CN110929239B - Terminal unlocking method based on lip language instruction - Google Patents
Terminal unlocking method based on lip language instruction Download PDFInfo
- Publication number
- CN110929239B CN110929239B CN201911045860.6A CN201911045860A CN110929239B CN 110929239 B CN110929239 B CN 110929239B CN 201911045860 A CN201911045860 A CN 201911045860A CN 110929239 B CN110929239 B CN 110929239B
- Authority
- CN
- China
- Prior art keywords
- representing
- lip
- image
- frame
- formula
- 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.)
- Active
Links
- 238000000034 method Methods 0.000 title claims abstract description 43
- 230000009471 action Effects 0.000 claims abstract description 19
- 238000012549 training Methods 0.000 claims abstract description 9
- 230000008569 process Effects 0.000 claims abstract description 8
- 238000012795 verification Methods 0.000 claims abstract description 6
- 239000013598 vector Substances 0.000 claims description 24
- 230000006870 function Effects 0.000 claims description 12
- 239000011159 matrix material Substances 0.000 claims description 8
- 210000002569 neuron Anatomy 0.000 claims description 7
- 230000008859 change Effects 0.000 claims description 6
- 238000012937 correction Methods 0.000 claims description 6
- 239000000284 extract Substances 0.000 claims description 6
- 230000001629 suppression Effects 0.000 claims description 6
- 230000009466 transformation Effects 0.000 claims description 6
- 238000012545 processing Methods 0.000 claims description 5
- 230000009467 reduction Effects 0.000 claims description 5
- 230000004913 activation Effects 0.000 claims description 3
- 238000013528 artificial neural network Methods 0.000 claims description 3
- 238000013527 convolutional neural network Methods 0.000 claims description 3
- 238000001514 detection method Methods 0.000 claims description 3
- 230000001815 facial effect Effects 0.000 claims description 3
- 238000010606 normalization Methods 0.000 claims description 3
- 238000011176 pooling Methods 0.000 claims description 3
- 230000009191 jumping Effects 0.000 claims description 2
- 230000000694 effects Effects 0.000 abstract description 5
- 238000009825 accumulation Methods 0.000 abstract description 3
- 238000013461 design Methods 0.000 abstract description 3
- 238000005286 illumination Methods 0.000 description 4
- 238000010586 diagram Methods 0.000 description 3
- 238000013135 deep learning Methods 0.000 description 2
- 230000007547 defect Effects 0.000 description 2
- 238000005516 engineering process Methods 0.000 description 2
- 230000001419 dependent effect Effects 0.000 description 1
- 230000010365 information processing Effects 0.000 description 1
- 230000003068 static effect Effects 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F21/00—Security arrangements for protecting computers, components thereof, programs or data against unauthorised activity
- G06F21/30—Authentication, i.e. establishing the identity or authorisation of security principals
- G06F21/31—User authentication
- G06F21/32—User authentication using biometric data, e.g. fingerprints, iris scans or voiceprints
-
- 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
-
- 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
- G06V40/168—Feature extraction; Face representation
- G06V40/171—Local features and components; Facial parts ; Occluding parts, e.g. glasses; Geometrical relationships
-
- 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
- G06V40/172—Classification, e.g. identification
Landscapes
- Engineering & Computer Science (AREA)
- Oral & Maxillofacial Surgery (AREA)
- Health & Medical Sciences (AREA)
- Theoretical Computer Science (AREA)
- General Health & Medical Sciences (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Human Computer Interaction (AREA)
- Multimedia (AREA)
- Computer Security & Cryptography (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Computer Hardware Design (AREA)
- Software Systems (AREA)
- General Engineering & Computer Science (AREA)
- Image Analysis (AREA)
- Image Processing (AREA)
Abstract
The invention relates to a terminal unlocking method based on a lip language instruction. In the verification process, key feature points needing face recognition are extracted in the same way, the Euclidean distance of face features is calculated by adopting facenet network, and comparison threshold judgment is carried out. The user can design the instruction action by himself during collection, and the same action can be made during identification, so that the action instruction is not easy to be stolen by others, and the authentication safety is improved. Meanwhile, the lip language instruction unlocking method does not need large-scale operation on the terminal, so that the hardware performance requirement is greatly reduced, and the recognition speed is increased. The invention can avoid the problem of overlarge gradient caused by accumulation of a certain quadrant in the space, improve the network learning and training efficiency, play the effect of actively learning and training the model and solve the problem that the traditional fixed instruction action is easy to expose.
Description
Technical Field
The invention relates to a terminal unlocking method based on a lip language instruction, and belongs to the technical field of image information processing.
Background
At present, the terminal unlocking mode mainly comprises: face, fingerprint, iris. However, the information is easy to forge, and the static identification method is easy to crack, so that the security is poor, and the leakage of private information is easy to cause. The invention adopts a lip language instruction unlocking method to realize dynamic unlocking and improve the safety of authentication.
