CN108921853A - Image partition method based on super-pixel and clustering of immunity sparse spectrums - Google Patents
Image partition method based on super-pixel and clustering of immunity sparse spectrums Download PDFInfo
- Publication number
- CN108921853A CN108921853A CN201810651506.7A CN201810651506A CN108921853A CN 108921853 A CN108921853 A CN 108921853A CN 201810651506 A CN201810651506 A CN 201810651506A CN 108921853 A CN108921853 A CN 108921853A
- Authority
- CN
- China
- Prior art keywords
- antibody
- super
- pixel
- pop
- indicate
- 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
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/10—Segmentation; Edge detection
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/23—Clustering techniques
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/004—Artificial life, i.e. computing arrangements simulating life
- G06N3/006—Artificial life, i.e. computing arrangements simulating life based on simulated virtual individual or collective life forms, e.g. social simulations or particle swarm optimisation [PSO]
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/10—Image acquisition modality
- G06T2207/10004—Still image; Photographic image
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Data Mining & Analysis (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- Life Sciences & Earth Sciences (AREA)
- Artificial Intelligence (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Evolutionary Computation (AREA)
- Bioinformatics & Computational Biology (AREA)
- Evolutionary Biology (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Health & Medical Sciences (AREA)
- Biomedical Technology (AREA)
- Biophysics (AREA)
- Computational Linguistics (AREA)
- General Health & Medical Sciences (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Image Analysis (AREA)
Abstract
The invention discloses a kind of image partition method based on super-pixel and clustering of immunity sparse spectrums mainly solves the problems, such as the accurate low and poor robustness of the segmentation of conventional images dividing method.Its method and step is:Firstly, carrying out super-pixel division to texture image, its textural characteristics is extracted, as characteristic data set;Then, immune clone algorithm and rarefaction representation is combined to find the similarity matrix of optimal characteristic data set during spectral clustering;Finally, marking original image according to cluster labels combination super-pixel, the segmentation to texture image is realized.The present invention extracts the super-pixel block of image as characteristic data set, is divided with the image partition method based on clustering of immunity sparse spectrums to characteristic data set, obtains more accurate segmentation result.
Description
Technical field
The invention belongs to technical field of image processing, in particular to a kind of image Segmentation Technology can be used for target identification.
Background technique
The common method of image segmentation has the dividing method based on threshold value, the dividing method based on edge and based on region
Dividing method etc. also obtains extensively with the theoretical developments of pattern-recognition and artificial intelligence in conjunction with the dividing method of specific theory
Application, such as the cutting techniques based on cluster, the cutting techniques based on artificial neural network, the segmentation skill based on genetic algorithm
Art etc..In existing clustering method, spectral clustering is a kind of simple, unrelated and global optimizing with dimension good with realizing
Characteristic clustering method.It is a popular research side in image segmentation field in recent years that spectral clustering, which is applied to image segmentation,
To.The process of image segmentation be exactly using each pixel as a data point, segmentation the result is that assigning one for these data points
A category.Pixel with same category is divided into one kind, to realize the segmentation to image.
The key of spectral clustering is the similarity matrix for calculating N × N, and N is number of samples, for large-scale data, calculation amount
It is too big.
Juanjuan Luo, Licheng Jiao, and Jose A.Lozano. are in paper " A Sparse Spectral
Clustering Framework via Multiobjective Evolutionary Algorithm”(IEEE
Transactions on Evolutionary Computation20.3(2016):418-433.) in propose one kind and be based on
The sparse spectral clustering frame of multiple-objection optimization.This method by rarefaction representation in conjunction with the genetic algorithm NSGA2 of non-dominated ranking,
The similarity matrix of spectral clustering is optimized.The method improve the generations of similarity matrix, improve the cluster of spectral clustering
Accuracy rate, but this method is not good enough in the diversity that convergence rate is conciliate, and is estimated using RC and chosen as last solution
Index, no NC estimates accurately.
Xian Electronics Science and Technology University is in patent " being based on clustering of immunity sparse spectrums image partition method " (patent Shen of its application
Please numbers 200910024374.6, publication number CN101673398A) in disclose a kind of image based on clustering of immunity sparse spectrums point
Segmentation method.It is combined the process employs immune clone algorithm and spectral clustering and image is clustered, realize point of image
It cuts.Though this method does not need priori knowledge, computation complexity is low, and there are still the not high disadvantages of segmentation precision.
Summary of the invention
It is an object of the invention to overcome the shortcomings of above-mentioned prior art, propose a kind of based on super-pixel and immunity sparse spectrums
The image partition method of cluster improves segmentation precision to extract more succinct effective feature.
Basic ideas of the invention are:By carrying out super-pixel division to texture image, its textural characteristics is extracted, is gathered
Class data set;By using immune clone algorithm and rarefaction representation during spectral clustering, optimal cluster data collection is searched out
Similarity matrix;Original image is marked by cluster labels combination super-pixel, texture image is split.
