CN103984941A - Face recognition checking-in method and device thereof - Google Patents

Abstract

The invention provides a face recognition checking-in method and a device thereof. Front and side face images of a person are obtained in the face checking-in process, then collected face images in two directions of the front and the side are matched, and the method can be switched into a normal identification process only if the collected pictures of two angles are successfully matched, so as to effectively overcome the problem of false attendance by using the picture. By adopting the scheme, not only is the recognition accuracy rate and efficiency of a face attendance machine not affected, but also the problem of false attendance by using the picture can be quickly and effectively overcome.

Description

Human face identification work-attendance checking method and device thereof

Technical field

The present invention relates to field of image recognition, refer in particular to a kind of human face identification work-attendance checking method and device thereof.

Background technology

The staff attendance management of strictly regulating is the important guarantee that management benefit improves in modern enterprises and institutions, and traditional work attendance form representing to check card, to block Zhong Wei, exist generation beat phenomenon, computing velocity is slow, magnetic card easily damages or leaves behind, maintenance of equipment high in cost of production drawback, has more and more been not suitable for the needs that modern enterprises and institutions develop.

Biological identification technology is by close combinations of high-tech means such as computing machine and optics, acoustics, biology sensor and biostatistics principles, utilizes the intrinsic physiological property of human body (as fingerprint, face picture, iris etc.) and behavioural characteristic (as person's handwriting, sound, gait etc.) to carry out the qualification of personal identification.Nowadays bio-identification, constantly universal, not only becomes the Work attendance management system of enterprise's first-selection, and along with the continuous upgrading of biological identification technology, its application is also in continuous expansion.Enterprise's work attendance is a large vital point of enterprise, and the continuous upgrading of its equipment is also representing the continuous upgrading of biological identification technology.The rise of biometric apparatus overcome be replicated, stolen, the series of problems such as pass into silence, be all widely used in every field taking fingerprint recognition, recognition of face, iris recognition as the biometric apparatus of representative.Fingerprint attendance system is current the most ripe and biological Time Attendance Device that price is more cheap, but fingerprint attendance machine is all very high to the requirement of environment and work attendance personnel skin, and when the situations such as air is dry, skin is dirty, cast off a skin are with regard to None-identified, and read head easily weares and teares.These reasons make the attendance recorder life-span short, and maintenance cost is high.The reliability of iris identification equipment is fine, but cost is high, cannot promote on a large scale.

Recognition of face, with respect to other biometric discrimination method, has obvious advantage, thereby becomes rapidly a market focus in the whole world in recent years.The advantage of face Time Attendance Device is as follows:

1) user easily accepts, transport simply to use, to user without particular/special requirement.

2) anti-counterfeiting performance is good, is difficult for forgery or stolen.

3) can carry, not worry omitting or losing, can use whenever and wherever possible.

4) noncontact, much cleaner, be not afraid of spread of germs.

5) convenient and swift, recognition time is less than 1 second.

6) camera is a large amount of universal, is easy to promote the use of.

7) safe and reliable, do not relate to individual privacy.

8) by network or USB flash disk communication.

Visible, application human face identification work-attendance checking system can effectively improve the attendance management mode of enterprise, and the operation of specification staff attendance prevents the behavior that generation checks card, practises fraud, also raising work attendance efficiency easily and effectively.Support TCP/IP networking mode, attendance data is uploaded administrative authority automatically simultaneously, management attendance data.Be widely used in enterprises and institutions, education of middle and primary schools mechanism, hotel, club, hospital etc.

But traditional two dimensional image face attendance recorder gathers facial image conventionally from front, if cribber uses other people photo to carry out work attendance, just can play the object that replaces other people work attendance, this just makes two dimensional image face attendance recorder lose the meaning existing.And 3-D view face attendance recorder cost is high, computation complexity is high, the general consumer group there is no need to select large, the slow-footed Time Attendance Device of this cost.

Summary of the invention

Technical matters to be solved by this invention is: the problem of avoiding the low and three-dimensional face work attendance high cost of common face work attendance reliability in the past, provide a kind of by front face image and Side Face Image, by increase face positive with the mating flow process and then realize the high reliability face work attendance under low cost of side.

