CN111340111A - Method for recognizing face image set based on wavelet kernel extreme learning machine - Google Patents

Method for recognizing face image set based on wavelet kernel extreme learning machine Download PDF

Info

Publication number
CN111340111A
CN111340111A CN202010119515.9A CN202010119515A CN111340111A CN 111340111 A CN111340111 A CN 111340111A CN 202010119515 A CN202010119515 A CN 202010119515A CN 111340111 A CN111340111 A CN 111340111A
Authority
CN
China
Prior art keywords
image set
learning machine
wavelet kernel
limit learning
wavelet
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Granted
Application number
CN202010119515.9A
Other languages
Chinese (zh)
Other versions
CN111340111B (en
Inventor
郝丽秀
于威威
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.)
Shanghai Maritime University
Original Assignee
Shanghai Maritime 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 Shanghai Maritime University filed Critical Shanghai Maritime University
Priority to CN202010119515.9A priority Critical patent/CN111340111B/en
Publication of CN111340111A publication Critical patent/CN111340111A/en
Application granted granted Critical
Publication of CN111340111B publication Critical patent/CN111340111B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Images

Classifications

    • 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
    • G06F18/21—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)
  • Data Mining & Analysis (AREA)
  • Physics & Mathematics (AREA)
  • General Physics & Mathematics (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Evolutionary Biology (AREA)
  • General Engineering & Computer Science (AREA)
  • Evolutionary Computation (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Bioinformatics & Computational Biology (AREA)
  • Artificial Intelligence (AREA)
  • Bioinformatics & Cheminformatics (AREA)
  • Health & Medical Sciences (AREA)
  • General Health & Medical Sciences (AREA)
  • Human Computer Interaction (AREA)
  • Oral & Maxillofacial Surgery (AREA)
  • Multimedia (AREA)
  • Image Analysis (AREA)
  • Image Processing (AREA)

Abstract

The invention discloses a method for recognizing a face image set based on a wavelet kernel extreme learning machine, which comprises the following steps: s1, inputting an image set; s2, preprocessing the image set; s3, modeling the image set: initializing a global wavelet kernel limit learning machine model; s4, training the global wavelet kernel limit learning machine model by using the training image set to obtain a wavelet kernel limit learning machine model of each type of image set; s5, respectively reconstructing the original test image set by using the wavelet kernel limit learning machine model corresponding to each type of image set, and outputting the reconstructed test image set; s6, calculating a reconstruction error between the reconstructed test image set and the original test image set; s7, obtaining the minimum reconstruction error, obtaining the category to which the minimum reconstruction error belongs, wherein the category to which the minimum reconstruction error belongs represents the category represented by the test image set; and S8, outputting the category to which the test image set belongs. The invention can efficiently realize the classification of the face image set, and has high learning speed and good generalization performance.

