CN110211126B - Image segmentation method based on intuitive fuzzy C-means clustering - Google Patents
Image segmentation method based on intuitive fuzzy C-means clustering Download PDFInfo
- Publication number
- CN110211126B CN110211126B CN201910507275.7A CN201910507275A CN110211126B CN 110211126 B CN110211126 B CN 110211126B CN 201910507275 A CN201910507275 A CN 201910507275A CN 110211126 B CN110211126 B CN 110211126B
- Authority
- CN
- China
- Prior art keywords
- image
- value
- pixel
- membership
- gray
- 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.)
- Active
Links
- 238000000034 method Methods 0.000 title claims abstract description 97
- 238000003709 image segmentation Methods 0.000 title claims abstract description 21
- 239000011159 matrix material Substances 0.000 claims abstract description 42
- 238000011156 evaluation Methods 0.000 claims abstract description 10
- 238000001514 detection method Methods 0.000 claims abstract description 6
- 230000011218 segmentation Effects 0.000 claims description 40
- 238000010276 construction Methods 0.000 claims description 5
- 238000004364 calculation method Methods 0.000 claims description 4
- 229910010888 LiIn Inorganic materials 0.000 claims description 3
- 230000017105 transposition Effects 0.000 claims description 3
- 238000012937 correction Methods 0.000 claims description 2
- 238000007781 pre-processing Methods 0.000 abstract description 2
- 238000012216 screening Methods 0.000 abstract 2
- 238000004422 calculation algorithm Methods 0.000 description 23
- 239000006002 Pepper Substances 0.000 description 19
- 230000000694 effects Effects 0.000 description 11
- 238000004088 simulation Methods 0.000 description 7
- 238000005516 engineering process Methods 0.000 description 4
- 235000002566 Capsicum Nutrition 0.000 description 3
- 241000722363 Piper Species 0.000 description 3
- 235000016761 Piper aduncum Nutrition 0.000 description 3
- 235000017804 Piper guineense Nutrition 0.000 description 3
- 235000008184 Piper nigrum Nutrition 0.000 description 3
- 230000007547 defect Effects 0.000 description 3
- 238000012545 processing Methods 0.000 description 3
- 150000003839 salts Chemical class 0.000 description 3
- 238000001914 filtration Methods 0.000 description 2
- 238000003909 pattern recognition Methods 0.000 description 2
- 230000035945 sensitivity Effects 0.000 description 2
- 230000001629 suppression Effects 0.000 description 2
- DOLVFFGLNYMIPV-UHFFFAOYSA-N 2,4-ditert-butyl-6-[3-[(3,5-ditert-butyl-2-hydroxyphenyl)methylideneamino]propyliminomethyl]phenol Chemical compound CC(C)(C)C1=CC(C(C)(C)C)=CC(C=NCCCN=CC=2C(=C(C=C(C=2)C(C)(C)C)C(C)(C)C)O)=C1O DOLVFFGLNYMIPV-UHFFFAOYSA-N 0.000 description 1
- 238000004458 analytical method Methods 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 210000004556 brain Anatomy 0.000 description 1
- 238000002372 labelling Methods 0.000 description 1
- 238000003672 processing method Methods 0.000 description 1
- 238000011160 research Methods 0.000 description 1
- 230000003595 spectral effect Effects 0.000 description 1
- 230000000007 visual effect Effects 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/23—Clustering techniques
- G06F18/232—Non-hierarchical techniques
- G06F18/2321—Non-hierarchical techniques using statistics or function optimisation, e.g. modelling of probability density functions
- G06F18/23213—Non-hierarchical techniques using statistics or function optimisation, e.g. modelling of probability density functions with fixed number of clusters, e.g. K-means clustering
-
- 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
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/10—Image acquisition modality
- G06T2207/10024—Color image
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Data Mining & Analysis (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Probability & Statistics with Applications (AREA)
- Life Sciences & Earth Sciences (AREA)
- Artificial Intelligence (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Bioinformatics & Computational Biology (AREA)
- Evolutionary Biology (AREA)
- Evolutionary Computation (AREA)
- General Engineering & Computer Science (AREA)
- Image Analysis (AREA)
Abstract
The invention discloses an image segmentation method based on intuitive fuzzy C-means clustering. The method mainly solves the problems that the image segmentation is sensitive to noise and easy to fall into local optimum, and the clustering number needs to be set, and has the scheme that: constructing an intuitive fuzzy set robust to noise using a switch mean based strategyPerforming curve fitting on the gray value distribution of the image, screening all peak points of a fitting curve as a cluster center initial value range set, and counting the number of the maximum cluster centers; in thatOn the basis, the position information and the gray level information of the pixels are utilized to construct a linear weighting function coefficient in an intuitive fuzzy target function, and a membership degree matrix U is obtained; and evaluating the index evaluation U according to the grading distance index to obtain an optimal membership matrix, and screening the misclassified pixels by using an error detection strategy to carry out correct classification. The invention enhances the robustness to noise, can self-adaptively determine the image clustering number, and can be used for image recognition and computer vision preprocessing.
Description
Technical Field
The invention belongs to the field of digital image processing, and particularly relates to an image segmentation method which can be used for image recognition and computer vision preprocessing.
Background
Since the seventies of the last century, many scholars continuously pay attention to image segmentation, the image segmentation technology has become a basic technology in many fields, the image segmentation technology cannot be lacked as long as the problem of extracting the content in the image is concerned, and the quality and effect of the image segmentation work directly or indirectly influence the subsequent image engineering. The existing image segmentation methods are mainly based on threshold value image segmentation methods and clustering image segmentation methods, and the existing image segmentation techniques are a method of using statistical information segmentation, and usually extracting appropriate threshold values to separate out interested target parts in images. However, it is only suitable for images with obvious gray scale difference, and in addition, noise and other interference elements also cause difficulty in accurate image segmentation. Until now, a unified theoretical system is not formed to be applied to different fields, and image segmentation still is a popular research field.
