from the cognitive phase that relies on single-mode.

The identity recognizing technology research that the data fusion technology that develops rapidly in recent years provides solid theory, many in the world scholars extensively to be devoted to multi-biological characteristic for multi-biological characteristic identification.Roberto has just proposed to utilize a plurality of biological characteristics to carry out the method for personal identification in nineteen ninety-five, and he discerns sound classifier and facial image grader respectively with the weighted geometric mean and the network integration of hyper-base function, obtained better effects.Bigun etc. (1997) propose to utilize supervised learning and merge sound in conjunction with the method for Bayes theory and look like to carry out the identity discriminating with face, have reached very high accuracy.People such as Dieckmann adopt sound and moving these the two kinds of behavioral characteristics of information of lip and this static nature of facial image to merge identification, and they utilize simple ballot algorithm to judge whether the decision-making of single grader is consistent with other two graders.Verlinde equals proposition in 1997 and merges vocal print and visual signature with the K-NN method, has also obtained result preferably.People such as Jain proposed in 1998 the result of fingerprint and recognition of face is merged, and proved quantitatively that theoretically the multi-biological characteristic authentication system is with respect to the improve of single kind biological characteristic authentication system on implementation efficiency, the fusion that has proved multi-biological characteristic in 1999 theoretically can improve the authentication rate, and propose to determine that in 2000 the method for each user's special parameter merges the recognition result of fingerprint, face picture and hand shape.People such as Kittler have proposed the blending theory framework and it have been divided into three layers, have compared addition criterion and the pluses and minuses of multiplication criterion scheduling algorithm in fusion simultaneously.Maes etc. have realized the system of a combining with biological characteristic (fingerprint) and abiotic feature (password) for the first time.In addition, in a lot of open source literatures, also show many other multi-biological characteristic recognition methods and combining with biological characteristic and abiotic feature methods of carrying out identification, all obtained reasonable recognition effect.Though the multi-biological characteristic identity recognizing technology also is in the elementary step, but a lot of achievements in research also is applied in the practice by commercial company, wherein the most famous system is exactly the BioID system of DCSAG company, it has used moving three the basic biological characteristics of face picture, sound and lip to merge and has discerned individual identity, has also obtained reasonable comprehensive recognition result.

Domestic oneself be that a collection of R﹠D institution of representative sets foot in this research field through having with the Chinese Academy of Sciences, Qinghua China university etc., and obtained a large amount of achievements in research.Leaders' such as Institute of Automation, CAS pattern recognition Tan of National Key Laboratory car pusher, Wang Yunhong problem group has succeeded in developing by multiple human body biological characteristics such as iris, face picture and sound and has carried out the new technology that identity is differentiated.The multimode biometric identity authentication recognition system " TH-ID " of Tsing-Hua University's independent development comprises two large divisions's content: four based on people's face (TH-FaceID), person's handwriting (TH-writerlD), the subsystem of signature (TH-SignID) and the authentication of iris (TH-IrisID) biometric identity (comprising identification and checking) merges identity authorization system with the multi-modal biological characteristic that utilizes multiple biological characteristic, this system constructing based on people's face of unified database, person's handwriting, signature, the multimode biometric identity identification Verification System of four kinds of biological characteristics of iris, can carry out the selection of fusion mode, carry out various possible patterns and merge.

At present, identity identification system based on biological characteristic also mainly rests on this level of single creature feature, mature system based on multiple biological characteristic is actually rare, and the ripe application system that comprehensive utilization fingerprint characteristic information, phonetic feature information and face characteristic information come the identity card holder to be carried out the identity discriminating does not nearly all have disclosed report at home and abroad.

Summary of the invention

Technical problem to be solved by this invention is that fingerprint, voice and the face characteristic information that the comprehensive utilization terminal equipment collects comes the true identity of identity card holder is differentiated, and has proposed a kind of identity identification system and control method thereof based on three mode in view of the above.

The concrete scheme of technical solution problem of the present invention is: a kind of three mode biological characteristic holder identity identification systems are provided, comprise and be used for terminal equipment hardware unit, terminal management software module, Terminal Service software module and center service software module four major parts that identity is differentiated, it is characterized in that the terminal equipment hardware unit is mainly used in ID card information, fingerprint characteristic, voice and the face characteristic of gathering the individual; The terminal management software module mainly is responsible for terminal management, fingerprint registration, voice registration, the registration of people's face, captured identity card information and data analysis etc.; The Terminal Service software module mainly is responsible for fingerprint, sound and face characteristic are compared and provided based on the fusion identity identification result that merges; The center service software module mainly is responsible for access fingerprint, sound and face characteristic data, and the use of terminal equipment is controlled.

The corresponding control method that a kind of three mode biological characteristic holder identity identification systems also are provided, comprise and be used for terminal equipment hardware unit, terminal management software module, Terminal Service software module and center service software module four major parts that identity is differentiated, it is characterized in that described control method comprises step:

(1) terminal equipment is by the network communication interface connecting system and connect central server, after the legitimate verification of terminal equipment identity process, obtain corresponding service request mandate, refuse unwarranted terminal equipment access system to central server request characteristic;

(2) ID card information, finger print information, sound and the video human face information of collection holder are carried out preliminary treatments such as pattern classification and noise remove;

(3) above-mentioned steps (2) is carried out preliminary treatment more respectively and extracted corresponding feature through fingerprint, sound and people's face data after the preliminary treatment;

(4) Terminal Service software obtains corresponding fingerprint, sound and face characteristic masterplate data according to the identity card numbering retrieval local data base LDB that the identity card card reader collects;

(5) fingerprint, sound and face characteristic comparing module are according to the matching algorithm searching database of the current apolegamy of system, and the similarity score value that obtains compared in record at every turn, and its value is respectively v1, v2 and v3;

(6) obtain respectively again v1, v2 and v3 value being carried out decision level fusion after the coupling score value of fingerprint, sound and face characteristic, calculate and merge later matching result.

The invention has the beneficial effects as follows: by multiple single creature feature is merged discriminating to identity, can effectively prevent to cause because of certain biological characteristic loses efficacy that system differentiated the problem of conclusion mistake, thereby improve the reliability of availability, adaptability and the identification result of identity identification system greatly.But this identity identification system and control method thereof can also provide Zai Xian off-line, the extendible independent operating of terminal also can networking operation identity differentiate service, can make the user not need to drop under the situation of substantial contribution and equipment, obtain the holder authentication service of high-quality and high reliability.

