CN106778496A - Biopsy method and device - Google Patents
Biopsy method and device Download PDFInfo
- Publication number
- CN106778496A CN106778496A CN201611039845.7A CN201611039845A CN106778496A CN 106778496 A CN106778496 A CN 106778496A CN 201611039845 A CN201611039845 A CN 201611039845A CN 106778496 A CN106778496 A CN 106778496A
- Authority
- CN
- China
- Prior art keywords
- lip
- character
- verification code
- random verification
- characteristic vector
- 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.)
- Pending
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
- 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
- G06V40/174—Facial expression recognition
- G06V40/176—Dynamic expression
Landscapes
- Engineering & Computer Science (AREA)
- Human Computer Interaction (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Multimedia (AREA)
- Theoretical Computer Science (AREA)
- Health & Medical Sciences (AREA)
- General Health & Medical Sciences (AREA)
- Oral & Maxillofacial Surgery (AREA)
- Image Analysis (AREA)
Abstract
The present invention provides a kind of biopsy method and device, and for the facial image of detection and identification object to recognize whether it is living person, the method includes：Collection differentiates that object reads the video image of random verification code；Obtain the characteristic vector of the lip image sequence of lip region in video image described in per frame；The characteristic vector according to continuous multiple frames calls the lip reading identification model of training in advance to recognize the lip information for differentiating object；Detect whether the lip information is consistent with character in random verification code；When the lip information is consistent with character in random verification code, discriminating object is live body.Relative to traditional discrimination method, lip feature is also acquired while picture pick-up device collection face, without additionally increasing any hardware device newly, reduce the cost of checking system, it is more convenient to use；Differentiate that object reads random verification code and can directly determine whether live body, not only increase the security and anti-counterfeit capability of identifying system, also improve the efficiency of live body checking.
Description
Technical field
The present invention relates to biometrics identification technology field, more particularly to a kind of live body inspection based on the identification of image lip reading
Survey method and device.
Background technology
With the development of biometrics identification technology, face identification method has become a kind of the conventional of confirmation user identity
Method.In the prior art, face live body mirror method for distinguishing is increased in some face identification methods, face can be preferably carried out
Detection and knowledge.
However, existing face recognition process, during being based particularly on face recognition, illegal registrant can be by puppet
Make children face " deception " camera or other image capture devices so that the photo that image capture device is obtained is not live body
Human face photo.For example before being placed in image capture device using the human face photo or face video fragment of registrant, image is adopted
The mug shot of the active user acquired in collection equipment comes from photo or video segment actually, or, illegal registrant
Three-dimensional face model can also be forged, before the three-dimensional face model is placed in into image capture device, now image capture device
Acquired human face photo is the photo of three-dimensional face model, but in the feature carried out based on face and the comparison being distributed,
This point cannot be discovered, the anti-counterfeit capability that this has resulted in identification system is weak, the low problem of security.
The content of the invention
The shortcoming of prior art in view of the above, it is an object of the invention to provide a kind of biopsy method and dress
Put, whether the face for solving that detection object cannot be determined in the prior art is live body, causes the false proof of identification system
Ability is weak, the low problem of security.
In order to achieve the above objects and other related objects, the present invention provides a kind of biopsy method, for detection and identification
To recognize whether it is living person, the biopsy method includes the facial image of object：
Collection differentiates that object reads the video image of random verification code；
Obtain the characteristic vector of the lip image sequence of lip region in video image described in per frame；
The characteristic vector according to continuous multiple frames calls the lip reading identification model of training in advance to recognize the lip for differentiating object
Information；
Detect whether the lip information is consistent with character in random verification code；When the lip information and random verification code
When middle character is consistent, discriminating object is live body.
Another object of the present invention is to provide living body detection device, for the facial image of detection and identification object recognizing
Whether it is living person, including：
Acquisition module, for gathering the video image for differentiating that object reads random verification code；
Characteristic extracting module, for obtain lip region in video image described in every frame lip image sequence feature to
Amount；
Lip reading identification module, the lip reading identification model for calling training in advance for the characteristic vector according to continuous multiple frames is known
Not Jian Bie object lip information；
Detection module, it is whether consistent with character in random verification code for detecting the lip information；When lip letter
When breath is consistent with character in random verification code, discriminating object is live body.
As described above, biopsy method of the invention and device, have the advantages that：
By gathering the video image for differentiating that object reads random verification code, the video image is pre-processed successively,
Segmentation, alignment, so as to extract the characteristic vector of the lip image sequence for differentiating object；Lip reading identification mould according to default training
The corresponding lip reading information of the characteristic vector of type identification lip image sequence, detects the lip reading information with word in random verification code
Whether whether symbol is consistent, differentiates whether object is live body according to the completely the same determination of character.Make relative to traditional discrimination method
Used time, lip feature is also acquired while picture pick-up device collection face, without additionally increasing any hardware device newly, reduce and test
The cost of card system, it is more convenient to use；Differentiate that object reads random verification code and can directly determine whether live body, not only carry
The high security and anti-counterfeit capability of identifying system, also improves the efficiency of live body checking.
