CN106997452A - Live body verification method and device - Google Patents
Live body verification method and device Download PDFInfo
- Publication number
- CN106997452A CN106997452A CN201610051911.6A CN201610051911A CN106997452A CN 106997452 A CN106997452 A CN 106997452A CN 201610051911 A CN201610051911 A CN 201610051911A CN 106997452 A CN106997452 A CN 106997452A
- Authority
- CN
- China
- Prior art keywords
- face
- eyes
- characteristic value
- video
- image
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Granted
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V40/00—Recognition of biometric, human-related or animal-related patterns in image or video data
- G06V40/40—Spoof detection, e.g. liveness detection
- G06V40/45—Detection of the body part being alive
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
- G06N3/084—Backpropagation, e.g. using gradient descent
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V40/00—Recognition of biometric, human-related or animal-related patterns in image or video data
- G06V40/10—Human or animal bodies, e.g. vehicle occupants or pedestrians; Body parts, e.g. hands
- G06V40/16—Human faces, e.g. facial parts, sketches or expressions
- G06V40/161—Detection; Localisation; Normalisation
- G06V40/166—Detection; Localisation; Normalisation using acquisition arrangements
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- Human Computer Interaction (AREA)
- General Engineering & Computer Science (AREA)
- Multimedia (AREA)
- Artificial Intelligence (AREA)
- Evolutionary Computation (AREA)
- General Health & Medical Sciences (AREA)
- Life Sciences & Earth Sciences (AREA)
- Health & Medical Sciences (AREA)
- Computing Systems (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Computational Linguistics (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Bioinformatics & Computational Biology (AREA)
- Molecular Biology (AREA)
- Evolutionary Biology (AREA)
- Biophysics (AREA)
- Biomedical Technology (AREA)
- Oral & Maxillofacial Surgery (AREA)
- Image Analysis (AREA)
- Collating Specific Patterns (AREA)
Abstract
The invention provides a kind of live body verification method and device, wherein, this method includes:Using the face video that acquisition is to be verified;Extract the multiframe face eyes image in the face video to be verified;Multiframe face eyes image is carried out to open and close eyes judgement, first is obtained and opens and closes eyes characteristic value;Multiframe face eyes image is carried out to block eye judgement, first is obtained and blocks a characteristic value;Block a characteristic value according at least to first characteristic value and first that opens and closes eyes and verify whether the corresponding face of face video to be verified is true man's living body faces.Solved by the present invention during carrying out live body checking by inspection blink motion in correlation technique, it can not exclude in eye formation local motion to simulate the camouflage blink motion of blink, cause the incorrect problem of live body the result, so as to eliminate the influence for forging blink situation to live body the result.
Description
Technical field
The present invention relates to field of biological recognition, and in particular to a kind of live body verification method and device.
Background technology
Face recognition technology has tended to be ripe.But in most applications, except to know to face
It is not outer, in addition it is also necessary to which that live body checking is carried out to face, to prevent disabled user from relying on papery photo or electronics
The deception such as screen display photo reaches a standard.
Existing live body verification method, there are many based on the motion of detection blink to carry out live body checking
's.But, these methods often simply devise how to detect the motion of blink (for example:CN
101216887A, CN103400122A), do not account for but how resisting by causing local fortune in eye
Move to simulate the deception mode of blink.In addition, the live body verification method based on light stream is (for example:CN
101908140A), by detect the light stream in living body faces image change, judge face to be measured whether be
Living body faces.But the calculating consumption of light stream is big, and this method can not be resisted by being made in eye
The deception mode of blink is simulated into local motion.
For in correlation technique, during checking that blink motion carries out live body checking, it is impossible to arrange
Blink and move except the camouflage in eye formation local motion to simulate blink, so as to cause live body checking knot
Really incorrect phenomenon, does not propose effective solution also.
The content of the invention
Therefore, the technical problem to be solved in the present invention is to overcome being blinked by checking in correlation technique
During motion carries out live body checking, it is impossible to exclude in eye formation local motion to simulate blink
Camouflage blink motion, so as to cause the incorrect phenomenon of live body the result, so as to provide a kind of live body
Verification method and device.
According to an aspect of the invention, there is provided a kind of live body verification method, including:Obtain to be tested
The face video of card;Extract the multiframe face eyes image in the face video to be verified;To institute
State multiframe face eyes image open and close eyes judgement, obtain first and open and close eyes characteristic value;To described many
Frame face eyes image carries out blocking eye judgement, obtains first and blocks a characteristic value;According at least to described
First characteristic value and described first that opens and closes eyes blocks characteristic value and verifies the face video pair to be verified
Whether the face answered is true man's living body faces.
Alternatively, it is described to block a feature according at least to described first characteristic value and described first that opens and closes eyes
Value verifies whether the corresponding face of the face video to be verified is that true man's living body faces include:Institute
State first characteristic value that opens and closes eyes and indicate that the face video to be verified has a blink action, and described the
In the case that one blocks a characteristic value instruction unobstructed phenomenon of multiframe face eyes image, it is determined that institute
The corresponding face of face video to be verified is stated for true man's living body faces.
Alternatively, the judgement that opens and closes eyes is carried out to the multiframe face eyes image, obtains first and open and close eyes
Characteristic value includes:The multiframe face eyes image is inputted to the deep neural network for the classification that opens and closes eyes,
Described first is obtained to open and close eyes characteristic value;Wherein, the deep neural network of the classification that opens and closes eyes is used for
Judge whether the multiframe face eyes image has the phenomenon opened eyes or closed one's eyes.
Alternatively, the multiframe face eyes image is carried out blocking eye judgement, obtains first and block eye
Characteristic value includes:The multiframe face eyes image is inputted to the deep neural network for blocking eye classification,
Obtain described first and block a characteristic value;Wherein, the deep neural network for blocking eye classification is used for
It is true man's eye image or the camouflage eye image that is blocked to judge the multiframe face eyes image.
Alternatively, block a characteristic value according at least to described first characteristic value and described first that opens and closes eyes and test
Whether demonstrate,prove the corresponding face of the face video to be verified is that true man's living body faces include:By described
One opens and closes eyes characteristic value and the described first blink grader for blocking characteristic value input video level, checking
Whether the corresponding face of the face video to be verified is true man's living body faces;Wherein, the video
The blink grader of level is used to verify whether the corresponding face of the face video to be verified is that true man live
Body face.
Alternatively, by the multiframe face eyes image input open and close eyes classification deep neural network it
Before, the deep neural network for the classification that opened and closed eyes described in training, wherein, open and close eyes classification described in training
Deep neural network includes:The eyes image opened eyes using multiple and multiple eyes images closed one's eyes are trained
The deep neural network of the classification that opens and closes eyes.
Alternatively, by the multiframe face eyes image input block eye classification deep neural network it
Before, the deep neural network of eye classification is blocked described in training, wherein, eye classification is blocked described in training
Deep neural network includes:Use the eyes image and multiple camouflage living body faces of multiple true man's faces
The deep neural network of eye classification is blocked described in eyes image training;Wherein, the camouflage living body faces
Eyes image be simulated by blocking the eye of living body faces blink action camouflage living body faces figure
Picture.
Alternatively, described first characteristic value and described first that opens and closes eyes is blocked into a characteristic value input video
Before the blink grader of level, the blink grader of the videl stage is trained, wherein, regarded described in training
The blink grader of frequency level includes:Obtaining includes the positive sample of video that true man normally blink and not including very
The video that people normally blinks bears sample;Regarded by the deep neural network of the classification that opens and closes eyes from described
The positive sample of frequency and the video are born the second of the multiple image extracted in sample in video and opened and closed eyes feature
Value, and deep neural network that eye classifies is blocked from the positive sample of the video and the video by described
Second that the multiple image in video is extracted in negative sample blocks a characteristic value;Open and close using described second
The characteristic value and described second of eye block the blink grader of the characteristic value training videl stage of eye.