Description of drawings

Fig. 1 is the topology diagram of system provided by the present invention.

Fig. 2 is functional module structure figure of the present invention.

Fig. 3 is a terminal equipment outline drawing of the present invention.

Fig. 4 is an independent operation mode structure chart of the present invention.

Fig. 5 is a networking operation model structure chart of the present invention.

Fig. 6 is the data flow diagram of authentication of the present invention.

Fig. 7 is the interface arrangement figure between each subsystem element in the system of the present invention.

Fig. 8 is a terminal device validity checking flow chart of the present invention.

Fig. 9 is that identity of the present invention is differentiated the processing detail flowchart.

Embodiment

The present invention will be further described in detail below in conjunction with accompanying drawing and specific embodiment:

Technical problem to be solved by this invention is how to fully utilize the fingerprint that terminal equipment collects, voice and face characteristic information come the true identity of identity card holder is differentiated, and make up a kind of identity identification system and control method thereof in view of the above based on three mode, this identity identification system overcomes the defective of prior art, Zai Xian off-line can be provided, terminal is extendible, but the identity that independent operating also can networking operation is differentiated service, can make the user not need to drop under the situation of substantial contribution and equipment, obtain the holder authentication service of high-quality and high reliability.

As Fig. 1, Fig. 2, Fig. 3, Fig. 4, Fig. 5, Fig. 6 and shown in Figure 7, technical problem proposed by the invention is to solve like this: a kind of holder identity identification system is provided, comprise the collection fingerprint, sound, the terminal equipment of people's face and ID card information (specifically seeing Fig. 3), terminal management software, Terminal Service software and center service software is totally four parts (specifically seeing Fig. 7), this solution thinking is characterized in that, terminal equipment is by the identity card card reader, the fingerprint instrument, microphone, parts such as video camera device are formed, and are mainly used to reading identity card information, fingerprint, sound and face characteristic signal; Terminal management software mainly is made up of functional modules such as system management, identity information acquisition, fingerprint registration, voice registration, the registration of people's face and simplation verifications, main also storage fingerprint characteristic, sound characteristic and the face characteristic of system parameters configuration, reading identity card information, registration of being responsible for, and to functions such as the simplation verification of holder identity and system's service data statistical analyses; Terminal Service software mainly is made up of communication module, collection apparatus module, signature verification module and decision-making Fusion Module etc., mainly is responsible for finishing functions such as collection fingerprint, sound and face characteristic data, searching database, checking fingerprint, sound and face characteristic and calculating decision level fusion result; Center service software mainly is made up of terminal Authority Verification, data access and interactive communication module, main be responsible between central server and terminal equipment network service, terminal equipment is inserted legitimate verification with service request and access fingerprint and functions such as face characteristic data and ID card information.

Wherein, terminal management software specifically constitutes (specifically seeing Fig. 2) by following functional module:

1. system management module: for registering legal terminal equipment information, authorizing, be provided with operational parameter value important in the system etc. for the use operating personnel of native system.

2. identity information typing module: manually add, delete, revise typing personnel identity essential information (pressing the identity card content displayed), and the EXCEL mode of identity information in enormous quantities imports.

2. fingerprint Registration Module: registrant's fingerprint characteristic information, and register information preserved into database (local data base or system database).

3. sound Registration Module: registrant's sound characteristic information, and the characteristic information of being registered preserved into database (local data base or system database).

5. people's face Registration Module: registrant's face characteristic information, and register information preserved into database (local data base or system database).

6. simplation verification module: behind input identity information in the management work station, dry run is based on the bimodal holder identity fusion identification system flow process of fingerprint and face characteristic.

7. data analysis module: the service data to whole system is added up, is analyzed, and excavates valuable potential information, differentiates record data or the like as system's running log record data and identity.

Terminal Service software specifically is made of following functional module:

1. interactive communication module: the main interactive communication module swap data of being responsible for " service for checking credentials center software ", with the legitimate verification of finishing terminal equipment, upload the biological attribute data that extracts, and receive the biological attribute data of downloading from central server.

2. data acquisition module: mainly finish based on the ID card information of identity card card reader gather automatically and manual typing, based on the fingerprint characteristic collection of fingerprint instrument with based on people's face video image acquisition of camera head etc.

3. fingerprint sound the face characteristic authentication module: be responsible for the fingerprint, sound and the people's face data that collect are carried out preliminary treatment, characteristic information extraction, the template characteristic of finishing and storing is carried out the 1:1 checking.

4. merge decision-making module: after people's face, sound and fingerprint characteristic mate the basis separately, carry out decision level fusion, differentiate according to the score value after merging whether the holder identity is consistent with accredited of institute.

Center service software specifically is made of following functional module:

1. interactive communication module: mainly be responsible for the equipment validity authorization information that receiving terminal apparatus sends, whether legal with the access request of determining current device as shown in Figure 8, and be responsible for receiving biological characteristic and the auxiliary data thereof that legal terminal is uploaded and downloaded.

2. data access module: be responsible for the terminal data that storage receives by communication module and (store system database into, SDB), and give communication module by the biological attribute data of ID card information searching terminal request and loopback and pass down to terminal equipment.

3. terminal equipment Authority Verification module: verify according to the information such as terminal equipment coding that receive whether the terminal equipment of institute's access service centring system is legal, and be the terminal distribution authority according to the checking result.

A kind of holder identity identification system and control method thereof based on fingerprint, sound and face characteristic is characterized in that, comprise following step (specifically seeing Fig. 6 and Fig. 9):

(1) terminal equipment is by the network communication interface connecting system and connect central server, after the legitimate verification of terminal equipment identity process, obtain corresponding service request mandate, refuse unwarranted terminal equipment access system to central server request characteristic (Fig. 8);

(2) ID card information, finger print information, sound and the video human face information of collection holder are carried out preliminary treatments such as pattern classification and noise remove;

(3) step (2) is carried out preliminary treatment more respectively and extracted corresponding feature through fingerprint, sound and people's face data after the preliminary treatment.

