CN114882545A - Multi-angle face recognition method based on three-dimensional intelligent reconstruction - Google Patents

Multi-angle face recognition method based on three-dimensional intelligent reconstruction Download PDF

Info

Publication number
CN114882545A
CN114882545A CN202210336825.5A CN202210336825A CN114882545A CN 114882545 A CN114882545 A CN 114882545A CN 202210336825 A CN202210336825 A CN 202210336825A CN 114882545 A CN114882545 A CN 114882545A
Authority
CN
China
Prior art keywords
face
representing
model
dimensional
characteristic quantity
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
Application number
CN202210336825.5A
Other languages
Chinese (zh)
Inventor
张晖
赵上辉
赵海涛
朱洪波
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Nanjing University of Posts and Telecommunications
Original Assignee
Nanjing University of Posts and Telecommunications
Priority date (The priority date 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 date listed.)
Filing date
Publication date
Application filed by Nanjing University of Posts and Telecommunications filed Critical Nanjing University of Posts and Telecommunications
Priority to CN202210336825.5A priority Critical patent/CN114882545A/en
Publication of CN114882545A publication Critical patent/CN114882545A/en
Pending legal-status Critical Current

Links

Images

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
    • G06V40/10—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
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T17/00—Three dimensional [3D] modelling, e.g. data description of 3D objects
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T19/00—Manipulating 3D models or images for computer graphics
    • G06T19/20—Editing of 3D images, e.g. changing shapes or colours, aligning objects or positioning parts
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V10/00—Arrangements for image or video recognition or understanding
    • G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
    • G06V10/74—Image or video pattern matching; Proximity measures in feature spaces
    • G06V10/761—Proximity, similarity or dissimilarity measures
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V10/00—Arrangements for image or video recognition or understanding
    • G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
    • G06V10/82—Arrangements for image or video recognition or understanding using pattern recognition or machine learning using neural networks
    • 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
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2219/00—Indexing scheme for manipulating 3D models or images for computer graphics
    • G06T2219/20—Indexing scheme for editing of 3D models
    • G06T2219/2012—Colour editing, changing, or manipulating; Use of colour codes
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2219/00—Indexing scheme for manipulating 3D models or images for computer graphics
    • G06T2219/20—Indexing scheme for editing of 3D models
    • G06T2219/2016—Rotation, translation, scaling

Abstract

The invention discloses a multi-angle face recognition method based on three-dimensional intelligent reconstruction. Inputting a face picture, extracting face characteristic points in the picture, calculating a face deflection angle and reconstructing the face characteristic points by combining an ASM algorithm; predicting a shape vector and a texture vector in the deformation model by using a cascade residual error network, establishing a three-dimensional face gray model according to the predicted shape vector, and matching with the feature points of the face; then coloring according to the texture vector to construct a complete three-dimensional face model; and rotating the three-dimensional face model according to the face deflection angle, and comparing the three-dimensional face model with the similarity of the face of the input picture to obtain a final three-dimensional face model so as to finish face identification. The method realizes multi-angle face recognition based on three-dimensional reconstruction, accelerates the operation process of the deformation model by using the cascade residual error network, forms a self-feedback training punishment of the self-feedback network on the deformation model, and has very wide application scenes.

