The content of the invention
For these reasons, the present invention provides a kind of micro- expression face and examines householder method, device and storage medium, can be in face
During examining, micro- expression of loan customer is identified, analyzed, it is objective, effectively screen suspicion of the client with the presence or absence of fraud
Doubt.
To achieve the above object, the present invention provides a kind of computing device, the computing device by network connection camera device,
Or camera device is included by network connection mobile terminal, the mobile terminal.Camera device shooting air control personnel face is examined credit money
The video of client, the computing device is sent to by network.Micro- expression face in the computing device memory of the computing device
Auxiliary program is examined, realizes following steps:
Emergent question and answer step:According to the essential information of client automatically generate magnanimity stress problem and these stress problem
Model answer;
First video acquisition step:Gather client answer stress problem the first video, by it is single stress be in units of problem
First video is cut to obtain it is each stress the first video segment corresponding to problem;
Fisrt feature extraction step:Extract the first motion characteristic and that client's face includes in each first video segment
Two motion characteristics;
Model training step:The spy formed with the first motion characteristic identified in each video segment and the second motion characteristic
Subset is levied, and face examines fraud of the personnel to the video segment and is labeled as sample data, and supporting vector machine model is trained,
Obtain cheating Rating Model;
Second video acquisition step:Gather client answer stress problem the second video, by it is single stress be in units of problem
Second video is cut to obtain it is each stress the second video segment corresponding to problem;
Second feature extraction step:Extracted from each second video segment the first motion characteristic for including of client's face and
The character subset that second motion characteristic is formed;
Score step:The fraud Rating Model that the character subset input training of extraction is obtained, output client answers each
Stress problem fraud appraisal result.
First motion characteristic is motor unit (action unit, AU), and second motion characteristic includes head court
To and eyeball direction.
Preferably, the fisrt feature extraction step, comprises the following steps:
Using Face datection algorithm facial image is extracted from the first video segment;
Human face characteristic point is extracted from facial image;
Dimension-reduction treatment is carried out to the human face characteristic point of extraction;
According to the first motion characteristic occurred in the human face characteristic point identification facial image after dimensionality reduction, it is dynamic to obtain each first
Make feature duration in first video segment, and according to the coordinate meter of human face characteristic point in each first motion characteristic
Calculate the second motion characteristic that first motion characteristic includes.
Further, the fisrt feature extraction step, can also comprise the following steps:
The time of first motion characteristic and the second motion characteristic while appearance, number are counted in each first video segment;
Count the first motion characteristic in each first video segment, two motion characteristics of the second motion characteristic only have one to go out
Existing time, number;
One feature set is generated according to statistical result;And
The feature in the feature set is screened using Feature Selection algorithm, obtains character subset.
The present invention also provides a kind of micro- expression face and examines householder method, and applied to computing device, this method includes:
Emergent question and answer step:According to the essential information of client automatically generate magnanimity stress problem and these stress problem
Model answer;
First video acquisition step:Gather client answer stress problem the first video, by it is single stress be in units of problem
First video is cut to obtain it is each stress the first video segment corresponding to problem;
Fisrt feature extraction step:Extract the first motion characteristic and that client's face includes in each first video segment
Two motion characteristics;
Model training step:The spy formed with the first motion characteristic identified in each video segment and the second motion characteristic
Subset is levied, and face examines fraud of the personnel to the video segment and is labeled as sample data, and supporting vector machine model is trained,
Obtain cheating Rating Model;
Second video acquisition step:Gather client answer stress problem the second video, by it is single stress be in units of problem
Second video is cut to obtain it is each stress the second video segment corresponding to problem;
Second feature extraction step:Extracted from each second video segment the first motion characteristic for including of client's face and
The character subset that second motion characteristic is formed;
Score step:The fraud Rating Model that the character subset input training of extraction is obtained, output client answers each
Stress problem fraud appraisal result.
Preferably, the fisrt feature extraction step, comprises the following steps:
Using Face datection algorithm facial image is extracted from the first video segment;
Human face characteristic point is extracted from facial image;
Dimension-reduction treatment is carried out to the human face characteristic point of extraction;
According to the first motion characteristic occurred in the human face characteristic point identification facial image after dimensionality reduction, it is dynamic to obtain each first
Make feature duration in first video segment, and according to the coordinate meter of human face characteristic point in each first motion characteristic
Calculate the second motion characteristic that first motion characteristic includes.