Wherein, the processing to fingerprint mainly comprises following step:

1. the fingerprint effective coverage is cut apart: fingerprint image is divided into the image block of a series of 16 * 16 non-intersections, and each piece is labeled as B (1,1) respectively, B (1,2) ..., B (i, j), utilize then following formula v (i, j)

$v(i,j)=\sqrt{\frac{{(x_1-\stackrel{\‾}{x})}^2+{(x_2-\stackrel{\‾}{x})}^2+...+{(x_n-\stackrel{\‾}{x})}^2}{N}}$

Calculate the grey scale pixel value variance of each image block, wherein x _nWith x represent respectively this figure fast in the gray value of pixel, N represents the pixel quantity that comprises in the segment segmentation threshold v to be set _θ=11.5, respectively with the variance yields and the v of each segment _θRelatively, if v (i, j)＞v _θ, then (i j) is judged as effective finger-print region to this segment B, otherwise is judged as the background area of fingerprint;

2. the field of direction of fingerprint is calculated: calculate respectively segment B (i, j) in each pixel gradient G in the x and y direction _xAnd G _y, utilize formula d (i, j)

$d(i,j)=\frac{1}{2}\arctan \left(\frac{\underset{i_b^{\′}=1}{\overset{w}{\mathrm{\Σ}}}\underset{j_b^{\′}=1}{\overset{w}{\mathrm{\Σ}}}2G_x(i_b^{\′},j_b^{\′})G_y(i_b^{\′},j_b^{\′})}{\underset{i_b^{\′}=1}{\overset{w}{\mathrm{\Σ}}}\underset{j_b^{\′}=1}{\overset{w}{\mathrm{\Σ}}}{G}_x^2(i_b^{\′},j_b^{\′})-G_y^2(i_b^{\′},j_b^{\′})}\right)$

Calculate respectively segment B (i, local direction j), in the formula (i ' _b, j ' _b) coordinate of pixel in the expression segment, w represents the pixel wide of segment, value like the segment local direction value that calculates is only got in four component values 0, π/4, π/3 and 3 π/4 recently, local direction d (the i that segment is final, j) ∈ { 0,4 π/4, π, 3 π/4}, by the consistency feature correction field of direction result of calculation of the field of direction, (i is in 5 * 5 neighborhood D scope j) at segment B, calculate its consistency value C (i, j)

$C(i,j)=\frac{1^{,}}{24}\sqrt{\underset{(i^{\′},j^{\′})\∈D}{\mathrm{\Σ}}{|d(i,j)-d(i^{\′},j^{\′})|}^2},$

Have in this formula Here d=mod (d (i ', j ')-d (i, j)+2 π)

If C (i, j)＜0.35, then with segment B (i, local direction j) are adjusted into the most significant direction of local direction in the neighborhood D, otherwise segment B (i, local direction j) remains unchanged;

3. use the M-PCNN network that the later image of preliminary treatment is carried out filtering;

4. fingerprint image is carried out feature extraction, comprises that global characteristics extracts and minutia is extracted:

Global characteristics is extracted as the Fourier spectrum feature: at first the fingerprint image with input is divided into 32 * 32 image block, and segment is done two dimensional discrete Fourier transform, and formula is

$G(m,n)=\frac{1}{N}\underset{i=0}{\overset{N-1}{\mathrm{\&Sigma;}}}\underset{K=0}{\overset{N-1}{\mathrm{\&Sigma;}}}\left(g(i,k)\exp (-\mathrm{<figure-callout\; id="2"\; label="j"\; filenames="H2009101679254E00041.png"\; state="\{\{state\}\}">j</figure-callout>}2\mathrm{\&pi;}(\frac{\mathrm{mi}}{N}+\frac{\mathrm{kn}}{N}))\right)$

In the formula, (m n) is the codomain coordinate of pixel, and (i k) is the frequency domain coordinate of pixel correspondence, and each subgraph piece has just obtained the fourier spectrum figure of full figure through after the Fourier transform, and it is quantized into the global characteristics vector of fingerprint by rule;

Minutia is extracted, and comprises core point, bifurcation and end points, and used minutia is extracted template respectively as following table:

P 3	P 2	P 1
			P 4	P?	P 0
P 5	P 6	P 7

Wherein,

$t=\underset{i=0}{\overset{7}{\mathrm{\&Sigma;}}}{P}_i$ And $m=\left\{\begin{array}{cc}1,& \mathrm{if\; t}=1\\ -1,& \mathrm{if\; t}>2\end{array}\right.$

If m=1 then represent that P is an end points, otherwise P is the branching point.

ASM models treated to the video human face image mainly comprises following step:

1. sampling obtains shape vector and profile point characteristic information to the gray scale facial image.

At first, set up model and need the manual training image of demarcating.Among the present invention, in order to ensure the accuracy of identification, select 256 width of cloth facial images (the difference expression and the attitude that comprise a plurality of people), each width of cloth image is manual demarcates 68 profile point as training data.Profile point is generally demarcated in the place that can represent objective contour, and the profile point of selecting among the present invention is marked at the exterior contour of face and the edge of organ.

Relevant calibration point is:

S _i＝(x _i，1，y _i，1，x _i，2，y _i，2，...，x _i，68，y _i，68) ^T，i＝1，2，...，256

Wherein, (xij yij) represents the coordinate of j (1≤j≤68) profile point of i width of cloth image; The Si of each width of cloth image represents a shape vector.And obtain near the characteristic information of each calibration point (profile point), these features are main foundations of utilizing the ASM model to mate.

2. set up the ASM model of people's face.

Because the difference of each sample image shooting condition, resolution, the coordinate that obtains shape vector has different proportional sizes, therefore also to make them have consistency when in the same coordinate system, representing by rotation, translation, convergent-divergent to the shape vector normalization of sample image.Utilize following method that two shape vectors are carried out and can align among the present invention: min D=[T (x)-x '] °

Wherein employed transformation rule T (x) is defined as:

$T\left(\begin{array}{c}x\\ y\end{array}\right)=\left(\begin{array}{cc}a& -b\\ b& a\end{array}\right)\left(\begin{array}{c}x\\ y\end{array}\right)+\left(\begin{array}{c}{t}_x\\ t_y\end{array}\right)$

And then change each shape vector to the positive tangent space (x of average _t-x) x _t=0, so just just in time x has been amplified 1/ (x.xt) doubly, have here | xt|=1.