The existing lip language unlocking technology is extremely dependent on deep learning, a specific single instruction model needs to be trained at a PC (personal computer) end and then deployed at a terminal for use, and a user needs to match a fixed instruction action. The method has poor effect, does not adapt to the data of the user, only can adapt to fixed command actions, and the commands are easy to be exposed.
Disclosure of Invention
The purpose of the invention is as follows: aiming at the defects of the existing unlocking technology, a terminal unlocking method based on a lip language instruction is provided.
The technical scheme is as follows: a terminal unlocking method based on a lip language instruction comprises the following steps:
step 2, extracting characteristic points of the lip video frame data set, matching the characteristic points of adjacent frames and marking position coordinates;
step 3, extracting the change characteristics of the positions of the characteristic points by using a frame difference method, namely the algebraic characteristics of the lip movement;
step 4, matching the human face in a database;
step 5, if matching is successful, people need to be identified to make the same lip language instruction action towards the terminal camera, the terminal extracts lip feature points similarly, and calculates algebraic features of lip movement, and whether matching is an unlocking instruction or not;
and 6, when the face is matched or the matching instruction is unsuccessful, prompting that the matching is failed, and jumping to the step 4.
In a further embodiment, the step 1 is further:
step 1-1, calculating a color histogram of an RGB space of each frame of a video segment, dividing each channel into 32 intervals according to pixel values, and carrying out normalization processing to obtain 96-dimensional features; forming a matrix by the characteristic vectors of each frame, performing dimensionality reduction on the matrix, and calculating an initialization clustering center:
in the formula, CnRepresenting the cluster center of the nth segment, fnFeature vector representing the nth frame, fn+1Represents the (n +1) th feature vector;
calculating the similarity of each new frame to the current cluster center, defining a threshold value sigma, and judging f when the similarity is greater than the threshold valuenBelonging to the cluster center CnAt this time, f isnAdding CnIn the method, a new clustering center C is obtained by updatingn′：
In the formula (f)nFeature vector representing the nth frame, CnRepresenting the cluster center of the nth segment, Cn′Representing and updating to obtain a new clustering center;
when the similarity is smaller than the threshold value, f is judgednMembership to a new cluster center, using fnInitializing a new cluster center Cn′：
Cn′＝fn
Step 1-2, firstly, recognizing the contour of a human face, removing a background, carrying out lip cutting on the human face in a video frame, positioning the position of facial feature contour points in the human face, including the coordinates of a nose tip, the leftmost coordinates of the lips, the rightmost coordinates of the lips and the coordinates of a central point of a mouth, cutting an image containing lip details according to the coordinates, and calculating the cutting size according to a formula:
in the formula, LMNDistance, x, between coordinates representing nose tip and coordinates of center point of mouthRight sideAbscissa, y, representing the rightmost feature point of the lipRight sideOrdinate, x, representing the rightmost feature point of the lipLeft side ofAbscissa, y, representing the leftmost feature point of the lipLeft side ofA vertical coordinate representing a feature point at the leftmost side of the lip;
step 1-3, performing deviation correction on the cut lip image, training the lip image based on a binary model of a convolutional neural network, and judging whether the extracted lip image is an effective image:
where l denotes the number of convolution layers, k denotes the convolution kernel, b denotes the convolution offset, MjRepresenting the local perceptual value of the input, beta the output parameter, and down () the pooling function.
In a further embodiment, the step 2 is further:
step 2-1, aiming at the cropped images extracted in the step 1, a D3D model is constructed to accelerate network convergence, and a loss function correction model is introduced:
in the formula,
denoted is the cross entropy loss, { yiK is an indicator function, local (pre) denotes the network output probability, σ is a scaling factor;
wherein, P({Z|X})＝∑k＝1P (pi | X), which is the sum of the probabilities formed by all paths after merging;
step 2-2, respectively extracting feature points from the images of two adjacent frames and obtaining two sets of feature point sets:
p＝{p1、p2、p3…pn}
p′＝{p1′、p2′、p3′…pn′}
and respectively calculating pixel interpolation values of the neighborhoods of the two groups of feature points by taking the pixel values of the windows W of the neighborhoods of the two groups of feature points as descriptors of the feature points according to the two adjacent groups of feature points as centers:
in the formula, S represents the pixel interpolation of two groups of characteristic point fields, x represents the abscissa of a pixel point, y represents the ordinate of the pixel point, W represents a field window, a descriptor is made in the formula, p represents a previous frame image, and p' represents a next frame image;
step 2-3, according to the pixel interpolation obtained in the step 2-2, finding a matching point according to a matching coefficient between the feature point and a neighborhood window:
in the formula, G represents the gray value of the previous frame image, G' represents the gray value of the next frame image, C represents the matching coefficient, and the other symbols have the same meanings as above.