Description

Multi-angle face recognition method based on three-dimensional intelligent reconstruction
Technical Field
The invention relates to the field of computer vision and the field of deep learning, in particular to a multi-angle face recognition method based on three-dimensional intelligent reconstruction.
Background
With the rapid development of the internet technology, the data processing capability of the algorithm is greatly improved, and with the rapid development of the big data technology, hidden dangers such as information safety and other related problems can be generated, so that information identification, detection and other related technical means are very important. The advent of face recognition technology has addressed these problems. Compared with the traditional means of using magnetic cards or identity cards and the like with magnetic passwords, the face recognition is used as an inherent characteristic of a human body, is fixed and unchangeable for a single person and can not be repeated for different persons, the face is a biological characteristic, is determined by the unique and unchangeable gene of each person, and the identity recognition through the face is simple, convenient and quick, so the rapid development of the face recognition is a necessary trend for the social progress.
With the rapid development of deep learning, the face recognition field has achieved extremely high achievements, but is still not perfect. Compared with front static face recognition, the development of multi-angle face recognition technology in social scenes has met with a lot of bottlenecks. One of the main reasons that multi-angle face recognition is not as developed as front face recognition is the imperfection of the data set. Various face data sets including google are all front faces, but the faces are three-dimensional features, and the deflection of face angles also serves as a main research field. Dynamic face recognition is crucial to realize the full potential of face recognition in life, because multi-angle face recognition is essentially a passive recognition technology that recognizes an uncooperative object. In recent years, the human face three-dimensional reconstruction technology is further developed, various three-dimensional reconstruction technologies are widely applied, and the traditional three-dimensional reconstruction based on the picture depth technology has high requirements on the illumination effect and is not suitable for real scenes. The three-dimensional reconstruction based on the multi-view stereo vision has good reconstruction effect, but is not an automatic process and needs manual control in a background. Therefore, the current three-dimensional face reconstruction using deformation models is still one of the main research directions.
Therefore, the multi-angle face recognition technology in the prior art has the problem of incomplete data collection; the traditional three-dimensional reconstruction based on the picture depth technology is not suitable for a real scene due to high illumination requirements; the traditional three-dimensional reconstruction based on multi-view stereo vision needs a background to consider the operation and control, and the realization is too troublesome.
Disclosure of Invention
Aiming at the problems, the invention provides a multi-angle face recognition method based on three-dimensional intelligent reconstruction.
In order to realize the aim of the invention, the invention provides a multi-angle face recognition method based on three-dimensional intelligent reconstruction, which comprises the following steps:
s 1: inputting a face picture, extracting face characteristic points of the face picture, calculating face characteristic quantity of each region of the face, and calculating face deflection angles in scenes according to the number of the extracted face characteristic points;
s 2: reconstructing the extracted face characteristic points into front face characteristic points by adopting an ASM algorithm based on the face characteristic quantity of each region of the face and the face deflection angle;
s 3: constructing an internal face database and a deformation model, predicting a shape vector and a texture vector of the deformation model based on the internal face database by using a sample-trained cascade residual error network, wherein the internal face database stores a three-dimensional face picture training data set for prediction;
s 4: constructing a three-dimensional face gray model according to the predicted shape vector, and then performing matching operation on the reconstructed front face feature points and the three-dimensional face gray model; if the matching is successful, coloring and detail supplementing operations are carried out on the three-dimensional face gray model based on the texture vector, and then a three-dimensional face model is established; if the matching fails, returning to the step s3 to perform self-feedback penalty training on the deformation model;
s 5: after the three-dimensional face model is rotated according to the face deflection angle, matching the input face picture with the rotated three-dimensional face model through face similarity, obtaining face similarity SIM, and performing scene operation according to the face similarity SIM to obtain a final three-dimensional face model;
s 6: and comparing the final three-dimensional face model with an external face database to finish final face recognition.
Further, in step s1, the total number of the face feature points is 68, and the face feature quantity of each region of the face includes: human eye characteristic quantity, nose bridge characteristic quantity, mouth characteristic quantity and mandible characteristic quantity;
the calculation formula of the human eye characteristic quantity is as follows:
Figure BDA0003574660920000021
wherein D is eye A characteristic quantity of a human eye is represented,
Figure BDA0003574660920000022
and
Figure BDA0003574660920000023
respectively represents the sum of the connecting line distances of the two characteristic points of the upper eyelid and the two corresponding characteristic points of the lower eyelid,
Figure BDA0003574660920000024
representing the distance of a transverse connecting line of the characteristic points of the left and right canthus;
the calculation formula of the nose bridge characteristic quantity is as follows:
Figure BDA0003574660920000025
wherein D is nose The nose bridge characteristic quantity is represented,
Figure BDA0003574660920000026
the vertical distance between the highest feature point and the lowest feature point of the bridge of the middle nose is represented,
Figure BDA0003574660920000027
representing a lateral link distance of a lowest feature point of the bridge of the middle nose and a rightmost feature point of the nose;
the calculation formula of the mouth characteristic quantity is as follows:
Figure BDA0003574660920000031
wherein D is mouth The characteristic quantity of the mouth is represented,
Figure BDA0003574660920000032
respectively represents the vertical distance of the connecting lines of 3 groups of upper and lower characteristic points of the upper lip,
Figure BDA0003574660920000033
respectively represents the vertical distance of the connecting lines of 3 groups of upper and lower characteristic points of the lower lip,
Figure BDA0003574660920000034
and
Figure BDA0003574660920000035
respectively showing the transverse connection distance between the left and right corners of the upper and lower lips;
the calculation formula of the mandible characteristic quantity is as follows:
Figure BDA0003574660920000036
wherein D is mandible The lower jaw characteristic quantity is expressed,
Figure BDA0003574660920000037
representing the vertical distance between the lowest mandibular feature point and the highest mandibular feature point,
Figure BDA0003574660920000038
the lateral distance between the lowest mandibular feature point and the highest mandibular feature point is represented.
Further, in step s1, the sub-scene calculation formula of the face deflection angle is as follows:
Figure BDA0003574660920000039
wherein the content of the first and second substances,
Figure BDA00035746609200000310
representing the deflection angle of the human face, omega representing the number of extracted characteristic points of the human face,
Figure BDA00035746609200000311
representing the distance between the eyes of the face after rotation,
Figure BDA00035746609200000312
indicating the distance, R, from the mouth to the line connecting the centers of the two eyes after deflection 31 、R 32 、R 33 Respectively representing variables in a rotation matrix, a being a constant, theta 1 Representing the angle of deflection, theta 2 Representing a horizontal deflection angle;
if the omega is more than or equal to 65 and less than or equal to 68, calculating the face deflection angle by using a first formula; if omega is more than or equal to 48 and less than or equal to 64, calculating the face deflection angle by using a second formula; if omega is less than 48, a third formula is used for calculating the face deflection angle.
Further, in the step s2, the reconstruction of the facial feature points is expressed by the following model:
Figure BDA00035746609200000313
wherein the content of the first and second substances,
Figure BDA00035746609200000314
representing an average shape vector, wherein pi represents a new human face sample characteristic point formed by the principal component shape vectors; b * A vector representing the weight of a control feature point,
Figure BDA0003574660920000041
representing and extracting principal component shape vectorThe covariance matrix of (a) is determined,
Figure BDA0003574660920000042
represents the average amount of human eye features,
Figure BDA0003574660920000043
the average nose bridge characteristic quantity is expressed,
Figure BDA0003574660920000044
the average mouth characteristic quantity is represented,
Figure BDA0003574660920000045
the average mandibular feature quantity is expressed.
Further, in step s3, the structure of the sample-trained cascade residual error network includes the following:
the three-dimensional face is uniformly represented in a vector form:
Figure BDA0003574660920000046
wherein the content of the first and second substances,
Figure BDA0003574660920000047
and
Figure BDA0003574660920000048
respectively representing the shape vector and the texture vector of the ith face in the internal face database,
Figure BDA0003574660920000049
representing a face gray model consisting of shape vectors,
Figure BDA00035746609200000410
representing details of a face consisting of texture vectors, a i And b i Respectively representing the shape vector and the texture vector of the ith face,
Figure BDA00035746609200000411
representing the number of the faces in the internal face database; suppose the rendered model face image is I model (x mat ，y mat ) The target face image is I input (x mat ，y mat ) Then the objective function is:
Figure BDA00035746609200000412
wherein | · | purple sweet 2 Denotes the square of the 1-norm, (x) mat ,y mat ) Representing pixel coordinates for three-dimensional face model matching, E I Representing the variance of the model face and the target face, I R,model (x mat ,y mat )、I G,model (x mat ,y mat )、I B,model (x mat ,y mat ) Respectively representing three model face images corresponding to three primary colors RGB channels, wherein T represents a transposed matrix;
in order to realize the cascade residual error network by using mathematical formula
Figure BDA00035746609200000413
Is an activation function,
Figure BDA00035746609200000414
For the convolution function, define the mth cascade residual block output as
Figure BDA00035746609200000415
And the residual errors after the two convolutional layers are added to obtain the following formula:
Figure BDA00035746609200000416
wherein, O m-1 Denotes the output, W, of the preceding 1 x 1 sized convolutional layer to which the mth cascaded residual block is connected m Denotes the convolutional layer to which the mth cascaded residual block is connected, W m,1 Denotes the first convolutional layer, W, to which the mth concatenated residual block is connected m ,2 Indicates the second convolution layer to which the mth residual block is connected, "" indicates a cascade operation;
using the loss function:
Figure BDA0003574660920000051
wherein l MSE Represents a loss function, r o Representing a scaling factor, I LR Representing low resolution face images, I HR Representing high resolution face images, W id Width of picture, H ig Which represents the height of the picture,
Figure BDA0003574660920000056
representing the generated frontal face image.
Further, in the step s5, after the three-dimensional face model is rotated according to the face deflection angle, the face similarity matching is performed between the face picture and the rotated three-dimensional face model, and a specific process of obtaining the face similarity SIM is as follows:
first, the RGB color space is converted into the Lup color space through the XYZ color space, the conversion formula is as follows:
Figure BDA0003574660920000052
wherein (X) col ,Y col ,Z col ) Coordinates representing XYZ color space, (R) col ,G col ,B col ) Coordinates representing an RGB color space;
then, the XYZ color space is converted to the Lup color space, the conversion formula being as follows:
Figure BDA0003574660920000053
Figure BDA0003574660920000054
wherein (L) c ,u c ,p c ) Coordinates representing the Lup color space, t c And
Figure BDA0003574660920000055
all represent intermediate variables;
the human face comparison process is as follows:
Figure BDA0003574660920000061
wherein, Pic 1 And Pic 2 Indicating that two pictures are to be compared and,
Figure BDA0003574660920000062
and
Figure BDA0003574660920000063
the pixel value variances of the two pictures are respectively represented,
Figure BDA0003574660920000064
and
Figure BDA0003574660920000065
respectively representing the pixel mean of the two pictures,
Figure BDA0003574660920000066
representing the covariance of the two pictures, LU representing the brightness similarity, CON representing the contrast similarity, STR representing the structural similarity;
the calculation formula of the face similarity SIM is as follows:
Figure BDA0003574660920000067
where SIM represents the final face similarity, in the range of [0,1 ].
Further, in the step s5, the specific process of performing scene division operation according to the face similarity SIM and obtaining a final three-dimensional face model includes the following steps:
when the face similarity SIM is less than or equal to 60 percent, firstly expanding the number of the three-dimensional face pictures in the internal face database, and then repeating the step s3 to carry out self-feedback punishment training on the deformation model;
when the face similarity is more than or equal to 60% and the SIM is less than or equal to 85%, repeating the step s3 to carry out self-feedback punishment training on the deformation model;
and when the face similarity SIM is more than or equal to 85 percent, the three-dimensional face model at the moment is the final three-dimensional face model.
Compared with the prior art, the invention has the following beneficial technical effects:
the invention can adopt a face three-dimensional intelligent reconstruction method to carry out multi-angle face recognition, integrates a face characteristic point reconstruction method, has higher recognition accuracy on the side face with larger angle deflection, improves the running speed by using a deep learning method, and has very wide application scenes.
Drawings
FIG. 1 is a schematic flow chart of a multi-angle face recognition method based on three-dimensional intelligent reconstruction according to an embodiment;
FIG. 2 is a schematic diagram of 68 points of a person's face according to one embodiment;
FIG. 3 is a diagram of eye distance characterization for one embodiment;
FIG. 4 is a nose bridge distance characterization graph of an embodiment;
FIG. 5 is a mouth distance characterization graph of an embodiment;
FIG. 6 is a mandible distance characterization graph for one embodiment;