Description

Method for recognizing face image set based on wavelet kernel extreme learning machine
Technical Field
The invention relates to the field of face image set identification, in particular to a method for identifying a face image set based on a wavelet Kernel Extreme Learning Machine (KELM).
Background
With the development of artificial intelligence, face recognition develops rapidly, which is particularly reflected in that from the initial single face image recognition to the nearest face image set recognition, the face has various angles and illumination changes due to video or photographing, so that the attention points of face recognition change. Image set identification may describe individuals in more detail. Meanwhile, the face image set identification also faces many challenges, such as how to define the distance between two individual face image sets and how to extract more effective individual features.
A series of studies on face image set recognition have appeared in recent decades, and one of the main difficulties is how to effectively model and extract features that can effectively express individuals. The existing methods greatly improve the recognition rate of the face image set, such as methods of establishing manifold models, establishing manifold subspaces and the like. Hu et al define the distance between sets as the distance between their Sparse Approximate Nearest Points (SANPs). For the representation of the image set on the manifold, suitable distance measures are used, such as geodesic distances and projection kernel measures on the Grassmann manifold, and log-mapped distance measures on the riemann manifold. However, the method of manifold modeling of an image requires that the manifold class of a face image be presupposed on the manifold of the image in advance. But it is assumed that the face data follows a gaussian distribution, such as Wang and Chen, which proposes Manifold Discriminant Analysis (MDA), modeling each image set with multiple local linear clusters; wang et al directly uses modeling of covariance matrices for each image set. They map the covariance matrix of each image set from the Riemann manifold to Euclidean space through a kernel function based on Log Euclidean distances. The image set is then classified according to learning a regression function using a kernel partial least squares method. In practice this does not apply to all images. Or other methods require the prior assumption that the image set can be represented by a linear subspace, such as Harandi et al model the image set of the linear subspace as points on a grassmann manifold. They define kernel functions that map the image set from the grassmann manifold to the euclidean space, where classification is by Graph Embedding Discriminant Analysis (GEDA); hayat et al learn the structure of each image set using a deep learning model. The labels of the test set are then estimated according to the minimum reconstruction error and the majority voting scheme. In general, structure-based algorithms require relatively more images in each set to accurately model the data structure. However, data may be located on complex manifolds, and to model more complex data structures, many methods have also been proposed to model image sets as convex or affine hulls of data samples.
Since the prior art is conceptually similar to the nearest neighbor classification (KNN) and certain constraints must be imposed to avoid finding neighboring points in certain low-dimensional spaces where the image sets may intersect. However, modeling more complex image set structures comes at the cost of increased algorithm complexity. Therefore, these algorithms cannot efficiently handle the large image set classification task. Therefore, it is necessary to develop a method for recognizing a face image set based on a wavelet Kernel Extreme Learning Machine (KELM).
Disclosure of Invention
The invention aims to provide a method for recognizing a face image set based on a wavelet Kernel Extreme Learning Machine (KELM), which realizes the classification and recognition of the face image set by using the KELM and can efficiently realize the classification of the face image set on the premise of not assuming a data structure in advance.
In order to achieve the purpose, the invention is realized by the following technical scheme:
a face image set recognition method based on a wavelet kernel limit learning machine comprises the following steps: s1, inputting an image set, wherein the image set is divided into a plurality of image sets; s2, preprocessing the input image set; s3, modeling the image set: initializing a global wavelet kernel limit learning machine model; s4, training the global wavelet kernel limit learning machine model by using the training image set to obtain a trained wavelet kernel limit learning machine model of each type of image set; s5, after obtaining the wavelet kernel limit learning machine model of each type of image set after training, respectively reconstructing the original test image set before reconstruction by using the wavelet kernel limit learning machine model after training corresponding to each type of image set, and outputting the reconstructed test image set; s6, calculating a reconstruction error between the reconstructed test image set and the original test image set; s7, acquiring the minimum reconstruction error, and acquiring the category to which the minimum reconstruction error belongs, wherein the category to which the minimum reconstruction error belongs represents the category represented by the test image set; and S8, outputting the category to which the test image set belongs.