Alternatively, the multiframe face eyes image extracted in the face video to be verified includes:
Using sliding window short-sighted frequency is obtained from the face video to be verified;Extract in the short-sighted frequency
The multiframe face eyes image.
Alternatively, methods described also includes:Calculate facial image in the face video to be verified
Locomotion speed value;It is described to block a feature according at least to described first characteristic value and described first that opens and closes eyes
Value verifies whether the corresponding face of the face video to be verified is true man's living body faces, including:Root
Opened and closed eyes characteristic value, the described first fortune for blocking a characteristic value and the facial image according to described first
Dynamic velocity amplitude verifies whether the corresponding face of the face video to be verified is true man's living body faces.
Alternatively, it is described according to described first open and close eyes characteristic value, described first block a characteristic value with
And whether the locomotion speed value of the facial image verifies the corresponding face of the face video to be verified
Include for true man's living body faces:Indicate that the face to be verified is regarded in described first characteristic value that opens and closes eyes
Frequently there are blink action, described first to block a characteristic value and indicate that multiframe face eyes image is unobstructed existing
As, and the facial image locomotion speed value be less than predetermined threshold in the case of, it is determined that it is described
The corresponding face of face video to be verified is true man's living body faces.
Alternatively, calculating the locomotion speed value of facial image in the face video to be verified includes:
Obtain the seat of the face key feature points of adjacent two frames facial image in the face video to be verified
Mark information;The face video to be verified is calculated according to the coordinate information of the face key feature points
The locomotion speed value of middle facial image.
According to another aspect of the present invention, a kind of live body checking device is additionally provided, including:Obtain
Module, the face video to be verified for obtaining;Extraction module, for extracting the people to be verified
Multiframe face eyes image in face video;The First Eigenvalue acquisition module, for the multiframe people
Face eyes image open and close eyes judgement, obtains first and opens and closes eyes characteristic value;Second Eigenvalue obtains mould
Block, for carrying out blocking eye judgement to the multiframe face eyes image, obtains first and blocks a feature
Value;Authentication module, for being opened and closed eyes characteristic value and described first to block eye special according at least to described first
Value indicative verifies whether the corresponding face of the face video to be verified is true man's living body faces.
Alternatively, the authentication module is treated specifically for being opened and closed eyes described first described in characteristic value instruction
The face video of checking has blink action, and described first blocks a characteristic value instruction multiframe face eye
In the case of the unobstructed phenomenon of portion's image, it is determined that the corresponding face of the face video to be verified
For true man's living body faces.
Alternatively, the First Eigenvalue acquisition module is specifically for by the multiframe face eyes image
Input opens and closes eyes the deep neural network of classification, obtains described first and opens and closes eyes characteristic value;Wherein, institute
Stating the deep neural network for the classification that opens and closes eyes is used to judge whether the multiframe face eyes image has eye opening
Or the phenomenon closed one's eyes.
Alternatively, the Second Eigenvalue acquisition module is specifically for by the multiframe face eyes image
The deep neural network of eye classification is blocked in input, is obtained first and is blocked a characteristic value;Wherein, it is described to hide
The deep neural network of gear eye classification is used to judge that the multiframe face eyes image is true man's eye image
Still the camouflage eye image being blocked.
Alternatively, the authentication module is specifically for described first is opened and closed eyes characteristic value and described first
A blink grader for characteristic value input video level is blocked, is opened and closed eyes characteristic value and institute according to described first
State first and block whether a corresponding face of the characteristic value checking face video to be verified is that true man live
Body face;Wherein, the blink grader of the videl stage is used to verify the face video to be verified
Whether corresponding face is true man's living body faces.
Alternatively, described device also includes:First training module, for the eye opened eyes using multiple
Image and the deep neural network for the classification that opened and closed eyes described in eyes images training of multiple eye closings.
Alternatively, described device also includes:Second training module, for using multiple true man's faces
The depth nerve of eye classification is blocked described in the eyes image training of eyes image and multiple camouflage living body faces
Network;Wherein, it is described camouflage living body faces eyes image be by block the eye of living body faces come
Simulate the camouflage living body faces image of blink action.
Alternatively, described device also includes:3rd training module, for described first to be opened and closed eyes spy
Value indicative and described first block the blink grader of characteristic value input video level before, regarded described in training
The blink grader of frequency level, wherein, the 3rd training module includes:First acquisition unit, is used for
Obtaining includes the positive sample of video that true man normally blink and the negative sample of video do not blinked normally including true man
Example;First extraction unit, for the deep neural network by the classification that opens and closes eyes from the video
Positive sample and the video bear the second feature opened and closed eyes of the multiple image extracted in sample in video
Value, and deep neural network that eye classifies is blocked from the positive sample of the video and the video by described
Second of the multiple image in the video characteristic value for blocking eye is extracted in negative sample;Training unit, is used for
The characteristic value for blocking eye using second characteristic value opened and closed eyes and described second trains the videl stage
Blink grader.
Alternatively, the extraction module includes:Second acquisition unit, for using sliding window from institute
State and short-sighted frequency is obtained in face video to be verified;Second extraction unit, for extracting the short-sighted frequency
In the multiframe face eyes image.
Alternatively, described device also includes:Computing module, is regarded for calculating the face to be verified
The locomotion speed value of facial image in frequency;The authentication module according to described first specifically for opening and closing eyes
Characteristic value, described first are blocked described in the locomotion speed value checking of a characteristic value and the facial image
Whether the corresponding face of face video to be verified is true man's living body faces.
Alternatively, the authentication module is treated specifically for being opened and closed eyes described first described in characteristic value instruction
The face video of checking has blink action, described first to block a characteristic value instruction multiframe face eye figure
In the case of being less than predetermined threshold as the locomotion speed value of unobstructed phenomenon, and the facial image,
It is true man's living body faces then to determine the corresponding face of the face video to be verified.
Alternatively, the computing module includes:3rd acquiring unit, it is described to be verified for obtaining
The coordinate information of the face key feature points of adjacent two frames facial image in face video;Computing unit,
For calculating people in the face video to be verified according to the coordinate information of the face key feature points
The locomotion speed value of face image.
According to a further aspect of the invention, a kind of live body checking system is additionally provided, including:Shooting
Device, the face video to be verified for catching;Processor, is connected with the camera device, is used for
Receive the face video to be verified and perform following steps:Extract the face video to be verified
In multiframe face eyes image;The judgement that opens and closes eyes is carried out to the multiframe face eyes image, obtained
First opens and closes eyes characteristic value;The multiframe face eyes image is carried out to block eye judgement, first is obtained
Block a characteristic value;A characteristic value is blocked according at least to described first characteristic value and described first that opens and closes eyes
Whether verify the corresponding face of the face video to be verified is true man's living body faces.
By the present invention, using the face video that acquisition is to be verified;Extract the face video to be verified
In multiframe face eyes image;The judgement that opens and closes eyes is carried out to multiframe face eyes image, first is obtained
Open and close eyes characteristic value;Multiframe face eyes image is carried out to block eye judgement, first is obtained and blocks eye spy
Value indicative;Block a characteristic value according at least to first characteristic value and first that opens and closes eyes and verify the people to be verified
Whether the corresponding face of face video is true man's living body faces.Solve in correlation technique, blinked by checking
During eye movement carries out live body checking, it is impossible to exclude in eye formation local motion to simulate blink
Camouflage blink motion, cause the incorrect problem of live body the result, thus eliminate forgery blink
Influence of the situation to live body the result.
Brief description of the drawings
In order to illustrate more clearly of the specific embodiment of the invention or technical scheme of the prior art, under
Face will be briefly described to the required accompanying drawing used in embodiment or description of the prior art,
It should be evident that drawings in the following description are some embodiments of the present invention, it is general for this area
For logical technical staff, on the premise of not paying creative work, it can also be obtained according to these accompanying drawings
Obtain other accompanying drawings.
