CN107122795B - Pedestrian re-identification method based on coring characteristics and random subspace integration - Google Patents

Pedestrian re-identification method based on coring characteristics and random subspace integration Download PDF

Info

Publication number
CN107122795B
CN107122795B CN201710212251.XA CN201710212251A CN107122795B CN 107122795 B CN107122795 B CN 107122795B CN 201710212251 A CN201710212251 A CN 201710212251A CN 107122795 B CN107122795 B CN 107122795B
Authority
CN
China
Prior art keywords
pedestrian
coring
sample
samples
features
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Expired - Fee Related
Application number
CN201710212251.XA
Other languages
Chinese (zh)
Other versions
CN107122795A (en
Inventor
赵才荣
陈亦鹏
王学宽
卫志华
苗夺谦
田元
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.)
Tongji University
Original Assignee
Tongji 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 Tongji University filed Critical Tongji University
Priority to CN201710212251.XA priority Critical patent/CN107122795B/en
Publication of CN107122795A publication Critical patent/CN107122795A/en
Application granted granted Critical
Publication of CN107122795B publication Critical patent/CN107122795B/en
Expired - Fee Related 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
    • G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
    • G06F18/2413—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches based on distances to training or reference patterns
    • G06F18/24133—Distances to prototypes

Landscapes

  • Engineering & Computer Science (AREA)
  • Data Mining & Analysis (AREA)
  • Theoretical Computer Science (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Bioinformatics & Cheminformatics (AREA)
  • Bioinformatics & Computational Biology (AREA)
  • Artificial Intelligence (AREA)
  • Evolutionary Biology (AREA)
  • Evolutionary Computation (AREA)
  • Physics & Mathematics (AREA)
  • General Engineering & Computer Science (AREA)
  • General Physics & Mathematics (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Image Analysis (AREA)
  • Traffic Control Systems (AREA)

Abstract

The invention relates to a pedestrian re-identification method based on nucleation features and random subspace integration, which comprises the following steps: s1, acquiring a training sample set and a testing sample set of the pedestrian image, and determining a coring function between the two samples; s2, converting the original features of the two sample sets into coring features respectively; s3, randomly selecting a plurality of different subspaces in the coring feature space of the training sample set, and respectively calculating covariance matrixes and inverse matrixes of coring feature differences of different and same pedestrian image pairs to obtain distribution functions of the coring feature differences of the image pairs; s4, calculating the probability that the sample is the same pedestrian and the probability that the sample is different pedestrians under each subspace, and taking the ratio of the two probabilities as the distance between the samples; and S5, integrating the distances to obtain the final distance between each sample pair. Compared with the prior art, the pedestrian re-identification method has good pedestrian re-identification capability, is suitable for various different characteristics, and has strong robustness.

