CN104834849B - Dual-factor identity authentication method and system based on Application on Voiceprint Recognition and recognition of face - Google Patents

Abstract

The present invention relates to the Dual-factor identity authentication method and system based on Application on Voiceprint Recognition and recognition of face, authentication method includes：Detect user's facial image；Mark key point；Face authentication unit verifies human face similarity degree；Generate password；Acquire voice data and User ID；To language data process；Content of text verification unit verifies password；Voiceprint unit verifies vocal print similarity.Verification System includes the Face datection unit set on client, facial key point calibration unit, face authentication unit, the first password generation unit, speech terminals detection unit, and voice recognition unit, content of text verification unit, vocal print feature vector extraction unit, voiceprint unit set on server end.The superior effect of the present invention is：Speech recognition is added on the basis of vocal print and recognition of face, improves the safety and reliability of certification；Password is in client and server end complete independently, safe and not with text or ciphertext in server end and client transmissions.

Description

Dual-factor identity authentication method and system based on Application on Voiceprint Recognition and recognition of face

Technical field

The invention belongs to mode identification technologies, are related to long-distance identity-certifying technology, and in particular to one kind being based on vocal printThe Dual-factor identity authentication method and system of identification and recognition of face.

Background technology

With mobile Internet high speed development and hand-held terminal device such as smart mobile phone, tablet computer it is universal, mutuallySecurity issues become increasingly urgent for networking.Currently, the either hardware digital certificate or E-token dynamic password card of bank, all only accomplish pairThe management of trusted terminal can not verify user identity.

Biometrics identification technology is the physiological characteristic or behavioural characteristic using people, to carry out the identification of personal identification.The biological characteristic for being used for bio-identification has sound, fingerprint, face, iris, retina etc., and microphone and camera are generally depositedIt is existing mobile terminal, therefore by sound and face come to carry out authentication be most convenient, most economical solution.

Recognition of face is a kind of biometrics identification technology that the facial feature information based on people carries out identification, mainlyIncluding two big module of face registration and face authentication.Recognition of face acquires image or video flowing containing face using camera,And automatic detect and track face in the picture, and then the face to detecting carries out a series of the relevant technologies processing of face.

The sound of people covers the information of multiple dimensions, such as speech content, the tone of speaking, sound speciality.Application on Voiceprint RecognitionIt is a kind of technology distinguishing different speakers by the sound speciality of people, different channel structures determines the unique of vocal printProperty.Application on Voiceprint Recognition includes mainly two big modules：Voiceprint registration module and vocal print authentication module.Voiceprint registration refers to using choosing in advanceFixed model models the speech samples of user, generates the sound-groove model of the user；When user asks authentication, profitRequest voice is authenticated with corresponding sound-groove model.User only Jing Guo voiceprint registration could use voiceprint work(Energy.Application on Voiceprint Recognition combination speech content, can effectively avoid Replay Attack.

Invention content

It is an object of the invention to overcome deficiency in the prior art, provide a kind of based on Application on Voiceprint Recognition and recognition of faceDual-factor identity authentication method and system.

The present invention is achieved by the following technical solutions：

Dual-factor identity authentication method based on Application on Voiceprint Recognition and recognition of face includes at least following steps：

S01：The human face region image of the Face datection unit detection request certification user；

S02:Unit, which is demarcated, by facial key point is detecting the facial key point of calibration in human face region；

S03:The face identification unit calculates the phase of the faceform for the registration that the face of the user is stored with clientLike degree, the face judging unit is used to judge whether human face similarity degree to be more than the threshold value of setting, if human face similarity degree is more than thresholdValue is then by entering S04, the authentification failure if human face similarity degree is less than threshold value；

S04:The first password generation unit is generated by random algorithm using current time and User ID as seed dynamicState password text, while triggering the second password generated unit and generating identical dynamic password text；

The first password generation unit and the second password generated unit utilize current time and the user for being accurate to minuteID generates dynamic password.

