CN110399814B - Face recognition method based on local linear representation field adaptive measurement - Google Patents
Face recognition method based on local linear representation field adaptive measurement Download PDFInfo
- Publication number
- CN110399814B CN110399814B CN201910628501.7A CN201910628501A CN110399814B CN 110399814 B CN110399814 B CN 110399814B CN 201910628501 A CN201910628501 A CN 201910628501A CN 110399814 B CN110399814 B CN 110399814B
- Authority
- CN
- China
- Prior art keywords
- matrix
- image
- vector data
- linear representation
- local linear
- 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
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
- 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
-
- 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)
- Health & Medical Sciences (AREA)
- Oral & Maxillofacial Surgery (AREA)
- General Health & Medical Sciences (AREA)
- Human Computer Interaction (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Multimedia (AREA)
- Theoretical Computer Science (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Image Analysis (AREA)
- Image Processing (AREA)
Abstract
A face recognition method based on local linear representation field adaptive measurement is disclosed, the technical scheme is as follows: vector data point X after preprocessing of a face image i Respectively fromSelecting vector data point X after preprocessing with one face image from vector data point forming matrix data X after preprocessing with one face image i The method comprises the steps of performing local linear representation on K adjacent points with the same category to obtain a local linear representation coefficient, establishing a local reconstruction error matrix based on the local linear representation, measuring divergence between multi-manifold with different categories by using logarithmic Euclidean distance, measuring dissimilarity between a training data domain and a testing data domain by using the logarithmic Euclidean distance, and finally searching a low-dimensional discrimination subspace by maximizing the divergence between the multi-manifold and minimizing the difference between source data and target data to realize face image discrimination feature extraction. The method extracts the classification characteristics of the face image by maximizing the multi-manifold divergence and the similarity of training and testing data distribution, and has the characteristic of improving the face image recognition effect.
Description
Technical Field
The invention belongs to the technical field of face recognition, and particularly relates to a face recognition method based on local linear representation field adaptive measurement.
Background
The face recognition is one of main product forms of machine vision biological authentication, and is a necessary technology for public safety, daily life and the like. In recent years, the range of applications of face recognition is increasing, and high accuracy is required. In the imaging process of the device, due to reasons such as illumination, shielding and posture labels, the face recognition algorithm cannot effectively extract features, the defects influence the algorithm recognition rate, the effect of the face recognition technology in actual use is reduced, and even potential safety hazards are brought. The stability of the face recognition algorithm is particularly important.
Face recognition algorithms have gone through several stages of development. If the low-dimensional representation is obtained, the face recognition based on local features and the local descriptor method based on learning are carried out. At present, a deep learning method based on a deep neural network is provided, the method is strong in identification capability and good in robustness, but the structure is complex, and the calculated amount is large. How to obtain a method which has strong recognition capability, good robustness and small time-space complexity and still keeps good recognition effect on a small amount of training set samples becomes a hot topic of face recognition.
Disclosure of Invention
The invention aims to overcome the defects of the prior art and provides a face recognition method for adaptive measurement in the field of local linear representation, which can improve the recognition capability and robustness of a model and reduce the complexity of time and space.