Fig. 1 is the flow chart of the live body verification method of the embodiment of the present invention 1;
Fig. 2 for the embodiment of the present invention 1 training open and close eyes classification deep neural network flow chart;
Fig. 3 blocks the flow chart of the deep neural network of eye classification for the training of the embodiment of the present invention 1;
Fig. 4 is the flow chart of the blink grader of the training video level of the embodiment of the present invention 1;
Fig. 5 treats checking face using the blink grader for the videl stage trained by the present embodiment and regarded
Frequency carries out the flow chart of live body checking;
Fig. 6 verifies the structured flowchart of device for the live body of the embodiment of the present invention 2;
Fig. 7 verifies the structured flowchart of the 3rd training module in device for the live body of the embodiment of the present invention 2;
Fig. 8 verifies the structured flowchart of the second extraction module in device for the live body of the embodiment of the present invention 2;
Fig. 9 verifies the structured flowchart of the preferred embodiment of device for the live body of the embodiment of the present invention 2;
Figure 10 verifies the structured flowchart of the computing module in device for the live body of the embodiment of the present invention 2.
Embodiment
Technical scheme is clearly and completely described below in conjunction with accompanying drawing, it is clear that
Described embodiment is a part of embodiment of the invention, rather than whole embodiments.Based on this hair
Embodiment in bright, those of ordinary skill in the art are obtained under the premise of creative work is not made
Every other embodiment, belong to the scope of protection of the invention.
In the description of the invention, it is necessary to explanation, term " " center ", " on ", " under ", " left side ",
The orientation or position relationship of the instruction such as " right side ", " vertical ", " level ", " interior ", " outer " are based on attached
Orientation or position relationship shown in figure, are for only for ease of the description present invention and simplify description, rather than
Indicate or imply signified device or element and must have specific orientation, with specific azimuth configuration and
Operation, therefore be not considered as limiting the invention.In addition, term " first ", " second ", "
Three " are only used for describing purpose, and it is not intended that indicating or implying relative importance.
Embodiment 1
A kind of live body verification method is provided in the present embodiment, and Fig. 1 is the live body of the embodiment of the present invention 1
The flow chart of verification method, as shown in figure 1, the flow comprises the following steps:
Step S11:Obtain face video to be verified.
Step S12:Extract the multiframe face eyes image in face video to be verified.
Step S13:Multiframe face eyes image is carried out to open and close eyes judgement, first is obtained and opens and closes eyes feature
Value.
Step S14:Multiframe face eyes image is carried out to block eye judgement, first is obtained and blocks a feature
Value.
Step S15:Opened and closed eyes characteristic value and first to block the checking of characteristic value to be verified according at least to first
The corresponding face of face video whether be true man's living body faces.
By above-mentioned steps, detecting whether the corresponding face of face video to be verified is living body faces
During, extract first and open and close eyes characteristic value and first to block a characteristic value the two factors common
Live body is verified.Compared in correlation technique, only by detecting that blink moves this factor pair
Live body is verified that above-mentioned steps are solved in correlation technique, by checking that blink motion carries out live body
During checking, it is impossible to exclude in eye formation local motion to simulate the camouflage blink motion of blink,
Cause the incorrect problem of live body the result, tied so as to eliminate forgery blink situation and live body is verified
The influence of fruit.
If because the very fast frame per second of blink action is relatively low, as a result can be deteriorated.Therefore, in an optional reality
Apply in example, it is not recommended that extract a frame every several frames.Above-mentioned multiframe face eyes image is to be verified to extract
Each frame face eyes image in face video.Wherein, first characteristic value is opened and closed eyes for judging many
Whether frame face eyes image has the phenomenon opened eyes or closed one's eyes;First, which blocks a characteristic value, is used to judge
Multiframe face eyes image is true man's eye image or the camouflage eye image that is blocked.
Above-mentioned steps S13 is related to acquisition first and opened and closed eyes characteristic value, in one alternate embodiment,
First is obtained by the deep neural network for the classification that opens and closes eyes to open and close eyes characteristic value, specifically, by multiframe
The input of face eyes image opens and closes eyes the deep neural network of classification, obtains first and opens and closes eyes characteristic value;
Wherein, the deep neural network of classification of opening and closing eyes is used to judge whether the multiframe face eyes image has and open
Eye or the phenomenon closed one's eyes.
Above-mentioned steps S13 is related to the depth nerve that multiframe face eyes image is inputted to the classification that opens and closes eyes
Network, obtains first and opens and closes eyes characteristic value, and multiframe face eyes image is inputted to the depth for the classification that opens and closes eyes
Neutral net is spent, first is obtained and opens and closes eyes before characteristic value, it is necessary to train the depth god for the classification that opens and closes eyes
Through network.It should be noted that deep neural network that can in several ways to the classification that opens and closes eyes
It is trained, this is illustrated below.In one alternate embodiment, multiple eye openings are gathered
Facial image and multiple eye closings facial images, extract the eye figure in the facial image of all collections
Picture, the eyes image that eyes image is divided into the eyes image of eye opening and closed one's eyes, uses the eye of eye opening
Image and the eyes image closed one's eyes train the deep neural network for the classification that opens and closes eyes, so that opening and closing eyes
Whether the deep neural network of classification can have blink action progress to the face video to be verified of input
Judge.
As a kind of preferred embodiment, before step S13, in addition to the depth for the classification that opens and closes eyes
Degree neutral net is trained, specifically, as shown in Fig. 2 the flow comprises the following steps:
Step S21:Different people, the eye opening under different illumination conditions and the people of eye closing of magnanimity are gathered respectively
Face photo.
Step S22:Extract the eyes image in all photos.
Step S23:Suitable deep neural network model is designed (for example, may be referred to article
Network knot in Gradient-Based Learning Applied to Document Recognition
Structure), using the eyes images of all eye openings as classification A, using the eyes images of all eye closings as
Classification B, these classifications A and classification B eyes image are used as designed deep neural network model
Input, trains the deep neural network for the classification that opens and closes eyes so that the deep neural network can be well
Differentiate and open eyes or close one's eyes.
Above-mentioned steps S14, which is related to, to be obtained first and blocks a characteristic value, in one alternate embodiment,
By blocking the deep neural network that eye is classified, obtain this and first block a characteristic value, specifically, will
The deep neural network of eye classification is blocked in multiframe face eyes image input, is obtained this and first is blocked eye
Characteristic value;Wherein, the deep neural network for blocking eye classification is used to judge the multiframe face eye figure
It seem the camouflage eye image that true man's eye image is still blocked.
Above-mentioned steps S14, which is related to, inputs multiframe face eyes image in the depth nerve for blocking eye classification
Network, obtains first and blocks a characteristic value, and multiframe face eyes image is inputted to the depth for blocking eye classification
Spend neutral net, obtain first block a characteristic value before, it is necessary to train block eye classification depth god
, can be in several ways to blocking the deep neural network that eye is classified through network, it is necessary to illustrate
It is trained, this is illustrated below.In one alternate embodiment, multiple are gathered to include
Open and close eyes action true man's facial image and multiple include by block camouflage living body faces eye come mould
Intend the camouflage living body faces image of blink action, extract the eyes image in the facial image of all collections,
Eyes image is divided into the eyes image of true man's face eyes image and camouflage living body faces, true man are used
The depth nerve net of eye classification is blocked in the eyes image training of the eyes image and camouflage living body faces of face
Network, so that the deep neural network for blocking eye classification can be to the face video to be verified of input
Whether there is the action for blocking eyes image to be judged.
Due to the grader accuracy rate decline that opens and closes eyes when eyes are blocked, can be caused, so
Criminal may by block means fraud system (such as finger photo eyes according to blinking
Eye frequency is rocked), cause system to be mistakenly considered true man's blink.In order to prevent this fraud, carry
Go out addition to block grader to resist such case.As a kind of preferred embodiment, in step S14
Before, in addition to the deep neural network for blocking eye classification it is trained, specifically, such as Fig. 3 institutes
Show, the flow comprises the following steps:
Step S31:Use papery human face photo or electronics screen display human face photo camouflage living body faces.Make
Blink action is simulated with different modes, these modes include but is not limited to finger in human face photo
Eye moves back and forth simulation blink etc..