Description

Pedestrian re-identification method based on coring characteristics and random subspace integration
Technical Field
The invention relates to a feature extraction and distance measurement learning method in intelligent analysis of surveillance videos, in particular to a pedestrian re-identification method based on nuclear feature and random subspace integration.
Background
The pedestrian re-identification refers to the problem of matching pedestrian images under different camera view angles in a system consisting of a plurality of cameras. The method provides key help for analyzing different aspects of pedestrian identity, behavior and the like, and develops into a key component in the field of intelligent video monitoring.
The main methods in the field of pedestrian re-identification can be divided into the following two categories: 1) a pedestrian re-identification method based on the feature representation; 2) a pedestrian re-identification method based on metric learning.
In the feature representation-based pedestrian re-recognition method, low-level visual features are the most commonly used features. Common low-level visual features include color histograms, textures, and the like, which are described in detail below: the color histogram describes the color distribution characteristics of the whole image or one small area of the image through the color distribution on the statistical image. It is robust to viewing angle variations but is susceptible to luminance changes such as illumination, and therefore it is usually extracted over a particular color space. The structural information of the whole image or one small area of the image described by the texture features is well complementary to the color information described by the color histogram features. Most pedestrian re-identification algorithms are realized based on the bottom layer visual features, however, when a human body performs a pedestrian re-identification task, whether two images belong to the same pedestrian is judged more by the attribute features with semantics rather than the bottom layer visual features. Such as: hair style, type of shirt, type of coat, shoes, etc. Compared with the underlying visual features, the semantic attribute feature-based method has natural advantages: the semantic attributes are more robust to the pedestrian appearance feature difference under different monitoring videos, and the description of the semantic attributes of the same pedestrian is usually unchanged under different monitoring videos; semantic attributes are closer to human comprehension, so that the result obtained by the characteristic method based on the semantic attributes is more in line with the requirements of people; the semantic attribute based approach is more convenient for human interaction.
After the feature representation method, how to measure the distances of different pedestrian images is also one of the key issues in the field of pedestrian re-identification. When calculating the similarity of feature vectors, the feature-based method usually uses classical distance functions such as euclidean distance, cosine distance, geodesic distance, and the like. These classical distance functions do not take into account the characteristics of the sample and therefore tend to perform poorly. In recent years, a large number of documents adopt a distance measurement method, and a distance function more conforming to the characteristics of a sample is obtained through training of a labeled sample, so that the performance is improved. These methods are implemented by learning a distance function in the form of a mahalanobis. Wherein a pedestrian re-identification method based on a simple and direct strategy (KISS) is top in performance. However, this method is based on the theoretical assumption that the distribution of samples follows the gaussian distribution, but in reality, the samples do not perfectly follow the gaussian distribution, and may even deviate seriously, thereby causing performance degradation. In addition, in practical situations, the sample size tends to be much smaller than the feature dimension, making the computation of mahalanobis distance in metric learning difficult or even impossible.
Disclosure of Invention
The present invention aims to overcome the defects of the prior art and provide a pedestrian re-identification method based on the nucleation feature and the random subspace integration, so that the extracted features more approximately obey the gaussian distribution energy and the contradiction between the sample size and the feature dimension, and the SSS (small scale sample) problem is avoided, thereby improving the performance.
The purpose of the invention can be realized by the following technical scheme:
a pedestrian re-identification method based on nucleation feature and random subspace integration comprises the following steps:
s1, acquiring a training sample set and a test sample set of images of pedestrians, determining a nucleation function between two samples, wherein the output value of the nucleation function is a one-dimensional real number, and the same pedestrian has a plurality of images in each sample set; in the training sample set, the corresponding relation between each pedestrian and a plurality of images of the pedestrian is known, and in the testing sample set, the corresponding relation is unknown;
s2, respectively converting the original features of the two sample sets into coring features, wherein the dimensionalities of the coring features are the number of samples in the training sample set;
s3, randomly selecting a plurality of different subspaces from the coring feature space of the training sample set, respectively calculating covariance matrixes and inverse matrixes of the coring feature differences of different pedestrian image pairs under each subspace, and calculating covariance matrixes and inverse matrixes of the coring feature differences of the same pedestrian image pair to obtain a distribution function of the coring feature differences of the image pairs; the purpose of using the training sample is to learn a suitable distribution function through the training sample;