According to above-mentioned thinking, realization step of the invention includes as follows:
(1) image to be split that one width size of input is 256*256;
(2) segmented image is treated using simple linear iteration cluster SLIC method divide super-pixel;
(3) feature of super-pixel is extracted using gray level co-occurrence matrixes and gabor wavelet transformation;
(4) all super-pixel characteristics are normalized between [0,1], to remove the influence of magnitude between data, is obtained
Super-pixel characteristic data set I after normalization;
(5) to the super-pixel characteristic data set I after normalization, non-dominant antibody collection is obtained using immune clonal selection method
P;
(6) the non-dominant antibody of each of non-dominant antibody collection P is successively used as to the similarity matrix of spectral clustering process, to super
Pixel characteristic data collection I carries out spectral clustering, obtains a series of cluster results about super-pixel characteristic data set I;
(7) NC for calculating each cluster result estimates, and chooses the smallest NC and estimates corresponding cluster result as super-pixel
The final cluster result of characteristic data set I;
It (8) is corresponding to it by the category label of each pixel according to the cluster result of super-pixel characteristic data set I
The classification of super-pixel point completes image segmentation as the cluster result of image to be split.
Compared with the prior art, the present invention has the following advantages:
First, since the present invention is in the cluster process of image segmentation, uses super-pixel and handle, obtain representative
Super-pixel block, overcome disadvantage when carrying out spectral clustering with primitive image features more than number of samples, reduce calculation amount;
Second, since the present invention is in the cluster process of image segmentation, use immune clone algorithm and sparse spectral clustering
It combines, widely distributed, various candidate solution can be obtained, improve the segmentation precision of image;
Third estimates selection last solution using NC, overcomes RC since the present invention is in the cluster process of image segmentation
Estimate the defect for not accounting for similar degree in the class, improves the accuracy of image segmentation.
Detailed description of the invention
Fig. 1 is implementation flow chart of the invention;
Fig. 2 is with the present invention and the existing segmentation comparison diagram there are two types of texture image segmenting method to a 2 class texture images;
Fig. 3 is to be compared with the present invention and existing two kinds of texture image segmenting methods to the segmentation of a 3 class texture images
Figure.
Specific embodiment
The present invention is described in further detail with reference to the accompanying drawing.
Step 1, image to be split is inputted.
The texture image that a width size is A=256*256, image to be split as input are chosen from database.
Step 2, it treats segmented image and divides super-pixel.
The existing method for dividing super-pixel has the method based on graph theory, the method based on entropy rate, based on gradient decline
Method Meanshift, simple linear iterative clustering methods SLIC and the Level Set Method Turbopixels based on geometry flow etc..
Image to be split is divided into n super-pixel point using simple linear iterative clustering methods SLIC by this example, and implementation step is such as
Under:
(2a) uniformly chooses n pixel as initial seed point, for each initial seed point point in image to be split
With a label, each seed point includes A/n pixel, and the distance between seed point is
(2b) reselects seed point in the 2S*2S neighborhood of seed point, that is, calculates the ladder of all pixels point in the neighborhood
Seed point is moved on to the smallest place of neighborhood inside gradient, it is avoided to fall in the Position disturbance cluster process of image boundary by angle value;
(2c) calculates the pixel and seed point distance D ' in neighborhood in the 2S*2S neighborhood of each seed point,Wherein, dcIndicate color distance lj、ajAnd bjIt respectively indicates
The colouring information of seed point, li、aiAnd biRespectively indicate the colouring information of pixel, dsRepresentation space distancexjAnd yjRespectively indicate the spatial information of seed point, xiAnd yiRespectively indicate the space letter of pixel
Breath, m indicates balance parameters, for measuring the specific gravity of color value and spatial information in measuring similarity, by each pixel
Label be set as the label with its most similar seed point;
(2d) constantly iterative process (2b) and (2c) obtains n seed point until convergence, uses these seed points as surpassing
Pixel obtains the set V={ v of super-pixel point1,v2,…,vi,…,vn, viIndicate i-th of super-pixel point.
Step 3, the characteristic data set of super-pixel is extracted.