FIG. 7 is a schematic view of an embodiment of a face deflection angle scene division;
fig. 8 is a schematic diagram of a cascaded residual network structure according to an embodiment.
Detailed Description
In order to make the objects, technical solutions and advantages of the present application more apparent, the present application is described in further detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative of the present application and are not intended to limit the present application.
Reference herein to "an embodiment" means that a particular feature, structure, or characteristic described in connection with the embodiment can be included in at least one embodiment of the application. The appearances of the phrase in various places in the specification are not necessarily all referring to the same embodiment, nor are separate or alternative embodiments mutually exclusive of other embodiments. It is explicitly and implicitly understood by one skilled in the art that the embodiments described herein can be combined with other embodiments.
Referring to fig. 1, fig. 1 is a schematic flow chart of a multi-angle face recognition method based on three-dimensional intelligent reconstruction according to an embodiment, and includes the following steps:
s 1: inputting a face picture, extracting face characteristic points of the face picture, calculating face characteristic quantity of each region of the face, and calculating face deflection angles in scenes according to the number of the extracted face characteristic points;
the face picture can be a front face or a non-front face.
s 2: reconstructing the extracted face characteristic points into front face characteristic points by adopting an ASM algorithm based on the face characteristic quantity of each region of the face and the face deflection angle;
s 3: constructing an internal face database and a deformation model, predicting a shape vector and a texture vector of the deformation model based on the internal face database by using a sample-trained cascade residual error network, wherein the internal face database stores a three-dimensional face picture training data set for prediction;
s 4: constructing a three-dimensional face gray model according to the predicted shape vector, and then performing matching operation on the reconstructed front face feature points and the three-dimensional face gray model; if the matching is successful, coloring and detail supplementing operations are carried out on the three-dimensional face gray model based on the texture vector, and then a three-dimensional face model is established; if the matching fails, returning to the step s3 to perform self-feedback penalty training on the deformation model;
s 5: after the three-dimensional face model is rotated according to the face deflection angle, matching the input face picture with the rotated three-dimensional face model through face similarity, obtaining face similarity SIM, and performing scene operation according to the face similarity SIM to obtain a final three-dimensional face model;
s 6: and comparing the final three-dimensional face model with an external face database to finish final face recognition.
In one embodiment, in the step s1, 68 feature points shown in fig. 2 are selected as the face feature points, and the facial feature quantities of the regions of the face include: the human eye characteristic quantity as shown in fig. 3, the nose bridge characteristic quantity as shown in fig. 4, the mouth characteristic quantity as shown in fig. 5, and the mandible characteristic quantity as shown in fig. 6;
the calculation formula of the human eye characteristic quantity is as follows:
Figure BDA0003574660920000081
wherein D is eye A characteristic quantity of a human eye is represented,
Figure BDA0003574660920000082
and
Figure BDA0003574660920000083
respectively represents the sum of the connecting line distances of the two characteristic points of the upper eyelid and the two corresponding characteristic points of the lower eyelid,
Figure BDA0003574660920000084
representing the distance of a transverse connecting line of the characteristic points of the left and right canthus;
the calculation formula of the nose bridge characteristic quantity is as follows:
Figure BDA0003574660920000085
wherein D is nose The nose bridge characteristic quantity is represented,
Figure BDA0003574660920000086
the vertical distance between the highest feature point and the lowest feature point of the bridge of the middle nose is represented,
Figure BDA0003574660920000087
representing a lateral link distance of a lowest feature point of the bridge of the middle nose and a rightmost feature point of the nose;
the calculation formula of the mouth characteristic quantity is as follows:
Figure BDA0003574660920000088
wherein D is mouth The characteristic quantity of the mouth is represented,
Figure BDA0003574660920000089
respectively represents the vertical distance of the connecting lines of 3 groups of upper and lower characteristic points of the upper lip,
Figure BDA00035746609200000810
respectively represents the vertical distance of the connecting lines of 3 groups of upper and lower characteristic points of the lower lip,
Figure BDA00035746609200000811
and
Figure BDA00035746609200000812
respectively showing the transverse connection distance between the left and right corners of the upper and lower lips;
the calculation formula of the mandible characteristic quantity is as follows:
Figure BDA00035746609200000813
wherein D is mandible The lower jaw characteristic quantity is expressed,
Figure BDA00035746609200000814
representing the vertical distance between the lowest mandibular feature point and the highest mandibular feature point,
Figure BDA00035746609200000815
the lateral distance between the lowest mandibular feature point and the highest mandibular feature point is represented.
In one embodiment, as shown in fig. 7, the detected face feature points are divided into three scenes according to the number of the detected face feature points, and when 65 to 68 face feature points are detected, the detected face feature points are classified as a scene (1), and the corresponding face deflection angle ranges from 0 to 23 degrees; when 48-64 personal face characteristic points are detected, classifying the points into a scene (2), wherein the deflection angle range of the corresponding face is 23-67 degrees; when less than 48 feature points are detected, they are classified as scene (3), corresponding to a face deflection angle in the range of 67-90 °.
Let the angle of rotation around the x-axis be α, the angle of rotation around the y-axis be β, the angle of rotation around the z-axis be γ, and the corresponding total rotation matrix be:
Figure BDA0003574660920000091
let the three-dimensional model vertex be (X) mod ,Y mod ,Z mod ) The translation vector is (T) x ,T y ,T z ) T The image feature point is pi (u) s ,v s )，
Comprises the following steps:
Figure BDA0003574660920000092
wherein s is r Is a constant number, K r Is the adjustment vector, noted as:
s r pi＝HI；
the matrix H is a conversion factor from two dimensions to three dimensions, called an identity matrix, pi is a feature vector of an image feature point, I is a feature vector of a three-dimensional model vertex, and the internal structure of the identity matrix is as follows:
Figure BDA0003574660920000093
wherein fu is a camera parameter, and solving the camera parameter fu by a least square method is as follows:
Figure BDA0003574660920000094
giving the nonlinear system of equations to be solved:
Figure BDA0003574660920000095
the sub-scene calculation formula of the face deflection angle is as follows:
Figure BDA0003574660920000101
wherein the content of the first and second substances,
Figure BDA0003574660920000102
representing the deflection angle of the human face, omega representing the number of extracted characteristic points of the human face,
Figure BDA0003574660920000103
representing the distance between the eyes of the face after rotation,
Figure BDA0003574660920000104