Preferably, the pre-treatment comprises: graying of an image: the graying process means that the RGB values of each pixel point are unified into the same value, and the grayed image is changed into a single channel from three channels; feature extraction: and extracting the characteristic vector of the grayed face image by using an HOG-NMF method.
Preferably, the training image set is selected from at least a portion of the input image set, the original test image set being the remaining portion of the input image set; the input image set is divided into N classes, and each class of training image set is marked as
Figure BDA0002392492060000031
Each type of test image is collected as
Figure BDA0002392492060000032
Wherein d represents the characteristic dimension of each type of face image in the training image set or the test image set, and i represents the ith type of face.
Preferably, the step S3 further includes: defining an initial global wavelet kernel limit learning machine KELM model, and recording the weight of the adopted global KELM model as
Figure BDA0002392492060000033
Wherein h represents the number of hidden nodes;
Figure BDA0002392492060000034
and representing the weight of the ith hidden node of the human face image set G in the global wavelet kernel limit learning machine model.
Preferably, the step S4 further includes: taking the face image of each type of test image set as the input of an initial global wavelet kernel limit learning machine and performing comparisonTraining is carried out, and a wavelet kernel limit learning machine model of each type of image set after training is obtained and recorded as
Figure BDA0002392492060000035
LjA wavelet kernel limit learning machine model representing the jth class face image set, wherein the weight of the jth class wavelet kernel limit learning machine model is defined as
Figure BDA0002392492060000036
Preferably, in step S5, each test image set is
Figure BDA0002392492060000037
Models using respective classes of image sets
Figure BDA0002392492060000038
Respectively reconstructing: taking each type of face image set in the test image set as corresponding each wavelet core extreme learning machine model
Figure BDA0002392492060000039
Respectively obtaining N reconstructed test image sets and recording the N reconstructed test image sets as N wavelet core extreme learning machine models obtained by training
Figure BDA00023924920600000310
Preferably, the step S6 further includes: calculating the reconstruction error between the test image set before reconstruction and the test image set after reconstruction as a squared Euclidean distance:
Figure BDA00023924920600000311
wherein, i represents the ith class of image,
Figure BDA00023924920600000312
set of images representing the ith class before reconstruction
Figure BDA00023924920600000313
And the reconstructed image set
Figure BDA00023924920600000314
The squared euclidean distance between.
Preferably, the step S7 further includes: test image set
Figure BDA00023924920600000315
Class Y of (2) depends on the minimum reconstruction error
Figure BDA00023924920600000316
The reconstruction error is minimum, the closer to a certain class, the test image class label is obtained
Figure BDA00023924920600000317
Compared with the prior art, the invention has the beneficial effects that: the invention provides a novel image set identification method, which uses a wavelet kernel extreme learning machine to identify a face image set, although the image set has nonlinear characteristics, the method does not need a plurality of assumptions of a data set structure or select an algorithm related to a structure; in addition, the kernel extreme learning machine has the characteristics of high learning speed and good generalization performance.
Drawings
FIG. 1 is a flow chart of a method for identifying a face image set according to the present invention;
fig. 2-3 are schematic diagrams illustrating the principle of training and testing a face image set by using a wavelet kernel extreme learning machine according to the present invention.
Detailed Description
To make the objects, technical solutions and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the drawings in the embodiments of the present invention, and it is obvious that the described embodiments are only a part of the embodiments of the present invention, and not all of the embodiments. All other embodiments, which can be derived by a person skilled in the art from the embodiments given herein without making any creative effort, shall fall within the protection scope of the present invention.
As shown in fig. 1-3, the present invention provides a method for recognizing a face image set based on a wavelet Kernel Extreme Learning Machine (KELM), the method comprising the following steps:
s1, inputting an image set; classifying an input image set, and selecting images in the input image set in a testing process to serve as a training image set and a testing image set of a subsequent step; for example, the classification here is classified according to a plurality of face types, for example, the image set includes N persons, each person includes a plurality of image sets, and the number of classes of the classification may be considered to be N at this time, that is, the input image set is divided into N types of face image sets.
S2, preprocessing the image set input in the step S1;
s3, modeling the image set: initializing a global wavelet kernel limit learning machine model;
s4, training the global wavelet kernel limit learning machine model by using a training image set, and respectively obtaining a trained wavelet kernel limit learning machine model of each type of image set;