The method for extracting the feature of texture image has gray level co-occurrence matrixes, wavelet transformation and undecimated wavelet transform etc.,
The method that this example uses is gray level co-occurrence matrixes and gabor wavelet transformation, is realized as follows:
(3a) obtains 8 features of all pixels point of image to be split by gray level co-occurrence matrixes:
Image vector to be split is turned to 16 gray levels by (3a1);
Each pixel that (3a2) treats segmented image is generated in the window of 5*5 in distance d=1, and direction is θ=0 °,
θ=45 °, θ=90 °, gray level co-occurrence matrixes p (i, j), i and the j of θ=135 ° four direction are respectively gray level co-occurrence matrixes transverse direction
With longitudinal index value;
(3a3) calculates the contrasts of gray level co-occurrence matrixes, energy, entropy, correlation this four statistics:
Contrast:
Energy:
Entropy:
Correlation:
Wherein p ' (i, j) is the gray level co-occurrence matrixes after normalization, μx、σxRespectively indicate p ' (i, j) along the x-axis direction equal
Value and variance, μy、σyTable shows the mean value and variance of p ' (i, j) along the y-axis direction respectively;
(3a4) calculates separately the mean value and variance of the contrasts of four gray level co-occurrence matrixes, energy, entropy, correlation, altogether
Obtain 8 feature vectors of image all pixels point to be split;
(3b) obtains 2 features of all pixels point of image to be split by gabor wavelet transformation:
Image to be split is transformed into domain space using two different gabor filter groups, extracts image to be split
Two different scale different directions textural characteristics, two different gabor filter groups are respectively G1And G2, indicate such as
Under:
G2=Gcos [2 π f (xcos θ+ysin θ)]
Wherein, G (x, y) indicates Gabor function, sx、sy
Respectively indicate the variance along x-axis and y-axis, f is the frequency of SIN function, and θ indicates the direction of gabor filter, one be obtained to
2 feature vectors of segmented image all pixels point;
Gray level co-occurrence matrixes are obtained 8 features of all pixels point and obtain by gabor wavelet transformation 2 by (3c)
Feature merges, and each pixel obtains 10 features, to the textural characteristics for several pixels that each super-pixel point includes
Data are averaged, and as the characteristic data value of the super-pixel point, one is obtained the super-pixel characteristic data set that size is n*10
H, n indicate the number of super-pixel point, and 10 indicate the number of feature.
Step 4, super-pixel characteristic data set H is normalized between [0,1], to remove the influence of magnitude between data, is obtained
Super-pixel characteristic data set I after to normalization.
(4a) chooses the either element in super-pixel characteristic data set H;
(4b) calculates the difference of least member in taken element and the element column;
(4c) calculates the difference of greatest member and least member in the element column in the element column;
The difference that (4d) obtains the difference that (4b) is obtained divided by (4c) obtains the normalization result of taken element;
(4e) judges whether the element in super-pixel characteristic data set H has taken, if so, by the normalizing of all elements
Change the super-pixel characteristic data set I that result is merged into after normalization, otherwise, returns (4a).
Step 5, non-dominant antibody population is obtained using immune clone algorithm.
(5a) initializes antibody population:
Maximum number of iterations T=20, antibody population scale N=30 are set, it is random to generate initial antibodies group POP (t)=(X1,
X2..., Xl..., XN), wherein t indicates the number of iterations, is initialized as 0, XlIndicate that first antibody l=1 ..., N, N indicate anti-
Body group's scale,xijIndicate the similitude of i-th of super-pixel point and j-th of super-pixel point, i=
1 ..., n, j=1 ..., n, n indicate the number of super-pixel point,Rand is between one 0 to 1
Random number;
(5b) calculates the objective function of each antibody:
Wherein f1(Xl) indicate antibody XlError project scalar functions, f2(Xl) indicate antibody XlSparse item objective function, I
Indicate the characteristic of super-pixel, | | | |2 2Indicate square of 2 norms, | | | |0Indicate 0 norm;
(5c) selection operation
Two antibody X are selected from antibody population POPlAnd Xm, judge first antibody XlWith secondary antibody XmBetween whether meet
(f1(Xl)≤f1(Xm)&f2(Xl)<f2(Xm)) or (f1(Xl)<f1(Xm)&f2(Xl)≤f2(Xm)):
If it is, antibody XlFor non-dominant antibody, it being capable of dominated antibody Xm;
Otherwise, antibody XlIt is not non-dominated antibody, it is unable to dominated antibody Xm;
Wherein, f1(Xl) and f2(Xl) respectively indicate first antibody XlError project scalar functions and sparse item objective function
Value, f1(Xm) and f2(Xm) respectively indicate second antibody XmError project scalar functions and sparse item objective function value;
All non-dominant antibody is selected from antibody population POP, forms non-dominant antibody population POP1;
(5d) clone operations
To non-dominant antibody population POP1The antibody population that the operation of 4 time clonings generates is carried out, is expressed as:
POP2={ POP1, POP1, POP1, POP1}。
(5e) crossover operation
Two 0 to 1 random numbers, i.e. the first random number α, the second random number β and third random number η is arranged in (5e1);
First random number and given threshold 0.5 are compared by (5e2):
If α>0.5, then from the antibody population POP after clone2In randomly select two antibody Xr1And Xr2Uniform crossover is carried out,
Generate new antibody:Xr3, Xr3In element
If α≤0.5, from the antibody population POP after clone2In randomly select two antibody Xr1And Xr2Carry out arithmetic friendship
Fork, generates new antibody:Xr4=η Xr1+(1-η)Xr2;
(5e3) repetitive operation (5e2) n times generate N number of new antibodies, the antibody population POP after composition intersection3。
(5f) mutation operation
(5f1) chooses the antibody population POP after intersecting3In any one antibody XlIf antibody XlIn i-th super-pixel
The similitude x of point and j-th of super-pixel pointijVariation criterion be:
The wherein random number that rand and γ is 0 to 1, h are neighbour time of j-th of the super-pixel o'clock about i-th of super-pixel point
Sequence;
(5f2) is to the antibody population POP after intersection3In N number of antibody according to variation criterion successively carry out mutation operation, generate
N number of new antibodies, the antibody population POP after Composition Variation4。
(5g) judges whether the number of iterations has reached maximum number of iterations T, if so, by antibody population POP4It is middle to be used as finally
Antibody population P, execute step (6), otherwise, return step (5b), carry out next iteration.