The clustering-based image segmentation method divides images into a plurality of categories according to the similarity between pixel points in the images, has the characteristics of good segmentation effect, short time consumption, stable method operation and the like, and becomes one of the commonly applied methods in recent years. Common clustering segmentation methods include: k-means, fuzzy C-means clustering, hierarchical clustering, and spectral clustering), etc., with fuzzy C-means clustering being one of the most widely used methods. The fuzzy C-means method was first proposed by dunn, see: dunn, a Fuzzy image processing method related to an iterative self-organization analysis algorithm and application thereof to detection of Compact and easily separable Clusters, report on the control theory 1973,3(3):32-57 (J.C. Dunn.A Fuzzy relationship of the ISODATA Process and Its Use in Detecting Compact Well-Separated Clusters J. Journal of Cybernetics 1973,3(3):32-57.), the method is sensitive to noise in an image due to no consideration of any spatial information in the image, so that the segmentation of a noisy image cannot achieve an ideal segmentation effect. Many scholars later published improved methods of combining image space neighborhood information. Ehamald et al add a spatial neighborhood information item to an objective function of a fuzzy C-means, and propose an FCM _ S algorithm, which improves robustness against noise, but has high computational complexity, and in order to reduce the computational complexity of the FCM _ S algorithm, cheng and zhao introduce neighborhood information to the objective function of the algorithm through mean filtering and median filtering, and propose FCM _ S1 and FCM _ S2 algorithms, see: volume 34, 1907-; the FCM _ S1 algorithm in the two algorithms has a good Gaussian noise processing effect but poor robustness to salt and pepper noise, and the two FCM _ S2 algorithms have a good processing effect on salt and pepper noise but poor robustness to Gaussian noise, so that the two algorithms cannot achieve robust universality to various types of noise. Chua-Weiling et al constructs a linear weighted sum image by combining spatial information and gray information of the image, and proposes a fast generation FCM algorithm, see: a Fast Robust Fuzzy C-means Clustering algorithm for Image Segmentation, which introduces Local Information, is characterized by Pattern recognition, volume 40,825-838,2007 (W.Cai, S.Chen, and D.Zhang, "Fast and Robust fusing Local Information for Image Segmentation," Pattern recognition, vol.40, No.3, pp.825-838, Mar.2007), which is Robust to Gaussian noise as well as to salt-pepper noise; however, the algorithms do not consider more fuzziness of data, and charles et al further find that more fuzziness of data can be considered by using an intuitive fuzzy set theory, the classification of the data is more accurate, and a fuzzy clustering method based on intuitive fuzzy data is provided, which is as follows: application of soft computing (T, Chaira. "A novel intuitionistic fuzzy C means clustering and its application to medical images." application. Soft. Compat. 11(2): 1711-; because the fuzzy clustering method based on the intuitive fuzzy data is sensitive to noise, Welman et al further introduces local spatial information into the intuitive fuzzy C-means algorithm, see: welmar, Aglaval, Salan. Improved intuitive Fuzzy C-means Algorithm for Brain Image Segmentation Incorporating Local spatial Information application Soft computing 543-.
Although the improved method optimizes the performance of the fuzzy clustering algorithm on the anti-noise performance to a certain extent, the improved method still has the defects of not strong enough robustness on noise, very sensitivity to the initial value of a clustering center, incapability of self-adaptively analyzing the image clustering number and the like.
Disclosure of Invention
The invention aims to provide an image segmentation method based on intuitive fuzzy C-means clustering to overcome the defects of the prior art, so as to enhance the robustness to noise, improve the segmentation accuracy, reduce the sensitivity to the initial value of a clustering center and determine the number of image clusters in a self-adaptive manner.
In order to achieve the above object, the technique of the present invention comprises the steps of:
(1) inputting an image to be segmented, and converting the image to be segmented into a gray image if the image to be segmented is a color image;
(2) initializing parameters: setting the maximum iteration number T as 100 and the stop threshold epsilon as 10-5The fuzzy weighting index m is 2, the radius omega of the neighborhood window is 3, the initial iteration time t is 1, and the default initial value Y of the clustering number is 2;
(3) constructing an intuitive fuzzy set robust to noise using a switch-mean based strategyWherein 1 ≦ i ≦ n, which represents the total number of pixels;
(4) performing curve fitting by using a least square method according to the gray value distribution of the input image, selecting the gray level corresponding to the peak value point of a fitting curve as an initialized clustering center value, and setting the range of the estimated clustering number as {2, …, YmaxThe initial value range set C of the cluster center isWherein, cYDenotes an initial value of the Yth cluster center, Y is 1. ltoreq. Y.ltoreq.Ymax,YmaxMaximum value representing the number of clusters, 2. ltoreq. YmaxN is less than or equal to n, and T represents transposition;
(5) iterative computation of linear weighting function coefficients HirIntuitive fuzzy distanceAnd an intuitive fuzzy objective function J:
(5a) construction of linear weighting function coefficient H using position information and gradation information of pixelir:
Wherein (a)i,bi) And (a)r,br) Spatial coordinate values, x, representing the central pixel i and the neighborhood pixel r, respectivelyiAnd xrRepresenting the gray values, u, of the central pixel i and the neighborhood pixel r, respectivelykiDenotes xiFor the k-th cluster center ckThe membership degree of k is more than or equal to 1 and less than or equal to Y, r is more than or equal to 1 and less than or equal to omega2;
(5b) Calculating pixel x under intuitive fuzzy setiTo the center of the cluster ckIntuitive fuzzy distance ofWherein,andrespectively represent pixels xiAnd cluster center ckThe intuitive fuzzy set of (1);
(5c) combining the linear weighting function coefficient H obtained in (5a)irThe intuitive blur distance obtained in (5b)Calculating an intuitive fuzzy objective function J and solving a membership function u in the intuitive fuzzy objective function JkiAnd a cluster center ckIntuition fuzzy set of
(6) Judging iteration stop condition | | C(t+1)-C(t)If | | < epsilon or the number of operation iterations T > T holds:
if not, making t equal to t +1, and returning to the step (5);
