CN107578049A - The symmetrical Gabor wavelet deep decomposition image classification method of circle - Google Patents
The symmetrical Gabor wavelet deep decomposition image classification method of circle Download PDFInfo
- Publication number
- CN107578049A CN107578049A CN201710775527.5A CN201710775527A CN107578049A CN 107578049 A CN107578049 A CN 107578049A CN 201710775527 A CN201710775527 A CN 201710775527A CN 107578049 A CN107578049 A CN 107578049A
- Authority
- CN
- China
- Prior art keywords
- mtd
- mrow
- csgw
- msub
- decomposition
- 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
Landscapes
- Information Retrieval, Db Structures And Fs Structures Therefor (AREA)
- Image Analysis (AREA)
Abstract
The present invention proposes a kind of image classification method, and this method uses the deep decomposition scheme based on the symmetrical Gabor wavelet (CSGW) of circle, referred to as DD CSGW.Because CSGW has invariable rotary characteristic, thus DD CSGW are also a kind of decomposition method of invariable rotary.Deep decomposition refers to the iteration and hierarchicabstract decomposition by using CSGW, it means that rougher subband can be by CSGW continuous decompositions into some finer subbands.Because DD CSGW transform domains have stronger dependence, therefore we are relied on using Copula models to capture these ratios, i.e., each layer of decomposition subband is portrayed with Copula models.Use the average of Copula model parameters and DD CSGW subbands and variance to be used for representing image simultaneously, and image classification is carried out using SVM.The deep decomposition method effect of the present invention has good performance, sorting phase speed in terms of image classification, and has selection invariant feature.
Description
Technical field
The present invention relates to image classification field, is revolved more particularly, to using a kind of of the symmetrical Gabor wavelet deep decomposition of circle
Turn constant image classification method.
Background technology
Image classification is the important research direction in computer vision.The mission critical of classification is how to use less area
Divide property information, the feature of image is generally referred to as, effectively to represent given image.For most of character representation sides
Method, if obtaining image at different directions visual angle, the feature extracted will be significantly different.Therefore, designed for graphical representation
Invariable rotary method is still an important and challenging job.It is known that the method based on wavelet transformation, including
Wavelet transform and Gabor wavelet conversion are not invariable rotaries, because different sub-band will produce difference in rotating condition
Feature.Both at home and abroad to have studied the technologies of some invariable rotaries based on wavelet transformation.Although having multi-resolution characteristics,
Method (such as discrete wavelet and Gabor wavelet) performance based on small echo still has much room for improvement.
The content of the invention
In consideration of it, the present invention designs the image classification method based on the symmetrical Gabor wavelet (CSGW) of circle.CSGW be by
The invariable rotary method based on Gabor filter of Porter and Canagarajah designs.Design CSGW purpose is to obtain
Rotational invariance.However, not as Gabor wavelet set direction characteristic, CSGW when applied to graphical analysis will produce compared with
Few identification information.Therefore, compared with Gabor wavelet, CSGW is relatively inefficient for non-rotating graphical representation.CSGW by
Expression determines：
hm(x, y)=λ-mhC(x ', y ')
WhereinIt is round symmetrical Gabor filters
Ripple device (CSGF).X '=λ-mX, y '=λ-my；λ-m, (m=0 ..., S-1) is scale parameter；M is scale parameter, and J is to decompose
Yardstick quantity.W is centre frequency, and σ is variance.
In order to obtain more distinction information, the present invention proposes a kind of deep decomposition method based on CSGW, referred to as DD-
CSGW.Deep decomposition refers to the iteration and hierarchicabstract decomposition by using CSGW, it means that rougher subband can pass through
CSGW continuous decompositions are into some finer subbands.It should be noted that deep decomposition is not simply to be divided image with CSGW
Solution is into more yardstick.Because with the increase of decomposition scale scale amounts, CSGW performance not always improves, and tests table
Bright, 5 yardstick sections are decomposed and are up to optimum performance.But deep decomposition proposed by the present invention is far better than the CSGW decomposabilities of 5 yardsticks
Can be with the performance of Gabor wavelet.