In a further embodiment, the step 3 is further:
step 3-1, recording images of three adjacent independent frames, respectively recording the images as f (n +1), f (n) and f (n-1), and respectively recording the gray values corresponding to the three frames of images as G (n +1)x,y、G(n)x,y、G(n-1)x,yAnd obtaining an image P' by adopting a frame difference method:
P′＝|G(n+1)x,y-G(n)x,y|∩|G(n)x,y-G(n-1)x,y|
comparing the image P' with a preset threshold value T to analyze the liquidity, and extracting a moving target, wherein the comparison conditions are as follows:
in the formula, N represents the total number of pixels in the region to be detected, τ represents the suppression coefficient of illumination, a represents the image of the entire frame, and T is a threshold.
In a further embodiment, the step 4 is further:
step 4-1, on a multi-user terminal, such as a safe case and a door lock, face recognition is required to be carried out, and whether the face of the user exists in a matching database or not is matched; on a single user private terminal, such as a mobile phone and a tablet, face recognition is not needed, face verification can be performed, the facenet network is adopted to calculate the Euclidean distance of face features, and comparison threshold judgment is performed:
in the formula,
a pair of positive samples is represented, and,
a pair of negative samples is represented, and,
representing a flat sample pair, alpha representing the constraint range between the positive sample pair and the negative sample pair, phi representing the set of triples;
introducing a neuron model:
hW,b(x)＝f(WTx)
wherein W represents a weight vector of a neuron, WTx represents the non-linear transformation of an input vector x，f(WTx) represents the activation function transformation of the weight vector;
assigning an input vector x to xiInto WTx：
In the formula, n represents the number of stages of the neural network, and b represents an offset.
In a further embodiment, the step 5 is further: the method comprises the following steps of establishing a coordinate axis by taking the center of a lip as a coordinate origin in an acquisition process, fitting an inner lip region in a lip gray image into two semi-ellipse combinations, enabling an upper inner lip to correspond to an upper ellipse, enabling a lower inner lip to correspond to a lower ellipse, and extracting change characteristics of corresponding characteristic point positions by using a frame difference method, namely algebraic characteristics of interframe lip motion:
recording images of two adjacent independent frames, respectively recording the images as f (n +1) and f (n), and respectively recording the gray values corresponding to the two frames of images as G (n +1)x,y、G(n)x,yObtaining an image P' by adopting a frame difference method:
P′＝|G(n+1)x,y-G(n)x,y|
comparing the image P' with a preset threshold value T to analyze the liquidity, and extracting a moving target, wherein the comparison conditions are as follows:
in the formula, N represents the total number of pixels in the region to be detected, τ represents the suppression coefficient of illumination, a represents the image of the entire frame, and T is a threshold.
Has the advantages that: the invention relates to a terminal unlocking method based on a lip language instruction, wherein a user can design instruction actions by himself during collection and only needs to make the same actions during identification, so that the action instructions are not easy to steal by others, and the authentication safety is improved. Meanwhile, the lip language instruction unlocking method does not need large-scale operation on the terminal, so that the hardware performance requirement is greatly reduced, and the recognition speed is increased. According to the invention, through carrying out matrix dimensionality reduction processing, extracting feature points, initializing a clustering center and adopting facenet network to calculate the Euclidean distance of face features, the problem of overlarge gradient caused by accumulation of a certain quadrant in a space can be avoided, the network learning and training efficiency is improved, the effect of actively learning a training model is achieved, and the problem that the traditional fixed instruction action is easy to expose is solved.
Drawings
FIG. 1 is a flow chart of the present invention.
FIG. 2 is a schematic diagram of establishing a coordinate system for lips according to the present invention.
FIG. 3 is a diagram illustrating an image containing details of a lip cut out from a lip unlock command according to the present invention.
FIG. 4 is a schematic diagram of the introduction of a neuron model according to the present invention.
Detailed Description
In the following description, numerous specific details are set forth in order to provide a more thorough understanding of the present invention. It will be apparent, however, to one skilled in the art, that the present invention may be practiced without one or more of these specific details. In other instances, well-known features have not been described in order to avoid obscuring the invention.
The applicant believes that in the field of lip language unlocking, the prior art extremely depends on deep learning, a specific single instruction model needs to be trained at a PC (personal computer) end and then deployed at a terminal for use, and a user needs to match a fixed instruction action. The method has poor effect, does not adapt to the data of the user, can only adapt to fixed command actions, and the commands are easy to expose, so that how to construct the lip language model and continuously improve the active learning of the machine are very important.