In order to solve the problems of the technologies described above, the technical solution used in the present invention is: a kind of human face identification work-attendance checking method, comprises coupling flow process and identification process;

Described coupling flow process comprises step,

S21) in front face image and the Side Face Image of synchronization collection destination object;

S22) adopt relatively front face image and Side Face Image of Elastic Matching method, forward identification process to if the match is successful, it fails to match returns to error message;

Described identification process comprises compares data in the front face image of collection and face database, if unanimously work attendance success of comparison is failed return to the step of error message.

The invention still further relates to a kind of human face identification work-attendance checking device, comprise at least two group cameras and a central processing unit; Two groups of cameras are arranged at respectively the both sides of face pickup area and are connected with described central processing unit;

Described two groups of cameras are respectively used to gather front face image and the Side Face Image of destination object;

Described central processing unit is used for adopting relatively front face image and Side Face Image of Elastic Matching method, forwards identification process to if the match is successful, and it fails to match returns to error message; Carry out identification process data in the front face image of collection and face database are compared, if comparison unanimously work attendance is successfully, failed return to error message.

Beneficial effect of the present invention is: by obtaining people's front and Side Face Image when the face work attendance, then the facial image of the front collecting and side both direction is mated, only have picture when two angles that collect the match is successful just to forward normal identification process to, thereby effectively overcome the problem of utilizing the false work attendance of photo.This scheme does not only affect recognition accuracy and the efficiency of face attendance recorder, and can overcome fast and effectively the problem of utilizing the false work attendance of photo.

Brief description of the drawings

Below in conjunction with accompanying drawing in detail concrete structure of the present invention is described in detail

Fig. 1 is concrete example flow diagram of the present invention.

Embodiment

By describing technology contents of the present invention, structural attitude in detail, being realized object and effect, below in conjunction with embodiment and coordinate accompanying drawing to be explained in detail.

The design of most critical of the present invention is: before recognition of face, increase a coupling identification, object is whether the facial image that the facial image that obtains of inspection front camera and side camera obtain is same person.After gathering respectively the facial image of and side positive until work attendance people, mate identification, only have and in the time that both mate, put capable identification into, do not mate and directly return to error message.If cribber uses photo to carry out work attendance, obviously, front scan to facial image and the facial image of side scanning be not same person, matching stage cannot pass through, face attendance recorder is refused to know.This has just effectively reached the object that prevents from using photo cheating.

The invention provides a kind of human face identification work-attendance checking method, comprise coupling flow process and identification process;

Described coupling flow process comprises step,

S21) in front face image and the Side Face Image of synchronization collection destination object;

S22) adopt relatively front face image and Side Face Image of Elastic Matching method, forward identification process to if the match is successful, it fails to match returns to error message;

Described identification process comprises compares data in the front face image of collection and face database, if unanimously work attendance success of comparison is failed return to the step of error message.

From foregoing description, beneficial effect of the present invention is: by increase coupling flow process before recognition of face, in the time of face work attendance, obtain people's front and Side Face Image, then the facial image of the front collecting and side both direction is mated, only have picture when two angles that collect the match is successful just to forward normal identification process to, thereby effectively overcome the problem of utilizing the false work attendance of photo.This scheme does not only affect recognition accuracy and the efficiency of face attendance recorder, and can overcome fast and effectively the problem of utilizing the false work attendance of photo.

Embodiment 1:

In above-mentioned a kind of human face identification work-attendance checking method, also include register flow path, this register flow path comprises step:

S11) for destination object distribute one No. ID;

S12) gather the facial image of multiple destination objects;

S13) facial image gathering is carried out to the feature extraction of the laggard pedestrian's face of pre-service;

S14) by corresponding with No. ID face characteristic of the extraction face database that deposits in.

Embodiment 2:

Further, in above-mentioned register flow path, pass through Adaboost algorithm performing step S13.

Adaboost algorithm is that Paul Viola and Michael Jones propose [10] in calendar year 2001.It is a kind of alternative manner, and its basic thought is to train same sorter (Weak Classifier) for different training sets, then the sorter obtaining on these different training sets is gathered, and forms a final strong classifier.

Embodiment 3:

In described coupling flow process, step S22 specifically comprises step,

S221) the two-dimensional grid F to front face image definition face template.

S222) use feature vector, X_irepresent near information node i in two-dimensional grid F.