Further, the fisrt feature extraction step, can also comprise the following steps:
The time of first motion characteristic and the second motion characteristic while appearance, number are counted in each first video segment;
Count the first motion characteristic in each first video segment, two motion characteristics of the second motion characteristic only have one to go out
Existing time, number;
One feature set is generated according to statistical result;And
The feature in the feature set is screened using Feature Selection algorithm, obtains character subset.
In addition, to achieve the above object, the present invention also provides a kind of computer-readable recording medium, the computer
Readable storage medium storing program for executing includes micro- expression face and examines auxiliary program 10, and micro- expression face is examined auxiliary program 10 and is executed by processor
When, realize the arbitrary steps that micro- expression face as described above is examined in householder method.
Householder method, device and storage medium are examined in micro- expression face provided by the invention, according to the essential information of loan customer
Automatically generate magnanimity stress problem and these stress problem model answer.In the model training stage, collection client answers should
Swash the first video of problem, by it is single the first video of collection stress be cut to obtain in units of problem it is each stress problem
Corresponding video segment, identify client's face includes in each video segment the first motion characteristic (AU) and the second motion characteristic
(such as head direction, eyeball direction), is formed with the first motion characteristic and the second motion characteristic that are identified in each video segment
Character subset, and face examine personnel the client in the video segment is answered stress the fraud of problem be labeled as sample data,
Supporting vector machine model is trained, obtains cheating Rating Model.Afterwards, the model trained is examined applied to real-time face
Link:Gather client answer stress problem the second video, first video of collection stress be carried out in units of problem by single
Cutting obtain it is each stress video segment corresponding to problem, from each video segment extract client's face include first action
The character subset that feature and the second motion characteristic are formed, the fraud Rating Model that the character subset input training of extraction is obtained,
Export client answer each stress problem fraud appraisal result.Using the present invention, can with it is objective, effectively whether screen client
In the presence of the suspicion of fraud.
Embodiment
It should be appreciated that the specific embodiments described herein are merely illustrative of the present invention, it is not intended to limit the present invention.
It is the applied environment figure that the preferred embodiment of householder method first is examined in the micro- expression face of the present invention shown in ginseng Fig. 1.In the reality
Apply in example, camera device 3 connects computing device 1 by network 2, and camera device 3 shoots air control personnel face and examines regarding for loan customer
Frequently (the positive face video for mainly shooting loan customer), computing device 1 is sent to by network 2, computing device 1 utilizes the present invention
Auxiliary program 10 (APP) is examined in the micro- expression face provided, and video is analyzed, a risk is provided to the possibility of Customer Fraud
Scoring, examine personnel for face and financial institution refers to.
Computing device 1 can be that server, smart mobile phone, tablet personal computer, pocket computer, desktop PC etc. have
The terminal device of calculation function.
The computing device 1 includes:Memory 11, network interface 12, processor 13 and communication bus 14.
Camera device 3 is installed on the particular place that link is examined in execution face, such as office space, monitor area, shoots air control people
Member examines in face the video of loan customer, in the present invention, mainly shoots the positive face video of loan customer, will be shot by network 2
Obtained transmission of video is to processor 13.Network interface 12 can include standard wireline interface, wave point (such as WI-FI connects
Mouthful).Communication bus 14 is used to realize the connection communication between these components.
Memory 11 includes the readable storage medium storing program for executing of at least one type.The readable storage medium storing program for executing of at least one type
Can be such as flash memory, hard disk, multimedia card, the non-volatile memory medium of card-type memory.In certain embodiments, it is described can
Read the internal storage unit that storage medium can be the computing device 1, such as the hard disk of the computing device 1.In other realities
Apply in example, the readable storage medium storing program for executing can also be the external memory storage 11 of the computing device 1, such as the computing device 1
The plug-in type hard disk of upper outfit, intelligent memory card (Smart Media Card, SMC), secure digital (Secure Digital,
SD) block, flash card (Flash Card) etc..
In the present embodiment, the memory 11 stores micro- expression face and examines the program code of auxiliary program 10, shooting
The video that device 3 is shot, and processor 13 perform micro- expression face examine the data that are applied to of program code of auxiliary program 10 with
And data finally exported etc..