indicating the distance, R, from the mouth to the line connecting the centers of the two eyes after deflection 31 、R 32 、R 33 Respectively representing variables in a rotation matrix, a being a constant, theta 1 Representing the angle of deflection, theta 2 Representing a horizontal deflection angle;
if the omega is more than or equal to 65 and less than or equal to 68, calculating the face deflection angle by using a first formula; if the omega is more than or equal to 48 and less than or equal to 64, calculating the face deflection angle by using a second formula; if omega is less than 48, a third formula is used for calculating the face deflection angle.
In one embodiment, in step s2, the reconstruction of the facial feature points is expressed by the following model:
Figure BDA0003574660920000105
wherein the content of the first and second substances,
Figure BDA0003574660920000106
representing an average shape vector, wherein pi represents a new human face sample characteristic point formed by the principal component shape vectors; b * A vector representing the weight of a control feature point,
Figure BDA0003574660920000107
representing a covariance matrix obtained after extracting the principal component shape vector,
Figure BDA0003574660920000108
represents the average amount of human eye features,
Figure BDA0003574660920000109
the average nose bridge characteristic quantity is expressed,
Figure BDA00035746609200001010
the average mouth characteristic quantity is represented,
Figure BDA00035746609200001011
the average mandibular feature quantity is expressed.
Using the constructed mapping model to test and match the new point set of the new image, i.e. using the model region to describe the feature point set in the image
Figure BDA00035746609200001012
Figure BDA00035746609200001013
Figure BDA00035746609200001014
Wherein the content of the first and second substances,
Figure BDA00035746609200001015
and
Figure BDA00035746609200001016
the amount of translational change of each feature point is represented,
Figure BDA00035746609200001017
a transformation matrix is represented that is,
Figure BDA00035746609200001018
represents a mapping manner for converting the feature points of the original image into the feature points of the face,
Figure BDA00035746609200001019
set of feature points, s, representing a new image u The scale of the scaling is represented by,
Figure BDA0003574660920000111
representing the face deflection angle, corresponding to the calculation method of the previous section under three different scenes, x ASM And y ASM Representing each feature point that needs to be transformed.
To describe
Figure BDA0003574660920000112
And
Figure BDA0003574660920000113
establishing an objective function to express the matching degree according to the matching similarity between the following steps:
Figure BDA0003574660920000114
wherein |. non chlorine 2 The square of the modulus is represented as,
Figure BDA0003574660920000115
for the newly established objective function, parameter b * Is set to 0, the face model is the average face model at the moment, and the transformation matrix is solved through the rotation translation transformation
Figure BDA0003574660920000116
Then will be
Figure BDA0003574660920000117
By transforming matrices
Figure BDA0003574660920000118
Mapping to a face model to obtain a set of points
Figure BDA0003574660920000119
Figure BDA00035746609200001110
Wherein the content of the first and second substances,
Figure BDA00035746609200001111
to represent
Figure BDA00035746609200001112
Inverse matrix of (2), new point set
Figure BDA00035746609200001113
Mapping to a set of mean points
Figure BDA00035746609200001114
In the tangent plane of (a), the following scale transformation is made:
Figure BDA00035746609200001115
wherein the content of the first and second substances,
Figure BDA00035746609200001116
representing a new set of points subjected to scale transformation, and solving new shape parameters through a shape model
Figure BDA00035746609200001117
Figure BDA00035746609200001118
Wherein the content of the first and second substances,
Figure BDA00035746609200001119
representing a covariance matrix
Figure BDA00035746609200001120
If updated
Figure BDA00035746609200001121
If the convergence reaches the given range, the positioning is successful, and if the convergence does not exist, the feature point set in the human face is continuously searched according to the solved new pi
Figure BDA00035746609200001122
Up to the shape parameter
Figure BDA00035746609200001123
And (6) converging.
When the human face measurement standard is judged, a judgment criterion is introduced, wherein the distance error between the point cloud of the target object shape model and the point cloud of the adjacent boundary is used as matching judgment criterion. The vector y represents the shape model point cloud, y' represents the neighboring edge point cloud, and the distance error is:
Figure BDA00035746609200001124
wherein the content of the first and second substances,
Figure BDA00035746609200001125
indicating the distance error, |, denotes modulo. Supposing that a certain face characteristic point is input, sampling points from the left side and the right side of the characteristic point and constructing a vector g face Then, a series set of (ordered) facial feature points { g } of the front face is obtained face }。
In one embodiment, the cascaded residual network is used to predict the shape vector and the texture vector of the deformation model, and an original algorithm flow of the deformation model for calculating the shape vector and the texture vector is eliminated, in step s3, the structure of the cascaded residual network after sample training includes the following steps:
the three-dimensional face is uniformly represented in a vector form:
Figure BDA0003574660920000121
wherein the content of the first and second substances,
Figure BDA0003574660920000122
and
Figure BDA0003574660920000123
respectively representing the shape vector and the texture vector of the ith face in the internal face database,
Figure BDA0003574660920000124
representing a face gray model consisting of shape vectors,
Figure BDA0003574660920000125
representing details of a face consisting of texture vectors, a i And b i Respectively representing the shape vector and the texture vector of the ith face,
Figure BDA0003574660920000126
representing the number of the faces in the internal face database; suppose the rendered model face image is I model (x mat ，y mat ) The target face image is I input (x mat ，y mat ) Then the objective function is:
Figure BDA0003574660920000127
wherein | · | purple sweet 2 Denotes the square of the 1-norm, (x) mat ,y mat ) Representing pixel coordinates for three-dimensional face model matching, E I To representVariance of model face and target face, I R,model (x mat ,y mat )、I G,model (x mat ,y mat )、I B,model (x mat ,y mat ) Respectively representing three model face images corresponding to three primary colors RGB channels, and T representing a transposed matrix.
As shown in FIG. 8, to implement the cascaded residual network using mathematical formulas, let
Figure BDA0003574660920000128
Is an activation function,
Figure BDA0003574660920000129
For the convolution function, define the mth cascade residual block output as
Figure BDA00035746609200001210
And the residual errors after the two convolutional layers are added to obtain the following formula:
Figure BDA00035746609200001211
wherein, O m-1 Denotes the output, W, of the preceding 1 x 1 sized convolutional layer to which the mth cascaded residual block is connected m Denotes the convolutional layer to which the mth cascaded residual block is connected, W m,1 Denotes the first convolutional layer, W, to which the mth concatenated residual block is connected m ,2 Indicates the second convolution layer to which the mth residual block is connected, "" indicates a cascade operation;
using the loss function:
Figure BDA0003574660920000131
wherein l MSE Represents a loss function, r o Representing a scaling factor, I LR Representing low resolution face images, I HR Representing high resolution face images, W id Width of picture, H ig Which represents the height of the picture,
Figure BDA0003574660920000132
