CN107862335A - A kind of texture image classification method based on multiple dimensioned feedback metric learning - Google Patents
A kind of texture image classification method based on multiple dimensioned feedback metric learning Download PDFInfo
- Publication number
- CN107862335A CN107862335A CN201711084549.3A CN201711084549A CN107862335A CN 107862335 A CN107862335 A CN 107862335A CN 201711084549 A CN201711084549 A CN 201711084549A CN 107862335 A CN107862335 A CN 107862335A
- Authority
- CN
- China
- Prior art keywords
- texture image
- texture
- mrow
- msub
- image
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Granted
Links
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
-
- 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
- G06V10/50—Extraction of image or video features by performing operations within image blocks; by using histograms, e.g. histogram of oriented gradients [HoG]; by summing image-intensity values; Projection analysis
Abstract
A kind of texture image classification method based on multiple dimensioned feedback metric learning, is related to pattern-recognition and technical field of image processing, comprises the following steps:Training texture image and test texture image are handled using profile wave convert, obtains training texture image and tests the profile marble band of texture image;By extracting different increment generation histogram feature in profile marble band;The weight of subband distance in texture image class distance definition is obtained by establishing model learningWAnd by circulating constantly amendmentW, utilizeWThe distance for calculating remaining test texture image and each classification training sample is classified so as to realize.Beneficial effect of the present invention:Solve the problems, such as that Texture classification accuracy caused by the imbalance of profile marble band is relatively low in current image texture expression and classification, has the characteristics that classification accuracy is high, robustness is strong, is of very high actual application value.
Description
Technical field
It is specifically a kind of to be measured based on multiple dimensioned feedback the present invention relates to pattern-recognition and technical field of image processing
The texture image classification method of study.
Background technology
Profile wave convert is a kind of very effective image direction multi-scale transform.Effective modeling of profile marble band and spy
Sign extraction is to build the important step that image texture represents.After profile wave convert is suggested, based on profile ripple
Video procession method gradually emerges.Profile wave convert is widely used in each point of video procession
In Zhi Wenti, major contribution is made that for horizontal improve of current video procession, is also mankind's Sci-Tech Level
Raising serves larger facilitation.Therefore, people are constantly subjected in past ten years, the research on profile ripple sum
Great attention.
Image texture represents and classified to be an important subject in Pattern recognition and image processing.Image texture point
Class refers to trains grader by the image texture block for giving label, then asks what image texture block to be sorted was classified
Topic, and a kind of problem of image recognition.But, Texture classification is important to notice that the classification of texture image, especially pays close attention to image
Texture information.In people's life and work, image texture analysis technique can be widely applied to image recognition, texture repair
Deng field, therefore, image texture represents and classification is always one of focus direction of video procession technical research.Recently
Many decades, researcher propose great amount of images texture representation and sorting technique, and image texture sorting technique substantially can be with present
It is divided into two major classes:Texture classifying method based on spatial domain and the texture classifying method based on transform domain.More chis are based in recent years
The texture method of degree transform domain is of great interest, and still, these texture classifying methods based on multi-scale transform domain are past
Toward the imbalance problem for ignoring multiple dimensioned subband, so as to cause the accuracy of constructed texture classifying method not ideal enough.
The content of the invention
The technical problems to be solved by the invention are to provide a kind of texture image point based on multiple dimensioned feedback metric learning
Class method, it is relatively low to solve Texture classification accuracy caused by the imbalance of profile marble band during current image texture is represented and classified
The problem of.
The present invention is that technical scheme is used by solving above-mentioned technical problem:One kind is based on multiple dimensioned anti-
The texture image classification method of metric learning is presented, is comprised the following steps:
Step 1: given C class texture images, will all be divided into two groups per a kind of texture image, one group is used to train, and referred to as trains line
Image is managed, another group is used to test, and texture image is referred to as tested, then using profile wave convert to training texture image and test
Texture image is handled, and is obtained training texture image and is tested the profile marble band of texture image;
Step 2: extract different increment in the profile marble band of the training texture image obtained by step 1 and test texture image
Generate histogram feature;
Step 3: by the method for integer subdivision, subdivision is done to test texture image number, and build feedback step-length vector V, V
=(v1,v2,…,vτ), wherein τ is the group number that subdivision is carried out to test texture image number;
Step 4: making m=1 to τ -1, feedback step-length vector V is traveled through successively, circulated as follows, until m >=τ -1:
(1) histogram feature generated for the different increment for training texture image profile marble band, solves following linear programming
Problem, to obtain the weight W of subband distance in texture image class distance definition:
So that
Wherein, W={ ωi,j, i=1,2 ..., L, j=1,2 ..., M, ωi,jRepresent the i-th yardstick, j-th of Directional Contour marble
The weight coefficient of band, L are the out to outs of profile Wave Decomposition, and M is the number of directional subband under each profile Wave Decomposition yardstick;
di,jThe i-th yardstick is represented, the different increment of j-th of Directional Contour marble band generates the distance between histogram feature, and c represents c-th of line
Manage class, c=1,2 ... C;SxAnd SyTwo texture images for belonging to c classes are represented respectively;μ is control parameter, for constraining in class
Distance;
(2) the weight W obtained using step (1) passes through formulaCalculate remaining test texture image with
Each the distance between training texture image, and using the minimum value of these distances as the test texture in c-th of texture classification
Distance of the image to texture classes c;
(3) according to principle of the test texture image to c-th of texture classification distance from small to large, v before retrievalmIndividual test texture maps
Picture, the test texture image that will be retrieved, belong to the other texture image of same class as with c classes training texture image,
And these test texture images are added the training texture image of c classes, among follow-up metric learning, carry out next time
Circulation;
Step 5: the subband distance weighting W for learning to obtain using step 4, calculate the individual test texture images of remaining v (τ) with it is each
The distance of classification training sample;
Step 6: texture image training sample institute corresponding with the minimum range in the distance of each classification training sample will be tested
The class label of category assigns test texture image and realizes classification.
Step (1) calculates the European of the different increment generation histogram feature of profile marble interband in step 4 of the present invention
Distance di,jFormula be:Wherein hkAnd gkThe i-th yardstick, j-th of Directional Contour marble are represented respectively
Histogram feature with the generation of different increment, D represent the dimension of histogram feature.
The beneficial effects of the invention are as follows:
(1) correlation analysis of the different increment generation histogram feature of profile marble band provided by the invention, is preferably solved
The modeling of profile marble band and dependency analysis problem;
(2) present invention is directed to the imbalance problem of image texture profile marble band, and the metric learning method established can be extensive
The fields such as image texture identification and retrieval are applied to, identification and retrieval are more accurate;
(3) quantitative comparison experiment proves the texture image classification method provided by the present invention based on multiple dimensioned feedback metric learning
Texture classification accuracy is relatively low caused by the imbalance of profile marble band in solving current image texture expression and classifying asks
Topic, have the characteristics that classification accuracy is high, robustness is strong, be of very high actual application value.
Brief description of the drawings
Fig. 1 is texture image classification method flow chart of the present invention based on multiple dimensioned feedback metric learning;
Fig. 2 is present invention classification accuracy rate and graph of a relation of training sample number on Set-1 data sets.
Embodiment
A kind of texture image classification method based on multiple dimensioned feedback metric learning, comprises the following steps:
Step 1: given C class texture images, will all be divided into two groups per a kind of texture image, one group is used to train, and referred to as trains line
Image is managed, another group is used to test, and texture image is referred to as tested, then using profile wave convert to training texture image and test
Texture image is handled, and is obtained training texture image and is tested the profile marble band of texture image;
Step 2: extract different increment in the profile marble band of the training texture image obtained by step 1 and test texture image
Generate histogram feature;
Step 3: by the method for integer subdivision, subdivision is done to test texture image number, and build feedback step-length vector V, V
=(v1,v2,…,vτ), wherein τ is the group number that subdivision is carried out to test texture image number;
Step 4: making m=1 to τ -1, feedback step-length vector V is traveled through successively, circulated as follows, until m >=τ -1:
(1) histogram feature generated for the different increment for training texture image profile marble band, solves following linear programming
Problem, to obtain the weight W of subband distance in texture image class distance definition:
So that
Wherein, W={ ωi,j, i=1,2 ..., L, j=1,2 ..., M, ωi,jRepresent the i-th yardstick, j-th of Directional Contour marble
The weight coefficient of band, L are the out to outs of profile Wave Decomposition, and M is the number of directional subband under each profile Wave Decomposition yardstick;
di,jThe i-th yardstick is represented, the different increment of j-th of Directional Contour marble band generates the distance between histogram feature,Wherein hkAnd gkThe i-th yardstick is represented respectively, the different increment generation of j-th of Directional Contour marble band
Histogram feature, D represent the dimension of histogram feature;C represents c-th of texture classes, c=1,2 ... C;SxAnd SyCategory is represented respectively
In two texture images of c classes;μ is control parameter, for constraining inter- object distance;
(2) the weight W obtained using step (1) passes through formulaCalculate remaining test texture image with
Each the distance between training texture image, and using the minimum value of these distances as the test texture in c-th of texture classification
Distance of the image to texture classes c;
(3) according to principle of the test texture image to c-th of texture classification distance from small to large, v before retrievalmIndividual test texture maps
Picture, the test texture image that will be retrieved, belong to the other texture image of same class as with c classes training texture image,
And these test texture images are added the training texture image of c classes, among follow-up metric learning, carry out next time
Circulation;
Step 5: the subband distance weighting W for learning to obtain using step 4, calculate the individual test texture images of remaining v (τ) with it is each
The distance of classification training sample;
Step 6: texture image training sample institute corresponding with the minimum range in the distance of each classification training sample will be tested
The class label of category assigns test texture image and realizes classification.
The effect of the present invention is described further by following emulation experiment:
1) simulated conditions
All experiment of the present invention be all central processing unit be@3.20GHZ of Intel (R) Xeon (R) CPU E5-1650 0, it is interior
Deposit in the operating system of 16GB, WINDOWS 10, with Matlab procedure simulations.The experimental data used in experiment is disclosed number
According to storehouse 61CUReT320, VisTex, with Brodatz.To two databases later, two data of our categorizing selections
Collection, so as to which we have five texture image datasets, are designated as Set-1, Set-2, Set-3, Set-4 and Set-5 respectively.To this
Each texture image that five data are concentrated, is all divided into 16 texture blocks, wherein 8 pieces are used for training sample, remaining 8 pieces are used for
Test sample.In our experiment, our in-service evaluation indexs are average correct classification rate (the average
classificationaccuracyrate(ACAR))。
2) emulation content
In order to verify the validity of the texture image classification method based on multiple dimensioned feedback metric learning of proposition, Wo Menxuan
Five conventional texture image datasets are selected, and five kinds of very typical texture classifying methods have been contrast experiment.This five kinds of allusion quotations
Type method is the BP methods based on bit plane probabilistic model, the LEH methods based on local energy histogram, based on Poisson hybrid guided mode
The PMM methods of type Bayes's classification, based on the CSC methods of profile marble band cluster, and the SLR side based on linear regression model (LRM)
Method.This five kinds of texture classifying methods are all nearest relatively typical methods.We include comparing result in Tables 1 and 2.
Table 1 is five kinds of typical texture classifying methods compared with classification accuracy rate of the inventive method on five data sets
Table;Table 2 is point of five kinds of typical texture classifying methods with the inventive method in the texture dataset containing 80 texture images
Class accuracy comparison sheet, from table 1, table 2, the classification of method (abbreviation MFML) proposed by the invention on five data sets
Effect is all more advantageous.Fig. 2 shows the classification accuracy rate and training sample of the method proposed on Set-1 data sets
The relation of this number.By table 1 and table 2 it can be seen that method proposed by the invention has compared with high-class accuracy and robustness
The features such as.
Classification accuracy rate of the table 1. on five data sets compares
Classification accuracy rate of the table 2. in the texture dataset containing 80 texture images compares
。
Claims (2)
- A kind of 1. texture image classification method based on multiple dimensioned feedback metric learning, it is characterised in that:Comprise the following steps:Step 1: given C class texture images, will all be divided into two groups per a kind of texture image, one group is used to train, and referred to as trains line Image is managed, another group is used to test, and texture image is referred to as tested, then using profile wave convert to training texture image and test Texture image is handled, and is obtained training texture image and is tested the profile marble band of texture image;Step 2: extract different increment in the profile marble band of the training texture image obtained by step 1 and test texture image Generate histogram feature;Step 3: by the method for integer subdivision, subdivision is done to test texture image number, and build feedback step-length vector V, V =(v1,v2,…,vτ), wherein τ is the group number that subdivision is carried out to test texture image number;Step 4: making m=1 to τ -1, feedback step-length vector V is traveled through successively, circulated as follows, until m >=τ -1:(1) histogram feature generated for the different increment for training texture image profile marble band, solves following linear programming Model, to obtain the weight W of subband distance in texture image class distance definition:<mrow> <munder> <mrow> <mi>m</mi> <mi>i</mi> <mi>n</mi> </mrow> <mi>W</mi> </munder> <munder> <mo>&Sigma;</mo> <mi>c</mi> </munder> <munder> <mo>&Sigma;</mo> <mrow> <msub> <mi>S</mi> <mi>x</mi> </msub> <mo>,</mo> <msub> <mi>S</mi> <mi>y</mi> </msub> </mrow> </munder> <mi>T</mi> <mi>D</mi> <mrow> <mo>(</mo> <msub> <mi>S</mi> <mi>x</mi> </msub> <mo>,</mo> <msub> <mi>S</mi> <mi>y</mi> </msub> <mo>)</mo> </mrow> </mrow>So that<mrow> <mi>T</mi> <mi>D</mi> <mrow> <mo>(</mo> <msub> <mi>S</mi> <mi>x</mi> </msub> <mo>,</mo> <msub> <mi>S</mi> <mi>y</mi> </msub> <mo>)</mo> </mrow> <mo>=</mo> <munderover> <mo>&Sigma;</mo> <mrow> <mi>i</mi> <mo>=</mo> <mn>1</mn> </mrow> <mi>L</mi> </munderover> <munderover> <mo>&Sigma;</mo> <mrow> <mi>j</mi> <mo>=</mo> <mn>1</mn> </mrow> <mi>M</mi> </munderover> <msup> <mn>2</mn> <mi>i</mi> </msup> <msub> <mi>&omega;</mi> <mrow> <mi>i</mi> <mo>,</mo> <mi>j</mi> </mrow> </msub> <msub> <mi>d</mi> <mrow> <mi>i</mi> <mo>,</mo> <mi>j</mi> </mrow> </msub> </mrow>Wherein, W={ ωi,j, i=1,2 ..., L, j=1,2 ..., M, ωi,jRepresent the i-th yardstick, j-th of Directional Contour marble The weight coefficient of band, L are the out to outs of profile Wave Decomposition, and M is the number of directional subband under each profile Wave Decomposition yardstick; di,jThe i-th yardstick is represented, the different increment of j-th of Directional Contour marble band generates the distance between histogram feature, and c represents c-th of line Manage class, c=1,2 ... C;SxAnd SyTwo texture images for belonging to c classes are represented respectively;μ is control parameter, for constraining in class Distance;(2) the weight W obtained using step (1) passes through formulaCalculate remaining test texture image and the Each the distance between training texture image, and using the minimum value of these distances as the test texture maps in c texture classification Distance as arriving texture classes c;(3) according to principle of the test texture image to c-th of texture classification distance from small to large, v before retrievalmIndividual test texture maps Picture, the test texture image that will be retrieved, belong to the other texture image of same class as with c classes training texture image, And these test texture images are added the training texture image of c classes, among follow-up metric learning, carry out next time Circulation;Step 5: the subband distance weighting W for learning to obtain using step 4, calculate the individual test texture images of remaining v (τ) with it is each The distance of classification training sample;Step 6: texture image training sample institute corresponding with the minimum range in the distance of each classification training sample will be tested The class label of category assigns test texture image and realizes classification.
- 2. a kind of texture image classification method based on multiple dimensioned feedback metric learning according to claim 1, its feature It is:Step (1) calculates the distance d between the different increment generation histogram feature of profile marble interband in the step 4i,jPublic affairs Formula is:Wherein hkAnd gkThe i-th yardstick, the different increment life of j-th of Directional Contour marble band are represented respectively Into histogram feature, D represent histogram feature dimension.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201711084549.3A CN107862335B (en) | 2017-11-07 | 2017-11-07 | Texture image classification method based on multi-scale feedback metric learning |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201711084549.3A CN107862335B (en) | 2017-11-07 | 2017-11-07 | Texture image classification method based on multi-scale feedback metric learning |
Publications (2)
Publication Number | Publication Date |
---|---|
CN107862335A true CN107862335A (en) | 2018-03-30 |
CN107862335B CN107862335B (en) | 2021-06-01 |
Family
ID=61701118
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201711084549.3A Active CN107862335B (en) | 2017-11-07 | 2017-11-07 | Texture image classification method based on multi-scale feedback metric learning |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN107862335B (en) |
Citations (8)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
KR100849691B1 (en) * | 2007-01-16 | 2008-07-31 | 광주과학기술원 | Method and Apparatus for Analyzing Image Texture, and Method and Apparatus for Image classification Using the Method and Apparatus |
US8064694B2 (en) * | 2006-06-21 | 2011-11-22 | Hewlett-Packard Development Company, L.P. | Nonhuman animal integument pixel classification |
CN103530647A (en) * | 2013-10-10 | 2014-01-22 | 哈尔滨工程大学 | Texture classification method on basis of fractional Fourier transform (FrFT) |
WO2015142923A1 (en) * | 2014-03-17 | 2015-09-24 | Carnegie Mellon University | Methods and systems for disease classification |
CN105224960A (en) * | 2015-11-04 | 2016-01-06 | 江南大学 | Based on the corn seed classification hyperspectral imagery model of cognition update method of clustering algorithm |
CN105320962A (en) * | 2015-10-21 | 2016-02-10 | 东南大学 | Pavement damage type identification method based on classifier ensemble |
CN106408029A (en) * | 2016-09-28 | 2017-02-15 | 河南科技大学 | Image texture classification method based on structural difference histogram |
