CN108256478A - A kind of video method for measuring similarity based on the study of QPSO Riemann manifolds - Google Patents

A kind of video method for measuring similarity based on the study of QPSO Riemann manifolds Download PDF

Info

Publication number
CN108256478A
CN108256478A CN201810045439.4A CN201810045439A CN108256478A CN 108256478 A CN108256478 A CN 108256478A CN 201810045439 A CN201810045439 A CN 201810045439A CN 108256478 A CN108256478 A CN 108256478A
Authority
CN
China
Prior art keywords
video
image
riemann
qpso
similarity
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
CN201810045439.4A
Other languages
Chinese (zh)
Inventor
王玉
吕颖达
黄永平
申铉京
马舒阳
沈哲
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.)
Jilin University
Original Assignee
Jilin 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 Jilin University filed Critical Jilin University
Priority to CN201810045439.4A priority Critical patent/CN108256478A/en
Publication of CN108256478A publication Critical patent/CN108256478A/en
Pending legal-status Critical Current

Links

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V20/00—Scenes; Scene-specific elements
    • G06V20/40—Scenes; Scene-specific elements in video content
    • 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
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/25—Fusion techniques
    • G06F18/253—Fusion techniques of extracted 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
    • 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
    • G06V10/46—Descriptors for shape, contour or point-related descriptors, e.g. scale invariant feature transform [SIFT] or bags of words [BoW]; Salient regional features
    • G06V10/467—Encoded features or binary features, e.g. local binary patterns [LBP]

Landscapes

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

Abstract

The present invention proposes a kind of video method for measuring similarity based on the study of QPSO Riemann manifolds.Video similarity is considered as image set measuring similarity problem by the algorithm, texture feature extraction and fusion treatment is carried out after image alignment, recycle the Riemann manifold with QPSO optimizations significantly brief dimension to obtain the intrinsic representation of video data, the measurement of similarity then using affine hull distance calibration method, finally obtains optimal classification recognition result using libSVM.By the contrast experiment carried out on associated video data library, the validity of put forward algorithm is demonstrated, carries that algorithm accuracy of identification is higher, and error is relatively low, and with stronger robustness.