Step S32:Extract the eyes image in all photos.
Step S33:Suitable deep neural network model is designed (for example, may be referred to article
Network knot in Gradient-Based Learning Applied to Document Recognition
Structure), using the eyes images of all true man's faces as classification C, the eyes of all camouflage living body faces
Image is used as designed depth nerve net as classification D, these classifications C and classification D eyes image
The deep neural network model of eye classification is blocked in the input of network model, training so that the model can be very
True man's human eye and the camouflage human eye blocked by (part) are differentiated well.
Above-mentioned steps S15 is related to acquisition live body the result, in one alternate embodiment, by the
One opens and closes eyes characteristic value and the first blink grader for blocking characteristic value input video level, according at least to
First characteristic value and first that opens and closes eyes blocks characteristic value and verifies the corresponding face of face video to be verified
Whether it is true man's living body faces;Wherein, the blink grader of videl stage is used to verify the people to be verified
Whether the corresponding face of face video is true man's living body faces.
Above-mentioned steps S15 involves how to open and close eyes characteristic value and first to block eye special according at least to first
Value indicative verifies whether the corresponding face of face video to be verified is true man's living body faces, optional at one
In embodiment, indicate that face video to be verified has blink action in first characteristic value that opens and closes eyes, and
In the case that first blocks a characteristic value instruction unobstructed phenomenon of multiframe face eyes image, it is determined that
The corresponding face of face video to be verified is true man's living body faces, is verified.That is,
First opens and closes eyes in the case that characteristic value indicates that face video to be verified has blink action, can't be true
Fixed blink action is exactly the blink action of true man, it may be possible to cause local motion to simulate in eye
The deception action of blink, actually or in order to judge that blink action true man blink acts the blink of simulation
Action is, it is necessary to which first blocks a characteristic value progress auxiliary judgment, if first blocks a characteristic value instruction
Multiframe face eyes image has eclipse phenomena, then it is the blink action of simulation to illustrate blink action, if
First, which blocks a characteristic value, indicates multiframe face eyes image without eclipse phenomena, then explanation blink is dynamic
Work is the action of true man's blink, and live body is verified.So as to which the blink action eliminated simulation is mistaken for
True man's blink is acted, the result of influence live body checking.
Also related in above-mentioned steps S15 and the blink grader of videl stage is trained, it is necessary to illustrate
, the blink grader of videl stage can be trained in several ways, this is carried out below
Illustrate.In one alternate embodiment, obtain include the positive sample of video that true man normally blink with
Do not include the video normally blinked of true man and bear sample, by the deep neural network of the classification that opens and closes eyes from regarding
The positive sample of frequency and video bear the second characteristic value opened and closed eyes that multiple image is extracted in sample, and by hiding
The deep neural network of gear eye classification extracts each two field picture from the positive sample of video and the negative sample of the video
The second characteristic value for blocking eye;The characteristic value of eye is blocked using the second characteristic value opened and closed eyes and second
The blink grader of training video level.Can be to input so as to the blink grader of the videl stage after training
The corresponding face of face video to be verified whether be that true man's face is judged.
As a kind of preferred embodiment, during being trained to the blink grader of videl stage,
As shown in figure 4, its specific steps includes:
Step S41:The positive sample of short-sighted frequency that the true man of generation magnanimity normally blink, this section of short-sighted positive sample of frequency
Example includes the continuous facial image of n frames, there is the action that a true man normally blink in this section of short-sighted frequency.
Step S42:The short-sighted frequency for generating the non-blink of magnanimity bears sample, and this section of short-sighted frequency is born sample and included
The continuous facial image of n frames, but there is no in this section of short-sighted frequency an action that true man normally blink, for example,
Eyes are opened always in this section of short-sighted frequency, or eyes closed always in this section of short-sighted frequency, or in the section
By using blink for moving back and forth forgery before human eye of the finger in photo etc. in short-sighted frequency.
Step S43:Extract the feature that opens and closes eyes of the face eye of each frame in video, its specific steps bag
Include:The face eye of each frame in video is extracted, eyes image is input to what is trained in step S2
Open and close eyes the deep neural network model of classification, the feature scores value opened and closed eyes.
Step S44:The face eye for extracting each frame in video blocks a feature, its specific steps bag
Include:The face eye of each frame in video is extracted, eyes image is input to what is trained in step S2
The deep neural network model of eye classification is blocked, obtains blocking the feature scores value of eye.
In order that live body checking result it is more accurate, except first is opened and closed eyes characteristic value and first
Block a characteristic value as basis for estimation outside, in one alternate embodiment, also by face to be verified
The locomotion speed value of facial image also serves as basis for estimation in video, that is to say, that the work of the present embodiment
Body verification method is also including each frame facial image in calculating face video to be verified with respect to former frame
The locomotion speed value of facial image, by first open and close eyes characteristic value, first block a characteristic value and face
The locomotion speed value of image is inputted to the blink grader of the videl stage, verifies face video to be verified
Whether corresponding face is true man's living body faces.
As a kind of preferred embodiment, step S45 specifically includes following sub-step:
Step S45.1:Using face tracking and face key feature points alignment schemes, before obtaining in video
The coordinate information of the face key feature points of two frame facial images afterwards.
Step S45.2:According to the coordinate letter of two frame facial image key feature points before and after in step S45.1
Breath, calculates the relative motion size scores value of face in front and rear two frame facial images.
Above-mentioned steps are related to the locomotion speed value for calculating facial image in face video to be verified, need
It is noted that calculating the method for the locomotion speed value of facial image can have a variety of, this is entered below
Row is illustrated.In one alternate embodiment, adjacent two frame in face video to be verified is obtained
The coordinate information of the face key feature points of facial image, according to the coordinate information of face key feature points
Calculate the locomotion speed value of facial image in face video to be verified.
Step S46:To every section of positive sample and each frame of the short-sighted frequency of negative sample, respectively repeat steps
S43-S45, calculates the feature scores value that opens and closes eyes per frame facial image, blocks feature scores value and a phase
To motion size scores value, these fractional values are stitched together, as total feature of this section of short-sighted frequency,
Dimension is 3n.
Make by the first open and close eyes characteristic value, the first movement velocity for blocking a characteristic value and facial image
In the case of foundation to judge live body checking, in one alternate embodiment, opened and closed eyes spy first
Value indicative indicates that the face video to be verified has blink action, first to block a characteristic value and indicates each frame
The unobstructed phenomenon of face eyes image, and the locomotion speed value of facial image is less than predetermined threshold
In the case of, it is determined that the corresponding face of face video to be verified is living body faces, is verified.
Step S47:To all positive and negative short-sighted frequencies of sample in step S41 and step S42 according to step
S46 calculates its feature, can use the blink grader of conventional classifier training videl stage, conventional
Grader such as linear classifier, SVMs, random forest etc..
The present embodiment proposes a kind of method of the live body checking based on blink detection.Its specific steps bag
The deep neural network model for training the classification that opens and closes eyes is included, the deep neural network of eye classification is blocked in training
Model;Use the disaggregated model that opens and closes eyes trained to extract the feature that opens and closes eyes, use that is trained to block
Eye disaggregated model, which is extracted, blocks a feature, and extracts the velocity characteristic of face motion;To whether blinking
Positive and negative sample extract first characteristic value and first that opens and closes eyes respectively and block the class of a characteristic value two or first open
The category feature of locomotion speed value three that eye closing characteristic value, first block a characteristic value and facial image, training
The blink grader of videl stage;Use the blink grader of trained videl stage to treat checking face to regard
Frequency carries out live body checking.