In order to realize the technical scheme, the technical scheme adopted by the invention comprises the following specific steps:
s1, preprocessing the collected face image to obtain a preprocessed face image vector data matrix X = [ X ] 1 ,X 2 ,…,X C ]Wherein: c is the number of the categories,m is the vectorized feature dimension, N c C is more than or equal to 1 and less than or equal to C;
s2, for the c class X c From X c Vector data point x after pre-processing of one image is selected i Selecting vector data points x preprocessed with an image i K vector data points x with the closest euclidean distance i1 ,x i2 ,…,x iK Vector data points x after preprocessing an image i Performing local linear expression, minimizing linear expression mean square error function to obtain minimum linear expression coefficient vector W c Weighting coefficient vector W c Extending to N dimension, wherein the minimum linearity between vector points with neighbor relation represents that the coefficient is not changed, and the others are 0;
s3, repeating the steps for the data points of the rest classes in the image vector data matrix X to obtain the weight coefficient matrix W = [ W ] of all classes 1 ,W 2 ,…,W C ]∈R N×N Where N is the total number of all pre-processed images,
s4, based on the calculated weight coefficient matrix W = [ W ] 1 ,W 2 ,…,W C ]∈R N×N Calculating the reconstruction error of each class to obtain a reconstruction error matrix R = [ R = [ R ] ] 1 ,R 2 ,…,R C ];
S5, carrying out data matrix X = [ X ] on the preprocessed face image vector data 1 ,X 2 ,…,X C ]Training data set X S And test data set X T Any data point x of j Repeating the steps S2, S3 and S4 to respectively obtain a training data set X S In (3) reconstructing the error matrix R s =[R S1 ,R S2 ,…,R SC ]And test data set X T Is reconstructed error matrix R T =[R T1 ,R T2 ,…,R TC ];
S6, constructing a target function based on the reconstruction error matrix to obtain a low-dimensional projection matrix A;
s7, based on the low-dimensional projection matrix, vector data points x after one image is preprocessed i Calculating the low-dimensional characteristic Yi of the image after linear projection;
and S8, classifying the low-dimensional features Yi by using a nearest neighbor algorithm to realize the recognition of the human face features.
Further, the preprocessing of the acquired face image in step S1 includes graying, smoothing, normalizing, and vectorizing, which are performed in sequence.
Further, in step S2, the minimum linear representation mean square error function is:
wherein: x is the number of i Representing pre-processed vector data points, i.e. X, of an image c The (c) th column of (a),
x ij (j =1,2.., K) represents X c Vector data point x after preprocessing of one image i The K vector data points that are closest in euclidean distance,
W i,j represents the ith webThe weighting factor of j neighbors of an image, W c Each column of the ith row of (1).
Further, in step S4, the reconstruction error matrix of each class is represented as:
wherein: t represents a transpose operation of the matrix;
i is and W c Identity matrix with same dimension.
Further, step S6 specifically includes:
s61, applying a tiny disturbance to enable R c Becomes a positive definite matrix:
δ represents the disturbance magnitude;
s62, measuring different symmetrical positive definite matrixes R by using log-Euclidean distance c Distance between them, establishing divergence S between the manifolds M :
S63, training data set X by using log-Euclidean distance measurement S And test data set X T Degree of dissimilarity S of d :
S64, establishing an objective function in a low-dimensional space:
s.t.A T A=I
wherein A ∈ R d×N A low-dimensional projection matrix is represented,
d is the dimension of the space after the dimension reduction,
log (-) represents the matrix logarithm based on e;
i denotes an identity matrix.
Further, in step S6, the low-dimensional feature Y after linear projection i Expressed as:
Y i =A T X i (7)。
due to the adoption of the technical scheme, the invention has the beneficial effects that:
the invention adopts a face recognition method of adaptive measurement in the field of local linear representation, on one hand, vector data points X after preprocessing of a face image i Selecting vector data points X preprocessed with one face image from the matrix data X composed of the vector data points preprocessed with all the face images i The method comprises the steps of establishing manifold local linear representation by K adjacent points with the same category, measuring multi-manifold divergence between the manifold local linear representations by using log-Euclidean distance, and searching a low-dimensional projection matrix A by maximizing the multi-manifold divergence on the basis of keeping the manifold local structure unchanged.
Therefore, the method has the characteristics of strong recognition capability, good robustness, small time and space complexity and good recognition effect on a small amount of training set samples by maximizing the similarity characteristic between the source data domain and the target data domain.
Detailed Description
The invention is further described with reference to specific embodiments, without limiting its scope.
Example 1
A face recognition method based on local linear representation field adaptive measurement. The method of the embodiment comprises the following specific steps:
step one, carrying out gray processing, smoothing processing, normalization processing and vectorization processing on the collected face image to obtain a preprocessed face image vector data matrix X = [ X ] 1 ,X 2 ,…,X C ]Wherein: c is the number of the categories,m is the vectorized feature dimension, N c C is more than or equal to 1 and less than or equal to C;
step two, for the c class X c From X c Selecting a vector data point x after image preprocessing i Selecting vector data points x preprocessed with an image i K vector data points x with the closest euclidean distance i1 ,x i2 ,…,x iK Vector data points x after preprocessing an image i Performing local linear expression, and minimizing linear expression to obtain a mean square error function:
wherein: x is the number of i Representing pre-processed vector data points, i.e. X, of an image c The (c) th column of (a),
x ij (j =1,2.., K denotes X c Vector data point x after preprocessing of one image i The K vector data points that are closest in euclidean distance,
W i,j j neighbors representing the ith imageWeight coefficient of a point, i.e. W c Each column of the ith row of (1).