S05:The voice data that user reads dynamic password is acquired by voice collecting unit；And it is examined by sound endThe starting endpoint and end caps of unit detection user speech are surveyed, and the voice data of detection and User ID are sent to serverEnd；

S06:The received server-side voice data and User ID, the voice recognition unit is to the voice number that receivesAccording to progress voice recognition processing, and convert voice data into password text；

S07:The dynamic that the content of text verification unit generates the password text of conversion and the second password generated unitPassword text is compared, by entering S08 if password text is identical, if password text difference authentification failure；

S08:The vocal print feature vector extraction unit extracts vocal print feature vector, the sound from the voice data of userLine recognition unit is calculated by inner product between the sound-groove model of the registration of user's vocal print feature vector and server end storageSimilarity, the vocal print judging unit be used for judge vocal print similarity whether be more than setting threshold value, if vocal print similarity is bigIn threshold value then authentication success, the authentification failure if vocal print similarity is less than threshold value.

The technical solution is preferably, in the S08, the vocal print feature vector extraction unit extract vocal print feature toWhen amount, it converts user speech to short-term spectrum characteristic sequence, calculates each frame frequency spectrum signature in each Gauss of global context modelPosterior probability on component obtains the gauss hybrid models of user using maximum posteriori criterion adaptive training, by GaussThe mean value of Gaussian component is spliced to form high dimension vector in mixed model, and the high dimension vector is vocal print feature vector.

The technical solution is preferably that the short-term spectrum feature is linear pre- using mel-frequency cepstrum coefficient or perceptionSurvey coefficient.

The technical solution is preferably, in the S01, the human face region of the Face datection unit detection registration userImage, and multiple human face region images are therefrom intercepted as face sample, and be stored in face modeling unit.

The technical solution is preferably the acquisition requirement of the face sample：Two adjacent human face region imagesTime interval is at least 500 milliseconds and difference of the adjacent two human face region images on gray value is more than preset thresholdValue.

The technical solution is preferably, in the S06, the voice data and use of the received server-side registration userFamily ID converts voice data to the vocal print feature vector of regular length by vocal print feature vector extraction unit, and with userID is that index is stored in vocal print modeling unit.

The technical solution is preferably the registration user one or many voice data of typing in registration.

The present invention provides a kind of Dual-factor identity authentication system based on Application on Voiceprint Recognition and recognition of face, using it is described it is double becausePlain identity identifying method, the Dual-factor identity authentication system include being set to client and sequentially connected Face datection listMember, facial key point calibration unit, face authentication unit, the first password generation unit, speech terminals detection unit, the faceKey point, which is demarcated, is equipped with face modeling unit between unit and the first password generation unit；Further include be set to server end and according toVoice recognition unit, content of text verification unit, vocal print feature vector extraction unit, the voiceprint unit of secondary connection, it is describedContent of text verification unit is connect with the second password generated unit, and the vocal print being connect with vocal print feature vector extraction unit is builtForm unit.

The technical solution is preferably that the Face datection unit is equipped with video acquisition device；The video acquisition dressIt sets for acquiring user's human face region image.

The Face datection unit, when sending out ID authentication request, face is acquired by video acquisition device for userArea image；The face key point calibration unit is used to determine the face position and profile in the human face region；DescribedOne password generation unit is for generating dynamic password text, by random algorithm using current time and User ID as seed, togetherWhen triggering be set to the second password generated unit of server end and generate identical dynamic password text；The speech terminals detectionUnit is used to detect the starting endpoint and end caps of user speech, and the voice data of detection is sent to server end.

The voice recognition unit is used to the voice data that client is sent being converted into content of text；The content of textThe content of text that verification unit is used to send the dynamic password text that the second password generated unit generates with voice recognition unitBe compared, if comparison result unanimously if pass through, if inconsistent authentification failure.The vocal print feature vector extraction unit is used forExtraction can represent the vocal print feature vector of user's vocal print from voice data.The face modeling unit is used for from registration userChoose multiple facial image samples in the video data of typing, and then establish faceform, the vocal print modeling unit be used for fromExtraction vocal print feature vector in the voice data of user's typing is registered, and then establishes sound-groove model.