In order to solve the problems in the prior art, the invention provides a terminal unlocking method based on a lip language instruction, a user can design instruction actions by himself during collection and only needs to make the same actions during identification, so that the action instructions are not easy to steal by others, and the authentication safety is improved.
The technical scheme of the invention is further explained by the embodiment and the corresponding attached drawings.
Firstly, a terminal camera collects a lip language instruction video frame unlocked by a user, the terminal carries out face detection and extracts face features, and meanwhile, a lip region video frame is extracted; calculating a color histogram of an RGB space of each frame of a video clip, dividing each channel into 32 intervals according to pixel values, and carrying out normalization processing to obtain 96-dimensional features; forming a matrix by the characteristic vectors of each frame, performing dimensionality reduction on the matrix, and calculating an initialization clustering center:
in the formula, CnRepresenting the cluster center of the nth segment, fnFeature vector representing the nth frame, fn+1Represents the (n +1) th feature vector;
calculating the similarity of each new frame to the current cluster center, defining a threshold value sigma, and judging f when the similarity is greater than the threshold valuenBelonging to the cluster center CnAt this time, f isnAdding CnIn the method, a new clustering center C is obtained by updatingn′：
In the formula (f)nFeature vector representing the nth frame, CnRepresenting the cluster center of the nth segment, Cn′Representing and updating to obtain a new clustering center;
when the similarity is smaller than the threshold value, f is judgednMembership to a new cluster center, using fnInitializing a new cluster center Cn′：
Cn′＝fn
Recognizing the outline of the face, removing the background, cutting the lips of the face in a video frame, positioning the positions of facial feature outline points in the face, including the coordinates of the nose tip, the leftmost coordinates of the lips, the rightmost coordinates of the lips and the coordinates of the center point of the mouth, cutting an image containing the details of the lips according to the coordinates, and calculating the cutting size according to a formula:
in the formula, LMNDistance, x, between coordinates representing nose tip and coordinates of center point of mouthRight sideAbscissa, y, representing the rightmost feature point of the lipRight sideOrdinate, x, representing the rightmost feature point of the lipLeft side ofAbscissa, y, representing the leftmost feature point of the lipLeft side ofA vertical coordinate representing a feature point at the leftmost side of the lip;
carrying out deviation correction on the cut lip images, training the lip images based on a binary model of a convolutional neural network, and judging whether the extracted lip images are effective images:
where l denotes the number of convolution layers, k denotes the convolution kernel, b denotes the convolution offset, MjRepresenting the local perceptual value of the input, beta the output parameter, and down () the pooling function.
Then, extracting characteristic points of the lip video frame data set, matching the characteristic points of adjacent frames, and marking position coordinates;
for the extracted cropped images, a D3D model is constructed to accelerate network convergence, and a loss function correction model is introduced:
in the formula,
denoted is the cross entropy loss, { yiK is an indicator function, and location (pre) representsThe network output probability, sigma is a proportionality coefficient;
where P ({ Z | X }) ═ Σk＝1P (pi | X), which is the sum of the probabilities formed by all paths after merging;
respectively extracting feature points from the images of two adjacent frames and obtaining two groups of feature point sets:
p＝{p1、p2、p3 … pn}
p′＝{p1′、p2′、p3′ … pn′}
and respectively calculating pixel interpolation values of the neighborhoods of the two groups of feature points by taking the pixel values of the windows W of the neighborhoods of the two groups of feature points as descriptors of the feature points according to the two adjacent groups of feature points as centers:
in the formula, S represents the pixel interpolation of two groups of characteristic point fields, x represents the abscissa of a pixel point, y represents the ordinate of the pixel point, W represents a field window, a descriptor is made in the formula, p represents a previous frame image, and p' represents a next frame image;
according to the pixel interpolation obtained above, finding a matching point according to the matching coefficient between the feature point and the neighborhood window:
in the formula, G represents the gray value of the previous frame image, G' represents the gray value of the next frame image, C represents the matching coefficient, and the other symbols have the same meanings as above.
Then, extracting the change characteristics of the positions of the characteristic points by using a frame difference method, namely the algebraic characteristics of the lip movement; recording images of three adjacent independent frames, respectively recording the images as f (n +1), f (n-1), and respectively recording the gray values corresponding to the three frames as G (n +1)x,y、G(n)x,y、G(n-1)x,yAnd obtaining an image P' by adopting a frame difference method:
P′＝|G(n+1)x,y-G(n)x,y|∩|G(n)x,y-G(n-1)x,y|
comparing the image P' with a preset threshold value T to analyze the liquidity, and extracting a moving target, wherein the comparison conditions are as follows:
in the formula, N represents the total number of pixels in the region to be detected, τ represents the suppression coefficient of illumination, a represents the image of the entire frame, and T is a threshold.