s5, after obtaining the trained wavelet kernel limit learning machine model of each type of image set, respectively reconstructing the test image set (the original test image set before reconstruction) by using the wavelet kernel limit learning machine model of each type of image set, and outputting the reconstructed test image set;
s6, calculating the reconstruction error between the output reconstructed test image set and the original test image set;
s7, obtaining the minimum reconstruction error and obtaining the category to which the minimum reconstruction error belongs; wherein the category to which the minimum reconstruction error belongs represents a category represented by the test image set;
and S8, outputting the category to which the test image set belongs.
In step S1, the method further includes:
(1.1) using an ORL image set for recognition, the ORL data set comprising N (e.g., 40) individuals, each person having M (e.g., M ═ 10) images;
(1.2) randomly selecting M1 (for example, M1 ≦ 6, and M1 ≦ M) images of each person as a training image set (training set), so that there are N × M1 (for example, N × M1 ≦ 40 × 6 ≦ 240) training images, where each training image set is denoted as
Figure BDA0002392492060000051
Illustratively, when the image set contains 40 types of face image sets, each type of training image set is recorded as
Figure BDA0002392492060000052
d represents the characteristic dimension of each type of face image in the training set, and i represents the face of the ith type.
(1.3) the remaining M2 (M2-M1, e.g., 10-6-4) images of each person are selected as a test image set (test set for short), and then N × M2 (e.g., N × M2-40 × 4-160) test images are available, where each test image set is designated as a test image set
Figure BDA0002392492060000053
Illustratively, when the image set contains 40 types of face image sets, each type of test image set is marked as
Figure BDA0002392492060000054
d represents the characteristic dimension of each type of face image in the test set, and i represents the face of the ith type.
(1.4) firstly, using the wavelet kernel extreme learning machine after random weight parameters as an initial model, and after the initial model is available, using the training sets to train the weight parameters of different training sets to obtain models of the wavelet kernel extreme learning machine of different training sets; and the test set is used to test the accuracy of the current model.
In step S2, the preprocessing of the face image set, such as graying the image and extracting features, further includes the following steps:
the method comprises the steps of preprocessing a face image set, namely graying the face, wherein the graying process means that RGB values of each pixel point are unified into the same value, an image after graying is changed from three channels into a single channel, data processing of the single channel is simpler than that of the three channels, then characteristic vectors are extracted from the face image after graying by using a HOG-NMF (Histogram of oriented Gradient-Non-negative Matrix Factorization) method (namely, after the characteristic vectors are extracted by using the HOG, the characteristics which can most represent the face are extracted by using the NMF method), so that characteristic dimensions are reduced, the calculation complexity is reduced, the calculation speed is improved, and the identification method is more accurate.
In step S3, modeling the preprocessed face image set, further includes the following steps:
firstly, an initial global wavelet kernel limit learning machine model (KELM model) is defined, and the global KELM model weight L is usedGIs marked as
Figure BDA0002392492060000055
h represents the number of hidden nodes; the model comprises an input layer, a hidden layer and an output layer, wherein the weights between the input layer and the hidden layer are initialized randomly, and the weights before the hidden layer and the output layer are initialized randomly; wherein,
Figure BDA0002392492060000061
and representing the weight of the ith hidden node of the human face image set G in the global wavelet kernel limit learning machine model.
The step S4 further includes: the face image of each type of test image set is used as the input of the initial global wavelet core limit learning machine model, the training is carried out on the face image, the trained wavelet core limit learning machine model of each type of image set is obtained respectively, and the model is marked as
Figure BDA0002392492060000062
Wherein L isjA wavelet kernel limit learning machine model representing the jth class face image set, wherein the weight of the jth class wavelet kernel limit learning machine model is defined as
Figure BDA0002392492060000063
In a specific embodiment, the training image set includes 40 classes of face image sets, N is 40, and each class includes 6 face images (i.e. as a test image set of each class), the face images of each class of test image set are used as the input of an initial global wavelet kernel limit learning machine, a gradient descent method is used to minimize a loss function until the loss function cannot be descended, a wavelet kernel limit learning machine model of each class of face image set is obtained, and due to the existence of 40 classes of face images, the number of the wavelet kernel limit learning machine models is 40, and these models are recorded as the models
Figure BDA0002392492060000064
In step S5, the method further includes:
each type of test image set
Figure BDA0002392492060000065
Model using corresponding class of face image set
Figure BDA0002392492060000066
Respectively reconstructing, namely taking each type of human face image set in the test image set as each wavelet kernel limit learning machine model
Figure BDA0002392492060000067
The input of (1) obtaining N wavelet kernel extreme learning machine models after modeling, and respectively obtaining N outputs after model operation, namely N reconstructed test image sets which are recorded as
Figure BDA0002392492060000068
I.e. to the original test image set
Figure BDA0002392492060000069
Can be obtained after reconstruction
Figure BDA00023924920600000610
Namely, it is
Figure BDA00023924920600000611
Through
Figure BDA00023924920600000612
Obtaining correspondences after model reconstruction
Figure BDA00023924920600000613
As a specific embodiment, the image set comprises a 40-class face image set, and if N is 40, the image set is tested
Figure BDA00023924920600000614
Model using corresponding each type of face image set
Figure BDA00023924920600000615
Respectively reconstructing, namely taking each type of human face image set in the test image set as each wavelet kernel limit learning machine model
Figure BDA00023924920600000616
Because 40 wavelet kernel extreme learning machine models are obtained after modeling and training, after model operation, each face image set can obtain one output, and 40 types of faces can obtain 40 outputs and are recorded as
Figure BDA00023924920600000617
The reconstructed test image set is obtained
Figure BDA00023924920600000618
In step S6, the method further includes:
to test the sample
Figure BDA00023924920600000619
The reconstruction error of (2) is calculated as the squared euclidean distance:
Figure BDA00023924920600000620
i represents the image of the i-th class,
Figure BDA00023924920600000621
set of images representing the ith class before reconstruction
Figure BDA00023924920600000622
And the reconstructed image set
Figure BDA00023924920600000623
The squared euclidean distance between. Each image set in the test image set has 4 images, for example, the reconstructed output of one image set and the non-reconstructed test image set are calculated by calculating the squared Euclidean distance
Figure BDA0002392492060000071
Similarly, reconstruction errors can be obtained after each type of image is reconstructed
Figure BDA0002392492060000072
In step S7, the method further includes:
test specimen
Figure BDA0002392492060000073
Class Y of (2) depends on the minimum reconstruction error
Figure BDA0002392492060000074
According to distance
Figure BDA0002392492060000075
Judging which classification the test set belongs to, wherein the judgment is carried out according to the minimum reconstruction error: the error is minimized, namely the closer to a certain class, the test image class label is obtained
Figure BDA0002392492060000076
In step S8, the method further includes:
reconstruction error
Figure BDA0002392492060000077
The smallest of them is recorded as
Figure BDA0002392492060000078
And further obtaining a category i to which the image of the test set belongs, wherein the category i represents which category the image of the test set belongs to, namely completing classification, and identifying the category to which the test face image set belongs.
In the present invention, the algorithm of the kernel limit learning machine for face image set classification is as follows:
inputting:
the face image set G comprises N types of faces, and each type of face image set is
Figure BDA0002392492060000079
Class label is Y ═ Ytrain}NTesting a set of human face images as
Figure BDA00023924920600000710
(II) outputting: test set XtestClass Y to which it belongs;
(III) training:
Figure BDA00023924920600000711
(initialize the global wavelet kernel limit learning machine model);
for j＝1:N do
Figure BDA00023924920600000712
(training wavelet kernel limit learning machine models for each class of face image set);
end for
and (IV) testing:
for
Figure BDA00023924920600000713
do
for j＝1:N do
Figure BDA00023924920600000714
(wherein，LjA wavelet kernel-limit learning model representing a j-th class of face image set,
Figure BDA00023924920600000715
representing the ith class of image set in the test image set,
Figure BDA00023924920600000716
to represent
Figure BDA00023924920600000717
Passing through a model LjOutputting the image set obtained after the reconstruction,
Figure BDA0002392492060000081
wavelet kernel formula representing a set of images);
Figure BDA0002392492060000082
end for
sample(s)
Figure BDA0002392492060000083
The category (2):
Figure BDA0002392492060000084
end for
class of test set:
Figure BDA0002392492060000085
in summary, the image set identification method provided by the invention uses a wavelet kernel limit learning machine to identify a face image set, and although the image set has nonlinear characteristics, the method provided by the invention does not need many assumptions of the structure of the data set or selects an algorithm related to the structure; in addition, the kernel extreme learning machine has the characteristics of high learning speed and good generalization performance.
While the present invention has been described in detail with reference to the preferred embodiments, it should be understood that the above description should not be taken as limiting the invention. Various modifications and alterations to this invention will become apparent to those skilled in the art upon reading the foregoing description. Accordingly, the scope of the invention should be determined from the following claims.