if yes, outputting a membership matrix U and judging that Y is less than YmaxIf yes, making Y equal to Y +1, taking the first Y numerical values of the initial value range set C of the clustering center, and making the initial value set of the clustering center asMaking the initial iteration number t equal to 1, returning to (5), if Y < YmaxIf not, performing (7), wherein,in the form of a set of cluster center values,represents the t-th set of intuitive ambiguities for the qth cluster-center iteration,denotes an initial value of the pth cluster center, 1. ltoreq. Q.ltoreq.Y, 1. ltoreq. P.ltoreq.Y, U ═ U ≦ Yki};
(7) Output (Y) from (6)max-1) membership matrices U and evaluating them using a hierarchical distance index evaluation index GD; setting the membership matrix with the maximum GD value as UmaxWill U ismaxCorresponding cluster number YfinAs the final cluster number of the input image, where 1. ltoreq. Yfin≤Ymax;
(8) Membership degree matrix U of (7) outputmaxAnd (3) carrying out pixel classification, carrying out error detection classification, and outputting a segmentation image:
(8a) according to the maximum membership degree principle and the final clustering number YfinMembership matrix U for the output of (7)maxIs classified, and a label matrix U is outputLWherein, ULIncluding only the class values {1,2, …, YW,…,Yfin},1≤YW≤Yfin;
(8b) Let the label class value of the center pixel i be LiIn the neighborhood of 5 by 5 of the central pixel i, the label type value L is judgediNumber of pixels nLWhether or not n is satisfiedL≤25/(Yfin+1) condition:
if yes, the central pixel i is judged to be a wrong pixel, and the label class value L of the wrong pixel is divided intoiCorrecting the label category value to be the label category value with the maximum number of the same label category values in the neighborhood, and executing (8 c);
if not, judging that the central pixel i is a correct distribution pixel, and executing (8 c);
(8c) judging tag matrix ULIf so, outputting a final label matrix UFExecuting (8d), if not, returning to (8b) to continue traversing;
(8d) for the final label matrix UFThe middle label class value is YWThe pixels of (2) are assigned different corresponding gray values xWAnd the matrix given different gray values is output as a divided image.
Compared with the prior art, the invention has the following beneficial technical effects:
firstly, an intuitive fuzzy set robust to noise is constructed by using a switch mean value-based strategy, a linear weighting function coefficient is constructed by using position information and gray information of pixels and is introduced into an intuitive fuzzy target function, and the problem that a clustering algorithm is sensitive to noise is solved.
Secondly, the invention uses the least square method to carry out curve fitting on the gray value distribution of the image, selects the gray levels corresponding to all peak points of the fitting curve as the initial value range of the clustering center, evaluates the membership matrix according to the grading distance index evaluation index, selects the clustering number suitable for the input image, realizes the self-adaptive determination of the image clustering number, and overcomes the defect that the traditional clustering algorithm is easy to fall into the local optimum.
Thirdly, the invention screens the misclassified pixels by using an error detection strategy and correctly distributes the misclassified elements, thereby improving the segmentation accuracy and ensuring the segmentation effect to be more ideal.
Drawings
FIG. 1 is a flow chart of an implementation of the present invention;
FIG. 2 is a comparison graph of the results of a simulation segmentation of a #3096 image in a Berkeley image database using the present invention and a prior art method;
FIG. 3 is a comparison of results of simulated segmentation of #238011 images in a Berkeley image database using the present invention and a prior art method.
Detailed Description
The embodiments and effects of the present invention are further described below with reference to the accompanying drawings:
referring to fig. 1, the implementation steps of the present invention are as follows:
step 1: an image to be segmented is input.
And inputting an image to be segmented, and converting the image to be segmented into a gray image if the image to be segmented is a color image.
Step 2: setting initial parameter values.
Setting the maximum iteration number T as 100 and the stop threshold epsilon as 10-5The fuzzy weighting index m is 2, the neighborhood window radius ω is 3, the initial iteration number t is 1, and the default initial value Y of the cluster number is 2.
The prior art method for constructing the intuitive fuzzy set includes an IFCM algorithm and an IIFCM algorithm, which both use Yager operator to construct the intuitive fuzzy set.
The example uses a switch mean strategy to correct the pixel gray value of the input image, and constructs an intuitive fuzzy set robust to noise through a Yager operator, wherein the switch mean strategy is to correct the gray value according to the pixel gray value distribution condition in the neighborhood of the pixel i, and the implementation is as follows:
3.1) first, it is determined whether the pixel gray values in the 3 by 3 neighborhood of the center pixel i are not all 0 or 255:
if yes, the gray value of the pixel in the neighborhood is represented to have other gray values except 0 or 255, the pixels with the gray values of 0 and 255 in the neighborhood are taken out, then the gray average value of the rest pixels in the neighborhood is taken and set as the correction value of the central pixel, and then 3.2) is executed;
if not, directly executing 3.2) without modifying the gray value of the central pixel;
3.2) judging whether all pixels in the input image traverse: if yes, outputting the image with the corrected gray value, executing 3.3), and if not, returning to 3.1) to continue traversing;
3.3) calculating the pixel gray value of the corrected image output in 3.2) as an intuitive fuzzy set
3.3.1) setting μ (x)i) Representing pixel grey value xiDegree of membership of, v (x)i) And pi (x)i) Are each xiDegree of non-membership and hesitation, μ (x), using Yager operator constructioni)、v(xi) And pi (x)i) The calculation formulas of (A) are respectively as follows:
π(xi)=1-μ(xi)-v(xi),
wherein 0 is less than or equal to mu (x)i)≤1,0≤v(xi)≤1,0≤μ(xi)+v(xi) 1, and alpha is 0.8 to represent a Yager operator coefficient;
3.3.2) calculating the resulting μ (x) according to 3.3.1)i)、v(xi) And pi (x)i) Structure intuition fuzzy set
And 4, step 4: and aiming at the gray value distribution of the input image, performing curve fitting by using a least square method, selecting the gray level corresponding to the peak value point of a fitting curve as an initialized clustering center value, and estimating the maximum clustering number and the initial value range of the clustering center.
The existing technology for selecting the initial value of the clustering center directly selects the peak point of the image gray level histogram as the initial clustering center.