Description

A kind of video method for measuring similarity based on the study of QPSO Riemann manifolds
Technical field
The present invention relates to computer vision field, more particularly to a kind of video method for measuring similarity.
Background technology
Video measuring similarity becomes the research hotspot and difficulties of computer vision field in recent years, and solution is regarded The problems such as frequency recognition of face, video object detection, video human Activity recognition, is of great significance and researching value.Relative to quiet State image, available characteristic information is more rich and varied in dynamic video, for example, the time multidate information of video helps to know The not promotion of rate；The relatively high image of resolution ratio can be chosen from video sequence so that recognition performance can be improved；It can be with Reconstruct target three-dimensional is learnt by video, video measuring similarity can be efficiently realized using these models.In short, when Between and movable information play the role of in based on video measuring similarity it is vital.Therefore, it is necessary to a kind of solutions of method The above problem.
Invention content
The technical problems to be solved by the invention be for how valid metric video similarity problem, propose one kind be based on QPSO optimizes the video method for measuring similarity of Riemann manifold.The method is solving video human face confirmation and video attribute judgement Etc. have important value.
In order to solve the above-mentioned technical problem, the technical solution adopted in the present invention is：
A kind of video method for measuring similarity based on the study of QPSO Riemann manifolds, includes the following steps：
Each video is indicated by step A by image collection, and image calibration is carried out to the frame image in image collection, Tri- kinds of features of LBP, CSLBP, FPLBP of image set are extracted, and these three features of acquisition are merged to obtain image set The high dimensional feature of conjunction represents；
Step B carries out Riemann manifold dimension-reduction treatment using its high dimensional feature to all image sets and obtains all image sets Respective low-dimensional feature, the low-dimensional feature vector of all pictures forms the character representation of this image set in same image set；
The image set feature vector of acquisition is represented to be described in the form of affine hull, obtains image collection by step C Subspace Distribution measures the similarity between the two image sets by the distance between two sub-spaces；
Step D chooses a part as training set in video database, and rest part is gathered as test, to training Each video in set is sent into SVM classifier to a similarity can be calculated, by all similarities, may finally A grader is obtained, and to the test video of input to predicting, so as to obtain the recognition result whether video matching.
Riemann manifold dimension-reduction treatment described in step B is by points not reachable in the Riemann manifold dimensionality reduction result on presenting set Amount is used as fitness, and Riemann manifold optimized parameter group is obtained with quantum particle swarm optimization strategy, final feature vector not only dimension It is relatively low, and can represent the inherent attribute of video.
In step D, the described grader can predict whether two videos of video centering are under the jurisdiction of same classification, Can further expand for solve video human face confirm, video attribute judge the problems such as.
Beneficial effects of the present invention：The present invention proposes a kind of video measuring similarity based on QPSO optimization Riemann manifolds Method, the algorithm propose a kind of video method for measuring similarity based on QPSO optimization Riemann manifolds, and the algorithm is by video Similarity is considered as image set measuring similarity problem, texture feature extraction and carries out fusion treatment after image alignment, recycles band Have QPSO optimize Riemann manifold significantly to obtain the intrinsic representation of video data, the measurement of similarity then uses brief dimension Affine hull distance calibration method finally obtains optimal classification recognition result using libSVM.By being carried out on Relational database Contrast experiment, demonstrate the validity of put forward algorithm, carry that algorithm accuracy of identification is higher, and error is relatively low, and with relatively strong Robustness.Carried algorithm efficiently solves key frame in video and is difficult to selection while higher accuracy of identification is obtained Problem, and with stronger antijamming capability, to illumination variation the problems such as, also have preferable robustness.
Description of the drawings
Fig. 1 is that the present invention is based on the flow charts of the video method for measuring similarity of QPSO optimization Riemann manifolds.
Specific embodiment
Below in conjunction with the accompanying drawings, to a kind of video measuring similarity side based on QPSO optimization Riemann manifolds proposed by the present invention Method is described in detail：
As shown in Figure 1, the video method for measuring similarity of the present invention, its step are as follows：
Step 101, the frame image in video is extracted and carries out image calibration；
Step 102, the pretreatments such as dry, smooth are carried out to the image in image collection；
Step 103, using LBP, CSLBP, FPLBP operator extraction characteristics of image, and by the three classes Fusion Features of acquisition with The high dimensional feature for obtaining video represents, to enhance Video Key information；
Step 104, dimensionality reduction is carried out to video high dimensional feature using the Riemann manifold dimension reduction method with QPSO, on the one hand may be used It is represented with the data for obtaining low dimensional, improves efficiency of algorithm, on the other hand, the inherent table of video can be obtained by Riemann manifold Show, enhance the expression ability of algorithm, obtain more accurate recognition result；
Step 105, grader is obtained by SVM in the feature space of training sample to realize the category to test video pair Property judge.
With reference to Fig. 1 video method for measuring similarity based on QPSO optimization Riemann manifolds that the present invention will be described in detail.
First, feature extraction is carried out.Especially field of face identification is identified in image based on the feature extracting method of texture It is widely used, wherein LBP (Local Binary Patterns) is the classical operators of texture description.LBP is a kind of description image The binary system expression method of single pixel property (the different channels of gray scale or chroma image) Local size relationship, the operator have The advantages that calculating is simple, Scale invariant, rotational invariance.CSLBP(Center-Symmetric Local Binary Patterns operator) is improved using the LBP that two pixels of diagonal position are encoded, in pedestrian's test problems, CSLBP is compiled The texture feature information of code is notable, is widely used.
For any pixel point in given image, LBP encoded radios can be calculated by equation below：
The coding of CSLBP at the kth pixel of the jth frame of i-th of image set is defined as follows shown：
Wherein, threshold value tCSUsually take smaller value.
FPLBP (Four-Patch LBP) is a kind of LBP texture description operators based on piecemeal.The operator passes through in observation Cross reference between heart piecemeal and edge segmentation obtains Local textural feature, has to image type and localized variation preferable Robustness.The coding of FPLBP operators is defined as：
Wherein, r1、r2, S, ω, α represent the strategy that FPLBP selects block, P1iRepresent i-th of piecemeal of first lap, dist () table Show certain distance (Euclidean distance of such as gray difference) of two piecemeals.
Secondly, the RML dimension-reduction treatment with QPSO optimizations is carried out to the expression of higher-dimension video features.The Riemann that the present invention uses Manifold learning based on the assumption that：Higher-dimension input sample collection is there are intrinsic dimension, and sample set is distributed in the Riemann of such dimension In manifold.Then, in this low-dimensional Riemann manifold represent again be exactly Riemann manifold dimensionality reduction main thought.
The main three classes of parameter of Riemann manifold：(1) datum mark p decides the position of low-dimensional coordinate system, selects suitable base Calculation amount can be significantly reduced on schedule；(2) target dimension d；(3) selection of neighboring regions, wherein, k is represented near datum mark Neighbour's number, m represents neighbour's number of other sample points, r represents neighbour's maximum radius.
For datum mark p, selection is using manifold center herein.Calculate sample point between Euclidean distance and build non-directed graph and Distance matrix D={ distij=| | xc,i-xc,j| |, acquire multi-source shortest path.Each sample point to other sample points most The longest distance of short path is defined as the geometric radius of the point, using the point of geometric radius minimum as manifold center.
QPSO parameter optimizations are employed to the selection of target dimension and neighboring regions size herein.In PSO algorithm, The speed of particle is always limited, and algorithm is caused to cannot ensure convergence with probability 1 to globally optimal solution.And QPSO algorithms can guarantee Global convergence, and have many advantages, such as that control parameter is few, evolutionary process is simple, fast convergence rate and operation are simple.
Population scale in QPSO algorithms selects 20 here, and i-th of particle is in the position P in t generationsi(t) representing one group can The parameter of energy, and can all substitute position after parameter iteration according to the fine-grained following movement of fitness distribution progress of institute every time, Also known as Evolution of Population：
In formula, N represents population scale, and α (t) represents convergent-divergent coefficient, is related to convergence rate, and mbest (t+1) is represented The average value of the optimum position in each generation in preceding t is for population,It is to obey equally distributed random number.PPi(t+1) For representing the history optimum position P of i-th of particle in preceding t generationsi(t) and preceding t generation in all particles history optimum position Pg (t) random point between.Wherein, the value of α (t) is particularly significant, and constant can be selected constant, also can be empirically using change Amount, usually：
Wherein, Maxtime is maximum allowable iterations, and m, n are constant, here, select m=1, the experience of n=0.5 Value.
Next, the distance between affine hull measuring similarity image collection.For including ncThe image collection of a imageFor, after above-mentioned Riemann manifold optimum choice process, the image that a low-dimensional sample reconstitutes can be obtained Set X 'c, then X 'cConvex closure form be CH (X 'c)：
If it enablesAbove formula can be rewritten as：
For two given image collection XiAnd Xj, the similarity distance between them can be by solving following constraint Convex optimization problem obtains：
XiWith XjThe distance between can be expressed as：
Finally, the determined property to test video pair is performed.A part is chosen in the database as training set, remaining part It is allocated as gathering for test.The similarity information of all videos pair during training is gathered is sent into Linear SVM grader and is trained, Final classification device is obtained, and to the test video of input to predicting, so as to obtain the recognition result whether video matching.
Pass through the above embodiment, it is seen that the invention has the advantages that：
The present invention efficiently solves key frame in video and is difficult to asking for selection while higher accuracy of identification is obtained Topic, and with stronger antijamming capability.
In addition, the present invention obtains textural characteristics using LBP operators and improvement operator, the class operator is to illumination variation, rotation Turn to wait and there is preferable robustness.

