CN102622588A - Dual-certification face anti-counterfeit method and device - Google Patents

Abstract

Translated fromChinese

本发明涉及一种双验证人脸防伪方法及装置，该方法包括：步骤1，对采集的目标人脸进行活体检测，判断目标人脸是否具有生物活性，如果目标人脸被认定具有活体特性，则转入步骤2；步骤2，如果是在人脸识别应用中，则计算采集到的目标人脸与识别结果对应的人脸之间的相似度，若大于某一阈值，则认为该目标人脸是真实有效的人脸；如果是在人脸验证应用中，则计算采集到的目标人脸与目标人脸所声称的身份对应的人脸之间的相似度，若大于某一阈值，则认为该目标人脸是真实有效的人脸。利用本发明的方法，通过活体检测与身份验证的结合，提供准确、可靠的人脸防伪检测结果。

The present invention relates to a double verification face anti-counterfeiting method and device. The method includes: step 1, performing liveness detection on the collected target face, judging whether the target face has biological activity, and if the target face is determined to have living body characteristics, Then turn to step 2; step 2, if it is in the face recognition application, then calculate the similarity between the collected target face and the face corresponding to the recognition result, if it is greater than a certain threshold, then consider the target person The face is a real and effective face; if it is in the face verification application, calculate the similarity between the collected target face and the face corresponding to the identity claimed by the target face, if it is greater than a certain threshold, then It is considered that the target face is a real and valid face. The method of the invention provides accurate and reliable face anti-counterfeiting detection results through the combination of living body detection and identity verification.

Description

Two identifier's face method for anti-counterfeit and device

Technical field

The present invention relates to a kind of people's face method for anti-counterfeit and device, the especially a kind of pair of identifier's face method for anti-counterfeit and device belong to the technical field of Flame Image Process and pattern-recognition.

Background technology

People's face anti-counterfeiting technology is related to the security of recognition of face authentication and authorization system, if nobody's face antiforge function, the recognition of face authentication and authorization system is vulnerable to the attack of false people's face, and then possibly cause serious safety problem.For example, the assailant can obtain the facial image of a certain specific objective (being the nominator) and process photo, video or mask etc. through certain means, is presented in face of the system, in the hope of obtaining illegal authority.Therefore, people's face anti-counterfeiting technology receives increasing concern.Existing in the world at present people's face anti-counterfeiting technology, mainly based on the man-machine interaction strategy: system sends specific instruction, requires the user to make specific behaviors such as nictation, pronunciation, and then judges the activity of input people face.Can be divided into following three kinds of modes according to common action: first kind of live body that is based on nictation detects; Disclosing this technological document has: 1) Gang Pan; Lin Sun; Zhaohui Wu and Shilong Lao.Eyeblink-based Anti-Spoofing in Face Recognition from a Generic Webcamera, International Conference on Computer Vision, 2007; 2) K.Kollreider; H.Fronthaler and J.Bigun.Verifying Liveness by Multiple Experts in Face Biometrics, IEEE Conference on Computer Vision and Pattern Recognition Workshop, 2008; 3) patent No. is ZL200710178088.6, and denomination of invention is the patent documentation of " a kind of biopsy method and system based on the motion of people's face physiological ".Second kind is based on the live body of shaking the head and detects; Disclosing this technological pertinent literature comprises: 1) K.Kollreider; H.Fronthaler and J.Bigun.Evaluating Liveness by Face Images and the Structure Tensor; IEEE Workshop on Automatic Identification Advanced Technologies, 2005,2) Wei Bao; Hong Li; Nan Li and Wei Jiang.A Liveness Detection Method for Face Recognition Based on Optical Flow Field, International Conference on Image Analysis and Signal Processing, 2009.The third live body that is based on voice and mouth action detects; Disclosing this technological pertinent literature has: G.Chetty and M.Wagner.Liveness Verification in Audio-Video Speaker Authentication.In 10th Australian Int.Conference on Speech Science and Technology, 2004.

This method based on man-machine interaction is owing to require the user to show specific behavior, so burden for users is heavier, user experience is not good, required time is longer.

In addition; The researcher who has starts with from multispectral angle, carries out live body through the reflectivity of analyzing skin under different spectrum and detects, and pertinent literature has: 1) Ioannis Pavlidis; Peter Symosek; The Imaging Issue in an Automatic Face/Disguise Detection System, IEEE workshop on Computer Vision Beyond the Visible Spectrum:Methods and Applications, 2000.2)Youngshin?Kim，Jaekeun?Na，Seongbeak?Yoon，and?Juneho?Yi.Masked?fake?face?detection?using?radiance?measurements，J.Opt.Soc.Am，vol.26，no.4，April?2009。But this kind method is also very coarse at present, and is also unsatisfactory on the precision, also has very big room for improvement.

Above-described people's face method for anti-counterfeit can also become the live body detection technique, because they only judge whether biologically active of target people face.Yet, in the practical application, the situation that true personnel go the bogus attack nominator might appear, and this moment, target people face was real human face really, but still belonged to the behavior of attacking face identification system.Therefore people's face anti-counterfeiting technology should only not comprise the live body detection.And said method ubiquity burden for users is heavy, the man-machine interaction time is long, the not high shortcoming of accuracy, and it is imperative therefore to develop people's face method for anti-counterfeit accurate, quick, applied widely.

Summary of the invention

The objective of the invention is to overcome the deficiency that exists in the prior art, a kind of pair of identifier's face method for anti-counterfeit and device are provided, it improves identification accuracy, and is convenient, safe and reliable.

According to technical scheme provided by the invention, a kind of pair of identifier's face method for anti-counterfeit, said method comprises:

Step 1 is carried out live body to target people's face of gathering and is detected, and judges whether biologically active of target people face, if target people face is had the live body characteristic by identification, then changesstep 2 over to;

Step 2 if in face recognition application, is then calculated the similarity between the target people's face collect people's face corresponding with recognition result, as if greater than a certain threshold value, thinks that then this target people face is authentic and valid people's face;

If in people's face checking is used, then calculate the similarity between the corresponding people's face of identity that target people's face of collecting and target people face claimed, as if greater than a certain threshold value, think that then this target people face is authentic and valid people's face,

Wherein step 1 is irrelevant with the nominator, andstep 2 is relevant with the nominator, and after the checking through step 1 andstep 2 simultaneously of target people face, just can be identified as is authentic and valid people's face, is false people's face otherwise be identified as.

If said pair of identifier's face method for anti-counterfeit is based on visible light, then step 1 further comprises:

Step 101; Target people face is carried out live body detect, at first gather true in a large number, false people's face sample, target people face is extracted various textural characteristics; Training living body detects texture classifier; If target people face is detected texture classifier by live body and regards as real human face, then get intostep 2, otherwise regard as false people's face;

Step 102; Confirm the validity of target people face through man-machine interaction, wherein system sends instruction, requires the user to make certain action; System constantly detects target people face and whether makes corresponding actions then; If detect the generation of above-mentioned action within a certain period of time, judge that then target people face is a real human face, otherwise be false people's face;

Have only target people face simultaneously throughstep 101 and 102, just be considered to detect through the live body of step 1.

