CN110837804A - Face identification method for sparse mixed dictionary learning - Google Patents

Face identification method for sparse mixed dictionary learning Download PDF

Info

Publication number
CN110837804A
CN110837804A CN201911082514.5A CN201911082514A CN110837804A CN 110837804 A CN110837804 A CN 110837804A CN 201911082514 A CN201911082514 A CN 201911082514A CN 110837804 A CN110837804 A CN 110837804A
Authority
CN
China
Prior art keywords
dictionary
class
sparse
data
training
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Pending
Application number
CN201911082514.5A
Other languages
Chinese (zh)
Inventor
狄岚
矫慧文
顾雨迪
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Jiangnan University
Original Assignee
Jiangnan University
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by Jiangnan University filed Critical Jiangnan University
Priority to CN201911082514.5A priority Critical patent/CN110837804A/en
Publication of CN110837804A publication Critical patent/CN110837804A/en
Pending legal-status Critical Current

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V40/00—Recognition of biometric, human-related or animal-related patterns in image or video data
    • G06V40/10—Human or animal bodies, e.g. vehicle occupants or pedestrians; Body parts, e.g. hands
    • G06V40/16—Human faces, e.g. facial parts, sketches or expressions
    • G06V40/172—Classification, e.g. identification
    • 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
    • G06F18/2135—Feature extraction, e.g. by transforming the feature space; Summarisation; Mappings, e.g. subspace methods based on approximation criteria, e.g. principal component analysis
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/24—Classification techniques
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/28—Determining representative reference patterns, e.g. by averaging or distorting; Generating dictionaries

Abstract

The invention discloses a feature extraction method based on sparse mixed dictionary learning, which is applied to face identification, and improves the accuracy of the face identification to a certain extent; and then, extracting category commonality by using the intra-class difference dictionary, capturing the same characteristics of different categories, and finally combining the category characteristic dictionary with the intra-class difference dictionary. The results of experiments on the face databases such as AR, CMU-PIE, LFW and the like by using the method show that the method can obtain higher identification precision under the condition of less sample training.