s4, calculating differences of nucleation characteristics of sample pairs in the test sample set under each subspace (the subspace is the subspace selected in the step S3), calculating the probability that the sample pairs are the same pedestrians and the probability that the sample pairs are different pedestrians according to the difference covariance matrix, the inverse matrix and the distribution function thereof, and taking the ratio of the two probabilities as the distance between the two samples;
and S5, integrating the distances calculated in different subspaces to obtain the final distance between each sample pair in the test sample set for pedestrian identification, wherein the smaller the final distance is, the higher the possibility that the sample pair is the same pedestrian is. Thus, the images of the same pedestrian in the test sample set can be identified.
In step S1, the coring function is a gaussian kernel function, and the distribution function obtained in step S3 is a gaussian distribution function.
In the step S1, the coring function is k (x)i,xj)，
Figure BDA0001261235230000031
Where σ is 1, xi、xjRespectively representing the ith and jth training samples.
In step S2, the specific process of converting the original features into the coring features includes:
converting training set X into coring features
Figure BDA0001261235230000032
Figure BDA0001261235230000033
Wherein m is the number of samples in the training set, and X belongs to Rd×mAnd d is the characteristic dimension of the sample,
Figure BDA0001261235230000034
converting test set Z into nucleated features
Figure BDA0001261235230000035
Figure BDA0001261235230000036
Wherein n is the number of the test set samples,
Figure BDA0001261235230000037
in the step S3, the covariance Σ of the coring feature difference values of the different pairs of pedestrian images0The calculation formula is as follows:
Figure BDA0001261235230000038
wherein, y ij0 represents all pairs of samples not belonging to the same pedestrian, N0The total number of samples meeting the condition;
covariance matrix Σ of nucleated feature difference values for identical pedestrian image pairs1The calculation formula is as follows:
Figure BDA0001261235230000039
wherein, y ij1 represents all pairs of samples not belonging to the same pedestrian, N1The total number of samples meeting this condition.
In step S4, the distance between two samples is calculated as:
Figure BDA00012612352300000310
compared with the prior art, the invention has the following advantages:
(1) original features are converted into coring features, and the difference distribution between feature pairs is more approximately subjected to Gaussian distribution, which is the basic theoretical assumption of distance measurement learning based on a simple and direct strategy;
(2) in a nonlinear space, the coring characteristics tend to have stronger identification capability;
(3) the complex characteristic vectors are projected into a plurality of subspaces with smaller dimensionalities to respectively calculate distances, and the randomly selected subspace projection mode is used for replacing the traditional distance calculation mode, so that the algorithm performance is obviously improved, the sample distance calculation process is optimized, and the cost of matrix operation is saved;
(4) the randomly selected number of the subspace dimensions is far smaller than the sample size, and the contradiction that the sample size is far smaller than the feature dimension number in practical application is reconciled, so that the distance calculation is more accurate and efficient.
Drawings
FIG. 1 is a flow chart of the method of the present invention;
FIGS. 2(a) -2(d) are probability distribution diagrams of feature differences between pairs of samples before and after using the method of the present embodiment, where: 2(a) is the difference distribution of the same pedestrian sample pair under the original characteristics; 2(b) is the difference distribution of different pedestrian sample pairs under the original characteristics; 2(c) is the difference distribution of the same pedestrian sample pair under the nucleation characteristic; and 2(d) is the difference distribution of different pedestrian sample pairs under the coring characteristic.
Fig. 3(a), 3(b) are CMC curves on a VIPeR (P316) pedestrian re-identification public data set using different parameters and different features in the method of the present embodiment, where: 3(a) CMC curves using different parameters under the LOMO characteristics; and 3(b) is a CMC curve using different parameters under the characteristics of kCCA.
Detailed Description
The invention is described in detail below with reference to the figures and specific embodiments. The present embodiment is implemented on the premise of the technical solution of the present invention, and a detailed implementation manner and a specific operation process are given, but the scope of the present invention is not limited to the following embodiments.
Examples
A pedestrian re-identification method based on nucleation feature and random subspace integration comprises the following steps:
the method comprises the following steps: converting the original features into a coring feature representation, which is described in detail as follows:
training set X belonging to R for obtaining pedestrian imaged×mAnd test set Z ∈ Rd×nWherein the characteristic dimension of the sample is d, the number of samples in the training set is m, the number of samples in the test set is n, and x is usediDenotes the ith training sample by ziThe ith test sample is indicated. Using coring functions
Figure BDA0001261235230000041
Converting training set X into coring features
Figure BDA0001261235230000042
Converting test set Z into nucleated features
Figure BDA0001261235230000051
Where σ is 1. The specific conversion process can be expressed as
Figure BDA0001261235230000052
Step two: in the coring feature space, L different subspaces are randomly selected. The specific description is as follows: after the step one is completed, the dimension of the coring feature space is the same as the number of samples of the training set, namely m-dimension. Randomly choosing L different subspaces D in the m-dimensionk(k ═ 1, 2.. times, L), the dimensions of the subspace being
Figure BDA0001261235230000053
Step three: calculating covariance matrixes sigma of feature differences between different pairs of pedestrian images under different subspaces respectively0And find the inverse matrix
