CN-121982311-A - Image segmentation method based on multi-target clustering

CN121982311ACN 121982311 ACN121982311 ACN 121982311ACN-121982311-A

Abstract

The invention discloses an image segmentation method based on multi-target clustering. The invention optimizes the clustering center pixels of image segmentation by using an improved multi-target evolution algorithm, and segments the image by using the clustering center pixels obtained by optimization. In the improved multi-objective evolution algorithm, the included angles between individual target values in the current population are calculated, the angle convergence index of the current population is calculated according to the included angles between the individual target values, the dominant relationship is adaptively selected according to the angle convergence index to perform non-dominant ranking on the population, the adaptive crowding degree of the individuals is calculated, and the individuals are selected to enter the new generation population according to the non-dominant ranking result and the adaptive crowding degree of the individuals. The invention effectively maintains the balance between population convergence and diversity by utilizing the angle convergence index and the adaptive crowding degree mechanism, and improves the precision of image segmentation.

Inventors

ZHAO QINGYU
Yi Zibin
GUO ZHAOLU
ZHANG SHIYAO
ZHANG WENSHENG

Assignees

江西理工大学

Dates

Publication Date: 20260505
Application Date: 20260122

Claims (3)

1. An image segmentation method based on multi-target clustering is characterized by comprising the following steps: Step1, inputting an image to be segmented ; Step 2, setting the number of clusters Setting the size of the population Maximum update iteration number Interval parameter Angle weight control parameter ; Step 3, setting the current iteration times ; Step 4, randomly generating population , wherein, Is a population Is the first of (2) Individual, each individual in the population storing Clustering center pixels; represent the first Individual first Clustering center pixels, individual subscripts Dimension subscript ; Step 5, calculating individuals in the population according to formulas (1) and (2) Adapted value of (2) : (1) (2) Wherein, the Representing a population Middle (f) A first target value for the individual person, Representing an image to be segmented Is a set of pixels of (a); For images to be segmented Is the first of (2) A set of pixels of a class; Representing a collection Pixels in (a); Representing a population Middle (f) Individual first A cluster center of the class; representing pixels And cluster center In the image to be segmented The Euclidean distance in (a); Representing a population Middle (f) A second goal of the individual is to have a second goal, For images to be segmented Is a pixel of (1); Is a pixel Is a set of eight neighborhood pixels; Is a collection Pixels in (a); representing pixels And pixel Cluster difference value, cluster category superscript ; Step 6, setting stock population ; Step 7, if the current iteration number Greater than Go to step 25, otherwise go to step 8; Step 8, making the interval remainder Where mod is a function of the remainder; step 9, if the remainder is spaced Equal to 0, go to step 10, otherwise go to step 12; Step 10, calculating a governance policy ; Step 11, setting stock population ; Step 12, for population Crossover and mutation are carried out to generate a product with the size of Is a offspring population of (a) Calculating offspring population Target vector of the individual in the population of offspring And parent population Merging into a population , wherein, Representing a population Is the first of (2) Individual, subscript ; Step 13, if the strategy is dominant Angle dominant, go to step 14, otherwise go to step 15; step 14, calculating population Angle vectors for each individual; step 15, setting population according to formula (3) Temporary target vector for individual: (3) Wherein, the Is a population Middle (f) The temporary target vector of the individual person, Is a population Middle (f) Angle vectors of individual individuals; Is a population Middle (f) A first component of the angle vector of the individual; Is a population Middle (f) A second component of the individual angle vectors; Is a population Middle (f) Target vectors for individual individuals; Is a population Middle (f) A first target value of the target vector of the individual; Is a population Middle (f) A second target value of the target vector of the individual; step 16, according to the population Temporary target vector of (a) To the population Non-dominant ranking, the result of ranking is recorded as Wherein The first to represent non-dominant ranking results Layer, subscript A kind of electronic device The number of layers that are the result of the non-dominant ordering; step 17, calculating population An adaptive congestion level for each individual; Step 18, selecting a population from the population based on the non-dominant ranking result Selecting individuals to enter a new generation population The specific process is as follows: step 18.1, population is collected Set as empty set, set ; Step 18.2 if The number of individuals in a population The number of individuals in (a) is smaller than Go to step 18.3, otherwise go to step 19; step 18.3, will Is put into population In (a) and (b); step 18.4, setting ; Step 18.5, go to step 18.2; step 19, for The individuals in (2) are ranked according to the adaptability crowding degree from big to small; step 20, calculating population Number of individuals in (3) ; Step 21, will Front of (a) Individual into a population In (a) and (b); Step 22, setting population ; Step 23, setting the current iteration times ; Step 24, turning to step 7; Step 25, population is collected Non-dominant individuals in (1) are saved to collection From the collection Selecting individuals as a clustering scheme; Step 26, extracting the clustering center pixel of the clustering scheme, and utilizing the obtained clustering center pixel to perform image matching Dividing pixels of the image to obtain an image Is a segmentation result of (a).
2. The method for image segmentation based on multi-objective clustering according to claim 1, wherein the step 10 calculates a dominating strategy The steps of (a) are as follows: step 10.1, current population is collected And inventory population Merging into a population ; Step 10.2, computing population Is the minimum target value of (2) And a maximum target value ; Step 10.3, according to the minimum target value And a maximum target value Respectively to populations Sum population Is normalized for individuals of (a); Step 10.4, calculating normalized population And population of Included angles of target vectors among individuals; Step 10.5, for population Individuals in (a) And population of The included angles of all the individuals are ordered from small to large, and the ordering result is recorded as follows: ; step 10.6, setting an individual With individuals Included angle between Ranking in the ranked results as Wherein the individual Is a population Is of the order of For stock population Is a subject of (2); step 10.7, calculating population according to formula (4) Angle weight of the middle individual : (4) Step 10.8, calculating population according to formulas (5) and (6) Angle convergence index value of (2) : (5) (6) Wherein, the And Respectively individual And individuals Is the first of (2) Normalized target values, superscript ; Representing a natural exponential function; Representing individuals For individuals Is a dominant quantity of (2); Representing a maximum function; step 10.9, calculating population according to formula (7) Angle convergence index value of (2) ; (7) Wherein, the Is a population Angle weight of the middle individual; Representing individuals For individuals Is a dominant quantity of (2); step 10.10 if Then set the governance policy For angle dominance, otherwise set dominance policy Is pareto dominant.
3. The method for image segmentation based on multi-objective clustering according to claim 1, wherein the step 17 calculates the population The steps of the adaptive congestion level of each individual are as follows: Step 17.1, calculating the individual And (3) with Minkowski distance of the target vector between all individuals; step 17.2, sorting the obtained Minkowski distances in order of decreasing size, and recording the sorting result as Wherein, the method comprises the steps of, Is that Number of individuals in (b) subscript ; Step 17.3, calculating the congestion factor according to equation (8) : (8) Wherein, the To optimize the target number; Step 17.4, calculating the individual according to formulas (9) and (10) and (11) Adaptive congestion degree of (2): (9) (10) (11) Wherein, the For individuals Is a congestion level of adaptability; is an adaptive distance; representing a normalized minkowski distance; And Respectively represent maximum value and minimum value in all Minkowski distances Positive integer ; Rounding to a rounding function; to take a minimum function.