In this example, a least square method is used to perform curve fitting on the image gray value distribution, and then a peak point is selected on the smooth fitting curve, which is implemented as follows:
4.1) setting the gray level image to 256 gray levels, wherein the gray level range is [0,255], and drawing a gray level histogram of the gray level image;
4.2) counting the number of pixels corresponding to 256 gray levels on the gray level histogram, setting the number as gray level distribution frequency f (l), and fitting a curve to f (l) by using a seven-order least square method to fit gray level data points into a gray level curve, wherein l is more than or equal to 0 and less than or equal to 255;
4.3) solving the first derivative f '(l) of f (l), and comparing f' (l) with the first derivatives f '(l-1) and f' (l +1) of the front and back gray levels of f (l);
4.4) judging whether the gray level l is under the condition that l is 0 or 255:
if l is 0 or 255 and l is 0, then judging whether f' (l) < 0 is true, if true, outputting the gray level l is 0 as the peak point, and executing 4.5 again);
if l is 0 or 255 and l is 255, then determining whether f' (l) > 0 is true, if true, outputting the gray level l is 255 as the peak point, and then executing 4.5);
if not under the condition that l is 0 or 255, namely, under the condition that 0 < l < 255, judging whether f ' (l) is 0 and f ' (l-1) > 0 and f ' (l +1) < 0 are true, if so, outputting the gray level l as a peak point, and executing 4.5 again);
4.5) judging whether 256 gray levels are completely traversed, if so, outputting peak points, executing 4.6), and if not, returning to 4.4) and continuing traversing;
4.6) sorting the peak points output by 4.5) according to the sequence of the gray level distribution frequency values from large to small, and counting the number of the peak points as YmaxAssuming that the number of clusters ranges from 2 to YmaxAnd counting the gray levels l corresponding to the peak points ordered from large to small to obtain the gray level value range ofThe corresponding value of the gray level is counted as the initial value of the cluster center, that isSetting the initial value range set C of the clustering center asWherein:
lYexpressing the gray level corresponding to the Y-th peak point gray level distribution frequency value in the descending order of the peak point gray level distribution frequency value, cYRepresents an initial value of the Yth cluster center, Y is not less than 1 and not more than Ymax,YmaxMaximum value representing the number of clusters, 2. ltoreq.YmaxN is less than or equal to n, and T represents transposition.
And 5: iterative computation of a linear weighting function coefficient HirIntuitive fuzzy distanceAnd an intuitive blur objective function J.
The existing FCM _ S1 algorithm and FCM _ S2 algorithm introduce local spatial information of an image into its objective function, wherein the FCM _ S1 algorithm considers the mean filter characteristic of an image and the FCM _ S2 algorithm considers the median filter characteristic of an image.
In the embodiment, the linear weighting function coefficient H is constructed by using the position information and the gray information of the pixelirThen H containing local spatial information of the imageirIntroducing an intuitive fuzzy objective function J, which is realized as follows:
5.1) construction of Linear weighting function coefficients H Using position information and Gray-level information of pixelsir:
Wherein (a)i,bi) And (a)r,br) Spatial coordinate values, x, representing the central pixel i and the neighborhood pixel r, respectivelyiAnd xrRepresenting the gray values, u, of the central pixel i and the neighborhood pixel r, respectivelykiDenotes xiFor the k-th cluster center ckThe membership degree of k is more than or equal to 1 and less than or equal to Y, r is more than or equal to 1 and less than or equal to omega2;
5.2) calculating the pixel gray value x under the intuitive fuzzy setiTo the center of the cluster ckIntuitive fuzzy distance of
5.2.1) setting μ (c)k) Representing the center of the cluster ckDegree of membership of v (c)k) And pi (c)k) Respectively represent the cluster centers ckDegree of non-membership and hesitation, μ (c), using Yager operator constructionk)、v(ck) And pi (c)k) The calculation formulas of (A) are respectively as follows:
π(ck)=1-μ(ck)-v(ck),
wherein,0≤μ(ck)≤1,0≤v(ck)≤1,0≤μ(ck)+v(ck)≤1;
Wherein, mu (x)i)、v(xi) And pi (x)i) Respectively representing pixel grey values xiDegree of membership, degree of non-membership, and degree of hesitation;
5.3) combining the linear weighting function coefficients H obtained in (5.1)irThe intuitive fuzzy distance obtained from (5.2)Calculating an intuitive fuzzy objective function J:
wherein u iskiIs a pixel xiTo the clustering center ckM represents a fuzzy weighting index;
5.4) solving an intuitive fuzzy target function J by using a Lagrange multiplier method and obtaining a membership function ukiAnd a cluster center ckIntuition fuzzy set of
5.4.1) introducing Lagrange multiplier lambda in the intuitive fuzzy target function JiConstructing an objective function L with a Lagrange multiplier:
5.4.2) separately calculating the objective function L with Lagrange multiplier versus the membership function ukiCluster center ckIntuition fuzzy set ofAnd lagrange multiplier λiPartial derivative of (2)And
5.4.3) in partial derivativesUnder the condition (1), the membership function u is obtained as followskiAnd cluster center ckIntuition fuzzy set of
Wherein k is 1,2, …, Y, i is 1,2, …, n, r is 1,2, …, ω2。
Step 6: and judging whether the iteration stop condition is met.
If the iteration stop condition C is satisfied(t+1)-C(t)If | is less than epsilon or the operation iteration times T is more than T, outputting a membership degree matrix U and executing 7), wherein,is a set of cluster center values that are,representing the visual fuzzy set of the t iteration of the Q clustering center;
if the iteration stop condition is not satisfied, let t be t +1, return 5.1).
And 7: and judging whether the current clustering number Y meets the condition or not.
If the current clustering number Y is less than the maximum clustering number YmaxI.e. Y < YmaxIf Y is Y +1, then taking the first Y values of the cluster center initial value range set C, outputting the values to the cluster center initial value set, and making the cluster center initial value set C(1)Is composed ofLet the initial number of iterations t be 1, return 5.1), where,representing the initial value of the P-th clustering center, wherein P is more than or equal to 1 and less than or equal to Y;
if Y < Y is not satisfiedmaxThen (8) is executed.
And 8: and evaluating the membership matrix by using the grading distance index evaluation index GD.