Brief description of the drawings
Fig. 1 is shown as a kind of biopsy method flow chart of present invention offer；
Fig. 2 is shown as the flow chart of step S2 in a kind of biopsy method of present invention offer；
Fig. 3 be shown as the present invention offer based on lip image segmentation alignment structures schematic block diagram；
Fig. 4 is shown as the moment feature extraction mode structured flowchart of present invention offer；
Fig. 5 is shown as the ISA network structure block diagrams of present invention offer；
Fig. 6 be shown as the present invention offer based on stacking convolution ISA network structure block diagrams；
Fig. 7 is shown as the flow chart based on stacking convolution ISA network calculations video features of present invention offer；
Fig. 8 is shown as the stream that observational networks are produced based on time series cutting and HMM of present invention offer
Cheng Tu；
Fig. 9 is shown as the Hidden Markov Model state transfer figure of present invention offer；
Figure 10 is shown as the flow chart of step S3 in a kind of biopsy method of present invention offer；
Figure 11 is shown as the flow chart of step S4 in a kind of biopsy method of present invention offer；
Figure 12 is shown as a kind of living body detection device structured flowchart of present invention offer；
Figure 13 is shown as the structured flowchart of characteristic extracting module in a kind of living body detection device of present invention offer；
Figure 14 is shown as the structured flowchart of lip reading identification module in a kind of living body detection device of present invention offer；
Figure 15 is shown as the structured flowchart of detection module in a kind of living body detection device of present invention offer.
Specific embodiment
Embodiments of the present invention are illustrated below by way of specific instantiation, those skilled in the art can be by this specification
Disclosed content understands other advantages of the invention and effect easily.The present invention can also be by specific realities different in addition
The mode of applying is embodied or practiced, the various details in this specification can also based on different viewpoints with application, without departing from
Various modifications or alterations are carried out under spirit of the invention.It should be noted that, in the case where not conflicting, following examples and implementation
Feature in example can be mutually combined.
It should be noted that the diagram provided in following examples only illustrates basic structure of the invention in a schematic way
Think, component count, shape and size when only display is with relevant component in the present invention rather than according to actual implementation in schema then
Draw, it is actual when the implementing kenel of each component, quantity and ratio can be a kind of random change, and its assembly layout kenel
It is likely more complexity.
Fig. 1 is referred to, the present invention provides a kind of biopsy method flow chart, for the facial image of detection and identification object
To recognize whether it is living person, the biopsy method includes：
Step S1, collection differentiates that object reads the video image of random verification code；
Specifically, the random verification code is according to random generation in advance training set, by image capture device
Collection differentiates that object reads the video image of random verification code, wherein it is desired to the image capture device for passing through correlation carries out face
The acquisition of image；Immediately obtained by image capture device, in this manner it is ensured that acquired image is to be currently at figure
As the image of the people of the image acquisition region of collecting device.
Step S2, obtains the characteristic vector of the lip image sequence of lip region in video image described in per frame；
Specifically, lip change includes lip shape, lip texture and lip color corresponding to lip region in the application
In any one, wherein, lip image sequence comprising in lip shape sequence, lip texture sequence and lip colour sequential arbitrarily
One kind, and lip change is corresponded with lip image sequence.
Step S3, the characteristic vector according to continuous multiple frames calls the lip reading identification model of training in advance to recognize discriminating object
Lip information；
Specifically, it is right using a large amount of characters institute based on HMM before using lip reading identification model
The lip image sequence (state-transition matrix) answered is trained to the lip reading identification model, then, will with random verification code
Differentiate the characteristic vector reversal of identification of the lip image sequence corresponding to object, identify that lip reads the lip of random verification code
Information.
Step S4, detects whether the lip information is consistent with character in random verification code；When the lip information with
When character is consistent in machine identifying code, discriminating object is live body.
Specifically, as shown in figure 11, the flow chart of step S4 in a kind of biopsy method for being provided for the present invention, bag
Include：
Whether step S401, the lip reading information of detection and identification object is consistent with character in random verification code；
Step S402, when the lip reading information is consistent with character in random verification code, discriminating object is live body；
Step S403, when character is inconsistent in the lip reading information and random verification code, differentiates that object is not live body.
Face live body discrimination method common at present is determined using infrared thermal imaging detection temperature by secondary light source
It is authenticated whether object is live body；Or by receiving preset instructions, head enters horizontal deflection, to confirm to differentiate whether object is living
Body.
In the present embodiment, when in use, lip feature is also acquired while picture pick-up device collection face, without extra
Increase any hardware device newly, reduce the cost of checking system, it is more convenient to use；Differentiate that object reading random verification code can
Directly determine whether live body, not only increase the security and anti-counterfeit capability of identifying system, also improve the effect of live body checking
Rate.
