CN104537660B - Image partition method based on Multiobjective Intelligent body evolution clustering algorithm - Google Patents

Abstract

The invention discloses a kind of image partition method based on Multiobjective Intelligent body evolution clustering algorithm, mainly solve to be easily trapped into the not high problem of local optimum, algorithm robustness in image Segmentation Technology.The problem of dividing the image into is converted into a global optimization clustering problem.Process includes：Extract image slices vegetarian refreshments half-tone information to be split；Parameter initialization simultaneously sets up image intelligent volume mesh；The energy of image intelligent body is calculated, non-dominated ranking is carried out；Carry out neighborhood contention operation；Carry out Gaussian mutation operation；Calculate the energy of image intelligent body, non-dominated ranking；Self study operation is carried out, optimum cluster result is selected according to crowding distance, cluster labels are exported；Realize that image is split.Image processing process multiple target, not only good in convergence effect, and enhance the robustness of method can be improved the quality of image segmentation and the stability of enhancing segmentation effect, be conducive to the extraction, identification and some other subsequent treatment of image object by the present invention.

Description

Image partition method based on Multiobjective Intelligent body evolution clustering algorithm

Technical field

The present invention is under the jurisdiction of technical field of image processing, relates generally to image partition method, is specifically that one kind is based on many mesh The image partition method of intelligent body evolution clustering algorithm is marked, available for fields such as pattern-recognition and computer visions.

Background technology

Image is a key technology in image processing, its image procossing research in have widely should With being all that segmentation result directly influences follow-up work based on image is split for example to target identification, target measurement Carry out, therefore the research tool of image segmentation is of great significance.Image segmentation is a kind of special image procossing skill in fact Art, it according to image pixel attribute is a process that gray scale, texture, color are classified that it, which is substantially,.

Method based on cluster is one kind of unsupervised segmentation, is widely used in work biomedicine, computer vision With the field such as remote sensing image processing.Cluster is that one group of data of unknown distribution are classified, and farthest makes same category In data there is identical property, and inhomogeneous data have different properties, and it is substantially to find to be hidden in Rule in data.

Based on this, various clustering algorithms are applied to image segmentation field and also achieve increasingly satisfaction Effect.But because the clustering algorithm of the particularity and diversity of view data, and not all can be applied directly to image point Field is cut, many algorithms are required for carrying out packaging improvement, and even a part of algorithm is not suitable for carrying out image point Cut.In current research, conventional clustering technique has following several：Hierarchical clustering algorithm, nearest-neighbor clustering algorithm is obscured poly- Class algorithm, artificial neural network clustering algorithm, genetic algorithm for clustering.Initial research personnel be frequently utilized for image segmentation cluster be Hierarchical cluster.Its advantage is simple, it is easy to operated, but it also brings many inconvenience simultaneously：Such as, algorithm robustness is poor, holds Easily it is absorbed in local optimum, cluster result is less desirable etc..Such issues that in order to solve, many researchers have been carried out a lot Attempt, propose to use Genetic Algorithms combined with it, obtained satisfied result, but because traditional genetic algorithm is global The limitation of evolution mechanism, the clustering method with reference to after still has cluster result unstable and is easily trapped into local optimum etc. Defect, causes the Quality Down of image segmentation result and the reduction of segmentation effect stability, be unfavorable for follow-up graphical analysis and Understand.

The content of the invention

It is an object of the invention to overcome the shortcomings of that above-mentioned local optimum and the algorithm robustness of being easily absorbed in be not high, one is proposed The image partition method based on Multiobjective Intelligent body evolution clustering algorithm is planted, to strengthen the robustness and image point of dividing method The stability of result is cut, improves image segmentation.

The technical scheme is that, based on the image partition method of Multiobjective Intelligent body evolution clustering algorithm, its feature It is：Including having the following steps：

Step 1 inputs image to be split, extracts the half-tone information of image to be split, is needed if input picture is not gray level image Pre-processed and it is changed into gray level image, half-tone information is labeled as data.