The technical solution is preferably that the face authentication unit includes sequentially connected face identification unit and faceJudging unit.The face identification unit be used for calculate user face and client storage registered face model it is similarDegree, the face judging unit are used to judge whether human face similarity degree to be more than the threshold value of setting.

The technical solution is preferably that the end-speech detection unit is equipped with voice collecting unit；The voice collectingUnit is used to acquire the voice data that user reads dynamic password.

The technical solution is preferably that the voiceprint unit includes sequentially connected Application on Voiceprint Recognition unit and vocal printJudging unit.The Application on Voiceprint Recognition unit is used to calculate the registration vocal print mould of the vocal print feature vector sum server end storage of userSimilarity between type, the vocal print judging unit are used to judge whether vocal print similarity to be more than the threshold value of setting.

The technical solution is preferably that the speech terminals detection unit is equipped with real time data transmission unit；Institute's predicateSound recognition unit is equipped with real time data receiving unit.The real time data transmission unit is for acquiring speech terminals detection unitVoice data be sent to server end, the real time data receiving unit is for receiving the voice data that client is sent.

The technical solution is preferably that number is equipped between the speech terminals detection unit and real time data transmission unitAccording to compression unit, data decompression unit is equipped between the voice recognition unit and real time data receiving unit.The data pressureFor compressing voice data, the data decompression unit is used for voice data decompression contracting unit.

The technical solution is preferably that the Verification System includes iterative testing unit, for checking request userWhether registered in client and server end, and whether registers corresponding faceform and sound-groove model.

The technical solution is preferably that the Verification System includes faceform's training unit and vocal print model training listMember；Faceform's training unit is used to train the face of the registration user according to facial image input by user is registeredModel, and it is stored in client, the sound-groove model training unit is used to, according to voice data input by user is registered, train instituteThe sound-groove model of registration user is stated, and is stored in server end.

Compared with prior art, superior effect of the invention is：On the basis of Application on Voiceprint Recognition and recognition of face, it is addedSpeech recognition verifies dynamic password, improves the safety and reliability of long-distance identity-certifying；The dynamic password is sameWhen in client and server end complete independently, and dynamic password need not be with text or ciphertext in server end and clientIt is transmitted between end, has ensured that the dynamic password is not stolen and attacked by the third party in network transmission；Playback is taken precautions againstSwarm into attack, enhance the safety of certification.

Description of the drawings

Fig. 1 is the structural schematic diagram of the Dual-factor identity authentication system based on Application on Voiceprint Recognition and recognition of face；

Fig. 2 is the Dual-factor identity authentication method implementation flow chart based on Application on Voiceprint Recognition and recognition of face；

Fig. 3 is registration process implementation flow chart in identity identifying method described in Fig. 2.

Attached drawing mark is as follows：

11- Face datections unit, 12- face key points calibration unit, 13- face authentications unit, 131- recognition of face listsMember, 132- faces judging unit, the first passwords of 14- generation unit, 15- speech terminals detections unit, 16- faces modeling unit,21- voice recognition units, 22- content of text verification unit, 23- vocal print feature vectors extraction unit, 24- voiceprints unit,241- Application on Voiceprint Recognition unit, 242- vocal prints judging unit, the second password generateds of 25- unit, 26- vocal print modeling units.

Specific implementation mode

The specific embodiment of the invention is described in further detail below in conjunction with the accompanying drawings.