Step 6, spectral clustering is carried out to the non-dominant antibody population P that immune clone algorithm obtains.
(6a) is for the antibody X in non-dominant antibody population Pl, l=1 ..., N, calculating degree matrixWhereinWherein xijIt indicates between i-th of super-pixel point and j-th of super-pixel point
Similitude, n indicate the number of super-pixel point;
(6b) calculates Laplacian Matrix:L=D-Xl;
(6c) carries out Eigenvalues Decomposition to Laplacian Matrix L according to Eigenvalues Decomposition formula Lu=λ u, obtains n spy
Vector λ, λ=[λ of value indicative composition1, λ2..., λi..., λn] and n feature vector composition matrix u, u=[u1, u2...,
ui..., un], wherein eigenvalue λiCorresponding to feature vector ui;
(6d) sorts from small to large to the characteristic value of vector λ, obtain n sequence after eigenvalue cluster at vector
λ ', λ '=[λ1', λ2' ..., λn'], h the smallest eigenvalue clusters are at vector λ ", λ "=[λ before choosing1', λ2' ..., λh'],
Corresponding feature vector forms matrix u ', u '=[u1', u2' ..., uh'], matrix u " is obtained to matrix u ' normalization;
(6e) divides u " using kmeans method, obtains the cluster result V, V=of super-pixel characteristic data set I
(V1..., Vi..., Vn), ViIndicate the super-pixel point set of the i-th class in cluster result.
Step 7, the NC for calculating each cluster result estimates, and chooses the smallest NC and estimates corresponding cluster result as super picture
The cluster result of plain characteristic data set I.
(7a) is estimated by the NC that following formula calculates each cluster result:
Wherein, c indicates the class number of cluster,Weight between expression class,
VhIndicate the super-pixel point set of h class in cluster result,Indicate the super-pixel point set that h class is not belonging in cluster result
It closes, wijIndicate the standard similitude between i-th of super-pixel point and j-th of super-pixel point, Indicate antibody Xl
In similitude between i-th of super-pixel point and j-th of super-pixel point, l=1 ..., N, N indicates the poly- of non-dominant antibody tormation
The number of class result, vol (Vh) indicate subset VhWeight,diI-th of element in degree of a representation matrix;
The NC that all cluster results have been calculated in (7b) estimates, and chooses the smallest NC and estimates corresponding cluster result as super picture
The cluster result of plain data set I.
Step 8, according to the cluster result of super-pixel characteristic data set I, the category label of each pixel is right for its
The classification for the super-pixel point answered completes image segmentation as the cluster result of image to be split.
Effect of the invention can be further illustrated by following emulation:
1. simulated environment:
It is core i5 3.30GHZ, 7 system of memory 8G, WINDOWS and meter that emulation of the invention, which is in allocation of computer,
Calculation machine software configuration is to carry out under Matlab R2014b environment.
2. emulation content:
Using the K Mean Method of the method for the present invention and the prior art, fuzzy C-mean algorithm method respectively to two classes and three classes line
The emulation experiment that reason image is split evaluates these image segmentations from segmentation precision and edge holding, details reservation etc.
The performance of method.
Emulation 1, is split using the texture image that the present invention may be partitioned into two classes to a selected width, as a result as schemed
2, wherein
Fig. 2 (a) is the texture image that a width may be partitioned into two classes;
Fig. 2 (b) is the emulation segmentation result figure being split using the method for the present invention to Fig. 2 (a);
Fig. 2 (c) is the emulation segmentation result figure being split using K mean value to Fig. 2 (a);
Fig. 2 (d) is the emulation segmentation result figure being split using Fuzzy C-Means Cluster Algorithm to Fig. 2 (a);
Fig. 2 (e) is the segmentation template that may be partitioned into the texture image figure of two classes.
Emulation 2, is split using the texture image that the present invention may be partitioned into three classes to a selected width, as a result as schemed
3, wherein:
Fig. 3 (a) is the texture image figure that a width may be partitioned into three classes;
Fig. 3 (b) is the emulation segmentation result figure being split using the method for the present invention to Fig. 3 (a);
Fig. 3 (c) is the emulation segmentation result figure being split using K mean value to Fig. 3 (a);
Fig. 3 (d) is the emulation segmentation result figure being split using Fuzzy C-Means Cluster Algorithm to Fig. 3 (a);
Fig. 3 (e) is the segmentation template that may be partitioned into the texture image figure of three classes.