Processor 13 can be in certain embodiments a central processing unit (Central Processing Unit,
CPU), microprocessor or other data processing chips.
Fig. 1 illustrate only the computing device 1 with component 11-14, it should be understood that being not required for implementing all show
The component gone out, what can be substituted implements more or less components.
Alternatively, the computing device 1 can also include user interface, and user interface can include input block such as keyboard
(Keyboard), speech input device such as microphone (microphone) etc. has the equipment of speech identifying function, voice defeated
Go out device such as sound equipment, earphone etc., alternatively user interface can also include wireline interface, the wave point of standard.
Alternatively, the electronic installation 1 can also include display.Display can be that LED is shown in certain embodiments
Device, liquid crystal display, touch-control liquid crystal display and OLED (Organic Light-Emitting Diode, organic light emission
Diode) touch device etc..Display is used for the information and visual user interface for showing that computing device 1 is handled.
Alternatively, the computing device 1 also includes touch sensor.What the touch sensor was provided is touched for user
The region for touching operation is referred to as touch area.In addition, touch sensor described here can be resistive touch sensor, electric capacity
Formula touch sensor etc..Moreover, the touch sensor not only includes the touch sensor of contact, proximity may also comprise
Touch sensor etc..In addition, the touch sensor can be single sensor, or such as multiple biographies of array arrangement
Sensor.Personnel are examined in user, such as face, can start the careful auxiliary program 10 in micro- expression face by touching.
It is the applied environment figure that the preferred embodiment of householder method second is examined in the micro- expression face of the present invention shown in ginseng Fig. 2.Loan visitor
Family, air control personnel are remotely performed face by terminal 3 and examine link, and the camera device 30 of terminal 3 shoots air control personnel face and examined credit money visitor
The video at family, and the computing device 1 is sent to by network 2, the processor 13 of computing device 1 performs what memory 11 stored
The program code of auxiliary program 10 is examined in micro- expression face, and video is analyzed, a risk is provided to the possibility of Customer Fraud
Scoring, examine personnel for face and financial institution refers to.
The component of computing device 1 in Fig. 2, such as memory 11, network interface 12, processor 13 and the communication shown in figure
Bus 14, and the component not shown in figure, please join the introduction on Fig. 1.
The terminal 3 can be that smart mobile phone, tablet personal computer, pocket computer, desktop PC etc. have computing work(
The terminal device of energy.
In some scenes, pending object itself is undesirable, but for reached purpose, such as obtain and trust,
Preferential policy, subsidy, loan etc. are obtained, can deliberately conceal truth, there is provided deceptive information, therefore needed in review process
These objects are screened with the presence or absence of fraud suspicion.For example, when the client for fraud of attempting examines in the face of loan, can be intentional
Its mood is hidden, as client can deliberately hide the mood having a guilty conscience, is difficult to only according to human eye.However, it is possible to by it is a variety of stress,
Stimulate client to show true emotional, then by camera device 30 shoot client answer stress be during problem video, utilize
Micro- expression face examines auxiliary program 10 and video is analyzed, and auxiliary judgment client whether there is fraud suspicion.
Auxiliary program 10 is examined in micro- expression face in Fig. 1 or Fig. 2, when being performed by processor 13, realizes following steps:
Emergent question and answer step:According to the essential information of client automatically generate magnanimity stress problem and these stress problem
Model answer;
First video acquisition step:Gather client answer stress problem the first video, by it is single stress be in units of problem
First video is cut to obtain it is each stress the first video segment corresponding to problem;
Fisrt feature extraction step:Extract the first motion characteristic and that client's face includes in each first video segment
Two motion characteristics;
Model training step:The spy formed with the first motion characteristic identified in each video segment and the second motion characteristic
Subset is levied, and face examines fraud of the personnel to the video segment and is labeled as sample data, to SVMs (support
Vector model, SVM) model is trained, obtains cheating Rating Model;
Second video acquisition step:Gather client answer stress problem the second video, by it is single stress be in units of problem
Second video is cut to obtain it is each stress the second video segment corresponding to problem;
Second feature extraction step:Extracted from each second video segment the first motion characteristic for including of client's face and
The character subset that second motion characteristic is formed;
Score step:The fraud Rating Model that the character subset input training of extraction is obtained, output client answers each
Stress problem fraud appraisal result.