Calculating to obtain the minimum linear expression coefficient vector W c Weighting coefficient vector W c Extending to N dimension, wherein the minimum linearity between vector points with neighbor relation represents that the coefficient is unchanged, and the others are 0;
step three, repeating the steps for the data points of the rest classes in the image vector data matrix X to obtain the weight coefficient matrix W = [ W ] of all classes 1 ,W 2 ,…,W C ]∈R N×N Where N is the total number of all pre-processed images,
step four, based on the calculated weight coefficient matrix W = [ W ] 1 ,W 2 ,…,W C ]∈R N×N The reconstruction error matrix for each class is represented as:
wherein: t represents a transpose operation of the matrix;
i is and W c Identity matrix with same dimension.
Calculating the reconstruction error of each class to obtain a reconstruction error matrix R = [) 1 ,R 2 ,…,R C ];
Step five, carrying out X = [ X ] on the preprocessed face image vector data matrix 1 ,X 2 ,…,X C ]Training data set X S And test data set X T Any data point x of j Repeating the steps S2, S3 and S4 to respectively obtain a training data set X S In (3) reconstructing the error matrix R s =[R S1 ,R S2 ,…,R SC ]And test data set X T Is reconstructed error matrix R T =[R T1 ,R T2 ,…,R TC ];
Step six, applying a small disturbanceMake R c Becomes a positive definite matrix:
δ represents the disturbance magnitude;
the different symmetric positive definite matrices R are then measured using log-Euclidean distance c Distance between them, establishing divergence S between the manifolds M :
Training dataset X using log-Euclidean distance metric S And test data set X T Degree of dissimilarity S of d :
In the low-dimensional space, an objective function is established:
s.t.A T A=I
wherein A ∈ R d×N A low-dimensional projection matrix is represented,
d is the dimension of the space after the dimension reduction,
log (-) represents the matrix logarithm based on e;
i denotes an identity matrix.
The matrix a is obtained by maximizing the objective function.
Seventhly, vector data points after one image is preprocessed based on the low-dimensional projection matrixx i Low dimensional feature Y after linear projection i Expressed as:
Y i =A T X i (7)
calculating the low-dimensional characteristic Y of the image after linear projection i ;
Step eight, using a nearest neighbor algorithm to the low-dimensional feature Y i And classifying to realize the recognition of the human face characteristics.
The beneficial effects of the embodiment are as follows:
the invention adopts a face recognition method based on the adaptive measurement of the local linear representation field, on one hand, the vector data point X after the preprocessing of a face image i Selecting vector data points X preprocessed with one face image from the matrix data X composed of the vector data points preprocessed with all the face images i The method comprises the steps of establishing manifold local linear representation by K adjacent points with the same category, measuring multi-manifold divergence between the manifold local linear representations by using log-Euclidean distance, maximizing the multi-manifold divergence to find a low-dimensional projection matrix A on the basis of keeping the manifold local structure unchanged, and minimizing the covariance distance between a source data domain and a target data domain in a projected space to ensure the similarity between the source data domain and the target data domain, so that the discriminant feature extraction of a face image is realized, and the face recognition effect is improved.
Therefore, the method has the characteristics of strong recognition capability, good robustness, small time and space complexity and good recognition effect on a small amount of training set samples by maximizing the similarity characteristic between the source data domain and the target data domain.