The live body verification method that the present embodiment is provided, using eyes image as input, builds many layer depths
Spend convolutional neural networks, many layer depth convolutional neural networks pass through convolutional layer, down-sampled layer, non-
Linear layer is sequentially connected, and last layer is the full articulamentum of a f dimension, and the state opened and closed eyes is as defeated
Go out layer;Using the eyes image in training set and the state that opens and closes eyes, to the depth convolutional Neural built
Network is trained, and the training is based on back-propagation algorithm, and stochastic gradient is utilized on the training data
Decline and update model parameter.
Sum it up, the live body verification method of the present embodiment needs to have both sides function:" detection
True man blink " and " preventing dummy from cheating ", it is that the live body verification method of this present embodiment is devised and regarded
The blink grader of frequency level, the comprehensive feature for the use of opening and closing eyes, blocking eye, speed etc. three.
So as to realize can either effective detection frame true man blink, can prevent criminal from using the hand such as photo again
Section fraud system.
When judging the action of obvious blink according to the grader that opens and closes eyes, and block the classification of a grader
Result is true man's human eye, and velocity sorting device is when judging without big speed, is only in this case very
Just effective blink.
The possible space-consuming of face video to be verified is larger, and in one alternate embodiment, extraction is treated
Each frame face eyes image in the face video of checking includes:Using sliding window to be verified
Short-sighted frequency is obtained in face video, each frame face eyes image in short-sighted frequency is extracted.If at it
In the action of true man's blink is detected in any one short-sighted frequency, then illustrate face video pair to be verified
The face answered is true man's face, is verified;If comprising short-sighted frequency in be not detected by true man
The action of blink, then it is not true man's face to illustrate the corresponding face of face video to be verified.
Fig. 5 treats checking face video by the present embodiment using the blink grader for the videl stage trained
The flow chart of live body checking is carried out, as shown in figure 5, the flow comprises the following steps:
Step S51:It is right every time using the sliding window that length is n for face video to be verified
The short-sighted frequency of n frames composition in window.
Step S52:Its feature is calculated according to step S43-46 to short-sighted frequency.
Step S53:Then the blink videl stage grader obtained using being trained in step S47 is to being calculated
Feature classified, so as to judge whether the short-sighted frequency in the sliding window is to have blink action.
Step S54:For face video to be verified, sliding window is constantly moved, if sliding window
Intraoral short-sighted frequency has been judged as blink action in step S53, then this section of video live body checking is logical
Cross;Otherwise, this section of video live body checking does not pass through.
Embodiment 2
The present embodiment provides a kind of live body checking device, as shown in fig. 6, including:Acquisition module 62,
The face video to be verified for obtaining;Extraction module 64, for extracting the face video to be verified
In multiframe face eyes image;The First Eigenvalue acquisition module 66, for the multiframe face eye
Image open and close eyes judgement, obtains first and opens and closes eyes characteristic value;Second Eigenvalue acquisition module 68,
For carrying out blocking eye judgement to multiframe face eyes image, obtain first and block a characteristic value;Checking
Module 70, for according at least to first open and close eyes characteristic value and this first block a characteristic value and verify that this is treated
Whether the corresponding face of face video of checking is true man's living body faces.
Solved by said apparatus in correlation technique, by checking that blink motion carries out live body checking
During, it is impossible to exclude in eye formation local motion to simulate the camouflage blink motion of blink, cause
The incorrect problem of live body the result, so as to eliminate forgery blink situation to live body the result
Influence.
Alternatively, authentication module 70 specifically for first open and close eyes characteristic value indicate it is described to be verified
Face video has blink action, and first blocks a characteristic value instruction equal nothing of multiframe face eyes image
In the case of eclipse phenomena, it is determined that the corresponding face of face video to be verified is true man's living body faces.
Alternatively, the First Eigenvalue acquisition module 66 is specifically for the input of multiframe face eyes image is opened
Close one's eyes the deep neural network of classification, obtain first and open and close eyes characteristic value;Wherein, open and close eyes classification
Deep neural network is used to judge whether multiframe face eyes image has the phenomenon opened eyes or closed one's eyes.
Alternatively, Second Eigenvalue acquisition module 68 hides specifically for multiframe face eyes image is inputted
The deep neural network of eye classification is kept off, first is obtained and blocks a characteristic value;Wherein, eye classification is blocked
Deep neural network is used to judge that multiframe face eyes image is true man's eye image or the puppet being blocked
Fill eye image.
Alternatively, authentication module 70 by first characteristic value and first that opens and closes eyes specifically for blocking a feature
It is worth the blink grader of input video level, a characteristic value is blocked according to first characteristic value and first that opens and closes eyes
Whether the checking corresponding face of face video to be verified is true man's living body faces;Wherein, videl stage
Blink grader is used to verify whether the corresponding face of face video to be verified is true man's living body faces.
Alternatively, device also includes:First training module 72, for the eye figure opened eyes using multiple
Picture and the eyes image of multiple eye closings train the deep neural network of the classification that opens and closes eyes.
Alternatively, the device also includes:Second training module 74, for using multiple true man's faces
The eyes image of eyes image and multiple camouflage living body faces trains the depth nerve net that this blocks eye classification
Network;Wherein, the eyes image of the camouflage living body faces is to be simulated by blocking the eye of living body faces
The camouflage living body faces image of blink action.
Alternatively, the device also includes:3rd training module 76, for first to be opened and closed eyes characteristic value
With this first block the blink grader of characteristic value input video level before, train blinking for the videl stage
Eye grader, as shown in fig. 7, the 3rd training module 76 includes:First acquisition unit 762, is used for
Obtaining includes the positive sample of video that true man normally blink and the negative sample of video do not blinked normally including true man
Example;First extraction unit 764, for the deep neural network by the classification that opens and closes eyes from the positive sample of video
The second characteristic value opened and closed eyes of the multiple image extracted in sample in video is born with video, and passes through this
The deep neural network for blocking eye classification is extracted in video from the positive sample of the video and the negative sample of the video
Multiple image the second characteristic value for blocking eye;Training unit 766, for what is opened and closed eyes using second
Characteristic value and second characteristic value for blocking eye train the blink grader of the videl stage.
As shown in figure 8, extraction module 64 also includes:Second acquisition unit 822, for using slip
Window obtains short-sighted frequency from face video to be verified;Second extraction unit 824, it is short-sighted for extracting
The multiframe face eyes image in frequency.
As shown in figure 9, the device also includes:Computing module 92, for calculating, face to be verified is regarded
The locomotion speed value of facial image in frequency;Authentication module 70 specifically for by first open and close eyes characteristic value,
First locomotion speed value for blocking a characteristic value and facial image is inputted to the blink classification of the videl stage
Device, verifies whether the corresponding face of face video to be verified is true man's living body faces.
Alternatively, authentication module 70 is additionally operable to indicate the people to be verified in first characteristic value that opens and closes eyes
Face video has blink action, this first blocks characteristic value and indicate that multiframe face eyes image is unobstructed
Phenomenon, and the facial image locomotion speed value be less than predetermined threshold in the case of, it is determined that this is treated
The corresponding face of face video of checking is true man's living body faces, is verified.
As shown in Figure 10, computing module 92 includes:3rd acquiring unit 922, it is to be tested for obtaining
The coordinate information of the face key feature points of adjacent two frames facial image in the face video of card;Meter
Unit 924 is calculated, for calculating face video to be verified according to the coordinate information of face key feature points
The locomotion speed value of middle facial image.
Embodiment 3
A kind of live body checking system is present embodiments provided, including:Additionally provide a kind of live body checking system
System, including:Camera device, the face video to be verified for catching;Processor, with camera device
Connection, face video to be verified and following steps are performed for receiving:Face to be verified is extracted to regard
Multiframe face eyes image in frequency;Multiframe face eyes image is carried out to open and close eyes judgement, the is obtained
One opens and closes eyes characteristic value;Multiframe face eyes image is carried out to block eye judgement, first is obtained and blocks eye
Characteristic value;According at least to this first open and close eyes characteristic value and this first block a characteristic value and verify that this is to be tested
Whether the corresponding face of face video of card is true man's living body faces.