Description

Face identification method for sparse mixed dictionary learning
Technical Field
The invention relates to the technical field of computer image processing, in particular to image feature extraction and classification.
Background
With the innovation of computer hardware technology and the development of software technology, face identification is gradually applied to different fields of economic engineering, social security and the like. The face identification method based on deep learning has high recognition rate, but depends on the requirement of a large number of data samples, expensive hardware definition equipment and training time of several days. Compared with the face identification training method, the face identification training based on sparse representation is simple, has strong robustness to noise, and has attracted wide attention of domestic and foreign scholars in recent years.
In 2009, j.wrigh et al proposed Sparse Representation-based classification (SRC). The method is based on an illumination model, supposing that any test sample can be represented by the reconstruction of the training sample set, and classifying the test sample by selecting the minimum reconstruction error, wherein the sparse representation method is introduced into the field of face identification for the first time. Then, many scholars have proposed an improvement on the basis of SRC. Zhang et al changes the constraint term in the SRC method into a CRC method, and the complexity of solving sparse coding is reduced while the recognition rate is ensured, so that the running speed of the method is obviously improved. Aharon et al generalize K-means clustering according to the principle of minimum error, and propose a K-SVD method to perform SVD on error terms.
The dictionaries in the methods are all directly formed by training data, and have the defects of insufficient dictionary distinguishability and sensitivity to noise. Aiming at dictionary distinguishability, Sprechmann et al propose an idea of identifying sub-dictionaries, and construct corresponding sub-dictionaries for each type of data by using sparse coding. Jiang et al add tag consistency constraints to the K-SVD method, improving dictionary discrimination.
Aiming at the code distinguishability, Yang et al propose an FDDL method by taking a Fisher criterion as a model, enhance the relevance between dictionary atoms and training labels, and restrict large inter-class variance and small intra-class variance of sparse coding. In 2014, Cai et al proposed an SVGDL method to adaptively determine the weight of each coding vector pair, so that a better classification effect can be obtained when training samples are insufficient. In 2018, Zhou et al propose a new sample expansion method, and Li et al introduce a kernel method into a cooperative neighbor method to enhance the classification capability of sparse coding.
While identifying sub-dictionaries can effectively extract distinctiveness and distinctiveness between categories, commonalities between different categories are not considered. In 2018, Wang et al propose significant feature extraction and shared dictionary construction, Li et al propose a CSICVDL method, construct an intra-class dictionary by using auxiliary data, and capture the same features among different classes.
Disclosure of Invention
This section is for the purpose of summarizing some aspects of embodiments of the invention and to briefly introduce some preferred embodiments. In this section, as well as in the abstract and the title of the invention of this application, simplifications or omissions may be made to avoid obscuring the purpose of the section, the abstract and the title, and such simplifications or omissions are not intended to limit the scope of the invention.
The present invention is proposed in view of the problems of requiring a large number of data samples, expensive hardware definition, and training time as long as several days in the above-mentioned face identification process.
Therefore, the invention provides a face identification method for sparse mixed dictionary learning, which can obtain higher identification precision under the condition of less sample training.
In order to solve the technical problems, the invention provides the following technical scheme: acquiring a face picture, and performing downsampling and dimensionality reduction to obtain a training sample set; constructing a class characteristic dictionary model by taking a Fisher criterion and a Laplace matrix as constraints; the class characteristic dictionary model is utilized to independently learn a sub-dictionary for each class of samples, so that the particularity among classes of the training sample set is extracted, the intra-class coding dispersion is reduced while the similarity of sparse coding data is kept, and the inter-class coding dispersion is increased; constructing an intra-class difference dictionary model, and learning a dictionary for all samples together, thereby extracting the class commonality of the training sample set and capturing the same characteristics of different classes; combining dictionaries respectively learned by the category characteristic dictionary model and the intra-category difference dictionary model into a new dictionary, and reserving sparse codes of the category characteristic dictionary to obtain a sparse mixed dictionary model; and inputting the face image data to be recognized into the sparse mixed dictionary model for residual calculation, completing recognition according to the minimum residual value of each type of face image, and obtaining a classification label through a classifier.
As a preferred scheme of the face identification method for sparse mixed dictionary learning according to the present invention, wherein: the method comprises the training step of the class characteristic dictionary model, and a training sample set A is defined: a ═ A1,A2,…,AK}∈Rm×NWherein each column of A represents an m-dimensional vector,
Figure BDA0002264390760000022
for the training data of the i-th class,
Figure BDA0002264390760000021
the total number of training samples; training data a ═ a1,A2,…,AKThe feature vectors of the class dictionary D are initialized to the atoms of the dictionary, and each class of the class dictionary D is normalized to l2Norm is 1; fixing the category dictionary D, and updating a sparse coefficient X; fixing the sparse coefficient X, and updating the category dictionary D; and repeating the cycle for two times until the value of the function of the two times reaches a threshold value or the maximum iteration number.
As a preferred scheme of the face identification method for sparse mixed dictionary learning according to the present invention, wherein: training the in-class difference dictionary model, defining standard data N, changing data X and initializing in-class dictionary DS(ii) a Fixed dictionary DSUpdate the sparse coefficient αi,βiFixed sparse coefficient αi,βiUpdating dictionary DS(ii) a And repeating the cycle for two times until the value of the function of the two times reaches a threshold value or the maximum iteration number.
As a preferred scheme of the face identification method for sparse mixed dictionary learning according to the present invention, wherein: the class dictionary D and the in-class dictionary D obtained by trainingSCombining to obtain the sparse mixed dictionary; inputting the face image data to be recognized into the sparse mixed dictionary to calculate the minimum residual value of each type of face image; and inputting the minimum residual value into a classifier to obtain a corresponding identified classification label.
As a preferred scheme of the face identification method for sparse mixed dictionary learning according to the present invention, wherein: the function for which the class-feature dictionary model is proposed is as follows,
Figure BDA0002264390760000031
wherein A isiRepresenting the i-th class of training samples, D representing the global dictionary, DiDenotes the i-th class dictionary, XiRepresenting the sparse coefficients reconstructed by the dictionary,
Figure BDA00022643907600000311
representation dictionary DiAnd (5) reconstructing sparse coefficients.
