CN110399814A - It is a kind of to indicate that field adapts to the face identification method of measurement based on local linear - Google Patents

It is a kind of to indicate that field adapts to the face identification method of measurement based on local linear Download PDF

Info

Publication number
CN110399814A
CN110399814A CN201910628501.7A CN201910628501A CN110399814A CN 110399814 A CN110399814 A CN 110399814A CN 201910628501 A CN201910628501 A CN 201910628501A CN 110399814 A CN110399814 A CN 110399814A
Authority
CN
China
Prior art keywords
matrix
indicate
vector data
data point
pretreated
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Granted
Application number
CN201910628501.7A
Other languages
Chinese (zh)
Other versions
CN110399814B (en
Inventor
李波
田逸辉
李颜瑞
张晓龙
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.)
Wuhan University of Science and Engineering WUSE
Original Assignee
Wuhan University of Science and Engineering WUSE
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 Wuhan University of Science and Engineering WUSE filed Critical Wuhan University of Science and Engineering WUSE
Priority to CN201910628501.7A priority Critical patent/CN110399814B/en
Publication of CN110399814A publication Critical patent/CN110399814A/en
Application granted granted Critical
Publication of CN110399814B publication Critical patent/CN110399814B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Classifications

    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06VIMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V40/00Recognition of biometric, human-related or animal-related patterns in image or video data
    • G06V40/10Human or animal bodies, e.g. vehicle occupants or pedestrians; Body parts, e.g. hands
    • G06V40/16Human faces, e.g. facial parts, sketches or expressions
    • G06V40/161Detection; Localisation; Normalisation
    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06VIMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V40/00Recognition of biometric, human-related or animal-related patterns in image or video data
    • G06V40/10Human or animal bodies, e.g. vehicle occupants or pedestrians; Body parts, e.g. hands
    • G06V40/16Human faces, e.g. facial parts, sketches or expressions
    • G06V40/168Feature extraction; Face representation
    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06VIMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V40/00Recognition of biometric, human-related or animal-related patterns in image or video data
    • G06V40/10Human or animal bodies, e.g. vehicle occupants or pedestrians; Body parts, e.g. hands
    • G06V40/16Human faces, e.g. facial parts, sketches or expressions
    • G06V40/172Classification, 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 kind of to indicate that field adapts to the face identification method of measurement based on local linear, technical solution is: vector data point X pretreated for a width facial imageiSelection and the pretreated vector data point X of a width facial image in matrix data X are constituted from the pretreated vector data point of all width facial images respectivelyiThe identical K Neighbor Points of classification carry out local linear expression, obtaining local linear indicates coefficient, establish the Partial Reconstruction error matrix indicated based on local linear, then the divergence between different classes of multiple manifold is measured with logarithm Euclidean distance, simultaneously, also using the diversity between logarithm Euclidean distance measurement training data domain and test data domain, finally by the divergence maximized between multiple manifold, minimize the difference between source data and target data, it finds low-dimensional and differentiates subspace, realize that facial image differentiates feature extraction.The present invention extracts the characteristic of division of facial image by maximizing the similitude that multiple manifold divergence and training, test data are distributed, and has the characteristics that improve facial image recognition effect.