Client can submit some essential informations, including ID card information, address letter when applying providing a loan to financial institution
Breath, telephone number, occupation etc..According to these essential informations, it is above-mentioned stress be in question and answer step, according to the basic letter of loan customer
Cease generation stress problem and model answer, including following classification:
A. according to the ID card information of client, preceding 2 representatives of the coding criterion of identity card, such as ID card No. are passed through
Province, municipality directly under the Central Government, autonomous region, 3-4 positions represent prefecture-level city, and 5-6 positions represent county, and 7-14 positions represent date of birth, 15-16
Position represents township's political affairs, and the 17th represents sex (1- men, 2- female), the generation birthday, constellation, the place where his residence is registered, place where his residence is registered postcode,
The problems such as area code, the Chinese zodiac, age and answer;
B. according to the address information of client, by inquiring about electronic map (such as Baidu map), generate which the address nearby has
From some address (such as X hospitals) in a little association address, such as hospital, market, park, restaurant, hotel, and these addresses
To another address (such as Y parks) mode of transportation and time the problems such as and answer;
C. according to the phone number of client, by the coding criterion of phone number, for example, cell-phone number preceding 3 bit digital generation
Table operator, 4-7 positions represent the ownership place of phone number, the problem of generation ownership place, operator and answer;
D. according to the occupation of client, relevant issues and answer are generated, such as:
Client applies for that the occupation that loan is filled in is " wholesale, retail charger baby ", then can generate problems with:
Charger baby has any brand
How much is Huawei's charger baby capacity maximum
Client applies for that the occupation that loan is filled in is " pig ", then can generate problems with:
How long is pig sleep daily
How long pig grows to 100 jin of needs
Client applies for that the occupation that loan is filled in is " JAVA engineer ", then can generate problems with:
What is Spring frameworksIt please say several design patterns that you commonly use
These stress problem and its model answer, be pre-stored in the memory 11 of computing device 1, or be stored in and pass through network
The other storage devices being connected with computing device 1.
It is above-mentioned on generation stress problem be only to provide partial example, fail exhaustion.
On being discussed in detail for above-mentioned steps, explanations of the Fig. 3 on method flow diagram please be join.
As shown in figure 3, the flow chart of householder method preferred embodiment is examined for the micro- expression face of the present invention.Needing to specific right
In the case of with the presence or absence of fraud suspicion, using the framework shown in Fig. 1 or Fig. 2, while start computing device 1, and at startup
Reason device 13 performs the program code that auxiliary program 10 is examined in micro- expression face, realizes following steps:
Step S10, according to the essential information of client automatically generate magnanimity stress problem and these stress problem standard
Answer.These stress problem and its model answer can realize and be ready to, be stored in storage device, such as computing device 1 is deposited
Reservoir 11, or other distal ends that can be read by computing device 1 or local memory device.How according to the basic of loan customer
Information generation stress problem and model answer, example above please be join, repeated no more secondary.
Step S20, collection client answer stress problem the first video, by it is single stress be in units of problem to the first video
Cut to obtain it is each stress the first video segment corresponding to problem.For example, during client examines in face, it may be necessary to return
Answer tens even it is greater number of stress problem.Stress problem it is more, can more stimulate client to expose true emotional.
Stress problem show the mode of client, can be that the mode that personnel on site puts question to client is examined in face:With computing device 1
The camera device 3 that is connected record client examined in face answer instantly stress problem video;Or personnel are examined with face and remotely putd question to
The mode of client:Client examines personnel with face and is remotely connected by terminal 3 and network 2, and face is examined personnel and remotely putd question to, and utilizes terminal
3 camera device 30 or network 2 connection camera device 30 record client answer stress problem video, recording video transmission
And it is stored in the memory 11 of computing device 1.In addition, it is described stress problem can be in a manner of questionnaire in the display of terminal 3
Showing interface to client, each stress problem default Reaction time can be set, then shown more than preset time next
Stress problem, the camera device 30 that terminal 3 or network 2 connect record client answer stress problem video, the video of recording passes
Send and be stored in the memory 11 of computing device 1.