Description

Image segmentation method based on multi-target clustering Technical Field The invention relates to the field of digital image processing, in particular to an image segmentation method based on multi-target clustering. Background Image segmentation is an important research direction in the field of digital image processing, and has wide application in engineering practice, such as medical image diagnosis, industrial defect detection, remote sensing image analysis and other engineering applications. In the image segmentation method based on clustering, the clustering center pixels of the image segmentation often affect the effect of the image segmentation to a great extent. However, there is no perfect theoretical guidance on how to optimize the cluster center pixels of the image segmentation. The multi-objective evolution algorithm is an intelligent optimization algorithm which refers to the natural evolution law, and a good optimization effect is obtained in a plurality of digital image processing problems. Currently, many researchers have attempted to optimize cluster center pixels of image segmentation using multi-objective evolutionary algorithms. However, the conventional multi-objective evolution algorithm is easy to have the defect of insufficient segmentation precision when optimizing the clustering center pixels of image segmentation. Disclosure of Invention The invention provides an image segmentation method based on multi-target clustering. The method overcomes the defect that the traditional multi-target evolution algorithm is easy to have insufficient segmentation precision when optimizing the clustering center pixels of image segmentation to a certain extent, and can improve the segmentation precision of the digital image. The technical scheme of the invention is that the image segmentation method based on multi-target clustering comprises the following steps: Step1, inputting an image to be segmented ; Step 2, setting the number of clustersSetting the size of the populationMaximum update iteration numberInterval parameterAngle weight control parameter; Step 3, setting the current iteration times; Step 4, randomly generating population, wherein,Is a populationIs the first of (2)Individual, each individual in the population storingClustering center pixels; represent the first Individual firstClustering center pixels, individual subscriptsDimension subscript; Step 5, calculating individuals in the population according to formulas (1) and (2)Adapted value of (2): Wherein, the Representing a populationMiddle (f)A first target value for the individual person,Representing an image to be segmentedIs a set of pixels of (a); For images to be segmented Is the first of (2)A set of pixels of a class; Representing a collection Pixels in (a); Representing a population Middle (f)Individual firstA cluster center of the class; representing pixels And cluster centerIn the image to be segmentedThe Euclidean distance in (a); Representing a population Middle (f)A second goal of the individual is to have a second goal,For images to be segmentedIs a pixel of (1); Is a pixel Is a set of eight neighborhood pixels; Is a collection Pixels in (a); representing pixels And pixelCluster difference value, cluster category superscript; Step 6, setting stock population; Step 7, if the current iteration numberGreater thanGo to step 25, otherwise go to step 8; Step 8, making the interval remainder Where mod is a function of the remainder; step 9, if the remainder is spaced Equal to 0, go to step 10, otherwise go to step 12; Step 10, calculating a governance policy ; Step 11, setting stock population; Step 12, for populationCrossover and mutation are carried out to generate a product with the size ofIs a offspring population of (a)Calculating offspring populationTarget vector of the individual in the population of offspringAnd parent populationMerging into a population, wherein,Representing a populationIs the first of (2)Individual, subscript; Step 13, if the strategy is dominantAngle dominant, go to step 14, otherwise go to step 15; step 14, calculating population Angle vectors for each individual; step 15, setting population according to formula (3) Temporary target vector for individual: Wherein, the Is a populationMiddle (f)The temporary target vector of the individual person,Is a populationMiddle (f)Angle vectors of individual individuals; Is a population Middle (f)A first component of the angle vector of the individual; Is a population Middle (f)A second component of the individual angle vectors; Is a population Middle (f)Target vectors for individual individuals; Is a population Middle (f)A first target value of the target vector of the individual; Is a population Middle (f)A second target value of the target vector of the individual; step 16, according to the population Temporary target vector of (a)To the populationNon-dominant ranking, the result of ranking is recorded asWhereinThe first to represent non-dominant ranki