The existing evaluation index for the membership matrix has a division coefficient VpcAnd dividing entropy VpeAll of which reflect the degree of ambiguity of the membership degree matrix, where VpcThe larger the numerical value, the smaller the ambiguity of the membership degree matrix, VpeThe smaller the value, the more accurate the pixel classification.
The membership matrix is evaluated using the hierarchical distance index evaluation index GD in this example, i.e. input (Y) from 7max-1) membership matrices U which are evaluated using a hierarchical distance index evaluation index GD, outputting a membership matrix with a maximum GD value, and setting the membership matrix with the maximum GD value as UmaxWill U ismaxCorresponding cluster number YfinAs the final cluster number of the input image, wherein: the formula of the grading distance index evaluation index GD is as follows:
wherein u isi1Representing the maximum membership value, u, of the ith pixeli2Represents the second largest membership value of the ith pixel, and is more than or equal to 1 and less than or equal to Yfin≤Ymax。
And step 9: to degree of membership matrix UmaxPixel classification is performed.
According to the maximum membership principle and the final clustering number YfinTo the membership matrix U of 8) outputmaxIs classified, and a label matrix U is outputLWherein the maximum membership criterion is based on Vk=argk{max(uki) Divide the pixel i into the category V with the largest degree of membershipk,ULIncluding only the class values {1,2, …, YW,…,Yfin},1≤YW≤Yfin。
Step 10: for tag matrix ULAnd performing error detection classification.
10.1) setting the tag matrix U of 9) outputLHas a label class value of LiIn the neighborhood of 5 by 5 of the central pixel i, the label type value L is judgediNumber of pixels nLWhether or not n is satisfiedL≤25(Yfin+1) condition:
if yes, the central pixel i is judged to be a wrong pixel, and the label class value L of the wrong pixel is divided intoiCorrecting the label category value to be the label category value with the maximum number of the same label category values in the neighborhood, and executing 10.2);
if not, judging that the central pixel i is a correctly distributed pixel, and executing 10.2);
10.2) judging the tag matrix ULIf the pixels in the image are all traversed, outputting a final label matrix UFExecution 11), if not, return to 10.1) continue traversal.
Step 11: for the final label matrix UFAnd assigning and outputting the image.
For the final label matrix U output in 10.2)FThe middle label class value is YWThe pixels of (2) are assigned different corresponding gray values xWAnd outputting the matrix with different gray values as a segmentation image, and finally labeling the matrix UFIncluding only the class values {1,2, …, YW,…,YfinThe assignment formula is as follows:
xW=[255/YW],
wherein [. cndot. ] represents an internal integer.
The technical effects of the invention are further explained by combining simulation experiments;
1. simulation conditions are as follows:
the simulation experiment is carried out in the software environment of computer Inter (R) core (TM) i5-3210M 2.50GHZ CPU, 8G memory and MATLAB R2014 a.
2. Simulation content:
2(a) is an original image of 3096 image;
2(b) is a standard segmentation map of the 3096 image;
2(c) is a gaussian noisy image of 3096 image, with a noise intensity of 0.006;
2(d) is the result of segmenting the gaussian noisy image of 3096 images by the existing FCM method;
2(e) is the result of segmentation of the gaussian noisy image of 3096 image using the existing FCM _ S1 method;
2(f) is the result of segmentation of the gaussian noisy image of 3096 image using the existing FCM _ S2 method;
2(g) is a result of segmenting the gaussian noisy image of 3096 image by the conventional FGFCM method;
2(h) is a result of segmenting the gaussian noisy image of 3096 image by the existing IFCM method;
2(i) is the result of segmentation of the gaussian noisy image of 3096 images using the existing IIFCM method;
2(j) is the result of the segmentation of the Gaussian noisy image of 3096 image using the present invention;
2(k) is a noisy image of salt and pepper of 3096 image, with a noise intensity of 0.01;
2(l) is the segmentation result of the salt-pepper noisy image of 3096 image by the existing FCM method;
2(m) is the segmentation result of the salt-pepper noisy image of 3096 image using the existing FCM _ S1 method;
2(n) is the segmentation result of the salt-pepper noisy image of 3096 image using the existing FCM _ S2 method;
2(o) is the segmentation result of the salt-and-pepper noisy image of 3096 image by the existing FGFCM method;
2(p) is the segmentation result of the salt-pepper noisy image of 3096 image by the existing IFCM method;
2(q) is the segmentation result of the salt-pepper noisy image of 3096 image by the existing IIFCM method;
2(r) is the segmentation result of the salt-pepper noisy image of 3096 image by the invention;
as can be seen from FIG. 2, the background and the target can be separated on the basis of noise suppression, the number of image clusters can be determined adaptively, and the method is insensitive to the initial value of the cluster center, so that the segmentation effect of the method on noisy images is superior to that of the existing FCM method, FCM _ S1 method, FCM _ S2 method, FGFCM method, IFCM method and IIFCM method.
Simulation 2, selecting an image with the number of 238011 in the Berkeley image database, and segmenting the image by using the method of the present invention and the existing FCM method, FCM _ S1 method, FCM _ S2 method, FGFCM method, IFCM method and IIFCM method respectively, wherein the result is shown in fig. 3, wherein:
reference numeral 3(a) denotes an original drawing of an 238011 image;
3(b) is a standard segmentation map of the 238011 image;
3(c) is a gaussian noisy image of 238011 images with a noise intensity of 0.006;
3(d) is the result of segmentation of the gaussian noisy image of 238011 image using the existing FCM method;
3(e) is the result of segmentation of the gaussian noisy image of 238011 images using the existing FCM _ S1 method;
3(f) is the result of segmentation of the gaussian noisy image of 238011 images using the existing FCM _ S2 method;
3(g) is a result of segmenting 238011 gaussian noisy images using the existing FGFCM method;
3(h) is a result of segmenting a gaussian noisy image of the 238011 image by the existing IFCM method;
3(i) is the result of segmentation of the gaussian noisy image of 238011 images using the existing IIFCM method;
3(j) is the result of the segmentation of the gaussian noisy image of 238011 images using the present invention;
3(k) is a salt-pepper noisy image of 238011 images, with a noise intensity of 0.01;
3(l) is the segmentation result of the salt-pepper noisy image of 238011 image by the existing FCM method;
3(m) is the segmentation result of the salt-pepper noisy image of 238011 image by the existing FCM _ S1 method;
3(n) is the segmentation result of the salt-and-pepper noisy image of 238011 image by the existing FCM _ S2 method;
3(o) is the segmentation result of the salt-pepper noisy image of 238011 image by the existing FGFCM method;
3(p) is the segmentation result of the salt-pepper noisy image of 238011 image by the existing IFCM method;
3(q) is the segmentation result of the salt-pepper noisy image of 238011 image by the existing IIFCM method;
3(r) is the segmentation result of the salt-pepper noisy image of 238011 image by the present invention;
as can be seen from FIG. 3, the invention can separate the background and the target on the basis of noise suppression, the number of image clusters can be determined adaptively, the method is insensitive to the initial value of the cluster center, and the segmentation effect on the noisy images is better than that of the existing FCM method, FCM _ S1 method, FCM _ S2 method, FGFCM method, IFCM method and IIFCM method.