As a preferred scheme of the face identification method for sparse mixed dictionary learning according to the present invention, wherein: the class characteristic dictionary model comprises a function updating step, wherein when the class dictionary D is fixed and the sparse coefficient X is updated; the objective function is converted into:
Figure BDA0002264390760000032
when the sparse coefficient X is fixed and the dictionary D is updated, updating the dictionary D by adopting a one-by-one updating method, namely when the jth sub-dictionary is updated, defaulting other sub-dictionaries Dj(i ≠ j) is updated; the objective function is converted into:
Figure BDA0002264390760000033
whereinXjAll data A is represented by a dictionary DjA sparse coefficient of representation; thus Q (D)i) Can be converted into:
Figure BDA0002264390760000035
wherein
Figure BDA0002264390760000036
In the same way, update one by one
Figure BDA0002264390760000037
Atom in (2), when updated
Figure BDA0002264390760000038
When the default is that other atoms are updated; updating
Figure BDA0002264390760000039
The steps and formula of (a) are as follows:
definition of Zi＝[z(1)；…；z(m)]Wherein z is(k)Is ZiLine k of (1), order
Figure BDA00022643907600000310
Thus:
Figure BDA0002264390760000041
order to
Figure BDA0002264390760000042
Obtaining:
Figure BDA0002264390760000043
at this time
Figure BDA0002264390760000044
I.e. DiUpdating the value of the kth atom, and finally obtaining the normalized value and the norm value of the dictionary
Figure BDA0002264390760000045
Normalization:
Figure BDA0002264390760000046
as a preferred embodiment of the face identification method for sparse mixed dictionary learning according to the present invention, the function of the intra-class difference dictionary model is proposed as follows,
Figure BDA0002264390760000047
wherein, a common k-class data training intra-class difference dictionary is defined, and standard data are as follows:
Figure BDA0002264390760000048
the change data is:
the standard data and the variation data are collectively called auxiliary data C, where DS∈Rm×rIs an in-class difference dictionary and is provided with a plurality of different dictionaries,
Figure BDA00022643907600000410
sparse coefficients of standard invariant data, βi∈Rr×1Is as a quilt DSAnd (5) reconstructing sparse coefficients.
As a preferred scheme of the face identification method for sparse mixed dictionary learning according to the present invention, wherein: the intra-class difference dictionary model comprises the following function updating step when a fixed dictionary DSUpdate the sparse coefficient αi,βiTime, update sparse coefficient αi,βiThe function of (d) is as follows:
Figure BDA00022643907600000412
wherein Di＝[Ni,DS]，ri＝[αi,βi]
When the sparse coefficient α is fixedi,βiDictionary DSWhen the current is over; the objective function is converted into:
Figure BDA00022643907600000413
definition of Ri＝Xi-Niαi，R＝[Ri，K，Rk]，
Figure BDA00022643907600000414
The objective function is converted into:
Figure BDA00022643907600000415
when updating dictionary atom djWhen the atom is updated, the rest atoms are assumed to be updated; therefore, there are:
Figure BDA00022643907600000416
definition of Y ═ R-Sigmai≠jdiβ (i, using Lagrange multiplier method; updating djThe steps and formula of (a) are as follows:
Figure BDA0002264390760000051
wherein γ is a scalar;
order to
Figure BDA0002264390760000052
Thus, there are: di＝Yβ(j,:)T(β(j,:)β(j,:)T-γ)-1
At this time djI.e. DSUpdating the value of the jth atom, and finally obtaining d to ensure the normalization and norm requirements of the dictionaryjNormalization: dj＝Yβ(j,:)T/||Yβ(j,:)T||2At this time, dictionary atom djAfter the updating is finished, all dictionary atoms are updated step by step to obtain the intra-class difference dictionary DS。
As a preferred scheme of the face identification method for sparse mixed dictionary learning according to the present invention, wherein: the class label acquisition includes the steps of,
Figure BDA0002264390760000053
in which the test data y is stored in a memory,
Figure BDA0002264390760000054
sparse coding representing the reconstruction of the class feature dictionary D,representing intra-class difference dictionaries DSSparse coding of the reconstruction; y from the i-th class dictionary DiThe error of reconstruction is:
Figure BDA0002264390760000056
wherein
Figure BDA0002264390760000057
Representation class sub-dictionary DiSparse coding of the reconstruction, w is used to balance the reconstruction term and constraint term,
the label for y is then: identification (y) argmini{ei}。
As a preferred scheme of the face identification method for sparse mixed dictionary learning according to the present invention, wherein: the Laplace matrix is a matrix based on graph theory, in order to better convert sparse representation into graph theory problem, an undirected graph G is defined as { V, E }, a vertex set V represents each sample, and a weighted edge represents similarity between sample codesqrAnd representing the weight between the vertexes q and r, wherein the weight formula is as follows according to the mutual k neighbor principle:
Figure BDA0002264390760000058
wherein x isq∈Nk(xr) Denotes xqIs xrK is a kernel bandwidth parameter, and an adjacency matrix W is defined as (W)qr) The sum of the weights u of all edges adjacent to a vertexqExpressed as:
Figure BDA0002264390760000061
a plurality of u form a degree matrix: u-diag (∑)q≠rwqr)
The idea of the fisher criterion is to project samples to a proper projection axis, so that the distance between the projection points of the same type of samples is as small as possible, and the distance between the projection points of the different type of samples is as large as possible. For sparse coding X, u is defined0Represents the center of all sparse coefficients uiRepresenting various sparse coding mean vectors:
Figure BDA0002264390760000062
within-class divergence matrix Sw(X), inter-class divergence matrix SB(X) can be defined as:
Figure BDA0002264390760000063
the invention has the beneficial effects that: the invention provides a face identification method for learning a sparse mixed dictionary, which is characterized in that a sparse mixed dictionary is obtained by utilizing training data to perform experiments, and classification labels are obtained under the operation of a classifier, so that the method can obtain higher-accuracy identification under fewer samples, the time is greatly shortened, unnecessary definition backup configuration is reduced, and the efficiency is improved.
Drawings
In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings needed to be used in the description of the embodiments will be briefly introduced below, and it is obvious that the drawings in the following description are only some embodiments of the present invention, and it is obvious for those skilled in the art to obtain other drawings based on these drawings without inventive exercise. Wherein:
fig. 1 is a schematic flow chart of a face identification method for sparse mixed dictionary learning according to the present invention.
FIG. 2 is a schematic diagram of an AR face database test of a face identification method for sparse hybrid dictionary learning according to the present invention;
FIG. 3 is a schematic diagram of a CMU-PIE database face test of a face identification method for sparse mixed dictionary learning according to the present invention;
FIG. 4 is a schematic diagram of an LFW database experimental sample of the face identification method for sparse mixed dictionary learning according to the present invention;
FIG. 5 is a schematic diagram of another experimental sample of the face identification method for sparse mixed dictionary learning according to the present invention;
FIG. 6 is a schematic diagram of a variation curve of dimensional influence accuracy of the face identification method for sparse mixed dictionary learning according to the present invention;
FIG. 7 is a schematic diagram of a variation curve of weight influence accuracy of the face identification method for sparse mixed dictionary learning according to the present invention;
FIG. 8 shows λ of a face identification method for sparse mixed dictionary learning according to the present invention1A schematic diagram of a variation curve affecting accuracy;