Shoot the obtained video of a client be it is continuous, for screen answer each stress problem when client true feelings
Thread, need to by it is each stress video segment corresponding to problem extract, analyze respectively.For example, by client answer stress problem " whether you
There are fixed stock assets", " your working stability", " if the funds are lent into you, whether you can be in the prescribed time-limit
It can refund" video segment, cut out one by one from total video, remain subsequent analysis.
Step S30, extract client's face includes in each first video segment the first motion characteristic and the second action spy
Sign.
First motion characteristic refers to Facial action unit (action unit, AU).No matter complete expression, or micro- table
Feelings, showed by the activity of human face's action.Borrow Ai Keman is therefore the activity of all facial muscles of the mankind
Split, numbering, be then proposed facial behavior coded system (Facial ActionCodingSystem, FACS).
Borrow Ai Keman sums up the mankind one and shares 39 main AU.Each AU, it is exactly a small group flesh of face
Meat shrinks code.For example AU1 just represents the action that brows are gathered and lifted to centre.Different AU combinations represent difference
Mood.
Second motion characteristic includes eyeball direction, head direction.
Studies have found that for most people, when the brain of people enters memory search state, that is, certain part is recalled
During the thing of necessary being, eyes meeting is first upward, turns left again.
Most people, when brain " is building " sound or image, in other words, and when people tell a lie, eye
The direction of motion of ball is probably upper right side, but if not rotating eyes, can not illustrate that it is not lied.If people are trying
Seal plays the thing occurred really, can be seen to upper left side.This " eye moves " is a kind of reflex action, is screwed up discipline unless having received,
Otherwise pretend not come.
Furthermore when fraudster is reviewed, if notice is concentrated very much, eyeball, which starts drying, can cause to blink too much, and this is
Individual fatal information leakage.
The expression of most of liar, act all with usually different, be that can find clue by effectively identifying, analyzing
's.
On the concrete principle of the first motion characteristic of extraction and the second motion characteristic, being discussed in detail for Fig. 4 and Fig. 5 please be join.
Step S40, feature formed with the first motion characteristic identified in each video segment and the second motion characteristic
Collection, and face examine fraud of the personnel to the video segment and are labeled as sample data, and supporting vector machine model is trained, obtained
Cheat Rating Model.
Personnel are examined in face can consider the data of client's application loan offer, face careful situation, history prestige of client etc., give
Each video segment distributes a fraud mark, and fraud mark represents that the client whether there is fraud suspicion, such as 1 indicates
Fraud suspicion, 0 represents without fraud suspicion.In the present embodiment, fraud mark considers the following aspects:
Client to stress problem answer whether mistake;
Client answer stress problem when mood it is whether abnormal;
Whether client has the record of overdue refund;
Electric core opinion (judging whether fraud suspicion after carrying out call-on back by phone to client, client kith and kin).
Step S10 to step S40, belong to and collect training sample, model training stage.Sample data is more, and training obtains
Fraud Rating Model it is more accurate.Afterwards, into the model application stage.
Step S50, collection client answer stress problem the second video, by it is single stress be in units of problem to the second video
Cut to obtain it is each stress the second video segment corresponding to problem.
Step S60, the first motion characteristic and the second action that client's face includes are extracted from each second video segment
The character subset that feature is formed.
Step S50 is corresponding with step S20, step S60 and step S30 are corresponding, and difference is that step S20, S30 is applied to mould
The type training stage, step S50, S60 be applied to the model application stage.
Step S70, the fraud Rating Model that the character subset input training of extraction is obtained, output client, which answers, each should
Swash the fraud appraisal result of problem.For example, output fraud appraisal result be 1 when, represent client answer stress problem when, performance
Go out is not true emotional, has fraud suspicion, when the fraud appraisal result of output is 0, represent client answer stress problem when it is anti-
Should be true.
As shown in figure 4, the particular flow sheet for the preferred embodiments of step S30 first in Fig. 3.
Step S31, facial image is extracted from the first video segment using Face datection algorithm.Face datection algorithm can be with
For the method based on geometric properties, Local Features Analysis method, eigenface method, the method based on elastic model, neutral net
Method, etc..