After the shape vector alignment, according to the following steps people's face shape is carried out the ASM modeling:

At first calculate the average x of each shape vector in 256 samples, utilize following formula to calculate the covariance S of 256 shape vectors again, concrete method is as follows:

$S=\frac{1}{s-1}\underset{i=1}{\overset{s}{\mathrm{\&Sigma;}}}(x_i-\stackrel{\&OverBar;}{x}){(x_i-\stackrel{\&OverBar;}{x})}^T$

Calculated characteristics vector φ i and corresponding eigenvalue i on these result of calculations, press the approximation of uniform rules estimation arbitrary shape vector, x ≈ x+ Φ again _iB wherein b is that one 180 vector of tieing up is by formula b=Φ _i ^T(x-x) calculate.In addition, also need by from the mode of the distribution probability P (b) of training collective estimation b with the b stipulations to $\left|b_i\right|\&le;3\sqrt{{\mathrm{\&lambda;}}_i}.$ In the present invention, suppose each component independence and Gaussian distributed of b, so P (b) just satisfies following equation,

$\log p\left(b\right)=-0.5\underset{i=1}{\overset{\mathrm{<figure-callout\; id="2"\; label="t\; b\; i"\; filenames="H2009101679254E00041.png"\; state="\{\{state\}\}">t</figure-callout>}}{\mathrm{\&Sigma;}}}\frac{b_i^{}}{}\frac{{}_2^{}}{{\mathrm{\&lambda;}}_i}+\mathrm{const}$

Wherein constant parameter const is chosen for the average of sample, and λ i is corresponding characteristic value component.

In addition, the processing to voice signal mainly comprises following step:

1. the preliminary treatment of voice signal comprises pre-filtering, preemphasis and branch frame.What pre-filtering was adopted is the band pass filter of 60Hz to 7.8Hz.Filtered signal uses 16ms frame length and the capable frame that divides of 8ms frame shift-in then by a preemphasis filter;

2. phonetic feature extracts, and phonetic feature adopts 14 rank linear prediction cepstrums (linear predictivecoding cep strum, LPCC) coefficient.In the training stage, the cepstrum coefficient feature of extracting frame by frame from twice training utterance (identified person reads out the name of oneself) is deposited with the form of vector sequence and is the phonetic feature template, and stores feature templates into database.At the stage of recognition, from voice (identified person reads out the name of oneself), extract the speaker phonetic feature of 14 rank linear prediction cepstrum coefficients as input;

3. phonetic feature coupling, the feature vector sequence of supposing template be X=(X1, X2 ..., XT), speaker's speech characteristic vector sequence of input be Y=(Y1, Y2 ..., YT), calculate both average distances, promptly

$D(X,Y)=\frac{1}{T}\underset{t=1}{\overset{T}{\mathrm{\&Sigma;}}}d\left(t\right)=\underset{Q(\&CenterDot;)}{\min }\left\{\frac{1}{T}\underset{t=1}{\overset{T}{\mathrm{\&Sigma;}}}d(X_{Q\left(t\right)},Y_t)\right\}$

Wherein: d (t) is the matching distance of t frame; Q (t) is a crooked integer function of dull time, and Q (1)=1, Q (T)=T, matching algorithm provide Q (t) makes average matching distance reach minimum in the value of other points.

Wherein the interframe matching distance adopt Euclidean distance square:

$d(X_i,Y_j)=\underset{n=1}{\overset{N}{\mathrm{\&Sigma;}}}{(x_{i,n}-y_{j,n})}^2$

Wherein: Xi=(xi, 1, xi, 2 ..., xi, N), Yj=(yj, 1, yj, 2 ..., yj, N), N characteristic vector dimension, N=14 in this patent (i.e. 14 rank).The matching distance of input voice Y and template X be calculated as D (X, Y).

4. calculate the similarity score value, the distance D that matching process is obtained (X Y) is converted to similarity score value V3, and the V3 here changes by following rule:

$V_3=\frac{D(X,Y)-\min D}{\max D-\min D}\&times;100$

Wherein, maxD and minD represent to test vector ultimate range and the minimum range that obtains respectively, are respectively 14 and 1.

(4) Terminal Service software obtains corresponding fingerprint, sound and face characteristic masterplate data according to the identity card numbering retrieval local data base (being LDB) that the identity card card reader collects.

(5) fingerprint, sound and face characteristic comparing module are according to the matching algorithm searching database of the current apolegamy of system, and the similarity score value that obtains compared in record at every turn, and its value is respectively V1, V2 and V3.

(6) obtain respectively after the coupling score value of fingerprint, sound and face characteristic, again V1 and V2, V3 value are carried out decision level fusion, calculate and merge later matching result.

In the present invention, employing is optimum Bayes decision-making fusion method.At first V1, V2 and V3 are carried out vectorial normalized, form one 3 dimensional vector V input vector as the Bayes decision networks, the output result according to this grader obtains final decision level fusion result.Can search for optimum multi-modal fusion rule adaptively, make the Bayes value-at-risk reach minimum value, thereby construct the multi-modal biological characteristic emerging system of an optimum.

Concrete step is as follows:

1. calculate FAR, the FRR of single mode biological characteristic (fingerprint, sound and people's face)

The wrong acceptance rate of fingerprint and people's face single mode living things feature recognition method and false rejection rate are to adopt statistical method to obtain, be to have how many illegal individualities to be accredited as legal (FAR) in the used whole individualities of statistical experiment, and have how many legal individualities to be accredited as illegally (FRR).

At the matching stage of living things feature recognition, biological characteristic to be identified and the template in the database are compared in recognition methods, find out and the most similar template of individuality to be verified.Then, judge further whether individuality to be verified is present in the database, if one illegal individual, this individual visiting demand can be refused by system.For this reason, corresponding threshold value should be set personal feature is divided (legal or illegal).