Step 4, matching the human face in a database: on a multi-user terminal, such as a safe case and a door lock, face recognition is required to be carried out, and whether the face of the user exists in a database or not is matched; on a single user private terminal, such as a mobile phone and a tablet, face recognition is not needed, face verification can be performed, the facenet network is adopted to calculate the Euclidean distance of face features, and comparison threshold judgment is performed:
in the formula,
a pair of positive samples is represented, and,
a pair of negative samples is represented, and,
representing a flat sample pair, alpha representing the constraint range between the positive sample pair and the negative sample pair, phi representing the set of triples;
introducing a neuron model:
hW,b(x)＝f(WTx)
wherein W represents a weight vector of a neuron, WTx denotes the nonlinear transformation of the input vector x, f (W)Tx) represents the activation function transformation of the weight vector;
assigning an input vector x to xiInto WTx：
In the formula, n represents the number of stages of the neural network, and b represents an offset.
Step 5, if matching is successful, people need to be identified to make the same lip language instruction action towards the terminal camera, the terminal extracts lip feature points similarly, and calculates algebraic features of lip movement, and whether matching is an unlocking instruction or not; the method comprises the following steps of establishing a coordinate axis by taking the center of a lip as a coordinate origin in an acquisition process, fitting an inner lip region in a lip gray image into two semi-ellipse combinations, enabling an upper inner lip to correspond to an upper ellipse, enabling a lower inner lip to correspond to a lower ellipse, and extracting change characteristics of corresponding characteristic point positions by using a frame difference method, namely algebraic characteristics of interframe lip motion:
recording images of two adjacent independent frames, respectively recording the images as f (n +1) and f (n), and respectively recording the gray values corresponding to the two frames of images as G (n +1)x,y、G(n)x,yObtaining an image P' by adopting a frame difference method:
P′＝|G(n+1)x,y-G(n)x,y|
comparing the image P' with a preset threshold value T to analyze the liquidity, and extracting a moving target, wherein the comparison conditions are as follows:
in the formula, N represents the total number of pixels in the region to be detected, τ represents the suppression coefficient of illumination, a represents the image of the entire frame, and T is a threshold.
And when the face matching or the matching instruction is unsuccessful, prompting that the matching is failed, continuing matching the face in the database, repeating the steps, and temporarily locking the terminal equipment when the matching is failed for more than three times.
In summary, aiming at the defects of the prior art, the invention provides a terminal unlocking method based on a lip language instruction, which is used for acquiring a face by taking several frames of images in the acquisition process and extracting part of key feature points. In the verification process, key feature points needing face recognition are extracted in the same way, the Euclidean distance of face features is calculated by adopting facenet network, and comparison threshold judgment is carried out. In the acquisition process, a coordinate axis is established by taking the center of the lip as a coordinate origin, an inner lip region in the lip gray image is fitted into two semiellipse combinations (an upper inner lip corresponds to an upper ellipse, and a lower inner lip corresponds to a lower ellipse), the variation characteristic of the position of the corresponding characteristic point, namely the algebraic characteristic of the interframe lip motion, is extracted by using a frame difference method, and a judgment threshold value is calculated. In the verification process, lip motion characteristics are extracted in the same way and are compared and judged. By carrying out matrix dimensionality reduction processing, extracting feature points, initializing a clustering center and adopting the facenet network to calculate the Euclidean distance of the face features, the problem of overlarge gradient caused by accumulation of a certain quadrant in a space can be avoided, the network learning and training efficiency is improved, the effect of actively learning a training model is achieved, and the problem that the traditional fixed instruction action is easy to expose is solved.
As noted above, while the present invention has been shown and described with reference to certain preferred embodiments, it is not to be construed as limited thereto. Various changes in form and detail may be made therein without departing from the spirit and scope of the invention as defined by the appended claims.
