CN109508697B - Face recognition method, system and storage medium based on semi-nonnegative matrix factorization of E auxiliary function - Google Patents
Face recognition method, system and storage medium based on semi-nonnegative matrix factorization of E auxiliary function Download PDFInfo
- Publication number
- CN109508697B CN109508697B CN201811536236.1A CN201811536236A CN109508697B CN 109508697 B CN109508697 B CN 109508697B CN 201811536236 A CN201811536236 A CN 201811536236A CN 109508697 B CN109508697 B CN 109508697B
- Authority
- CN
- China
- Prior art keywords
- function
- matrix
- module
- auxiliary function
- training sample
- 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
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V40/00—Recognition of biometric, human-related or animal-related patterns in image or video data
- G06V40/10—Human or animal bodies, e.g. vehicle occupants or pedestrians; Body parts, e.g. hands
- G06V40/16—Human faces, e.g. facial parts, sketches or expressions
- G06V40/168—Feature extraction; Face representation
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/213—Feature extraction, e.g. by transforming the feature space; Summarisation; Mappings, e.g. subspace methods
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
-
- 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)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- Computer Vision & Pattern Recognition (AREA)
- General Physics & Mathematics (AREA)
- Oral & Maxillofacial Surgery (AREA)
- Health & Medical Sciences (AREA)
- Bioinformatics & Computational Biology (AREA)
- Evolutionary Computation (AREA)
- General Engineering & Computer Science (AREA)
- Evolutionary Biology (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Artificial Intelligence (AREA)
- General Health & Medical Sciences (AREA)
- Life Sciences & Earth Sciences (AREA)
- Human Computer Interaction (AREA)
- Multimedia (AREA)
- Collating Specific Patterns (AREA)
- Image Analysis (AREA)
Abstract
The invention provides a face recognition method, a system and a storage medium based on E auxiliary function semi-nonnegative matrix decomposition, wherein the face recognition method comprises the following steps: converting training sample images into a training sample matrixSetting an error thresholdMaximum number of iterationsAnd inputting the training sample matrixError threshold and maximum number of iterations(ii) a The second step is as follows: to the base image matrix and coefficient matrixCarrying out initialization; the fourth step: updating the base image matrix according to equation (6)Sum coefficient matrix(ii) a A sixth step: determining an objective functionOr whether the iteration number n reaches the maximum iteration numberAnd if so, outputting the base image matrixSum coefficient matrixOtherwise, executing the fourth step. The invention has the beneficial effects that: the face recognition method has the advantages of high recognition performance and low calculation complexity, and the results show that the method developed by the patent has certain superiority by experimental comparison with related algorithms in a public face database.
Description
Technical Field
The invention relates to the technical field of data processing, in particular to a face recognition method, a face recognition system and a storage medium based on E auxiliary function semi-nonnegative matrix decomposition.
Background
With the advent of the information age, biometric identification technology for identifying an individual's identity using physiological and behavioral characteristics inherent to a human body has become one of the most active research fields. Among the many branches of biometric technology, one of the most well accepted techniques is face recognition technology, because face recognition is non-invasive, non-mandatory, non-contact, and concurrent with respect to other biometric technologies.
The face recognition technology comprises two stages, wherein the first stage is feature extraction, namely extracting face feature information in a face image, and the first stage directly determines the quality of the face recognition technology; the second stage is identity authentication, and personal identity authentication is carried out according to the extracted characteristic information. Principal Component Analysis (PCA) and Singular Value Decomposition (SVD) are more classical feature extraction methods, but feature vectors proposed by the two methods usually contain negative elements, so that the methods are not reasonable and interpretable under the condition that an original sample is non-negative data. non-Negative Matrix Factorization (NMF) is a feature extraction method for processing non-negative data, and its application is very wide, such as hyperspectral data processing, face image recognition, etc. The NMF algorithm has non-negative limitation on the extracted features in the original sample non-negative data matrix decomposition process, namely all components after decomposition are non-negative, so that non-negative sparse features can be extracted. The essence of the NMF algorithm is that the non-negative matrix X is approximately decomposed as the product of the base image matrix W and the coefficient matrix H, i.e. X ≈ WH, and both W and H are non-negative matrices. Thus each column of the matrix X can be represented as a non-negative linear combination of the vectors of the columns of the matrix W, which also follows the construction basis of the NMF algorithm-the perception of the whole is made up of the perception of the parts that make up the whole (purely additive).
However, NMF requires the original sample to be non-negative data, which limits the range of application of the algorithm to some extent. In order to expand the range of NMF applications, in recent years, researchers have proposed many algorithms for NMF deformation, such as Semi-non-negative matrix factorization (Semi-NMF) and Convex non-negative matrix factorization (Convex-NMF). Both algorithms can be used for non-negative data as well as for other data. However, since they allow the original sample to contain negative numbers, it becomes a complicated problem to prove the convergence of the iterative formula. Meanwhile, the existing Semi-NMF iteration method has the problems of low convergence rate, low algorithm recognition rate and the like.
1. Auxiliary Function (Auxiliary Function)
The auxiliary function is a common tool for proving the convergence of the algorithm, and is defined and characterized as follows: definition 1: for arbitrary matrices H and H(t)If the condition is satisfied
G(H,H(t)) ≧ f (H), and G (H)(t),H(t))=f(H(t))
Then called G (H, H)(t)) Is an auxiliary function of the function f (h).
Introduction 1: if G (H, H)(t)) Is an auxiliary function of f (H), then f (H) is as followsThe new rule is that the operation is monotonous and does not increase,
2. semi non-negative matrix factorization (Semi-NMF)
Semi-non-negative matrix factorization (Semi-NMF) does not limit the sign of the sample matrix, thereby expanding the application range of NMF. For matrix decomposition X ≈ WH, the optimization problem that Semi-NMF needs to solve is:
the updating iteration formula is as follows:
3. Convex non-negative matrix factorization (Convex-NMF)
The main idea of Convex non-negative matrix factorization (Convex-NMF) is to restrict the columns in the base image matrix to one Convex combination of columns in the original matrix. For matrix decomposition X ≈ XWH, the optimization problem that the Convex-NMF needs to solve is:
the updating iteration formula is as follows:
as can be seen from the above Semi-NMF and Convex-NMF iterative formulas, the square root of the open arithmetic appears therein, which greatly increases the computational complexity of the algorithm and reduces the computational efficiency of the algorithm.
To sum up:
1. the application range of the non-negative matrix factorization algorithm (NMF) is narrow, and the non-negative matrix factorization algorithm is only suitable for non-negative data.
2. Although the application range of the non-negative matrix factorization algorithm is expanded by the traditional Semi-non-negative matrix factorization algorithm (Semi-NMF), the effect and the convergence rate of the non-negative matrix factorization algorithm still need to be improved.
Disclosure of Invention
The invention provides a face recognition method based on E auxiliary function semi-nonnegative matrix decomposition, which comprises a training step, wherein the training step comprises the following steps:
the first step is as follows: converting the training sample image into a training sample matrix X, and setting an error threshold epsilon and a maximum iteration number ImaxAnd inputting a training sample matrix X, an error threshold epsilon and a maximum iteration number Imax
The second step is as follows: initializing a base image matrix W and a coefficient matrix H;
the third step: setting the iteration number n to be 0;
the fourth step: updating the base image matrix W and the coefficient matrix H according to formula (6);
the fifth step: n is n + 1;
a sixth step: judging whether the objective function F (W, H) is less than or equal to epsilon or the iteration number n reaches the maximum iteration number ImaxIf yes, outputting a base image matrix W and a coefficient matrix H, otherwise, executing a fourth step;
in the fourth step, equation (6) is as follows:
W←XHT(HHT)-1
in formula (6), W represents a base image matrix, H represents a coefficient matrix, and X represents a training sample matrix.
As a further improvement of the invention: the face recognition method further comprises a recognition step which is executed after the training step, wherein the recognition step comprises the following steps:
a seventh step of: calculating the average characteristic vector m of each class in the training samplej(j=1,…,c);
The eighth step is that the face image y to be recognized is input and the characteristic vector h of the face image y is calculatedy=W-1y;
The ninth step is that the characteristic vector h of the face image to be recognized is calculatedyAverage feature vector m to each classjIf h isyAnd mjIf the distance is the minimum, the face image y to be recognized is classified into the P-th class;
a tenth step: and outputting the category P.
The invention also provides a face recognition system based on the semi-nonnegative matrix factorization of the E auxiliary function, which comprises a training module, wherein the training module comprises:
an input module: for converting training sample image into training sample matrix X, setting error threshold epsilon and maximum iteration number ImaxAnd inputting a training sample matrix X, an error threshold epsilon and a maximum iteration number Imax;
An initialization module: the method comprises the steps of initializing a base image matrix W and a coefficient matrix H;
an assignment module: setting the iteration number n to be 0;
an update module: for updating the base image matrix W and the coefficient matrix H according to equation (6); a counting module: n is n + 1;
a judging module: judging whether the objective function F (W, H) is less than or equal to epsilon or the iteration number n reaches the maximum iteration number ImaxIf yes, outputting a base image matrix W and a coefficient matrix H, otherwise, executing an updating module;
in the update module, equation (6) is as follows:
W←XHT(HHT)-1
in formula (6), W represents a base image matrix, H represents a coefficient matrix, and X represents a training sample matrix.
As a further improvement of the invention: the face recognition system further comprises a recognition module executed after the training module, the recognition module comprising:
an average feature vector calculation module: for calculating an average feature vector m for each class in a training samplej(j=1,…,c);
The characteristic vector calculation module is used for inputting the face image y to be recognized and calculating the characteristic vector h thereofy=W-1y;
Distance calculating module for calculating characteristic vector h of face image to be recognizedyAverage feature vector m to each classjIf h isyAnd mjIf the distance is the minimum, the face image y to be recognized is classified into the P-th class;
an output module: for outputting the category P.
The invention also discloses a computer-readable storage medium, in which a computer program is stored which, when being invoked by a processor, is configured to carry out the steps of the method as claimed in the claims.
The invention has the beneficial effects that: the face recognition method has the advantages of high recognition performance and low calculation complexity, and the results show that the method developed by the patent has certain superiority by experimental comparison with related algorithms in a public face database.
Drawings
FIG. 1 is a flow chart of the algorithm construction process of the present invention;
FIG. 2 is a flow chart of a method of the present invention;
FIG. 3 shows the method and related algorithms (Convex-NMF and Semi-NMF) proposed by the present invention
Comparison of recognition rates on ORL face database, where FSNMF represents the method of the invention;
FIG. 4 is a graph comparing the convergence curves of the inventive method and the Semi-NMF algorithm, where FSNMF represents the inventive method.
Detailed Description
The invention firstly provides a new concept of the E auxiliary function of the objective function, and accordingly provides a new basic theory and framework for constructing the auxiliary function, which greatly expand the selection range of the auxiliary function and also provide a powerful tool for flexibly constructing the auxiliary function to design a new high-performance non-negative characteristic algorithm. According to the method for the E auxiliary function, provided by the invention, a new auxiliary function is constructed, and a new fast semi-nonnegative matrix factorization (FSNMF) algorithm is deduced according to the new auxiliary function. The FSNMF algorithm obtained by the invention is convergent according to the property of the auxiliary function. The experimental result shows that compared with other NMF-based algorithms, the FSNMF iteration method provided by the invention has better recognition rate and faster convergence rate. Meanwhile, the E auxiliary function method provided by the invention is also suitable for developing other biological feature identification iterative algorithms and proving the convergence of the algorithms.
Keyword interpretation:
1. description of the symbols:
x matrix
XTTranspose of X
Taking absolute value of every element of matrix X
XijThe ijth element of the matrix X
xjJ column of matrix X
AdD powers of each element in the matrix A
2. Non-negative Matrix Factorization (Non-negative Matrix Factorization, NMF)
The basic idea of NMF is to use a non-negative sample matrixThe approximate decomposition is the product of two non-negative matrices, namely:
X≈WH,
wherein the content of the first and second substances,andreferred to as base image matrix and coefficient matrix, respectively. And, by constructing a loss function metric that measures the degree of approximation between X and WH, the loss function is typically defined based on the F-norm, written as:
F(W,H)=||X-WH||2 (1)
where | · | | represents the F-norm.
3. Semi-Non-negative Matrix Factorization (Semi-Non-negative Matrix Factorization, Semi-NMF)
The Semi-NMF does not limit the sign of the sample matrix, thereby expanding the application range of the NMF. For sample matrixConsidering the decomposition:
X≈WH,
wherein the content of the first and second substances,andi.e., Semi-NMF only limits the coefficient matrix to non-negative.
The specific technical scheme is as follows:
1. the main objects of the present invention are:
(1) the application range of the NMF is expanded, so that the NMF still has effectiveness on data containing negative numbers;
(2) a new concept of the E-helper function is proposed, whereby a basic theoretical framework and method of constructing a new helper function is presented.
(3) According to the proposed E-helper function method, a new fast semi-nonnegative matrix factorization (FSNMF) face recognition algorithm with high recognition performance and low computational complexity is developed.
2. The technical scheme is as follows:
(1) definition and properties of the E-helper function:
for many objective functions, it is difficult to construct a suitable helper function. For example, the auxiliary function constructed in the conventional semi-nonnegative matrix factorization algorithm is extremely complex, and the convergence rate of the obtained iterative formula is slow. In order to overcome the defects, the patent firstly provides the concept of the E auxiliary function, and obtains a new fast semi-nonnegative matrix factorization (FSNMF) face recognition algorithm with high recognition performance and low computation complexity based on the E auxiliary function. The definition and nature of the E-helper function is as follows:
definition 2: for arbitrary matrices H and H(t)If E (H, H)(t)) Satisfies the conditions
E(H,H(t)) Not less than 0, and E (H)(t),H(t))=0
Then called E (H, H)(t)) An E helper function for the function f (H).
Theorem 1: if the function G (H, H)(t)) Is an auxiliary function of the function f (H), if the function E (H, H)(t)) Is an E-helper function of the function f (h). Then there are:
is still an auxiliary function of f (H), where λ ≧ 0.
And (3) proving that: from the definition of the helper function and the E helper function, we have:
(2) Proposal of new FSNMF:
construction of an objective function
X≈WH
wherein the content of the first and second substances,andthe objective function to obtain the fast semi-nonnegative matrix factorization algorithm (FSNMF) is:
F(W,H)=||X-WH||2 (3)
to solve the two unknown matrices W and H in the objective function (3), we convert the objective function into two sub-objective functions, which are:
f1(W)=||X-WH||2in which H is fixed
f2(H)=||X-WH||2Where W is fixed, problem (3) also evolves into two sub-problems, respectively:
minf1(W) (4)
minf2(H)s.t.H≥0 (5)
learning of the base image matrix W
The H is fixed, and the reaction is carried out,due to f1(W) is convex, and the sign of W is not limited, so that the extreme point is the optimal solution. The gradient is calculated and made 0 with:
the closed form solution of W can be solved:
W=XHT(HHT)-1
learning of coefficient matrix H
Fixing W, we construct the objective function f first2(H) An auxiliary function G (H, H)(t)). An objective function f2(H) The expansion is as follows:
f2(H)=tr(XTX)-2tr(HT(WTX)+-HT(WTX)-)+tr((WTW)+-(WTW)-)HHT
for the last two terms on the right of the equation, we have:
2, leading: function(s)
Is tr ((W)TW)+HHT) An auxiliary function of (2).
And 3, introduction: function(s)
G2(H,H(t))=-2tr((WTW)-H(t)HT)+tr((WTW)-H(t)H(t)T)
Is-tr ((W)TW)-HHT) An auxiliary function of (2).
From the above two lemmas, we can get theorem 2.
Theorem 2: function(s)
Is f2(H) An auxiliary function of (2).
And (3) proving that: as can be seen from lemons 2 and 3,is tr ((W)TW)+HHT) The auxiliary function of (a) is selected,is-tr ((W)TW)-HHT) The auxiliary function of (2).
Therefore, there is G (H, H)(t)) Is f2(H) An auxiliary function of (2).
Generally, updating the iterative formula can be derived directly from the auxiliary function. But here by an auxiliary function G (H, H)(t)) The derived formula does not satisfy the non-negativity, so that the method needs to reconstruct a new auxiliary function by means of the E auxiliary function.
Theorem 3: function(s)
Is f2(H) An E auxiliary function of.
And (3) proving that: it is clear that,
E(H,H(t)) Not less than 0, and E (H)(t),H(t))=0
Thus, E (H, H) is known(t)) Is still f2(H) An E auxiliary function of.
Theorem 4: defining a function G (H, H)(t)) And E (H, H)(t)) The following were used:
And (3) proving that: from theorems 2 and 3, it can be seen that: function G (H, H)(t)) Is a function f2(H) An auxiliary function of, and a function E (H, H)(t)) Is a function f2(H) An E auxiliary function of. With theorem 1, we can directly deduce that λ is 1Is still f2(H) An auxiliary function of (2).
From definition 1 and lemma 1, we know the functionIs a function f2(H) An upper limit of (2), andto obtainWe solve its derivative and make it 0, have
From this equation, the updated iterative equation for H can be solved:
through simple derivation, the iterative formula can be proved to meet the KKT condition.
To sum up, the updating iterative formula of FSNMF proposed by this patent is:
compared with the traditional Semi-NMF iterative formula (2), the updated iterative formula (6) provided by the patent not only greatly improves the convergence rate and the calculation rate, but also has a better recognition effect.
To sum up:
as shown in fig. 2, the present invention provides a face recognition method based on E-helper function semi-nonnegative matrix factorization, which includes a training step, where the training step includes the following steps:
the first step is as follows: converting the training sample image into a training sample matrix X, and setting an error threshold epsilon and a maximum iteration number ImaxAnd inputting a training sample matrix X, an error threshold epsilon and a maximum iteration number Imax;
The second step is as follows: initializing a base image matrix W and a coefficient matrix H;
the third step: setting the iteration number n to be 0;
the fourth step: updating the base image matrix W and the coefficient matrix H according to formula (6);
the fifth step: n is n + 1;
a sixth step: judging whether the objective function F (W, H) is less than or equal to epsilon or the iteration number n reaches the maximum iteration number ImaxIf yes, outputting a base image matrix W and a coefficient matrix H, otherwise, executing a fourth step;
in the fourth step, equation (6) is as follows:
W←XHT(HHT)-1
in formula (6), W represents a base image matrix, H represents a coefficient matrix, X represents a training sample matrix, and T represents a transpose of the matrix.
The face recognition method further comprises a recognition step which is executed after the training step, wherein the recognition step comprises the following steps:
a seventh step of: calculating each class in the training sampleAverage feature vector m ofj(j ═ 1, …, C), j denotes the jth class, C denotes a total of C different face classes;
the eighth step is that the face image y to be recognized is input and the characteristic vector h of the face image y is calculatedy=W-1y, W represent the base image matrix;
the ninth step is that the characteristic vector h of the face image to be recognized is calculatedyAverage feature vector m to each classjIf h isyAnd mjIf the distance is the minimum, the face image y to be recognized is classified into the P-th class;
a tenth step: and outputting the category P.
And outputting the type P, namely the face image y to be recognized is recognized as the No. P personal face type, so that after the type P is output, the face recognition is finished.
The invention also provides a face recognition system based on the semi-nonnegative matrix factorization of the E auxiliary function, which comprises a training module, wherein the training module comprises:
an input module: for converting training sample image into training sample matrix X, setting error threshold epsilon and maximum iteration number ImaxAnd inputting a training sample matrix X, an error threshold epsilon and a maximum iteration number Imax;
An initialization module: the method comprises the steps of initializing a base image matrix W and a coefficient matrix H;
an assignment module: setting the iteration number n to be 0;
an update module: for updating the base image matrix W and the coefficient matrix H according to equation (6); a counting module: n is n + 1;
a judging module: judging whether the objective function F (W, H) is less than or equal to epsilon or the iteration number n reaches the maximum iteration number ImaxIf yes, outputting a base image matrix W and a coefficient matrix H, otherwise, executing an updating module;
in the update module, equation (6) is as follows:
W←XHT(HHT)-1
in formula (6), W represents a base image matrix, H represents a coefficient matrix, X represents a training sample matrix, and T represents a transpose of the matrix.
The face recognition system further comprises a recognition module executed after the training module, the recognition module comprising:
an average feature vector calculation module: for calculating an average feature vector m for each class in a training samplej(j ═ 1, …, C), j denotes the jth class, C denotes a total of C different face classes;
the characteristic vector calculation module is used for inputting the face image y to be recognized and calculating the characteristic vector h thereofy=W-1y, W represent the base image matrix;
distance calculating module for calculating characteristic vector h of face image to be recognizedyAverage feature vector m to each classjIf h isyAnd mjIf the distance is the minimum, the face image y to be recognized is classified into the P-th class;
an output module: for outputting the category P.
The invention also discloses a computer-readable storage medium, in which a computer program is stored which, when being invoked by a processor, is configured to carry out the steps of the method as claimed in the claims.
Table 1 shows the recognition rate (%) of the Method proposed in this patent (Our Method) in ORL face database compared with Convex non-negative matrix factorization (Convex-NMF) and Semi-non-negative matrix factorization (Semi-NMF) (TN indicates the number of training samples per class)
TN | 3 | 4 | 5 | 6 | 7 | 8 | 9 |
Convex-NMF | 72.21 | 76.79 | 83.70 | 85.75 | 87.17 | 88.50 | 88.50 |
Semi-NMF | 75.93 | 80.92 | 85.85 | 86.00 | 89.00 | 90.13 | 89.00 |
Our Method | 78.29 | 82.92 | 88.05 | 88.56 | 91.83 | 91.63 | 91.75 |
Table 2 shows the comparison of the recognition speed(s) of the Method proposed by this patent (Our Method) with the Convex non-negative matrix factorization (Convex-NMF) and the Semi-non-negative matrix factorization (Semi-NMF) on the ORL face database (TN indicates the number of training samples per class)
TN | 3 | 4 | 5 | 6 | 7 | 8 | 9 |
Convex-NMF | 0.49 | 0.76 | 1.18 | 1.53 | 1.70 | 2.39 | 2.72 |
Semi-NMF | 0.53 | 0.63 | 0.95 | 1.08 | 1.09 | 1.24 | 1.38 |
Our Method | 0.46 | 0.56 | 0.88 | 0.89 | 0.90 | 1.09 | 1.21 |
The invention has the beneficial effects that:
1. the invention allows the input matrix to contain negative numbers, and expands the application range of NMF.
2. The patent proposes a new concept of the E-helper function, from which a basic theoretical framework and method of constructing a new helper function is presented.
3. The face recognition method has the advantages of high recognition performance and low calculation complexity.
4. An iterative formula with better recognition effect and faster convergence rate is obtained by constructing a new E auxiliary function.
5. The convergence of the algorithm provided by the patent is not only proved theoretically by using the auxiliary function, but also verified in experiments, and the algorithm has higher convergence.
6. The results of experiments and comparisons of the disclosed face database and related algorithms show that the method developed by the patent has certain superiority.
The foregoing is a more detailed description of the invention in connection with specific preferred embodiments and it is not intended that the invention be limited to these specific details. For those skilled in the art to which the invention pertains, several simple deductions or substitutions can be made without departing from the spirit of the invention, and all shall be considered as belonging to the protection scope of the invention.
Claims (3)
1. A face recognition method based on E auxiliary function semi-nonnegative matrix decomposition is characterized by comprising a training step, wherein the training step comprises the following steps:
the first step is as follows: converting the training sample image into a training sample matrix X, and setting an error threshold epsilon and a maximum iteration number ImaxAnd inputting a training sample matrix X, an error threshold epsilon and a maximum iteration number Imax;
The second step is as follows: initializing a base image matrix W and a coefficient matrix H;
the third step: setting the iteration number n to be 0;
the fourth step: updating the base image matrix W and the coefficient matrix H according to formula (6);
the fifth step: n is n + 1;
a sixth step: judging whether the objective function F (W, H) is less than or equal to epsilon or the iteration number n reaches the maximum iteration number ImaxIf yes, outputting a base image matrix W and a coefficient matrix H, otherwise, executing a fourth step; the face recognition method further comprises a recognition step which is executed after the training step, wherein the recognition step comprises the following steps:
a seventh step of: calculating the average characteristic vector m of each class in the training samplej(j ═ 1, …, c), j denotes the jth class, c denotes a total of c different face classes;
the eighth step is that the face image y to be recognized is input and the characteristic vector h of the face image y is calculatedy=W-1y, W represent the base image matrix;
the ninth step is that the characteristic vector h of the face image to be recognized is calculatedyAverage feature vector m to each classjIf h isyAnd mjIf the distance is the minimum, the face image y to be recognized is classified into the P-th class;
a tenth step: outputting the class P, thereby completing face recognition;
in the fourth step, equation (6) is as follows:
W←XHT(HHT)-1
in formula (6), W represents a base image matrix, H represents a coefficient matrix, X represents a training sample matrix, and T represents a transpose of the matrix;
the definition and theorem of the E auxiliary function is as follows:
defining: for arbitrary matrices H and H(t)If E (H, H)(t)) Satisfies the conditions
E(H,H(t)) Not less than 0, and E (H)(t),H(t))=0
Then called E (H, H)(t)) An E helper function that is a function f (H);
theorem: if the function G (H, H)(t)) Is an auxiliary function of the function f (H), if the function E (H, H)(t)) Is an E helper function for function f (H) then:
is still an auxiliary function of (f) (H), where λ ≧ 0;
the iterative formula derivation process for updating the coefficient matrix H is as follows:
sub-targeting function f2(H)=||X-WH||2Wherein W is fixed;
function(s)
Is f2(H) An auxiliary function of
Function(s)
E(H,H(t)) Not less than 0, and E (H)(t),H(t))=0
Thus, E (H, H) is known(t)) Is still f2(H) An E helper function of;
thenIs f2(H) An auxiliary function of, function G (H, H)(t)) Is a function f2(H) An auxiliary function of, and a function E (H, H)(t)) Is a function f2(H) Taking λ as 1, an auxiliary function of (1) can be directly derivedIs still f2(H) An auxiliary function of (2); to obtainSolving its derivative to 0, an updated iterative formula for H can be solved.
2. A face recognition system based on semi-non-negative matrix factorization of E-helper functions, comprising a training module comprising:
an input module: for converting training sample image into training sample matrix X, setting error threshold epsilon and maximum iteration number ImaxAnd inputting a training sample matrix X, an error threshold epsilon and a maximum iteration number Imax;
An initialization module: the method comprises the steps of initializing a base image matrix W and a coefficient matrix H;
an assignment module: setting the iteration number n to be 0;
an update module: for updating the base image matrix W and the coefficient matrix H according to equation (6);
a counting module: n is n + 1;
judgment ofA module: judging whether the objective function F (W, H) is less than or equal to epsilon or the iteration number n reaches the maximum iteration number ImaxIf yes, outputting a base image matrix W and a coefficient matrix H, otherwise, executing an updating module; the face recognition system further comprises a recognition module executed after the training module, the recognition module comprising:
an average feature vector calculation module: for calculating an average feature vector m for each class in a training samplej(j ═ 1, …, c), j denotes the jth class, c denotes a total of c different face classes;
the characteristic vector calculation module is used for inputting the face image y to be recognized and calculating the characteristic vector h thereofy=W-1y, W represent the base image matrix;
distance calculating module for calculating characteristic vector h of face image to be recognizedyAverage feature vector m to each classjIf h isyAnd mjIf the distance is the minimum, the face image y to be recognized is classified into the P-th class;
an output module: for outputting the class P, thereby completing face recognition;
in the update module, equation (6) is as follows:
W←XHT(HHT)-1
in formula (6), W represents a base image matrix, H represents a coefficient matrix, X represents a training sample matrix, and T represents a transpose of the matrix;
the definition and theorem of the E auxiliary function is as follows:
defining: for arbitrary matrices H and H(t)If E (H, H)(t)) Satisfies the conditions
E(H,H(t)) Not less than 0, and E (H)(t),H(t))=0
Then called E (H, H)(t)) An E helper function that is a function f (H);
theorem: if the function G (H, H)(t)) Is an auxiliary function of the function f (H)Number E (H, H)(t)) Is an E helper function for function f (H) then:
is still an auxiliary function of (f) (H), where λ ≧ 0;
the iterative formula derivation process for updating the coefficient matrix H is as follows:
sub-targeting function f2(H)=||X-WH||2Wherein W is fixed;
function(s)
Is f2(H) An auxiliary function of (2);
function(s)
E(H,H(t)) Not less than 0, and E (H)(t),H(t))=0
Thus, E (H, H) is known(t)) Is still f2(H) An E helper function of;
thenIs f2(H) An auxiliary function of, function G (H, H)(t)) Is a function f2(H) An auxiliary function of, and a function E (H, H)(t)) Is a function f2(H) Taking λ as 1, an auxiliary function of (1) can be directly derivedIs still f2(H) An auxiliary function of (2); to obtainSolving its derivative to 0, an updated iterative formula for H can be solved.
3. A computer-readable storage medium, characterized in that the computer-readable storage medium stores a computer program configured to, when invoked by a processor, implement the steps of the method of claim 1.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201811536236.1A CN109508697B (en) | 2018-12-14 | 2018-12-14 | Face recognition method, system and storage medium based on semi-nonnegative matrix factorization of E auxiliary function |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201811536236.1A CN109508697B (en) | 2018-12-14 | 2018-12-14 | Face recognition method, system and storage medium based on semi-nonnegative matrix factorization of E auxiliary function |
Publications (2)
Publication Number | Publication Date |
---|---|
CN109508697A CN109508697A (en) | 2019-03-22 |
CN109508697B true CN109508697B (en) | 2022-03-08 |
Family
ID=65752628
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201811536236.1A Active CN109508697B (en) | 2018-12-14 | 2018-12-14 | Face recognition method, system and storage medium based on semi-nonnegative matrix factorization of E auxiliary function |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN109508697B (en) |
Families Citing this family (6)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN110070028B (en) * | 2019-04-17 | 2023-03-07 | 深圳大学 | Method, system and storage medium for representing and identifying non-negative features of face image based on conjugate gradient method |
CN110210343B (en) * | 2019-05-20 | 2023-07-11 | 深圳供电局有限公司 | Big data face recognition method and system and readable storage medium thereof |
CN110378262B (en) * | 2019-07-08 | 2022-12-13 | 深圳大学 | Additive Gaussian kernel based kernel nonnegative matrix factorization face recognition method, device and system and storage medium |
WO2021003637A1 (en) * | 2019-07-08 | 2021-01-14 | 深圳大学 | Kernel non-negative matrix factorization face recognition method, device and system based on additive gaussian kernel, and storage medium |
WO2021042328A1 (en) * | 2019-09-05 | 2021-03-11 | 深圳大学 | Method, device, and system for presenting and recognizing non-negative features of facial data in self-configured cosine kernel space, and storage medium |
CN110705368B (en) * | 2019-09-05 | 2023-04-18 | 深圳大学 | Method, device and system for representing and identifying non-negative characteristics of face data in self-constructed cosine kernel space and storage medium |
Citations (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN102592148A (en) * | 2011-12-29 | 2012-07-18 | 华南师范大学 | Face identification method based on non-negative matrix factorization and a plurality of distance functions |
CN107341485A (en) * | 2017-07-28 | 2017-11-10 | 江汉大学 | Face identification method and device |
CN107480636A (en) * | 2017-08-15 | 2017-12-15 | 深圳大学 | Face identification method, system and storage medium based on core Non-negative Matrix Factorization |
Family Cites Families (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US8515879B2 (en) * | 2010-08-11 | 2013-08-20 | Seiko Epson Corporation | Supervised nonnegative matrix factorization |
US8498949B2 (en) * | 2010-08-11 | 2013-07-30 | Seiko Epson Corporation | Supervised nonnegative matrix factorization |
CN105893954B (en) * | 2016-03-30 | 2019-04-23 | 深圳大学 | A kind of Non-negative Matrix Factorization face identification method and system based on nuclear machine learning |
-
2018
- 2018-12-14 CN CN201811536236.1A patent/CN109508697B/en active Active
Patent Citations (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN102592148A (en) * | 2011-12-29 | 2012-07-18 | 华南师范大学 | Face identification method based on non-negative matrix factorization and a plurality of distance functions |
CN107341485A (en) * | 2017-07-28 | 2017-11-10 | 江汉大学 | Face identification method and device |
CN107480636A (en) * | 2017-08-15 | 2017-12-15 | 深圳大学 | Face identification method, system and storage medium based on core Non-negative Matrix Factorization |
Non-Patent Citations (4)
Title |
---|
Convex and Semi-Nonnegative Matrix Factorizations;Chris Ding等;《IEEE Transactions on Pattern Analysis and Machine Intelligence》;20100131;第32卷(第1期);全文 * |
Measuring the Degree of Face Familiarity Based on Extended NMF;CE ZHAN 等;《ACM Transactions on Applied Perception》;20130531;第10卷(第2期);全文 * |
基于非负矩阵分解算法的人脸识别技术的研究;刘昱昊;《中国博士学位论文全文数据库 信息科技辑》;20150115(第1期);全文 * |
改进的非负矩阵分解算法及其在人脸识别中的应用;邵强;《中国优秀硕士学位论文全文数据库 信息科技辑》;20160415(第4期);全文 * |
Also Published As
Publication number | Publication date |
---|---|
CN109508697A (en) | 2019-03-22 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN109508697B (en) | Face recognition method, system and storage medium based on semi-nonnegative matrix factorization of E auxiliary function | |
Cong et al. | Fast simulation of hyperplane-truncated multivariate normal distributions | |
Ho | Nonnegative matrix factorization algorithms and applications | |
Li et al. | A scalable end-to-end gaussian process adapter for irregularly sampled time series classification | |
Arora et al. | Stochastic optimization for PCA and PLS | |
Krishnapuram et al. | Sparse multinomial logistic regression: Fast algorithms and generalization bounds | |
WO2018149133A1 (en) | Method and system for face recognition by means of dictionary learning based on kernel non-negative matrix factorization, and sparse feature representation | |
CN110070028B (en) | Method, system and storage medium for representing and identifying non-negative features of face image based on conjugate gradient method | |
Li et al. | Classification approach based on non-negative least squares | |
Pernkopf et al. | Maximum margin Bayesian network classifiers | |
WO2020118708A1 (en) | E auxiliary function based semi-non-negative matrix factorization facial recognition method and system, and storage medium | |
Singla et al. | Fantastic four: Differentiable bounds on singular values of convolution layers | |
Conroy et al. | Fast, exact model selection and permutation testing for l2-regularized logistic regression | |
CN109002794B (en) | Nonlinear non-negative matrix factorization face recognition construction method, system and storage medium | |
Le Brigant et al. | Fisher-Rao geometry of Dirichlet distributions | |
Shen et al. | One-hot graph encoder embedding | |
Nguyen et al. | EPEM: efficient parameter estimation for multiple class monotone missing data | |
Nguyen | A Gyrovector space approach for symmetric positive semi-definite matrix learning | |
Darnell et al. | Adaptive randomized dimension reduction on massive data | |
Khouja et al. | Tensor decomposition for learning Gaussian mixtures from moments | |
Kong et al. | Efficient algorithms for selecting features with arbitrary group constraints via group lasso | |
Sun et al. | A stable approach for model order selection in nonnegative matrix factorization | |
Hosseini-Asl et al. | Nonnegative matrix factorization for document clustering: A survey | |
Zhao et al. | A block coordinate descent approach for sparse principal component analysis | |
CN110378262B (en) | Additive Gaussian kernel based kernel nonnegative matrix factorization face recognition method, device and system and storage medium |
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 |