CN106803093A (en) * | 2016-12-06 | 2017-06-06 | 同济大学 | A kind of plant species recognition methods based on blade textural characteristics and iOS platforms |
-
2017
- 2017-11-07 CN CN201711084549.3A patent/CN107862335B/en active Active
Patent Citations (8)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US8064694B2 (en) * | 2006-06-21 | 2011-11-22 | Hewlett-Packard Development Company, L.P. | Nonhuman animal integument pixel classification |
KR100849691B1 (en) * | 2007-01-16 | 2008-07-31 | 광주과학기술원 | Method and Apparatus for Analyzing Image Texture, and Method and Apparatus for Image classification Using the Method and Apparatus |
CN103530647A (en) * | 2013-10-10 | 2014-01-22 | 哈尔滨工程大学 | Texture classification method on basis of fractional Fourier transform (FrFT) |
WO2015142923A1 (en) * | 2014-03-17 | 2015-09-24 | Carnegie Mellon University | Methods and systems for disease classification |
CN105320962A (en) * | 2015-10-21 | 2016-02-10 | 东南大学 | Pavement damage type identification method based on classifier ensemble |
CN105224960A (en) * | 2015-11-04 | 2016-01-06 | 江南大学 | Based on the corn seed classification hyperspectral imagery model of cognition update method of clustering algorithm |
CN106408029A (en) * | 2016-09-28 | 2017-02-15 | 河南科技大学 | Image texture classification method based on structural difference histogram |
CN106803093A (en) * | 2016-12-06 | 2017-06-06 | 同济大学 | A kind of plant species recognition methods based on blade textural characteristics and iOS platforms |
Non-Patent Citations (2)
Title |
---|
YONGSHENGDONG等: "Feature extraction through contourlet subband clustering for texture classification", 《NEUROCOMPUTING》 * |
杨晓鸣: "基于多尺度几何分析的图像特征提取与分类", 《中国优秀硕士学位论文全文数据库信息科技辑》 * |
Also Published As
Publication number | Publication date |
---|---|
CN107862335B (en) | 2021-06-01 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN109344736B (en) | Static image crowd counting method based on joint learning | |
CN108564129B (en) | Trajectory data classification method based on generation countermeasure network | |
Zhang et al. | MCnet: Multiple context information segmentation network of no-service rail surface defects | |
CN109190524B (en) | Human body action recognition method based on generation of confrontation network | |
CN105809672B (en) | A kind of image multiple target collaboration dividing method constrained based on super-pixel and structuring | |
CN104537647A (en) | Target detection method and device | |
CN110991532B (en) | Scene graph generation method based on relational visual attention mechanism | |
CN105389583A (en) | Image classifier generation method, and image classification method and device | |
CN105261004A (en) | Mean shift and neighborhood information based fuzzy C-mean image segmentation method | |
CN108416347A (en) | Well-marked target detection algorithm based on boundary priori and iteration optimization | |
CN105740404A (en) | Label association method and device | |
CN105930846A (en) | Neighborhood information and SVGDL (support vector guide dictionary learning)-based polarimetric SAR image classification method | |
CN103177265B (en) | High-definition image classification method based on kernel function Yu sparse coding | |
CN107133640A (en) | Image classification method based on topography's block description and Fei Sheer vectors | |
CN107944428A (en) | A kind of indoor scene semanteme marking method based on super-pixel collection | |
CN103745233B (en) | The hyperspectral image classification method migrated based on spatial information | |
CN105046714A (en) | Unsupervised image segmentation method based on super pixels and target discovering mechanism | |
CN103955709A (en) | Weighted synthetic kernel and triple markov field (TMF) based polarimetric synthetic aperture radar (SAR) image classification method | |
CN114692741A (en) | Generalized face counterfeiting detection method based on domain invariant features | |
Liu et al. | Unveiling patterns: A study on semi-supervised classification of strip surface defects | |
CN111144462A (en) | Unknown individual identification method and device for radar signals | |
Wang et al. | Learning latent semantic attributes for zero-shot object detection | |
CN105184285A (en) | Posture-spanning colored image facial expression recognition of direct push type migration group sparse discriminant analysis | |
CN113159142A (en) | Loss function variable hyper-parameter determination method for fine-grained image classification | |
CN106095811A (en) | A kind of image search method of the discrete Hash of supervision based on optimum code |
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 |