Fig. 2 is referred to, the flow chart of step S2 in a kind of biopsy method provided for the present invention, details are as follows：
Step S201, carries out pre-processing the video image for obtaining default specification to video image described in every frame；
Step S202, the video image of the default specification of segmentation obtains lip region, and affine change is carried out to the lip region
Get the lip image sequence of alignment in return；
Step S203, the feature extraction algorithm based on stacking convolution independence subspace analysis network calculates the lip image
The characteristic vector of sequence.
In the present embodiment, based on human face detection tech and key point extractive technique, face in the video image is positioned
Lip；As shown in figure 3, the key point of lip is 13 in the facial image, two key points of the corners of the mouth are angle point, using two
Individual key point calculates the translation and twiddle factor relative to standard mouth.Due to different people and the facial image mouth size of different frame
Different, in order to exclude influence of the size to recognizing, residing pretreatment, will also be to all in addition to rotation and translation is converted
Image carries out the change of scale based on standard mouth.In the prior art, affine transformation only needs 3 key points of mouth, according to
Above-mentioned angle point can complete normalization alignment using geometrical relationship.However, the interframe which will lose is with respect to variability, because
The mouth entered after line translation can make all segmentations align based on corners of the mouth key point has identical width.The alignment difficult point of lip exists
In how retaining the relative change of interframe, and the mouth of different frame is normalized under identical yardstick.
Be the problem that the lip thickness and width for overcoming different people differs greatly, using eye spacing as benchmark come by difference
The lip of people transforms to identical yardstick；Specifically, in figure 3 based on lip image segmentation alignment structures schematic block diagram, canthus
Two key points and scale factor is calculated according to eye spacing, obtained translation, rotation and the scale factor of affine transformation；It is right
Facial image carries out the lip image sequence just alignd after the above-mentioned lip segmentation based on key point and affine transformation.
In order to obtain, with distinction characteristic vector between uniformity and class, being easy to follow-up recognition feature vector；For example, traditional
The dynamic visual signature of lip include：The feature being made up of the position and its difference vector of key point, e.g., the image statisticses such as HoG, SIFT
Feature etc..The algorithm that traditional characteristic is extracted is manually to analyze to design rule of thumb and to problem, due to effect characteristicses table
The factor of Danone power is very more, thus the robustness of features described above all has obvious limitation.Because in the video image of collection
Simultaneously comprising face and mouth image and its movable information, therefore, a characteristic vector is formed from multiframe consecutive image
Subimage sequence in extract, as shown in figure 4, for the present invention in moment feature extraction mode structured flowchart, in order to ensure feature
Comprising as far as possible many useful informations, there is the overlap of certain frame number between subimage sequence, i.e. each characteristic vector sequence is corresponding
Multiple subimage sequence F have frame number overlap with further feature sequence vector.
As shown in figure 5, being the ISA network structure block diagrams for providing of the invention, constituting the basis of stacking convolution ISA networks is
ISA networks, it is a two-layer neutral net, and the non-linear node of ground floor and the second layer performs quadratic sum evolution fortune respectively
Calculate, connected by weight matrix W and V respectively between input node and the first node layer, between the second layer and third layer node.
It is x to make input vectort, then the response of i-th node of the second layer isIts
In, W, V are respectively weight matrix, xtIt is input vector, m, n, i, j are respectively the integer more than 1.Parameter W in ISA networks and
V solves optimization problem by Projected descent method, specific as follows：It is minimum value, T is sample, subject
To is constraints.
subject to WWT=I
Employ the feature extraction algorithm based on stacking convolution independence subspace analysis network.Stacking convolution Independent subspace
Analysis (ISA) network is a kind of deep learning algorithm designed to extract video features, and its advantage includes：1) spy for extracting
Levy simultaneously comprising image and movable information, it is adaptable to the identification of lip motion；2) non-linear unit is simple, fast operation, fits
For extracting feature from video higher-dimension big data；3) network structure is simply clear, it is easy to accomplish；4) training method is unsupervised
, it is very convenient without artificial mark mass data.
As shown in fig. 6, based on stacking convolution ISA network structure block diagrams, when input video pixel is high dimensional data, ISA
The training process of network is very slow, and the problem can be overcome using stacking convolution ISA networks.Stacking convolution ISA networks by ISA and
Principal component analysis (PCA) successively stacks composition.The mode of convolution ISA network calculations video features is stacked, as shown in fig. 7, being this
The flow chart based on stacking convolution ISA network calculations video features that invention is provided, first, by the pixel in less video block
Ground floor ISA networks are input to after pulling into a vector, then, the ISA outputs of adjacent video block are combined in bigger region
Get up, by after the pretreatment of PCA dimensionality reductions, being input to second layer ISA networks, and so on, finally, every layer of ISA network it is defeated
Go out to be connected into characteristic vector of the vector as the video block.Due to each layer of dimension of the input data of ISA networks all
Will not be too high, and can successively train, therefore the training speed of stacking convolution ISA networks can be improved.