Claims (7)
1. The image segmentation method based on the intuitive fuzzy C-means clustering is characterized by comprising the following steps:
(1) inputting an image to be segmented, and converting the image to be segmented into a gray image if the image to be segmented is a color image;
(2) initializing parameters: setting the maximum iteration number T as 100 and the stop threshold epsilon as 10-5The fuzzy weighting index m is 2, the radius omega of the neighborhood window is 3, the initial iteration time t is 1, and the default initial value Y of the clustering number is 2;
(3) constructing an intuitive fuzzy set robust to noise using a switch mean based strategyWherein i is not less than 1 and not more than n, and n represents the total number of pixels; the implementation is as follows:
(3a) first, it is determined whether the pixel gradation values in the 3 by 3 neighborhood of the ith pixel are not all 0 or 255:
if yes, taking out the pixels with the gray values of 0 and 255, then taking the mean value of the gray values of the rest pixels, setting the mean value as the correction value of the central pixel, and executing (3 b);
if not, not modifying the gray value of the central pixel, and executing (3 b);
(3b) judging whether all pixels in the input image traverse: if yes, outputting the corrected image, executing (3c), and if not, returning to (3a) to continue traversing;
(3c1) Let u (x)i) Representing pixel grey value xiDegree of membership of, v (x)i) And pi (x)i) Are each xiThe non-membership degree and the hesitation degree constructed by using the Yager operator respectively have the following calculation formulas:
π(xi)=1-μ(xi)-v(xi),
wherein, 0 is less than or equal to mu (x)i)≤1,0≤v(xi)≤1,0≤μ(xi)+v(xi) 1, and alpha is 0.8 to represent a Yager operator coefficient;
(4) Performing curve fitting by using a least square method according to the gray value distribution of the input image, selecting the gray level corresponding to the peak value point of a fitting curve as an initialized clustering center value, and setting the range of the estimated clustering number as {2, …, YmaxThe initial value range set C of the cluster center isWherein, cYDenotes an initial value of the Yth cluster center, Y is 1. ltoreq. Y.ltoreq.Ymax,YmaxMaximum value representing the number of clusters, 2. ltoreq.YmaxN is less than or equal to n, and T represents transposition;
(5) iterative computation of linear weighting function coefficients HirIntuitive fuzzy distanceAnd an intuitive fuzzy objective function J:
(5a) construction of linear weighting function coefficient H using position information and gradation information of pixelir:
Wherein (a)i,bi) And (a)r,br) Individual watchSpatial coordinate values, x, of a center pixel i and a neighborhood pixel riAnd xrRepresenting the gray values, u, of the central pixel i and the neighborhood pixels r, respectivelykiDenotes xiFor the k-th cluster center ckThe membership degree of k is more than or equal to 1 and less than or equal to Y, r is more than or equal to 1 and less than or equal to omega2;
(5b) Calculating pixel x under intuitive fuzzy setiTo the center of the cluster ckIntuitive fuzzy distance ofWherein,andrespectively represent pixels xiAnd cluster center ckAn intuitive fuzzy set of;
(5c) combining the linear weighting function coefficient H obtained in (5a)irThe intuitive blur distance obtained in (5b)Calculating an intuitive fuzzy objective function J and solving a membership function u in the intuitive fuzzy objective function JkiAnd a cluster center ckIntuition fuzzy set of
(6) Judging iteration stop condition | | C(t+1)-C(t)If | | < epsilon or the number of operation iterations T > T holds:
if not, making t equal to t +1, and returning to the step (5);
if yes, outputting a membership matrix U and judging that Y is less than YmaxIf yes, making Y equal to Y +1, taking the first Y numerical values of the initial value range set C of the clustering center, and making the initial value set of the clustering center asMaking the initial iteration number t equal to 1, returning to (5), if Y < YmaxIf not, performing (7), wherein,in the form of a set of cluster center values,represents the set of intuitive ambiguities for the qth cluster-center iteration,denotes an initial value of the pth cluster center, 1. ltoreq. Q.ltoreq.Y, 1. ltoreq. P.ltoreq.Y, U ═ U ≦ Yki};
(7) Output (Y) from (6)max-1) membership matrices U and evaluating them using a hierarchical distance index evaluation index GD; setting the membership matrix with the maximum GD value as UmaxWill U ismaxCorresponding cluster number YfinAs the final cluster number of the input image, wherein 1 ≦ Yfin≤Ymax(ii) a The formula of the graded distance index evaluation index GD is expressed as follows:
wherein u isi1Represents the maximum membership value, u, of the ith pixeli2Represents the second largest membership value for the ith pixel;
(8) membership degree matrix U of (7) outputmaxAnd (3) carrying out pixel classification, carrying out error detection classification, and outputting a segmentation image:
(8a) according to the maximum membership degree principle and the final clustering number YfinMembership matrix U for the output of (7)maxIs classified, and a label matrix U is outputLWherein, ULIncluding only the class values {1,2, …, YW,…,Yfin},1≤YW≤Yfin;
(8b) Is provided withThe label class value of the center pixel i is LiIn the neighborhood of 5 by 5 of the central pixel i, the label type value L is judgediNumber of pixels nLWhether or not n is satisfiedL≤25/(Yfin+1) condition:
if yes, the central pixel i is judged to be a wrong pixel, and the label class value L of the wrong pixel is divided intoiCorrecting the label category value to be the label category value with the maximum number of the same label category values in the neighborhood, and executing (8 c);
if not, judging that the central pixel i is a correctly distributed pixel, and executing (8 c);
(8c) judging tag matrix ULIf so, outputting a final label matrix UFExecuting (8d), if not, returning to (8b) to continue traversing;
(8d) for the final label matrix UFThe middle label class value is YWThe pixels of (2) are assigned different corresponding gray values xWAnd the matrix given different gray values is output as a divided image.