Calculate the threshold value of living things feature recognition

Suppose that have M individuality in the template database, each individuality has K feature samples.So entire database can be expressed as X={x _Mk: m=1,2 ..., M; K=1,2 ..., K}.Remember similarity degree between two biological characteristic x, the y be l (x, y).If need to identify the identity of individual x, and found and the highest feature templates of x similarity degree Identify individual legitimacy according to following rule:

$\mathrm{output}=\left\{\begin{array}{cc}{m}^{*}& \mathrm{ifl}(x_{m^{*}{k}^{*}},x)\&le;T\\ \mathrm{reject}& \mathrm{otherwise}\end{array}\right.$

Some living creature characteristic recognition system is provided with two threshold values for each individuality: a low threshold value L _mWith a high threshold U _m(m=1,2 ..., M), the evaluation rule of system becomes following form [i]:

$\mathrm{output}=\left\{\begin{array}{cc}{m}^{*}& \mathrm{ifl}(x_{m^{*}{k}^{*}},x)\&le;L_{m^{*}}\\ \mathrm{reject},& \mathrm{ifl}(x_{m^{*}{k}^{*}},x)>U_{m^{*}}\\ \mathrm{applyheuristic},& \mathrm{if}{L}_{m^{*}}\&le;l(x_{m^{*}{k}^{*}},x)\&le;U_{m^{*}}\end{array}\right.$

In above-mentioned rule, low threshold value be used for determining personal feature x to be verified whether with feature templates

" enough approaching ": if both matching degrees are enough high, the identity of then identifying x is m ^*High threshold is used for clearly getting rid of two correlations between the individuality, when similarity surpasses high threshold, then thinks illegal individuality.Distance between two individualities

In the time of between the height threshold value, then need to adopt method for distinguishing, judge individual identity as heuristic.

The threshold calculations process is divided into following a few step:

At first, the biological attribute data storehouse is divided into two parts: legal individual training set X and illegal individual training set Y.Data set X is used to regulate discriminating and the classification capacity of recognition system to legal individuality; Data set Y is used for the ability that regulating system is refused legal individuality.Again training set X is divided into mutually disjoint two parts: X ₁And X ₂, suppose that each individuality has K feature samples, then X in the training set ₁And X ₂Be expressed as following form: X respectively ₁={ x _Mk∈ X:m=1,2 ..., M; K=1,2 ..., K/2}X ₂={ x _Mk∈ X:m=1,2 ..., M; K=K/2+1 ..., K}

Then, distance in the class of calculation training collection X:, select individual m at X for the individual m in the class ₁In a feature x _Mk, calculate x successively _MkWith X ₂In distance between the feature of all similar individual m.Calculate the mean value of these distances again, and this mean value as the low threshold value L of recognition system at individual m _m

At last, the between class distance between calculation training collection X and the Y: select a biological characteristic x among the X _Mk, calculated characteristics x successively _MkAnd the distance between each feature among the Y.With these the distance in minimum value as the high threshold U of recognition system at individual m _m

Note, L may occur in the Practical Calculation _m＞U _mSituation, need be to U _mAdjust, for example make U _m=L _m* 1.2.Generally can adjust according to actual needs.

3. calculate overall FAR, the FRR of emerging system

The global error rate FAR of emerging system _FusAnd FRR _FusCan calculate acquisition by FAR, the FRR of fusion rule and each single mode.Suppose that decision rule represents that with f the decision rule of bi-mode biology feature emerging system is reduced to as follows:

d 1	d 2	f?
			0?	0?	f 0
0?	1?	f 1
			1?	0?	f 2
1?	1?	f 3

D in the last table ₁, d ₂Represent two kinds of single mode recognition methodss respectively, result or 0 of each single mode identification (individual illegal) or 1 (individual legal), the output f of corresponding emerging system _i(i ∈ 0,1,2, also be 0 or 1 3}).The global error rate FAR of emerging system _FusAnd FRR _FusCan be expressed as the function of f:

${\mathrm{FRR}}_{\mathrm{fus}}=\underset{i=0}{\overset{s-1}{\mathrm{\&Sigma;}}}((1-f_i)\&times;\left(\underset{j=1}{\overset{N}{\mathrm{\&Pi;}}}\mathrm{\&Psi;}{\mathrm{RR}}_j\right))$ ${\mathrm{FAR}}_{\mathrm{fus}}=\underset{i=0}{\overset{s-1}{\mathrm{\&Sigma;}}}(f_i\&times;\left(\underset{j=1}{\overset{N}{\mathrm{\&Pi;}}}\mathrm{\&Psi;}{\mathrm{AR}}_j\right))$

Wherein:

${\mathrm{\&Psi;AR}}_j=\left\{\begin{array}{cc}1-{\mathrm{FAR}}_j& (d_j=0)\\ {\mathrm{FAR}}_j& (d_j=1)\end{array}\right.$ ${\mathrm{\&Psi;RR}}_j=\left\{\begin{array}{cc}{\mathrm{FRR}}_j& (d_j=0)\\ 1-{\mathrm{FRR}}_j& (d_j=1)\end{array}\right.$

Variable S in the above-mentioned formula represents the length of fusion rule, and in fingerprint people face bi-mode biology feature emerging system of the present invention, the fixed value that S gets is 4.

(7) show current holder authentication result, by or do not pass through.

Control method according to fingerprint recognition provided by the present invention is characterized in that, detail characteristics of fingerprints extraction step in the step (3), and it is on the basis of direction of fingerprint field that core point extracts, and utilizes the poincare method to extract this value

Wherein Δ (k) meets

$\mathrm{\&Delta;}\left(k\right)=\left\{\begin{array}{cc}\mathrm{\&delta;}\left(k\right),& \mathrm{if}\left|\mathrm{\&delta;}\left(k\right)\right|<\frac{\mathrm{\&pi;}}{2}\\ \mathrm{\&pi;}+\mathrm{\&delta;}\left(k\right),& \mathrm{if\&delta;}\left(k\right)\&le;-\frac{\mathrm{\&pi;}}{2}\\ \mathrm{\&pi;}-\mathrm{\&delta;}\left(k\right),& \mathrm{otherwise}\end{array}\right.,$

δ(k)＝O′(ψ _x(i′)，ψ _y(i′))-O′(ψ _x(i)，ψ _y(i))，

i′＝(i+1)mod?N _ψ，

After calculate finishing, judge again p (i, value j) is if (i j)=0.5, then is a core point to P; The extraction of bifurcation and end points detects by the sliding form method and obtains, and at first defines the dot structure template of bifurcation and end points, respectively with this template by from left to right, from top to bottom order traversal full figure.The every slip of template once, just the matching degree of calculation template and image corresponding region, when meeting value greater than 0.7 the time, the minutia corresponding with using template with regard to the current pixel of process decision chart picture is consistent, if promptly meet the bifurcated template, then this pixel is a bifurcation, if meet the end points template, then this pixel is exactly an end points.