As shown in Fig. 1, the Dual-factor identity authentication system of the present invention based on Application on Voiceprint Recognition and recognition of face, includingIt is set to client and sequentially connected Face datection unit 11, facial key point calibration unit 12, face authentication unit 13, theOne password generation unit 14, speech terminals detection unit 15, the face key point calibration unit 12 and the first password generated listFace modeling unit 16 is equipped between member 14；Further include being set to server end and sequentially connected voice recognition unit 21, textThis content authentication unit 22, vocal print feature vector extraction unit 23, voiceprint unit 24, the content of text verification unit 2225 are connect with the second password generated unit, and the vocal print modeling unit 26 being connect with vocal print feature vector extraction unit 23.

The Face datection unit 11 is equipped with video acquisition device (not shown)；The video acquisition device is for acquiringUser's human face region image.The Face datection unit 11, when sending out ID authentication request, passes through video acquisition for userDevice acquires human face region image；The face key point calibration unit 12 is used to determine the face position in the human face regionIt sets, eye position, eyebrow position, nose shape and profile；The first password generation unit 14 is for generating dynamic password textThis, by random algorithm using current time and User ID as seed, while triggering is set to the second password life of server endIdentical dynamic password text is generated at unit 25；The speech terminals detection unit 15 is used to detect the initiating terminal of user speechPoint and end caps, server end is sent to by the voice data of detection.The face modeling unit 16 is used for from registration userMultiple facial image samples are chosen in the video data of typing, and then establish faceform.

The end-speech detection unit 15 is equipped with voice collecting unit (not shown), such as microphone；The voice collectingUnit is used to acquire the voice data that user reads dynamic password.When the speech terminals detection unit 15 detects, every 20 millisSecond calculates the energy of current speech segment, if present energy is higher than preset threshold value, by current clip labeled as effectiveOtherwise current clip is labeled as invalid voice segment by sound bite.Record the number of efficient voice segment and invalid voice segmentMesh shows that user reads dynamic password knot if 10 invalid voice segments and efficient voice segments continuously occur is more than 50Beam, the voice collecting unit stop acquisition voice data.Then it is by the position mark that first efficient voice segment occursVoice starting endpoint, the position mark that the last one efficient voice segment is occurred are voice end caps.The vocal print modelingUnit 26 is used for the extraction vocal print feature vector from the voice data of registration user's typing, and then establishes sound-groove model.

The face authentication unit 13 includes sequentially connected face identification unit 131 and face judging unit 132.It is describedFace identification unit 131 is used to calculate the similarity of the faceform of the face of user and the registration of client storage, the peopleFace judging unit 132 is used to judge whether human face similarity degree to be more than the threshold value of setting.

The voiceprint unit 24 includes sequentially connected Application on Voiceprint Recognition unit 241 and vocal print judging unit 242.It is describedApplication on Voiceprint Recognition unit 241 is used to calculate between the sound-groove model of the registration of the vocal print feature vector sum server end storage of userSimilarity, the vocal print judging unit 242 are used to judge whether vocal print similarity to be more than the threshold value of setting.

The speech terminals detection unit 15 is equipped with real time data transmission unit (not shown)；The voice recognition unit21 are equipped with real time data receiving unit (not shown).The real time data transmission unit is used for speech terminals detection unit 15The voice data of acquisition is sent to server end, and the real time data receiving unit is for receiving the voice number that client is sentAccording to.Further, data compression unit is equipped between the speech terminals detection unit 15 and real time data transmission unit (in figureDo not show), data decompression unit (not shown) is equipped between the voice recognition unit 21 and real time data receiving unit.It is describedFor compressing voice data, the data decompression unit is used for voice data decompression data compression unit.By to clientEnd further reduced the bandwidth requirement of network transmission to input compress speech and real-time network transmission technology；It is servicing simultaneouslyDevice end real-time reception and data decompression greatly improve user's registration and the response speed of verification process.