Step 2 further comprises:

Step 201 is at first gathered a large amount of real human face images, and every facial image is extracted its textural characteristics;

Step 202; Then the proper vector of the face images of gathering is subtracted each other in twos; Whether belong to same individual according to two images; In proper vector type of being divided into after will subtracting each other, type between two types, utilize two types of sorters of machine learning algorithm training, the sorter that obtains of training can judge whether two proper vectors of input belong to same individual thus;

Step 203 if in face recognition application, if the target facial image facial image corresponding with recognition result, is regarded as by the sorter in thestep 202 and to be belonged to same people, thinks that then target people face is authentic and valid, otherwise is false people's face;

If in people's face checking is used, then target facial image and the corresponding facial image of being claimed of nominator's identity are regarded as by the sorter in thestep 202 and to be belonged to same people, think that then target people face is authentic and valid, otherwise are false people's face.

If it is multi-modal that the said pair of identifier's face method for anti-counterfeit is based on, then step 1 further comprises:

Step 101 is judged the biologically active of target people face roughly, wherein judges according in the following mode one or more: judge through thermal infrared to judge whether the temperature of target people face near 37 degree; Through the depth information of 3D rendering judgement people face, judge whether face is the 3D object; Through the ultrasonic reflections rate of ultrasonic reflections evaluating objects people face, judge whether the ultrasonic reflections rate of skin is similar with real human face; Through the reflectivity of multispectral imaging evaluating objects people face under different spectrum; Whether the multispectral reflectivity of judging skin is similar with real human face; If judge that through above-mentioned one or more modes the information index of target people face is similar with real human face, then get intostep 102;

Step 102 is accurately judged the biologically active of target people face, with the multispectral facial image that collects, utilizes mutual quotient images algorithm to carry out accurately live body and judges,

Have only target people face simultaneously throughstep 101 and 102, just be considered to detect through the live body of step 1.

The quotient images algorithm comprises the steps: mutually

Step 1021; Gather a large amount of true man people's faces and the false people's face multispectral imaging composing training data set under different distance; Image under any two different spectrum of same individual carries out being divided by of Pixel-level, forms mutual quotient images group, supposes to select arbitrarily two spectrum lambda₁, λ₂, the image of same individual face under two spectrum does

With

Its mutual quotient images defines as follows:

${\mathrm{MQI}}_{\mathrm{\λ}1,\mathrm{\λ}2}(x,y)=\frac{I_{\mathrm{\λ}1}}{I_{\mathrm{\λ}}2}=\frac{{\mathrm{\ρ}}_{\mathrm{\λ}1}(x,y){\mathrm{\κ}}_{\mathrm{\λ}1}\left(z\right)}{{\mathrm{\ρ}}_{\mathrm{\λ}2}(x,y){\mathrm{\κ}}_{\mathrm{\λ}2}\left(z\right)}$

Wherein, ρ representes the reflectivity of people's face, and κ represents the intensity of light source in people's face surface, and z representative face is apart from the distance of light source, and (x y) represents coordinate on the facial image;

Step 1022 for all mutual quotient images, is divided into a plurality of overlapping or nonoverlapping fritters on a plurality of yardsticks, extract the proper vector of each fritter, the proper vector of all fritters is made up, as the proper vector of the overall situation;

Step 1023, based on statistical learning method, training classifier on training dataset is used to distinguish true, false people's face.

Step 2 further comprises:

Step 201 is gathered the multi-modality images of a large amount of real human face, and every image is extracted its textural characteristics;

Step 202; The proper vector of image is subtracted each other in twos; Whether belong to same individual according to two images; In proper vector type of being divided into after will subtracting each other, type between two types, utilize two types of sorters of machine learning algorithm training, the sorter that training obtains can judge whether two proper vectors of input belong to same individual;

Step 203 if in face recognition application, if the target facial image facial image corresponding with recognition result, is regarded as by the sorter in thestep 202 and to be belonged to same people, thinks that then target people face is authentic and valid, otherwise is false people's face;

If in people's face checking is used, if target facial image and the corresponding facial image of being claimed of nominator's identity, regarded as by the sorter in thestep 202 and to belong to same people, think that then target people face is authentic and valid, otherwise be false people's face.

Every kind of different imaging type is called as a mode, and imaging type comprises visual light imaging, near infrared imaging, near ultraviolet imaging, thermal infrared imaging or ultrasonic imaging.

A kind of pair of identifier's face false proof device, this device comprises:

Sensing unit is used for using one or more of near infrared, ultrasound wave, RF-wise or visible image capturing head, through the mode of real-time monitoring, and the existence of induction people face;

Multi-modal generation source comprises active light source under a plurality of spectrum, is used for one or more of required 3D structured light of 3D imaging or ultrasonic generator;

Multi-modal data acquisition equipment is used to gather the multispectral imaging of people's face, the thermal infrared photoimaging that human body itself is sent, the 3D rendering of people's face or in the ultrasonic imaging one or more;

Multi-modal people's face detecting unit is used for detecting people's face position of multi-modality images, and detected facial image is sent to the anti-dummy unit of multi-modal pair of identifier's face;

The multi-modal pair of anti-dummy unit of identifier's face is used to verify whether target people face is authentic and valid people's face;

Display unit is used to show the false proof result of people's face,

Wherein, the multi-modal pair of anti-dummy unit of identifier's face further comprises: multi-modal people's face live body detecting unit is used for that target people face is carried out live body and detects; Multi-modal people's face authentication unit is used for target people face is carried out authentication.

When said multi-modal people's face live body detecting unit carries out the live body detection to target people face; At first; The rough biologically active of judging target people face is wherein judged according in the following mode one or more: judge through thermal infrared to judge whether the temperature of target people face near 37 degree; Through the depth information of 3D rendering judgement people face, judge whether face is the 3D object; Through the ultrasonic reflections rate of ultrasonic reflections evaluating objects people face, judge whether the ultrasonic reflections rate of skin is similar with real human face; Through the reflectivity of multispectral imaging evaluating objects people face under different spectrum; Whether the multispectral reflectivity of judging skin is similar with real human face; If judge that through above-mentioned one or more modes the information index of target people face is similar with real human face; Then continue accurately to judge the biologically active of target people face,, utilize mutual quotient images algorithm to carry out live body judgement accurately the multispectral facial image that collects.

The quotient images algorithm comprises the steps: mutually

Gather a large amount of true man people's faces and the false people's face multispectral imaging composing training data set under different distance, carry out being divided by of Pixel-level, form mutual quotient images group, suppose to select arbitrarily two spectrum lambda for the image under any two different spectrum of same individual₁, λ₂, the image of same individual face under two spectrum does

With

Its mutual quotient images defines as follows:

${\mathrm{MQI}}_{\mathrm{\λ}1,\mathrm{\λ}2}(x,y)=\frac{I_{\mathrm{\λ}1}}{I_{\mathrm{\λ}2}}=\frac{{\mathrm{\ρ}}_{\mathrm{\λ}1}(x,y){\mathrm{\κ}}_{\mathrm{\λ}1}\left(z\right)}{{\mathrm{\ρ}}_{\mathrm{\λ}2}(x,y){\mathrm{\κ}}_{\mathrm{\λ}2}\left(z\right)}$

Wherein, ρ representes the reflectivity of people's face, and κ represents the intensity of light source in people's face surface, and z representative face is apart from the distance of light source, and (x y) represents coordinate on the facial image;

For all mutual quotient images, on a plurality of yardsticks, be divided into a plurality of overlapping or nonoverlapping fritters, extract the proper vector of each fritter, the proper vector of all fritters is made up, as the proper vector of the overall situation;

Based on statistical learning method, training classifier on training dataset is used to distinguish true, false people's face.