2. The method according to claim 1, wherein (4) a curve fitting is performed using a least square method with respect to the gray value distribution of the input image, and a corresponding gray level is selected as an initialized cluster center value, which is implemented as follows:
(4a) setting the gray level image to 256 gray levels, wherein the range of the gray levels is [0,255], and drawing a gray level histogram of the gray level image;
(4b) counting the gray value distribution frequency f (l) of a gray level l on the gray level histogram, and performing seven-order least square curve fitting, wherein l is more than or equal to 0 and less than or equal to 255;
(4c) calculating a first derivative f ' (l) of f (l), comparing f ' (l) with first derivatives f ' (l-1) and f ' (l +1) of gray levels before and after f (l), and judging whether f ' (l) is less than or greater than 0 only when l is 0 or 255:
when l is 0, judging whether f' (l) < 0 is true, if yes, outputting the gray level l is 0 as a peak point;
when 0 < l < 255, judging whether f ' (l) ═ 0 and f ' (l-1) > 0 and f ' (l +1) < 0 hold, if yes, outputting the gray level l as the peak value point;
when l is 255, judging whether f' (l) > 0 is established, if so, outputting the gray level l is 255 as a peak point;
(4d) judging whether 256 gray levels are traversed completely:
if yes, outputting the peak point, executing (4e), and if not, returning to (4c) to continue traversing;
(4e) sorting the peak points output in the step (4d) according to the sequence of the gray distribution frequency values from large to small, wherein the number of the statistical peak points is YmaxAnd counting the gray levels l corresponding to the peak points ordered from large to small to obtain the gray level value range ofCounting the gray level correspondence values as initial values of cluster centers, e.g.Setting the initial value range set C of the clustering center asWherein lYAnd expressing the gray level corresponding to the Y-th peak point gray level distribution frequency value in the descending order of the peak point gray level distribution frequency value.
3. The method of claim 1, wherein (5b) the gray value x of the pixel under the intuitive blur set is calculatediTo the center of the cluster ckIntuitive fuzzy distance ofThe implementation is as follows:
(5b1) let u (c)k) Representing the center of clustering ckDegree of membership of v (c)k) And pi (c)k) Respectively represent the cluster centers ckThe non-membership degree and the hesitation degree constructed by using the Yager operator respectively have the following calculation formulas:
π(ck)=1-μ(ck)-v(ck),
wherein 0 is not more than mu (c)k)≤1,0≤v(ck)≤1,0≤μ(ck)+v(ck)≤1;
Wherein, mu (x)i)、v(xi) And pi (x)i) Respectively representing pixel grey values xiDegree of membership, degree of non-membership, and degree of hesitation.
4. The method according to claim 1, wherein (5c) local spatial information of the pixels is utilized in combination with the linear weighting function coefficients H obtained in (5a)irThe intuitive blur distance obtained in (5b)Calculating an intuitive fuzzy objective function J, wherein the formula is as follows:
wherein u iskiIs a pixel xiTo the clustering center ckM represents a fuzzy degree of membership function ofAnd (4) weighting the indexes.
5. The method of claim 4, wherein the membership function u is derived from the intuitive fuzzy objective function J using the Lagrangian multiplier methodkiAnd a cluster center ckIntuition fuzzy set ofThe implementation is as follows:
(5c1) introducing Lagrange multiplier lambda into intuition fuzzy target function JiConstructing an objective function L with a Lagrange multiplier:
(5c2) respectively calculating L to membership function ukiCluster center ckIntuition fuzzy set ofAnd lagrange multiplier λiPartial derivatives ofAndand in partial derivativesUnder the conditions of (1), u is obtained as followskiAnd
wherein k is 1,2, …, Y, i is 1,2, …, n, r is 1,2, …, ω2。
6. The method of claim 1, wherein the maximum membership criterion in (8a) is based on Vk=argk{max(uki) Divide the pixel i into the category V with the largest degree of membershipk。
7. The method of claim 1, wherein (8d) is performed on the final tag matrix UFThe middle label class value is YWThe pixels of (2) are assigned different corresponding gray values xWThe formula is as follows:
xW=[255/YW]
wherein [ ·]Indicating the integers taken inside, the final label matrix UFIncluding only the class values {1,2, …, YW,…,Yfin}。
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201910507275.7A CN110211126B (en) | 2019-06-12 | 2019-06-12 | Image segmentation method based on intuitive fuzzy C-means clustering |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201910507275.7A CN110211126B (en) | 2019-06-12 | 2019-06-12 | Image segmentation method based on intuitive fuzzy C-means clustering |
Publications (2)
Publication Number | Publication Date |
---|---|
CN110211126A CN110211126A (en) | 2019-09-06 |
CN110211126B true CN110211126B (en) | 2022-06-03 |
Family
ID=67792371
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201910507275.7A Active CN110211126B (en) | 2019-06-12 | 2019-06-12 | Image segmentation method based on intuitive fuzzy C-means clustering |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN110211126B (en) |
Families Citing this family (12)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN110634141B (en) * | 2019-09-19 | 2022-02-11 | 南京邮电大学 | Image segmentation method based on improved intuitionistic fuzzy c-means clustering and storage medium |
CN110956637B (en) * | 2019-12-02 | 2022-08-02 | 兰州交通大学 | Adaptive fuzzy C-means clustering noise image segmentation method and system |
CN111402272B (en) * | 2020-03-19 | 2023-09-29 | 西安石油大学 | Neighborhood weighted fuzzy clustering image segmentation method |