Claims (6)
1. A face recognition method based on local linear representation field adaptive measurement is characterized by comprising the following steps:
s1, preprocessing the collected face image to obtain a preprocessed face image vector data matrix X = [ X ] 1 ,X 2 ,…,X C ]Wherein: c is the number of the categories,m is the vectorized feature dimension, N c C is more than or equal to 1 and less than or equal to C;
s2, for the c class X c From X c Vector data point x after pre-processing of one image is selected i Selecting vector data points x preprocessed with an image i K vector data points x with the closest euclidean distance i1 ,x i2 ,…,x iK Vector data points x after preprocessing an image i Performing local linear expression, minimizing linear expression mean square error function to obtain minimum linear expression coefficient vector W c Weighting coefficient vector W c Extending to N dimension, wherein the minimum linearity between vector points with neighbor relation represents that the coefficient is unchanged, and the others are 0;
s3, repeating the steps for the data points of the rest classes in the image vector data matrix X to obtain the weight coefficient matrix W = [ W ] of all classes 1 ,W 2 ,…,W C ]∈R N×N Where N is the total number of all pre-processed images,
s4, based on the calculated weight coefficient matrix W = [ W = [ W ] 1 ,W 2 ,…,W C ]∈R N×N Calculating the reconstruction error of each class to obtain a reconstruction error matrix R = [ R = [ R ] ] 1 ,R 2 ,…,R C ];
S5, preprocessing the face image vector data matrix X = [ X ] 1 ,X 2 ,…,X C ]Training data set X S And test data set X T Any data point x of j Repeating the steps S2, S3 and S4 to respectively obtain a training data set X S In (3) reconstructing the error matrix R s =[R S1 ,R S2 ,…,R SC ]And test data set X T Is reconstructed to the error matrix R T =[R T1 ,R T2 ,…,R TC ];
S6, constructing a target function based on the reconstruction error matrix to obtain a low-dimensional projection matrix A;
s7, based on the low-dimensional projection matrix, vector data points x after one image is preprocessed i Calculating its low-dimensional feature Y after linear projection i ;
S8, using a nearest neighbor algorithm to the low-dimensional feature Y i And classifying to realize the recognition of the human face characteristics.
2. The method for recognizing the human face based on the local linear representation domain adaptive metric of claim 1, wherein the preprocessing of the collected human face image in the step S1 comprises a graying process, a smoothing process, a normalization process and a vectorization process which are sequentially performed.
3. The method for face recognition based on adaptive measure in local linear representation domain of claim 1, wherein in step S2, the minimum linear representation mean square error function is:
wherein: x is the number of i Representing pre-processed vector data points, i.e. X, of an image c The (c) th column of (a),
x ij (j =1,2.., K) represents X c Vector data point x after preprocessing of one image i The K vector data points that are closest in euclidean distance,
W i,j weight coefficient representing j neighbors of the ith image, i.e. W c Each column of the ith row of (1).
5. The face recognition method based on local linear representation domain adaptation metric according to claim 1, wherein step S6 specifically comprises:
s61, applying a tiny perturbation to enable R c Becomes a positive definite matrix:
δ represents the disturbance magnitude;
s62, measuring different symmetrical positive definite matrixes R by using log-Euclidean distance c Distance between them, establishing divergence S between the manifolds M :
S63, training a data set X by using log-Euclidean distance measurement S And test data set X T Degree of dissimilarity S of d :
S64, establishing an objective function in a low-dimensional space:
wherein A ∈ R d×N A low-dimensional projection matrix is represented,
d is the spatial dimension after the dimension reduction,
log (-) represents the matrix logarithm based on e;
i denotes an identity matrix.