When multi-modal people's face authentication unit carries out authentication to target people face, at first gather the multi-modality images of a large amount of real human face, every image is extracted its textural characteristics; Secondly; The proper vector of image is subtracted each other in twos; Whether belong to same individual according to two images; In proper vector type of being divided into after will subtracting each other, type between two types, utilize two types of sorters of machine learning algorithm training, the sorter that training obtains can judge whether two proper vectors of input belong to same individual; If in face recognition application, if the target facial image facial image corresponding, regarded as by above-mentioned two types of sorters and to belong to same people with recognition result, think that then target people face is authentic and valid, otherwise be false people's face; If in people's face checking is used, if target facial image and the corresponding facial image of being claimed of nominator's identity, regarded as by above-mentioned two types of sorters and to belong to same people, think that then target people face is authentic and valid, otherwise be false people's face.

Every kind of different imaging type is called as a mode, and imaging type comprises visual light imaging, near infrared imaging, near ultraviolet imaging, thermal infrared imaging or ultrasonic imaging.

Advantage of the present invention: detect and the combining of authentication through live body, provide accurately, reliable people's face anti-counterfeiting detection result.

Description of drawings

The two identifier's face method for anti-counterfeit process flow diagrams under visible light that Fig. 1 proposes for the present invention;

Fig. 2 is put the fake method process flow diagram for the two identifier's faces under multi-modal that the present invention proposes;

The two identifier's face false proof device structured flowcharts under multi-modal that Fig. 3 proposes for the present invention;

The workflow diagram of the two identifier's face false proof devices under multi-modal that Fig. 4 proposes for the present invention;

The light source coverage synoptic diagram of the two identifier's face false proof devices under multi-modal that Fig. 5 proposes for the present invention;

In two identifier's face false proof device one instances that Fig. 6 proposes for the present invention human face region gradation of image average and people's face apart from harvester apart between concern synoptic diagram;

Fig. 7 is people's face reflectance curve synoptic diagram of Black people and white man in certain spectral range;

Fig. 8 is the reflectance curve synoptic diagram of several frequently seen fraud people face in certain spectral range;

Fig. 9 is the panel synoptic diagram of multi-modal harvester in two identifier's face false proof device one instances of the present invention's proposition;

Figure 10 is people's face imaging synoptic diagram of three kinds of different spectrum, from left to right is followed successively by: visible light, 850nm near infrared light and 400nm purple light;

Figure 11 face thermal infrared imaging synoptic diagram of behaving;

Figure 12 face 3D imaging synoptic diagram of behaving;

Figure 13 is that ultrasound wave is at people's reflection wave synoptic diagram on the face.

Embodiment

Below in conjunction with concrete accompanying drawing and embodiment the present invention is described further.

The ultimate principle of people's face method for anti-counterfeit that the present invention proposes is based on the thought of two checkings.So-called two identifier's face is false proof, comprises following two steps: step 1, input people face is carried out the judgement of live body, non-living body, and this step and people's identity is irrelevant, and promptly the nominator is irrelevant;Step 2 is carried out the authentication of people's face to the input facial image, has only when importing facial image and pairing identity and be complementary, and just regards as authentic and valid people's face, and this step is relevant with the nominator.Have only simultaneously that the input people face of judging through above two steps just is considered to effectively, people's face really.

What use in the step 1 is people's face live body detection technique, promptly people's face is carried out live body, non-living body judgement, differentiates whether be true man's living body faces;Step 2 is actually the checking of people's face being carried out nominator, non-nominator.Wherein instep 2, if conduct oneself the face recognition application, then recognition result is the pairing identity of target people face, and the similarity of having only both is greater than certain threshold value, and just through this step card, threshold value can be set up on their own by managerial personnel according to the actual requirements; The face checking is used if conduct oneself, and then the pairing identity of target people face is the identity that target people face is claimed, the input facial image must just be thought through the authentication of people's face greater than certain threshold value with the similarity between the corresponding identityfacial image.Step 2 is the classification to nominator and non-nominator's image.Through the information of fusion steps 1 andstep 2, reach false proof purpose reliably.

Why method of the present invention adopts above-mentioned steps 1 andstep 2 simultaneously, is that potential false people's face type can't be estimated because in practical application, and simple live body detects can't keep high-accuracy always.And on the other hand, even false people's face has passed through the live body detection, also have reason to believe this falseness people's face and have certain difference between counterfeit everybody face of appointment, therefore can further strengthen the false proof precision of people's face through extracting, differentiating this difference.Therefore the facial image that proposes input needs to detect and the authentication of people's face through people's face live body simultaneously, just can regard as authentic and valid people's face.

Traditional false proof research of people's face also rests on people's face live body and detects, and has ignored the checking to input people face.In fact, people's face live body of step 1 detects, and is irrelevant with the nominator; People's face authentication ofstep 2 then is the authentication to the nominator.People's face anti-counterfeiting technology of the present invention detects in conjunction with authentication of people's face and live body, can effectively improve the false proof reliability of people's face.

In of the present invention pair of identifier's face method for anti-counterfeit, concrete application form and the concrete application form under multi-modal under visible light have been proposed further.

Two identifier's face method for anti-counterfeit under the visible light are applicable to traditional visible light recognition of face, people's face verification system, need not additional hardware and can accomplish the false proof task of people's face.False facial image can be regarded as the real human face image at the image through obtaining after certain aftertreatment, and therefore comparing its picture quality of true picture will have certain loss.Through the target people's face that captures is extracted polytype texture information, can fully excavate the appearance features of target people face on the dermatoglyph details, and then further classify according to pre-set evaluation criteria.In addition, also can further strengthen its accuracy through introducing traditional man-machine interaction process.

For the present invention propose multi-modal under two identifier's face method for anti-counterfeit, then further adopt multiple modalities fully to excavate people's face essential characteristic.Existing recognition of face, people's face verification technique also only rest on utilizes a kind of mode (for example visible light or near infrared) to obtain facial image.Therefore we think that this data acquisition modes can not fully excavate the skin properties of people's face, can not reach higher anti-spurious accuracy, propose and designed the two identifier's face false proof devices under multi-modal.This device comprises multispectral imaging device under the different spectrum, thermal infrared imaging device, ultrasonic imaging apparatus or the like, fully excavates the essential physical characteristics of people's face skin from different aspects.Through anatomizing true man people's face and the characteristic of typical false people's face under different modalities, choose the combination of suitable mode, the characteristic of tool resolving ability is provided for the follow-up false proof algorithm of people's face.

Two identifier's face method for anti-counterfeit under the visible light that the present invention proposes can through to the analysis of dermatoglyph details and/or the motion of people's face, accurately be judged the true and false of target people face on the basis that does not rely on additional hardware.And the present invention proposes based on multi-modal two identifier's face method for anti-counterfeit, with respect to existing people's face live body detection algorithm, not only can defend more attack type, and have few, characteristics such as user experience is good, accuracy rate height of time spent.Through multi-modal people's face information of obtaining abundanter people's face information can be provided, fully excavate the essential characteristic of people's face, increase the discrimination of true man people's face and false people's face, can effectively solve the false proof difficult problem of people's face.

The concrete applicating flow chart of two identifier's face method for anti-counterfeit under visible light that Fig. 1 proposes for the present invention.With reference to Fig. 1, detect in thestep 101 at live body, the people's face live body that uses dermatoglyph to combine with facial movement detectsstrategy.In authentication step 102; The pairing identity of target people face (checking of people's face is the identity of being claimed in using, and is the corresponding identity of recognition result in the face recognition application) is verified, if matching similarity is greater than certain threshold value; Then think real human face, otherwise be false people's face.Having only target people face to pass through 101 and 102 liang of step ability assertive goal people faces simultaneously is real human face.