Segmentation accuracy result of three kinds of methods to both above-mentioned texture images is subjected to counting statistics, as a result such as table 1
It is shown.
Segmentation accuracy of the 1 three kinds of methods of table to both texture images
The present invention | K mean value | Fuzzy C-mean algorithm | |
Two class texture images | 0.9804 | 0.9629 | 0.9485 |
Three classes texture image | 0.8964 | 0.8026 | 0.7655 |
3. analysis of simulation result:
From in Fig. 2 and table 1 as can be seen that Fig. 2 (b) is better than the segmentation result in Fig. 2 (c) and Fig. 2 (d), Fig. 2 (c) and Fig. 2
(d) more in miscellaneous point ratio Fig. 2 (b) in, but the edge of Fig. 2 (c) does not have Fig. 2 (b) clearly, shows the present invention and existing K mean value
It is compared with Fuzzy C-Means Cluster Algorithm with stronger robustness and more accurate image segmentation result can be obtained.
From in Fig. 3 and table 1 as can be seen that Fig. 3 (b) is better than the segmentation result in Fig. 3 (c) and Fig. 3 (d), Fig. 3 (c) and Fig. 3
(d) more in miscellaneous point ratio Fig. 3 (b) in, but the edge of Fig. 3 (c) does not have Fig. 3 (b) clearly, shows the present invention and existing K mean value
Comparing with Fuzzy C-Means Cluster Algorithm also has stronger robustness in the segmentation to three classes texture image and can obtain more quasi-
True image segmentation result.
To sum up, compared with existing K mean value and Fuzzy C-Means Cluster Algorithm, accuracy rate has significantly to be promoted the method for the present invention.
Claims (7)
1. a kind of image partition method based on super-pixel and clustering of immunity sparse spectrums, including it is as follows:
(1) image to be split that one width size of input is 256*256;
(2) segmented image is treated using simple linear iteration cluster SLIC method divide super-pixel;
(3) feature of super-pixel is extracted using gray level co-occurrence matrixes and gabor wavelet transformation;
(4) all super-pixel characteristics are normalized between [0,1], to remove the influence of magnitude between data, obtains normalizing
Super-pixel characteristic data set I after change;
(5) to the super-pixel characteristic data set I after normalization, non-dominant antibody collection P is obtained using immune clonal selection method;
(6) the non-dominant antibody of each of non-dominant antibody collection P is successively used as to the similarity matrix of spectral clustering process, to super-pixel
Characteristic data set I carries out spectral clustering, obtains a series of cluster results about super-pixel characteristic data set I;
(7) NC for calculating each cluster result estimates, and chooses the smallest NC and estimates corresponding cluster result as super-pixel feature
The final cluster result of data set I;
It (8) is the super picture corresponding to it by the category label of each pixel according to the cluster result of super-pixel characteristic data set I
The classification of vegetarian refreshments completes image segmentation as the cluster result of image to be split.
2. according to the method described in claim 1, the wherein step (5), carries out as follows:
(5a) initializes antibody population:
It is random to generate initial antibodies group POP (t)=(X1, X2..., Xl..., XN), wherein t indicates the number of iterations, it is initialized as 0,
XlIndicate that first antibody l=1 ..., N, N indicate antibody population scale,xijExpression i-th surpasses picture
The similitude of vegetarian refreshments and j-th of super-pixel point, i=1 ..., n, j=1 ..., n, n indicate the number of super-pixel point,Rand is the random number between one 0 to 1;
(5b) calculates the objective function of each antibody:
Wherein f1(Xl) indicate antibody XlError project scalar functions, f2(Xl) indicate antibody XlSparse item objective function, I indicate
The characteristic of super-pixel,Indicate square of 2 norms, | | | |0Indicate 0 norm;
(5c) selection operation
Two antibody X are selected from antibody population POPlAnd Xm, judge antibody XlWith antibody XmBetween whether meet (f1(Xl)≤f1
(Xm)&f2(Xl)<f2(Xm)) or (f1(Xl)<f1(Xm)&f2(Xl)≤f2(Xm)):
If it is, antibody XlFor non-dominant antibody, it being capable of dominated antibody Xm;
Otherwise, antibody XlIt is not non-dominated antibody, it is unable to dominated antibody Xm;
Wherein, f1(Xl) and f2(Xl) respectively indicate first antibody XlError project scalar functions and sparse item objective function
Value, f1(Xm) and f2(Xm) respectively indicate second antibody XmError project scalar functions and sparse item objective function value;
All non-dominant antibody is selected from antibody population POP, forms non-dominant antibody population POP1;
(5d) immune operation
To non-dominant antibody population POP1Clone operations are carried out, the antibody population POP after generating clone2;
To antibody population POP2Crossover operation is carried out, the antibody population POP after intersecting is generated3;
To antibody population POP3Mutation operation is carried out, the antibody population POP after generating variation4;
(5e) judges whether the number of iterations has reached maximum number of iterations T, if so, by antibody population POP4It is middle to resist as final
Body group P is executed step (6), and otherwise, return step (5b) carries out next iteration.