Claims (3)

1. a kind of video method for measuring similarity based on the study of QPSO Riemann manifolds, includes the following steps：
Each video is indicated by step A by image collection, and image calibration, extraction are carried out to the frame image in image collection Tri- kinds of features of LBP, CSLBP, FPLBP of image set, and these three features of acquisition are merged to obtain image collection High dimensional feature represents；
Step B carries out Riemann manifold dimension-reduction treatment using its high dimensional feature to all image sets and obtains all image sets respectively Low-dimensional feature, the low-dimensional feature vector of all pictures forms the character representation of this image set in same image set；
The image set feature vector of acquisition is represented to be described in the form of affine hull by step C, and the son for obtaining image collection is empty Between be distributed, measure the similarity between the two image sets by the distance between two sub-spaces；
Step D chooses a part as training set in video database, and rest part is gathered as test, and training is gathered In each video to a similarity can be calculated, by all similarities be sent into SVM classifier, may finally obtain One grader, and to the test video of input to predicting, so as to obtain the recognition result whether video matching.
2. a kind of video method for measuring similarity based on the study of QPSO Riemann manifolds according to claim 1, feature It is, Riemann manifold dimension-reduction treatment described in step B, is that will not used in the Riemann manifold dimensionality reduction result on presenting set up to points amount To make fitness, Riemann manifold optimized parameter group is obtained with quantum particle swarm optimization strategy, not only dimension is relatively low for final feature vector, And it can represent the inherent attribute of video.
3. a kind of video method for measuring similarity based on the study of QPSO Riemann manifolds according to claim 1, feature It is, in step D, the described grader can predict whether two videos of video centering are under the jurisdiction of same classification, can With further expand for solve video human face confirm, video attribute judge the problems such as.
CN201810045439.4A 2018-01-17 2018-01-17 A kind of video method for measuring similarity based on the study of QPSO Riemann manifolds Pending CN108256478A (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN201810045439.4A CN108256478A (en) 2018-01-17 2018-01-17 A kind of video method for measuring similarity based on the study of QPSO Riemann manifolds

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN201810045439.4A CN108256478A (en) 2018-01-17 2018-01-17 A kind of video method for measuring similarity based on the study of QPSO Riemann manifolds