representing the generated frontal face image.
In an embodiment, in the step s5, after the three-dimensional face model is rotated according to the face deflection angle, the face similarity matching between the face picture and the rotated three-dimensional face model is performed, and a specific process of obtaining the face similarity SIM is as follows:
first, the RGB color space is converted into the Lup color space through the XYZ color space, the conversion formula is as follows:
Figure BDA0003574660920000133
wherein (X) col ,Y col ,Z col ) Coordinates representing XYZ color space, (R) col ,G col ,B col ) Coordinates representing an RGB color space;
then, the XYZ color space is converted to the Lup color space, the conversion formula being as follows:
Figure BDA0003574660920000134
Figure BDA0003574660920000135
wherein (L) c ,u c ,p c ) Coordinates representing the Lup color space, t c And
Figure BDA0003574660920000136
all represent intermediate variables;
the human face comparison process is as follows:
Figure BDA0003574660920000141
wherein, Pic 1 And Pic 2 Indicating that two pictures are to be compared and,
Figure BDA0003574660920000142
and
Figure BDA0003574660920000143
the pixel value variances of the two pictures are respectively represented,
Figure BDA0003574660920000144
and
Figure BDA0003574660920000145
respectively representing the pixel mean of the two pictures,
Figure BDA0003574660920000146
representing the covariance of the two pictures, LU representing the brightness similarity, CON representing the contrast similarity, STR representing the structural similarity;
the calculation formula of the face similarity SIM is as follows:
Figure BDA0003574660920000147
where SIM represents the final face similarity, in the range of [0,1 ].
In an embodiment, in the step s5, the specific process of performing scene division operation according to the face similarity SIM and obtaining a final three-dimensional face model includes the following steps:
when the face similarity SIM is less than or equal to 60 percent, firstly expanding the number of the three-dimensional face pictures in the internal face database, and then repeating the step s3 to carry out self-feedback punishment training on the deformation model;
when the face similarity is more than or equal to 60% and the SIM is less than or equal to 85%, repeating the step s3 to carry out self-feedback punishment training on the deformation model;
and when the face similarity SIM is more than or equal to 85 percent, the three-dimensional face model at the moment is the final three-dimensional face model.
When the face similarity is less than or equal to 85% SIM, the deep learning prediction result is accurate, and self-feedback training punishment is not needed; when the face similarity SIM is less than or equal to 60 percent, the face three-dimensional reconstruction result is seriously inconsistent with the input picture, and the main reason is presumed to be that the three-dimensional face database data set is insufficient and the number of faces in the internal face database needs to be expanded; when the face similarity meets the condition that SIM is more than or equal to 60% and less than or equal to 85%, the deep learning prediction result is inaccurate, and self-feedback training punishment needs to be carried out:
as can be seen from the above, the matching of the three-dimensional face model is a multiple iteration process, and the iteration form of the k-th layer three-dimensional face model error is represented as:
Figure BDA0003574660920000151
wherein (x) mat ,y mat ) Representing pixel value coordinates for three-dimensional face model matching,
Figure BDA0003574660920000152
representing the target face image over k iterations,
Figure BDA0003574660920000153
representing a three-dimensional model face image after k iterations,
Figure BDA0003574660920000154
represents the variance over k iterations;
the form of the combined face vector after k iterations is as follows:
Figure BDA0003574660920000155
wherein the content of the first and second substances,
Figure BDA0003574660920000156
representing the three-dimensional model face shape vector after k iterations,
Figure BDA0003574660920000157
representing the three-dimensional model face texture vector after k iterations,
Figure BDA0003574660920000158
representing the average face shape over k iterations,
Figure BDA0003574660920000159
representing the average face texture over k iterations, a k Representing the shape vector over k iterations, b k Representing the texture vector over k iterations,
Figure BDA00035746609200001510
representing the shape vector principal components over k iterations,
Figure BDA00035746609200001511
representing the principal components of the texture vector after k iterations;
the iterative relationship of the texture vector and the shape vector from the lower layer to the upper layer is as follows:
Figure BDA00035746609200001512
wherein the content of the first and second substances,
Figure BDA00035746609200001513
a transformation matrix representing the shape vector over k iterations,
Figure BDA00035746609200001514
a transformation matrix representing the texture vector over k iterations,
Figure BDA00035746609200001515
and
Figure BDA00035746609200001516
to represent
Figure BDA00035746609200001517
And
Figure BDA00035746609200001518
the transposed matrix of (2);
the iteration times are determined according to the face similarity SIM, and the lower the face similarity SIM is, the more the iteration times are; the higher the face similarity SIM is, the fewer the iteration times are;
the calculation formula of the number of iterations is as follows:
Figure BDA00035746609200001519
wherein S is ta Sample standard deviation, λ, representing an internal face database cap Denotes the sample size, SIM denotes the degree of similarity of human face, ω denotes the level of significance, ξ denotes a constant,
Figure BDA0003574660920000161
the level of confidence is indicated and,
Figure BDA0003574660920000162
representing the historical iterative mean, k up Represents the upper limit of the number of iterations, k down Represents the lower limit of the number of iterations, k final Representing the final number of iterations.
The technical features of the above embodiments can be arbitrarily combined, and for the sake of brevity, all possible combinations of the technical features in the above embodiments are not described, but should be considered as the scope of the present specification as long as there is no contradiction between the combinations of the technical features.
It should be noted that the terms "first \ second \ third" referred to in the embodiments of the present application merely distinguish similar objects, and do not represent a specific ordering for the objects, and it should be understood that "first \ second \ third" may exchange a specific order or sequence when allowed. It should be understood that "first \ second \ third" distinct objects may be interchanged under appropriate circumstances such that the embodiments of the application described herein may be implemented in an order other than those illustrated or described herein.
The terms "comprising" and "having" and any variations thereof in the embodiments of the present application are intended to cover non-exclusive inclusions. For example, a process, method, apparatus, product, or device that comprises a list of steps or modules is not limited to the listed steps or modules but may alternatively include other steps or modules not listed or inherent to such process, method, product, or device.
The above-mentioned embodiments only express several embodiments of the present application, and the description thereof is more specific and detailed, but not construed as limiting the scope of the invention. It should be noted that, for a person skilled in the art, several variations and modifications can be made without departing from the concept of the present application, which falls within the scope of protection of the present application. Therefore, the protection scope of the present patent shall be subject to the appended claims.