Figure BDA0001261235230000054
Calculating a covariance matrix Σ of feature differences between pairs of identical pedestrian images1And find the inverse matrix
Figure BDA0001261235230000055
The specific description is as follows: calculating covariance matrixes sigma of feature differences between different pairs of pedestrian images under different subspaces respectively0The concrete calculation mode is
Figure BDA0001261235230000056
Wherein y isij0 representsAll pairs of samples, N, not belonging to the same pedestrian0The total number of samples meeting this condition. Calculating covariance matrix covariance of feature differences between pairs of identical pedestrian images1，Σ1Is calculated in a specific manner
Figure BDA0001261235230000057
Wherein y isij1 represents all pairs of samples belonging to the same pedestrian, N1The total number of samples meeting this condition.
Step four: based on Gaussian distribution function and two inverse matrixes respectively under different subspaces
Figure BDA0001261235230000058
Calculating the probability that the difference value of the two characteristics belongs to the same pedestrian and the probability that the difference value of the two characteristics belongs to different pedestrians, and regarding the ratio of the two probabilities as the distance between the samples, which is specifically described as follows: by using
Figure BDA0001261235230000059
Representing test samples under subspaces
Figure BDA00012612352300000510
And
Figure BDA00012612352300000511
by difference of (a) from (b), using (H)0The hypothesis that the ith test sample and the jth test sample belong to different pedestrians is represented by H1Representing the hypothesis that the ith and jth test samples belong to the same pedestrian. When the difference between the sample pairs obeys Gaussian distribution, the difference under two assumed conditions can be obtained
Figure BDA00012612352300000512
Probability of (2)
Figure BDA00012612352300000513
By using
Figure BDA00012612352300000514
To represent
Figure BDA00012612352300000515
The logarithm of the ratio of the probabilities obeying both assumptions can be obtained
Figure BDA0001261235230000061
Equation (5) can be converted into
Figure BDA0001261235230000062
Namely, it is
Figure BDA0001261235230000063
Removing the constant, the distance between the ith and jth test samples in the kth subspace can be obtained and expressed as
Figure BDA0001261235230000064
Step five: integrating the distances calculated in the L different subspaces to obtain a final distance, which is described in detail as follows: for the L distances obtained in step four, the distances are integrated by using a weighted average method, and the final distance formula can be expressed as:
Figure BDA0001261235230000065
as shown in fig. 1, it is a flowchart of this embodiment, and the specific implementation manner is as follows:
1) determining a kernel function;
2) converting the original features of the training sample into coring features;
3) converting the original features of the test sample into coring features;
4) to judge ziAnd zjWhether they belong to the same pedestrian or not, using H0Indicating that they are dissimilar, i.e. not belonging to the same pedestrian, by H1Indicating that they are similar, i.e., belong to the same pedestrian;
5) randomly selecting L subspaces D in the nucleated feature spacek(k＝1,2,...,L)；
6) Calculating covariance matrices Σ of feature differences between pairs of different pedestrian images in different subspaces, respectively0And find the inverse matrix
Figure BDA0001261235230000066
7) Calculating covariance matrices Σ of feature differences between pairs of identical pedestrian images in different subspaces, respectively1And find the inverse matrix
Figure BDA0001261235230000067
8) Calculating feature difference values in different subspaces respectively
Figure BDA0001261235230000068
Obeying the probability distribution function of two hypotheses and applying the logarithm of the probability ratio
Figure BDA0001261235230000069
As the distance between samples;
9) respectively converting the probability ratio distance into the Mahalanobis distance d in different subspacesk(zi,zj)；
10) And integrating the calculated distances in the L subspaces to obtain the final sample distance.
Fig. 2(a) -2(d) are probability distribution diagrams of feature differences between sample pairs before and after using the method of the present embodiment, where the histogram is actual probability distribution, the linear graph is a gaussian distribution curve drawn according to data variance, and four original features to be compared are LOMO, kCCA, SCNCD, and ELF18, which are features widely applied in the pedestrian re-identification method, where: 2(a) is the difference distribution of the same pedestrian sample pair under the original characteristics; 2(b) is the difference distribution of different pedestrian sample pairs under the original characteristics; 2(c) is the difference distribution of the same pedestrian sample pair under the nucleation characteristic; and 2(d) is the difference distribution of different pedestrian sample pairs under the coring characteristic.
Fig. 3(a) and 3(b) show the rank-1 matching rate of the present embodiment in the VIPeR (P316) pedestrian re-identification public data set under different parameters and different features, and compare with the conventional regularization method. Wherein: 3(a) is the rank-1 matching rate using different parameters under the LOMO characteristic; and 3(b) is the rank-1 matching rate under the characteristics of kCCA and using different parameters.
TABLE 1
Figure BDA0001261235230000071
Table 1 shows the performance of the method of this embodiment compared to other metric-based learning algorithms on a VIPeR (P316) pedestrian re-identification public data set.
TABLE 2
Figure BDA0001261235230000072
Table 2 shows the performance of this example method compared to other metric-based learning algorithms on the PRID 450S (P225) pedestrian re-identification public data set.
TABLE 3
KRKISS NFST MLAPG XQDA MFA kLFDA KISS LFDA
Time of day 5.04 2.48 40.9 3.86 2.58 2.74 7.41 229.3
Table 3 shows the comparison between the training time overhead of the method of the present embodiment and other metric learning based algorithms.