S223) definition centre frequency difference, bandwidth difference, two-dimensional Gabor filter that direction is different, be expressed as G=(g₁, g₂..., g_m)^t.

S224) to Side Face Image definition two-dimensional grid F'.

S225) use feature vector, X_i' represent near information node i in two-dimensional grid F', and X_i' and X_ifor proper vector of the same type

S226) adopt the matching value E (f) between Euclidean distance compute vector:

$\begin{array}{c}E\left(f\right)=\underset{i}{\mathrm{\Σ}}\frac{\⟨X_i,{X_i}^{\′}\⟩}{\left|\right|X_i\left|\right|\left|\right|{X_i}^{\′}\left|\right|}-{\mathrm{\λ}}_1\underset{i_1{j}_1{i}_2{j}_2}{\mathrm{\Σ}}{\left|\right|(P\left(i_1\right)-P\left(i_2\right))-(Q\left(j_1\right)-Q\left(j_2\right))\left|\right|}^2\\ -{\mathrm{\λ}}_2\underset{i,j}{\mathrm{\Σ}}\left|\right|(P\left(i\right)-Q\left(j\right))-\frac{1}{k}\underset{k=1}{\overset{K}{\mathrm{\Σ}}}(P\left(I_k\right)-Q\left(J_k\right))\left|\right|\end{array}$

In formula, P (i) is node i coordinate in front face image in grid F, and Q (j) is each node coordinate in Side Face Image in grid F', and K represents the number of got key point, P (I_k), Q (J_k) represent respectively the coordinate of k key point in front face image and Side Face Image, λ₁and λ₂for weighting coefficient;

S227) setting threshold thresholdd;

S228), in the time of E (f) >=thresholdd, the match is successful forwards identification process to; In the time of E (f) <thresholdd, it fails to match returns to error message.

What in the coupling flow process of the present embodiment, adopt is the method for Elastic Matching.The method represents face with sparse matrix figure, the proper vector mark that stage in figure obtains by the Gabor wavelet decomposition of picture position, the distance vector mark of connected node for image border, because Elastic Matching method is insensitive to illumination, displacement, dimensional variation, distortion is had to certain stability, be suitable for very much the positive face of this programme and mating of side face.This stage mates front face image and Side Face Image by Elastic Matching method, only has successfully coupling, just can carry out next step cognitive phase; Otherwise attendance recorder is refused to know.

Embodiment 4:

Described identification process specifically comprises step,

S31) front face image collecting is carried out to pre-service and feature extraction;

Described pre-service comprises carries out the adjustment of illumination, the removal of noise, the unification of dimension to image;

Described feature extraction is PAC feature extraction, and it is mainly that face characteristic is extracted, all characteristic storage of each object be an one-dimensional vector as a training sample, and all training samples are stored in a two-dimensional matrix.

S32) adopt and based on presentation class method, the test sample book in the comparison sample in the front face image collecting and face database is compared, obtain comparing score value;

S33) absolute difference of the comparison sample of comparison in front face image and any two test specimens in face database, if be not more than first threshold recognition failures return to error message, otherwise continuation step;

S34) judge whether two differences of comparing score values are less than Second Threshold, are that recognition failures returns to error message, otherwise continue step;

S35) judging that whether comparison result is consistent with No. ID, be to continue step, otherwise recognition failures returns to error message;

In this step, No. ID is by the process of information acquisition with the associated of result, has collected the id number of corresponding face.In work attendance process, matching stage success, identifies.What identifying adopted is the method based on presentation class, when having passed through S33 and S34 judgement, also judges that whether recognition result is consistent with No. ID that collects, if inconsistent or cannot be by work attendance.

S36) work attendance success.