In order to obtain characteristic vector Ft, multiple is divided into such as Fig. 7 institutes by the video block of multiple continuous lip reading image constructions
The small video block for showing, each small video block extracts characteristic vector in being imported into stacking convolution ISA networks, finally, each is small
The characteristic vector of video block is connected into the characteristic vector F for obtainingt。
Preferably, the lip reading identification model of the training in advance, including：
Training set include several characters, the random verification code be training set in generate at random；It is every in the training set
Individual character is based on HMM and is trained to obtain the unique forecast model sequence of correspondence, the Hidden Markov mould
Type includes N number of forecast model, N >=1；The state-transition matrix and mixed Gauss model of HMM are calculated, obtains pre-
If the lip reading identification model of training；Wherein, the corresponding time series fragment of each character is produced by HMM；And institute
Each moment corresponds to a hidden state in stating time series fragment, by the change of the hidden state by state-transition matrix table
Show, each hidden state also corresponds to an observational networks, and the observational networks are modeled as into mixed Gauss model.
In the present embodiment, as shown in figure 8, for the present invention provide based on time series cutting and HMM
The flow chart of observational networks is produced, identifies that the multiple individual characters for wherein including are needed to time sequence from one section of lid speech characteristic sequence
Row are segmented, and identify the individual character corresponding to each section, and the process is similar to speech recognition, therefore using in speech recognition
Conventional hidden Markov model (HMM) come realize lip reading recognize.Specifically, the time series fragment corresponding to each individual character by
One HMM is produced, and each moment t correspondences one hidden state St, such as S1 in fragment are original state, and St is end-state.Such as
Shown in Fig. 9, Hidden Markov Model state transfer figure, in HMM, the generation probability and upper of the hidden state St at each moment
The hidden state St-1 at individual moment is relevant, and the hidden state of adjacent moment is coupled by state-transition matrix A, the i-th row jth row in A
Element aij represent the probability that state j is transferred to by state i, state-transition matrix can carry out more intuitively table by state transition diagram
Show, as shown in Figure 9.Each state possesses an one's own observational networks, and therefrom produces characteristic vector, the observational networks
Mixed Gauss model (GMM) is modeled as, it can represent complicated multi-modal distribution.
Lip reading identification model the destination of study is to estimate the state-transition matrix and GMM parameters in HMM model.Lip reading is online
Identification is then the optimum state path that characteristic sequence to be identified is estimated in the case where model parameter has determined,
And individual character path is combined into by state path, this task is completed by famous Viterbi (Viterbi) decoding algorithm.It is based on
The study of lip reading identification model and identification of HMM can come by the ripe tool box of some of field of speech recognition such as Kaldi and HTK
Complete.
Figure 10 is shown as the flow chart of step S3 in a kind of biopsy method of present invention offer, including：
Step S301, the characteristic vector of lip image sequence is matched according to HMM, according to time series meter
The optimum state path of the characteristic vector of the lip image sequence is calculated, according to the single character of optimum state Path Recognition；
In the present embodiment, with the continuous HMM of time series, voice is parsed frame by frame, according to every frame language
The static nature of sound and be behavioral characteristics relative to the dynamic change of former frame, judge character corresponding to present frame and
The residing time state in the standard voice signals of space symbol, analysis result of the series connection per frame, that is, obtain audio user signal
Voice messaging be identified as correspondence character.Wherein, the continuous HMM of the time series for being used is two-stage knot
Structure:The first order is the other Hidden Markov time series models of character level, and the received pronunciation of each character includes four by one
The single order time series models of voice status represent that the change of each voice status is only related to previous voice status, such as
Shown in Fig. 9, wherein 0.4 is initial state, 1 is final state, and the Gaussian Mixture degree of each state is 4；The second level is character string
The continuous HMM of rank, the received pronunciation of random verification code is by character of arbitrarily connecting " Hidden Markov " rank
Model is constituted, wherein containing N number of forecast model, N >=1.
Step S302, combination producing differentiates object to the single character that will be recognized corresponding to the characteristic vector in temporal sequence
Lip reading information.
According to the single character of characteristic vector correspondence identification, single character is arranged according to time series, you can reflected
Other object reads the lip reading information of random verification code.
Figure 12 is referred to, another object of the present invention is to provide living body detection device, for the people of detection and identification object
Face image recognizing whether it is living person, including：
Acquisition module 1, for gathering the video image for differentiating that object reads random verification code；
Characteristic extracting module 2, the feature of the lip image sequence for obtaining lip region in video image described in every frame
Vector；
Lip reading identification module 3, the lip reading identification model of training in advance is called for the characteristic vector according to continuous multiple frames
Identification differentiates the lip information of object；
Detection module 4, it is whether consistent with character in random verification code for detecting the lip information；When lip letter
When breath is consistent with character in random verification code, discriminating object is live body.
Figure 13 is referred to, the structured flowchart of characteristic extracting module in a kind of living body detection device provided for the present invention, bag
Include：
Pretreatment unit 21, for carrying out pre-processing the video image for obtaining default specification to video image described in every frame；
Segmentation alignment unit 22, the video image that treatment has been preset for splitting obtains lip region, to the lip area
Domain carries out the lip image sequence that affine transformation is alignd；
Feature extraction unit 23, institute is calculated for the feature extraction algorithm based on stacking convolution independence subspace analysis network
State the characteristic vector of lip image sequence.