3. according to the method described in claim 2, the antibody population POP after wherein being cloned in step (5d)2, it is to non-dominant antibody
Group POP1The antibody population that the operation of 4 time clonings generates is carried out, is expressed as:
POP2={ POP1, POP1, POP1, POP1}。
4. according to the method described in claim 2, wherein generating the antibody population POP after intersecting in step (5d)3, realize as follows:
The first step, the random number that setting is two 0 to 1, i.e. the first random number α, the second random number β and third random number η;
First random number and given threshold 0.5 are compared by second step:
If α>0.5, then from the antibody population POP after clone2In randomly select two antibody Xr1And Xr2Uniform crossover is carried out, is generated
New antibody:Xr3, antibody Xr3In i-th of super-pixel point and the similitude of j-th of super-pixel point can be calculated as
If α≤0.5, from the antibody population POP after clone2In randomly select two antibody Xr1And Xr2Arithmetic crossover is carried out, is produced
Raw new antibody:Xr4=η Xr1+(1-η)Xr2;
Third step, repetitive operation second step n times generate N number of new antibodies, the antibody population POP after composition intersection3。
5. according to the method described in claim 2, wherein generating the antibody population POP after variation in step (5d)4, realize as follows:
Firstly, choosing the antibody population POP after intersecting3In any one antibody XlIf antibody XlIn i-th super-pixel point and
The similitude x of j-th of super-pixel pointijVariation criterion be:
The wherein random number that rand and γ is 0 to 1, h is neighbour order of j-th of the super-pixel o'clock about i-th of super-pixel point;
Then, to the antibody population POP after intersection3In N number of antibody successively carry out mutation operation according to variation criterion, generate N number of new
Antibody, the antibody population POP after Composition Variation4。
6. according to the method described in claim 1, wherein spectral clustering is carried out to super-pixel characteristic data set I in step (6), by such as
Lower step carries out:
(6a) is for the antibody X in non-dominant antibody population Pl, l=1 ..., N, calculating degree matrixIts
InWherein xijIndicate the similitude between i-th of super-pixel point and j-th of super-pixel point, n indicates super-pixel point
Number;
(6b) calculates Laplacian Matrix L=D-Xl;
(6c) carries out Eigenvalues Decomposition to Laplacian Matrix L according to formula Lu=λ u, obtains eigenvalue λ and feature vector u=
[u1, u2..., un];
(6d) chooses the corresponding feature vector of first h the smallest characteristic value and forms u '=[u1', u2' ..., uh'], to u ' normalizing
Change, the feature vector u " after being normalized;
(6e) divides u " using kmeans method, obtains the cluster result of super-pixel characteristic data set I:
V=(V1..., Vi..., Vc),
Wherein ViIndicate the super-pixel point set for belonging to the i-th class in cluster result, c indicates the class number of cluster.
7. according to the method described in claim 1, the NC that wherein step (7) calculates each cluster result estimates, by following public
Formula calculates:
Wherein, c indicates the class number of cluster,Weight between expression class,VhTable
Show the super-pixel point set of h class in cluster result,Indicate the super-pixel point set that h class is not belonging in cluster result, wij
Indicate the standard similitude between i-th of super-pixel point and j-th of super-pixel point,xij lIndicate antibody XlIn i-th
Similitude between a super-pixel point and j-th of super-pixel point, l=1 ..., N, N indicate the cluster result of non-dominant antibody tormation
Number, vol (Vh) indicate subset VhWeight,diI-th of element in degree of a representation matrix.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201810651506.7A CN108921853B (en) | 2018-06-22 | 2018-06-22 | Image segmentation method based on super-pixel and immune sparse spectral clustering |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201810651506.7A CN108921853B (en) | 2018-06-22 | 2018-06-22 | Image segmentation method based on super-pixel and immune sparse spectral clustering |
Publications (2)
Publication Number | Publication Date |
---|---|
CN108921853A true CN108921853A (en) | 2018-11-30 |
CN108921853B CN108921853B (en) | 2022-03-04 |
Family
ID=64422288
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201810651506.7A Active CN108921853B (en) | 2018-06-22 | 2018-06-22 | Image segmentation method based on super-pixel and immune sparse spectral clustering |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN108921853B (en) |