According to fingerprint characteristic discrimination method provided by the present invention, it is characterized in that in the step (6), the fingerprint characteristic comparison may further comprise the steps:

1. fingerprint alignment: input fingerprint feature point set Vi={v1, v2 ..., Vn}, wherein vi=(x, y, θ), Tj={t1, t2 ... tm} is the characteristic point set of template fingerprint, and at first to the core point alignment, after the alignment, the coordinate of Vi point set is done as down conversion:

X ' _i=x _i+ (x _To-x _Io) and y ' _i=y _i+ (y _To-y _Io),

(x wherein _To, y _To) and (x _Io, y _Io) difference template fingerprint and the core point of importing fingerprint; And then the rotation alignment, the rotation alignment is then by asking a reflection transformation t to obtain:

$t\left(v\right)=\left\{\begin{array}{ccc}\cos a& -\sin a& \mathrm{\&Delta;x}\\ \sin a& \cos a& \mathrm{\&Delta;y}\\ 0& 0& 1\end{array}\right\}\left\{\begin{array}{c}x\\ \mathrm{<figure-callout\; id="1"\; label="y"\; filenames="H2009101679254E00041.png"\; state="\{\{state\}\}">y</figure-callout>}\\ 1\end{array}\right\}$

In above-mentioned conversion, the core point alignment, Δ x and Δ y determine that a is by making expression formula | v _t-t (v _i) | ²＜θ sets up, and asks Vi and Vt match point logarithm maximum to obtain, and wherein θ is a preset threshold, is set to 0.05;

2. similarity is calculated: mate for global characteristics, the feature of its input fingerprint and template fingerprint is the fourier spectrum feature, characteristic vector is exactly spectrogram pixel serialization result, similarity between the two is not wait the Euclidean distance between the long vector to realize that similarity is pressed column count by calculating:

$s=(1-\frac{\mathrm{dis}}{\max (v_i,v_t)})\&times;100\%$

In the formula, dis is both Euclidean distance values, max (v _i, v _t) expression gets in input vector and the template vector maximum vector length; For the minutia coupling, be to be benchmark with the template, will import point set Vi by (Δ x, Δ y, a) conversion projects to after template point set Vt goes up, calculate range error less than 5 location of pixels in, all match points determine that to quantity the similarity value of this moment is calculated by following formula:

$s=\frac{n_{\mathrm{couple}}}{\max (v_i,v_t)}\&times;100\%$

In the formula, n _CoupleRepresent the successfully quantity of the point of pairing, max (v _i, v _t) still expression input point set and template point are concentrated the minutiae point number of maximum;

3. fingerprint authentication: calculate after the similarity data S between input fingerprint and the template fingerprint, again with S and default recognition threshold T _sRelatively, the T here _sAdjusted according to different security requirements, if S 〉=T _sJudge that then two fingerprints successfully mate, promptly come from same finger, otherwise judge that two fingerprints are not to come from same finger.

According to face characteristic discrimination method provided by the present invention, it is characterized in that in the step (6), the face characteristic comparison may further comprise the steps:

1. with the facial image in the average Viola Jones people face monitor detection video sequence, finish the mark of 68 characteristic points automatically.

2. each characteristic point normal direction is respectively got 7 points (i.e. 2K+1 point altogether with center, characteristic point position in facial image, K=7), these 15 points form a vectorial gi, the vectorial gi of 68 characteristic point correspondences on the facial image form one group of sequence vector g1, g2 ..., g68.

Normalized vector sequence gi, computation of mean values g and covariance Sg.

F (g by formula iteratively _s)=(g _s-g) ^TS _g ^-1(g _sThe mahalanobis distance value of-face characteristic template that g) calculating input sample is corresponding with identity card numbering in the database.

If the most little mahalanobis distance is less than given threshold value T, then face characteristic is compared successfully, otherwise aspect ratio to the failure, promptly current holder and feature registrant are not same people.

According to holder identity identification system provided by the present invention, it is characterized in that identity identification system provides two kinds of operational modes, i.e. independent operation mode (specifically seeing Fig. 4) and networking operation pattern (specifically seeing Fig. 5):

1. independent operation mode: independently finish the coupling checking of characteristic collection, local storage, data retrieval and characteristic in terminal, and provide and merge later checking conclusion.

2. networking operation pattern: terminal is connected with remote validation service centre, the independent biological attribute data that extracts, characteristic during registration is carried out this locality storage respectively and is submitted to the central server remote storage, there is not register information during checking such as among the local data base LDB, then by center service software inquiry system data SDB and download fingerprint and the face characteristic data to terminal equipment, carry out signature verification again and handle.

According to holder identity identification system provided by the present invention, it is characterized in that system runs on following hardware environment (specifically seeing Fig. 1):

Terminal management end software, Terminal Service software and center service software are all supported Windows 2000/XP/vista operating system, and the assistant software back-up environment that needs is MS SQL SERVER2000 and OFFICE2000; The hardware running environment of terminal management end software is common desktop computer, the hardware environment of center service software is the power PC server, the Terminal Service running software is in terminal equipment, and terminal equipment is made of parts such as panel computer (band touch-screen), fingerprint acquisition instrument, camera and identity card card reader.

The present invention mainly provides a kind of fingerprint, people's face and card identity card of utilizing to number the application system of carrying out the holder authentication, and its main feature is summarized as follows:

1. come the holder identity is effectively verified by fingerprint, face characteristic information.

2. the mode of operation of system is changeable, both can configuration effort in single creature signature verification pattern, also can work in two kinds of biological characteristics and merge Validation Modes;

3. system both can unit work, the work of also can networking;

4. terminal equipment can be connected by the Internet with central server;

5. central server provides the function of terminal equipment being carried out legitimate verification, avoids unauthorized terminal equipment access system, and the overload of central server load.

6. the management end of system separates with service end, finish the running status of identity information registration, physical characteristics collecting, Operational Data Analysis, supervisory control system separately by terminal management software, and dispose important functions such as system operational parameters, finish identity by Terminal Service software and differentiate that registration separates with checking.