Claims (7)

1. The multi-angle face recognition method based on three-dimensional intelligent reconstruction is characterized by comprising the following steps:
s 1: inputting a face picture, extracting face characteristic points of the face picture, calculating face characteristic quantity of each region of the face, and calculating face deflection angles in scenes according to the number of the extracted face characteristic points;
s 2: reconstructing the extracted face characteristic points into front face characteristic points by adopting an ASM algorithm based on the face characteristic quantity of each region of the face and the face deflection angle;
s 3: constructing an internal face database and a deformation model, predicting a shape vector and a texture vector of the deformation model based on the internal face database by using a sample-trained cascade residual error network, wherein the internal face database stores a three-dimensional face picture training data set for prediction;
s 4: constructing a three-dimensional face gray model according to the predicted shape vector, and then performing matching operation on the reconstructed front face feature points and the three-dimensional face gray model; if the matching is successful, coloring and detail supplementing operations are carried out on the three-dimensional face gray model based on the texture vector, and then a three-dimensional face model is established; if the matching fails, returning to the step s3 to perform self-feedback penalty training on the deformation model;
s 5: after the three-dimensional face model is rotated according to the face deflection angle, matching the input face picture with the rotated three-dimensional face model through face similarity, obtaining face similarity SIM, and performing scene operation according to the face similarity SIM to obtain a final three-dimensional face model;
s 6: and comparing the final three-dimensional face model with an external face database to finish final face recognition.
2. The method according to claim 1, wherein in step s1, there are 68 feature points in the face, and the facial feature quantities of each region of the face include: human eye characteristic quantity, nose bridge characteristic quantity, mouth characteristic quantity and mandible characteristic quantity;
the calculation formula of the human eye characteristic quantity is as follows:
Figure FDA0003574660910000011
wherein D is eye A characteristic quantity of a human eye is represented,
Figure FDA0003574660910000012
and
Figure FDA0003574660910000013
respectively represents the sum of the connecting line distances of the two characteristic points of the upper eyelid and the two corresponding characteristic points of the lower eyelid,
Figure FDA0003574660910000014
representing the distance of a transverse connecting line of the characteristic points of the left and right canthus;
the calculation formula of the nose bridge characteristic quantity is as follows:
Figure FDA0003574660910000015
wherein D is nose The nose bridge characteristic quantity is represented,
Figure FDA0003574660910000016
the vertical distance between the highest feature point and the lowest feature point of the bridge of the middle nose is represented,
Figure FDA0003574660910000017
representing a lateral link distance of a lowest feature point of the bridge of the middle nose and a rightmost feature point of the nose;
the calculation formula of the mouth characteristic quantity is as follows:
Figure FDA0003574660910000021
wherein D is mouth The characteristic quantity of the mouth is represented,
Figure FDA0003574660910000022
respectively represents the vertical distance of the connecting lines of 3 groups of upper and lower characteristic points of the upper lip,
Figure FDA0003574660910000023
respectively represents the vertical distance of the connecting lines of 3 groups of upper and lower characteristic points of the lower lip,
Figure FDA0003574660910000024
and
Figure FDA0003574660910000025
respectively showing the transverse connection distance between the left and right corners of the upper and lower lips;
the calculation formula of the mandible characteristic quantity is as follows:
Figure FDA0003574660910000026
wherein D is mandible The lower jaw characteristic quantity is expressed,
Figure FDA0003574660910000027
representing the vertical distance between the lowest mandibular feature point and the highest mandibular feature point,
Figure FDA0003574660910000028
the lateral distance between the lowest mandibular feature point and the highest mandibular feature point is represented.
3. The method for multi-angle face recognition based on three-dimensional intelligent reconstruction of claim 2, wherein in the step s1, the sub-scene calculation formula of the face deflection angle is as follows:
Figure FDA0003574660910000029
wherein the content of the first and second substances,
Figure FDA00035746609100000210
representing the deflection angle of the human face, omega representing the number of extracted characteristic points of the human face,
Figure FDA00035746609100000211
representing the distance between the eyes of the face after rotation,
Figure FDA00035746609100000212
indicating the distance, R, from the mouth to the line connecting the centers of the two eyes after deflection 31 、R 32 、R 33 Respectively representing variables in a rotation matrix, a being a constant, theta 1 Representing the angle of deflection, theta 2 Representing a horizontal deflection angle;
if the omega is more than or equal to 65 and less than or equal to 68, calculating the face deflection angle by using a first formula; if the omega is more than or equal to 48 and less than or equal to 64, calculating the face deflection angle by using a second formula; if omega is less than 48, a third formula is used for calculating the face deflection angle.
4. The method for multi-angle face recognition based on three-dimensional intelligent reconstruction of claim 3, wherein in the step s2, the reconstruction of the front face feature points is expressed by the following model:
Figure FDA0003574660910000031
Figure FDA0003574660910000032
wherein the content of the first and second substances,
Figure FDA0003574660910000033
representing an average shape vector, wherein pi represents a new human face sample characteristic point formed by the principal component shape vectors; b * A vector representing the weight of a control feature point,
Figure FDA0003574660910000034
representing a covariance matrix obtained after extracting the principal component shape vector,
Figure FDA0003574660910000035
represents the average amount of human eye features,
Figure FDA0003574660910000036
the average nose bridge characteristic quantity is expressed,
Figure FDA0003574660910000037
the average mouth characteristic quantity is represented,
Figure FDA0003574660910000038
the average mandibular feature quantity is expressed.
5. The method for multi-angle face recognition based on three-dimensional intelligent reconstruction as claimed in claim 4, wherein in said step s3, the structure of said sample-trained cascade residual error network includes the following:
the three-dimensional face is uniformly represented in a vector form:
Figure FDA0003574660910000039
wherein the content of the first and second substances,
Figure FDA00035746609100000310
and
Figure FDA00035746609100000311
respectively representing the shape vector and the texture vector of the ith face in the internal face database,
Figure FDA00035746609100000312
representing a face gray model consisting of shape vectors,
Figure FDA00035746609100000313
representing details of a face consisting of texture vectors, a i And b i Respectively representing the shape vector and the texture vector of the ith face,
Figure FDA00035746609100000314
representing the number of the faces in the internal face database; suppose the rendered model face image is I model (x mat ，y mat ) The target face image is I input (x mat ，y mat ) Then the objective function is:
Figure FDA00035746609100000315
wherein | · | purple sweet 2 Denotes the square of the 1-norm, (x) mat ,y mat ) Representing pixel coordinates for three-dimensional face model matching, E I Representing the variance of the model face and the target face, I R,model (x mat ,y mat )、I G,model (x mat ,y mat )、I B,model (x mat ,y mat ) Respectively representing three model face images corresponding to three primary colors RGB channels, wherein T represents a transposed matrix;
in order to realize the cascade residual error network by using mathematical formula
Figure FDA0003574660910000041
Is an activation function,
Figure FDA0003574660910000042
For the convolution function, define the mth cascade residual block output as
Figure FDA0003574660910000043
And the residual errors after the two convolutional layers are added to obtain the following formula:
Figure FDA0003574660910000044
wherein, O m-1 Denotes the output, W, of the preceding 1 x 1 sized convolutional layer to which the mth cascaded residual block is connected m Denotes the convolutional layer to which the mth cascaded residual block is connected, W m,1 Denotes the first convolutional layer, W, to which the mth concatenated residual block is connected m,2 Indicates the second convolution layer to which the mth residual block is connected, "" indicates a cascade operation;
using the loss function:
Figure FDA0003574660910000045
wherein l MSE Represents a loss function, r o Representing a scaling factor, I LR Representing low resolution face images, I HR Representing high resolution face images, W id Width of picture, H ig Which represents the height of the picture,
Figure FDA0003574660910000046
representing the generated frontal face image.
6. The method according to claim 5, wherein in step s5, after the three-dimensional face model is rotated according to the face deflection angle, the specific process of matching the input face image with the rotated three-dimensional face model to obtain the face similarity SIM is as follows:
first, the RGB color space is converted into the Lup color space through the XYZ color space, the conversion formula is as follows:
Figure FDA0003574660910000047
wherein (X) col ,Y col ,Z col ) Coordinates representing XYZ color space, (R) col ,G col ,B col ) Coordinates representing an RGB color space;
then, the XYZ color space is converted to the Lup color space, the conversion formula being as follows:
Figure FDA0003574660910000051
Figure FDA0003574660910000052
wherein (L) c ,u c ,p c ) Coordinates representing the Lup color space, t c And
Figure FDA0003574660910000053
all represent intermediate variables;
the human face comparison process is as follows:
Figure FDA0003574660910000054
Figure FDA0003574660910000055
Figure FDA0003574660910000056
wherein, Pic 1 And Pic 2 Indicating that two pictures are to be compared and,
Figure FDA0003574660910000057
and
Figure FDA0003574660910000058
the pixel value variances of the two pictures are respectively represented,
Figure FDA0003574660910000059
and
Figure FDA00035746609100000510
respectively representing the pixel mean of the two pictures,
Figure FDA00035746609100000511
representing the covariance of the two pictures, LU representing the brightness similarity, CON representing the contrast similarity, STR representing the structural similarity;
the calculation formula of the face similarity SIM is as follows:
Figure FDA00035746609100000512
where SIM represents the final face similarity, in the range of [0,1 ].
7. The method for multi-angle face recognition based on three-dimensional intelligent reconstruction as claimed in claim 6, wherein in the step s5, the specific process of obtaining the final three-dimensional face model according to the face similarity SIM scene operation comprises the following steps:
when the face similarity SIM is less than or equal to 60 percent, firstly expanding the number of the three-dimensional face pictures in the internal face database, and then repeating the step s3 to carry out self-feedback punishment training on the deformation model;
when the face similarity is more than or equal to 60% and the SIM is less than or equal to 85%, repeating the step s3 to carry out self-feedback punishment training on the deformation model;
and when the face similarity SIM is more than or equal to 85 percent, the three-dimensional face model at the moment is the final three-dimensional face model.
CN202210336825.5A 2022-03-31 2022-03-31 Multi-angle face recognition method based on three-dimensional intelligent reconstruction Pending CN114882545A (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202210336825.5A CN114882545A (en) 2022-03-31 2022-03-31 Multi-angle face recognition method based on three-dimensional intelligent reconstruction