6. The face recognition method based on local linear representation domain adaptation metric of claim 1, wherein in step S6, the low-dimensional feature Y after linear projection i Expressed as:
Y i =A T X i (7)。
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201910628501.7A CN110399814B (en) | 2019-07-12 | 2019-07-12 | Face recognition method based on local linear representation field adaptive measurement |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201910628501.7A CN110399814B (en) | 2019-07-12 | 2019-07-12 | Face recognition method based on local linear representation field adaptive measurement |
Publications (2)
Publication Number | Publication Date |
---|---|
CN110399814A CN110399814A (en) | 2019-11-01 |
CN110399814B true CN110399814B (en) | 2023-03-24 |
Family
ID=68325341
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201910628501.7A Active CN110399814B (en) | 2019-07-12 | 2019-07-12 | Face recognition method based on local linear representation field adaptive measurement |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN110399814B (en) |
Families Citing this family (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN110956098B (en) * | 2019-11-13 | 2023-05-12 | 深圳数联天下智能科技有限公司 | Image processing method and related equipment |
CN111144240B (en) * | 2019-12-12 | 2023-02-07 | 深圳数联天下智能科技有限公司 | Image processing method and related equipment |
CN115019368B (en) * | 2022-06-09 | 2023-09-12 | 南京审计大学 | Face recognition feature extraction method in audit investigation |
Citations (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN103336960A (en) * | 2013-07-26 | 2013-10-02 | 电子科技大学 | Human face identification method based on manifold learning |
CN107463920A (en) * | 2017-08-21 | 2017-12-12 | 吉林大学 | A kind of face identification method for eliminating partial occlusion thing and influenceing |
WO2017219391A1 (en) * | 2016-06-24 | 2017-12-28 | 深圳市唯特视科技有限公司 | Face recognition system based on three-dimensional data |
-
2019
- 2019-07-12 CN CN201910628501.7A patent/CN110399814B/en active Active
Patent Citations (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN103336960A (en) * | 2013-07-26 | 2013-10-02 | 电子科技大学 | Human face identification method based on manifold learning |
WO2017219391A1 (en) * | 2016-06-24 | 2017-12-28 | 深圳市唯特视科技有限公司 | Face recognition system based on three-dimensional data |
CN107463920A (en) * | 2017-08-21 | 2017-12-12 | 吉林大学 | A kind of face identification method for eliminating partial occlusion thing and influenceing |
Also Published As
Publication number | Publication date |
---|---|
CN110399814A (en) | 2019-11-01 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN106295124B (en) | The method of a variety of image detecting technique comprehensive analysis gene subgraph likelihood probability amounts | |
Li et al. | Overview of principal component analysis algorithm | |
Yang et al. | Feature fusion: parallel strategy vs. serial strategy | |
CN110399814B (en) | Face recognition method based on local linear representation field adaptive measurement | |
MageshKumar et al. | Gabor features and LDA based face recognition with ANN classifier | |
CN108875459B (en) | Weighting sparse representation face recognition method and system based on sparse coefficient similarity | |
CN109063757A (en) | It is diagonally indicated based on block and the multifarious multiple view Subspace clustering method of view | |
CN110659665A (en) | Model construction method of different-dimensional features and image identification method and device | |
CN112488205A (en) | Neural network image classification and identification method based on optimized KPCA algorithm | |
CN104573672B (en) | A kind of discriminating kept based on neighborhood is embedded in face identification method | |
CN109190511B (en) | Hyperspectral classification method based on local and structural constraint low-rank representation | |
CN106980809B (en) | Human face characteristic point detection method based on ASM | |
CN106599833B (en) | A kind of face identification method adapted to based on field and manifold distance is measured | |
CN110516557A (en) | Multisample facial expression recognizing method based on low-rank tensor resolution | |
CN106503633A (en) | The method for building up in face characteristic storehouse in a kind of video image | |
CN108932501B (en) | Face recognition method based on multi-core association integration dimension reduction | |
CN113011243A (en) | Facial expression analysis method based on capsule network | |
CN112836671A (en) | Data dimension reduction method based on maximization ratio and linear discriminant analysis | |
CN108564061A (en) | A kind of image-recognizing method and system based on two-dimensional principal component analysis | |
Zuobin et al. | Feature regrouping for cca-based feature fusion and extraction through normalized cut | |
CN112966735B (en) | Method for fusing supervision multi-set related features based on spectrum reconstruction | |
CN111611963B (en) | Face recognition method based on neighbor preservation canonical correlation analysis | |
Zhang et al. | Compound structure classifier system for ear recognition | |
CN112465062A (en) | Clustering method based on manifold learning and rank constraint | |
CN111950429A (en) | Face recognition method based on weighted collaborative representation |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
PB01 | Publication | ||
PB01 | Publication | ||
SE01 | Entry into force of request for substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
GR01 | Patent grant | ||
GR01 | Patent grant |