Step S32, extracts human face characteristic point from facial image.It is, for example, possible to use dlib algorithms are from facial image
Extract 68 human face characteristic points.Further, it is also possible to extract eyeball characteristic point.Such as one eyeball characteristic point of every eyeglasses detection.
Step S33, dimension-reduction treatment, such as PCA dimensionality reductions are carried out to the human face characteristic point of extraction.
Step S34, according to the first motion characteristic occurred in the human face characteristic point identification facial image after dimensionality reduction, obtain every
Individual first motion characteristic duration in first video segment, and according to human face characteristic point in each first motion characteristic
Coordinate calculate the second motion characteristic that first motion characteristic includes.For example, it is assumed that the first motion characteristic is AU1, i.e., " wrinkle
Eyebrow ", the information that AU1 is acquired from the video segment of client's answer living problem are as follows:" client is answering residential problem
When, occur frowning once altogether, the duration, aggregated duration 1.39 seconds ", from the video of client's answer identity verification class problem
The information that AU1 is acquired in fragment is as follows:" client occurs frowning once altogether when answering identity verification class problem, accumulative to hold
The continuous 3.64 seconds time ".Second motion characteristic, can be according to human face characteristic point in video segment such as eyeball direction, head direction
Change in location determine.
As shown in figure 5, the particular flow sheet for the preferred embodiments of step S30 second in Fig. 3.
Step S31-S34 is identical with Fig. 4, please join explanation above.
Step S35, count that the first motion characteristic and the second motion characteristic in each first video segment occur simultaneously when
Between, number.
Step S36, count two the first motion characteristic in each first video segment, the second motion characteristic motion characteristics only
There are time, the number of an appearance.
Step S35, S36 is to carry out latent structure.The video segment of process is examined in face, typically at least there is tens,
The assemblage characteristic for constructing to obtain by step S35, S36, number are usually many thousands of.
Step S37, a feature set is generated according to statistical result.
Step S38, the feature in the feature set is screened using Feature Selection algorithm, obtains character subset.It is special
It is a kind of data preprocessing method to levy selection.When the characteristic of sample is very big, may only have small part feature can be to knot
Fruit has an impact, therefore can use the quantity of feature selecting algorithm reduction feature.
Assuming that sample has n feature, then, it has 2n- a kind of possible character subset, if feature selecting needs to go thoroughly
Lift all 2nThe possible character subset of kind, in the case of n is bigger, the cost of calculating is too big, can not really realize.Therefore can
To realize the selection of feature by some algorithms.Here Feature Selection algorithm can be forward lookup/reverse search
(forward/backwardsearch), filtering characteristic selects (filter feature selection), or other are available
Feature Selection algorithm.
In addition, the embodiment of the present invention also proposes a kind of computer-readable recording medium, the computer-readable recording medium
Including micro- expression face and examine auxiliary program 10, the function of being realized when auxiliary program 10 is executed by processor is examined in micro- expression face,
It please join the above-mentioned introduction on Fig. 3~Fig. 5, will not be repeated here.
It should be noted that herein, term " comprising ", "comprising" or its any other variant are intended to non-row
His property includes, so that process, device, article or method including a series of elements not only include those key elements, and
And also include the other element being not expressly set out, or also include for this process, device, article or method institute inherently
Key element.In the absence of more restrictions, the key element limited by sentence "including a ...", it is not excluded that including this
Other identical element also be present in the process of key element, device, article or method.
The embodiments of the present invention are for illustration only, do not represent the quality of embodiment.Embodiment party more than
The description of formula, it is required general that those skilled in the art can be understood that above-described embodiment method can add by software
The mode of hardware platform is realized, naturally it is also possible to which by hardware, but the former is more preferably embodiment in many cases.It is based on
Such understanding, the part that technical scheme substantially contributes to prior art in other words can be with software products
Form embody, the computer software product is stored in a storage medium (such as ROM/RAM, magnetic disc, light as described above
Disk) in, including some instructions are make it that a station terminal equipment (can be mobile phone, computer, server, or the network equipment
Deng) perform method described in each embodiment of the present invention.
The preferred embodiments of the present invention are these are only, are not intended to limit the scope of the invention, it is every to utilize this hair
The equivalent structure or equivalent flow conversion that bright specification and accompanying drawing content are made, or directly or indirectly it is used in other related skills
Art field, is included within the scope of the present invention.