CN106599801A - Face recognition method based on intra-class average maximum likelihood cooperative expressions - Google Patents

Face recognition method based on intra-class average maximum likelihood cooperative expressions Download PDF

Info

Publication number
CN106599801A
CN106599801A CN201611057179.XA CN201611057179A CN106599801A CN 106599801 A CN106599801 A CN 106599801A CN 201611057179 A CN201611057179 A CN 201611057179A CN 106599801 A CN106599801 A CN 106599801A
Authority
CN
China
Prior art keywords
sample
neighbour
class
kinds
cooperation
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
CN201611057179.XA
Other languages
Chinese (zh)
Inventor
施志刚
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.)
Individual
Original Assignee
Individual
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 Individual filed Critical Individual
Priority to CN201611057179.XA priority Critical patent/CN106599801A/en
Publication of CN106599801A publication Critical patent/CN106599801A/en
Pending legal-status Critical Current

Links

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V40/00—Recognition of biometric, human-related or animal-related patterns in image or video data
    • G06V40/10—Human or animal bodies, e.g. vehicle occupants or pedestrians; Body parts, e.g. hands
    • G06V40/16—Human faces, e.g. facial parts, sketches or expressions
    • G06V40/168—Feature extraction; Face representation
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
    • G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/22—Matching criteria, e.g. proximity measures
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V10/00—Arrangements for image or video recognition or understanding
    • G06V10/40—Extraction of image or video features
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V10/00—Arrangements for image or video recognition or understanding
    • G06V10/40—Extraction of image or video features
    • G06V10/513—Sparse representations

Landscapes

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

Abstract

The invention discloses a face recognition method based on intra-class average maximum-likelihood cooperative expressions. Samples similar to a test sample are selected from each class of training samples to form a neighbor sample set; cooperative expression and reconstruction are carried out on intra-class average images of each class of neighbor samples; and classes of neighbor cooperative expressions of maximum likelihood are selected according to reconstruction errors of the different intra-class neighbor average images. Thus, cooperative expression, aimed at reducing a target class, of the neighbor samples reduces the computational complexity to certain extent, and further improves the recognition rate.