Preferably, lip reading identification model is specially described in training in advance：
Training set include several characters, the random verification code be training set in generate at random；It is every in the training set
Individual character is based on HMM and is trained to obtain the unique forecast model sequence of correspondence, the Hidden Markov mould
Type includes N number of forecast model, N >=1；The state-transition matrix and mixed Gauss model of HMM are calculated, obtains pre-
If the lip reading identification model of training；Wherein, the corresponding time series fragment of each character is produced by HMM；And institute
Each moment corresponds to a hidden state in stating time series fragment, by the change of the hidden state by state-transition matrix table
Show, each hidden state also corresponds to an observational networks, and the observational networks are modeled as into mixed Gauss model.
Figure 14 is referred to, the structured flowchart of lip reading identification module in a kind of living body detection device provided for the present invention, bag
Include：
Recognition unit 31, the characteristic vector for matching lip image sequence according to HMM calculates described
The optimum state path of the characteristic vector of lip image sequence, state path is combined determine according to the time series of IMAQ
Individual character path is recognizing single character；
Assembled unit 32, for will corresponding to the characteristic vector recognize single character in temporal sequence combination producing mirror
The lip reading information of other object.
Figure 15 is referred to, the structured flowchart of detection module in a kind of living body detection device provided for the present invention, including：
Whether detection unit 42, the lip reading information for detection and identification object is consistent with character in random verification code；
First confirmation unit 42, for when the lip reading information is consistent with character in random verification code, differentiates that object is
Live body；
Second confirmation unit 43, for when character is inconsistent in the lip reading information and random verification code, differentiating object
It is not live body.
In sum, the present invention differentiates the video image of object reading random verification code by gathering, successively to the video
Image is pre-processed, is split, is alignd, so as to extract the characteristic vector of the lip image sequence for differentiating object；According to default
The corresponding lip reading information of the characteristic vector of the lip reading identification model identification lip image sequence of training, detects the lip reading information
It is whether consistent with character in random verification code, differentiate whether object is live body according to the whether completely the same determination of character.Relative to
Traditional discrimination method when in use, lip feature is also acquired while picture pick-up device collection face, any without additionally increasing newly
Hardware device, reduces the cost of checking system, more convenient to use；Differentiate that object reads random verification code and can directly judge
Whether it is live body, not only increases the security and anti-counterfeit capability of identifying system, also improves the efficiency of live body checking.So,
The present invention effectively overcomes various shortcoming of the prior art and has high industrial utilization.
The above-described embodiments merely illustrate the principles and effects of the present invention, not for the limitation present invention.It is any ripe
The personage for knowing this technology all can carry out modifications and changes under without prejudice to spirit and scope of the invention to above-described embodiment.Cause
This, those of ordinary skill in the art is complete with institute under technological thought without departing from disclosed spirit such as
Into all equivalent modifications or change, should be covered by claim of the invention.
Claims (10)
1. a kind of biopsy method, for the facial image of detection and identification object to recognize whether it is living person, the live body
Detection method includes：
Collection differentiates that object reads the video image of random verification code；
Obtain the characteristic vector of the lip image sequence of lip region in video image described in per frame；
The characteristic vector according to continuous multiple frames calls the lip reading identification model of training in advance to recognize the lip information for differentiating object；
Detect whether the lip information is consistent with character in random verification code；When word in the lip information and random verification code
When according with consistent, discriminating object is live body.
2. biopsy method according to claim 1, it is characterised in that mouth in acquisition video image described in per frame
The step of characteristic vector of the lip image sequence in lip region, including：
Video image described in every frame is carried out to pre-process the video image for obtaining default specification；
The video image of the default specification of segmentation obtains lip region, and the mouth that affine transformation is alignd is carried out to the lip region
Lip image sequence；
Feature extraction algorithm based on stacking convolution independence subspace analysis network calculate the feature of the lip image sequence to
Amount.
3. biopsy method according to claim 1, it is characterised in that the lip reading identification model of the training in advance,
Including：
Training set include several characters, the random verification code be training set in generate at random；Each word in the training set
Symbol is based on HMM and is trained to obtain the unique forecast model sequence of correspondence, the HMM bag
Containing N number of forecast model, N >=1；The state-transition matrix and mixed Gauss model of HMM are calculated, default instruction is obtained
Experienced lip reading identification model；Wherein, the corresponding time series fragment of each character is produced by HMM；And when described
Between in sequence fragment each moment correspond to a hidden state, the change of the hidden state is represented by state-transition matrix, often
Individual hidden state also corresponds to an observational networks, and the observational networks are modeled as into mixed Gauss model.
4. biopsy method according to claim 1, it is characterised in that described to be called in advance according to the characteristic vector
The step of lip reading identification model identification of training differentiates the lip information of object, including：
The characteristic vector of lip image sequence is matched according to HMM, the lip image is calculated according to time series
The optimum state path of the characteristic vector of sequence, according to the single character of optimum state Path Recognition；
Will corresponding to the characteristic vector recognize single character in temporal sequence combination producing differentiate object lip reading information.