Live body detectsstep 101 and further comprisesstep 1011 and step 1012:step 1011; At first target people face is extracted various textural characteristics; For example LBP (Local Binary Pattern), HOG (Histograms of Oriented Gradients) characteristic etc. obtain the liveness detector based on dermatoglyph through gathering true false people's face sample through machine learning algorithm (like SVMs SVM) training then.If judge it is real human face, get intostep 1012.

An instance ofstep 1011 is; Utilize the LBP descriptor of different scale; For example

carries out filtering to the target facial image, then image carried out multiple dimensioned division, for example is divided into 1 * 1; 3 * 3; 5 * 5 fritter at the histogram of three kinds of LBP descriptors of each piece the inside statistics, is linked at together the textural characteristics as target people face to all histograms.

Gather the image of true in a large number, false people's face then, for example, gather 50 people's real human face image, utilize its facial image to be made into the photo of different sizes then, and then gather photograph image.Take out human face region, extracted characteristic according to the operation in a last step.Utilize the SVM algorithm training to obtain a sorter then.

Instep 1012, utilize man-machine interaction further to detect the biologically active of target people face.For example, can give the instruction of selling user's nictation or shaking the head through face identification system.Whether made corresponding actions through detecting target people face, thereby judged whether target people face is real human face.In this step, can utilize estimation or template matching algorithm to carry out facial movement and estimate.For example, if adopt the form of nictation, can utilize optical flow method to calculate the motion vector of target people face eye areas, and then judge whether to have taken place action nictation.Perhaps template matching algorithm, what a sorter of opening eyes, closing one's eyes of training in advance carries out motion detection then.

The mutual instance of one personal-machine is that face identification system provides the command request user within a certain period of time, for example 5 seconds, blinks.Through the human eye state sorter that trains, detect and whether to have occurred in the period at this section opening eyes-to close one's eyes-process of opening eyes.If occur, then think real human face, otherwise then think false people's face, get into step 1021.Human eye state sorter wherein above-mentioned can be collected in a large number the image of opening eyes, close one's eyes in advance, utilizes the training of svm classifier device to obtain the sorter of eye state then, is used for above-mentioned blink detection.

Inauthentication step 102; Facial image in the database is extracted characteristic (for example, LBP and Gabor characteristic), then the proper vector of the face images of gathering is subtracted each other in twos; Whether belong to same individual according to two images; In proper vector type of being divided into after will subtracting each other, type between two types, utilize two types of sorters of machine learning algorithm training, the sorter that obtains of training can judge whether two proper vectors of input belong to same individual thus;

Through above step, the similarity that belongs between same people's the face characteristic should be greater than the similarity between the face characteristic of different people.Through setting a rational threshold value, can be used for authentication: if the similarity betweenstep 102 target people's face and its identity of claiming is then thought and passed through authentication greater than threshold value; Otherwise failure.

Fig. 2 is the two applicating flow charts of identifier's face method for anti-counterfeit under multi-modal form.This method adopts multi-modal as carrier; Gather multi-modal facial image; Utilize abundant information that multi-modality images provides and utilize different biological features to have the characteristics of different physical characteristicss, through multi-modal information fusion, designed rationally, reliably two false proof algorithms of identifier's face.

Two identifier's face method for anti-counterfeit under multi-modal form comprise livebody verification step 201 andauthentication step 202 liang step.

In biological information checking 201, adopt by slightly going on foot strategy to two of essence.

At first, utilize multi-modal people's face information of being obtained, the live body characteristic of input people face is judged roughly in the first step 2011.An instance is: at first carry out temperature detection through thermal infrared images; If meet the temperature range (whether being 37 degree for example) of real human body, then carry out the judgement of people's face depth information, if judge that input people face is a three-dimensional body through the 3D facial image; Then continue to utilize the ultrasonic reflections rate of ultrasonic reflections wave measurement input people face; If whether in the reasonable scope reflectivity and true man's human face similarity then check its multispectral the average image brightness, if rationally; Then be judged as true man people's face, otherwise be false people's face.In this step, can choose people's face live body characteristic of judging as rough according to specific people's face modal dynamic.

After thefirst step 2011 was regarded as true man people's face, insecond step 2012, to the multi-modality imaging of people's face, the present invention proposed the people's face live body detection algorithm based on mutual quotient images, provides more accurate, meticulous testing result; If the quotient images algorithm judges that this person's face is true man people's face mutually, then people's face biologically active is imported in explanation.If be judged as true man people's face insecond step 2012, then be true man people's face, otherwise be false people's face.

Instep 2012, utilize mutual quotient images algorithm to carry out the Human Face live body and detect.Mutually quotient images be meant image under any two spectrum carry out the relevant position pixel value do the resulting image of division (Mutual Quotient Image, MQI).Quotient images can reflect the relation of shooting people face between two wave band reflectivity mutually, and irrelevant with the shape of people's face.According to the definition of mutual quotient images, suppose to select arbitrarily two spectrum lambda₁, λ₂, the image of same individual face under two spectrum does

With

Its mutual quotient images defines as follows:

${\mathrm{MQI}}_{\mathrm{\λ}1,\mathrm{\λ}2}(x,y)=\frac{I_{\mathrm{\λ}1}}{I_{\mathrm{\λ}2}}=\frac{{\mathrm{\ρ}}_{\mathrm{\λ}1}(x,y){\mathrm{\κ}}_{\mathrm{\λ}1}\left(z\right)}{{\mathrm{\ρ}}_{\mathrm{\λ}2}(x,y){\mathrm{\κ}}_{\mathrm{\λ}2}\left(z\right)}---\left(4\right)$

Wherein, ρ representes the reflectivity of people's face, and κ represents the intensity of light source in people's face surface, the distance of z representative face between the light source, and (x y) represents coordinate on the facial image.

If guarantee λ₁, λ₂The light source luminescent power of two spectrum is consistent, then in suitable distance range, and λ₁, λ₂It is 1 that the ratio of the intensity of two kinds of light sources waits approximately, and therefore (4) formula can approximate

${\mathrm{MQI}}_{\mathrm{\λ}1,\mathrm{\λ}2}(x,y)=\frac{I_{\mathrm{\λ}1}}{I_{\mathrm{\λ}2}}\≈\frac{{\mathrm{\ρ}}_{\mathrm{\λ}1}(x,y)}{{\mathrm{\ρ}}_{\mathrm{\λ}2}(x,y)}---\left(5\right)$

Can find out that this moment, quotient images reflected that people's face is at λ mutually₁, λ₂Therefore the ratio of the reflectivity under two kinds of spectrum is a characteristic that can reflect people's face intrinsic propesties, can be used for designing the live body detection algorithm.

In the derivation of formula (5), suppose in suitable distance range λ₁, λ₂It is 1 that the ratio of the intensity of two kinds of light sources waits approximately.Through the appropriate design light source, can in reality, satisfy this hypothesis.For example; The present invention has gathered 480nm and 850nm two kinds of light sources under the consistent situation of luminous power; Apart from light source 40cm between the 90cm, the situation of change of same individual's facial image gray average, as shown in Figure 6; Can find out that it is that 1 hypothesis is rational that the ratio of the intensity of two kinds of light sources waits approximately.