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202210336825.5A CN114882545A (en) 2022-03-31 2022-03-31 Multi-angle face recognition method based on three-dimensional intelligent reconstruction

Publications (1)

Publication Number Publication Date
CN114882545A true CN114882545A (en) 2022-08-09

Family

ID=82670320

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202210336825.5A Pending CN114882545A (en) 2022-03-31 2022-03-31 Multi-angle face recognition method based on three-dimensional intelligent reconstruction

Country Status (1)

Country Link
CN (1) CN114882545A (en)

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN116978102A (en) * 2023-08-04 2023-10-31 深圳市英锐存储科技有限公司 Face feature modeling and recognition method, chip and terminal

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN116978102A (en) * 2023-08-04 2023-10-31 深圳市英锐存储科技有限公司 Face feature modeling and recognition method, chip and terminal

Similar Documents

Publication Publication Date Title
CN112766160B (en) Face replacement method based on multi-stage attribute encoder and attention mechanism
Dai et al. A 3d morphable model of craniofacial shape and texture variation
CN107229920B (en) Behavior identification method based on integration depth typical time warping and related correction
CN111223133A (en) Registration method of heterogeneous images
KR20060058147A (en) Image collation system and image collation method
CN108470178B (en) Depth map significance detection method combined with depth credibility evaluation factor
CN109711268B (en) Face image screening method and device
CN112733627B (en) Finger vein recognition method based on fusion local and global feature network
CN111783748A (en) Face recognition method and device, electronic equipment and storage medium
CN110598638A (en) Model training method, face gender prediction method, device and storage medium
CN113435282B (en) Unmanned aerial vehicle image ear recognition method based on deep learning
CN110969171A (en) Image classification model, method and application based on improved convolutional neural network
CN113610046B (en) Behavior recognition method based on depth video linkage characteristics
CN111428689A (en) Multi-pool information fusion human face image feature extraction method
CN111652273A (en) Deep learning-based RGB-D image classification method
CN114140623A (en) Image feature point extraction method and system
CN112288758B (en) Infrared and visible light image registration method for power equipment
CN113378812A (en) Digital dial plate identification method based on Mask R-CNN and CRNN
CN116958420A (en) High-precision modeling method for three-dimensional face of digital human teacher
CN114882545A (en) Multi-angle face recognition method based on three-dimensional intelligent reconstruction
CN110321452A (en) A kind of image search method based on direction selection mechanism
CN111815620B (en) Fabric defect detection method based on convolution characteristic and low-rank representation
CN108765384B (en) Significance detection method for joint manifold sequencing and improved convex hull
CN111915735A (en) Depth optimization method for three-dimensional structure contour in video
CN111428555A (en) Joint-divided hand posture estimation method

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