Embodiment 5:

It is as follows based on presentation class method in step S32 in described coupling flow process,

Describedly based on presentation class method be: first utilize a linear combination of all training samples to represent test sample book.Suppose to have C class, every class has the training sample of n column vector form.Make x₁..., x_nfor all N training sample (N=Cn).X_{(i-1) n+k}represent k training sample of i object, i=1,2 ..., C.Make column vector z represent test sample book.Training sample from all classes is represented to test sample book, and test sample book can approximate representation be

y＝XB……(4)

Wherein X=[x₁... x_n] expression training sample, B=[b₁... b_n]^trepresent matrix of coefficients.(4) Xie Tongwei of formula

$\hat{B}={(X^{T}X+\mathrm{\&mu;I})}^{-1}{X}^{T}y......\left(5\right)$

μ is a little positive constant, and I is unit matrix.Order, obviously, if X^tx is nonsingular, and the solution of formula (3) can be expressed as following formula

$\hat{B}={\left(X^{T}X\right)}^{-1}{X}^{T}y......\left(6\right)$

The deviation that CRC calculates between i class training sample and test sample book (is compared score value r_i) be:

$r_i=\left|\right|z-X_i{\hat{B}}_i\left|\right|......\left(7\right)$

Wherein X_i=[x_{(i-1) * n+1}... x_i*n],${\hat{B}}_i={[{\hat{b}}_{(i-1)*n+1}...{\hat{b}}_{i*n}]}^T$。If$k=\underset{i}{\arg \min }{r}_i,$Test sample book is assigned to k class by CRC so.

Embodiment 6:

In described coupling flow process, in step S33, concrete operations are as follows,

Work as r_iwhen>=thresholdd1, system refuses to know.Because in the time that deviation score is more than or equal to thresholdd1 (first threshold), this test sample book is described and exists the object in training sample database to differ too large, the object of this test sample book is not registered personnel;

Embodiment 7:

In described coupling flow process, in step S34, concrete operations are as follows,

Be provided with p test sample book, make r_i1, r_i2..., r_iprepresent respectively i training sample respectively with the difference of p test sample book.So, as | r_ia-r_ib| (thresholdd2 is Second Threshold to≤thresholdd2, r for 1≤a≤p, 1≤b≤p_iaand r_ibrepresent the difference of i training sample and any two test sample books) time, system refuses to know.Because when the deviation between test sample book and two objects obtains phase-splitting difference when very little, illustrative system be can not distinguish this two objects.Two threshold value thresholdd1 (first threshold) are set for we and thresholdd2 (Second Threshold) refuses to know to surveying the testee who does not satisfy condition.Refuse to know function by increase, our face telltale system is more safe and reliable.

Concrete exemplifying embodiment:

Provide a concrete application example with regard to the technical scheme of this patent said method below.

This programme improves face attendance recorder, utilizes photo to carry out the generation of the cheating of work attendance to overcome attempt.Attendance recorder use procedure after improvement divides three parts: register flow path, coupling flow process and identification process.Illustrate the ins and outs of each flow process below:

Register flow path

This flow process is the information of all measurands (name, No. ID, department etc.) and facial image (being for example generally everyone 10 width) will be deposited in attendance recorder in advance, to afterwards measurand is carried out to work attendance and record.This stage mainly comprises that face detection, image pre-service, feature extraction, face characteristic deposit face database in.We adopt and conventional carry out face detection based on adaboost algorithm.Adaboost algorithm is that Paul Viola and Michael Jones proposed in calendar year 2001.It is a kind of alternative manner, and its basic thought is to train same sorter (Weak Classifier) for different training sets, then the sorter obtaining on these different training sets is gathered, and forms a final strong classifier.

Coupling flow process

Adopt the method for Elastic Matching.The method represents face with sparse matrix figure, then the proper vector mark that the stage in this figure is obtained by the Gabor wavelet decomposition of picture position, the distance vector mark of connected node for image border.Elastic Matching method is insensitive to illumination, displacement, dimensional variation, and distortion is had to certain stability, is suitable for very much the positive face of this programme and mating of side face.This stage mates front face image and Side Face Image by Elastic Matching method, only has successfully coupling, just can carry out next step cognitive phase; Otherwise attendance recorder is refused to know.

Idiographic flow is:

1) user inputs oneself No. ID;

2) in front face image and the Side Face Image of synchronization collection destination object;

3) judge whether face detects successfully, whether achievement collects front face image and the Side Face Image of object, otherwise returns to error message, is to continue;

4) adopt relatively front face image and Side Face Image of Elastic Matching method, forward identification process to if the match is successful, it fails to match returns to error message.