In multi-modal people's face live body detection algorithm, after having obtained the mutual quotient images of any two spectrum, can characteristic reasonable in design, detect so that carry out live body.Proper vector is extracted can adopt several different methods, as: intensity histogram, Gabor wave filter etc., likelihood ratio (Likelihood Ratio) etc.After selected characteristic type; Can carry out piecemeal to mutual quotient images, and do multiple dimensioned processing, obtain on the different scale, the mutual quotient images proper vector of people's face of diverse location; True, the dummy's face sample of a large amount of then collections utilizes the Boosting algorithm to carry out the training that live body detects sorter.

In multi-modal people's face live body detection algorithm, should take into full account difference in reflectivity true, fraud people face, carry out light source and select.Fig. 7 illustration people's face reflectance curve of multispectral down Black people and white man.Fig. 8 illustration the reflectance curve of multispectral down several frequently seen fraud people face, comprise two kinds of different silica gel and photo.According to this two width of cloth curve, can select to provide foundation for the spectrum in multi-modal people's face live body detection.

Idiographic flow based on people's face live body detection algorithm of mutual quotient images is following:

(1), gather a large amount of true man people's faces and the reflection intensity data composing training data set of fraud people face under different distance, carry out MQI calculating for the image any two different spectrum of same individual under.

(2), on all MQI images, on a plurality of yardsticks, be divided into a plurality of fritters (overlapping or not overlapping), extract the proper vector of each fritter, the proper vector of all fritters is made up, as the proper vector of the overall situation.

(3), based on statistical learning method, training classifier on training dataset is like SVM (SVMs), LDA (linear discriminant analysis), Boosting etc.

Further specify the mutual quotient images algorithm that live body detectsstep 2012 below by way of example.

For example, adopt two kinds of light sources of 480nm and 940nm to be carried out to picture, the facial image of acquisition is respectively I₄₈₀, I₉₄₀Stipulate I then₄₈₀Be reference picture, the mutual quotient images that calculates under these two wave bands is MQI_940,480(x, y)=I₉₄₀(x, y)/I₄₈₀(x, y).The present invention has only provided the situation of two kinds of wave bands by way of example at this, also can select the light source of any multiple wave band according to actual conditions.

128 * 128 MQI image carries out multiple dimensioned processing through after the pre-service, is divided into 5 yardsticks, and its size is respectively 128 * 128 pixels, 64 * 64 pixels, 32 * 32 pixels, 16 * 16 pixels, 8 * 8 pixels.Be based on the probability model that obtains through statistical learning on the training set; For the every bit on the mutual quotient images; Can calculate its likelihood that belongs to live body and non-living body

wherein the G representative image from live body;

representative image is from non-living body; (x y) is image coordinate.These two amounts are divided by, can obtain the local likelihood ratio of mutual quotient images:

$r(x,y,\mathrm{\σ})=\frac{p(x,y,\mathrm{\σ}|G)}{p(x,y,\mathrm{\σ}|\stackrel{\‾}{G})}---\left(6\right)$

For the mutual quotient images of above-mentioned multiresolution, all local likelihood ratios can constitute a live body proper vector, and its dimension is 21824.

In order to make characteristic have more discrimination and to have higher operation efficiency, the live body feature extraction algorithm utilizes Boosting to carry out feature selecting, from original high-dimensional characteristic, selects 3000 dimensional features of tool resolving ability.

Gather a large amount of true, dummy's face samples then; Set up tranining database; Characteristic label after selecting according to above-mentioned Boosting carries out feature extraction; And (Support Vector Machine, SVM) methodology acquistion to one two types of sorters are used for the proper vector of input is carried out the judgement of live body, non-living body to utilize support vector machine.

Inauthentication step 202, need carry out the similarity checking to input people face and its pairing identity.Concrete verification algorithm and the verification method under thevisible light 102 are similar, and difference is the summation that is characterized as all characteristics on the multi-modality images imported.

Have only the biological information of working as checking and two steps of authentication to assert that all input people face is true man people's face, input people face is just calculated through the false proof judgement of people's face.

Further specify the people's face verification algorithm in theauthentication step 202 below by way of example, wherein be applied as example with the checking of people's face.

Suppose that everyone has N to open the image of different modalities, at first every facial image is carried out LBP characteristic and Gabor feature extraction, form the proper vector f of this image_k, k=1:N.

The proper vector that will belong to every image in the same multi-modality images combination then is concatenated into forms unified proper vector F=[f₁...; f_N], then F is everyone multi-modal proper vector.

In the training process of people's face checking sorter, positive sample is multispectral proper vector F poor that belongs to same individual, and negative sample is poor for the multispectral proper vector F's that do not belong to same individual.Utilize the Boosting algorithm to carry out characteristic and select, obtain a character subset.

Everyone multi-modality images that training data is concentrated, the sample of selecting according to Boosting carries out feature extraction, and utilizes the LDA algorithm to carry out discriminatory analysis.

Through above step, the similarity that belongs between same people's the face characteristic should be greater than the face characteristic similarity between the different people.If the similarity instep 202 between target people face and its identity of claiming is then thought and has been passed through authentication greater than threshold value; Otherwise failure.

The invention allows for a kind of pair of identifier's face false proof device.Fig. 3 is the structured flowchart that the present invention is based on multi-modal two identifier's face false proof devices.Fig. 4 is the workflow diagram based on multi-modal two identifier's face false proof devices of the present invention.

Of the present invention based on multi-modal two identifier's face false proof devices in, one or more in the mode such as that wherein multi-modal comprises is multispectral, 3D, ultrasound wave.Because people's face skin has different reflectivity under different spectrum; Therefore the present invention introduces multispectral people's face imaging system; Be used to gather, the imaging of analyst's face under different spectrum, the intrinsic propesties that fully excavates people's face, thus abundant face characteristic is provided for follow-up people's face is false proof.Choosing of spectrum can comprise near infrared light, mid-infrared light, far infrared (thermal infrared), black light or the like, to reflect the different reflection characteristics of people's face as far as possible.Special, thermal infrared images refers to the infrared light imaging that the human body self heat is given out, relevant with individual's physique, biological properties, the remarkable individual difference of tool is suitable as the false proof foundation of people's face.Above light source all needs multispectral acquisition system that initiatively light source is provided except that TIR.

The present invention introduces the 3D facial image simultaneously, and ultrasonic imaging, with multispectral image, has constituted multi-modal facial image jointly and has obtained system.The depth information of the people face position that obtains through 3D rendering is the false proof important evidence of people's face, can resist the attack of common false people's face, for example photo, video etc.The method of ultrasonic imaging through measuring people's face skin for hyperacoustic reflectivity, can provide the physical characteristics tolerance means of other a kind of people's face skin, the demand that further auxiliary people's face live body detects.

In conjunction with Fig. 3 and Fig. 4; Of the present inventionly comprise sensingunit 301 based on multi-modal two identifier's face false proof devices;Multi-modal generation source 302, multi-modaldata acquisition equipment 303; Multi-modal people'sface detecting unit 304, the multi-modal pair of anti-dummy unit 305 (comprising multi-modal people's face livebody detecting unit 3051, multi-modal people's face identity authenticating unit 3052) of identifier's face,control module 306 anddisplay unit 307.

Sensing unit 301 is used to use near infrared, ultrasound wave or RF-wise to carry out the biological characteristic induction, perhaps uses the visible image capturing head to monitor in real time.This unit if sensed people's face, then sends the signal that object exists to controlmodule 306 in order to the existence at specific induction region internal induction people face.In fact, sensingunit 301 can not judge that what sense is people's face, as long as there is object to appear in the induction zone, just thinks to have sensed people'sface.Sensing unit 301 can use modes such as near infrared, ultrasound wave or radio frequency to carry out the induction of people's face, also can simply use the visible image capturing head to monitor in real time.Be set at the size of specific induction region and optimum seeking site and can catch whole people's face.

