CN106446806B - Semi-supervised face identification method based on the sparse reconstruct of fuzzy membership and system - Google Patents
Semi-supervised face identification method based on the sparse reconstruct of fuzzy membership and system Download PDFInfo
- Publication number
- CN106446806B CN106446806B CN201610811185.3A CN201610811185A CN106446806B CN 106446806 B CN106446806 B CN 106446806B CN 201610811185 A CN201610811185 A CN 201610811185A CN 106446806 B CN106446806 B CN 106446806B
- Authority
- CN
- China
- Prior art keywords
- fuzzy membership
- matrix
- test sample
- class label
- sparse
- 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.)
- Expired - Fee Related
Links
- 238000000034 method Methods 0.000 title claims abstract description 46
- 238000012360 testing method Methods 0.000 claims abstract description 135
- 239000011159 matrix material Substances 0.000 claims abstract description 112
- 238000012549 training Methods 0.000 claims abstract description 70
- 238000013480 data collection Methods 0.000 claims abstract description 43
- 230000001815 facial effect Effects 0.000 claims description 16
- 239000000470 constituent Substances 0.000 claims 1
- 230000006870 function Effects 0.000 description 8
- 238000005516 engineering process Methods 0.000 description 6
- 238000011524 similarity measure Methods 0.000 description 5
- 230000008569 process Effects 0.000 description 4
- 230000008901 benefit Effects 0.000 description 3
- 238000002474 experimental method Methods 0.000 description 3
- 230000006698 induction Effects 0.000 description 3
- 238000005259 measurement Methods 0.000 description 3
- 238000012986 modification Methods 0.000 description 3
- 230000004048 modification Effects 0.000 description 3
- 230000000644 propagated effect Effects 0.000 description 3
- 230000005540 biological transmission Effects 0.000 description 2
- 238000010276 construction Methods 0.000 description 2
- 238000002790 cross-validation Methods 0.000 description 2
- 238000010586 diagram Methods 0.000 description 2
- 230000000694 effects Effects 0.000 description 2
- 230000014509 gene expression Effects 0.000 description 2
- 230000009286 beneficial effect Effects 0.000 description 1
- 238000004364 calculation method Methods 0.000 description 1
- 230000008859 change Effects 0.000 description 1
- 238000004590 computer program Methods 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 235000013399 edible fruits Nutrition 0.000 description 1
- 238000000691 measurement method Methods 0.000 description 1
- 238000005457 optimization Methods 0.000 description 1
- 238000003909 pattern recognition Methods 0.000 description 1
- 238000012545 processing Methods 0.000 description 1
- 230000001902 propagating effect Effects 0.000 description 1
- 238000011160 research Methods 0.000 description 1
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
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/40—Extraction of image or video features
- G06V10/513—Sparse representations
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
The invention discloses a kind of semi-supervised face identification method and system based on the sparse reconstruct of fuzzy membership, the recognition methods include obtaining face image data collection;Face image data collection includes training sample subset and test sample subset;According to class label known to training sample, the fuzzy membership initialization matrix of training sample subset is obtained;Reconstructed residual according to test sample about class label, the fuzzy membership for obtaining test sample subset initialize matrix;The fuzzy membership that face image data collection is further obtained about class label initializes matrix;The sparse coefficient of all test samples is solved, and then obtains the sparse dematrix of test sample subset;Matrix and the sparse dematrix of test sample subset, fuzzy membership matrix of the iterative solution update face image data collection about class label are initialized according to fuzzy membership;It obtains the class label corresponding to the maximum membership degree of each test sample and classification can be completed.
Description
Technical field
The invention belongs to area of pattern recognition more particularly to a kind of semi-supervised faces based on the sparse reconstruct of fuzzy membership
Recognition methods and system.
Background technology
With the fast development of computer technology and image processing techniques, recognition of face is obtained since it is widely applied
The concern of numerous researchers becomes an importance in modern mode identification technology research.
However, label facial image is cumbersome and time consuming, it is contemplated that unmarked facial image simple and easy to get, it can be abundant
Using the label information in label facial image and characteristic information in unmarked facial image to unmarked facial image into
Row classification, this Learning Scheme are known as semi-supervised study.
Wherein the semi-supervised learning based on figure is one type Learning Scheme, and the semi-supervised learning point of use based on figure indicates
Data, side indicate the weight between data and respective classes.The probability that the larger class of data weights belongs to such is more than other classes,
Image classification can be carried out on the basis of this assumption, and the semi-supervised learning based on figure meets manifold it is assumed that and in the sheet of data
It levies lower dimensional space and realizes that effect is fine.
The effect of semi-supervised learning based on figure depends primarily on the definition of the similarity between data pair.One of the most common
Similarity Measures are the measures based on gaussian kernel function of theorem in Euclid space, and the scheme based on this method has LNP (linear
Neighbour propagates) it can be gone to calculate linear space by its approximated linear reconstruction of k neighbour based on each data of this hypothesis-
Data similarity;SIS (sparse induction similarity measure) goes to calculate sparse similarity, SIS and LNP phases using rarefaction representation technology
Neighbour's number need not be pre-defined when than using sparse complete coding excessively.
In view of rarefaction representation technology can be to avoid the problem of parameter selection of KNN (k neighbours) and ε-ball.SIS is (sparse
Induction similarity measure) it is used for marking propagation using the method for sparse induction similarity measurement, and redefine similarity measurement side
Method, this method go sparse linear to reconstruct data itself, the sparse coefficient that will be obtained using the remaining data of some data itself
As the similarity between data pair;LPSN (label propagates sparse neighbour) goes to obtain in luv space using rarefaction representation technology
Sparse coding, and sparse coding used in the reconstruct of data label;MLRR (manifold low-rank representation) uses the low-rank enhanced
Sparse data representation method carries out realizing data similarity measurement.
The above several method can fully obtain part and global feature between data, but data is sparse heavy
Structure residual error is ignored, and the residual error of data is the key factor of the semi-supervised learning based on rarefaction representation, can be dropped in this way
The classification and recognition of low semi-supervised method.
Invention content
In order to solve the disadvantage that the prior art, the present invention provide a kind of semi-supervised people based on the sparse reconstruct of fuzzy membership
Face recognition method and system.The method of the present invention has used membership function and has iterated to calculate test data using sparse reconstructed residual
Degree of membership, and maintain the manifold structure of data.
A kind of semi-supervised face identification method based on the sparse reconstruct of fuzzy membership, including:
Obtain face image data collection;Face image data collection includes training sample subset and test sample subset;Training
The sample of sample set is facial image known to class label, and the sample of test sample subset is the unknown face of class label
Image;
According to class label known to training sample, the fuzzy membership initialization matrix of training sample subset is obtained;Meter
Reconstructed residual of the test sample about class label is calculated, the fuzzy membership for obtaining test sample subset initializes matrix;Into one
Step obtains face image data collection and initializes matrix about the fuzzy membership of class label;
The sparse coefficient of all test samples is solved, and then obtains the sparse dematrix of test sample subset;
Matrix and the sparse dematrix of test sample subset, iterative solution update face are initialized according to fuzzy membership
Fuzzy membership matrix of the image data set about class label;
Obtain the classification that test sample can be completed in the class label corresponding to the maximum membership degree of each test sample.
The present invention is carried out using the fuzzy membership function and reconstructed residual usually ignored is put into membership function
The degree of membership for calculating test sample, data structure is implicitly kept in manifold space, fully consider part between data and
Global characteristics improve image recognition accuracy.
During the fuzzy membership for obtaining training sample subset initializes matrix, classification mark is predefined first
Label, then whether the training sample in training of judgement sample set belongs to predetermined class label, if so, corresponding training
Sample fuzzy membership is 1;Otherwise, corresponding training sample fuzzy membership is 0.
In the fuzzy membership initialization matrix of test sample subset, test sample is about the other fuzzy membership of tag class
Make the ratio of quotient for two numbers;Wherein, divisor is inverse of the test sample about respective labels classification residual error, and dividend is test
Tired reciprocal about the other reconstructed residual of all tag class of sample and.
During iterative solution updates fuzzy membership matrix of the face image data collection about class label,
According to the sparse dematrix of test sample subset, reconstructed residual of each test sample about class label is calculated;
Reconstructed residual according to each test sample about class label obtains fuzzy membership after the update of test sample
Degree, further obtains the updated fuzzy membership matrix of test sample subset;
Training sample subset fuzzy membership initialization matrix it is constant, merge training sample subset fuzzy membership and
The updated fuzzy membership matrix of test sample subset, obtains updated mould of the face image data collection about class label
Paste subordinated-degree matrix.
The reconstructed residual usually ignored is put into the degree of membership for carrying out calculating test sample in membership function by the present invention,
Manifold implicitly keeps data structure in space, fully considers the part between data and global characteristics.The method of the present invention uses
Membership function and using the degree of membership of sparse reconstructed residual iterative calculation test data, and maintain the manifold knot of data
Structure.
During iterative solution updates fuzzy membership matrix of the face image data collection about class label, if full
This fuzzy membership matrix sought of foot makees the institute in the matrix obtained after difference with the preceding fuzzy membership matrix once sought
There is the absolute value of element to be respectively less than predetermined threshold value, then iteration terminates;Or iterations are more than maximum iterations, then iteration
Terminate.
A kind of semi-supervised face identification system based on the sparse reconstruct of fuzzy membership, including recognition of face server, institute
Stating recognition of face server includes:
Face image data collection acquisition module, is used to obtain face image data collection;Face image data collection includes instruction
Practice sample set and test sample subset;The sample of training sample subset is facial image known to class label, test sample
The sample of subset is the unknown facial image of class label;
Fuzzy membership initializes matrix initialisation module, is used to, according to class label known to training sample, obtain
The fuzzy membership of training sample subset initializes matrix;Reconstructed residual of the test sample about class label is calculated, is surveyed
The fuzzy membership for trying sample set initializes matrix;Further obtain fuzzy person in servitude of the face image data collection about class label
Category degree initializes matrix;
Sparse solution Matrix Solving module, is used to solve the sparse coefficient of all test samples, and then obtains test sample
The sparse dematrix of subset;
Fuzzy membership matrix update module is used to initialize matrix and test sample subset according to fuzzy membership
Sparse dematrix, fuzzy membership matrix of the iterative solution update face image data collection about class label;
Class label acquisition module is used to obtain class label corresponding to the maximum membership degree of each test sample i.e.
It can complete the classification of test sample.
The fuzzy membership initialization matrix initialisation module includes the fuzzy membership initialization of training sample subset
Matrix initialisation module is used to predefine class label first, and then the training sample in training of judgement sample set is
It is no to belong to predetermined class label, if so, corresponding training sample fuzzy membership is 1;Otherwise, corresponding training sample mould
It is 0 to paste degree of membership.
The fuzzy membership initialization matrix initialisation module further includes that the fuzzy membership of test sample subset is initial
Change matrix initialisation module, is used to two numbers seeking test sample about the other fuzzy membership of tag class as quotient;Wherein,
Divisor is inverse of the test sample about respective labels classification residual error, and dividend is that test sample is other heavy about all tag class
Reciprocal tired of structure residual error and.
The fuzzy membership matrix update module is additionally operable to the sparse dematrix according to test sample subset, calculates each
Reconstructed residual of the test sample about class label;
Reconstructed residual according to each test sample about class label obtains fuzzy membership after the update of test sample
Degree, further obtains the updated fuzzy membership matrix of test sample subset;
Training sample subset fuzzy membership initialization matrix it is constant, merge training sample subset fuzzy membership and
The updated fuzzy membership matrix of test sample subset, obtains updated mould of the face image data collection about class label
Paste subordinated-degree matrix.
The semi-supervised face identification system based on the sparse reconstruct of fuzzy membership further includes display device, the display
Device is connected with recognition of face server;The display device is for showing corresponding to the maximum membership degree of each test sample
Class label.
The beneficial effects of the invention are as follows:
The present invention takes full advantage of the characteristics of image of global and local and by sparse residual error for obscuring sparse degree of membership
In calculating.Since face digital picture itself has manifold structure, reconstructed residual is combined solution with data fuzzy membership
During data membership degree can propagate reconstructed residual in view of label, consider the label of reconstructed residual than only consider sparse coding or
Person only considers that the label of data similarity is propagated to have and more judges information, therefore can improve image recognition accuracy.
Description of the drawings
Fig. 1 is the semi-supervised face identification method flow chart based on the sparse reconstruct of fuzzy membership of the present invention;
Fig. 2 is initialization sample data fuzzy membership matrix construction schematic diagram;
Fig. 3 is that semi-supervised one structure of face identification system embodiment based on the sparse reconstruct of fuzzy membership of the present invention is shown
It is intended to;
Fig. 4 is that semi-supervised two structure of face identification system embodiment based on the sparse reconstruct of fuzzy membership of the present invention is shown
It is intended to.
Specific implementation mode
The present invention will be further described with embodiment below in conjunction with the accompanying drawings:
The semi-supervised recognition methods for the face digital picture based on figure that the present invention provides a kind of.Traditional based on figure
In semi-supervised method or only use simple similarity measurement method or when executing label and propagating at one point
Two kinds of structure features of global and local of data are simply considered in generic task.Scheme proposed by the present invention uses fuzzy membership
Function and be put into the reconstructed residual usually ignored carries out calculating the degree of membership of test sample in membership function, in manifold sky
Between in implicitly keep data structure, fully consider the part between data and global characteristics.
The thinking of semi-supervised recognition methods of the face digital picture based on figure of the present invention is:
First, the subordinated-degree matrix about all corresponding class labels of all samples is initialized;
Then sparse reconstructed residual of the test sample about all categories label is calculated;
The person in servitude for re-defining the method for iterative modifications degree of membership and obtaining test sample about all categories by iterative solution
Category degree matrix obtains the classification that test sample can be completed in the class label corresponding to the maximum membership degree of each test sample.
Fig. 1 is the semi-supervised face identification method flow chart based on the sparse reconstruct of fuzzy membership of the present invention, as schemed institute
The semi-supervised face identification method based on the sparse reconstruct of fuzzy membership shown, including:
Step 1:Obtain face image data collection;Face image data collection includes training sample subset and test specimens book
Collection;The sample of training sample subset is facial image known to class label, the sample of test sample subset be class label not
The facial image known.
In specific implementation process:
Obtain face image data collection X=[Xl Xu], each sample is xi∈Rq{ i=1,2 ..., n }, n is sample
Number, sample dimension are q, whereinFor training sample set,For test sample
Collection, nlFor training sample number, nuFor test sample number;
Step 2:According to class label known to training sample, the fuzzy membership initialization square of training sample subset is obtained
Battle array;Reconstructed residual of the test sample about class label is calculated, the fuzzy membership for obtaining test sample subset initializes matrix;
The fuzzy membership that face image data collection is further obtained about class label initializes matrix.
Fig. 2 is initialization sample data fuzzy membership matrix construction schematic diagram, and specific method is:
(1) fuzzy membership matrix is initialized to training sample and test sample respectively first, defines category set C=
{ 1,2 ..., c } and training sample set fuzzy membership matrixWherein for training sample set data xi
∈Xl, c is the positive integer more than 1;The fuzzy membership matrix M of following formula initialization training sample can be used0l, wherein mij0For
Fuzzy membership matrix M0lElement:
Wherein mij0Indicate that the degree of membership of the jth class of i-th of training sample when the 0th iteration, p are certain classifications.
(2) for test sample collection data xk∈Xu, the fuzzy membership initialization of test sample can be obtained according to the following formula
Matrix M0u, wherein mkj0For fuzzy membership matrix M0lElement:
Wherein mkj0Indicate k-th of test sample x when the 0th iterationkJth class degree of membership, errkj0It is k-th
Test sample xkCorresponding jth class sample set XjInitial reconstitution residual error, errkjo=| | xk-Xjα′||2;Jth class sample set XjInterior
The class label of sample is jth class;α ' is k-th of test sample xkRelative to jth class sample set XjSparse coefficient.
In specific implementation process, k-th of test sample x is solved using following formulakRelative to jth class sample set XjIt is sparse
Coefficient:
min||xk-Xjα′||2+γ′||α′||1(k∈1,2,....nu;j∈C)
It is constant coefficient that wherein γ ' expressions prevent over-fitting coefficient, γ ' to the constraint of α '.
(3) merge M0lAnd M0uThe fuzzy membership that face image data collection is obtained about class label initializes matrix
M0:
Wherein
Step 3:The sparse coefficient of all test samples is solved, and then obtains the sparse dematrix of test sample subset.
In specific implementation process, to each test sample xk, sparse coefficient is solved using following formula:
min||xk-Xα||2+γ||α1||(k∈1,2,...,nu)
Wherein xkIt is removed from X, the sparse coefficient α of all acquisitions is formed into a coefficient matrix Scoeff ∈ Rn×n,
Wherein factor alpha ∈ Rn×1Indicate each sample in X to xkContribution rate;γ expressions prevent over-fitting coefficient, γ to the constraint of α
For constant coefficient.
Step 4:Matrix and the sparse dematrix of test sample subset are initialized according to fuzzy membership, iterative solution is more
New fuzzy membership matrix of the facial image data set about class label.
In implementation process, the fuzzy membership of iterative modifications test sample, wherein t are iterations, and t is to be more than or wait
In 1 positive integer.
Test sample x is calculated using coefficient matrix Scoeff obtained abovekWith every relevant reconstructed residual of one kind.
K-th of test sample xkCorresponding jth class sample set XjReconstructed residual errkjt:
errkjt=| | xk-XjWjt-1α2||
Wherein Wjt-1Indicate x when the t-1 times iterationkThe degree of membership of corresponding jth class;XjFor jth class sample set.
According to reconstructed residual err derived abovekjtRemove optimization following formula:
It can obtain the degree of membership of updated test sample:
Update subordinated-degree matrix:
Use mkjtRemove update Mtu, Mtl=M0l, new subordinated-degree matrix can be obtained
Until meeting Mt-Mt-1The absolute value of all elements in the matrix obtained afterwards is respectively less than predetermined threshold value, then iteration knot
Beam;Or current iteration number is more than maximum iterations, then iteration terminates.
Step 5:It obtains the class label corresponding to the maximum membership degree of each test sample and test sample can be completed
Classification.
The subordinated-degree matrix of the test sample obtained to the t times iteration is handled to obtain final classification knot according to the following formula
Fruit.
The face identification method of the present invention is verified with specific experiment below:
By taking used YaleB data sets as an example, everyone is a classification in used data set, at this time
C=1,2 ..., 38 }.5 times of cross validations are used in the present invention, sample data set are divided into 5 parts uniformly at random, often
It is secondary to take a copy of it as test sample collection Xu, remaining four parts are then used as Xl, experiment can be repeated 5 times.
It is verified with the image data in human face data collection (YaleB), includes 38 people in YaleB face databases,
Share the sample of 2414 known types, pixel 32*32.Experiment uses 5 times of cross validations, and all data are equal at random
It is even to be divided into 10 parts, one group is chosen every time and is used as test data, remaining is used as training data, tests to be repeated 5 times to take for 5 times and is averaged
For value as final recognition accuracy, accuracy rate is as shown in table 1.
1 recognition methods of table is compared
The present invention is more of the invention respectively to use rate in the case of 10%, 20%, 30%, 40%, 50% and 60%
The recognition methods and LNP, SIS, LPSN and MLRR recognition methods discrimination, identification side of the invention as can be seen from Table 1
Method has higher discrimination compared with other recognition methods.Wherein, LNP is Local Neighborhood Patterns'
Referred to as, full name is local neighborhood mode method;SIS is Sparsity Induced Similarity measure for
The abbreviation of label propagation, full name are the label transmission method based on sparse similarity measure;LPSN is Label
The abbreviation of propagation through sparse neighborhood, full name are sparse close to label transmission method;MLRR
For the abbreviation of manifold low-rank representation, full name is manifold low-rank representation method.
The present invention takes full advantage of the characteristics of image of global and local and by sparse residual error for obscuring sparse degree of membership
In calculating.Since face digital picture itself has manifold structure, reconstructed residual is combined solution with data fuzzy membership
During data membership degree can propagate reconstructed residual in view of label, consider the label of reconstructed residual than only consider sparse coding or
Person only considers that the label of data similarity is propagated to have and more judges information, therefore can improve image recognition accuracy.
Fig. 3 is that semi-supervised one structure of face identification system embodiment based on the sparse reconstruct of fuzzy membership of the present invention is shown
It is intended to.Semi-supervised face identification system based on the sparse reconstruct of fuzzy membership as shown in Figure 3, including recognition of face service
Device, the recognition of face server include:Face image data collection acquisition module, fuzzy membership initialize matrix initialisation mould
Block, sparse solution Matrix Solving module, fuzzy membership matrix update module and class label acquisition module.
(1) face image data collection acquisition module is used to obtain face image data collection;Face image data collection includes
Training sample subset and test sample subset;The sample of training sample subset is facial image known to class label, test specimens
The sample of this subset is the unknown facial image of class label.
(2) fuzzy membership initializes matrix initialisation module, is used to, according to class label known to training sample, obtain
The fuzzy membership for obtaining training sample subset initializes matrix;Reconstructed residual of the test sample about class label is calculated, is obtained
The fuzzy membership of test sample subset initializes matrix;Face image data collection is further obtained about the fuzzy of class label
Degree of membership initializes matrix.
Fuzzy membership initialization matrix initialisation module includes the fuzzy membership initialization matrix of training sample subset
Initialization module is used to predefine class label first, and then whether the training sample in training of judgement sample set belongs to
In predetermined class label, if so, corresponding training sample fuzzy membership is 1;Otherwise, corresponding training sample is fuzzy is subordinate to
Category degree is 0.
Fuzzy membership initialization matrix initialisation module further includes the fuzzy membership initialization square of test sample subset
Battle array initialization module, is used to two numbers seeking test sample about the other fuzzy membership of tag class as quotient;Wherein, divisor
Inverse for test sample about respective labels classification residual error, dividend are that test sample is residual about the other reconstruct of all tag class
It is poor reciprocal tired and.
(3) sparse solution Matrix Solving module, is used to solve the sparse coefficient of all test samples, and then obtains test specimens
The sparse dematrix of this subset.
(4) fuzzy membership matrix update module is used to initialize matrix and test sample according to fuzzy membership
The sparse dematrix of subset, fuzzy membership matrix of the iterative solution update face image data collection about class label.
Fuzzy membership matrix update module is additionally operable to the sparse dematrix according to test sample subset, calculates each test
Reconstructed residual of the sample about class label;
Reconstructed residual according to each test sample about class label obtains fuzzy membership after the update of test sample
Degree, further obtains the updated fuzzy membership matrix of test sample subset;
Training sample subset fuzzy membership initialization matrix it is constant, merge training sample subset fuzzy membership and
The updated fuzzy membership matrix of test sample subset, obtains updated mould of the face image data collection about class label
Paste subordinated-degree matrix.
(5) class label acquisition module is used to obtain the classification mark corresponding to the maximum membership degree of each test sample
The classification of test sample can be completed in label.
Fig. 4 is that semi-supervised two structure of face identification system embodiment based on the sparse reconstruct of fuzzy membership of the present invention is shown
It is intended to.Semi-supervised face identification system based on the sparse reconstruct of fuzzy membership as shown in Figure 4, including recognition of face server
And display device, the display device are connected with recognition of face server;
Recognition of face server includes:Face image data collection acquisition module, fuzzy membership initialize matrix initialisation
Module, sparse solution Matrix Solving module, fuzzy membership matrix update module and class label acquisition module.
Display device is used to show the class label corresponding to the maximum membership degree of each test sample.Wherein, display dress
It sets including various types of display screens.
The present invention takes full advantage of the characteristics of image of global and local and by sparse residual error for obscuring sparse degree of membership
In calculating.Since face digital picture itself has manifold structure, reconstructed residual is combined solution with data fuzzy membership
During data membership degree can propagate reconstructed residual in view of label, consider the label of reconstructed residual than only consider sparse coding or
Person only considers that the label of data similarity is propagated to have and more judges information, therefore can improve image recognition accuracy.
One of ordinary skill in the art will appreciate that realizing all or part of flow in above-described embodiment method, being can be with
Relevant hardware is instructed to complete by computer program, the program can be stored in computer read/write memory medium
In, the program is when being executed, it may include such as the flow of the embodiment of above-mentioned each method.Wherein, the storage medium can be magnetic
Dish, CD, read-only memory (Read-Only Memory, ROM) or random access memory (Random
AccessMemory, RAM) etc..
Above-mentioned, although the foregoing specific embodiments of the present invention is described with reference to the accompanying drawings, not protects model to the present invention
The limitation enclosed, those skilled in the art should understand that, based on the technical solutions of the present invention, those skilled in the art are not
Need to make the creative labor the various modifications or changes that can be made still within protection scope of the present invention.
Claims (10)
1. a kind of semi-supervised face identification method based on the sparse reconstruct of fuzzy membership, which is characterized in that including:
Obtain face image data collection;Face image data collection includes training sample subset and test sample subset;Training sample
The sample of subset is facial image known to class label, and the sample of test sample subset is the unknown face figure of class label
Picture;
According to class label known to training sample, the fuzzy membership initialization matrix of training sample subset is obtained;According to survey
Reconstructed residual of the sample sheet about class label, the fuzzy membership for obtaining test sample subset initialize matrix;Further
Fuzzy membership to face image data collection about class label initializes matrix;
The sparse coefficient of all test samples is solved, and then obtains the sparse dematrix of test sample subset;
Matrix is initialized according to the fuzzy membership of training sample subset and the sparse dematrix of test sample subset, iteration are asked
Fuzzy membership matrix of the solution update face image data collection about class label;
It obtains the class label corresponding to the maximum membership degree of each test sample and classification can be completed.
2. a kind of semi-supervised face identification method based on the sparse reconstruct of fuzzy membership as described in claim 1, feature
It is, during the fuzzy membership for obtaining training sample subset initializes matrix, predefines class label first, so
Whether the training sample in training of judgement sample set belongs to predetermined class label afterwards, if so, corresponding training sample
Fuzzy membership is 1;Otherwise, corresponding training sample fuzzy membership is 0.
3. a kind of semi-supervised face identification method based on the sparse reconstruct of fuzzy membership as described in claim 1, feature
It is, in the fuzzy membership initialization matrix of test sample subset, test sample is about the other fuzzy membership of tag class
Two numbers make the ratio of quotient;Wherein, divisor is inverse of the test sample about respective labels classification residual error, and dividend is test specimens
This it is reciprocal about the other reconstructed residual of all tag class tired and.
4. a kind of semi-supervised face identification method based on the sparse reconstruct of fuzzy membership as described in claim 1, feature
It is, during iterative solution updates fuzzy membership matrix of the face image data collection about class label,
According to the sparse dematrix of test sample subset, reconstructed residual of each test sample about class label is calculated;
Reconstructed residual according to each test sample about class label obtains fuzzy membership after the update of test sample, into
One step obtains the updated fuzzy membership matrix of test sample subset;
The fuzzy membership initialization matrix of training sample subset is constant, merges fuzzy membership and the test of training sample subset
The updated fuzzy membership matrix of sample set obtains updated fuzzy person in servitude of the face image data collection about class label
Category degree matrix.
5. a kind of semi-supervised face identification method based on the sparse reconstruct of fuzzy membership as described in claim 1, feature
It is, during iterative solution updates fuzzy membership matrix of the face image data collection about class label, if meeting
This fuzzy membership matrix sought is made all in the matrix obtained after difference with the preceding fuzzy membership matrix once sought
The absolute value of element is respectively less than predetermined threshold value, then iteration terminates;Or iterations are more than maximum iterations, then iteration knot
Beam.
6. a kind of semi-supervised face identification system based on the sparse reconstruct of fuzzy membership, which is characterized in that including recognition of face
Server, the recognition of face server include:
Face image data collection acquisition module, is used to obtain face image data collection;Face image data collection includes training sample
This subset and test sample subset;The sample of training sample subset is facial image known to class label, test sample subset
Sample be the unknown facial image of class label;
Fuzzy membership initializes matrix initialisation module, is used to, according to class label known to training sample, be trained
The fuzzy membership of sample set initializes matrix;Reconstructed residual according to test sample about class label obtains test specimens
The fuzzy membership of this subset initializes matrix;Further obtain fuzzy membership of the face image data collection about class label
Initialize matrix;
Sparse solution Matrix Solving module, is used to solve the sparse coefficient of all test samples, and then obtains test sample subset
Sparse dematrix;
Fuzzy membership matrix update module is used to initialize matrix and survey according to the fuzzy membership of training sample subset
Try the sparse dematrix of sample set, fuzzy membership square of the iterative solution update face image data collection about class label
Battle array;
Class label acquisition module, the class label for being used to obtain corresponding to the maximum membership degree of each test sample can be complete
Constituent class.
7. a kind of semi-supervised face identification system based on the sparse reconstruct of fuzzy membership as claimed in claim 6, feature
It is, the fuzzy membership initialization matrix initialisation module includes the fuzzy membership initialization matrix of training sample subset
Initialization module is used to predefine class label first, and then whether the training sample in training of judgement sample set belongs to
In predetermined class label, if so, corresponding training sample fuzzy membership is 1;Otherwise, corresponding training sample is fuzzy is subordinate to
Category degree is 0.
8. a kind of semi-supervised face identification system based on the sparse reconstruct of fuzzy membership as claimed in claim 6, feature
It is, the fuzzy membership initialization matrix initialisation module further includes the fuzzy membership initialization square of test sample subset
Battle array initialization module, is used to two numbers seeking test sample about the other fuzzy membership of tag class as quotient;Wherein, divisor
Inverse for test sample about respective labels classification residual error, dividend are that test sample is residual about the other reconstruct of all tag class
It is poor reciprocal tired and.
9. a kind of semi-supervised face identification system based on the sparse reconstruct of fuzzy membership as claimed in claim 6, feature
It is, the fuzzy membership matrix update module is additionally operable to the sparse dematrix according to test sample subset, calculates each survey
Reconstructed residual of the sample sheet about class label;
Reconstructed residual according to each test sample about class label obtains fuzzy membership after the update of test sample, into
One step obtains the updated fuzzy membership matrix of test sample subset;
The fuzzy membership initialization matrix of training sample subset is constant, merges fuzzy membership and the test of training sample subset
The updated fuzzy membership matrix of sample set obtains updated fuzzy person in servitude of the face image data collection about class label
Category degree matrix.
10. a kind of semi-supervised face identification system based on the sparse reconstruct of fuzzy membership as claimed in claim 6, feature
It is, the semi-supervised face identification system based on the sparse reconstruct of fuzzy membership further includes display device, the display dress
It sets and is connected with recognition of face server;The display device is used to show the class corresponding to the maximum membership degree of each test sample
Distinguishing label.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201610811185.3A CN106446806B (en) | 2016-09-08 | 2016-09-08 | Semi-supervised face identification method based on the sparse reconstruct of fuzzy membership and system |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201610811185.3A CN106446806B (en) | 2016-09-08 | 2016-09-08 | Semi-supervised face identification method based on the sparse reconstruct of fuzzy membership and system |
Publications (2)
Publication Number | Publication Date |
---|---|
CN106446806A CN106446806A (en) | 2017-02-22 |
CN106446806B true CN106446806B (en) | 2018-10-12 |
Family
ID=58164609
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201610811185.3A Expired - Fee Related CN106446806B (en) | 2016-09-08 | 2016-09-08 | Semi-supervised face identification method based on the sparse reconstruct of fuzzy membership and system |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN106446806B (en) |
Families Citing this family (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN107194314B (en) * | 2017-04-20 | 2020-09-25 | 江苏大学 | Face recognition method fusing fuzzy 2DPCA and fuzzy 2DLDA |
CN112131446B (en) * | 2020-08-19 | 2023-11-17 | 深圳云天励飞技术股份有限公司 | Graph node classification method and device, electronic equipment and storage medium |
CN112052187B (en) * | 2020-10-13 | 2022-07-12 | 苏州浪潮智能科技有限公司 | Distribution method of test matrix |
CN114330574A (en) * | 2021-12-31 | 2022-04-12 | 广东泰迪智能科技股份有限公司 | Fuzzy labeling method for pattern recognition |
CN115131610B (en) * | 2022-06-13 | 2024-02-27 | 西北工业大学 | Robust semi-supervised image classification method based on data mining |
Citations (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN102609681A (en) * | 2012-01-12 | 2012-07-25 | 北京大学 | Face recognition method based on dictionary learning models |
CN102831390A (en) * | 2012-07-02 | 2012-12-19 | 北京科技大学 | Human ear authenticating system and method |
CN104573726A (en) * | 2015-01-12 | 2015-04-29 | 山东师范大学 | Facial image identification method for reconstructing optimal error combination based on quartering and components |
-
2016
- 2016-09-08 CN CN201610811185.3A patent/CN106446806B/en not_active Expired - Fee Related
Patent Citations (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN102609681A (en) * | 2012-01-12 | 2012-07-25 | 北京大学 | Face recognition method based on dictionary learning models |
CN102831390A (en) * | 2012-07-02 | 2012-12-19 | 北京科技大学 | Human ear authenticating system and method |
CN104573726A (en) * | 2015-01-12 | 2015-04-29 | 山东师范大学 | Facial image identification method for reconstructing optimal error combination based on quartering and components |
Non-Patent Citations (1)
Title |
---|
"基于稀疏表示的人脸特征提取与识别算法研究";王永欣;《中国优秀硕士学位论文全文数据库-信息科技辑》;20180115(第1期);I138-1385 * |
Also Published As
Publication number | Publication date |
---|---|
CN106446806A (en) | 2017-02-22 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN106446806B (en) | Semi-supervised face identification method based on the sparse reconstruct of fuzzy membership and system | |
Qu et al. | Improving nuclei/gland instance segmentation in histopathology images by full resolution neural network and spatial constrained loss | |
Kim et al. | Learning full pairwise affinities for spectral segmentation | |
Shakhnarovich et al. | Nearest-neighbor methods in learning and vision | |
Ali et al. | A hybrid geometric spatial image representation for scene classification | |
Rahaman et al. | An efficient multilevel thresholding based satellite image segmentation approach using a new adaptive cuckoo search algorithm | |
CN109615014A (en) | A kind of data sorting system and method based on the optimization of KL divergence | |
Zhang et al. | Semantic classification of heterogeneous urban scenes using intrascene feature similarity and interscene semantic dependency | |
Chakraborty et al. | Simultaneous variable weighting and determining the number of clusters—A weighted Gaussian means algorithm | |
CN103955709B (en) | Weighted synthetic kernel and triple markov field (TMF) based polarimetric synthetic aperture radar (SAR) image classification method | |
Blanchart et al. | A semi-supervised algorithm for auto-annotation and unknown structures discovery in satellite image databases | |
CN106228027A (en) | A kind of semi-supervised feature selection approach of various visual angles data | |
Zhao et al. | Soft label based linear discriminant analysis for image recognition and retrieval | |
CN106250918B (en) | A kind of mixed Gauss model matching process based on improved soil-shifting distance | |
Deng et al. | Efficient 3D face recognition using local covariance descriptor and Riemannian kernel sparse coding | |
Wang et al. | Building correlations between filters in convolutional neural networks | |
CN107578063B (en) | Image Spectral Clustering based on fast selecting landmark point | |
Chen et al. | A robust automatic clustering algorithm for probability density functions with application to categorizing color images | |
Leclaire et al. | A fast multi-layer approximation to semi-discrete optimal transport | |
US12026873B2 (en) | Bone age assessment method for bone image | |
Khurana et al. | A unified approach to change detection using an adaptive ensemble of extreme learning machines | |
CN105844299B (en) | A kind of image classification method based on bag of words | |
Yang et al. | Robust Semisupervised Kernelized Fuzzy Local Information C‐Means Clustering for Image Segmentation | |
CN116524243A (en) | Classification method and device for fossil images of penstones | |
CN104050486B (en) | Polarimetric SAR image classification method based on maps and Wishart distance |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
C06 | Publication | ||
PB01 | Publication | ||
C10 | Entry into substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
GR01 | Patent grant | ||
GR01 | Patent grant | ||
CF01 | Termination of patent right due to non-payment of annual fee |
Granted publication date: 20181012 |