Cited By (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN109858529A (en) * | 2019-01-11 | 2019-06-07 | 广东工业大学 | A kind of image clustering method of scalability |
CN112396619A (en) * | 2019-08-15 | 2021-02-23 | 电子科技大学 | Small particle segmentation method based on semantic segmentation and internally complex composition |
CN112784894A (en) * | 2021-01-18 | 2021-05-11 | 西南石油大学 | Automatic labeling method for rock slice microscopic image |
CN113298821A (en) * | 2021-04-22 | 2021-08-24 | 江苏大学 | Hyperpixel matting method based on Nystrom spectral clustering |
Citations (16)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN101299243A (en) * | 2008-06-27 | 2008-11-05 | 西安电子科技大学 | Method of image segmentation based on immune spectrum clustering |
CN101667292A (en) * | 2009-09-25 | 2010-03-10 | 西安电子科技大学 | SAR image segmentation system and segmentation method based on immune clone and projection pursuit |
CN101673398A (en) * | 2009-10-16 | 2010-03-17 | 西安电子科技大学 | Method for splitting images based on clustering of immunity sparse spectrums |
CN101866489A (en) * | 2010-06-30 | 2010-10-20 | 西安电子科技大学 | Image dividing method based on immune multi-object clustering |
CN102184524A (en) * | 2011-04-13 | 2011-09-14 | 西安电子科技大学 | Neighbourhood learning meme image segmentation method based on standard cut |
CN102254326A (en) * | 2011-07-22 | 2011-11-23 | 西安电子科技大学 | Image segmentation method by using nucleus transmission |
US20120275702A1 (en) * | 2011-04-29 | 2012-11-01 | Cuneyt Oncel Tuzel | Method for Segmenting Images Using Superpixels and Entropy Rate Clustering |
CN102867307A (en) * | 2012-09-10 | 2013-01-09 | 西安电子科技大学 | SAR image segmentation method based on feature vector integration spectral clustering |
CN102968796A (en) * | 2012-11-30 | 2013-03-13 | 西安电子科技大学 | SAR (Synthetic Aperture Radar) image segmentation method based on sampling learning |
CN103310441A (en) * | 2013-05-16 | 2013-09-18 | 西安电子科技大学 | Texture image segmentation method based on immunity cloning and multitarget optimizing |
CN103366365A (en) * | 2013-06-18 | 2013-10-23 | 西安电子科技大学 | SAR image varying detecting method based on artificial immunity multi-target clustering |
CN104156943A (en) * | 2014-07-14 | 2014-11-19 | 西安电子科技大学 | Multi-target fuzzy cluster image variance detecting method based on non-control-neighborhood immune algorithm |
CN104217436A (en) * | 2014-09-16 | 2014-12-17 | 西安电子科技大学 | SAR image segmentation method based on multiple feature united sparse graph |
US20160358338A1 (en) * | 2015-06-05 | 2016-12-08 | Canon Kabushiki Kaisha | Image recognition apparatus, image recognition method, and recording medium |
CN106780639A (en) * | 2017-01-20 | 2017-05-31 | 中国海洋大学 | Hash coding method based on the sparse insertion of significant characteristics and extreme learning machine |
CN107784284A (en) * | 2017-10-24 | 2018-03-09 | 哈尔滨工业大学深圳研究生院 | Face identification method and system |
-
2018
- 2018-06-22 CN CN201810651506.7A patent/CN108921853B/en active Active
Patent Citations (16)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN101299243A (en) * | 2008-06-27 | 2008-11-05 | 西安电子科技大学 | Method of image segmentation based on immune spectrum clustering |
CN101667292A (en) * | 2009-09-25 | 2010-03-10 | 西安电子科技大学 | SAR image segmentation system and segmentation method based on immune clone and projection pursuit |
CN101673398A (en) * | 2009-10-16 | 2010-03-17 | 西安电子科技大学 | Method for splitting images based on clustering of immunity sparse spectrums |
CN101866489A (en) * | 2010-06-30 | 2010-10-20 | 西安电子科技大学 | Image dividing method based on immune multi-object clustering |
CN102184524A (en) * | 2011-04-13 | 2011-09-14 | 西安电子科技大学 | Neighbourhood learning meme image segmentation method based on standard cut |
US20120275702A1 (en) * | 2011-04-29 | 2012-11-01 | Cuneyt Oncel Tuzel | Method for Segmenting Images Using Superpixels and Entropy Rate Clustering |
CN102254326A (en) * | 2011-07-22 | 2011-11-23 | 西安电子科技大学 | Image segmentation method by using nucleus transmission |
CN102867307A (en) * | 2012-09-10 | 2013-01-09 | 西安电子科技大学 | SAR image segmentation method based on feature vector integration spectral clustering |
CN102968796A (en) * | 2012-11-30 | 2013-03-13 | 西安电子科技大学 | SAR (Synthetic Aperture Radar) image segmentation method based on sampling learning |
CN103310441A (en) * | 2013-05-16 | 2013-09-18 | 西安电子科技大学 | Texture image segmentation method based on immunity cloning and multitarget optimizing |
CN103366365A (en) * | 2013-06-18 | 2013-10-23 | 西安电子科技大学 | SAR image varying detecting method based on artificial immunity multi-target clustering |
CN104156943A (en) * | 2014-07-14 | 2014-11-19 | 西安电子科技大学 | Multi-target fuzzy cluster image variance detecting method based on non-control-neighborhood immune algorithm |