FIG. 9 shows λ of a face identification method for sparse mixed dictionary learning according to the present invention2A schematic diagram of a variation curve affecting accuracy;
fig. 10 is a schematic diagram of an accuracy rate change curve under macro-average of the face identification method for sparse mixed dictionary learning according to the present invention.
Detailed Description
In order to make the aforementioned objects, features and advantages of the present invention comprehensible, specific embodiments accompanied with figures are described in detail below, and it is apparent that the described embodiments are a part of the embodiments of the present invention, not all of the embodiments. All other embodiments, which can be obtained by a person skilled in the art without making creative efforts based on the embodiments of the present invention, shall fall within the protection scope of the present invention.
In the following description, numerous specific details are set forth in order to provide a thorough understanding of the present invention, but the present invention may be practiced in other ways than those specifically described and will be readily apparent to those of ordinary skill in the art without departing from the spirit of the present invention, and therefore the present invention is not limited to the specific embodiments disclosed below.
Furthermore, reference herein to "one embodiment" or "an embodiment" means that a particular feature, structure, or characteristic described in connection with the embodiment is included in at least one implementation of the invention. The appearances of the phrase "in one embodiment" in various places in the specification are not necessarily all referring to the same embodiment, nor are separate or alternative embodiments mutually exclusive of other embodiments.
The present invention will be described in detail with reference to the drawings, wherein the cross-sectional views illustrating the structure of the device are not enlarged partially in general scale for convenience of illustration, and the drawings are only exemplary and should not be construed as limiting the scope of the present invention. In addition, the three-dimensional dimensions of length, width and depth should be included in the actual fabrication.
Meanwhile, in the description of the present invention, it should be noted that the terms "upper, lower, inner and outer" and the like indicate orientations or positional relationships based on the orientations or positional relationships shown in the drawings, and are only for convenience of describing the present invention and simplifying the description, but do not indicate or imply that the referred device or element must have a specific orientation, be constructed in a specific orientation and operate, and thus, cannot be construed as limiting the present invention. Furthermore, the terms first, second, or third are used for descriptive purposes only and are not to be construed as indicating or implying relative importance.
The terms "mounted, connected and connected" in the present invention are to be understood broadly, unless otherwise explicitly specified or limited, for example: can be fixedly connected, detachably connected or integrally connected; they may be mechanically, electrically, or directly connected, or indirectly connected through intervening media, or may be interconnected between two elements. The specific meanings of the above terms in the present invention can be understood in specific cases to those skilled in the art.
Example 1
Referring to fig. 1, a schematic flow chart of a face identification method for sparse mixed dictionary learning is provided for a first embodiment of the present invention, and as shown in fig. 1, the face identification method for sparse mixed dictionary learning includes obtaining a face picture, performing downsampling and dimensionality reduction as a training sample set, constructing a class-specific dictionary model with fisher criterion and laplacian matrix as constraints, and using the class-specific dictionary model to learn a sub-dictionary for each class of samples individually, thereby extracting the specificity among classes of the training sample set, reducing intra-class coding dispersion while preserving the similarity of sparse coding data, and increasing inter-class coding dispersion; constructing an intra-class difference dictionary model, learning a dictionary for all samples together, extracting the class commonality of the training sample set, capturing the same characteristics of different classes, combining the dictionaries respectively learned by the class characteristic dictionary model and the intra-class difference dictionary model into a new dictionary, and reserving class characteristic dictionary sparse codes to obtain a sparse mixed dictionary model; and inputting the face image data to be recognized into the sparse mixed dictionary model for residual calculation, completing recognition according to the minimum residual value of each type of face image, and obtaining a classification label through a classifier.
Specifically, the face identification method for sparse mixed dictionary learning comprises a training step of a class characteristic dictionary model, wherein a training sample set A is defined as follows: a ═ A1,A2,…,AK}∈Rm×NWherein each column of A represents an m-dimensional vector,for the training data of the i-th class,
Figure BDA0002264390760000081
the total number of training samples; training data a ═ a1,A2,…,AKThe feature vectors of the class dictionary D are initialized to the atoms of the dictionary, and each class of the class dictionary D is normalized to l2Norm is 1; fixing the category dictionary D, and updating a sparse coefficient X; fixing the sparse coefficient X, and updating the category dictionary D; and repeating the cycle for two times until the value of the function of the two times reaches a threshold value or the maximum iteration number. Sparse hybrid dictionaryThe face identification method also comprises the training step of the in-class difference dictionary model, namely defining standard data N, changing data X and initializing an in-class dictionary DS(ii) a Fixed dictionary DSUpdate the sparse coefficient αi,βiFixed sparse coefficient αi,βiUpdating dictionary DS(ii) a And repeating the cycle for two times until the value of the function of the two times reaches a threshold value or the maximum iteration number.
Preferably, the sparse mixed dictionary is obtained by combining the class dictionary D and the class dictionary Ds which are obtained by utilizing the class characteristic dictionary model and the class difference dictionary model during training; the category characteristic dictionary model maintains the similarity of sparse coding data, simultaneously reduces intra-category coding dispersion, increases inter-category coding dispersion, enlarges the information difference of different categories and improves the discriminativity of the dictionary and the sparse coding; the intra-class difference dictionary model captures the same characteristics of different classes, and the discrimination capability of the method for the test samples under the conditions of illumination, shielding change and the like is enhanced on the basis of keeping the data common characteristics; and the data of the face image to be recognized is input into the sparse mixed dictionary to calculate the minimum residual value of each type of face image, and the corresponding recognized classification label is obtained by utilizing the distinguishing performance in the class characteristic dictionary model and the same feature identification capability of different classes in the intra-class difference dictionary model.
It should be recognized that embodiments of the present invention can be realized and implemented by computer hardware, a combination of hardware and software, or by computer instructions stored in a non-transitory computer readable memory. The methods may be implemented in a computer program using standard programming techniques, including a non-transitory computer-readable storage medium configured with the computer program, where the storage medium so configured causes a computer to operate in a specific and predefined manner, according to the methods and figures described in the detailed description. Each program may be implemented in a high level procedural or object oriented programming language to communicate with a computer system. However, the program(s) can be implemented in assembly or machine language, if desired. In any case, the language may be a compiled or interpreted language. Furthermore, the program can be run on a programmed application specific integrated circuit for this purpose.
Further, the operations of processes described herein can be performed in any suitable order unless otherwise indicated herein or otherwise clearly contradicted by context. The processes described herein (or variations and/or combinations thereof) may be performed under the control of one or more computer systems configured with executable instructions, and may be implemented as code (e.g., executable instructions, one or more computer programs, or one or more applications) collectively executed on one or more processors, by hardware, or combinations thereof. The computer program includes a plurality of instructions executable by one or more processors.
Further, the method may be implemented in any type of computing platform operatively connected to a suitable interface, including but not limited to a personal computer, mini computer, mainframe, workstation, networked or distributed computing environment, separate or integrated computer platform, or in communication with a charged particle tool or other imaging device, and the like. Aspects of the invention may be embodied in machine-readable code stored on a non-transitory storage medium or definition device, whether removable or integrated into a computing platform, such as a hard disk, optically read and/or write storage medium, RAM, ROM, or the like, such that it may be read by a programmable computer, which when read by the storage medium or definition device, may be used to configure and operate the computer to perform the procedures described herein. Further, the machine-readable code, or portions thereof, may be transmitted over a wired or wireless network. The invention described herein includes these and other different types of non-transitory computer-readable storage media when such media include instructions or programs that implement the steps described above in conjunction with a microprocessor or other data processor. The invention also includes the computer itself when programmed according to the methods and techniques described herein. A computer program can be applied to input data to perform the functions described herein to transform the input data to generate output data that is stored to non-volatile memory. The output information may also be applied to one or more output definitions such as a display. In a preferred embodiment of the invention, the transformed data represents physical and tangible objects, including particular visual depictions of physical and tangible objects produced on a display.
Example 2
Referring to fig. 2 to 10, a second embodiment of the present invention, which is different from the first embodiment, is: the experimental environment is a 64-bit Window 10 operating system, a memory 32GB, an Intel (R) Xeon (R) CPU E5-2620 v4@2.10GHZ and is realized by MatlabR2016b software programming. The experimental images are subjected to standardization processing, a CMU-PIE face database, an AR face database and an LFW face database are selected for experiments, and the comparison method comprises the following steps: SRC, FDDL, CRC, SVGDL and CSICVDL. 1. AR database experiments: and randomly selecting 100 persons from the AR face database to perform experiments, dividing each person into 5 sets, referring to FIG. 2, taking two faces without illumination expression change as training pictures, and dividing the rest into 4 sets to be respectively used as test pictures. Set S1Test data containing expression changes; set S2Test data containing illumination changes; set S3Test data of the glasses occlusion; set S4Test data for scarf shading.
Table 1 experimental results of method on AR library (%)
Figure BDA0002264390760000101
In the experiment, 80 persons are selected as a training set and a testing set, the other 20 persons are used for training an intra-class difference dictionary, the down-sampling of each picture is 60 multiplied by 80, and the sample data is reduced to 160 dimensions by adopting PCA. The recognition rate of each method in the AR database is shown in table 1. As can be seen from table 1, the best classification results were obtained with scheme four under different sets. The third scheme is suitable for testing the conditions of illumination change and sunglasses shielding of the sample, and the second scheme is stable in performance and good in performance when the expression change and the scarf shielding of the test sample exist. CSICVDL, the recognition rate of the method is higher than FDDL, and the necessity of learning the intra-class difference dictionary is illustrated. 2. CMU-PIE database experiments: experiment 4760 pictures of each person were selected from 68 persons in the CMU-PIE database. Each person will beThe 70 pictures are divided into 4 sets, the training samples are 2 posture front and normal lighting pictures, and the testing samples are divided into 4 sets. Referring to FIG. 3, S1For 10 turning pictures, S2For 38 front face containing partial lighting and shading change pictures, S310 lower pictures, S410 head-up pictures. The experiment selected 18 persons as auxiliary data, the remaining 50 persons were used for training and testing, the pictures were reduced in dimension by PCA, and the experimental results are shown in table 2.
TABLE 2 experimental results of the method on the CMU-PIE library (%)
Figure BDA0002264390760000102
As can be seen from table 2, the scheme four obtains the optimal classification effect under each set. For set S2～S4The method has high classification accuracy and is suitable for set S1The classification accuracy of each method is low, and the primary speculation is caused by the fact that important information is lost in the turning pictures. 3. LFW database experiments: the method selects an unlimited face database LFW for experiment, and utilizes 3d correction to fill up the feature information lost due to steering and shielding. 139 persons with more than 10 single pictures are selected as experimental data. 10 pictures of each person are used for carrying out experiments, 5 pictures are used as training samples, and the rest are used as test samples. FIG. 4 shows a training sample and a test sample of one of the persons. In order to verify the influence of the intra-class difference dictionary on the method, 19, 39, 59 and 79 individuals are randomly selected as auxiliary data in experiments and are respectively used as S1、S2、S3、S4The four sets, the rest for training and testing, are compared with the basic FDDL method and the CSSVDL method including the auxiliary dictionary, and the recognition rate of each method on each set is shown in Table 3. As can be seen from table 3, as the auxiliary data of the intra-structure-class difference dictionary increases, the recognition rate of the method increases. On the non-limited face database, the four-classification effect of the scheme is optimal, and the scheme has strong data dependency and unstable performance.
Table 3 results of experiments on LFW library by method (%)
database
Figure BDA0002264390760000111
Further, referring to fig. 2 to 4, experiments show that, among the four schemes, the scheme four has better stability and recognition rate than the other schemes, so the scheme four is used as the text method. To further verify the robustness of the method, the influence of each parameter on the classification accuracy is analyzed. In the experiment, 100 persons in the AR face database are selected as experimental data, 80 persons are randomly selected for training and testing, and 20 persons are used for constructing auxiliary data. Referring to fig. 5, 2 normal illumination and expression pictures are selected from each person by a training sample, and 5 pictures containing expressions and illumination changes are selected from each person by a testing sample. (1) Influence of dimensionality on accuracy: to explore the influence of data dimensionality on accuracy, we performed experiments taking the original 120 × 165 picture and downsampling 60 × 80, 30 × 40 pictures, and reduced the dimensionality to 100, 150, 200, 250, 300, 350, 400, 450, 500, respectively. Referring to fig. 6, the overall discrimination accuracy of the method for the face picture is enhanced as the sampling resolution is improved. With the increase of the dimensionality reduction of the PCA, the classification accuracy of the method under each resolution condition is firstly increased and then reduced. In the case of both resolutions 30 x 40, 60 x 80, the present method identifies the highest accuracy at a dimensionality reduction of 300, while in the case of resolution 120 x 165, the present method identifies the highest accuracy at a dimensionality reduction of 250. (2) influence of weights on accuracy: to investigate the influence of the weight w on the accuracy, a sample 30 x 40 was taken and tested, PCA was used to reduce the dimensions to 100, 200, 300, 400, 500, respectively, and the weights w were taken to be 0.1, 0.2, 0.3, 0.4, 0.5, 0.6, 0.7, 0.8, 0.9, respectively, referring to fig. 7, the identification accuracy of the method herein increased with increasing PCA dimension reduction. The general trend that the recognition rate of the method changes along with the weight under different dimensions is the same, when the weight is 0.1-0.5, the accuracy rate is increased along with the increase of the weight, and when the weight is 0.5-0.9, the accuracy rate of the method tends to be stable. (3) Parameter lambda1、λ2Influence on accuracy: face identification method for sparse mixed dictionary learning provided by the text needs definitionSetting two constraint parameters, and evaluating the parameter lambda by 5 times of verification1、λ2Defining the verification subset as {0.1,0.05,0.01,0.005,0.001 }. To explore the parameter lambda1、λ2For the influence of accuracy, we take the 30 × 40 samples and perform experiments to reduce the dimensions to 100, 200, 300, 400, and 500 dimensions using PCA, respectively. In FIGS. 8 and 9 is λ1、λ2The influence of 0.001, 0.005, 0.01, 0.05 and 0.1 on the accuracy of the experiment is taken. Referring to fig. 8 and 9, the recognition rate of the method according to the invention is a function of λ for different dimensions1、λ2The general trend of change of value is the same, wherein lambda1、λ2The highest accuracy is achieved when the value is 0.05. (4) complexity analysis: the complexity of the method is calculated by updating sparse coding and dictionary. Defining the number of training samples as n, the characteristic dimension of the samples as q, and the time complexity for updating the sparse coefficient as nO (q)2nr) Wherein r ≧ 1.2 is a constant. Time complexity of updating dictionary is sigmajnjO (2nq), wherein n isjRepresents DiThe number of atoms of (c). Thus, the overall complexity of the methods herein is:
nO(q2nr)+∑jnjO(2nq)
(5) and (3) multivariate classification evaluation: in the actual binary classification problem, four cases occur for the predicted value and the actual value: true Positive (TP), False Positive (FP), False Negative (FN), True Negative (TN). Besides the accuracy, the common classification evaluation indexes include Precision (P), Recall (R), and comprehensive evaluation index (F1 measure, F). In the multivariate classification evaluation, the macroaveraging (Macro-averaging) calculation formula is as follows:
Figure BDA0002264390760000121
Figure BDA0002264390760000122
Figure BDA0002264390760000123
wherein the content of the first and second substances,
Figure BDA0002264390760000124
and randomly selecting 120 LFW face databases after 3d correction for experiment, wherein 10 pictures and 5 pictures of each person are used as training samples, and the rest are used as test samples.
Table 4 results of experiments on LFW library by method (%)
Figure BDA0002264390760000125
As can be seen from Table 4 and FIG. 10, the accuracy and F1 index of the method are superior to those of other methods under macro-average, and the ROC curve is more convex than other curves, so that the classification effect is better.
Specifically, as shown in fig. 1, a face identification method for sparse mixed dictionary learning includes obtaining a face picture, performing downsampling and dimensionality reduction to obtain a training sample set, constructing a class characteristic dictionary model by using fisher criterion and a laplace matrix as constraints, and learning a sub-dictionary for each class of samples by using the class characteristic dictionary model, so as to extract the particularity among classes of the training sample set, reduce intra-class coding dispersion while preserving the similarity of sparse coding data, and increase inter-class coding dispersion; constructing an intra-class difference dictionary model, learning a dictionary for all samples together, extracting the class commonality of the training sample set, capturing the same characteristics of different classes, combining the dictionaries respectively learned by the class characteristic dictionary model and the intra-class difference dictionary model into a new dictionary, and reserving class characteristic dictionary sparse codes to obtain a sparse mixed dictionary model; and inputting the face image data to be recognized into the sparse mixed dictionary model for residual calculation, completing recognition according to the minimum residual value of each type of face image, and obtaining a classification label through a classifier.
Further, the face identification method for sparse mixed dictionary learning comprises a training step of a class characteristic dictionary model,defining a training sample set A: a ═ A1,A2,…,AK}∈Rm×NWherein each column of A represents an m-dimensional vector,for the training data of the i-th class,
Figure BDA0002264390760000127
the total number of training samples; training data a ═ a1,A2,…,AKThe feature vectors of the class dictionary D are initialized to the atoms of the dictionary, and each class of the class dictionary D is normalized to l2Norm is 1; fixing the category dictionary D, and updating a sparse coefficient X; fixing the sparse coefficient X, and updating the category dictionary D; and repeating the cycle for two times until the value of the function of the two times reaches a threshold value or the maximum iteration number. The face identification method for sparse mixed dictionary learning further comprises the training step of the in-class difference dictionary model, wherein standard data N, change data X and an initialization in-class dictionary D are definedS(ii) a Fixed dictionary DSUpdate the sparse coefficient αi,βiFixed sparse coefficient αi,βiUpdating dictionary DS(ii) a And repeating the cycle for two times until the value of the function of the two times reaches a threshold value or the maximum iteration number.
Preferably, the identification method can improve the accuracy of face identification under the condition of few samples, and provides a basis for optimizing complexity and accurately classifying; in the experiment, different testing methods are respectively adopted to compare which is more suitable for experiment conditions, on the basis of obtaining suitable conditions, a class characteristic dictionary model and an intra-class difference dictionary model are combined to perform the experiment, and a classifier is used for obtaining a classification label; the experiment is the verification of the face identification method for sparse mixed dictionary learning, and the effectiveness, feasibility and accuracy of the method are proved.
It should be noted that the above-mentioned embodiments are only for illustrating the technical solutions of the present invention and not for limiting, and although the present invention has been described in detail with reference to the preferred embodiments, it should be understood by those skilled in the art that modifications or equivalent substitutions may be made on the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, which should be covered by the claims of the present invention.

Claims (10)

1. A face identification method for sparse mixed dictionary learning is characterized in that: comprises the following steps of (a) carrying out,
acquiring a face picture, and performing downsampling and dimensionality reduction to obtain a training sample set;
constructing a class characteristic dictionary model by taking a Fisher criterion and a Laplace matrix as constraints;
the class characteristic dictionary model is utilized to independently learn a sub-dictionary for each class of samples, so that the particularity among classes of the training sample set is extracted, the intra-class coding dispersion is reduced while the similarity of sparse coding data is kept, and the inter-class coding dispersion is increased;
constructing an intra-class difference dictionary model, and learning a dictionary for all samples together, thereby extracting the class commonality of the training sample set and capturing the same characteristics of different classes;
combining dictionaries respectively learned by the category characteristic dictionary model and the intra-category difference dictionary model into a new dictionary, and reserving sparse codes of the category characteristic dictionary to obtain a sparse mixed dictionary model;
and inputting the face image data to be recognized into the sparse mixed dictionary model for residual calculation, completing recognition according to the minimum residual value of each type of face image, and obtaining a classification label through a classifier.
2. The sparse mixed dictionary learning face identification method of claim 1, wherein: comprising a training step of the class feature dictionary model,
defining a training sample set A: a ═ A1,A2,…,AK}∈Rm×NWherein each column of A represents an m-dimensional vector,
Figure FDA0002264390750000011
for the training data of the i-th class,
Figure FDA0002264390750000012
the total number of training samples;
training data a ═ a1,A2,…,AKThe feature vectors of the class dictionary D are initialized to the atoms of the dictionary, and each class of the class dictionary D is normalized to l2Norm is 1;
fixing the category dictionary D, and updating a sparse coefficient X;
fixing the sparse coefficient X, and updating the category dictionary D;
and repeating the cycle for two times until the value of the function of the two times reaches a threshold value or the maximum iteration number.
3. The sparse mixed dictionary learning face identification method of claim 1 or 2, wherein: comprising a training step of the intra-class difference dictionary model,
defining standard data N, change data X and initializing dictionary in class DS；
Fixed dictionary DSUpdate the sparse coefficient αi,βi；
Fixed sparse coefficient αi,βiUpdating dictionary DS；
And repeating the cycle for two times until the value of the function of the two times reaches a threshold value or the maximum iteration number.
4. The sparse mixed dictionary learning face identification method as claimed in claim 3, comprising the steps of,
the class dictionary D and the in-class dictionary D obtained by trainingSCombining to obtain the sparse mixed dictionary;
inputting the face image data to be recognized into the sparse mixed dictionary to calculate the minimum residual value of each type of face image;
and inputting the minimum residual value into a classifier to obtain a corresponding identified classification label.
5. The sparse mixed dictionary learning face identification method of claim 4, wherein: the function for which the class-feature dictionary model is proposed is as follows,
Figure FDA0002264390750000021
wherein A isiRepresenting the i-th class of training samples, D representing the global dictionary, DiDenotes the i-th class dictionary, XiRepresenting the sparse coefficients reconstructed by the dictionary,
Figure FDA0002264390750000022
representation dictionary DiAnd (5) reconstructing sparse coefficients.
6. The sparse mixed dictionary learning face identification method of claim 5, wherein: the category characteristic dictionary model includes a function updating step of,
when the class dictionary D is fixed and the sparse coefficient X is updated;
the objective function is converted into:
Figure FDA0002264390750000023
when the sparse coefficient X is fixed and the dictionary D is updated, updating the dictionary D by adopting a one-by-one updating method, namely when the jth sub-dictionary is updated, defaulting other sub-dictionaries Dj(i ≠ j) is updated;
the objective function is converted into:
Figure FDA0002264390750000024
wherein
Figure FDA0002264390750000025
XjAll data A is represented by a dictionary DjA sparse coefficient of representation;
thus Q (D)i) Can be converted into:
wherein
Figure FDA0002264390750000027
In the same way, update one by one
Figure FDA0002264390750000028
Atom in (2), when updated
Figure FDA0002264390750000029
When the default is that other atoms are updated;
updatingThe steps and formula of (a) are as follows:
definition of Zi＝[z(1)；…；z(m)]Wherein z is(k)Is ZiLine k of (1), order
Figure FDA00022643907500000211
Thus:
Figure FDA0002264390750000031
order to
Figure FDA0002264390750000032
Obtaining:
at this timeI.e. DiUpdating the value of the kth atom, and finally obtaining the normalized value and the norm value of the dictionary
Figure FDA0002264390750000035
Normalization:
Figure FDA0002264390750000036
7. the face identification method for sparse mixed dictionary learning as claimed in claim 5 or 6, wherein: the function for the intra-class difference dictionary model is proposed as follows,
Figure FDA0002264390750000037
wherein, a common k-class data training intra-class difference dictionary is defined, and standard data are as follows:
Figure FDA0002264390750000038
the change data is:
Figure FDA0002264390750000039
the standard data and the variation data are collectively called auxiliary data C, where DS∈Rm×rIs an in-class difference dictionary and is provided with a plurality of different dictionaries,
Figure FDA00022643907500000310
sparse coefficients of standard invariant data, βi∈Rr×1Is as a quilt DSAnd (5) reconstructing sparse coefficients.
8. The sparse mixed dictionary learning face identification method as claimed in claim 7, wherein: the intra-class difference dictionary model includes a function update step,
when fixing dictionary DSUpdate the sparse coefficient αi,βiWhen the current is over;
updating sparse coefficients αi,βiThe function of (d) is as follows:
Figure FDA00022643907500000311
wherein Di＝[Ni,DS]，ri＝[αi,βi]
When the sparse coefficient α is fixedi,βiDictionary DSWhen the current is over;
the objective function is converted into:
Figure FDA00022643907500000312
definition of Ri＝Xi-Niαi,R＝[Ri,K,Rk],
Figure FDA00022643907500000313
The objective function is converted into:
Figure FDA00022643907500000314
when updating dictionary atom djWhen the atom is updated, defining that the rest atoms are updated;
therefore, there are:
Figure FDA0002264390750000041
definition of Y ═ R-Sigmai≠jdiβ (i, i), using the Lagrangian multiplier method;
update djThe steps and formula of (a) are as follows:
Figure FDA0002264390750000042
wherein γ is a scalar;
order to
Figure FDA0002264390750000043
Thus, there are: di＝Yβ(j,:)T(β(j,:)β(j,:)T-γ)-1
At this time djI.e. DSUpdating the value of the jth atom, and finally obtaining d to ensure the normalization and norm requirements of the dictionaryjNormalization: dj＝Yβ(j,:)T/||Yβ(j,:)T||2At this time, dictionary atom djAfter the updating is finished, all dictionary atoms are updated step by step to obtain the intra-class difference dictionary DS。
9. The sparse mixed dictionary learning face identification method of claim 8, wherein: the class label acquisition includes the steps of,
Figure FDA0002264390750000044
in which the test data y is stored in a memory,
Figure FDA0002264390750000045
sparse coding representing the reconstruction of the class feature dictionary D,representing intra-class difference dictionaries DSSparse coding of the reconstruction; y from the i-th class dictionary DiThe error of reconstruction is:
Figure FDA0002264390750000047
wherein
Figure FDA0002264390750000048
Representation class sub-dictionary DiSparse coding of the reconstruction, w forThe reconstruction term and the constraint term are balanced,
the label for y is then: identification (y) argmini{ei}。
10. The sparse mixed dictionary learning face identification method of claim 1, wherein:
the Laplace matrix is a matrix based on graph theory, in order to better convert sparse representation into graph theory problem, an undirected graph G is defined as { V, E }, a vertex set V represents each sample, and weighted edges represent similarity between sample codesqrAnd representing the weight between the vertexes q and r, wherein the weight formula is as follows according to the mutual k neighbor principle:
Figure FDA0002264390750000051
wherein x isq∈Nk(xr) Denotes xqIs xrK is a kernel bandwidth parameter, and an adjacency matrix W is defined as (W)qr) The sum of the weights u of all edges adjacent to a vertexqExpressed as:
Figure FDA0002264390750000052
a plurality of u form a degree matrix: u-diag (∑)q≠rwqr)
The idea of the Fisher criterion is to project samples to a proper projection axis, so that the distance between projection points of the same type of samples is as small as possible, the distance between projection points of different types of samples is as large as possible, and for sparse coding X, u is defined0Represents the center of all sparse coefficients uiRepresenting various sparse coding mean vectors:
within-class divergence matrix Sw(X), inter-class divergence matrix SB(X) can be defined as:
Figure FDA0002264390750000054
。
CN201911082514.5A 2019-11-07 2019-11-07 Face identification method for sparse mixed dictionary learning Pending CN110837804A (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN201911082514.5A CN110837804A (en) 2019-11-07 2019-11-07 Face identification method for sparse mixed dictionary learning