Publications (1)

Publication Number Publication Date
CN108256478A true CN108256478A (en) 2018-07-06

Family

ID=62726733

Family Applications (1)

Application Number Title Priority Date Filing Date
CN201810045439.4A Pending CN108256478A (en) 2018-01-17 2018-01-17 A kind of video method for measuring similarity based on the study of QPSO Riemann manifolds

Country Status (1)

Country Link
CN (1) CN108256478A (en)

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN112733758A (en) * 2021-01-15 2021-04-30 哈尔滨工业大学（深圳） Curve-based three-dimensional face recognition method and system under Riemann geometric invariance

Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN102779269A (en) * 2012-06-13 2012-11-14 合肥工业大学 Human face identification algorithm based on image sensor imaging system
CN103136736A (en) * 2013-03-19 2013-06-05 哈尔滨工业大学 Hyperspectral remote sensing data non-linear dimension descending method
CN104008397A (en) * 2014-06-09 2014-08-27 华侨大学 Target tracking algorithm based on image set
WO2015021381A1 (en) * 2013-08-08 2015-02-12 University Of Florida Research Foundation, Incorporated Real-time reconstruction of the human body and automated avatar synthesis

Patent Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN102779269A (en) * 2012-06-13 2012-11-14 合肥工业大学 Human face identification algorithm based on image sensor imaging system
CN103136736A (en) * 2013-03-19 2013-06-05 哈尔滨工业大学 Hyperspectral remote sensing data non-linear dimension descending method
WO2015021381A1 (en) * 2013-08-08 2015-02-12 University Of Florida Research Foundation, Incorporated Real-time reconstruction of the human body and automated avatar synthesis
CN104008397A (en) * 2014-06-09 2014-08-27 华侨大学 Target tracking algorithm based on image set

Non-Patent Citations (3)

* Cited by examiner, † Cited by third party
Title
BO LEI ET AL: "Parameter Selection of Generalized Fuzzy Entropy-based Thresholding Method with Quantum-Behavior Particle Swarm Optimization", 《ICALIP2008》 *
林国军: "基于流形学习和协同表示的人脸识别算法研究", 《国博士学位论文全文数据库 信息科技辑》 *
王玉: "基于图像集合和视频序列的视频人脸识别算法研究", 《中国博士学位论文全文数据库 信息科技辑》 *

Cited By (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN112733758A (en) * 2021-01-15 2021-04-30 哈尔滨工业大学（深圳） Curve-based three-dimensional face recognition method and system under Riemann geometric invariance
CN112733758B (en) * 2021-01-15 2023-09-01 哈尔滨工业大学（深圳） Curve-based three-dimensional face recognition method and system under Riemann geometric invariance

Similar Documents

Publication Publication Date Title
CN106228185B (en) A kind of general image classifying and identifying system neural network based and method
Gilani et al. Deep, dense and accurate 3D face correspondence for generating population specific deformable models
CN107633226B (en) Human body motion tracking feature processing method
CN109165540B (en) Pedestrian searching method and device based on prior candidate box selection strategy
CN110879961B (en) Lane detection method and device using lane model
CN109284779A (en) Object detecting method based on the full convolutional network of depth
CN110188763B (en) Image significance detection method based on improved graph model
CN108764019A (en) A kind of Video Events detection method based on multi-source deep learning
Wang et al. An unequal deep learning approach for 3-D point cloud segmentation
CN104077742B (en) Human face sketch synthetic method and system based on Gabor characteristic
CN113706547B (en) Unsupervised domain adaptive semantic segmentation method based on category dissimilarity guidance
CN113032613B (en) Three-dimensional model retrieval method based on interactive attention convolution neural network
CN105631469A (en) Bird image recognition method by multilayer sparse coding features
CN112364881B (en) Advanced sampling consistency image matching method
CN106919884A (en) Human facial expression recognition method and device
CN113313123A (en) Semantic inference based glance path prediction method
CN107123138A (en) Based on vanilla R points to rejecting tactful point cloud registration algorithm
Meng et al. Merged region based image retrieval
CN108256478A (en) A kind of video method for measuring similarity based on the study of QPSO Riemann manifolds
US11521427B1 (en) Ear detection method with deep learning pairwise model based on contextual information
CN116993760A (en) Gesture segmentation method, system, device and medium based on graph convolution and attention mechanism
CN113723468B (en) Object detection method of three-dimensional point cloud
CN113469270B (en) Semi-supervised intuitive clustering method based on decomposition multi-target differential evolution superpixel
CN111461130B (en) High-precision image semantic segmentation algorithm model and segmentation method
Zhou et al. Multi-attribute joint point cloud super-resolution with adversarial feature graph networks

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