Description

Meansigma methodss maximum comparability cooperation representation face identification method in class
Technical field
The present invention relates to a kind of face identification method, particularly a kind of to be represented based on meansigma methodss maximum comparability cooperation in class Face identification method.
Background technology
In computer vision system and area of pattern recognition, feature extraction and classifying is mainly asking for recognition of face research Topic.The linear algorithm that wherein traditional feature based is extracted, such as principal component analysiss (PCA), linear discriminant analysis (LDA), also have Some innovatory algorithms etc. are intended to from higher dimensional space extract the low-dimensional global characteristics that can effectively reflect image, to classify. Although there is good effect in actual applications, the local nonlinearity that linear algorithm can not well extract dimensional images is special Levy, many popular learnings method, such as locality preserving projections (LPP), border Fisher analyses (MFA) are generated for this.At present, Also do not have transformation criterion of the authoritative dimensional images to lower dimensional space.
If to have preferably represent to target image in higher dimensional space just can more accurately be classified, then, WRIGHT Deng rarefaction representation classification (SRC) is proposed, by sparse constraint linear group of all kinds of training samples that can characterize target sample is solved Close, and target sample is integrated into into the apoplexy due to endogenous wind of most nonzero coefficients.Due to SRC to illumination present in image, attitude, angle, The factor that may be interfered to classification such as even block insensitive, therefore achieve good recognition effect, caused The concern of more and more scholars, is the robustness of boosting algorithm, there is many improvement projects again.
But in recognition of face, the sample for training is often incomplete.So SRC is based in actual applications l1-The sparse solution of norm, due to needing continuous iteration, causes the complexity for calculating higher.ZHANG et al. is pointed out by analysis The effect that the similarity of sample is represented for cooperating between class, by l2Replace l1-Norm weakens the openness of norm, proposes to be based on The cooperation of regularization least square method represents (CRC_RLS), and the recognition effect of this algorithm and SRC is suitable, but run time is significantly Reduction.On this basis, there are many improvement.Some research worker implement CRC classification, such as XIE using the low-dimensional feature of image Proposition is merged image after Shearlet multi-scale transforms, is carried with uniform local binary patterns (ULBP) with reference to piecemeal Take feature and represented with cooperating, improve recognition effect, but the complexity of algorithm is higher；WEI[24]Gray level image is resolved in proposition 8 bit planes, are weighted by the effective identification information for wherein including, and are represented with constructing virtual image cooperation, are effectively lifted Discrimination.Additionally, LIN proposes that robust cooperation represents (RCR), computation complexity is substantially reduced.LU, FAN point out having for sample Importance of the effect local message to rarefaction representation, builds weighting matrix, respectively by the similarity of training sample and target sample Weighting rarefaction representation is proposed, classification performance is effectively enhanced.But the method passes through the phase of each training sample of balance and target sample Like spending, weight coefficient is embedded in and is cooperated based on whole data set in the coefficient solution for representing, can reduce calculating to a certain extent Efficiency.Inspired by this, WANG introduces section thinking, by the maximum for extracting each training sample subimage corresponding with test sample Analog information is embedded in rarefaction representation, the uncontrolled environment such as blocks for existing, and discrimination has larger improvement.Further, XIONG etc. differs greatly to illumination, expression, or has the imagery exploitation low-rank matrix blocked to recover, with this using piecemeal maximum phase Like property method, identification robustness is obviously improved.LIN is unable to effectively utilizes pair for information such as illumination, angle, the attitudes in sample The interference of classification, constructs virtual sample and is embedded in all kinds of training samples, although have good recognition effect by different scenes. But increase per class sample number, will certainly dilated data set scale so that the efficiency that coefficient is solved declines.XU etc. passes through the stage Property reduce the other secondary classification method of target class and significantly improve discrimination, but the first stage is represented based on the cooperation of whole data set, Therefore the complexity for calculating is higher, without advantage equally in operational efficiency.