A kind of Dual-factor identity authentication method based on Application on Voiceprint Recognition and recognition of face of the present invention, as shown in Fig. 2, using upperDual-factor identity authentication system is stated, is included the following steps：

S01：The human face region image of the detection of the Face datection unit 11 request certification user；The Face datection unitThe human face region image of 11 detection registration users, and multiple human face region images are therefrom intercepted as face sample, and be stored inIn face modeling unit 16；The acquisition requirement of the face sample：The time interval of two adjacent human face region images is at leastFor 500 milliseconds and difference of the adjacent two human face region images on gray value is more than preset threshold value.

S02:Unit 12, which is demarcated, by facial key point is detecting the facial key point of calibration in human face region.

S03:The face identification unit 131 calculates the face of the user with the faceform's of the registration of client storageSimilarity, the face judging unit 132 are used to judge whether human face similarity degree to be more than the threshold value of setting, if human face similarity degree is bigIn threshold value then by entering S04, the authentification failure if human face similarity degree is less than threshold value.

S04:The first password generation unit 14 is generated using current time and User ID as seed by random algorithmDynamic password text, while triggering the second password generated unit 25 and generating identical dynamic password text.The first password lifeAt unit 14 and the second password generated unit 25 dynamic password text is generated using the current time and User ID for being accurate to minute.

S05:The voice data that user reads dynamic password is acquired by voice collecting unit；And it is examined by sound endIt surveys unit 15 and detects the starting endpoint and end caps of user speech, and the voice data of detection and User ID are sent to serviceDevice end.

S06:The received server-side voice data and User ID, the voice recognition unit 21 is to the voice that receivesData carry out voice recognition processing, and convert voice data into password text.The language of the received server-side registration userSound data and User ID, by vocal print feature vector extraction unit 23 convert voice data to the vocal print feature of regular length toAmount, and be that index is stored in vocal print modeling unit 26 with User ID；Registration user typing in registration is one or manyVoice data.

S07:The content of text verification unit 22 generates the password text of conversion and the second password generated unit 25Dynamic password text is compared, by entering S08 if password text is identical, if password text difference authentification failure.

S08:The vocal print feature vector extraction unit 23 extracts vocal print feature vector from the voice data of user, describedThe sound-groove model of user's vocal print feature vector and the registration of server end storage is calculated by inner product for Application on Voiceprint Recognition unit 241Between similarity, the vocal print judging unit 242 be used for judge vocal print similarity whether be more than setting threshold value, if vocal print phaseIt is more than threshold value then authentication success, the authentification failure if vocal print similarity is less than threshold value like degree.

When the vocal print feature vector extraction unit 23 extracts vocal print feature vector, it converts user speech to short-term spectrumCharacteristic sequence calculates posterior probability of each frame frequency spectrum signature in each Gaussian component of global context model, utilizes maximum a posterioriCanon of probability adaptive training obtains the gauss hybrid models of user, and the mean value of Gaussian component in gauss hybrid models is spliced shapeAt high dimension vector, the high dimension vector is vocal print feature vector.The short-term spectrum feature is using mel-frequency cepstrum coefficient, senseKnow linear predictor coefficient.

For the present invention by principal component analysis to the vocal print feature vector dimensionality reduction, compression sound-groove model size improves vocal printSimilarity calculation module speed；It specifically includes and is based on background sound training dataset, utilize vocal print feature vector extraction unit 23Extract data set in each voice vocal print feature vector, using principal component analysis calculate corresponding characteristic value and feature toAmount, characteristic value is sorted from big to small, is chosen the feature vector corresponding to top n characteristic value and is constituted the feature son sky after dimensionality reductionBetween, dimensionality reduction is carried out to registration vocal print feature vector sum certification vocal print feature vector using the subspace, obtains N-dimensional vocal print featureVector.

Present invention is not limited to the embodiments described above, without departing from the essence of the present invention, this field skillAny deformation, improvement, the replacement that art personnel are contemplated that each fall within the scope of the present invention.