In summary, the embodiment of the present invention proposes a kind of robust and is efficiently based on blinking judging that live body is tested
Card method, has used deep neural network model, not only trained the classification for efficiently judging to open and close eyes
Device, also trained can effectively judge whether to use papery photo or electronics screen display by way of forgery
Human eye in photo simulates the situation of blink.Meanwhile, the global information of face motion is also added into,
Trained based on short-sighted frequency videl stage blink judge module so that live body verify robust and it is efficient.This
Invention had both been adapted to active live body checking (user makes blink action according to prompting), also was adapted for
(user is noninductive, as long as detecting user in use for the live body checking of passive type (silent formula)
There is blink action, just live body is verified).Open and close eyes and hide using deep neural network model training
Eye grader is kept off, respectively to being whether eye opening or eye closing, eye have blocked good classifying quality.
In addition to the feature that opens and closes eyes, it is also added into blocking a feature and velocity characteristic, for being effective against various shapes
The forgery blink situation of formula has significant effect.Deep neural network model occupies little space, whole to calculate
Method flow operand expense is small, can on general mobile intelligent terminal process flow operation, and blink inspection
Survey effect good.
Obviously, above-described embodiment is only intended to clearly illustrate example, and not to embodiment party
The restriction of formula.For those of ordinary skill in the field, may be used also on the basis of the above description
To make other changes in different forms.There is no need and unable to give all embodiments
With exhaustion.And the obvious changes or variations thus extended out is still in the guarantor of the invention
Protect among scope.
Claims (25)
1. a kind of live body verification method, it is characterised in that including:
Obtain face video to be verified;
Extract the multiframe face eyes image in the face video to be verified;
The multiframe face eyes image is carried out to open and close eyes judgement, first is obtained and opens and closes eyes characteristic value;
The multiframe face eyes image is carried out to block eye judgement, first is obtained and blocks a characteristic value;
Block and treated described in the checking of characteristic value according at least to described first characteristic value and described first that opens and closes eyes
Whether the corresponding face of face video of checking is true man's living body faces.
2. according to the method described in claim 1, it is characterised in that described according at least to described first
The characteristic value that opens and closes eyes and described first blocks characteristic value and verifies that the face video to be verified is corresponding
Whether face is that true man's living body faces include:
Indicate that the face video to be verified has blink action in described first characteristic value that opens and closes eyes, and
And in the case that described first blocks a characteristic value instruction unobstructed phenomenon of multiframe face eyes image,
It is true man's living body faces then to determine the corresponding face of the face video to be verified.
3. according to the method described in claim 1, it is characterised in that to the multiframe face eye figure
As carrying out opening and closing eyes judgement, obtaining first characteristic value that opens and closes eyes includes:
The multiframe face eyes image is inputted to the deep neural network for the classification that opens and closes eyes, obtains described
First opens and closes eyes characteristic value;Wherein, the deep neural network of the classification that opens and closes eyes is used to judge described
Whether multiframe face eyes image has the phenomenon opened eyes or closed one's eyes.
4. according to the method described in claim 1, it is characterised in that to the multiframe face eye figure
As carrying out blocking eye judgement, obtain first and block a characteristic value including:
The multiframe face eyes image is inputted to the deep neural network for blocking eye classification, obtains described
First blocks a characteristic value;Wherein, the deep neural network for blocking eye classification is used to judge described
Multiframe face eyes image is true man's eye image or the camouflage eye image that is blocked.
5. according to the method described in claim 1, it is characterised in that open and close according at least to described first
Eye characteristic value and described first blocks a corresponding face of the characteristic value checking face video to be verified
Whether it is that true man's living body faces include:
Described first is opened and closed eyes characteristic value and described first blink for blocking characteristic value input video level
Grader, verifies whether the corresponding face of the face video to be verified is true man's living body faces;Its
In, the blink grader of the videl stage is used to verify the corresponding face of the face video to be verified
Whether it is true man's living body faces.
6. method according to claim 3, it is characterised in that by the multiframe face eye figure
As inputting before the deep neural network for the classification that opens and closes eyes, the depth nerve for the classification that opened and closed eyes described in training
Network, wherein, the deep neural network for the classification that opened and closed eyes described in training includes:
Opened and closed eyes classification described in the eyes image opened eyes using multiple and multiple eyes image closed one's eyes training
Deep neural network.
7. method according to claim 4, it is characterised in that by the multiframe face eye figure
Before the deep neural network that eye classification is blocked as input, the depth nerve of eye classification is blocked described in training
Network, wherein, the deep neural network of eye classification is blocked described in training to be included:
Institute is trained using the eyes image of the eyes image of multiple true man's faces and multiple camouflage living body faces
State the deep neural network for blocking eye classification;Wherein, the eyes image of the camouflage living body faces is logical
Cross and block the eye of living body faces to simulate the camouflage living body faces image of blink action.
8. method according to claim 5, it is characterised in that described first is opened and closed eyes feature
Value and described first block the blink grader of characteristic value input video level before, train the video
The blink grader of level, wherein, training the blink grader of the videl stage includes:
Obtain the video for including the positive sample of video that true man normally blink and not blinking normally including true man negative
Sample;
It is negative from the positive sample of the video and the video by the deep neural network of the classification that opens and closes eyes
Second that the multiple image in video is extracted in sample opens and closes eyes characteristic value, and blocks eye point by described
The deep neural network of class extracts many in video from the positive sample of the video and the negative sample of the video
The second of two field picture blocks a characteristic value;
Blocked and regarded described in the characteristic value training of eye using second characteristic value opened and closed eyes and described second
The blink grader of frequency level.
9. according to the method described in claim 1, it is characterised in that described to extract described to be verified
Multiframe face eyes image in face video includes:
Using sliding window short-sighted frequency is obtained from the face video to be verified;
Extract the multiframe face eyes image in the short-sighted frequency.
10. according to the method described in claim 1, it is characterised in that methods described also includes:
Calculate the locomotion speed value of facial image in the face video to be verified;
It is described to block characteristic value according at least to described first characteristic value and described first that opens and closes eyes and verify an institute
Whether state the corresponding face of face video to be verified is true man's living body faces, including:
According to described first open and close eyes characteristic value, described first block a characteristic value and the face figure
The locomotion speed value of picture verifies whether the corresponding face of the face video to be verified is true man live body people
Face.
11. method according to claim 10, it is characterised in that described to be opened according to described first
The locomotion speed value checking that eye closing characteristic value, described first block a characteristic value and the facial image
Whether the corresponding face of the face video to be verified is that true man's living body faces include:
Indicate that the face video to be verified has blink action, institute in described first characteristic value that opens and closes eyes
State first and block a characteristic value instruction unobstructed phenomenon of multiframe face eyes image, and the face
In the case that the locomotion speed value of image is less than predetermined threshold, it is determined that the face video to be verified
Corresponding face is true man's living body faces.
12. the method according to claim 10 or 11, it is characterised in that calculate described to be verified
Face video in the locomotion speed value of facial image include:
Obtain the face key feature points of the adjacent two frames facial image in the face video to be verified
Coordinate information;
People in the face video to be verified is calculated according to the coordinate information of the face key feature points
The locomotion speed value of face image.
13. a kind of live body verifies device, it is characterised in that including:
Acquisition module, the face video to be verified for obtaining;
Extraction module, for extracting the multiframe face eyes image in the face video to be verified;
The First Eigenvalue acquisition module, for carrying out the judgement that opens and closes eyes to the multiframe face eyes image,
First is obtained to open and close eyes characteristic value;
Second Eigenvalue acquisition module, for carrying out blocking eye judgement to the multiframe face eyes image,
Obtain first and block a characteristic value;
Authentication module, for being opened and closed eyes characteristic value and described first to block eye special according at least to described first
Value indicative verifies whether the corresponding face of the face video to be verified is true man's living body faces.
14. device according to claim 13, it is characterised in that the authentication module is specifically used
In indicating that the face video to be verified has blink action in described first characteristic value that opens and closes eyes, and
In the case that described first blocks a characteristic value instruction unobstructed phenomenon of multiframe face eyes image, then
It is true man's living body faces to determine the corresponding face of the face video to be verified.
15. device according to claim 13, it is characterised in that the First Eigenvalue is obtained
Module specifically for the multiframe face eyes image is inputted into the deep neural network of classification of opening and closing eyes,
Described first is obtained to open and close eyes characteristic value;Wherein, the deep neural network of the classification that opens and closes eyes is used for
Judge whether the multiframe face eyes image has the phenomenon opened eyes or closed one's eyes.
16. device according to claim 13, it is characterised in that the Second Eigenvalue is obtained
Module blocks the deep neural network of eye classification specifically for the multiframe face eyes image is inputted,
Obtain first and block a characteristic value;Wherein, the deep neural network for blocking eye classification is used to judge
The multiframe face eyes image is true man's eye image or the camouflage eye image that is blocked.
17. device according to claim 13, it is characterised in that the authentication module is specifically used
Opened and closed eyes characteristic value and described first blink point for blocking characteristic value input video level in by described first
Class device, opens and closes eyes characteristic value and described first to block the checking of characteristic value described to be tested according to described first
Whether the corresponding face of face video of card is true man's living body faces;Wherein, the blink of the videl stage
Grader is used to verify whether the corresponding face of the face video to be verified is true man's living body faces.
18. device according to claim 15, it is characterised in that described device also includes:
First training module, for the eyes image and multiple eyes images closed one's eyes opened eyes using multiple
Open and close eyes the deep neural network of classification described in training.
19. device according to claim 16, it is characterised in that described device also includes:
Second training module, for the eyes image using multiple true man's faces and multiple camouflage live body people
The deep neural network of eye classification is blocked described in the eyes image training of face;Wherein, the camouflage live body
The eyes image of face is the camouflage live body people that blink action is simulated by blocking the eye of living body faces
Face image.
20. device according to claim 17, it is characterised in that described device also includes:
3rd training module, for described first characteristic value and described first that opens and closes eyes to be blocked into a feature
Before the blink grader for being worth input video level, the blink grader of the videl stage is trained, wherein,
3rd training module includes:
First acquisition unit, includes the positive sample of video that true man normally blink and including very for obtaining
The video that people normally blinks bears sample;
First extraction unit, for the deep neural network by the classification that opens and closes eyes from the video
Positive sample and the video bear the second feature opened and closed eyes of the multiple image extracted in sample in video
Value, and deep neural network that eye classifies is blocked from the positive sample of the video and the video by described
Second of the multiple image in the video characteristic value for blocking eye is extracted in negative sample;
Training unit, the spy for blocking eye using second characteristic value opened and closed eyes and described second
Value indicative trains the blink grader of the videl stage.
21. device according to claim 13, it is characterised in that the extraction module includes:
Second acquisition unit, for obtaining short from the face video to be verified using sliding window
Video;
Second extraction unit, for extracting the multiframe face eyes image in the short-sighted frequency.
22. device according to claim 13, it is characterised in that described device also includes:
Computing module, the locomotion speed value for calculating facial image in the face video to be verified;
The authentication module specifically for according to described first open and close eyes characteristic value, described first block eye
The locomotion speed value of characteristic value and the facial image verifies that the face video to be verified is corresponding
Whether face is true man's living body faces.
23. device according to claim 22, it is characterised in that the authentication module is specifically used
In described first open and close eyes characteristic value indicate the face video to be verified have blink action, it is described
First, which blocks a characteristic value, indicates the unobstructed phenomenon of multiframe face eyes image, and the face figure
In the case that the locomotion speed value of picture is less than predetermined threshold, it is determined that the face video pair to be verified
The face answered is true man's living body faces.
24. the device according to claim 22 or 23, it is characterised in that the computing module bag
Include:
3rd acquiring unit, for obtaining the adjacent two frames face figure in the face video to be verified
The coordinate information of the face key feature points of picture;
Computing unit, for calculating described to be verified according to the coordinate information of the face key feature points
Face video in facial image locomotion speed value.
25. a kind of live body verifies system, it is characterised in that including:
Camera device, the face video to be verified for catching;
Processor, is connected with the camera device, for receiving the face video to be verified and holding
Row following steps:
Extract the multiframe face eyes image in the face video to be verified;
The multiframe face eyes image is carried out to open and close eyes judgement, first is obtained and opens and closes eyes characteristic value;
The multiframe face eyes image is carried out to block eye judgement, first is obtained and blocks a characteristic value;
Block and treated described in the checking of characteristic value according at least to described first characteristic value and described first that opens and closes eyes
Whether the corresponding face of face video of checking is true man's living body faces.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201610051911.6A CN106997452B (en) | 2016-01-26 | 2016-01-26 | Living body verification method and device |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201610051911.6A CN106997452B (en) | 2016-01-26 | 2016-01-26 | Living body verification method and device |
Publications (2)
Publication Number | Publication Date |
---|---|
CN106997452A true CN106997452A (en) | 2017-08-01 |
CN106997452B CN106997452B (en) | 2020-12-29 |
Family
ID=59428347
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201610051911.6A Active CN106997452B (en) | 2016-01-26 | 2016-01-26 | Living body verification method and device |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN106997452B (en) |
Cited By (11)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN108334817A (en) * | 2018-01-16 | 2018-07-27 | 深圳前海华夏智信数据科技有限公司 | Living body faces detection method and system based on three mesh |
CN108921117A (en) * | 2018-07-11 | 2018-11-30 | 北京市商汤科技开发有限公司 | Image processing method and device, electronic equipment and storage medium |
CN109635554A (en) * | 2018-11-30 | 2019-04-16 | 努比亚技术有限公司 | A kind of red packet verification method, terminal and computer storage medium |
CN109840406A (en) * | 2017-11-29 | 2019-06-04 | 百度在线网络技术(北京)有限公司 | Living body verification method, device and computer equipment |
CN109902667A (en) * | 2019-04-02 | 2019-06-18 | 电子科技大学 | Human face in-vivo detection method based on light stream guide features block and convolution GRU |
CN110287900A (en) * | 2019-06-27 | 2019-09-27 | 深圳市商汤科技有限公司 | Verification method and verifying device |
CN111274846A (en) * | 2018-12-04 | 2020-06-12 | 北京嘀嘀无限科技发展有限公司 | Method and system for identifying opening and closing actions |
CN111340014A (en) * | 2020-05-22 | 2020-06-26 | 支付宝(杭州)信息技术有限公司 | Living body detection method, living body detection device, living body detection apparatus, and storage medium |
CN111860056A (en) * | 2019-04-29 | 2020-10-30 | 北京眼神智能科技有限公司 | Blink-based in-vivo detection method and device, readable storage medium and equipment |
WO2020252740A1 (en) * | 2019-06-20 | 2020-12-24 | 深圳市汇顶科技股份有限公司 | Convolutional neural network, face anti-spoofing method, processor chip, and electronic device |
WO2024032574A1 (en) * | 2022-08-12 | 2024-02-15 | 北京字跳网络技术有限公司 | Image processing method and apparatus, and electronic device and storage medium |
Citations (7)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN101999900A (en) * | 2009-08-28 | 2011-04-06 | 南京壹进制信息技术有限公司 | Living body detecting method and system applied to human face recognition |
CN102004904A (en) * | 2010-11-17 | 2011-04-06 | 东软集团股份有限公司 | Automatic teller machine-based safe monitoring device and method and automatic teller machine |
CN103400122A (en) * | 2013-08-20 | 2013-11-20 | 江苏慧视软件科技有限公司 | Method for recognizing faces of living bodies rapidly |
CN103440479A (en) * | 2013-08-29 | 2013-12-11 | 湖北微模式科技发展有限公司 | Method and system for detecting living body human face |
US8856541B1 (en) * | 2013-01-10 | 2014-10-07 | Google Inc. | Liveness detection |
CN104348778A (en) * | 2013-07-25 | 2015-02-11 | 信帧电子技术(北京)有限公司 | Remote identity authentication system, terminal and method carrying out initial face identification at handset terminal |
FR2915008B1 (en) * | 2007-04-12 | 2015-04-17 | Sagem Defense Securite | METHOD FOR DETECTING THE LIVING CHARACTER OF A BODY AREA AND OPTICAL DEVICE FOR CARRYING OUT SAID METHOD |
-
2016
- 2016-01-26 CN CN201610051911.6A patent/CN106997452B/en active Active
Patent Citations (7)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
FR2915008B1 (en) * | 2007-04-12 | 2015-04-17 | Sagem Defense Securite | METHOD FOR DETECTING THE LIVING CHARACTER OF A BODY AREA AND OPTICAL DEVICE FOR CARRYING OUT SAID METHOD |
CN101999900A (en) * | 2009-08-28 | 2011-04-06 | 南京壹进制信息技术有限公司 | Living body detecting method and system applied to human face recognition |
CN102004904A (en) * | 2010-11-17 | 2011-04-06 | 东软集团股份有限公司 | Automatic teller machine-based safe monitoring device and method and automatic teller machine |
US8856541B1 (en) * | 2013-01-10 | 2014-10-07 | Google Inc. | Liveness detection |
CN104348778A (en) * | 2013-07-25 | 2015-02-11 | 信帧电子技术(北京)有限公司 | Remote identity authentication system, terminal and method carrying out initial face identification at handset terminal |
CN103400122A (en) * | 2013-08-20 | 2013-11-20 | 江苏慧视软件科技有限公司 | Method for recognizing faces of living bodies rapidly |
CN103440479A (en) * | 2013-08-29 | 2013-12-11 | 湖北微模式科技发展有限公司 | Method and system for detecting living body human face |
Non-Patent Citations (3)
Title |
---|
ASAD ALI ETAL.: "Liveness Detection using Gaze Collinearity", 《2012 THIRD INTERNATIONAL CONFERENCE ON EMERGING SECURITY TECHNOLOGIES》 * |
GANG PAN ETAL.: "Eyeblink-based Anti-Spoofing in Face Recognition from a Generic Webcamera", 《2007 IEEE 11TH INTERNATIONAL CONFERENCE ON COMPUTER VISION》 * |
孙霖等: "人脸识别中的活体检测技术研究", 《万方数据库》 * |
Cited By (18)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN109840406A (en) * | 2017-11-29 | 2019-06-04 | 百度在线网络技术(北京)有限公司 | Living body verification method, device and computer equipment |
CN108334817A (en) * | 2018-01-16 | 2018-07-27 | 深圳前海华夏智信数据科技有限公司 | Living body faces detection method and system based on three mesh |
CN108921117A (en) * | 2018-07-11 | 2018-11-30 | 北京市商汤科技开发有限公司 | Image processing method and device, electronic equipment and storage medium |
WO2020010927A1 (en) * | 2018-07-11 | 2020-01-16 | 北京市商汤科技开发有限公司 | Image processing method and apparatus, electronic device, and storage medium |
JP2021516405A (en) * | 2018-07-11 | 2021-07-01 | ベイジン センスタイム テクノロジー デベロップメント カンパニー, リミテッド | Image processing methods and devices, electronic devices and storage media |
CN109635554A (en) * | 2018-11-30 | 2019-04-16 | 努比亚技术有限公司 | A kind of red packet verification method, terminal and computer storage medium |
CN111274846A (en) * | 2018-12-04 | 2020-06-12 | 北京嘀嘀无限科技发展有限公司 | Method and system for identifying opening and closing actions |
CN111274846B (en) * | 2018-12-04 | 2023-09-19 | 北京嘀嘀无限科技发展有限公司 | Method and system for identifying opening and closing actions |
CN109902667A (en) * | 2019-04-02 | 2019-06-18 | 电子科技大学 | Human face in-vivo detection method based on light stream guide features block and convolution GRU |
CN111860056B (en) * | 2019-04-29 | 2023-10-20 | 北京眼神智能科技有限公司 | Blink-based living body detection method, blink-based living body detection device, readable storage medium and blink-based living body detection equipment |
CN111860056A (en) * | 2019-04-29 | 2020-10-30 | 北京眼神智能科技有限公司 | Blink-based in-vivo detection method and device, readable storage medium and equipment |
CN112424795A (en) * | 2019-06-20 | 2021-02-26 | 深圳市汇顶科技股份有限公司 | Convolutional neural network, face anti-counterfeiting method, processor chip and electronic equipment |
WO2020252740A1 (en) * | 2019-06-20 | 2020-12-24 | 深圳市汇顶科技股份有限公司 | Convolutional neural network, face anti-spoofing method, processor chip, and electronic device |
CN112424795B (en) * | 2019-06-20 | 2024-05-28 | 深圳市汇顶科技股份有限公司 | Face anti-counterfeiting method, processor chip and electronic equipment |
CN110287900A (en) * | 2019-06-27 | 2019-09-27 | 深圳市商汤科技有限公司 | Verification method and verifying device |
CN111340014B (en) * | 2020-05-22 | 2020-11-17 | 支付宝(杭州)信息技术有限公司 | Living body detection method, living body detection device, living body detection apparatus, and storage medium |
CN111340014A (en) * | 2020-05-22 | 2020-06-26 | 支付宝(杭州)信息技术有限公司 | Living body detection method, living body detection device, living body detection apparatus, and storage medium |
WO2024032574A1 (en) * | 2022-08-12 | 2024-02-15 | 北京字跳网络技术有限公司 | Image processing method and apparatus, and electronic device and storage medium |
Also Published As
Publication number | Publication date |
---|---|
CN106997452B (en) | 2020-12-29 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN106997452A (en) | Live body verification method and device | |
CN103679118B (en) | A kind of human face in-vivo detection method and system | |
CN105631439B (en) | Face image processing process and device | |
CN104966070B (en) | Biopsy method and device based on recognition of face | |
CN108549854B (en) | A kind of human face in-vivo detection method | |
CN110516616A (en) | A kind of double authentication face method for anti-counterfeit based on extensive RGB and near-infrared data set | |
CN106557726A (en) | A kind of band is mourned in silence the system for face identity authentication and its method of formula In vivo detection | |
CN108182409A (en) | Biopsy method, device, equipment and storage medium | |
CN105844206A (en) | Identity authentication method and identity authentication device | |
CN108304858A (en) | Fight specimen discerning model generating method, verification method and its system | |
CN108229325A (en) | Method for detecting human face and system, electronic equipment, program and medium | |
CN102483851B (en) | Method and apparatus for prominent facial features recognition | |
CN110223322B (en) | Image recognition method and device, computer equipment and storage medium | |
CN109886244A (en) | A kind of recognition of face biopsy method and device | |
KR20200091799A (en) | Method and system for integrity verification of fake video created through deep learning | |
CN108596041A (en) | A kind of human face in-vivo detection method based on video | |
CN108369785A (en) | Activity determination | |
CN109359502A (en) | False-proof detection method and device, electronic equipment, storage medium | |
CN109376608A (en) | A kind of human face in-vivo detection method | |
TW200842733A (en) | Object image detection method | |
CN107368769A (en) | Human face in-vivo detection method, device and electronic equipment | |
CN109508706B (en) | Silence living body detection method based on micro-expression recognition and non-sensory face recognition | |
CN106919921A (en) | With reference to sub-space learning and the gait recognition method and system of tensor neutral net | |
CN109766785A (en) | A kind of biopsy method and device of face | |
CN107798279A (en) | Face living body detection method and device |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
PB01 | Publication | ||
PB01 | Publication | ||
SE01 | Entry into force of request for substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
GR01 | Patent grant | ||
GR01 | Patent grant |