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN201911082514.5A CN110837804A (en) 2019-11-07 2019-11-07 Face identification method for sparse mixed dictionary learning

Publications (1)

Publication Number Publication Date
CN110837804A true CN110837804A (en) 2020-02-25

Family

ID=69576371

Family Applications (1)

Application Number Title Priority Date Filing Date
CN201911082514.5A Pending CN110837804A (en) 2019-11-07 2019-11-07 Face identification method for sparse mixed dictionary learning

Country Status (1)

Country Link
CN (1) CN110837804A (en)

Cited By (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN111339960A (en) * 2020-02-28 2020-06-26 江南大学 Face recognition method based on discrimination low-rank regression model
CN111783615A (en) * 2020-06-28 2020-10-16 南京工程学院 Non-constrained face recognition method based on weighted block tensor sparse graph mapping

Citations (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN104318261A (en) * 2014-11-03 2015-01-28 河南大学 Graph embedding low-rank sparse representation recovery sparse representation face recognition method
CN105844223A (en) * 2016-03-18 2016-08-10 常州大学 Face expression algorithm combining class characteristic dictionary learning and shared dictionary learning
CN108985177A (en) * 2018-06-21 2018-12-11 南京师范大学 A kind of facial image classification method of the quick low-rank dictionary learning of combination sparse constraint

Patent Citations (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN104318261A (en) * 2014-11-03 2015-01-28 河南大学 Graph embedding low-rank sparse representation recovery sparse representation face recognition method
CN105844223A (en) * 2016-03-18 2016-08-10 常州大学 Face expression algorithm combining class characteristic dictionary learning and shared dictionary learning
CN108985177A (en) * 2018-06-21 2018-12-11 南京师范大学 A kind of facial image classification method of the quick low-rank dictionary learning of combination sparse constraint

Non-Patent Citations (3)

* Cited by examiner, † Cited by third party
Title
MIAO ZHENG ET AL.: "Graph Regularized Sparse Coding for Image Representation", 《IEEE TRANSACTIONS ON IMAGE PROCESSING》 *
WENJING LI ET AL.: "Adaptive face representation via class-specific and intra-class variation dictionaries for recognition", 《MULTIMED TOOLS APPL》 *
魏彩娜 等: "域间F-范数正则化迁移谱聚类方法", 《计算机科学与探索》 *

Cited By (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN111339960A (en) * 2020-02-28 2020-06-26 江南大学 Face recognition method based on discrimination low-rank regression model
CN111339960B (en) * 2020-02-28 2023-08-11 江南大学 Face recognition method based on discrimination low-rank regression model
CN111783615A (en) * 2020-06-28 2020-10-16 南京工程学院 Non-constrained face recognition method based on weighted block tensor sparse graph mapping
CN111783615B (en) * 2020-06-28 2023-08-22 南京工程学院 Unconstrained face recognition method based on weighting block tensor sparse map mapping

Similar Documents

Publication Publication Date Title
CN109522956B (en) Low-rank discriminant feature subspace learning method
CN110659665B (en) Model construction method of different-dimension characteristics and image recognition method and device
CN110674748A (en) Image data processing method, image data processing device, computer equipment and readable storage medium
JP4618098B2 (en) Image processing system
Wang et al. Expression of Concern: Facial feature discovery for ethnicity recognition
Díaz et al. An algorithm for an accurate detection of anomalies in hyperspectral images with a low computational complexity
CN113360701B (en) Sketch processing method and system based on knowledge distillation
JP4376145B2 (en) Image classification learning processing system and image identification processing system
CN105528620B (en) method and system for combined robust principal component feature learning and visual classification
CN108229295A (en) A kind of figure based on indication of multiple local constraint optimizes Dimensionality Reduction method
Wang et al. Penalized fisher discriminant analysis and its application to image-based morphometry
CN111428557A (en) Method and device for automatically checking handwritten signature based on neural network model
CN108460400A (en) A kind of hyperspectral image classification method of combination various features information
CN110837804A (en) Face identification method for sparse mixed dictionary learning
CN113158777A (en) Quality scoring method, quality scoring model training method and related device
CN111325275A (en) Robust image classification method and device based on low-rank two-dimensional local discriminant map embedding
CN111652138A (en) Face recognition method, device and equipment for wearing mask and storage medium
JP6166981B2 (en) Facial expression analyzer and facial expression analysis program
Tu et al. Feature extraction using multidimensional spectral regression whitening for hyperspectral image classification
CN112966629B (en) Remote sensing image scene classification method based on image transformation and BoF model
CN111723759B (en) Unconstrained face recognition method based on weighted tensor sparse graph mapping
Dong et al. A supervised dictionary learning and discriminative weighting model for action recognition
CN114882288B (en) Multi-view image classification method based on hierarchical image enhancement stacking self-encoder
CN110287973A (en) A kind of image characteristic extracting method based on low-rank robust linear discriminant analysis
CN110717547A (en) Learning algorithm based on regression hypergraph

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
RJ01 Rejection of invention patent application after publication
RJ01 Rejection of invention patent application after publication

Application publication date: 20200225