Description

It is a kind of to indicate that field adapts to the face identification method of measurement based on local linear
Technical field
The invention belongs to technical field of face recognition, indicate that field adapts to measurement based on local linear more particularly to one kind Face identification method.
Background technique
Recognition of face is one of machine vision biological identification major product form, is the indispensabilities such as public safety, daily life Technology.In recent years, face recognition application range is continuously increased, and requires high accuracy.And in its equipment imaging process, Illumination, block and posture label etc. due to, cause face recognition algorithms that can not effectively extract feature, these defective effects Algorithm discrimination reduces the effect of face recognition technology in actual use, or even brings security risk.Therefore recognition of face is calculated Method stability is particularly important.
Face recognition algorithms experienced multiple developing stage.The characterization for such as obtaining low dimensional, the face based on local feature Identification, the partial descriptions submethod based on study.There is the deep learning method based on deep neural network again at present, this method is known Other ability is strong, robust is good, but structure is complicated, and calculation amount is very big.How to obtain that a kind of recognition capability is strong, robust is good, with constantly Between space complexity it is smaller, for the method that a small amount of training set sample still maintains preferable recognition effect, become recognition of face Hot topic.
Summary of the invention
The present invention is directed to overcome the shortage of prior art, it is therefore an objective to provide it is a kind of can improve model recognition capability and robustness, A kind of local linear for reducing time space complexity indicates that field adapts to the face identification method of measurement.
To realize that above-mentioned technical proposal, the technical solution adopted by the present invention comprise the concrete steps that:
S1, the facial image of acquisition is pre-processed, obtains pretreated facial image vector data matrix X= [X1, X2..., XC], in which: C is classification number,M is the characteristic dimension after vectorization, NcIt is c-th of class Image category quantity, 1≤c≤C;
S2, for c-th of class Xc, from XcThe middle pretreated vector data point x of selection piece imagei, choose and a width figure As pretreated vector data point xiK nearest vector data point x of Euclidean distancei1, xi2..., xiK, pre- to piece image Treated vector data point xiLocal linear expression is carried out, linear expression mean square error function is minimized, obtains minimal linear Indicate coefficient vector Wc, by weight coefficient vector WcN-dimensional is extended to, wherein minimal linear between the vector point with neighbor relationships Indicate that coefficient is constant, other are 0;
S3, to the data point of remaining class in image vector data matrix X, repeat the above steps, obtain the power of all categories Weight coefficient matrix W=[W1, W2..., WC]∈RN×N, wherein N is the total number of all pretreated images,
S4, it is based on calculated weight coefficient matrix W=[W1, W2..., WC]∈RN×N, the reconstruction error of each class is calculated, Obtain reconstruction error matrix R=[R1, R2..., RC];
S5, to pretreated facial image vector data matrix X=[X1, X2..., XC] training dataset XSAnd survey Try data set XTAny data point xj, step S2, S3, S4 are repeated, training dataset X is respectively obtainedSIn reconstructed error matrix RS=[RS1, RS2..., RSc] and test data set XTReconstructed error matrix RT=[RT1, RT2..., RTc];
S6, it is based on the reconstruction error matrix, constructs objective function, obtains low dimension projective matrix A;
S7, the low dimension projective matrix, vector data point x pretreated for piece image are based oni, calculate its process Low-dimensional characteristic Y i after linear projection;
S8, using nearest neighbor algorithm to the low-dimensional characteristic YiClassify, realizes the identification of face characteristic.
Further, according to claim 1 to indicate that field adapts to the recognition of face side of measurement based on local linear Method, which is characterized in that carrying out pretreatment to the facial image of acquisition in step S1 includes that the gray processing successively carried out is handled, smoothly Processing, normalized and vectorization processing.
Further, according to claim 1 to indicate that field adapts to the recognition of face side of measurement based on local linear Method, which is characterized in that in step S2, the minimal linear indicates mean square error function are as follows:
Wherein: xiIndicate the pretreated vector data point of piece image, i.e. XcI-th column,
xij(j=1,2 ..., K) indicates XcIn with the pretreated vector data point x of piece imageiEuclidean distance is nearest K vector data point,
WI, jIndicate the weight coefficient of j Neighbor Points of the i-th width image, i.e. WcIn the i-th row each column.
Further, according to claim 1 to indicate that field adapts to the recognition of face side of measurement based on local linear Method, which is characterized in that in step S4, the reconstruction error matrix of each class is expressed as:
Wherein: the transposition operation of T representing matrix;
I is and WcThe identical unit matrix of dimension.
Further, according to claim 1 to indicate that field adapts to the recognition of face side of measurement based on local linear Method, which is characterized in that step S6 is specifically included:
One S61, application small disturbance make RcAs positive definite matrix:
Wherein: IcBe withThe identical unit matrix of quasi- number;
δ indicates disturbance size;
S62, log-Euclidean distance metric difference symmetric positive definite matrix R is usedcThe distance between, establish multiple manifold Between divergence SM:
S63, log-Euclidean distance metric training dataset X is usedSWith test data set XTDistinctiveness ratio Sd:
S64, in lower dimensional space, establish objective function:
s.t.ATA=I
Wherein, A ∈ Rd×NIndicate low dimension projective matrix,
D is the Spatial Dimension after dimensionality reduction,
Log () is indicated using e as the logm operation at bottom;
I indicates unit matrix.
Further, according to claim 1 to indicate that field adapts to the recognition of face side of measurement based on local linear Method, which is characterized in that the low-dimensional characteristic Y in step S6, after linear projectioniIt indicates are as follows:
Yi=ATXi (7)
Due to the adoption of the above technical scheme, the beneficial effects of the present invention are:
The present invention indicates that field adapts to the face identification method of measurement using local linear, on the one hand for a width face figure As pretreated vector data point Xi, matrix data is constituted from the pretreated vector data point of all width facial images respectively In X, selection and the pretreated vector data point X of a width facial imagei, it is local that the identical K Neighbor Points of classification establish manifold Linear expression, then with log-Euclidean distance metric manifold local linear indicate between multiple manifold divergence, keep flow On the basis of shape partial structurtes are constant, multiple manifold divergence is maximized to find low dimension projective matrix A, on the other hand, in order to guarantee Similitude in space in the projected between source data domain and target data domain minimizes the covariance distance of the two, with this The differentiation feature extraction for realizing facial image, improves the effect of recognition of face.
Therefore, the present invention has identification energy by maximizing the similarity feature between source data domain and target data domain Power is strong, robust is good, while time space complexity is smaller, and the spy of preferable recognition effect is still maintained for a small amount of training set sample Point.
Specific embodiment
The invention will be further described With reference to embodiment, not to the limitation of its protection scope.
Embodiment 1
It is a kind of to indicate that field adapts to the face identification method of measurement based on local linear.The present embodiment the method it is specific Step is:
Step 1: at the gray processing processing, smoothing processing, normalized and the vectorization that are carried out to the facial image of acquisition Reason, obtains pretreated facial image vector data matrix X=[X1, X2..., XC], in which: C is classification number,M is the characteristic dimension after vectorization, NcIt is the image category quantity of c-th of class, 1≤c≤C;
Step 2: for c-th of class Xc, from XcThe middle pretreated vector data point x of selection piece imagei, choose and one Vector data point x after width image preprocessingiK nearest vector data point x of Euclidean distancei1, xi2..., xiK, to a width figure As pretreated vector data point xiLocal linear expression is carried out, linear expression mean square error function, minimal linear are minimized Indicate mean square error function are as follows:
Wherein: xiIndicate the pretreated vector data point of piece image, i.e. XcI-th column,
xij(j=1,2 ..., K indicate XcIn with the pretreated vector data point x of piece imageiEuclidean distance is nearest K vector data point,
WI, jIndicate the weight coefficient of j Neighbor Points of the i-th width image, i.e. WcIn the i-th row each column.
Minimal linear, which is calculated, indicates coefficient vector Wc, by weight coefficient vector WcN-dimensional is extended to, wherein having neighbour Minimal linear indicates that coefficient is constant between the vector point of relationship, other are 0;
Step 3: repeating the above steps to the data point of remaining class in image vector data matrix X, obtaining all categories Weight coefficient matrix W=[W1, W2..., WC]∈RN×N, wherein N is the total number of all pretreated images,
Step 4: being based on calculated weight coefficient matrix W=[W1, W2..., WC]∈RN×N, the reconstruction error of each class Matrix is expressed as:
Wherein: the transposition operation of T representing matrix;
I is and WcThe identical unit matrix of dimension.
The reconstruction error for calculating each class obtains reconstruction error matrix R=[R1, R2..., RC];
Step 5: to pretreated facial image vector data matrix, X=[X1, X2..., XC] training dataset XS With test data set XTAny data point xj, step S2, S3, S4 are repeated, training dataset X is respectively obtainedSIn reconstructed error Matrix Rs=[RS1, RS2..., RSC] and test data set XTReconstructed error matrix RT=[RT1, RT2..., RTC];
Step 6: applying a small disturbance makes RcAs positive definite matrix:
Wherein: IcBe withThe identical unit matrix of dimension;
δ indicates disturbance size;
Then log-Euclidean distance metric difference symmetric positive definite matrix R is usedcThe distance between, establish multiple manifold Between divergence SM:
Use log-Euclidean distance metric training dataset XSWith test data set XTDistinctiveness ratio Sd:
In lower dimensional space, objective function is established:
Wherein, A ∈ Rd×NIndicate low dimension projective matrix,
D is the Spatial Dimension after dimensionality reduction,
Log () is indicated using e as the logm operation at bottom;
I indicates unit matrix.
It maximizes objective function and obtains matrix A.
Step 7: being based on the low dimension projective matrix, vector data point x pretreated for piece imagei, by line Property projection after low-dimensional characteristic YiIt indicates are as follows:
Yi=ATXi (7)
Calculate its low-dimensional characteristic Y after linear projectioni
Step 8: using nearest neighbor algorithm to the low-dimensional characteristic YiClassify, realizes the identification of face characteristic.
The beneficial effect of present embodiment is:
The present invention, which is used, indicates that field adapts to the face identification method of measurement based on local linear, on the one hand for a width people The pretreated vector data point X of face imagei, matrix is constituted from the pretreated vector data point of all width facial images respectively In data X, selection and the pretreated vector data point X of a width facial imagei, the identical K Neighbor Points of classification establish manifold Local linear indicate, then with log-Euclidean distance metric manifold local linear indicate between multiple manifold divergence, protecting Hold manifold partial structurtes it is constant on the basis of, maximize multiple manifold divergence and find low dimension projective matrix A, on the other hand, in order to Guarantee the similitude in space in the projected between source data domain and target data domain, minimize the covariance distance of the two, The differentiation feature extraction that facial image is realized with this, improves the effect of recognition of face.
Therefore, the present invention has identification energy by maximizing the similarity feature between source data domain and target data domain Power is strong, robust is good, while time space complexity is smaller, and the spy of preferable recognition effect is still maintained for a small amount of training set sample Point.