Those of ordinary skill in the art will appreciate that embodiment described here is in order to help reader understanding's principle of the present invention, should to be understood that the protection range of inventing is not limited to such special statement and embodiment.Everyly make various possible being equal to according to foregoing description and replace or change, all be considered to belong to the protection range of claim of the present invention.

Claims (10)

1. mode biological characteristic holder identity identification system, comprise and be used for terminal equipment hardware unit, terminal management software module, Terminal Service software module and center service software module four major parts that identity is differentiated, it is characterized in that the terminal equipment hardware unit is mainly used in ID card information, fingerprint characteristic, voice and the face characteristic of gathering the individual; The terminal management software module mainly is responsible for terminal management, fingerprint registration, voice registration, the registration of people's face, captured identity card information and data analysis etc.; The Terminal Service software module mainly is responsible for fingerprint, sound and face characteristic are compared and provided based on the fusion identity identification result that merges; The center service software module mainly is responsible for access fingerprint, sound and face characteristic data, and the use of terminal equipment is controlled.

2. a kind of three mode biological characteristic holder identity identification systems according to claim 1 is characterized in that described terminal management software module comprises following module again:

S21. system management module: for registering legal terminal equipment information, authorizing, be provided with operational parameter value important in the system etc. for the use operating personnel of native system;

S22. identity information typing module: be used to manually add, delete, revise typing personnel identity essential information (pressing the identity card content displayed), and the EXCEL of identity information in enormous quantities imports;

S23. fingerprint Registration Module: registrant's fingerprint characteristic information, and register information preserved into database (local data base or system database);

S24. sound Registration Module: registrant's sound characteristic information, and the characteristic information of being registered preserved into database (local data base or system database);

S25. people's face Registration Module: registrant's face characteristic information, and register information preserved into database (local data base or system database);

S26. simplation verification module: behind input identity information in the management work station, dry run is based on the bimodal holder identity fusion identification system flow process of fingerprint and face characteristic;

S27. data analysis module: the service data to whole system is added up, is analyzed, and excavates valuable potential information, differentiates record data etc. as system's running log record data and identity.

3. a kind of three mode biological characteristic holder identity identification systems according to claim 1 is characterized in that described Terminal Service software module comprises following module again:

S31. interactive communication module: the main interactive communication module swap data of being responsible for " service for checking credentials center software ", with the legitimate verification of finishing terminal equipment, upload the biological attribute data that extracts, and receive the biological attribute data of downloading from central server;

S32. data acquisition module: mainly finish based on the ID card information of identity card card reader gather automatically and manual typing, based on the fingerprint characteristic collection of fingerprint instrument with based on people's face video image acquisition of camera head etc.;

S33. fingerprint, sound, face characteristic authentication module: be responsible for the fingerprint, sound and the people's face data that collect are carried out preliminary treatment, characteristic information extraction, finish checking in 1: 1 with holder corresponding templates feature;

S34. merge decision-making module: mainly finish on people's face, sound and fingerprint characteristic coupling gained score value basis, carry out decision level fusion to differentiate whether the holder identity is consistent with accredited of institute according to the decision-making fusion method.

4. a kind of three mode biological characteristic holder identity identification systems according to claim 1 is characterized in that described center service software module comprises following module again:

S41. interactive communication module: mainly be responsible for the equipment validity authorization information that receiving terminal apparatus sends, whether legal with the access request of determining current device, and be responsible for receiving biological characteristic and the auxiliary data thereof that legal terminal is uploaded and downloaded;

S42. data access module: be responsible for the terminal data (storing system database SDB into) that storage receives by communication module, and give communication module by the biological attribute data of ID card information searching terminal request and loopback and pass down to terminal equipment;

S43. terminal equipment Authority Verification module: verify according to the information such as terminal equipment coding that receive whether the terminal equipment of institute's access service centring system is legal, and be the terminal distribution authority according to the checking result.

5. the control method of a mode biological characteristic holder identity identification system, comprise and be used for terminal equipment hardware unit, terminal management software module, Terminal Service software module and center service software module four major parts that identity is differentiated, it is characterized in that described control method comprises step:

(1) terminal equipment is by the network communication interface connecting system and connect central server, after the legitimate verification of terminal equipment identity process, obtain corresponding service request mandate, refuse unwarranted terminal equipment access system to central server request characteristic;

(2) ID card information, finger print information, sound and the video human face information of collection holder are carried out preliminary treatments such as pattern classification and noise remove;

(3) above-mentioned steps (2) is carried out preliminary treatment more respectively and extracted corresponding feature through fingerprint, sound and people's face data after the preliminary treatment;

(4) Terminal Service software obtains corresponding fingerprint, sound and face characteristic masterplate data according to the identity card numbering retrieval local data base LDB that the identity card card reader collects;

(5) fingerprint, sound and face characteristic comparing module are according to the matching algorithm searching database of the current apolegamy of system, and the similarity score value that obtains compared in record at every turn, and its value is respectively v1, v2 and v3;

(6) obtain respectively again v1, v2 and v3 value being carried out decision level fusion after the coupling score value of fingerprint, sound and face characteristic, calculate and merge later matching result.

6. the control method of a kind of three mode biological characteristic holder identity identification systems according to claim 5 is characterized in that, described fingerprint is handled and be may further comprise the steps:

(11) the fingerprint effective coverage is cut apart: fingerprint image is divided into the image block of a series of 16 * 16 non-intersections, and each piece is labeled as B (1,1) respectively, B (1,2) ..., B (i, j), utilize then following formula v (i, j)

Calculate the grey scale pixel value variance of each image block, wherein x _nWith x represent respectively this figure fast in the gray value of pixel, N represents the pixel quantity that comprises in the segment segmentation threshold v to be set _θ=11.5, respectively with the variance yields and the v of each segment _θRelatively, if v (i, j)＞v _θ, then (i j) is judged as effective finger-print region to this segment B, otherwise is judged as the background area of fingerprint;

(12) field of direction of fingerprint is calculated: calculate respectively segment B (i, j) in each pixel gradient G in the x and y direction _xAnd G _y, utilize formula d (i, j)