Claims (5)

1. the authentication method of the Dual-factor identity authentication system based on Application on Voiceprint Recognition and recognition of face, the Dual-factor identity authenticationSystem include be set to client and sequentially connected Face datection unit, facial key point calibration unit, face authentication unit,First password generation unit, speech terminals detection unit, it is described face key point calibration unit and the first password generation unit itBetween be equipped with face modeling unit；Further include being set to server end and sequentially connected voice recognition unit, content of text verificationUnit, vocal print feature vector extraction unit, voiceprint unit, the content of text verification unit and the second password generated unitIt connects, and the vocal print modeling unit being connect with vocal print feature vector extraction unit；The Face datection unit is adopted equipped with videoAcquisition means；The speech terminals detection unit is equipped with voice collecting unit；The voiceprint unit includes sequentially connected soundLine recognition unit and vocal print judging unit；The Verification System includes faceform's training unit and vocal print model training unit；

It is characterized in that, including at least following steps：

S01：The human face region image of Face datection unit detection request certification user；

S02:Unit, which is demarcated, by facial key point is detecting the facial key point of calibration in human face region；

S03:Face identification unit calculates the similarity of the face and the faceform of the registration of client storage of the user, faceJudging unit is used to judge whether human face similarity degree to be more than the threshold value of setting, passes through entrance if human face similarity degree is more than threshold valueS04, the authentification failure if human face similarity degree is less than threshold value；

S04:First password generation unit generates dynamic password text using current time and User ID by random algorithm as seedThis, while triggering the second password generated unit and generating identical dynamic password text；

S05:The voice data that user reads dynamic password is acquired by voice collecting unit；And pass through speech terminals detection listThe starting endpoint and end caps of member detection user speech, and the voice data of detection and User ID are sent to server end；

S06:The received server-side voice data and User ID, voice recognition unit carry out language to the voice data receivedSound identifying processing, and convert voice data into password text；

S07:The dynamic password text that content of text verification unit generates the password text of conversion and the second password generated unitIt is compared, by entering S08 if password text is identical, if password text difference authentification failure；

S08:Vocal print feature vector extraction unit extracts vocal print feature vector from the voice data of user, and Application on Voiceprint Recognition unit is logicalCross the similarity between the sound-groove model for the registration that user's vocal print feature vector and server end storage is calculated in inner product, vocal printJudging unit be used for judge vocal print similarity whether be more than setting threshold value, if vocal print similarity be more than threshold value if authentication atWork(, the authentification failure if vocal print similarity is less than threshold value.

2. the authenticating party of the Dual-factor identity authentication system according to claim 1 based on Application on Voiceprint Recognition and recognition of faceMethod, which is characterized in that in the S01, the human face region image of the Face datection unit detection registration user, and therefrom interceptMultiple human face region images are stored in as face sample in face modeling unit.

3. the authenticating party of the Dual-factor identity authentication system according to claim 1 based on Application on Voiceprint Recognition and recognition of faceMethod, which is characterized in that in the S06, the voice data and User ID of the received server-side registration user pass through vocal print spyThe vectorial extraction unit of sign converts voice data to the vocal print feature vector of regular length, and is that index is stored in sound with User IDIn line modeling unit.

4. the authenticating party of the Dual-factor identity authentication system according to claim 3 based on Application on Voiceprint Recognition and recognition of faceMethod, which is characterized in that the registration user one or many voice data of typing in registration.

5. the authenticating party of the Dual-factor identity authentication system according to claim 1 based on Application on Voiceprint Recognition and recognition of faceMethod, which is characterized in that in the S08, when the vocal print feature vector extraction unit extracts vocal print feature vector, by user speechIt is converted into short-term spectrum characteristic sequence, it is general to calculate posteriority of each frame frequency spectrum signature in each Gaussian component of global context modelRate obtains the gauss hybrid models of user using maximum posteriori criterion adaptive training, by Gauss in gauss hybrid modelsThe mean value of component is spliced to form high dimension vector, and the high dimension vector is vocal print feature vector.