CN111476303A (en) * | 2020-04-09 | 2020-07-31 | 国网河北省电力有限公司电力科学研究院 | Line loss analysis method of fuzzy C-means clustering based on quantum optimization particle swarm |
CN111539966A (en) * | 2020-04-16 | 2020-08-14 | 华北电力大学 | Colorimetric sensor array image segmentation method based on fuzzy c-means clustering |
CN111932578B (en) * | 2020-08-12 | 2022-06-07 | 河北地质大学 | Pixel block clustering nondestructive testing image segmentation method integrating class index suppression factors |
CN112541528A (en) * | 2020-12-02 | 2021-03-23 | 国家电网有限公司 | Power transmission and transformation project cost prediction index optimization method based on fuzzy clustering |
CN112733825B (en) * | 2020-12-18 | 2024-07-05 | 北京城市排水集团有限责任公司 | Space extraction method for swing arm radar pipeline image diseases |
CN113269796B (en) * | 2021-06-10 | 2023-08-25 | 河北地质大学 | Image segmentation method and device and terminal equipment |
CN113409335B (en) * | 2021-06-22 | 2023-04-07 | 西安邮电大学 | Image segmentation method based on strong and weak joint semi-supervised intuitive fuzzy clustering |
CN114119503A (en) * | 2021-11-05 | 2022-03-01 | 华南师范大学 | Retina blood vessel segmentation method and device based on IFCM clustering |
CN115131588B (en) * | 2022-06-13 | 2024-02-23 | 西北工业大学 | Image robust clustering method based on fuzzy clustering |
Citations (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN108062757A (en) * | 2018-01-05 | 2018-05-22 | 北京航空航天大学 | It is a kind of to utilize the method for improving Intuitionistic Fuzzy Clustering algorithm extraction infrared target |
WO2018119808A1 (en) * | 2016-12-29 | 2018-07-05 | 浙江工商大学 | Stereo video generation method based on 3d convolutional neural network |
WO2019000455A1 (en) * | 2017-06-30 | 2019-01-03 | 上海联影医疗科技有限公司 | Method and system for segmenting image |
CN109145921A (en) * | 2018-08-29 | 2019-01-04 | 江南大学 | A kind of image partition method based on improved intuitionistic fuzzy C mean cluster |
CN109741330A (en) * | 2018-12-21 | 2019-05-10 | 东华大学 | A kind of medical image cutting method of mixed filtering strategy and fuzzy C-mean algorithm |
-
2019
- 2019-06-12 CN CN201910507275.7A patent/CN110211126B/en active Active
Patent Citations (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
WO2018119808A1 (en) * | 2016-12-29 | 2018-07-05 | 浙江工商大学 | Stereo video generation method based on 3d convolutional neural network |
WO2019000455A1 (en) * | 2017-06-30 | 2019-01-03 | 上海联影医疗科技有限公司 | Method and system for segmenting image |
CN108062757A (en) * | 2018-01-05 | 2018-05-22 | 北京航空航天大学 | It is a kind of to utilize the method for improving Intuitionistic Fuzzy Clustering algorithm extraction infrared target |
CN109145921A (en) * | 2018-08-29 | 2019-01-04 | 江南大学 | A kind of image partition method based on improved intuitionistic fuzzy C mean cluster |
CN109741330A (en) * | 2018-12-21 | 2019-05-10 | 东华大学 | A kind of medical image cutting method of mixed filtering strategy and fuzzy C-mean algorithm |
Non-Patent Citations (2)
Title |
---|
"A Modified Intuitionistic Fuzzy Clustering Algorithm for Medical Image Segmentation";S.V. Aruna Kumar;《DE GRUYTER》;20190429;第1-15页 * |
"显著信息引导的直觉空间模糊聚类图像分割";赵凤;《西安邮电大学学报》;20180731;第1-7页 * |
Also Published As
Publication number | Publication date |
---|---|
CN110211126A (en) | 2019-09-06 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN110211126B (en) | Image segmentation method based on intuitive fuzzy C-means clustering | |
CN111340824B (en) | Image feature segmentation method based on data mining | |
CN106875395B (en) | Super-pixel-level SAR image change detection method based on deep neural network | |
CN109509196B (en) | Tongue diagnosis image segmentation method based on fuzzy clustering of improved ant colony algorithm | |
Wu et al. | Adaptive entropy weighted picture fuzzy clustering algorithm with spatial information for image segmentation | |
CN110766058B (en) | Battlefield target detection method based on optimized RPN (resilient packet network) | |
CN106991686B (en) | A kind of level set contour tracing method based on super-pixel optical flow field | |
CN111242878B (en) | Mine image enhancement method based on cuckoo search | |
CN113050042A (en) | Radar signal modulation type identification method based on improved UNet3+ network | |
CN113408605A (en) | Hyperspectral image semi-supervised classification method based on small sample learning | |
CN110569782A (en) | Target detection method based on deep learning | |
CN109801305B (en) | SAR image change detection method based on deep capsule network | |
CN103761726B (en) | Block adaptive image partition method based on FCM | |
CN116091455A (en) | Steel mesh surface defect judging method based on machine vision | |
CN108846845B (en) | SAR image segmentation method based on thumbnail and hierarchical fuzzy clustering | |
CN107423771B (en) | Two-time-phase remote sensing image change detection method | |
Song et al. | A novel self-learning weighted fuzzy local information clustering algorithm integrating local and non-local spatial information for noise image segmentation | |
CN116912568A (en) | Noise-containing label image recognition method based on self-adaptive class equalization | |
CN113989256A (en) | Detection model optimization method, detection method and detection device for remote sensing image building | |
Ayech et al. | Image segmentation based on adaptive Fuzzy-C-Means clustering | |
CN113657394A (en) | Remote sensing image segmentation method based on Markov random field and evidence theory | |
CN111368625B (en) | Pedestrian target detection method based on cascade optimization | |
CN110910497B (en) | Method and system for realizing augmented reality map | |
CN102930558A (en) | Real-time tracking method for infrared image target with multi-feature fusion | |
CN115100406B (en) | Weight information entropy fuzzy C-means clustering method based on superpixel processing |
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 |