Calculate respectively segment B (i, local direction j), in the formula (i ' _b, j ' _b) coordinate of pixel in the expression segment, w represents the pixel wide of segment, value like the segment local direction value that calculates is only got in four component values 0, π/4, π/3 and 3 π/4 recently, local direction d (the i that segment is final, j) ∈ { 0,4 π/4, π, 3 π/4}, by the consistency feature correction field of direction result of calculation of the field of direction, (i is in 5 * 5 neighborhood D scope j) at segment B, calculate its consistency value C (i, j)

Have in this formula

Here d=mod (d (i ', j ')-d (i, j)+2 π) if C (i, j)＜0.35, then with segment B (i, local direction j) are adjusted into the most significant direction of local direction in the neighborhood D, otherwise segment B (i, local direction j) remains unchanged;

(13) use the M-PCNN network that the later image of preliminary treatment is carried out filtering;

(14) fingerprint image is carried out feature extraction, comprise that global characteristics extracts and minutia is extracted: global characteristics is extracted as the Fourier spectrum feature: at first the fingerprint image of importing is divided into 32 * 32 image block, and segment done two dimensional discrete Fourier transform, formula is

In the formula, (m n) is the codomain coordinate of pixel, and (i k) is the frequency domain coordinate of pixel correspondence, and each subgraph piece has just obtained the fourier spectrum figure of full figure through after the Fourier transform, and it is quantized into the global characteristics vector of fingerprint by rule;

Minutia is extracted, and comprises core point, bifurcation and end points, and used minutia is extracted template respectively as following table:

P ₃ P ₂ P ₁ P ₄ P? P ₀ P ₅ P ₆ P ₇

Wherein,

And

If m=1 then represent that P is an end points, otherwise P is the branching point.

7. the control method of a kind of three mode biological characteristic holder identity identification systems according to claim 5 is characterized in that, described fingerprint is handled and be may further comprise the steps:

(21) sampling obtains shape vector and profile point characteristic information to the gray scale facial image;

(22) set up the ASM model of people's face.

8. the control method of a kind of three mode biological characteristic holder identity identification systems according to claim 5, it is characterized in that in the described minutia extraction step, its core point is on the basis of direction of fingerprint field, utilize the poincare method to extract this value

Wherein Δ (k) meets

δ(k)＝O′(Ψ _x(i′)，Ψ _y(i′))-O′(Ψ _x(i)，Ψ _y(i))，

i′＝(i+1)mod?N _Ψ，

After calculate finishing, judge again p (i, value j) is if (i j)=0.5, then is a core point to P; The extraction of bifurcation and end points detects by the sliding form method and obtains, and at first defines the dot structure template of bifurcation and end points, respectively with this template by from left to right, from top to bottom order traversal full figure; The every slip of template once, just the matching degree of calculation template and image corresponding region, when meeting value greater than 0.7 the time, the minutia corresponding with using template with regard to the current pixel of process decision chart picture is consistent, if promptly meet the bifurcated template, then this pixel is a bifurcation, if meet the end points template, then this pixel is exactly an end points.

9. the control method of a kind of three mode biological characteristic holder identity identification systems according to claim 5 is characterized in that, the fingerprint characteristic contrast may further comprise the steps:

(31) fingerprint alignment: input fingerprint feature point set Vi={v1, v2 ..., Vn}, wherein vi=(x, y, θ), Tj={t1, t2 ... tm} is the characteristic point set of template fingerprint, and at first to the core point alignment, after the alignment, the coordinate of Vi point set is done as down conversion:

X ' _i=x _i+ (x _To-x _Io) and y ' _i=y _i+ (y _To-y _Io),

(x wherein _To, y _To) and (x _Io, y _Io) difference template fingerprint and the core point of importing fingerprint; And then the rotation alignment, the rotation alignment is then by asking a reflection transformation t to obtain:

In above-mentioned conversion, the core point alignment, Δ x and Δ y determine that a is by making expression formula | v _t-t (v _i) | ²＜θ sets up, and asks Vi and Vt match point logarithm maximum to obtain, and wherein θ is a preset threshold, is set to 0.05;

(32) similarity is calculated: mate for global characteristics, the feature of its input fingerprint and template fingerprint is the fourier spectrum feature, characteristic vector is exactly spectrogram pixel serialization result, similarity between the two is not wait the Euclidean distance between the long vector to realize by calculating, and wherein similarity is calculated by following formula:

In the formula, dis is both Euclidean distance values, max (v _i, v _t) expression gets in input vector and the template vector maximum vector length; For the minutia coupling, be to be benchmark with the template, will import point set Vi by (Δ x, Δ y, a) conversion projects to after template point set Vt goes up, calculate range error less than 5 location of pixels in, all match points determine that to quantity the similarity value of this moment is calculated by following formula:

In the formula, n _CoupleRepresent the successfully quantity of the point of pairing, max (v _i, v _t) still expression input point set and template point are concentrated the minutiae point number of maximum;

(33) fingerprint authentication: calculate after the similarity data S between input fingerprint and the template fingerprint, again with S and default recognition threshold T _sRelatively, the T here _sAdjusted according to different security requirements, if S 〉=T _sJudge that then two fingerprints successfully mate, promptly come from same finger, otherwise judge that two fingerprints are not to come from same finger.

10. the control method of a kind of three mode biological characteristic holder identity identification systems according to claim 5 is characterized in that, described face characteristic comparison may further comprise the steps:

(41), finish the mark of 68 characteristic points automatically with the facial image in the average Viola Jones people face monitor detection video sequence;

(42) in facial image each characteristic point normal direction with center, characteristic point position respectively get 7 points (i.e. 2K+1 point altogether, K=7), these 15 points form a vectorial gi, the vectorial gi of 68 characteristic point correspondences on the facial image forms one group of sequence vector g ₁, g ₂..., g ₆₈

(43) normalized vector sequence gi, computation of mean values g and covariance Sg.

(44) f (g by formula iteratively _s)=(g _s-g) ^TS _g ^-1(g _sThe mahalanobis distance value of-face characteristic template that g) calculating input sample is corresponding with identity card numbering in the database;

(45) if the most little mahalanobis distance less than given threshold value T, then face characteristic is compared successfully, otherwise aspect ratio to the failure, promptly current holder and feature registrant are not same people.

Images