Claims (6)

1. a kind of indicate that field adapts to the face identification method of measurement based on local linear, which comprises the steps of:
S1, the facial image of acquisition is pre-processed, obtains pretreated facial image vector data matrix X=[X1, X2..., XC], in which: C is classification number,M is the characteristic dimension after vectorization, NcIt is the image of c-th of class Categorical measure, 1≤c≤C;
S2, for c-th of class Xc, from XcThe middle pretreated vector data point x of selection piece imagei, choose pre- with piece image Treated vector data point xiK nearest vector data point x of Euclidean distancei1, xi2..., xiK, piece image is pre-processed Vector data point x afterwardsiLocal linear expression is carried out, linear expression mean square error function is minimized, obtains minimal linear expression Coefficient vector Wc, by weight coefficient vector WcN-dimensional is extended to, wherein minimal linear indicates between the vector point with neighbor relationships Coefficient is constant, other are 0;
S3, to the data point of remaining class in image vector data matrix X, repeat the above steps, obtain the weight system of all categories Matrix number W=[W1, W2..., WC]∈RN×N, wherein N is the total number of all pretreated images,
S4, it is based on calculated weight coefficient matrix W=[W1, W2..., WC]∈RN×N, the reconstruction error of each class is calculated, is obtained Reconstruction error matrix R=[R1, R2..., RC];
S5, to pretreated facial image vector data matrix X=[X1, X2..., XC] training dataset XSAnd test data Collect XTAny data point xj, step S2, S3, S4 are repeated, training dataset X is respectively obtainedSIn reconstructed error matrix Rs= [RS1, RS2..., RSC] and test data set XTReconstructed error matrix RT=[RT1, RT2..., RTC];
S6, it is based on the reconstruction error matrix, constructs objective function, obtains low dimension projective matrix A;
S7, the low dimension projective matrix, vector data point x pretreated for piece image are based oni, it is calculated by linear Low-dimensional characteristic Y after projectioni
S8, using nearest neighbor algorithm to the low-dimensional characteristic YiClassify, realizes the identification of face characteristic.
2. according to claim 1 indicate that field adapts to the face identification method of measurement based on local linear, feature exists In carrying out pretreatment to the facial image of acquisition in step S1 includes the gray processing processing successively carried out, smoothing processing, normalization Processing and vectorization processing.
3. according to claim 1 indicate that field adapts to the face identification method of measurement based on local linear, feature exists In in step S2, the minimal linear indicates mean square error function are as follows:
Wherein: xiIndicate the pretreated vector data point of piece image, i.e. XcI-th column,
xij(j=1,2 ..., K) indicates XcIn with the pretreated vector data point x of piece imageiEuclidean distance nearest K Vector data point,
WI, jIndicate the weight coefficient of j Neighbor Points of the i-th width image, i.e. WcIn the i-th row each column.
4. according to claim 1 indicate that field adapts to the face identification method of measurement based on local linear, feature exists In in step S4, the reconstruction error matrix of each class is expressed as:
Wherein: the transposition operation of T representing matrix;
I is and WcThe identical unit matrix of dimension.
5. according to claim 1 indicate that field adapts to the face identification method of measurement based on local linear, feature exists In step S6 is specifically included:
One S61, application small disturbance make RcAs positive definite matrix:
Wherein: IcBe withThe identical unit matrix of dimension;
δ indicates disturbance size;
S62, log-Euclidean distance metric difference symmetric positive definite matrix R is usedcThe distance between, it establishes between multiple manifold Divergence SM:
S63, log-Euclidean distance metric training dataset X is usedSWith test data set XTDistinctiveness ratio Sd:
S64, in lower dimensional space, establish objective function:
s.t.ATA=I
Wherein, A ∈ Rd×NIndicate low dimension projective matrix,
D is the Spatial Dimension after dimensionality reduction,
Log () is indicated using e as the logm operation at bottom;
I indicates unit matrix.
6. according to claim 1 indicate that field adapts to the face identification method of measurement based on local linear, feature exists In low-dimensional characteristic Y in step S6, after linear projectioniIt indicates are as follows:
Yi=ATXi。 (7)
CN201910628501.7A 2019-07-12 2019-07-12 Face recognition method based on local linear representation field adaptive measurement Active CN110399814B (en)