Claims (8)

1. A face image set recognition method based on a wavelet kernel limit learning machine is characterized by comprising the following steps:
s1, inputting an image set, wherein the image set is divided into a plurality of image sets;
s2, preprocessing the input image set;
s3, modeling the image set: initializing a global wavelet kernel limit learning machine model;
s4, training the global wavelet kernel limit learning machine model by using the training image set to obtain a trained wavelet kernel limit learning machine model of each type of image set;
s5, after obtaining the wavelet kernel limit learning machine model of each type of image set after training, respectively reconstructing the original test image set before reconstruction by using the wavelet kernel limit learning machine model after training corresponding to each type of image set, and outputting the reconstructed test image set;
s6, calculating a reconstruction error between the reconstructed test image set and the original test image set;
s7, acquiring the minimum reconstruction error, and acquiring the category to which the minimum reconstruction error belongs, wherein the category to which the minimum reconstruction error belongs represents the category represented by the test image set;
and S8, outputting the category to which the test image set belongs.
2. The method for recognizing a face image set based on a wavelet kernel limit learning machine according to claim 1,
the pretreatment comprises the following steps:
graying of an image: the graying process means that the RGB values of each pixel point are unified into the same value, and the grayed image is changed into a single channel from three channels;
feature extraction: and extracting the characteristic vector of the grayed face image by using an HOG-NMF method.
3. The method for recognizing a face image set based on a wavelet kernel limit learning machine according to claim 1,
the training image set is selected from at least a portion of the input image set, the original test image set being the remaining portion of the input image set;
the input image set is divided into N classes, and each class of training image set is marked as
Figure FDA0002392492050000011
Each type of test image is collected as
Figure FDA0002392492050000012
Wherein d represents the characteristic dimension of each type of face image in the training image set or the test image set, and i represents the ith type of face.
4. The method for recognizing a face image set based on a wavelet kernel limit learning machine according to claim 3,
the step S3 further includes:
defining an initial global wavelet kernel limit learning machine KELM model, and recording the weight of the adopted global KELM model as
Figure FDA0002392492050000021
Wherein h represents the number of hidden nodes;
Figure FDA0002392492050000022
and representing the weight of the ith hidden node of the human face image set G in the global wavelet kernel limit learning machine model.
5. The method for recognizing a face image set based on a wavelet kernel limit learning machine according to claim 4,
the step S4 further includes: each one is to beThe human face image of the class test image set is used as the input of the initial global wavelet core limit learning machine model and trained, and the trained wavelet core limit learning machine model of each class of image set is obtained and recorded as
Figure FDA0002392492050000023
LjA wavelet kernel limit learning machine model representing the jth class face image set, wherein the weight of the jth class wavelet kernel limit learning machine model is defined as
Figure FDA0002392492050000024
6. The method for recognizing a face image set based on a wavelet kernel limit learning machine according to claim 5,
in the step S5, each type of test image set
Figure FDA0002392492050000025
Models using respective classes of image sets
Figure FDA0002392492050000026
Respectively reconstructing: taking each type of face image set in the test image set as corresponding each wavelet core extreme learning machine model
Figure FDA0002392492050000027
Respectively obtaining N reconstructed test image sets and recording the N reconstructed test image sets as N wavelet core extreme learning machine models obtained by training
Figure FDA0002392492050000028
7. The method for recognizing a face image set based on a wavelet kernel limit learning machine according to claim 6,
the step S6 further includes:
calculating the reconstruction error between the test image set before reconstruction and the test image set after reconstruction as a squared Euclidean distance:
Figure FDA0002392492050000029
wherein, i represents the ith class of image,
Figure FDA00023924920500000210
set of images representing the ith class before reconstruction
Figure FDA00023924920500000211
And the reconstructed image set
Figure FDA00023924920500000212
The squared euclidean distance between.
8. The method for recognizing a set of human faces based on a wavelet kernel limit learning machine according to claim 7,
the step S7 further includes:
test image set
Figure FDA0002392492050000031
Class Y of (2) depends on the minimum reconstruction error
Figure FDA0002392492050000032
The reconstruction error is minimum, the closer to a certain class, the test image class label is obtained
Figure FDA0002392492050000033
CN202010119515.9A 2020-02-26 2020-02-26 Method for recognizing face image set based on wavelet kernel extreme learning machine Active CN111340111B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202010119515.9A CN111340111B (en) 2020-02-26 2020-02-26 Method for recognizing face image set based on wavelet kernel extreme learning machine