Claims (1)
1. A terminal unlocking method based on a lip language instruction is characterized by comprising the following steps:
step 1, a terminal camera collects a lip language instruction video frame of unlocking by a user, the terminal carries out face detection and extracts face features, and meanwhile, a lip region video frame is extracted;
step 1-1, calculating a color histogram of an RGB space of each frame of a video segment, dividing each channel into 32 intervals according to pixel values, and carrying out normalization processing to obtain 96-dimensional features; forming a matrix by the characteristic vectors of each frame, performing dimensionality reduction on the matrix, and calculating an initialization clustering center:
in the formula,
the cluster center of the nth segment is represented,
a feature vector representing the n-th frame,
represents the (n +1) th feature vector;
calculating the similarity of each new frame to the current cluster center, and defining a threshold value
When the similarity is larger than the threshold value, judging
Belonging to the cluster center
At this time, the
Adding into
In the method, a new clustering center is obtained by updating
：
In the formula,
a feature vector representing the n-th frame,
the cluster center of the nth segment is represented,
representing and updating to obtain a new clustering center;
when the similarity is smaller than the threshold value, judging
Belonging to a new cluster centre, in this case
Initializing new cluster centers
：
Step 1-2, firstly, recognizing the contour of a human face, removing a background, carrying out lip cutting on the human face in a video frame, positioning the position of facial feature contour points in the human face, including the coordinates of a nose tip, the leftmost coordinates of the lips, the rightmost coordinates of the lips and the coordinates of a central point of a mouth, cutting an image containing lip details according to the coordinates, and calculating the cutting size according to a formula:
in the formula,
representing the distance between the coordinates of the tip of the nose and the coordinates of the centre point of the mouth,
the abscissa representing the rightmost feature point of the lip,
the ordinate representing the rightmost feature point of the lip,
the abscissa representing the leftmost feature point of the lip,
a vertical coordinate representing a feature point at the leftmost side of the lip;
step 1-3, performing deviation correction on the cut lip image, training the lip image based on a binary model of a convolutional neural network, and judging whether the extracted lip image is an effective image:
in the formula,lthe number of layers of convolution is represented,kwhich represents the kernel of the convolution,bwhich represents the offset of the convolution,
a local perceptual value representing the input is represented,
the output parameters are represented by a number of output parameters,
representing a pooling function;
step 2, extracting characteristic points of the lip video frame data set, matching the characteristic points of adjacent frames and marking position coordinates;
step 2-1, aiming at the cropped images extracted in the step 1, a D3D model is constructed to accelerate network convergence, and a loss function correction model is introduced:
in the formula,
it is indicated that the cross-entropy loss is,
in order to indicate the function,
the probability of the network output is represented,
is a proportionality coefficient;
wherein,
i.e. the sum of the probabilities formed after all paths are merged;
step 2-2, respectively extracting feature points from the images of two adjacent frames and obtaining two sets of feature point sets:
and respectively calculating pixel interpolation values of the neighborhoods of the two groups of feature points by taking the pixel values of the windows W of the neighborhoods of the two groups of feature points as descriptors of the feature points according to the two adjacent groups of feature points as centers:
in the formula,Spixel interpolation representing the domain of two sets of feature points,xthe abscissa representing the pixel point,yThe ordinate of the pixel point is represented,Wrepresenting a domain window, in this formula a descriptor, p represents the previous frame image,
representing the next frame of image;
step 2-3, according to the pixel interpolation obtained in the step 2-2, finding a matching point according to a matching coefficient between the feature point and a neighborhood window:
in the formula,
representing the gray value of the image of the previous frame,
representing the gray-scale value of the image of the next frame,Crepresenting the matching coefficient, and the other symbols have the same meanings as above;
step 3, extracting the change characteristics of the positions of the characteristic points by using a frame difference method, namely the algebraic characteristics of the lip movement;
step 3-1, recording images of three adjacent independent frames, and respectively recording the images
、
、
The gray values corresponding to the three frames of images are respectively recorded as
、
、
Obtaining images by frame difference method
：
The image is processed
With a predetermined threshold valueTComparing and analyzing the circulation, and extracting a moving target under the comparison conditions:
in the formula,
representing the total number of pixels in the region to be detected,
the suppression coefficient of the light irradiation is represented,
an image representing a complete frame of the image,Tis a threshold value;
step 4, matching the human face in a database;