5. biopsy method according to claim 1, it is characterised in that the detection lip information with test at random
Whether character is consistent in card code；When the lip information is consistent with character in random verification code, differentiate that object is the step of live body
Suddenly, including：
Whether the lip reading information of detection and identification object is consistent with character in random verification code；When the lip reading information and accidental validation
When character is consistent in code, discriminating object is live body；When character is inconsistent in the lip reading information and random verification code, it is right to differentiate
As not being live body.
6. a kind of living body detection device, it is characterised in that for the facial image of detection and identification object recognizing whether it is living
People, including：
Acquisition module, for gathering the video image for differentiating that object reads random verification code；
Characteristic extracting module, the characteristic vector of the lip image sequence for obtaining lip region in video image described in every frame；
Lip reading identification module, the lip reading identification model for calling training in advance for the characteristic vector according to continuous multiple frames recognizes mirror
The lip information of other object；
Detection module, it is whether consistent with character in random verification code for detecting the lip information；When the lip information with
When character is consistent in random verification code, discriminating object is live body.
7. living body detection device according to claim 6, it is characterised in that the characteristic extracting module includes：
Pretreatment unit, for carrying out pre-processing the video image for obtaining default specification to video image described in every frame；
Segmentation alignment unit, the video image for splitting default specification obtains lip region, the lip region is imitated
Penetrate the lip image sequence that conversion is alignd；
Feature extraction unit, the lip is calculated for the feature extraction algorithm based on stacking convolution independence subspace analysis network
The characteristic vector of image sequence.
8. living body detection device according to claim 6, it is characterised in that the default training of the lip reading identification module is specific
Including：
Training set include several characters, the random verification code be training set in generate at random；
Each character is based on HMM and is trained to obtain the unique forecast model sequence of correspondence in the training set
Row, the HMM includes N number of forecast model, N >=1；
The state-transition matrix and mixed Gauss model of HMM are calculated, the lip reading identification mould of default training is obtained
Type；
Wherein, the corresponding time series fragment of each character is produced by HMM；And in the time series fragment
Each moment corresponds to a hidden state, and the change of the hidden state is represented by state-transition matrix, and each hidden state is also right
An observational networks are answered, the observational networks are modeled as mixed Gauss model.
9. living body detection device according to claim 6, it is characterised in that the lip reading identification module includes：
Recognition unit, the characteristic vector for matching lip image sequence according to HMM, according to time series meter
The optimum state path of the characteristic vector of the lip image sequence is calculated, according to the single character of optimum state Path Recognition；
Assembled unit, for will corresponding to the characteristic vector recognize single character in temporal sequence combination producing differentiate object
Lip reading information.
10. living body detection device according to claim 6, it is characterised in that the detection module is specifically included：
Whether detection unit, the lip reading information for detection and identification object is consistent with character in random verification code；
First confirmation unit, for when the lip reading information is consistent with character in random verification code, discriminating object to be live body；
Second confirmation unit, for when character is inconsistent in the lip reading information and random verification code, differentiating object not to live
Body.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201611039845.7A CN106778496A (en) | 2016-11-22 | 2016-11-22 | Biopsy method and device |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201611039845.7A CN106778496A (en) | 2016-11-22 | 2016-11-22 | Biopsy method and device |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN106778496A true CN106778496A (en) | 2017-05-31 |
Family
ID=58973805
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201611039845.7A Pending CN106778496A (en) | 2016-11-22 | 2016-11-22 | Biopsy method and device |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN106778496A (en) |