Because CSGW transform domains have stronger dependence, therefore we capture these ratios using Copula models
Rely on, i.e., each layer of decomposition subband is portrayed with Copula models.Copula models belong to multidimensional statistics model, including Copula
Two parts of function and some marginal distribution functions.Copula model h (x) represent as follows：
Wherein, fi(xi) and Fi(xi) respectively represent model marginal distribution density function and cumulative distribution function.c(F1
(x1),…,Fd(xd)) it is Copula density functions, the function is by Fi(xi) determine.The present invention by the use of Gaussian Copula as
Copula density functions；Edge distribution is used as by the use of the density function of Weibull distributions.Gaussian Copula represent as follows：
Wherein, ξ=[ξ1..., ξd], ξi=Φ-1(ui), i=1 ... d, Φ and Φ-1Represent normal distribution and its inverse function.R is
Gaussian Copula correlation matrix.The density function and cumulative distribution function difference of Weibull distributions are as follows：
Wherein, α and β is form parameter and scale parameter respectively.
In order to further improve performance, except using Copula model parameters, distribution is bright while decomposes son using DD-CSGW
The feature of average and standard deviation with coefficient as image.
For the image classification stage, the parameter attribute of the Copula models based on DD-CSGW, and DD-CSGW average
And standard deviation characteristic, normalize to [0,1], SVM (SVM) is used as grader.
Brief description of the drawings
Fig. 1 is the flow chart of the inventive method
Fig. 2 is the DD-CSGW decomposing schematic representations in the inventive method
Embodiment
The inventive method specific implementation step is as follows (see Fig. 1)：
Step 1, image is subjected to deep decomposition with CSGW, that is, carries out DD-CSGW decomposition and (see Fig. 2, given in figure two layers
The scheme of decomposition).
Step 1.1, first layer decomposes.With CSGW decompose input picture I (x, y), by picture breakdown into J yardsticks subband simultaneously
Its amplitude is taken, represents decomposition scale with S [i], i=1,2 ..., J respectively.The present invention takes J=5.It is formulated as：
S [i]=| hm(x, y) * I (x, y) |
Step 1.2, the second layer decomposes.Continue that subband S [i] is carried out into J Scale Decompositions respectively with CSGW, and calculate the layer point
Solve the amplitude S [i, j] of subband.
S [i, j]=| hm(x, y) * s [i] |
Step 1.3, L layers decompose.Continue that subband S [i, j ..., k] is carried out into J Scale Decompositions respectively with CSGW, and calculate
Decompose the amplitude S [i, j ..., k, l] of subband.The present invention takes L=3, that is, carries out the decomposition of 3 levels.
S [i, j ..., k, l]=| hm(x, y) * s [i, j ..., k] |
Step 2, characteristics of image is calculated.
Step 2.1, Copula models are built.Each layer of decomposition subband is portrayed with Copula models first.In model
Copula density functions Gaussian Copula, the density function that Marginal density function, is distributed with Weibull.So L layers point
Solution will L Copula model of parameter.To a sub-picture, deep decomposition of the invention will produce 3 Copula models.Copula
The parameter of model includes Copula density functions parameter and Marginal density function, parameter.With two benches (two-step) maximum likelihood
Method estimates the parameter of Copula models：First stage estimates the parameter of marginal density；Second stage estimation Copula functions
Parameter.Because the parameter R estimated is matrix (symmetrical matrix), it is necessary to stretch into vector, represent as follows：
Thus Copula model parameters XCPIt can be expressed as：
Wherein L represents Decomposition order,WithIt is expressed as the parameter of the edge distribution of l layers decomposition；Represent l layers
Element in the Copula density function parameters R of decomposition.
Step 2.2, the average and standard deviation that DD-CSGW decomposes subband are calculated.Each subband of every layer of decomposition is calculated respectively
The average and standard deviation of coefficient.The average and standard deviation characteristic X of modelenRepresent as follows：
Step 3, combinations of features.By Copula model parameter features XCP, and the average of model and standard deviation characteristic XenEnter
Row, which merges, obtains the feature X of image, represents as follows：
X=[XCP, Xes]
Step 4, svm classifier is used.Characteristics of image in being gathered with step 1-3 method extraction training, and to train SVM
Grader.After SVM classifier has been completed in training, the feature input SVM of present image extraction is subjected to discriminant classification.Carry out
When SVM is trained with differentiating, by feature normalization to [0,1].
Claims (3)
1. the symmetrical Gabor wavelet deep decomposition image classification method of circle, including：
Step 1, image is subjected to deep decomposition with CSGW, that is, carries out DD-CSGW decomposition.