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202010119515.9A CN111340111B (en) 2020-02-26 2020-02-26 Method for recognizing face image set based on wavelet kernel extreme learning machine

Publications (2)

Publication Number Publication Date
CN111340111A true CN111340111A (en) 2020-06-26
CN111340111B CN111340111B (en) 2023-03-24

Family

ID=71181881

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202010119515.9A Active CN111340111B (en) 2020-02-26 2020-02-26 Method for recognizing face image set based on wavelet kernel extreme learning machine

Country Status (1)

Country Link
CN (1) CN111340111B (en)

Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN101968882A (en) * 2010-09-21 2011-02-09 重庆大学 Multi-source image fusion method
WO2016138838A1 (en) * 2015-03-02 2016-09-09 华为技术有限公司 Method and device for recognizing lip-reading based on projection extreme learning machine
CN106127240A (en) * 2016-06-17 2016-11-16 华侨大学 A kind of classifying identification method of plant image collection based on nonlinear reconstruction model
CN106326843A (en) * 2016-08-15 2017-01-11 武汉工程大学 Face recognition method
WO2018149133A1 (en) * 2017-02-17 2018-08-23 深圳大学 Method and system for face recognition by means of dictionary learning based on kernel non-negative matrix factorization, and sparse feature representation

Patent Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN101968882A (en) * 2010-09-21 2011-02-09 重庆大学 Multi-source image fusion method
WO2016138838A1 (en) * 2015-03-02 2016-09-09 华为技术有限公司 Method and device for recognizing lip-reading based on projection extreme learning machine
CN106127240A (en) * 2016-06-17 2016-11-16 华侨大学 A kind of classifying identification method of plant image collection based on nonlinear reconstruction model
CN106326843A (en) * 2016-08-15 2017-01-11 武汉工程大学 Face recognition method
WO2018149133A1 (en) * 2017-02-17 2018-08-23 深圳大学 Method and system for face recognition by means of dictionary learning based on kernel non-negative matrix factorization, and sparse feature representation

Non-Patent Citations (2)

* Cited by examiner, † Cited by third party
Title
李张红等: "小波核极限学习机在人脸识别中的应用", 《无线互联科技》 *
高明哲等: "基于自适应核参数优化的小波核相关向量机算法", 《计算机工程》 *

Also Published As

Publication number Publication date
CN111340111B (en) 2023-03-24

Similar Documents

Publication Publication Date Title
CN110443143B (en) Multi-branch convolutional neural network fused remote sensing image scene classification method
CN106599797B (en) A kind of infrared face recognition method based on local parallel neural network
CN111680614B (en) Abnormal behavior detection method based on video monitoring
CN110413924B (en) Webpage classification method for semi-supervised multi-view learning
CN103605972B (en) Non-restricted environment face verification method based on block depth neural network
CN111428071B (en) Zero-sample cross-modal retrieval method based on multi-modal feature synthesis
CN111126482B (en) Remote sensing image automatic classification method based on multi-classifier cascade model
CN113011357B (en) Depth fake face video positioning method based on space-time fusion
CN111191660A (en) Rectal cancer pathology image classification method based on multi-channel collaborative capsule network
CN112101426A (en) Unsupervised learning image anomaly detection method based on self-encoder
CN112580590A (en) Finger vein identification method based on multi-semantic feature fusion network
CN113887661B (en) Image set classification method and system based on representation learning reconstruction residual analysis
CN112633382A (en) Mutual-neighbor-based few-sample image classification method and system
CN108154133B (en) Face portrait-photo recognition method based on asymmetric joint learning
CN108520215B (en) Single-sample face recognition method based on multi-scale joint feature encoder
CN112052772A (en) Face shielding detection algorithm
CN109255289A (en) A kind of across aging face identification method generating model based on unified formula
CN115953666B (en) Substation site progress identification method based on improved Mask-RCNN
CN111639697B (en) Hyperspectral image classification method based on non-repeated sampling and prototype network
CN111371611B (en) Weighted network community discovery method and device based on deep learning
Yu et al. Exemplar-based recursive instance segmentation with application to plant image analysis
CN114037001A (en) Mechanical pump small sample fault diagnosis method based on WGAN-GP-C and metric learning
CN114897764A (en) Pulmonary nodule false positive elimination method and device based on standardized channel attention
CN111242028A (en) Remote sensing image ground object segmentation method based on U-Net
CN110751005B (en) Pedestrian detection method integrating depth perception features and kernel extreme learning machine

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