CN104217436A (en) * | 2014-09-16 | 2014-12-17 | 西安电子科技大学 | SAR image segmentation method based on multiple feature united sparse graph |
US20160358338A1 (en) * | 2015-06-05 | 2016-12-08 | Canon Kabushiki Kaisha | Image recognition apparatus, image recognition method, and recording medium |
CN106780639A (en) * | 2017-01-20 | 2017-05-31 | 中国海洋大学 | Hash coding method based on the sparse insertion of significant characteristics and extreme learning machine |
CN107784284A (en) * | 2017-10-24 | 2018-03-09 | 哈尔滨工业大学深圳研究生院 | Face identification method and system |
Non-Patent Citations (7)
Title |
---|
CONGCONG GONG等: "Improved multi-objective clustering algorithm using particle swarm optimization", 《PLOS ONE》 * |
ECKART ZITZLER等: "Multiobjective evolutionary algorithms: A comparative case study and he strength Pareto approach", 《IEEE TRANSACTIONS ON EVOLUTIONARY COMPUTATION》 * |
JUANJUAN LUO等: "A Sparse Spectral Clustering Framework via Multi-Objective Evolutionary Algorithm", 《IEEE TRANSACTIONS ON EVOLUTIONARY COMPUTATION》 * |
李小平等: "图像分割的改进稀疏子空间聚类方法", 《系统工程与电子技术》 * |
杨咚咚: "基于人工免疫系统的多目标优化与SAR图像分割", 《中国优秀博硕士学位论文全文数据库(博士) 信息科技辑》 * |
翟艳鹏等: "遗传算法优化归一化划分准则的图像分割", 《计算机工程与应用》 * |
邹小林: "融合SLIC的DCUT改进图像分割算法", 《新疆大学学报(自然科学版)》 * |
Cited By (6)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN109858529A (en) * | 2019-01-11 | 2019-06-07 | 广东工业大学 | A kind of image clustering method of scalability |
CN109858529B (en) * | 2019-01-11 | 2022-11-01 | 广东工业大学 | Scalable image clustering method |
CN112396619A (en) * | 2019-08-15 | 2021-02-23 | 电子科技大学 | Small particle segmentation method based on semantic segmentation and internally complex composition |
CN112784894A (en) * | 2021-01-18 | 2021-05-11 | 西南石油大学 | Automatic labeling method for rock slice microscopic image |
CN112784894B (en) * | 2021-01-18 | 2022-11-15 | 西南石油大学 | Automatic labeling method for rock slice microscopic image |
CN113298821A (en) * | 2021-04-22 | 2021-08-24 | 江苏大学 | Hyperpixel matting method based on Nystrom spectral clustering |
Also Published As
Publication number | Publication date |
---|---|
CN108921853B (en) | 2022-03-04 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
Caelli et al. | An eigenspace projection clustering method for inexact graph matching | |
Unnikrishnan et al. | Toward objective evaluation of image segmentation algorithms | |
Ghosh et al. | Comparative analysis of k-means and fuzzy c-means algorithms | |
CN104732545B (en) | The texture image segmenting method with quick spectral clustering is propagated with reference to sparse neighbour | |
CN108921853A (en) | Image partition method based on super-pixel and clustering of immunity sparse spectrums | |
CN110163258A (en) | A kind of zero sample learning method and system reassigning mechanism based on semantic attribute attention | |
Bespalov et al. | Scale-space representation of 3d models and topological matching | |
CN103593855B (en) | The image partition method of cluster is estimated based on particle group optimizing and space length | |
Hess et al. | The spectacl of nonconvex clustering: A spectral approach to density-based clustering | |
CN107203785A (en) | Multipath Gaussian kernel Fuzzy c-Means Clustering Algorithm | |
Tang et al. | One-step multiview subspace segmentation via joint skinny tensor learning and latent clustering | |
de Arruda et al. | A complex networks approach for data clustering | |
Chen et al. | LABIN: Balanced min cut for large-scale data | |
Wang et al. | Model-based clustering of censored data via mixtures of factor analyzers | |
EP1353295B1 (en) | Non-linear data mapping and dimensionality reduction system | |
CN106022359A (en) | Fuzzy entropy space clustering analysis method based on orderly information entropy | |
CN108256569B (en) | Object identification method under complex background and used computer technology | |
CN105844299A (en) | Image classification method based on bag of words | |
CN113469270B (en) | Semi-supervised intuitive clustering method based on decomposition multi-target differential evolution superpixel | |
Maggu et al. | Transformed locally linear manifold clustering | |
CN109766910A (en) | Figure clustering method based on similarity transmitting | |
CN115310491A (en) | Class-imbalance magnetic resonance whole brain data classification method based on deep learning | |
Bhattacharyya et al. | Spectral clustering and block models: A review and a new algorithm | |
Xie et al. | Non-convex clustering via proximal alternating linearized minimization method | |
CN112884023A (en) | Community discovery method for system evolution transplantation subareas |
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 |