Claims (6)

1. A pedestrian re-identification method based on nucleation feature and random subspace integration is characterized by comprising the following steps:
s1, acquiring a training sample set and a test sample set of images of pedestrians, and determining a coring function between two samples, wherein in each sample set, the same pedestrian has a plurality of images;
s2, respectively converting the original features of the two sample sets into coring features, wherein the dimensionalities of the coring features are the number of samples in the training sample set;
s3, randomly selecting a plurality of different subspaces from the coring feature space of the training sample set, respectively calculating covariance matrixes and inverse matrixes of the coring feature differences of different pedestrian image pairs under each subspace, and calculating covariance matrixes and inverse matrixes of the coring feature differences of the same pedestrian image pair to obtain a distribution function of the coring feature differences of the image pairs;
s4, calculating differences of the nucleation characteristics of the sample pairs in the test sample set under each subspace, calculating the probability that the sample pairs are the same pedestrian and the probability that the sample pairs are different pedestrians according to the difference covariance matrix and the inverse matrix and the distribution function thereof, and taking the ratio of the two probabilities as the distance between the two samples;
and S5, integrating the distances calculated in different subspaces to obtain the final distance between each sample pair in the test sample set for pedestrian identification, wherein the smaller the final distance is, the higher the possibility that the sample pair is the same pedestrian is.
2. The method for re-identifying pedestrians according to claim 1, wherein in the step S1, the coring function is a gaussian kernel function, and the distribution function obtained in the step S3 is a gaussian distribution function.
3. The method for re-identifying pedestrians according to claim 1 or 2, wherein the kernel function is k (x) in step S1i,xj)，
Figure FDA0002321783670000011
Where σ is 1, xi、xjRespectively representing the ith and jth training samples.
4. The method for re-identifying pedestrians based on the integration of the nucleated features and the random subspace as claimed in claim 3, wherein in the step S2, the specific process of converting the original features into the nucleated features includes:
converting training set X into coring features
Figure FDA0002321783670000012
Figure FDA0002321783670000013
Wherein m is the number of samples in the training set, and X belongs to Rd×mAnd d is the characteristic dimension of the sample,
Figure FDA0002321783670000021
converting test set Z into nucleated features
Figure FDA0002321783670000022
Figure FDA0002321783670000023
Wherein n is the number of the test set samples,
Figure FDA0002321783670000024
5. the method as claimed in claim 4, wherein the step S3 is performed by using a covariance matrix Σ of differences between features of different pedestrian image pairs0The calculation formula is as follows:
Figure FDA0002321783670000025
wherein, yij0 represents all pairs of samples not belonging to the same pedestrian, N0The total number of sample pairs which do not belong to the same pedestrian;
covariance matrix Σ of nucleated feature difference values for identical pedestrian image pairs1The calculation formula is as follows:
Figure FDA0002321783670000026
wherein, yij1 represents all pairs of samples belonging to the same pedestrian, N1Is the total number of pairs of samples belonging to the same pedestrian.
6. The method for re-identifying pedestrians based on the integration of the nucleated features and the random subspace as claimed in claim 5, wherein in the step S4, the distance between two samples is calculated as:
Figure FDA0002321783670000027
wherein the content of the first and second substances,
Figure FDA0002321783670000028
covariance matrix Σ of nucleated feature difference values for the same pedestrian image pair1The inverse of the matrix of (a) is,
Figure FDA0002321783670000029
covariance matrix Σ of nucleated feature differences for different pedestrian image pairs0The inverse matrix of (c).
CN201710212251.XA 2017-04-01 2017-04-01 Pedestrian re-identification method based on coring characteristics and random subspace integration Expired - Fee Related CN107122795B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN201710212251.XA CN107122795B (en) 2017-04-01 2017-04-01 Pedestrian re-identification method based on coring characteristics and random subspace integration

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN201710212251.XA CN107122795B (en) 2017-04-01 2017-04-01 Pedestrian re-identification method based on coring characteristics and random subspace integration

Publications (2)

Publication Number Publication Date
CN107122795A CN107122795A (en) 2017-09-01
CN107122795B true CN107122795B (en) 2020-06-02

Family

ID=59724629

Family Applications (1)

Application Number Title Priority Date Filing Date
CN201710212251.XA Expired - Fee Related CN107122795B (en) 2017-04-01 2017-04-01 Pedestrian re-identification method based on coring characteristics and random subspace integration