Identification process

Adopt classification (RBC) method based on representing, the method for conventional RBC is also referred to as the method for " rarefaction representation classification (SRC) ".In the middle of numerous face identification methods, SRC has been subject to extensive concern.SRC method is a method based on all samples, and its basic thought is to combine to represent given test sample book by the sparse linear of whole training samples, and sparse non-null representation coefficient supposition should concentrate on the true class label of test sample book place.Verified SRC method is very effective aspect identification face, and blocks and all have suitable robustness for expression shape change, illumination condition and face.In recognition of face problem, conventional RBC comprise based on be the classification of Norm minimum, the classification based on Norm minimum, the classification based on Norm minimum.Wherein, the RBC based on Norm minimumization constraint has two types.To utilize to represent a test sample book from the training sample of all classes, as: cooperation presentation class (CRC); Another kind of utilization represents respectively test sample book from the training sample of every class, as: linear regression classification (LRC).These two kinds of methods are all finally to utilize expression result to classify.Classification based on Norm minimum is little with its computation complexity, degree of accuracy advantages of higher is widely used.

Idiographic flow is:

5) front face image collecting is carried out to pre-service and feature extraction;

6) compare with face database, obtain comparing score value;

7) judge whether comparison score value is not less than first threshold, is that recognition failures returns to error message, otherwise continue step;

8) judge whether the difference of comparing score value is less than Second Threshold, is that recognition failures returns to error message, otherwise continue step;

9) judging that whether comparison result is consistent with No. ID, be to continue step, otherwise recognition failures returns to error message;

10) work attendance success.

The invention still further relates to a kind of human face identification work-attendance checking device, comprise at least two group cameras and a central processing unit; Two groups of cameras are arranged at respectively the both sides of face pickup area and are connected with described central processing unit;

Described two groups of cameras are respectively used to gather front face image and the Side Face Image of destination object;

Described central processing unit is used for adopting relatively front face image and Side Face Image of Elastic Matching method, forwards identification process to if the match is successful, and it fails to match returns to error message; Carry out identification process data in the front face image of collection and face database are compared, if comparison unanimously work attendance is successfully, failed return to error message.

Said apparatus also can be equipped with a display screen if desired, for showing the facial image of positive and side and processing sequence of operations instruction.A kind of typical scheme, the face attendance recorder after improvement can be designed to have in the dull and stereotyped side tilting the flat board perpendicular to ground of installation, and camera has been installed in the above.Can gather the facial image of positive and side both direction simultaneously, and be presented on display screen.

Face Work attendance device after improvement has following several feature:

(1) gather the facial image of positive and side both direction simultaneously, and mate.Thereby prevent from cheating by photo the phenomenon of attendance recorder cheating.

(2) display screen can show the facial image of positive and side simultaneously, makes testee see whether camera clearly collects the face-image of oneself.

(3) face recognition result will be presented on screen, and checking-in result is recorded in database.

(4) flat board that infrared camera and screen is housed tilts, and is low no matter user is height, and all needing bows slightly looks down camera, gets final product the complete facial image that collects.

(5) in column, single-chip microcomputer is housed, is used for carrying out view data storage and view data processing, convenient for maintaining and upgrading.

(6) supvr can preserve process information by the mode of wireless network or USB flash disk, convenient management.

In sum, the invention provides a kind of human face identification work-attendance checking method and device that overcomes the cheating of face Time Attendance Device, by the existing technical increase coupling recognition function of face attendance recorder, make cribber cannot use photo to carry out false work attendance.This improvement project increases a camera in the side of attendance recorder, gathers the image of work attendance person front and side simultaneously.First, attendance recorder will carry out face coupling, guarantees that the front face image and the Side Face Image that collect are same persons.After the match is successful, facial image and face database that attendance recorder collects front are compared.In comparison process, not only non-registered personnel are refused to know, also to and two objects between deviation obtain object that phase-splitting difference is very little and refuse to know, thereby ensured face attendance recorder safety, carry out work attendance efficiently.Through not only safety, reliable of improved recognition of face, and counting yield is high, more convenient and practical.

The foregoing is only embodiments of the invention; not thereby limit the scope of the claims of the present invention; every equivalent structure or conversion of equivalent flow process that utilizes instructions of the present invention and accompanying drawing content to do; or be directly or indirectly used in other relevant technical fields, be all in like manner included in scope of patent protection of the present invention.