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 true CN110399814A (en) 2019-11-01
CN110399814B 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)

Cited By (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN110956098A (en) * 2019-11-13 2020-04-03 深圳和而泰家居在线网络科技有限公司 Image processing method and related equipment
CN111144240A (en) * 2019-12-12 2020-05-12 深圳数联天下智能科技有限公司 Image processing method and related equipment
CN115019368A (en) * 2022-06-09 2022-09-06 南京审计大学 Face recognition feature extraction method in audit investigation based on 2DESDLPP

Citations (3)

* Cited by examiner, † Cited by third party
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

Patent Citations (3)

* Cited by examiner, † Cited by third party
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

Cited By (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN110956098A (en) * 2019-11-13 2020-04-03 深圳和而泰家居在线网络科技有限公司 Image processing method and related equipment
CN111144240A (en) * 2019-12-12 2020-05-12 深圳数联天下智能科技有限公司 Image processing method and related equipment
CN115019368A (en) * 2022-06-09 2022-09-06 南京审计大学 Face recognition feature extraction method in audit investigation based on 2DESDLPP
CN115019368B (en) * 2022-06-09 2023-09-12 南京审计大学 Face recognition feature extraction method in audit investigation

Also Published As

Publication number Publication date
CN110399814B (en) 2023-03-24

Similar Documents

Publication Publication Date Title
CN105956582B (en) A kind of face identification system based on three-dimensional data
Zuo et al. Bidirectional PCA with assembled matrix distance metric for image recognition
CN100375108C (en) Automatic positioning method for characteristic point of human faces
CN101777131B (en) Method and device for identifying human face through double models
CN105447441B (en) Face authentication method and device
Li et al. Overview of principal component analysis algorithm
CN103279768B (en) A kind of video face identification method based on incremental learning face piecemeal visual characteristic
CN110399814A (en) It is a kind of to indicate that field adapts to the face identification method of measurement based on local linear
CN102930300B (en) Method and system for identifying airplane target
Fellenz et al. Comparing template-based, feature-based and supervised classification of facial expressions from static images
CN106599833B (en) A kind of face identification method adapted to based on field and manifold distance is measured
Huang et al. Complete local Fisher discriminant analysis with Laplacian score ranking for face recognition
CN108564061A (en) A kind of image-recognizing method and system based on two-dimensional principal component analysis
CN109766748A (en) A kind of pedestrian based on projective transformation and dictionary learning knows method for distinguishing again
CN103714340A (en) Self-adaptation feature extracting method based on image partitioning
CN108520201A (en) A kind of robust human face recognition methods returned based on weighted blend norm
Tao et al. Frontal-centers guided face: Boosting face recognition by learning pose-invariant features
CN111611963B (en) Face recognition method based on neighbor preservation canonical correlation analysis
Wang et al. Action recognition using linear dynamic systems
Goncharova et al. Greedy algorithms of feature selection for multiclass image classification
Ming et al. Efficient 3d face recognition with Gabor patched spectral regression
CN110399885B (en) Image target classification method based on local geometric perception
Lei et al. Student action recognition based on multiple features
CN112633229A (en) Pedestrian re-identification system based on SPD manifold
Wang et al. Bayesian face recognition based on markov random field modeling

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