Cited By (18)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107992833A (en) * | 2017-12-08 | 2018-05-04 | 北京小米移动软件有限公司 | Image-recognizing method, device and storage medium |
| CN107992812A (en) * | 2017-11-27 | 2018-05-04 | 北京搜狗科技发展有限公司 | A kind of lip reading recognition methods and device |
| CN108416595A (en) * | 2018-03-27 | 2018-08-17 | 百度在线网络技术（北京）有限公司 | Information processing method and device |
| CN108470169A (en) * | 2018-05-23 | 2018-08-31 | 国政通科技股份有限公司 | Face identification system and method |
| CN109271915A (en) * | 2018-09-07 | 2019-01-25 | 北京市商汤科技开发有限公司 | False-proof detection method and device, electronic equipment, storage medium |
| CN109409204A (en) * | 2018-09-07 | 2019-03-01 | 北京市商汤科技开发有限公司 | False-proof detection method and device, electronic equipment, storage medium |
| CN109461437A (en) * | 2018-11-28 | 2019-03-12 | 平安科技（深圳）有限公司 | The verifying content generating method and relevant apparatus of lip reading identification |
| CN110113319A (en) * | 2019-04-16 | 2019-08-09 | 深圳壹账通智能科技有限公司 | Identity identifying method, device, computer equipment and storage medium |
| WO2019223102A1 (en) * | 2018-05-22 | 2019-11-28 | 平安科技（深圳）有限公司 | Method and apparatus for checking validity of identity, terminal device and medium |
| CN110807356A (en) * | 2019-09-15 | 2020-02-18 | 成都恒道智融信息技术有限公司 | Living body detection method based on image lip language identification verification code |
| CN111339806A (en) * | 2018-12-19 | 2020-06-26 | 马上消费金融股份有限公司 | Training method of lip language recognition model, living body recognition method and device |
| CN111401134A (en) * | 2020-02-19 | 2020-07-10 | 北京三快在线科技有限公司 | Living body detection method, living body detection device, electronic apparatus, and storage medium |
| CN112257561A (en) * | 2020-10-20 | 2021-01-22 | 广州云从凯风科技有限公司 | Human face living body detection method and device, machine readable medium and equipment |
| CN112287722A (en) * | 2019-07-23 | 2021-01-29 | 北京中关村科金技术有限公司 | In-vivo detection method and device based on deep learning and storage medium |
| CN112287723A (en) * | 2019-07-23 | 2021-01-29 | 北京中关村科金技术有限公司 | In-vivo detection method and device based on deep learning and storage medium |
| CN112417925A (en) * | 2019-08-21 | 2021-02-26 | 北京中关村科金技术有限公司 | In-vivo detection method and device based on deep learning and storage medium |
| CN112560554A (en) * | 2019-09-25 | 2021-03-26 | 北京中关村科金技术有限公司 | Lip language-based living body detection method, device and storage medium |
| CN112633211A (en) * | 2020-12-30 | 2021-04-09 | 海信视像科技股份有限公司 | Service equipment and man-machine interaction method |
Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101046959A (en) * | 2007-04-26 | 2007-10-03 | 上海交通大学 | Identity identification method based on lid speech characteristic |
| CN104808794A (en) * | 2015-04-24 | 2015-07-29 | 北京旷视科技有限公司 | Method and system for inputting lip language |
| CN104834900A (en) * | 2015-04-15 | 2015-08-12 | 常州飞寻视讯信息科技有限公司 | Method and system for vivo detection in combination with acoustic image signal |
| CN104966086A (en) * | 2014-11-14 | 2015-10-07 | 深圳市腾讯计算机系统有限公司 | Living body identification method and apparatus |
| CN106096519A (en) * | 2016-06-01 | 2016-11-09 | 腾讯科技（深圳）有限公司 | Live body discrimination method and device |
-
2016
- 2016-11-22 CN CN201611039845.7A patent/CN106778496A/en active Pending
Patent Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101046959A (en) * | 2007-04-26 | 2007-10-03 | 上海交通大学 | Identity identification method based on lid speech characteristic |
| CN104966086A (en) * | 2014-11-14 | 2015-10-07 | 深圳市腾讯计算机系统有限公司 | Living body identification method and apparatus |
| CN104834900A (en) * | 2015-04-15 | 2015-08-12 | 常州飞寻视讯信息科技有限公司 | Method and system for vivo detection in combination with acoustic image signal |
| CN104808794A (en) * | 2015-04-24 | 2015-07-29 | 北京旷视科技有限公司 | Method and system for inputting lip language |
| CN106096519A (en) * | 2016-06-01 | 2016-11-09 | 腾讯科技（深圳）有限公司 | Live body discrimination method and device |
Non-Patent Citations (2)
| Title |
|---|
| YUQINGW520: "高安全性人脸识别系统中的唇语识别算法研究", 《HTTPS://WWW.DOC88.COM/P-2068920386740.HTML》 * |
| 任玉强 等: "高安全性人脸识别系统中的唇语识别算法研究", 《计算机应用研究》 * |
Cited By (21)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107992812A (en) * | 2017-11-27 | 2018-05-04 | 北京搜狗科技发展有限公司 | A kind of lip reading recognition methods and device |
| CN107992833A (en) * | 2017-12-08 | 2018-05-04 | 北京小米移动软件有限公司 | Image-recognizing method, device and storage medium |
| CN108416595A (en) * | 2018-03-27 | 2018-08-17 | 百度在线网络技术（北京）有限公司 | Information processing method and device |
| WO2019223102A1 (en) * | 2018-05-22 | 2019-11-28 | 平安科技（深圳）有限公司 | Method and apparatus for checking validity of identity, terminal device and medium |
| CN108470169A (en) * | 2018-05-23 | 2018-08-31 | 国政通科技股份有限公司 | Face identification system and method |
| CN109271915A (en) * | 2018-09-07 | 2019-01-25 | 北京市商汤科技开发有限公司 | False-proof detection method and device, electronic equipment, storage medium |
| CN109409204A (en) * | 2018-09-07 | 2019-03-01 | 北京市商汤科技开发有限公司 | False-proof detection method and device, electronic equipment, storage medium |
| CN109271915B (en) * | 2018-09-07 | 2021-10-08 | 北京市商汤科技开发有限公司 | Anti-counterfeiting detection method and device, electronic equipment and storage medium |
| CN109461437A (en) * | 2018-11-28 | 2019-03-12 | 平安科技（深圳）有限公司 | The verifying content generating method and relevant apparatus of lip reading identification |
| CN109461437B (en) * | 2018-11-28 | 2023-05-09 | 平安科技（深圳）有限公司 | Verification content generation method and related device for lip language identification |
| CN111339806A (en) * | 2018-12-19 | 2020-06-26 | 马上消费金融股份有限公司 | Training method of lip language recognition model, living body recognition method and device |
| CN111339806B (en) * | 2018-12-19 | 2021-04-13 | 马上消费金融股份有限公司 | Training method of lip language recognition model, living body recognition method and device |
| CN110113319A (en) * | 2019-04-16 | 2019-08-09 | 深圳壹账通智能科技有限公司 | Identity identifying method, device, computer equipment and storage medium |
| CN112287722A (en) * | 2019-07-23 | 2021-01-29 | 北京中关村科金技术有限公司 | In-vivo detection method and device based on deep learning and storage medium |
| CN112287723A (en) * | 2019-07-23 | 2021-01-29 | 北京中关村科金技术有限公司 | In-vivo detection method and device based on deep learning and storage medium |
| CN112417925A (en) * | 2019-08-21 | 2021-02-26 | 北京中关村科金技术有限公司 | In-vivo detection method and device based on deep learning and storage medium |
| CN110807356A (en) * | 2019-09-15 | 2020-02-18 | 成都恒道智融信息技术有限公司 | Living body detection method based on image lip language identification verification code |
| CN112560554A (en) * | 2019-09-25 | 2021-03-26 | 北京中关村科金技术有限公司 | Lip language-based living body detection method, device and storage medium |
| CN111401134A (en) * | 2020-02-19 | 2020-07-10 | 北京三快在线科技有限公司 | Living body detection method, living body detection device, electronic apparatus, and storage medium |
| CN112257561A (en) * | 2020-10-20 | 2021-01-22 | 广州云从凯风科技有限公司 | Human face living body detection method and device, machine readable medium and equipment |
| CN112633211A (en) * | 2020-12-30 | 2021-04-09 | 海信视像科技股份有限公司 | Service equipment and man-machine interaction method |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN106778496A (en) | Biopsy method and device | |
| CN110348319B (en) | Face anti-counterfeiting method based on face depth information and edge image fusion | |
| CN111563417B (en) | Pyramid structure convolutional neural network-based facial expression recognition method | |
| CN108182409A (en) | Biopsy method, device, equipment and storage medium | |
| CN104504362A (en) | Face detection method based on convolutional neural network | |
| CN106778506A (en) | A kind of expression recognition method for merging depth image and multi-channel feature | |
| CN107330444A (en) | A kind of image autotext mask method based on generation confrontation network | |
| CN106599883A (en) | Face recognition method capable of extracting multi-level image semantics based on CNN (convolutional neural network) | |
| CN106295506A (en) | A kind of age recognition methods based on integrated convolutional neural networks | |
| CN107180234A (en) | The credit risk forecast method extracted based on expression recognition and face characteristic | |
| CN106570491A (en) | Robot intelligent interaction method and intelligent robot | |
| CN109377429A (en) | A kind of recognition of face quality-oriented education wisdom evaluation system | |
| CN109117755A (en) | A kind of human face in-vivo detection method, system and equipment | |
| CN106960181A (en) | A kind of pedestrian's attribute recognition approach based on RGBD data | |
| CN110796101A (en) | Face recognition method and system of embedded platform | |
| Paul et al. | Extraction of facial feature points using cumulative histogram | |
| CN112966736B (en) | Vehicle re-identification method based on multi-view matching and local feature fusion | |
| CN112200176B (en) | Method and system for detecting quality of face image and computer equipment | |
| CN114662497A (en) | False news detection method based on cooperative neural network | |
| CN110110603A (en) | A kind of multi-modal labiomaney method based on facial physiologic information | |
| CN106599834A (en) | Information pushing method and system | |
| Shi et al. | Visual speaker authentication by ensemble learning over static and dynamic lip details | |
| CN113343198B (en) | Video-based random gesture authentication method and system | |
| CN105160285A (en) | Method and system for recognizing human body tumble automatically based on stereoscopic vision | |
| RU2005100267A (en) | METHOD AND SYSTEM OF AUTOMATIC VERIFICATION OF THE PRESENCE OF A LIVING FACE OF A HUMAN IN BIOMETRIC SECURITY SYSTEMS |
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 | ||
| CB02 | Change of applicant information | ||
| CB02 | Change of applicant information |
Address after: 401122 5 stories, Block 106, West Jinkai Avenue, Yubei District, Chongqing Applicant after: Chongqing Zhongke Yuncong Technology Co., Ltd. Address before: 401122, Chongqing North New District, Mount Huangshan Avenue, mercury science and technology building, B District, are six floor Applicant before: CHONGQING ZHONGKE YUNCONG TECHNOLOGY CO., LTD. |