Claims (6)

1. a human face identification work-attendance checking method, is characterized in that: comprise coupling flow process and identification process;

Described coupling flow process comprises step,

S21) in front face image and the Side Face Image of synchronization collection destination object;

S22) adopt relatively front face image and Side Face Image of Elastic Matching method, forward identification process to if the match is successful, it fails to match returns to error message;

Described identification process comprises compares data in the front face image of collection and face database, if unanimously work attendance success of comparison is failed return to the step of error message.

2. human face identification work-attendance checking method as claimed in claim 1, is characterized in that: also comprise register flow path; Described register flow path comprises step,

S11) for destination object distribute one No. ID;

S12) gather the facial image of multiple destination objects;

S13) facial image gathering is carried out to the feature extraction of the laggard pedestrian's face of pre-service;

S14) by corresponding with No. ID face characteristic of the extraction face database that deposits in.

3. human face identification work-attendance checking method as claimed in claim 2, is characterized in that: in described register flow path, pass through Adaboost algorithm performing step S13.

4. human face identification work-attendance checking method as claimed in claim 1, is characterized in that: in described coupling flow process, step S22 specifically comprises step,

S221) the two-dimensional grid F to front face image definition face template;

S222) represent near information node i in two-dimensional grid F with feature vector, X i;

S223) definition centre frequency difference, bandwidth difference, two-dimensional Gabor filter that direction is different, be expressed as G=(g₁, g₂..., g_m)^t;

S224) to Side Face Image definition two-dimensional grid F',

S225) use feature vector, X_i' represent near information node i in two-dimensional grid F', and X_i' and X_ifor proper vector of the same type

S226) adopt the matching value E (f) between Euclidean distance compute vector:

$\begin{array}{c}E\left(f\right)=\underset{i}{\mathrm{\&Sigma;}}\frac{\&lang;X_i,{X_i}^{\&prime;}\&rang;}{\left|\right|X_i\left|\right|\left|\right|{X_i}^{\&prime;}\left|\right|}-{\mathrm{\&lambda;}}_1\underset{i_1{j}_1{i}_2{j}_2}{\mathrm{\&Sigma;}}{\left|\right|(P\left(i_1\right)-P\left(i_2\right))-(Q\left(j_1\right)-Q\left(j_2\right))\left|\right|}^2\\ -{\mathrm{\&lambda;}}_2\underset{i,j}{\mathrm{\&Sigma;}}\left|\right|(P\left(i\right)-Q\left(j\right))-\frac{1}{k}\underset{k=1}{\overset{K}{\mathrm{\&Sigma;}}}(P\left(I_k\right)-Q\left(J_k\right))\left|\right|\end{array}$

In formula, P (i) is node i coordinate in front face image in grid F, and Q (j) is each node coordinate in Side Face Image in grid F', and K represents the number of got key point, P (I_k), Q (J_k) represent respectively the coordinate of k key point in front face image and Side Face Image, λ₁and λ₂for weighting coefficient;

S227) setting threshold thresholdd;

S228), in the time of E (f) >=thresholdd, the match is successful forwards identification process to; In the time of E (f) <thresholdd, it fails to match returns to error message.

5. human face identification work-attendance checking method as claimed in claim 1, is characterized in that: described identification process specifically comprises step,

S31) front face image collecting is carried out to pre-service and feature extraction;

S32) adopt and based on presentation class method, the test sample book in the comparison sample in the front face image collecting and face database is compared, obtain comparing score value;

S33) absolute difference of the comparison sample of comparison in front face image and any two test specimens in face database, if be not more than first threshold recognition failures return to error message, otherwise continuation step;

S34) judge whether two differences of comparing score values are less than Second Threshold, are that recognition failures returns to error message, otherwise continue step;

S35) judging that whether comparison result is consistent with No. ID, be to continue step, otherwise recognition failures returns to error message;

S36) work attendance success.

6. a human face identification work-attendance checking device, is characterized in that: comprise at least two group cameras and a central processing unit; Two groups of cameras are arranged at respectively the both sides of face pickup area and are connected with described central processing unit;

Described two groups of cameras are respectively used to gather front face image and the Side Face Image of destination object;

Described central processing unit is used for adopting relatively front face image and Side Face Image of Elastic Matching method, forwards identification process to if the match is successful, and it fails to match returns to error message; Carry out identification process data in the front face image of collection and face database are compared, if comparison unanimously work attendance is successfully, failed return to error message.