step 4-1, on a multi-user terminal, such as a safe case and a door lock, face recognition is required to be carried out, and whether the face of the user exists in a matching database or not is matched; on a single user private terminal, such as a mobile phone and a tablet, face recognition is not needed, face verification can be performed, the facenet network is adopted to calculate the Euclidean distance of face features, and comparison threshold judgment is performed:
in the formula,
a pair of positive samples is represented, and,
a pair of negative samples is represented, and,
a pair of flat samples is represented, and,
representing the range of constraints between pairs of positive and negative samples,
representing a set of triples;
introducing a neuron model:
in the formula,
a weight vector representing the neuron and,
representing input vectorxThe non-linear transformation is carried out,
representing an activation function transformation of the weight vector;
inputting vectorxIs assigned a value of
In a belt
：
In the formula,nthe number of stages of the neural network is represented,brepresents an offset;
step 5, if matching is successful, people need to be identified to make the same lip language instruction action towards the terminal camera, the terminal extracts lip feature points similarly, and calculates algebraic features of lip movement, and whether matching is an unlocking instruction or not;
the method comprises the following steps of establishing a coordinate axis by taking the center of a lip as a coordinate origin in an acquisition process, fitting an inner lip region in a lip gray image into two semi-ellipse combinations, enabling an upper inner lip to correspond to an upper ellipse, enabling a lower inner lip to correspond to a lower ellipse, and extracting change characteristics of corresponding characteristic point positions by using a frame difference method, namely algebraic characteristics of interframe lip motion:
recording images of two adjacent separate frames, respectively
、
The gray values corresponding to the two frames of images are respectively recorded as
、
Obtaining an image by adopting a frame difference method
：
The image is processed
With a predetermined threshold valueTComparing and analyzing the circulation, and extracting a moving target under the comparison conditions:
in the formula,
representing the total number of pixels in the region to be detected,
the suppression coefficient of the light irradiation is represented,
an image representing a complete frame of the image,Tis a threshold value, and is,
indicating the corresponding gray value of the nth frame image,
representing the corresponding gray value of the n +1 th frame image;
and 6, when the face is matched or the matching instruction is unsuccessful, prompting that the matching is failed, and jumping to the step 4.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201911045860.6A CN110929239B (en) | 2019-10-30 | 2019-10-30 | Terminal unlocking method based on lip language instruction |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201911045860.6A CN110929239B (en) | 2019-10-30 | 2019-10-30 | Terminal unlocking method based on lip language instruction |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN110929239A CN110929239A (en) | 2020-03-27 |
| CN110929239B true CN110929239B (en) | 2021-11-19 |
Family
ID=69849882
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201911045860.6A Active CN110929239B (en) | 2019-10-30 | 2019-10-30 | Terminal unlocking method based on lip language instruction |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN110929239B (en) |
Families Citing this family (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN112733807A (en) * | 2021-02-22 | 2021-04-30 | 佳都新太科技股份有限公司 | Face comparison graph convolution neural network training method and device |
| CN114220142B (en) * | 2021-11-24 | 2022-08-23 | 慧之安信息技术股份有限公司 | Face feature recognition method of deep learning algorithm |
| CN114220177B (en) * | 2021-12-24 | 2024-06-25 | 湖南大学 | Lip syllable recognition method, device, equipment and medium |
Citations (9)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO2016061780A1 (en) * | 2014-10-23 | 2016-04-28 | Intel Corporation | Method and system of facial expression recognition using linear relationships within landmark subsets |
| WO2016201679A1 (en) * | 2015-06-18 | 2016-12-22 | 华为技术有限公司 | Feature extraction method, lip-reading classification method, device and apparatus |
| CN106570461A (en) * | 2016-10-21 | 2017-04-19 | 哈尔滨工业大学深圳研究生院 | Video frame image extraction method and system based on lip movement identification |
| KR101767234B1 (en) * | 2016-03-21 | 2017-08-10 | 양장은 | System based on pattern recognition of blood vessel in lips |
| CN107358085A (en) * | 2017-07-28 | 2017-11-17 | 惠州Tcl移动通信有限公司 | A kind of unlocking terminal equipment method, storage medium and terminal device |
| CN108960103A (en) * | 2018-06-25 | 2018-12-07 | 西安交通大学 | The identity identifying method and system that a kind of face and lip reading blend |
| CN109409195A (en) * | 2018-08-30 | 2019-03-01 | 华侨大学 | A kind of lip reading recognition methods neural network based and system |
| CN110276230A (en) * | 2018-03-14 | 2019-09-24 | 阿里巴巴集团控股有限公司 | The method, apparatus and electronic equipment that user authentication, lip reading identify |
| CN110276259A (en) * | 2019-05-21 | 2019-09-24 | 平安科技（深圳）有限公司 | Lip reading recognition methods, device, computer equipment and storage medium |
Family Cites Families (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US9159321B2 (en) * | 2012-02-27 | 2015-10-13 | Hong Kong Baptist University | Lip-password based speaker verification system |