The analysis of computer vision and the CRC algorithm of image processing field is to sum up widely used in recent years, is using whole Individual data set cooperation is represented, therefore the scale of data set can directly influence the timeliness of sample coefficient calculating.With above cooperation table The improvement project shown is different, and the present invention is more concerned with the operational efficiency of algorithm, best certainly on the premise of discrimination is not affected Also it has a certain upgrade.LI selects the sample more like with test sample to cooperate with expression based on whole training set, significantly reduces Run time, but the method is vulnerable to the interference of picture noise, when illumination, attitude, angle etc. are changed greatly, recognition effect one As.On this basis, YIN proposes improvement project, i.e., in training set, by relatively more all training samples and test sample system Several similaritys, selected part sample local cooperation represents that identification robustness is higher, but the method is by the similar of sample coefficient Sexual behavior mode neighbour, because the coefficient of each sample is calculated whole training set is based on, if training sample increases, computational complexity will Improve, therefore the actual effect of algorithm is general.Enlightened by above-mentioned algorithm, can be by selecting effective and small number of sample cooperation Represent to improve recognition performance.
The content of the invention
The technical problem to be solved be to provide it is a kind of for illumination, angle, attitude etc. it is uncontrolled in the case of, have The higher face identification method represented based on the maximum similar cooperation of meansigma methodss in class of robustness and operational efficiency is recognized more by force.
To solve above-mentioned technical problem, the technical solution adopted in the present invention is：One kind is based on meansigma methodss maximum phase in class Cooperate the face identification method for representing like property, it is characterised in that comprises the steps of：
Step1：Face database contains the image of C people, and everyone has niWidth image, each image size be m × N, defines training sample setTest sample is Y ∈ Rm×n, by all kinds of instructions Practice sampleVector is turned toThe matrix of such i-th class training sample composition isC classes The matrix that training sample is constituted is Χ=[Χ1,…,XC]∈Rm×N, test sample Y vector is turned to into y ∈ Rm×1；
Step2：Calculate each training sampleThe distance between with y, and find out in all kinds of training samples with test sample y away from From K nearest neighbour's sample, the matrix of the i-th class K neighbour sample composition isWherein σ1, σ2,…,σK∈[1,ni], then new neighbour's sample matrix
Step3：In neighbour's sample matrixIt is middle to calculate each class neighbour sampleClass in meansigma methodssIn composition class Neighbour's the average image matrix
Step4：Based on cooperation table representation model, willAs encoder dictionary, code coefficient α is solved with method of least square, obtained
Step5：All kinds of interior neighbour's the average images are corresponded in usage factor vector αSparse coefficient αiReconstruct respectively is surveyed Sample this y, obtains belonging to the reconstructed sample of all kinds of interior neighbour's the average imagesNeighbour's the average image reconstructed sample collection in composition class
Step6：Calculate all kinds of neighbour's the average image reconstructed samplesWith the error e of test sample yi, i.e.,
Step7：According to reconstructed error ei, from neighbour's sample setThe minimum front S classes neighbour's sample of middle Select Error, wherein φiClass neighbour's sample matrix isNew neighbour's sample set is constituted with this
Step8：By front S classes neighbour's sample set that reconstructed error is minimumAs encoder dictionary, solved with method of least square Code coefficient based on cooperation table representation model
Step9：Usage factor vectorIn correspond to all kinds of neighbour's samples of front S classesCoefficientReconstruct test respectively Sample y, obtains belonging to the reconstructed sample of all kinds of neighbours
Step10：All kinds of neighbour's reconstructed samples of S classes before calculatingError between test sample y, i.e.,
Step11：According toMinima, judge the ownership of test sample y.
Further, Different categories of samples meansigma methodssPass throughIt is calculated, and maximum comparability depends on Neighbour's number K in class, with the neighbour of test sample y is formula according to Euclidean distance in original training set ΧMeter Obtain.
Further, all kinds of interior neighbour's the average imagesIt is not direct basis Euclidean distance with the error of test sample yIt is calculated, but based on neighbour's mean chart image setCooperation passes through reconstructed error after representing Obtain.
Further, before classification, ＆ apos, according to all kinds of interior neighbour's the average imagesWith the error e of test sample yi, cast out easily The uncorrelated classification interfered to classification, only remains with the front S classes neighbour's sample beneficial to classification, i.e., Cooperation is represented.
Further, the ownership of test sample y passes through formulaJudge, lead to first Cross and obtain φ in the second sparse reconstruct to S class neighbour's samplesiValue, then according to φiObtain calculating all kinds of in first time Neighbour's the average image reconstructed sampleWith the minimum error e of test sample yiCorresponding i values, that is, affiliated classification.
The present invention compared with prior art, with advantages below and effect：
1st, by all kinds of training samples, selecting the neighbour of test sample, it is ensured that the cooperation of lesser amt sample is represented The complexity of computing is reduced to a great extent；
2 and test sample arest neighbors class in some samples meansigma methodss tend to farthest include its feature, it is ensured that When cooperation is represented, test sample is sufficiently small with the error of sample in class, while the error of sample is sufficiently large between class, makes classification More effectively.
If the 3, retaining Ganlei neighbour sample use with the error of test sample according to sparse reconstruct of all kinds of interior neighbour's the average images Classify in last cooperation, further reduce run time, it reduces class scope, makes identification more accurate.
Specific embodiment
Below by embodiment, the present invention is described in further detail, following examples be explanation of the invention and The invention is not limited in following examples.
Face identification method of the present invention based on meansigma methodss maximum comparability cooperation expression in class is by all kinds of training samples The neighbour of test sample is selected in this, it is ensured that the efficiency of computing is improved when cooperation is represented；And it is flat according to all kinds of interior neighbours If image sparse reconstruct retains Ganlei neighbour sample and represents for last cooperating with the error of test sample, to a certain degree On reduce class object, make identification more accurate.Method is comprised the following steps that：
Step1：If face database contains the image of C people, everyone has ni(i=1,2 ..., C) width image, by This defines training sample set It is the i-th class jth width image pattern, Each image size is m × n, and test sample is Y ∈ Rm×n, by all kinds of training samplesVector is turned toSuch I classes training sample composition matrix beC classes training sample constitute matrix be Χ= [Χ1,…,XC]∈Rm×N, test sample Y vector is turned to into y ∈ Rm×1。
Step2：I-th class training sample is calculated according to Euclidean distanceWith the distance of test sample yI.e.Wherein i=1,2 ... C, j=1,2 ..., ni。
Step3：The K neighbour sample closest with test sample y is found out in the i-th class training sampleρ∈[1, ni], constitute i-th class neighbour's sample matrixWherein The matrix that then new neighbour's sample is constituted is
Step4：In neighbour's sample setIt is middle to calculate each class neighbour sampleClass in meansigma methodssI.e.
Neighbour's mean chart image set in C class of compositionWherein i=1,2 ..., C, σ1, σ2,…,σK∈[1,ni]。
Step5：By neighbour's the average image collectionAs the encoder dictionary of cooperation presentation class, solved with method of least square and compiled Code factor alpha, obtainsWherein λ is regularization parameter, stable and openness with keep sample to reconstruct.
Step6：All kinds of neighbour's the average images are corresponded in usage factor vector αSparse coefficient αiReconstruct test respectively Sample y, obtains belonging to all kinds of reconstructed samplesI.e.Wherein i=1,2 ... C,It is average for the i-th class neighbour ImageReconstruct.Then C neighbour's the average image reconstructed sampleThe matrix of composition is
Step7：I-th class neighbour's the average image reconstructed sample is calculated according to Euclidean distanceWith the error e of test sample yi, I.e.Wherein i=1,2 ... C.
Step8：According to all kinds of neighbour's the average imagesSparse reconstructed error ei, from neighbour's sample setMiddle Select ErrorMinimum front S classes neighbour's sample, whereinφiClass neighbour's sample Matrix isNew neighbour's sample set is constituted with this Wherein φ1,φ2,…,φS∈ [1, C], η1,η2,…,ηK,…,ε1,ε2,…, εK,…,β1,β2,…,βK∈[1,ni]。
Step9：WillAs encoder dictionary, based on cooperation table representation model, with method of least square code coefficient is solved
Step10：Usage factor vectorIn correspond to front S classes neighbour's sampleSparse coefficientReconstruct test respectively Sample y, obtains S class neighbour's reconstructed samplesI.e.Wherein φi=φ1,φ2,…,φS∈ [1, C],For φiClass neighbourReconstruct.
Step11：Test sample y φ each with front S classes is calculated according to Euclidean distanceiClass neighbour's reconstructed sampleError I.e.Wherein φi∈[1,C]。
Step12：According toDifferentiate the ownership of test sample y.
CRC has good effect for recognition of face, but the coefficient based on whole data set is calculated and causes operational efficiency bright It is aobvious to decline.Therefore how to represent that discrimination preferably also has one on the premise of computational complexity is reduced using the cooperation of less sample It is fixed to be lifted, it is still at present a problem especially when the noise variance such as illumination, angle, attitude is larger in image.Class is introduced for this Interior neighbour's average face, proposes meansigma methodss maximum comparability cooperation representation in a species.It is of the invention by simulation results show Face identification method is compared existing some CRC methods and is had a distinct increment on recognition performance.The method is first in training set And Different categories of samples of moderate number most like with test sample is found, by neighbour's meansigma methodss in class to can at utmost reflect The feature of test sample；Then will reconstruct after all kinds of most like the average image rarefaction representations, and if selecting reconstructed error minimum The cooperation of Ganlei neighbour sample is represented, so as to complete pattern classification.This maximum phase by constructing neighbour's sample mean in class Like feature while high discrimination is kept during cooperation is represented, due to the reduction of sample number and class object, fortune is substantially reduced The row time.
Above content described in this specification is only illustration made for the present invention.Technology belonging to of the invention The technical staff in field can be made various modifications to described specific embodiment or supplement or substituted using similar mode, only Without departing from the content of description of the invention or to surmount scope defined in the claims, the guarantor of the present invention all should be belonged to Shield scope.

Claims (5)

1. it is a kind of based in class meansigma methodss maximum comparability cooperation represent face identification method, it is characterised in that comprising following step Suddenly：
Step1：Face database contains the image of C people, and everyone has niWidth image, each image size is m × n, is defined Training sample setTest sample is Y ∈ Rm×n, by all kinds of training samples ThisVector is turned toThe matrix of such i-th class training sample composition isC classes The matrix that training sample is constituted is Χ=[Χ1,…,XC]∈Rm×N, test sample Y vector is turned to into y ∈ Rm×1；
Step2：Calculate each training sampleThe distance between with y, and find out in all kinds of training samples with test sample y distance most K near neighbour's sample, wherein the matrix of the i-th class k nearest neighbor sample composition isThen new neighbour Sample matrix
Step3：In neighbour's sample matrixIt is middle to calculate each class neighbour sampleClass in meansigma methodssThen neighbour puts down in class Equal image array
Step4：By neighbour's the average image matrixAs encoder dictionary, solved based on cooperation table representation model with method of least square Code coefficient α, obtains
Step5：All kinds of interior neighbour's the average images are corresponded in usage factor vector αSparse coefficient αiTest specimens are reconstructed respectively This y, obtains belonging to the reconstructed sample of all kinds of interior neighbour's the average imagesNeighbour's the average image reconstructed sample collection in composition class
Step6：Calculate all kinds of neighbour's the average image reconstructed samplesWith the error e of test sample yi, i.e.,
Step7：According to reconstructed error ei, from neighbour's sample setThe minimum front S classes neighbour's sample of middle Select Error, wherein φi Class neighbour's sample matrix isNew neighbour's sample set is constituted with this
Step8：By front S classes neighbour's sample set that reconstructed error is minimumAs encoder dictionary, solved with method of least square and be based on The code coefficient of cooperation table representation model
Step9：Usage factor vectorIn correspond to all kinds of neighbour's samples of front S classesCoefficientTest sample y is reconstructed respectively, Obtain belonging to the reconstructed sample of all kinds of neighbours
Step10：All kinds of neighbour's reconstructed samples of S classes before calculatingError between test sample y, i.e.,
Step11：According toJudge the ownership of test sample y.
2., according to the face identification method represented based on meansigma methodss maximum comparability cooperation in class described in claim 1, it is special Levy and be：Different categories of samples meansigma methodssPass throughIt is calculated, and maximum comparability depends on neighbour in class Number K, with the neighbour of test sample y are formula according to Euclidean distance in original training set ΧIt is calculated.
3., according to the face identification method represented based on meansigma methodss maximum comparability cooperation in class described in claim 1, it is special Levy and be：All kinds of interior neighbour's the average imagesIt is not that direct basis Euclidean distance is calculated with the error of test sample y, but Based on neighbour's mean chart image setCooperation passes through reconstructed error after representingObtain.
4. according to the face identification method represented based on meansigma methodss maximum comparability cooperation in class described in claim 1, its It is characterised by：According to all kinds of interior neighbour's the average imagesWith the error e of test sample yi, cast out easily classification is interfered Uncorrelated classification, only remains with the front S classes neighbour's sample beneficial to classification, i.e., Cooperation is represented.
5., according to the face identification method represented based on meansigma methodss maximum comparability cooperation in class described in claim 1, it is special Levy and be：The method belong to reduce target classification for the purpose of classification method, due to calculating reconstructed error twice, therefore judge to survey The ownership of sample this y passes through formulaJudge, first by near to S classes at second φ is obtained in the sparse reconstruct of adjacent sampleiValue, then according to φiObtain calculating all kinds of neighbour's the average image reconstruct in first time SampleWith the minimum error e of test sample yiCorresponding i values, that is, affiliated classification.
CN201611057179.XA 2016-11-26 2016-11-26 Face recognition method based on intra-class average maximum likelihood cooperative expressions Pending CN106599801A (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN201611057179.XA CN106599801A (en) 2016-11-26 2016-11-26 Face recognition method based on intra-class average maximum likelihood cooperative expressions

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN201611057179.XA CN106599801A (en) 2016-11-26 2016-11-26 Face recognition method based on intra-class average maximum likelihood cooperative expressions

Publications (1)

Publication Number Publication Date
CN106599801A true CN106599801A (en) 2017-04-26

Family

ID=58593332

Family Applications (1)

Application Number Title Priority Date Filing Date
CN201611057179.XA Pending CN106599801A (en) 2016-11-26 2016-11-26 Face recognition method based on intra-class average maximum likelihood cooperative expressions

Country Status (1)

Country Link
CN (1) CN106599801A (en)

Cited By (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN110659700A (en) * 2019-10-10 2020-01-07 西南石油大学 KNN-based image sample generation method
CN111209818A (en) * 2019-12-30 2020-05-29 新大陆数字技术股份有限公司 Video individual identification method, system, equipment and readable storage medium
CN115429289A (en) * 2022-09-01 2022-12-06 天津大学 Brain-computer interface training data amplification method, device, medium and electronic equipment
CN115429289B (en) * 2022-09-01 2024-05-31 天津大学 Brain-computer interface training data amplification method, device, medium and electronic equipment

Cited By (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN110659700A (en) * 2019-10-10 2020-01-07 西南石油大学 KNN-based image sample generation method
CN110659700B (en) * 2019-10-10 2022-10-11 西南石油大学 KNN-based image sample generation method
CN111209818A (en) * 2019-12-30 2020-05-29 新大陆数字技术股份有限公司 Video individual identification method, system, equipment and readable storage medium
CN115429289A (en) * 2022-09-01 2022-12-06 天津大学 Brain-computer interface training data amplification method, device, medium and electronic equipment
CN115429289B (en) * 2022-09-01 2024-05-31 天津大学 Brain-computer interface training data amplification method, device, medium and electronic equipment

Similar Documents

Publication Publication Date Title
CN108960140B (en) Pedestrian re-identification method based on multi-region feature extraction and fusion
CN109389608B (en) There is the fuzzy clustering image partition method of noise immunity using plane as cluster centre
Prince et al. Probabilistic models for inference about identity
CN102682302B (en) Human body posture identification method based on multi-characteristic fusion of key frame
Xu et al. Consistent instance false positive improves fairness in face recognition
CN108509854B (en) Pedestrian re-identification method based on projection matrix constraint and discriminative dictionary learning
CN105184298B (en) A kind of image classification method of quick local restriction low-rank coding
CN105224937B (en) Fine granularity semanteme color pedestrian recognition methods again based on human part position constraint
CN108875459B (en) Weighting sparse representation face recognition method and system based on sparse coefficient similarity
CN109063757A (en) It is diagonally indicated based on block and the multifarious multiple view Subspace clustering method of view
CN107145841B (en) Low-rank sparse face recognition method and system based on matrix
Shen et al. Image recognition method based on an improved convolutional neural network to detect impurities in wheat
CN110889865A (en) Video target tracking method based on local weighted sparse feature selection
CN102722578B (en) Unsupervised cluster characteristic selection method based on Laplace regularization
CN112215268A (en) Method and device for classifying disaster weather satellite cloud pictures
Wang et al. Towards calibrated hyper-sphere representation via distribution overlap coefficient for long-tailed learning
CN108491719A (en) A kind of Android malware detection methods improving NB Algorithm
Li et al. A hierarchical framework for image-based human age estimation by weighted and OHRanked sparse representation-based classification
CN106599801A (en) Face recognition method based on intra-class average maximum likelihood cooperative expressions
CN108388918B (en) Data feature selection method with structure retention characteristics
CN108121964B (en) Matrix-based joint sparse local preserving projection face recognition method
Li et al. Spatial and temporal information fusion for human action recognition via Center Boundary Balancing Multimodal Classifier
CN116051924A (en) Divide-and-conquer defense method for image countermeasure sample
CN110135363A (en) Based on differentiation dictionary insertion pedestrian image search method, system, equipment and medium
Liu et al. Discriminative self-adapted locality-sensitive sparse representation for video semantic analysis

Legal Events

Date Code Title Description
PB01 Publication
PB01 Publication
WD01 Invention patent application deemed withdrawn after publication
WD01 Invention patent application deemed withdrawn after publication

Application publication date: 20170426