Country Status (1)

Country Link
CN (1) CN107122795B (en)

Families Citing this family (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN107909049B (en) * 2017-11-29 2020-07-31 广州大学 Pedestrian re-identification method based on least square discriminant analysis distance learning

Citations (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN103295242A (en) * 2013-06-18 2013-09-11 南京信息工程大学 Multi-feature united sparse represented target tracking method
CN103885049A (en) * 2014-03-06 2014-06-25 西安电子科技大学 Meter-wave radar low elevation estimating method based on minimum redundancy linear sparse submatrix
CN104008380A (en) * 2014-06-16 2014-08-27 武汉大学 Pedestrian detection method and system based on salient regions
CN104408705A (en) * 2014-09-23 2015-03-11 西安电子科技大学 Anomaly detection method of hyperspectral image
CN104616319A (en) * 2015-01-28 2015-05-13 南京信息工程大学 Multi-feature selection target tracking method based on support vector machine
WO2016026370A1 (en) * 2014-08-22 2016-02-25 Zhejiang Shenghui Lighting Co., Ltd. High-speed automatic multi-object tracking method and system with kernelized correlation filters

Patent Citations (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN103295242A (en) * 2013-06-18 2013-09-11 南京信息工程大学 Multi-feature united sparse represented target tracking method
CN103885049A (en) * 2014-03-06 2014-06-25 西安电子科技大学 Meter-wave radar low elevation estimating method based on minimum redundancy linear sparse submatrix
CN104008380A (en) * 2014-06-16 2014-08-27 武汉大学 Pedestrian detection method and system based on salient regions
WO2016026370A1 (en) * 2014-08-22 2016-02-25 Zhejiang Shenghui Lighting Co., Ltd. High-speed automatic multi-object tracking method and system with kernelized correlation filters
CN104408705A (en) * 2014-09-23 2015-03-11 西安电子科技大学 Anomaly detection method of hyperspectral image
CN104616319A (en) * 2015-01-28 2015-05-13 南京信息工程大学 Multi-feature selection target tracking method based on support vector machine

Also Published As

Publication number Publication date
CN107122795A (en) 2017-09-01

Similar Documents

Publication Publication Date Title
CN110443143B (en) Multi-branch convolutional neural network fused remote sensing image scene classification method
CN107066559B (en) Three-dimensional model retrieval method based on deep learning
CN109410184B (en) Live broadcast pornographic image detection method based on dense confrontation network semi-supervised learning
CN111709909A (en) General printing defect detection method based on deep learning and model thereof
CN111639564B (en) Video pedestrian re-identification method based on multi-attention heterogeneous network
CN108960404B (en) Image-based crowd counting method and device
CN108154133B (en) Face portrait-photo recognition method based on asymmetric joint learning
CN109447123B (en) Pedestrian re-identification method based on label consistency constraint and stretching regularization dictionary learning
CN106778814B (en) Method for removing SAR image spots based on projection spectral clustering algorithm
CN112836671B (en) Data dimension reduction method based on maximized ratio and linear discriminant analysis
CN111079847A (en) Remote sensing image automatic labeling method based on deep learning
CN107862680B (en) Target tracking optimization method based on correlation filter
JPH09134432A (en) Pattern recognition method
CN104408728A (en) Method for detecting forged images based on noise estimation
CN110852292B (en) Sketch face recognition method based on cross-modal multi-task depth measurement learning
CN111079543B (en) Efficient vehicle color identification method based on deep learning
CN109741380B (en) Textile picture fast matching method and device
CN112070116B (en) Automatic artistic drawing classification system and method based on support vector machine
CN107122795B (en) Pedestrian re-identification method based on coring characteristics and random subspace integration
CN113313179A (en) Noise image classification method based on l2p norm robust least square method
CN108596906A (en) It is a kind of to refer to screen image quality evaluating method entirely based on sparse locality preserving projections
CN111275687A (en) Fine-grained image stitching detection method based on connected region marks
He et al. Fast image quality assessment via supervised iterative quantization method
CN113034454B (en) Underwater image quality evaluation method based on human visual sense
Hong et al. Saliency-based feature learning for no-reference image quality assessment

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
CF01 Termination of patent right due to non-payment of annual fee

Granted publication date: 20200602

CF01 Termination of patent right due to non-payment of annual fee