| US20150279364A1 (en) * | 2014-03-29 | 2015-10-01 | Ajay Krishnan | Mouth-Phoneme Model for Computerized Lip Reading |
-
2019
- 2019-10-30 CN CN201911045860.6A patent/CN110929239B/en active Active
Patent Citations (9)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO2016061780A1 (en) * | 2014-10-23 | 2016-04-28 | Intel Corporation | Method and system of facial expression recognition using linear relationships within landmark subsets |
| WO2016201679A1 (en) * | 2015-06-18 | 2016-12-22 | 华为技术有限公司 | Feature extraction method, lip-reading classification method, device and apparatus |
| KR101767234B1 (en) * | 2016-03-21 | 2017-08-10 | 양장은 | System based on pattern recognition of blood vessel in lips |
| CN106570461A (en) * | 2016-10-21 | 2017-04-19 | 哈尔滨工业大学深圳研究生院 | Video frame image extraction method and system based on lip movement identification |
| CN107358085A (en) * | 2017-07-28 | 2017-11-17 | 惠州Tcl移动通信有限公司 | A kind of unlocking terminal equipment method, storage medium and terminal device |
| CN110276230A (en) * | 2018-03-14 | 2019-09-24 | 阿里巴巴集团控股有限公司 | The method, apparatus and electronic equipment that user authentication, lip reading identify |
| CN108960103A (en) * | 2018-06-25 | 2018-12-07 | 西安交通大学 | The identity identifying method and system that a kind of face and lip reading blend |
| CN109409195A (en) * | 2018-08-30 | 2019-03-01 | 华侨大学 | A kind of lip reading recognition methods neural network based and system |
| CN110276259A (en) * | 2019-05-21 | 2019-09-24 | 平安科技（深圳）有限公司 | Lip reading recognition methods, device, computer equipment and storage medium |
Non-Patent Citations (6)
| Title |
|---|
| A Large-Scale Hierarchical Multi-View RGB-D Object Dataset;Kevin Lai;《Proceedings - IEEE International Conference on Robotics and Automation·May 2011》;20140528;文章全文 * |
| FaceNet: A Unified Embedding for Face Recognition and Clustering;Florian Schroff;《2015 IEEE Conference on Computer Vision and Pattern Recognition》;20151015;文章全文 * |
| 基于帧差法和边缘检测法的视频分割算法;陈春雨;《济南大学学报》;20120316;第26卷(第1期);第31-36页 * |
| 声纹识别（说话人识别）;weixin_30596343;《https://blog.csdn.net/weixin_30596343/details/99112089》;20180726;文章全文 * |
| 帧差法在运动目标实时跟踪中的应用;邱道尹;《华北水利水电学院学报》;20100111;第30卷(第3期);第45-46、64页 * |
| 视频镜头分割和关键帧提取关键技术研究;郝会芬;《中国优秀硕士学位论文全文数据库 信息科技辑》;20160215;第2-4章 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN110929239A (en) | 2020-03-27 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN111401257B (en) | Face recognition method based on cosine loss under non-constraint condition | |
| CN108009520B (en) | Finger vein identification method and system based on convolution variational self-encoder network | |
| CN110929239B (en) | Terminal unlocking method based on lip language instruction | |
| Shen et al. | Finger vein recognition algorithm based on lightweight deep convolutional neural network | |
| CN109949278B (en) | Hyperspectral anomaly detection method based on antagonistic self-coding network | |
| CN111160533B (en) | Neural network acceleration method based on cross-resolution knowledge distillation | |
| CN111368683B (en) | Face image feature extraction method and face recognition method based on modular constraint CenterFace | |
| Tian et al. | Ear recognition based on deep convolutional network | |
| CN108268859A (en) | A kind of facial expression recognizing method based on deep learning | |
| CN107103281A (en) | Face identification method based on aggregation Damage degree metric learning | |
| Liu et al. | Gait recognition based on outermost contour | |
| CN112036383B (en) | Hand vein-based identity recognition method and device | |
| CN106355138A (en) | Face recognition method based on deep learning and key features extraction | |
| CN107239741B (en) | Single-sample face recognition method based on sparse reconstruction | |
| CN110633655A (en) | Attention-attack face recognition attack algorithm | |
| CN111666845A (en) | Small sample deep learning multi-mode sign language recognition method based on key frame sampling | |
| CN113591747A (en) | Multi-scene iris recognition method based on deep learning | |
| CN109325472B (en) | Face living body detection method based on depth information | |
| CN108875645A (en) | A kind of face identification method under the conditions of underground coal mine complex illumination | |
| CN103745242A (en) | Cross-equipment biometric feature recognition method | |
| Huang et al. | Human emotion recognition based on face and facial expression detection using deep belief network under complicated backgrounds | |
| Chen et al. | Robust gender recognition for uncontrolled environment of real-life images | |
| Fang et al. | Deep belief network based finger vein recognition using histograms of uniform local binary patterns of curvature gray images | |
| CN114998966B (en) | Facial expression recognition method based on feature fusion | |
| CN112800959B (en) | Difficult sample mining method for data fitting estimation in face recognition |
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 |
Address after: 211000 floor 3, building 3, Qilin artificial intelligence Industrial Park, 266 Chuangyan Road, Nanjing, Jiangsu Applicant after: Zhongke Nanjing artificial intelligence Innovation Research Institute Applicant after: INSTITUTE OF AUTOMATION, CHINESE ACADEMY OF SCIENCES Address before: 211000 3rd floor, building 3, 266 Chuangyan Road, Jiangning District, Nanjing City, Jiangsu Province Applicant before: NANJING ARTIFICIAL INTELLIGENCE CHIP INNOVATION INSTITUTE, INSTITUTE OF AUTOMATION, CHINESE ACADEMY OF SCIENCES Applicant before: INSTITUTE OF AUTOMATION, CHINESE ACADEMY OF SCIENCES |
|
| CB02 | Change of applicant information | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |