CN110443133A - Based on the face human ear characteristic fusion recognition algorithm for improving rarefaction representation - Google Patents

Based on the face human ear characteristic fusion recognition algorithm for improving rarefaction representation Download PDF

Info

Publication number
CN110443133A
CN110443133A CN201910592320.3A CN201910592320A CN110443133A CN 110443133 A CN110443133 A CN 110443133A CN 201910592320 A CN201910592320 A CN 201910592320A CN 110443133 A CN110443133 A CN 110443133A
Authority
CN
China
Prior art keywords
algorithm
human ear
rarefaction representation
fusion
face human
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
CN201910592320.3A
Other languages
Chinese (zh)
Inventor
郑秋梅
王璐璐
马茂东
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.)
China University of Petroleum East China
Original Assignee
China University of Petroleum East China
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 China University of Petroleum East China filed Critical China University of Petroleum East China
Priority to CN201910592320.3A priority Critical patent/CN110443133A/en
Publication of CN110443133A publication Critical patent/CN110443133A/en
Pending legal-status Critical Current

Links

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
    • G06F18/213—Feature extraction, e.g. by transforming the feature space; Summarisation; Mappings, e.g. subspace methods
    • G06F18/2135—Feature extraction, e.g. by transforming the feature space; Summarisation; Mappings, e.g. subspace methods based on approximation criteria, e.g. principal component analysis
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/25—Fusion techniques
    • 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
    • G06V40/171—Local features and components; Facial parts ; Occluding parts, e.g. glasses; Geometrical relationships
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V40/00—Recognition of biometric, human-related or animal-related patterns in image or video data
    • G06V40/10—Human or animal bodies, e.g. vehicle occupants or pedestrians; Body parts, e.g. hands
    • G06V40/16—Human faces, e.g. facial parts, sketches or expressions
    • G06V40/172—Classification, e.g. identification

Landscapes

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

Abstract

The present invention proposes to be broken limitation of the single creature feature in identification process based on the face human ear characteristic fusion recognition algorithm (P-SRC) for improving rarefaction representation, belonged to computer research technical field.It is the following steps are included: (1) extracts face human ear fusion feature using the PCA extraction algorithm that computation complexity can be effectively reduced；(2) Fusion Features are carried out to face human ear characteristic using weighting fused in tandem method；(3) using iteration speed, than faster orthogonal matching pursuit algorithm, to test sample, the rarefaction representation coefficient in training sample is solved；(4) Classification and Identification is carried out using minimum residual method.Experiments verify that the present invention can be effectively reduced computation complexity and improve recognition accuracy.

Description

Based on the face human ear characteristic fusion recognition algorithm for improving rarefaction representation
Technical field
The present invention proposes to break single based on the face human ear characteristic fusion recognition algorithm (P-SRC) for improving rarefaction representation Limitation of the biological characteristic in identification process, belongs to computer research technical field.
Background technique
Biological identification technology becomes one of the important technology of protection information security.Currently, single mode living things feature recognition is anti- The fields such as theft system, mobile payment, financial service have obtained extensive use, but single mode biological identification technology noise jamming, Discrimination and safety etc. have obvious deficiency.In order to solve single mode bio-identification in security performance and recognition performance Drawback, multimodality fusion identification were receiving the concern of extensive scholar in recent years.Multimode biological characteristic fusion identifying technology has The advantages that reliability of uniting is higher, range of applicability is wider, safety is stronger, the biological characteristic of selection should have generality, Yi Cai Collection, anti-fraud etc., and the recognition performance for the biological characteristic chosen will be got well, and computation complexity is also low as far as possible, obtains good Good recognition result, hereafter more and more researchers start to explore multimode bio-identification.Select the category of fusion It is a kind of key, every kind of biological characteristic has a respective advantage and disadvantage in identification, how using giving up theirs the advantages of them Disadvantage, and the topic discussed instantly.
Face human ear fusion recognition algorithm relatively conventional in recent years, predominantly Principal Component Analysis Algorithm (PCA), typical phase Parser (CCA), kernel canonical correlation analysis algorithm (KCCA) etc. are closed, these algorithms are in image irradiation variation, expression shape change, bat It is not strong to take the photograph angle change etc. robustness, and the sorting algorithm (SRC) based on rarefaction representation can efficiently use the spy of subspace Property, under illumination, expression shape change complex environment, there is better recognition effect, but the calculating that SRC how is effectively reduced is complicated Degree becomes research direction.
Summary of the invention
In view of the above-mentioned problems, of the invention is that improved rarefaction representation algorithm is introduced into face human ear fusion recognition, Purpose is to improve the robustness and discrimination of recognizer.
The present invention adopts the following technical scheme that: a kind of innovatory algorithm packet of the face human ear fusion recognition based on rarefaction representation Include following steps:
(1) face human ear fusion feature is extracted using the PCA extraction algorithm that computation complexity can be effectively reduced；
(2) Fusion Features are carried out to face human ear characteristic using weighting fused in tandem method；
(3) using iteration speed than faster orthogonal matching pursuit algorithm to test sample the rarefaction representation in training sample Coefficient is solved；
(4) Classification and Identification is carried out using minimum residual method.
The step (1) is that classification and identification algorithm (SRC) computation complexity based on rarefaction representation is effectively reduced, using master Constituent analysis algorithm (PVC) improves existing SRC, forms the low P-SRC algorithm of calculation amount.
Step (2) feature and fusion can time redundancy information be effectively compressed, moreover it is possible to utilize difference to the full extent The ga s safety degree of mode biological characteristic.And fused in tandem method is more simple and efficient, it is easier to be extended to the multimode more than both modalities which Biology fusion.
The orthogonal matching pursuit algorithm that the step (3) uses is compared to base tracking and match tracing scheduling algorithm, convergence speed Degree faster, it is more sparse to the decomposition of object vector.
The step (4) carries out the classification of rarefaction representation using the minimum residual method that a kind of comparison is popularized
The invention adopts the above technical scheme, which has the following advantages:
1, it is identified compared to single mode feature, multimode biological characteristic integration technology is wider with more system reliability, the scope of application And the advantage that safety is stronger.
2, in common multimode living things feature recognition algorithm, the sorting algorithm (SRC) based on rarefaction representation can be effectively sharp There is better recognition effect under illumination, expression shape change complex environment with the characteristic of subspace.But due to its calculation amount It is larger, therefore present invention introduces after Principal Component Analysis Algorithm (PCA) formation improvement rarefaction representation classification and identification algorithm (P-SRC), Tests prove that the present invention can be effectively reduced computation complexity.
Detailed description of the invention
Fig. 1 is based on the face human ear fusion recognition flow chart for improving rarefaction representation；
The corresponding discrimination of Fig. 2 weight coefficient α；
The discrimination of Fig. 3 different mode classification compares；
The various multimodality fusion algorithm discriminations of Fig. 4 compare.
Specific embodiment
Step 1: PCA algorithm basic thought is the feature vector corresponding to characteristic value larger in training sample come structure A projection observing matrix P is made, for an arbitrary sample vector x, feature vector z can be by projection observing matrix P to vector x It is projected to obtain, it may be assumed that z=PTx。
If the test object of face and human ear shares c classification, there are m face test sample and m in each class respectively Human ear test sample.Face training sample and human ear training sample are used respectivelyWithIt indicates, wherein Ai=[ai,1,ai,2,Λ,ai,m] (i=1,2, Λ, c) represent i-th of class M test sample.Then, the feature vector of face training sample and the feature vector of human ear training sample can be by Df=(Pf)TAf, De=(Pe)TAeIt is calculated, wherein PfObserving matrix is projected by the face that PCA algorithm obtains for face training sample, PeObserving matrix, D are projected by the human ear that PCA algorithm obtains for human ear training samplefFor the eigenmatrix of face training sample, DeFor the eigenmatrix of human ear training sample.The corresponding feature vector of face human ear test sample can pass through zf=(Pf)Tyf, ze= (Pe)TyeIt is calculated, wherein yf,yeRespectively indicate the test sample vector of face, human ear, zf,zeRespectively indicate face, human ear The feature vector of test sample.
Step 2: in view of face human ear characteristic information may have different recognition capabilities to identification, so this hair It is bright to joined weight coefficient when carrying out the fused in tandem of feature vector, face and human ear characteristic information pair are made full use of with this The capability of influence of identification.
Fusion Features detailed process:
(1) normalization of feature vector indicates
In order to make the feature vector of face human ear that there is same expressive force in identification, so the present invention is in people Before face and human ear characteristic Vector Fusion, normalized has been carried out to both feature vectors.From step 1: Df, DePoint Not Wei the eigenmatrix of face training sample and the eigenmatrix of human ear training sample, ifFor DfIn jth in i-th of classification The feature vector of a sample is right herein belowIt is normalized,Wherein, μfFor DfMiddle institute There are the mean vector of column vector, σfFor DfIn all column vectors variance vectors.After normalization, all face samples are special The mean value for levying vector is 0, variance 1.Similarly, then with identical method to the feature vector D of human ear training sampleeCarry out normalizing Change processing,WhereinFor DeIn column vector, μeFor DeAll column vectors mean vector, σe For DeAll column vectors variance vectors.
(2) feature vector weights fused in tandem
If D is the eigenmatrix after the feature vector fusion of face and human ear test sample, di,jFor i-th of classification in D In j-th of samples fusion after feature vector, shown in the following formula of Weighted Fusion method of face human ear characteristic vector:
Wherein, α and β will meet constraint condition alpha+beta=1.
In specific experiment, in conjunction with Fig. 2, in order to make full use of the Classification and Identification ability of different modalities, weight coefficient can lead to Training obtains, method particularly includes: as unit of 0.1, weight coefficient α is gradually adjusted to 0.9 since 0.1, corresponding weight Factor beta is also gradually adjusted under the limitation for meeting constraint alpha+beta=1, is respectively completed entire identification process, wherein corresponding highest The weight coefficient of discrimination is optimal weight coefficient.
Step 3: the present invention solves rarefaction representation coefficient using orthogonal matching algorithm.Face and human ear characteristic Fused training sample matrix is D, and the corresponding face of test sample and the fused vector of human ear characteristic are z, by orthogonal It is n (n=c that rarefaction representation coefficient x, x of the test vector z in the dictionary matrix D that training sample is constituted, which can be acquired, with tracing algorithm × m) vector in dimension space, rarefaction representation coefficient x will meet following conditionWherein,For iteration threshold model It encloses.
Step 4: sending carry out Classification and Identification using relatively more universal least residual.By training sample in a certain classification On linear combination reconstruct test sample, linear combination coefficient is that test sample corresponds in classification in training sample matrix Rarefaction representation coefficient, reconstructed sample and the smallest classification of test sample residual error are the classification where test sample.
If xi, (i=1,2, Λ, m) ∈ RnFor coefficient corresponding to i-th of classification in rarefaction representation coefficient x, remaining classification Corresponding coefficient is 0, so passing through rarefaction representation coefficient xiThe available test sample reconstructed by the i-th classification training sample,WhereinFor the test sample vector of the i-th classification training sample reconstruct.The reconstructed residual of i-th classification training sample isWherein the smallest classification of reconstructed residual is the classification where test sample, therefore can be determined that test sample z The classification at place are as follows:

Claims (5)

1. it includes following steps based on the face human ear characteristic fusion recognition algorithm for improving rarefaction representation
1) face human ear fusion feature is extracted using the PCA extraction algorithm that computation complexity can be effectively reduced；
2) Fusion Features are carried out to face human ear characteristic using weighting fused in tandem method；
3) using iteration speed than faster orthogonal matching pursuit algorithm to test sample the rarefaction representation coefficient in training sample It is solved；
4) Classification and Identification is carried out using minimum residual method.
2. if claim 1 is based on the face human ear fusion identification method for improving rarefaction representation, it is characterised in that: the step 1) classification and identification algorithm (SRC) computation complexity based on rarefaction representation, is effectively reduced, using Principal Component Analysis Algorithm (PVC) Existing SRC is improved, the low P-SRC algorithm of calculation amount is formed.
3. if claim 1 is based on the face human ear fusion identification method for improving rarefaction representation, it is characterised in that: the step 2), feature and fusion can time redundancy information be effectively compressed, moreover it is possible to the full extent using different modalities biological characteristics can Distinction.And fused in tandem method is more simple and efficient, it is easier to be extended to the multimode biology fusion more than both modalities which.
4. such as claim 1) and 2) as described in feature extraction and biological characteristic blending algorithm, it is characterised in that: the step 3), The orthogonal matching pursuit algorithm of use compared to base tracking and match tracing scheduling algorithm, iteration speed is fast, convergence rate faster, it is right The decomposition of object vector is more sparse, can be used for that rarefaction representation coefficient solves in training sample to test sample.
5. such as claim 1) and 2) and 3) as described in, it is characterised in that: the step 4), the minimum popularized using a kind of comparison Residual error method carries out the classification of rarefaction representation.
CN201910592320.3A 2019-07-03 2019-07-03 Based on the face human ear characteristic fusion recognition algorithm for improving rarefaction representation Pending CN110443133A (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN201910592320.3A CN110443133A (en) 2019-07-03 2019-07-03 Based on the face human ear characteristic fusion recognition algorithm for improving rarefaction representation

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN201910592320.3A CN110443133A (en) 2019-07-03 2019-07-03 Based on the face human ear characteristic fusion recognition algorithm for improving rarefaction representation

Publications (1)

Publication Number Publication Date
CN110443133A true CN110443133A (en) 2019-11-12

Family

ID=68429498

Family Applications (1)

Application Number Title Priority Date Filing Date
CN201910592320.3A Pending CN110443133A (en) 2019-07-03 2019-07-03 Based on the face human ear characteristic fusion recognition algorithm for improving rarefaction representation

Country Status (1)

Country Link
CN (1) CN110443133A (en)

Citations (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
WO2012078702A1 (en) * 2010-12-10 2012-06-14 Eastman Kodak Company Video key frame extraction using sparse representation
CN103268485A (en) * 2013-06-09 2013-08-28 上海交通大学 Sparse-regularization-based face recognition method capable of realizing multiband face image information fusion
CN103345621A (en) * 2013-07-09 2013-10-09 东南大学 Face classification method based on sparse concentration index

Patent Citations (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
WO2012078702A1 (en) * 2010-12-10 2012-06-14 Eastman Kodak Company Video key frame extraction using sparse representation
CN103268485A (en) * 2013-06-09 2013-08-28 上海交通大学 Sparse-regularization-based face recognition method capable of realizing multiband face image information fusion
CN103345621A (en) * 2013-07-09 2013-10-09 东南大学 Face classification method based on sparse concentration index

Non-Patent Citations (1)

* Cited by examiner, † Cited by third party
Title
张明 等: ""基于核稀疏表示的人脸人耳融合识别算法的研究"", 《现代电子技术》 *

Similar Documents

Publication Publication Date Title
Soltane et al. Face and speech based multi-modal biometric authentication
CN109255289B (en) Cross-aging face recognition method based on unified generation model
Kapoor et al. Facial gesture recognition using correlation and Mahalanobis distance
CN107239741B (en) Single-sample face recognition method based on sparse reconstruction
CN102737633A (en) Method and device for recognizing speaker based on tensor subspace analysis
CN107220627B (en) Multi-pose face recognition method based on collaborative fuzzy mean discrimination analysis
CN105117708A (en) Facial expression recognition method and apparatus
CN105810199A (en) Identity verification method and device for speakers
Kryszczuk et al. Improving classification with class-independent quality measures: Q-stack in face verification
CN110969073A (en) Facial expression recognition method based on feature fusion and BP neural network
CN111259780B (en) Single-sample face recognition method based on block linear reconstruction discriminant analysis
CN110956113B (en) Robust face recognition method based on secondary cooperation representation identification projection
Sun et al. [Retracted] Research on Face Recognition Algorithm Based on Image Processing
Alsaedi et al. Dynamic Audio-Visual Biometric Fusion for Person Recognition.
Zhang et al. View-invariant action recognition in surveillance videos
CN111950429A (en) Face recognition method based on weighted collaborative representation
Boncolmo et al. Gender Identification Using Keras Model Through Detection of Face
CN110287973B (en) Image feature extraction method based on low-rank robust linear discriminant analysis
Soltane et al. Soft decision level fusion approach to a combined behavioral speech-signature biometrics verification
Yuan et al. Research on intelligent algorithm of identity authentication based on facial features
CN110443133A (en) Based on the face human ear characteristic fusion recognition algorithm for improving rarefaction representation
Khan et al. Securing voice biometrics: One-shot learning approach for audio deepfake detection
Richiardi et al. Quality measures in unimodal and multimodal biometric verification
Singh Review on multibiometrics: classifications, normalization and fusion levels
Song et al. A bimodal biometric verification system based on deep learning

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
WD01 Invention patent application deemed withdrawn after publication

Application publication date: 20191112

WD01 Invention patent application deemed withdrawn after publication