Sensing unit 301 is sensed the existence of people's face, operation below concrete the execution: if the current people's face that do not detect of step 1. exists, then continue cycle detection; If detect the existence of people's face, then changestep 2 over to; Step 2 is waited for certain hour, and then detects people's face, if people's face still exists, then thinks effective people's face, and concurrent feed signals is givencontrol module 306; If people's face no longer exists, then think invalid people's face, change step 1 over to and restart to detect.

In one example, sensingunit 301 is the visible image capturing head, and its mode with monitoring is carried out the induction of people's face.A visible image capturing circle collection image also detects whether there is people's face.If there is not people's face, then continues to gather visible images and carry out the detection of people's face; If there is people's face, then wait for images acquired and detect people's face once more in 0.5 second.If this moment, people's face also existed, then explanation has stable, effective people's face to occur, and sends signal then and givescontrol module 306, begins corresponding IMAQ work; If people's face disappears after waiting for, explain that this person's face probably is not the people's face that carries out the multi-modality images collection, think noise and ignore.Continue to gather visible images and detect whether there is people's face.

Multi-modal generation source 302 can include, but is not limited to following one or more equipment: the active light source under a plurality of spectrum (providing multispectral imaging required illumination) is used for the required 3D structured light of 3D imaging, ultrasonic generator (in order to the emission ultrasound wave).In multispectral light source; Spectral combination can comprise visible light (need not provide visible light source this moment); But must comprise the combination of one or more non-visible light light source, the light source light spectrum scope can be near-infrared band (740nm-4000nm), or near ultraviolet band (360-400nm).Also can comprise thermal infrared imaging, this moment, TIR was sent by human body, needn't set up additional light source again.But spectral combination should not comprise harmful light, for example medium ultraviolet light (290-320nm wavelength) or black light (200nm-290nm wavelength).The 3D structured light can dispose according to the actual requirements, for example line laser or 3DNIR structured light.The frequency of ultrasonic generator is set according to the actual requirements, for example, can be made as 50kHz.

For multispectral light source wherein, the light that light source sends should meet two principles: 1, in the suitable distance scope, in plane, 302 dead ahead, multi-modal generation source, certain area planted agent keeps light intensity roughly even.As shown in Figure 5, certain distance (d) is located in the collecting device dead ahead, and (circular shown in the figure) light intensity should keep even in certain area.2, luminous intensity should keep in the reasonable scope, makes imaging device can collect facial image clearly, and it is too big and cause that the user's is uncomfortable to be unlikely to light intensity again.

Multi-modaldata acquisition equipment 303 is used to gather the multispectral light that the active light source irradiation reflects the people on the face then, also is used to gather the thermal infrared light that human body itself is sent in addition, the 3D rendering of people's face, and the ultrasonic imaging of people's face.This collecting device is including, but not limited to following one or more units: respond the camera of each source light, the receiving tube that responds each spectrum light or photodiode, thermal infrared induction camera or inductor, 3D rendering collecting device, supersonic imaging device or receiver.

Multi-modaldata acquisition unit 303 comprises that at first imaging device corresponding to each spectrum in 302 in order to gather the multispectral light of people's face reflection, comprises imaging device and corresponding filter disc, comprises thermal infrared, 3D, supersonic imaging device or inductor in addition.The camera of multispectral imaging evaluation method selecting optimal equipment good response multispectral light source light, this moment, the return data type was an image.If condition is limited, also can use other receiving equipment, for example respond the receiving tube, photodiode of multispectral light etc., this moment, the return data type was the reflection strength scalar.A kind of light source in the multispectral light source can corresponding camera, also can utilize single camera to respond the multispectral light source of a plurality of wave bands.Camera should have higher sensitivity at the spectrum place that is responded.For supersonic imaging device, should be consistent with the ultrasonic generator holding frequency in 302; If condition does not allow, also can select ultrasonic receiver for use.For thermal infrared, preferred thermal infrared camera, also can select for use can temperature sensor inductor.For the 3D camera, what then collect is the image of reflection people face depth information.

In the multispectral imaging equipment of multi-modaldata acquisition unit 303, need to be equipped with the filter disc of corresponding wave band, in order to eliminate the interference of surround lighting and other wave band light to this wave band.Filter disc should be placed on the imaging device front of corresponding wave band, and is close to cam lens or receiving equipment, gets into to prevent veiling glare.

The multi-modal people's face of sensingunit detecting unit 304; Be used for the facial image of multi-modality images imaging device collection is carried out pre-service; Then to detecting through pretreated facial image, all be detected when face images and think the detected people's of being face under the situation of people's face and eyes.

Multi-modal two identifier's faces are preventeddummy unit 305, comprise 3052 liang of sub-cells of multi-modal people's facelive body detection 3051 and multi-modal people's face checking.Wherein detect in 3051 at multi-modal people's face live body, that mentions in the employing preceding text designs suitable multi-modal people's face live body sorter by slightly going on foot strategy to two of essence; Multi-modal people's faceidentity authenticating unit 3052 is extracted from multi-modal facial image and can be confirmed that the information of target identities carries out the authentication of people's face.Wherein, multi-modal people's face livebody detecting unit 3051 and multi-modal people's faceidentity authenticating unit 3052 have been formed therealization unit 305 based on the multi-modal false proof algorithm of two identifier's faces of the present invention jointly.

Control module 306 is used to control the duty of each unit, the work such as information communication between the unit;Display unit 307 is used on output medium, showing intermediate result, makes things convenient for user inquiring.

Control module 306 is in order to the duty that realizesmulti-modal generation source 302 and the control of multi-modal data acquisition unit 303.Can use Single-chip Controlling, also can adopt PC to connect control.

With reference to Fig. 3 and Fig. 4; The control mode ofcontrol module 306 is: after there is signal in the people's face that receives sensingunit 301 transmissions; At first provide control signal, open the light source of spectrum 1, wait for that then the regular hour gives the camera exposure; Gather picture signal then, close the light source of spectrum 1 then corresponding to the camera of spectrum 1.Provide signal then, open the light source ofspectrum 2, wait for certain time shutter, hold and gather picture signal, close the light source ofspectrum 2 then corresponding to the camera ofspectrum 2, and the like, finish up to the image data acquiring of all spectrum.If do not use camera under the some spectrum, and be to use other receiving equipment,, then read corresponding receiving intensity numerical value like receiving tube, photodiode etc.Control the thermal infrared camera then and carry out IMAQ.After thermal infrared, control 3D video camera carries out the 3D man face image acquiring.Control the ultrasonic emitting ultrasound wave then, and be carried out to picture with supersonic imaging device.

An instance is:control module 306 is made up of host computer PC endsoftware.Control module 306 at first provides the open command of light source 1 after the signal that receives sensingunit 301 transmissions, wait for that 50ms provides the acquisition of the camera (or receiving tube) of corresponding light source 1 then, by camera (or receiving tube) image data.Make light source 1 extinguish then, provide the open command oflight source 2, wait for 50ms, make the camera (or receiving tube) oflight source 2 carry out data acquisition.And the like, till the camera of all light sources all collects data.When gathering thermal infrared and 3D rendering then, need not wait for and can directly gather this moment.Open ultrasonic transmitter then, and echo is received and form images through supersonic imaging device.Then controlmodule 306 can with each camera collection to view data send into multi-modal people'sface detecting unit 304.

The facial image that displayunit 307 is gathered by multi-modaldata acquisition unit 303 in order to demonstration, and provide various intermediate results or feedback information, make things convenient for man-machine interaction.

It should be noted that if certain above-mentioned mode does not have corresponding image data acquiring equipment, also can use other non-image formula induction instrument aratus to replace.

Fig. 9 has provided the synoptic diagram of multi-modal generation source and multi-modal data acquisition unit by way of example.Wherein, multi-modality images harvester panel 804 plays the effect of device frame.Panel is divided into two parts up and down; The first half is multi-modal generation source 901 and multi-modal data acquisition unit 902, and the latter half is a display unit 905, is made up of a lcd screen; Be a visible image capturing 903 between two parts, use as sensing unit.In the first half, three multi-modal emissive sources are respectively 800nm multispectral light source, 3D structure light source and ultrasonic emitting source.Three kinds of emissive source cross arrangements, and form rectangle, can guarantee that like this each emissive source can form even distribution in the certain limit of device the place ahead.Central authorities are four imaging devices (or receiving equipment) launching in a steady stream; Comprise multispectral imaging equipment (camera the place ahead all covers the filter disc of corresponding wave band, to prevent the interference of visible light or other spectrum light), thermal infrared camera (being used to gather thermal infrared images), 3D and supersonic imaging device.When testing, people's face should be positive in the face of this harvester.Control module is not comprised on the face version of multispectral harvester, but an independent one-tenth part (can be single-chip microcomputer, also can be upper computer software) is connected through control signal wire with multispectral harvester panel.

Multi-modal people's face detecting unit, the multi-modal pair of anti-dummy unit of identifier's face are the application program of host computer, after receiving the multi-modal facial image that collects, deliver to above two unit respectively, and provide corresponding results.

Above-mentioned workflow based on multi-modal two identifier's face false proof devices is as shown in Figure 4.With reference to Fig. 4, at first respond to the existence of people's faces by sensing unit 401; If there is not people's face, then continue cycle detection, and in fact, sensing unit 401 can not be judged the detected people's of being face, just when having sensed the object existence, promptly think to sense people's face; If there is people's face, then give an order to control module 402, send control command by control module 402, instruct multi-modal generation source 403 to open, close, and multi-modal data acquisition unit 404 image data; Get into multi-modal people's face detecting unit 405 then and carry out the detection of people's face,, then signal to display unit 407 if do not detect people's face in the image that has.Output detects the information of failure, and returns sensing unit 401, carries out IMAQ again; If all mode images all detect people's face, then get into the anti-dummy unit 406 of multi-modal pair of identifier's face, subsequent signal is given display unit 407, so that output people face detects certain facial image of information or display capture; Carry out people's face live body detection judgement 4061 and people's face authentication 4062 after getting into the anti-dummy unit 406 of multi-modal pair of identifier's face; If provide the information of corresponding live body detection failure for fraud people face then through display unit 407; And return sensing unit 401, carry out the IMAQ of a new round; If also provide by display unit 407 for true man people's face, wait for a period of time then, return sensing unit 401, people's face of a beginning new round detects.

In one example, multi-modal people'sface detecting unit 405 is the application program of host computer PC end, and the corresponding people's face of every image call that is used for thatmulti-modal data collector 403 is collected detects sorter and carries out the detection of people's face.If all images all detects people's face, carry certain facial image to be used for showing (for example, selecting the facial image under the visible light for use) fordisplay unit 407, and the facial image under detected all spectrum is inputed to theanti-dummy unit 406 of multi-modal pair of identifier's face.If all do not detect people's face, then carry the result who detects failure to show fordisplay unit 407, and returnsensing unit 401, restart image sensing.

At last, the present invention must point out, the two identifier's face method for anti-counterfeit and the device thereof that utilize the present invention to propose, and the user can be applicable to different biological mode, for example people's face, iris etc. according to the needs of oneself.And can freely select the mode combination according to actual conditions, for example, can select the different spectrum combination separately for use, also can combine thermal infrared light, 3D rendering or ultrasonic imaging to unite use.

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

Claims (12)

1. two identifier's face method for anti-counterfeit is characterized in that said method comprises:

Step 1 is carried out live body to target people's face of gathering and is detected, and judges whether biologically active of target people face, if target people face is had the live body characteristic by identification, then changes step 2 over to;

Step 2 if in face recognition application, is then calculated the similarity between the target people's face collect people's face corresponding with recognition result, as if greater than a certain threshold value, thinks that then this target people face is authentic and valid people's face;

If in people's face checking is used, then calculate the similarity between the corresponding people's face of identity that target people's face of collecting and target people face claimed, as if greater than a certain threshold value, think that then this target people face is authentic and valid people's face,

Wherein step 1 is irrelevant with the nominator, and step 2 is relevant with the nominator, and after the checking through step 1 and step 2 simultaneously of target people face, just can be identified as is authentic and valid people's face, is false people's face otherwise be identified as.

2. according to claim 1 pair of identifier's face method for anti-counterfeit is characterized in that, if said pair of identifier's face method for anti-counterfeit is based on visible light, then step 1 further comprises:

Step 101; Target people face is carried out live body detect, at first gather true in a large number, false people's face sample, target people face is extracted various textural characteristics; Training living body detects texture classifier; If target people face is detected texture classifier by live body and regards as real human face, then get into step 2, otherwise regard as false people's face;

Step 102; Confirm the validity of target people face through man-machine interaction, wherein system sends instruction, requires the user to make certain action; System constantly detects target people face and whether makes corresponding actions then; If detect the generation of above-mentioned action within a certain period of time, judge that then target people face is a real human face, otherwise be false people's face;

Have only target people face simultaneously through step 101 and 102, just be considered to detect through the live body of step 1.

3. according to claim 2 pair of identifier's face method for anti-counterfeit is characterized in that step 2 further comprises:

Step 201 is at first gathered a large amount of real human face images, and every facial image is extracted its textural characteristics;

Step 202; Then the proper vector of the face images of gathering is subtracted each other in twos; Whether belong to same individual according to two images; In proper vector type of being divided into after will subtracting each other, type between two types, utilize two types of sorters of machine learning algorithm training, the sorter that obtains of training can judge whether two proper vectors of input belong to same individual thus;

Step 203 if in face recognition application, if the target facial image facial image corresponding with recognition result, is regarded as by the sorter in the step 202 and to be belonged to same people, thinks that then target people face is authentic and valid, otherwise is false people's face;

If in people's face checking is used, then target facial image and the corresponding facial image of being claimed of nominator's identity are regarded as by the sorter in the step 202 and to be belonged to same people, think that then target people face is authentic and valid, otherwise are false people's face.

4. according to claim 1 pair of identifier's face method for anti-counterfeit is characterized in that, multi-modal if the said pair of identifier's face method for anti-counterfeit is based on, then step 1 further comprises:

Step 101 is judged the biologically active of target people face roughly, wherein judges according in the following mode one or more: judge through thermal infrared to judge whether the temperature of target people face near 37 degree; Through the depth information of 3D rendering judgement people face, judge whether face is the 3D object; Through the ultrasonic reflections rate of ultrasonic reflections evaluating objects people face, judge whether the ultrasonic reflections rate of skin is similar with real human face; Through the reflectivity of multispectral imaging evaluating objects people face under different spectrum; Whether the multispectral reflectivity of judging skin is similar with real human face; If judge that through above-mentioned one or more modes the information index of target people face is similar with real human face, then get into step 102;

Step 102 is accurately judged the biologically active of target people face, with the multispectral facial image that collects, utilizes mutual quotient images algorithm to carry out accurately live body and judges,

Have only target people face simultaneously through step 101 and 102, just be considered to detect through the live body of step 1.