Step 1.1, first layer decomposes.Input picture I (x, y) is decomposed with CSGW, picture breakdown into the subband of J yardsticks and is taken it
Amplitude, represent decomposition scale with S [i], i=1,2 ..., J respectively.The present invention takes J=5.It is formulated as：
S [i]=| hm(x, y) * I (x, y)
Step 1.2, the second layer decomposes.Continue that subband S [i] is carried out into J Scale Decompositions respectively with CSGW, and calculate and decompose subband
Amplitude S [i, j].
S [i, j]=| hm(x, y) * s [i] |
Step 1.3, L layers decompose.Continue that subband S [i, j ..., k] is carried out into J Scale Decompositions respectively with CSGW, and calculate decomposition
The amplitude S [i, j ..., k, l] of subband.The present invention takes L=3, that is, carries out the decomposition of 3 levels.
S [i, j ..., k, l]=| hm(x, y) * s [i, j ..., k] |
Step 2, characteristics of image is calculated.
Step 2.1, Copula models are built.Each layer of decomposition subband is portrayed with Copula models first.In model
Copula density functions Gaussian Copula, the density function that Marginal density function, is distributed with Weibull.So L layers point
Solution will L Copula model of parameter.To a sub-picture, deep decomposition of the invention will produce 3 Copula models.Copula
The parameter of model includes Copula density functions parameter and Marginal density function, parameter.With two benches maximal possibility estimation Copula
The parameter of model：First stage estimates the parameter of marginal density；Second stage estimates the parameter of Copula functions.Due to estimating
Parameter R be matrix (symmetrical matrix), it is necessary to stretch into vector, represent as follows：
<mrow>
<mi>R</mi>
<mo>=</mo>
<mfenced open = "[" close = "]">
<mtable>
<mtr>
<mtd>
<mn>1</mn>
</mtd>
<mtd>
<msub>
<mi>r</mi>
<mrow>
<mn>1</mn>
<mo>,</mo>
<mn>2</mn>
</mrow>
</msub>
</mtd>
<mtd>
<mn>...</mn>
</mtd>
<mtd>
<msub>
<mi>r</mi>
<mrow>
<mn>1</mn>
<mo>,</mo>
<mi>d</mi>
</mrow>
</msub>
</mtd>
</mtr>
<mtr>
<mtd>
<msub>
<mi>r</mi>
<mrow>
<mn>2</mn>
<mo>,</mo>
<mn>1</mn>
</mrow>
</msub>
</mtd>
<mtd>
<mn>1</mn>
</mtd>
<mtd>
<mn>...</mn>
</mtd>
<mtd>
<msub>
<mi>r</mi>
<mrow>
<mn>2</mn>
<mo>,</mo>
<mi>d</mi>
</mrow>
</msub>
</mtd>
</mtr>
<mtr>
<mtd>
<mn>...</mn>
</mtd>
<mtd>
<mn>...</mn>
</mtd>
<mtd>
<mn>...</mn>
</mtd>
<mtd>
<mn>...</mn>
</mtd>
</mtr>
<mtr>
<mtd>
<msub>
<mi>r</mi>
<mrow>
<mi>d</mi>
<mo>,</mo>
<mn>1</mn>
</mrow>
</msub>
</mtd>
<mtd>
<msub>
<mi>r</mi>
<mrow>
<mi>d</mi>
<mo>,</mo>
<mn>2</mn>
</mrow>
</msub>
</mtd>
<mtd>
<mn>...</mn>
</mtd>
<mtd>
<mn>1</mn>
</mtd>
</mtr>
</mtable>
</mfenced>
<mo>&RightArrow;</mo>
<mfenced open = "[" close = "]">
<mtable>
<mtr>
<mtd>
<msub>
<mi>r</mi>
<mrow>
<mn>1</mn>
<mo>,</mo>
<mn>2</mn>
</mrow>
</msub>
</mtd>
<mtd>
<mn>...</mn>
</mtd>
<mtd>
<msub>
<mi>r</mi>
<mrow>
<mi>d</mi>
<mo>-</mo>
<mn>1</mn>
<mo>,</mo>
<mi>d</mi>
</mrow>
</msub>
</mtd>
</mtr>
</mtable>
</mfenced>
</mrow>
Thus Copula model parameters XCPIt can be expressed as：
<mrow>
<msub>
<mi>X</mi>
<mrow>
<mi>C</mi>
<mi>P</mi>
</mrow>
</msub>
<mo>=</mo>
<msubsup>
<mrow>
<mo>&lsqb;</mo>
<mn>...</mn>
<mo>,</mo>
<msubsup>
<mi>&alpha;</mi>
<mi>k</mi>
<mi>l</mi>
</msubsup>
<mo>,</mo>
<msubsup>
<mi>&beta;</mi>
<mi>k</mi>
<mi>l</mi>
</msubsup>
<mo>,</mo>
<mn>...</mn>
<mo>,</mo>
<msubsup>
<mi>r</mi>
<mrow>
<mi>i</mi>
<mo>,</mo>
<mi>j</mi>
</mrow>
<mi>l</mi>
</msubsup>
<mo>,</mo>
<mn>...</mn>
<mo>&rsqb;</mo>
</mrow>
<mrow>
<mi>l</mi>
<mo>=</mo>
<mn>1</mn>
</mrow>
<mi>L</mi>
</msubsup>
</mrow>
Wherein L represents Decomposition order,WithIt is expressed as the parameter of the edge distribution of l layers decomposition；Represent l layers
Element in the Copula density function parameters R of decomposition.
Step 2.2, the average and standard deviation that DD-CSGW decomposes subband are calculated.Each sub-band coefficients of every layer of decomposition are calculated respectively
Average and standard deviation.The average and standard deviation characteristic X of modelenRepresent as follows：
<mrow>
<msub>
<mi>X</mi>
<mrow>
<mi>e</mi>
<mi>n</mi>
</mrow>
</msub>
<mo>=</mo>
<msubsup>
<mrow>
<mo>&lsqb;</mo>
<mn>...</mn>
<mo>,</mo>
<msubsup>
<mi>m</mi>
<mi>k</mi>
<mi>l</mi>
</msubsup>
<mo>,</mo>
<msubsup>
<mi>e</mi>
<mi>k</mi>
<mi>l</mi>
</msubsup>
<mo>,</mo>
<msubsup>
<mi>s</mi>
<mi>k</mi>
<mi>l</mi>
</msubsup>
<mn>...</mn>
<mo>&rsqb;</mo>
</mrow>
<mrow>
<mi>l</mi>
<mo>=</mo>
<mn>1</mn>
</mrow>
<mi>L</mi>
</msubsup>
</mrow>
Step 3, combinations of features.By the parameter attribute X of Copula modelsCP, and the average of model and standard deviation characteristic XenCarry out
Merging obtains the feature X of image, represents as follows：
X=[XCP, Xes]
Step 4, svm classifier is used.Characteristics of image in being gathered with step 1-3 method extraction training, and to train svm classifier
Device.After SVM classifier has been completed in training, the feature input SVM of present image extraction is subjected to discriminant classification.Carrying out SVM instructions
When practicing with differentiating, by feature normalization to [0,1].
A kind of 2. symmetrical Gabor wavelet deep decomposition image classification method of circle as claimed in claim 1, it is characterised in that：Profit
Characteristics of image is extracted with the symmetrical Gabor wavelet deep decomposition method of circle (being referred to as DD-CSGW), and DD-CSGW has invariable rotary
Characteristic.
A kind of 3. symmetrical Gabor wavelet deep decomposition image classification method of circle as claimed in claim 1, it is characterised in that：Together
When make use of DD-CSGW decompose subband Copula models parameter attribute, and average and standard deviation characteristic represent image.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201710775527.5A CN107578049B (en) | 2017-09-01 | 2017-09-01 | Circular symmetry Gabor wavelet depth decomposition image classification method |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201710775527.5A CN107578049B (en) | 2017-09-01 | 2017-09-01 | Circular symmetry Gabor wavelet depth decomposition image classification method |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN107578049A true CN107578049A (en) | 2018-01-12 |
| CN107578049B CN107578049B (en) | 2020-06-05 |
Family
ID=61031016
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201710775527.5A Expired - Fee Related CN107578049B (en) | 2017-09-01 | 2017-09-01 | Circular symmetry Gabor wavelet depth decomposition image classification method |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN107578049B (en) |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108256581A (en) * | 2018-01-19 | 2018-07-06 | 宜宾学院 | Gabor wavelet domain copula model image sorting techniques |
| CN108280470A (en) * | 2018-01-21 | 2018-07-13 | 宜宾学院 | Discrete wavelet domain copula model image sorting techniques |
Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106022218A (en) * | 2016-05-06 | 2016-10-12 | 浙江工业大学 | Palm print palm vein image layer fusion method based on wavelet transformation and Gabor filter |
| CN106570183A (en) * | 2016-11-14 | 2017-04-19 | 宜宾学院 | Color picture retrieval and classification method |
-
2017
- 2017-09-01 CN CN201710775527.5A patent/CN107578049B/en not_active Expired - Fee Related
Patent Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106022218A (en) * | 2016-05-06 | 2016-10-12 | 浙江工业大学 | Palm print palm vein image layer fusion method based on wavelet transformation and Gabor filter |
| CN106570183A (en) * | 2016-11-14 | 2017-04-19 | 宜宾学院 | Color picture retrieval and classification method |
Non-Patent Citations (3)
| Title |
|---|
| CHAORONG LI 等: "Color texture image retrieval based on Gaussian copula models of Gabor wavelets", 《PATTERN RECOGNITION》 * |
| 徐婉莹 等: "一种基于Gabor小波的局部特征尺度提取方法", 《中国图像图形学报》 * |
| 李朝荣 等: "小波域Copula多维模型纹理检索", 《中国科学》 * |
Cited By (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108256581A (en) * | 2018-01-19 | 2018-07-06 | 宜宾学院 | Gabor wavelet domain copula model image sorting techniques |
| CN108256581B (en) * | 2018-01-19 | 2021-06-11 | 宜宾学院 | Gabor wavelet domain copula model image classification method |
| CN108280470A (en) * | 2018-01-21 | 2018-07-13 | 宜宾学院 | Discrete wavelet domain copula model image sorting techniques |
| CN108280470B (en) * | 2018-01-21 | 2021-06-04 | 宜宾学院 | Discrete wavelet domain copula model image classification method |
Also Published As
| Publication number | Publication date |
|---|---|
| CN107578049B (en) | 2020-06-05 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN103093441B (en) | Based on the non-local mean of transform domain and the image de-noising method of two-varaible model | |
| Luu et al. | Contourlet appearance model for facial age estimation | |
| CN105701477B (en) | A kind of fabric defect detection method based on Stationary Wavelet Transform vision significance | |
| CN102063708B (en) | Image denoising method based on Treelet and non-local means | |
| CN102930149B (en) | Sensor network sensing information denoising processing method based on principal component analysis (PCA) and empirical mode decomposition (EMD) | |
| Gao et al. | CW-SSIM based image classification | |
| CN100573559C (en) | Self-adapting multi-dimension veins image segmenting method based on small echo and average drifting | |
| CN106570183B (en) | A kind of Color Image Retrieval and classification method | |
| CN102750385A (en) | Correlation-quality sequencing image retrieval method based on tag retrieval | |
| CN107133640A (en) | Image classification method based on topography's block description and Fei Sheer vectors | |
| CN105184744A (en) | Prior fuzzy kernel estimation method based on standardized sparse measurement image block | |
| CN104008537A (en) | Novel noise image fusion method based on CS-CT-CHMM | |
| CN104484890A (en) | Video target tracking method based on compound sparse model | |
| CN101493935A (en) | Synthetic aperture radar image segmentation method based on shear wave hidden Markov model | |
| CN107578049A (en) | The symmetrical Gabor wavelet deep decomposition image classification method of circle | |
| CN107895354A (en) | A kind of SAR image method for reducing speckle of high score three based on non-lower sampling Shearlet conversion | |
| CN101566688B (en) | Method for reducing speckle noises of SAR image based on neighborhood directivity information | |
| CN105913383A (en) | Image noise reduction method based on image block prior estimation mixed framework | |
| CN1975762A (en) | Skin detecting method | |
| CN104732498A (en) | Thresholding image denoising method based on nonsubsampled Contourlet transformation | |
| CN103247052A (en) | Image segmentation algorithm for local region characteristics through nonsubsampled contourlet transform | |
| Han et al. | Noise reduction method for chaotic signals based on dual-wavelet and spatial correlation | |
| Viswanathan et al. | Fingerprint enhancement and compression method using Morlet wavelet | |
| CN103345739B (en) | A kind of high-resolution remote sensing image building area index calculation method based on texture | |
| CN106204495A (en) | Image de-noising method based on non-down sampling contourlet transform and device |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant | ||
| CF01 | Termination of patent right due to non-payment of annual fee | ||
| CF01 | Termination of patent right due to non-payment of annual fee |
Granted publication date: 20200605 Termination date: 20210901 |