5. according to claim 4 pair of identifier's face method for anti-counterfeit is characterized in that, the quotient images algorithm comprises the steps: mutually

Step 1021; Gather a large amount of true man people's faces and the false people's face multispectral imaging composing training data set under different distance; Image under any two different spectrum of same individual carries out being divided by of Pixel-level, forms mutual quotient images group, supposes to select arbitrarily two spectrum lambda₁, λ₂, the image of same individual face under two spectrum does

With

Its mutual quotient images defines as follows:

${\mathrm{MQI}}_{\mathrm{\λ}1,\mathrm{\λ}2}(x,y)=\frac{I_{\mathrm{\λ}1}}{I_{\mathrm{\λ}2}}=\frac{{\mathrm{\ρ}}_{\mathrm{\λ}1}(x,y){\mathrm{\κ}}_{\mathrm{\λ}1}\left(z\right)}{{\mathrm{\ρ}}_{\mathrm{\λ}2}(x,y){\mathrm{\κ}}_{\mathrm{\λ}2}\left(z\right)}$

Wherein, ρ representes the reflectivity of people's face, and κ represents the intensity of light source in people's face surface, and z representative face is apart from the distance of light source, and (x y) represents coordinate on the facial image;

Step 1022 for all mutual quotient images, is divided into a plurality of overlapping or nonoverlapping fritters on a plurality of yardsticks, extract the proper vector of each fritter, the proper vector of all fritters is made up, as the proper vector of the overall situation;

Step 1023, based on statistical learning method, training classifier on training dataset is used to distinguish true, false people's face.

6. according to claim 4 pair of identifier's face method for anti-counterfeit is characterized in that step 2 further comprises:

Step 201 is gathered the multi-modality images of a large amount of real human face, and every image is extracted its textural characteristics;

Step 202; The proper vector of image is subtracted each other in twos; Whether belong to same individual according to two images; In proper vector type of being divided into after will subtracting each other, type between two types, utilize two types of sorters of machine learning algorithm training, the sorter that training obtains can judge whether two proper vectors of input belong to same individual;

Step 203 if in face recognition application, if the target facial image facial image corresponding with recognition result, is regarded as by the sorter in the step 202 and to be belonged to same people, thinks that then target people face is authentic and valid, otherwise is false people's face;

If in people's face checking is used, if target facial image and the corresponding facial image of being claimed of nominator's identity, regarded as by the sorter in the step 202 and to belong to same people, think that then target people face is authentic and valid, otherwise be false people's face.

7. according to claim 4 pair of identifier's face method for anti-counterfeit is characterized in that, every kind of different imaging type is called as a mode, and imaging type comprises visual light imaging, near infrared imaging, near ultraviolet imaging, thermal infrared imaging or ultrasonic imaging.

8. two identifier's face false proof device, this device comprises:

Sensing unit is used for using one or more of near infrared, ultrasound wave, RF-wise or visible image capturing head, through the mode of real-time monitoring, and the existence of induction people face;

Multi-modal generation source comprises active light source under a plurality of spectrum, is used for one or more of required 3D structured light of 3D imaging or ultrasonic generator;

Multi-modal data acquisition equipment is used to gather the multispectral imaging of people's face, the thermal infrared photoimaging that human body itself is sent, the 3D rendering of people's face or in the ultrasonic imaging one or more;

Multi-modal people's face detecting unit is used for detecting people's face position of multi-modality images, and detected facial image is sent to the anti-dummy unit of multi-modal pair of identifier's face;

The multi-modal pair of anti-dummy unit of identifier's face is used to verify whether target people face is authentic and valid people's face;

Display unit is used to show the false proof result of people's face,

Wherein, the multi-modal pair of anti-dummy unit of identifier's face further comprises: multi-modal people's face live body detecting unit is used for that target people face is carried out live body and detects; Multi-modal people's face authentication unit is used for target people face is carried out authentication.

9. according to claim 8 pair of identifier's face false proof device; It is characterized in that; When said multi-modal people's face live body detecting unit carries out the live body detection to target people face, at first, judge the biologically active of target people face roughly; Wherein judge: judge through thermal infrared to judge whether the temperature of target people face near 37 degree according in the following mode one or more; Through the depth information of 3D rendering judgement people face, judge whether face is the 3D object; Through the ultrasonic reflections rate of ultrasonic reflections evaluating objects people face, judge whether the ultrasonic reflections rate of skin is similar with real human face; Through the reflectivity of multispectral imaging evaluating objects people face under different spectrum; Whether the multispectral reflectivity of judging skin is similar with real human face; If judge that through above-mentioned one or more modes the information index of target people face is similar with real human face; Then continue accurately to judge the biologically active of target people face,, utilize mutual quotient images algorithm to carry out live body judgement accurately the multispectral facial image that collects.

10. according to claim 9 pair of identifier's face false proof device is characterized in that, the quotient images algorithm comprises the steps: mutually

Gather a large amount of true man people's faces and the false people's face multispectral imaging composing training data set under different distance, carry out being divided by of Pixel-level, form mutual quotient images group, suppose to select arbitrarily two spectrum lambda for the image under any two different spectrum of same individual₁, λ₂, the image of same individual face under two spectrum does

With

Its mutual quotient images defines as follows:

${\mathrm{MQI}}_{\mathrm{\λ}1,\mathrm{\λ}2}(x,y)=\frac{I_{\mathrm{\λ}1}}{I_{\mathrm{\λ}2}}=\frac{{\mathrm{\ρ}}_{\mathrm{\λ}1}(x,y){\mathrm{\κ}}_{\mathrm{\λ}1}\left(z\right)}{{\mathrm{\ρ}}_{\mathrm{\λ}2}(x,y){\mathrm{\κ}}_{\mathrm{\λ}2}\left(z\right)}$

Wherein, ρ representes the reflectivity of people's face, and κ represents the intensity of light source in people's face surface, and z representative face is apart from the distance of light source, and (x y) represents coordinate on the facial image;

For all mutual quotient images, on a plurality of yardsticks, be divided into a plurality of overlapping or nonoverlapping fritters, extract the proper vector of each fritter, the proper vector of all fritters is made up, as the proper vector of the overall situation;

Based on statistical learning method, training classifier on training dataset is used to distinguish true, false people's face.

11. according to claim 9 pair of identifier's face false proof device is characterized in that, when multi-modal people's face authentication unit carries out authentication to target people face, at first gathers the multi-modality images of a large amount of real human face, and every image is extracted its textural characteristics; Secondly; The proper vector of image is subtracted each other in twos; Whether belong to same individual according to two images; In proper vector type of being divided into after will subtracting each other, type between two types, utilize two types of sorters of machine learning algorithm training, the sorter that training obtains can judge whether two proper vectors of input belong to same individual; If in face recognition application, if the target facial image facial image corresponding, regarded as by above-mentioned two types of sorters and to belong to same people with recognition result, think that then target people face is authentic and valid, otherwise be false people's face; If in people's face checking is used, if target facial image and the corresponding facial image of being claimed of nominator's identity, regarded as by above-mentioned two types of sorters and to belong to same people, think that then target people face is authentic and valid, otherwise be false people's face.

12. according to each described pair of identifier's face false proof device of claim 9-11; It is characterized in that: every kind of different imaging type is called as a mode, and imaging type comprises visual light imaging, near infrared imaging; The near ultraviolet imaging, thermal infrared imaging or ultrasonic imaging.

Images