Step 2 set initial clustering number c and iteration upper limit number of times T, to gradation of image information carry out clustering prototype with Machine is initialized, and clustering prototype is the initial clustering result of gradation of image information, and the random initializtion of clustering prototype refers at random It is cluster centre to specify c pixel value, by the pixel classifications of image to margin of image element away from minimum cluster centre, each gray scale The clustering prototype of information is as an image intelligent body, then for image intelligent body L_y, L subscript y refers to image intelligent body Numbering, y=1,2 ..., L_size×L_size, L_y={ e_p, e_pIt is image intelligent body L_yIn element, p is element in image intelligent body Subscript, 1≤p≤c, setting image intelligent body existence grid L^tSize is L_size×L_size, competition probability is P₀, mutation probability is P_m, make current evolutionary generation t=0.

Step 3 calculates image intelligent body L_yENERGY E rengyL_y=(E₁,E₂), y refers to the numbering of image intelligent body, y= 1,2,…,L_size×L_size, ENERGY E rengyL_yIncluding the component L of energy 1_y(E₁) and the component L of energy 2_y(E₂)。

Step 4 is according to the energy values of all image intelligent bodies, to image intelligent volume mesh L^tMiddle image intelligent body L_yCarry out non- Dominated Sorting is operated, and the image intelligent body that order is not dominated by any other image intelligent body is first layer, is saved in set In level1, remaining all image intelligent body is made to be saved in set level2；

Step 5 acts on neighborhood competition operator on the image intelligent body in set level2, and by obtained new figure New image intelligent volume mesh L is collectively constituted as intelligent body merges with the image intelligent body in set level1^t+1/2。

Step 6 actively produces the random number R (0,1) of one 0 to 1, by mutation probability P_mCompared with R (0,1), if variation Probability p_mMore than R (0,1), then Gaussian mutation operator is acted on to image intelligent volume mesh L successively^t+1/2In intelligent body on, obtain To image intelligent volume mesh L of future generation^t+1。

Step 7 pair next generation's image intelligent volume mesh L^t+1Middle image intelligent body performs step 3 and arrives step 4, obtain one group it is non- Dominating figure is as intelligent body L_a, a refers to the numbering of non-dominant image intelligent body, a=1,2 ..., A, 1≤A≤L_size×L_size, this Group non-dominant image intelligent body refers to the image intelligent body in ruleless first layer.

Step 8 is to non-dominant image intelligent body L_aSelf-learning operator operation is carried out successively, obtains second generation non-dominant disaggregation, Include second generation non-dominant image intelligent body L_b, b refers to the numbering of second generation non-dominant image intelligent body, b=1,2 ..., A, 1 ≤A≤L_size×L_size；

Step 9 judges whether iteration terminates：If evolutionary generation t is not less than algebraically T, jump procedure 10 is terminated, if evolving generation Number t, which is less than, terminates algebraically T, t=t+1, return to step 3；

Step 10 is according to the second generation non-dominant image intelligent body L obtained in step 8_bB=1,2 ..., A, 1≤A≤ L_size×L_size, the crowding distance dis of each image intelligent body is calculated successively,

In formula, b=1,2,3 ..., A select optimum image intelligent body to crowding distance according to maximum crowding distance principle L_BestAnd export the corresponding optimum cluster label of the image intelligent body；

Step 11 carries out pixel classifications according to optimum cluster label to input picture, regard similar pixel as a pixel Block, all block of pixels constitute segmentation figure picture by image original position, export the segmentation result of input picture.

Compared with prior art, it is an advantage of the invention that overcoming the limitation of traditional images segmentation, realize：

1) because the present invention has used for reference design in the algorithm frame of multiple target thought, cluster process and used quick non-dominant Sort the operation such as operator and crowding distance selection, it is single to overcome optimal result in iterative process, the shortcomings of population sample is single, So that the population of image segmentation process is more diversified, segmentation result is closer to optimal result.

2) because being calculated present invention employs Agent Grid structure, neighborhood competition operator, Gaussian mutation operator and self study Son etc. is operated, therefore overcomes the shortcomings of being easily absorbed in local optimum and unstable segmentation result so that image segmentation result Accuracy is higher, also improves the stability of image segmentation.

The population more diversification of the cluster process in image segmentation can be realized by the present invention, and the invention is broken The limitation of normal image dividing method, it is initiative to be combined multiple target thought with multi-Agent evolutionary Algorithm, propose A kind of evolution clustering algorithm based on Multiobjective Intelligent body, realizes the raising to segmentation result accuracy, and enhances point Cut the stability of effect.

Brief description of the drawings

The implementation process figure of Fig. 1 the inventive method；

The grid environment of Fig. 2 intelligent bodies existence；

Fig. 3 the inventive method is applied to the emulation segmentation result of two gray-like image pictures；

Fig. 4 the inventive method, Kmeans cluster segmentations and FCM cluster segmentations are applied to the simulation result of sar rivers image；

Fig. 5 the inventive method, Kmeans cluster segmentations and FCM cluster segmentations are applied to the simulation result of sar Airport Images；

Embodiment

Below in conjunction with the accompanying drawings to the detailed description of the invention：

Based on the image partition method of Multiobjective Intelligent body evolution clustering algorithm, its committed step is multiple target thought and many The combination of intelligent body evolution algorithm, and complete image segmentation for clustering optimization.

Image partition method based on Multiobjective Intelligent body evolution clustering algorithm is characterized in：Image to be split is extracted first Half-tone information, then it is clustered and optimized with Multiobjective Intelligent body evolution algorithm, then according to the poly- of final output Class label is classified to pixel and then realizes the segmentation of image.

Embodiment 1

Reference picture 1, the present invention is a kind of image partition method based on Multiobjective Intelligent body evolution clustering algorithm, specific real Existing step is as follows：

Step 1, image to be split is inputted, the half-tone information of image to be split is extracted, if input picture is not gray level image It need to be pre-processed and it is changed into gray level image, half-tone information is labeled as data.

Step 2, set initial clustering number c and iteration upper limit number of times T, to gradation of image information carry out clustering prototype with Machine is initialized, and clustering prototype is the initial clustering result of gradation of image information, and the random initializtion of clustering prototype refers at random It is cluster centre to specify c pixel value, is saved in image intelligent body, by the pixel classifications of image to margin of image element away from minimum Cluster centre, each clustering prototype is as an image intelligent body, then for image intelligent body L_y, L subscript y refers to image The numbering of intelligent body, y=1,2 ..., L_size×L_size, L_y={ e_p, e_pIt is image intelligent body L_yIn element, p is image intelligent Element subscript in body, 1≤p≤c, setting image intelligent body existence grid L^t, t refers to current evolutionary generation, and sizing grid is L_size ×L_size, competition probability is P₀, mutation probability is P_m, make current evolutionary generation t=0；In this example, clusters number c=2, in iteration Limit number of times T=100, Agent Grid size L_size=8, compete probability P₀=0.2, Gaussian mutation probability P_m=0.1；

Image intelligent volume mesh is a grid environment of image intelligent body existence, is designated as L, size is defined as L_size× L_size, wherein L_sizeFor integer, each image intelligent body is fixed on a lattice point, the image intelligence that note is arranged in the i-th i rows, jth j Energy body is L_ii,jj, ii, jj=1,2 ..., L_size, image intelligent body L_ii,jjNeighborhood be：

Wherein,

Each image intelligent body is immovable, can only be interacted with its neighborhood,For image intelligent body L_ii,jj The maximum image intelligent body of neighboring region energy；

Fig. 2 gives each circle in the structural representation of image intelligent volume mesh, figure and represents an image intelligent body, circle Numeral in circle represents the position of image intelligent body within a grid, and two image intelligent bodies for having mutual line could occur Interaction.

Step 3, all image intelligent body L in population are calculated using following two formula_yENERGY E rengyL_y=(E₁, E₂), L subscript y refers to the numbering of image intelligent body, y=1,2 ..., L_size×L_size, calculating process include 1 component L_y(E₁) Calculate and 2 component L_y(E₂) calculating, wherein：Image intelligent body L_y1 component of energy is calculated

X is cluster image intensity value, z in formula_jFor correspondence cluster centre, image intelligent body L_y2 components of energy are calculated

I is a data point in formula, and m is the number of data point, and j is i nearest neighbor point, and s is nearest neighbor point number, x_i,j For i-th of data point and the relation value of j-th of nearest neighbor point, wherein i, j is similar pixel then value 0, and inhomogeneity then takes

Step 4, according to the energy value of all image intelligent bodies, to image intelligent volume mesh L^tMiddle image intelligent body, y=1, 2,…,L_size×L_sizeCarry out non-dominated ranking operation, the image L that order is not dominated by any other image intelligent body_yIntelligent body is First layer, is saved in set level1, makes remaining all image intelligent body be saved in set level2.Detailed process bag Include：

(4.1) initiation parameter, order dominates intelligent body L_yIntelligent body number n_Ly=0.

(4.2) n is updated as follows_Ly：

For image intelligent body L_y, y=1,2 ..., L_size×L_size, have image intelligent body, i refers to the volume of image intelligent body Number, i=1,2 ..., L_size×L_size, and L_iCan not be L_yIf, image intelligent L_i L_yEnergy meet Then image intelligent body L_yBy image intelligent body L_iDominated, make n_Ly=n_Ly+ 1, if meetingThen claim image intelligence Can body L_yDominating figure is as intelligent body L_i。

(4.3) if image intelligent body L_y, y=1,2 ..., L_size×L_sizeN_Ly=0 makes image intelligent body L_yInto One layer, it is saved in set level1, there is n_level1Individual, if its n_Ly≠ 0, make L_yInto the second layer, level2 is saved in In.The image intelligent body of different levels is retained separately by the present invention, is subsequently to use for convenience.

Step 5, by neighborhood competition operator act on successively on the image intelligent body in set level2, by obtain newly Image intelligent body is merged with the image intelligent body in set level1, and new image intelligent volume mesh is collectively constituted after merging L^t+1/2。

Step 6, the random number R (0,1) of one 0 to 1 is actively produced, mutation probability and random number R (0,1) are compared, If mutation probability p_mMore than random number R (0,1), then Gaussian mutation operator is acted on to image intelligent volume mesh L successively^t+1/2In On intelligent body, image intelligent volume mesh L of future generation is obtained^t+1。

Gaussian mutation operator is to produce a variation image intelligent body by following formula

Wherein p=1,2 ..., c, c are clusters number, l_pFor intelligent body L_ii,jjIn be located at p at element value, G (0,1/t) It is the random number of Gaussian Profile；And random number R (0,1) is the random number between 0 to 1, T is total evolutionary generation, and t is to work as evolution Algebraically.

Step 7, to image intelligent volume mesh L of future generation^t+1Middle image intelligent body performs step 3 and arrives step 4, obtains one group Non-dominant image intelligent body L_a, a refers to the numbering of non-dominant image intelligent body, a=1,2 ..., A, 1≤A≤L_size×L_size, This group of non-dominant image intelligent body refers to the image intelligent body in ruleless first layer.

Step 8, to non-dominant image intelligent body L_a, a=1,2 ... A, 1≤A≤L_size×L_sizeSelf study is carried out successively Operator is operated, and is obtained second generation non-dominant disaggregation, is included second generation non-dominant image intelligent body L_b, b refers to second generation non-dominant The numbering of image intelligent body, b=1,2 ..., A, 1≤A≤L_size×L_size。

Step 9, judge whether iteration terminates：If evolutionary generation t is not less than algebraically T, jump procedure (10) is terminated, if evolving Algebraically t, which is less than, terminates algebraically T, t=t+1, return to step 3.

Step 10, according to the second generation non-dominant image intelligent body L obtained in step 8_b, each image intelligent body is calculated successively Crowding distance dis,

In formula, b=1,2,3 ..., A select optimum image intelligent body to crowding distance according to maximum crowding distance principle L_BestAnd export the corresponding optimum cluster label of the image intelligent body.

Step 11, pixel classifications are carried out to input picture according to optimum cluster label, regard similar pixel as a picture Plain block, all block of pixels constitute segmentation figure picture by image original position, export the segmentation result of input picture.

Quick non-dominant has been used to arrange because the present invention has used for reference design in the algorithm frame of multiple target thought, cluster process The operation such as sequence operator and crowding distance selection, overcomes that optimal result in iterative process is single, the shortcomings of population sample is single makes Image segmentation process population it is more diversified, segmentation result is closer to optimal result.

Embodiment 2

Constitute new based on the image partition method be the same as Example 1 of Multiobjective Intelligent body evolution clustering algorithm, wherein in step 5 Image intelligent volume mesh L^t+1/2Process be that neighborhood is competed into the image intelligent body that acts on successively in set level2 of operator On, and obtained new image intelligent body is merged with the image intelligent body in set level1, collectively constituted after merging new Image intelligent volume mesh L^t+1/2.Neighborhood competition process is to compete operator by neighborhood with the new image intelligent of two kinds of strategy generations Body, total process includes：

Neighborhood competition operator is to produce a new image intelligent body by one of the following two kinds strategy

Strategy 1, produces image intelligent body as the following formula

Wherein p=1,2 ..., c, e_pForIn element,For the lower bound of all image intelligent element of volume values,For The upper bound of all image intelligent element of volume values, m_pIt is located at the element value at p, l for any image intelligent body in level1_pFor image Intelligent body L_ii,jjIn be located at the value that p locates, c is clusters number, and R (- 1,1) is the random number between -1 to+1.

Strategy 2, produces image intelligent body as follows

The first step, as the following formula willMiddle all elements m_pIt is mapped on interval [0,1], obtains new member

Element:Constituted with these elements middle for image intelligent body

Wherein p=1,2 ..., c, 1<i₁<C, 1<i₂<C, i₁<i₂, c is optimum cluster number, m_pFor any figure in level1 As element value of the intelligent body at p,For the lower bound of all image intelligent element of volume values,For all image intelligent volume elements The upper bound of element value；

Second step, according to following formula by image intelligent bodyMap back intervalOn, obtain image intelligent body

Wherein, e_pForIn element,For the lower bound of all image intelligent element of volume values,For all image intelligents The upper bound of element of volume value, c is clusters number.

Two kinds of above-mentioned Different Strategies are according to competition probability P₀Selected：

First, the random number R (0,1) between one 0 to 1 is actively produced；

Secondly, random number R (0,1) and competition probability P are made₀Compare, if random number R (0,1) ＞ P₀, then selection strategy 1, otherwise, selection strategy 2.

Embodiment 3

Image partition method be the same as Example 1-2 based on Multiobjective Intelligent body evolution clustering algorithm, is wherein obtained in step 8 Second generation non-dominant image intelligent body L_b, b be second generation non-dominant image intelligent body numbering, b=1,2 ..., A, 1≤A≤ L_size×L_sizeProcess be by non-dominant image intelligent body L_a, a=1,2 ..., A, 1≤A≤L_size×L_sizeEnter successively The operation of row self-learning operator is obtained.Self-learning operator operating process is to produce a new image intelligent by self-learning operator Body, total process includes：

Self-learning operator is to produce an image intelligent body as follows：

(8.1) a self study image intelligent volume mesh sL is produced using the method for image intelligent volume mesh generation, its is big Small is sL_size×sL_size, sL_sizeFor integer, all image intelligent body sL thereon_i′,j′, i ', j '=1,2 ..., sL_size Produced according to following formula：

Wherein p=1,2 ..., c,For the lower bound of all image intelligent body gray values；For all image intelligent bodies ash The upper bound of angle value；l_pFor intelligent body L_a, a=1,2 ... A, 1≤A≤L_size×L_sizeIn be located at p at gray value, R (1-sR, 1 + sR) represent 1-sR to 1+sR between random number, sR ∈ [0,1] represent search radius.

(8.2) neighborhood is competed into operator and mutation operator iteration is acted on self study image intelligent volume mesh sL, it is maximum Iterative algebra is sG, and present image intelligent body is substituted with the maximum image intelligent body of energy in self study image intelligent volume mesh sL L_a, a=1,2 ... A, 1≤A≤L_size×L_size。

In this example, sL_size=4, sR=0.3, sG=8.

Because present invention employs Agent Grid structure, neighborhood competition operator, Gaussian mutation operator and self-learning operator Deng operation, therefore overcome the shortcomings of being easily absorbed in local optimum and unstable segmentation result so that the standard of image segmentation result True property is higher, also improves the stability of image segmentation.

The effect of the present invention can be further illustrated by following simulation result.

Embodiment 4

Image partition method be the same as Example 1-3 based on Multiobjective Intelligent body evolution clustering algorithm, passes through emulation experiment pair The implementation process and technique effect of the present invention is further illustrated.

Parameter setting：

Clusters number c=2, maximum evolutionary generation T=100, Agent Grid size L_size=8, compete probability P₀=0.2, Gaussian mutation probability P_m=0.1, self study Agent Grid size sL_size=4, search radius sR=0.3, self study variation are general Rate sP_m=0.1, self study algebraically sG=8.

Emulation content：

Using image partition method of the present invention, the gray-like image picture of a width two and a width SAR rivers image are divided respectively Cut, and give result to illustrate that the inventive method is applied to the performance that image is split.

Emulation content 1, the image partition method of the present invention is split applied to two gray-like image pictures, and its result is as schemed Shown in 8, wherein：Fig. 3 (a) is original composograph, and object gray value is 255, and background gray levels are that 51, Fig. 3 (b) ties for segmentation Fruit is schemed.From result it can be seen from the figure that present invention obtains correct segmentation result, the two classes ash of synthesis has successfully been partitioned into The circular object spent in image, illustrates that the inventive method can obtain correct segmentation result.

Emulation content 2, the emulation experiment that SAR rivers image is split is applied to by the image partition method of the present invention, And the emulation experiment of conventional segmentation methods Kmeans cluster segmentations method and FCM cluster segmentation methods, its result such as Fig. 4 institutes Show, Fig. 4 (a) is original SAR rivers image, Fig. 4 (b) is the segmentation result figure of Kmeans cluster segmentation methods, and Fig. 4 (c) is poly- for FCM The segmentation result figure of class dividing method, Fig. 4 (d) is the segmentation result figure of dividing method of the present invention.From Fig. 4 (d) as can be seen that pair In SAR rivers image, the segmentation result of this method has preferable region consistency, compared to traditional cluster segmentation side in Fig. 4 (c) FCM cluster segmentation results in method Kmeans cluster segmentations result and Fig. 4 (d), the inventive method can not only successfully be partitioned into figure River as in, the mistake point to land building also reduces a lot, greatly reduces the influence that noise spot is split to river.

Embodiment 5

Image partition method be the same as Example 1-3 based on Multiobjective Intelligent body evolution clustering algorithm, simulated conditions and emulation Parameter setting be the same as Example 4.

Emulation content 3, the emulation experiment that SAR Airport Images are split is applied to by the image partition method of the present invention, And the emulation experiment of conventional segmentation methods Kmeans cluster segmentations method and FCM cluster segmentation methods, its result such as Fig. 5 institutes Show, Fig. 5 (a) is original SAR rivers image, Fig. 5 (b) is the segmentation result figure of Kmeans cluster segmentation methods, and Fig. 5 (c) is poly- for FCM The segmentation result figure of class dividing method, Fig. 5 (d) is the segmentation result figure of dividing method of the present invention.In terms of Fig. 5 (d) simulation results Go out, for SAR airfield runway images, compared to the FCM clusters point in the Kmeans cluster segmentations result and Fig. 5 (c) in Fig. 5 (b) Result is cut, the segmentation result of the inventive method has more preferable region consistency, and inhibits the noise pair on airfield runway The influence of image segmentation, then gives simultaneously for the weak signal target in addition to primary runway and ignores, be more clearly from partitioned into airport master Runway.

In brief, the image partition method of the invention based on Multiobjective Intelligent body evolution clustering algorithm, is mainly solved Local optimum is easily trapped into conventional images cutting techniques, the problems such as algorithm robustness is not high.The present invention asks the segmentation of image Topic is converted into the clustering problem of a global optimization, and its segmentation step is：Extract the half-tone information of image slices vegetarian refreshments to be split；Ginseng Number initializes and sets up image intelligent volume mesh；Calculate the energy of all image intelligent bodies and carry out non-dominated ranking；It is right successively Second layer image intelligent body carries out neighborhood competition operator and operates and merge the new image intelligent of composition with first layer image intelligent body Volume mesh；Gaussian mutation operation is carried out to new Agent Grid to calculate；Energy to all image intelligent bodies and progress again Non-dominated ranking；Carry out self study operation；Optimal cluster result is selected according to crowding distance, cluster labels are exported；Realize figure As segmentation.The processing procedure multiple target that the present invention is divided the image into, not only good in convergence effect, and enhance the robust of method Property, so the quality of image segmentation and the stability of enhancing segmentation effect can be improved, be conducive to extraction, the identification of image object And some other subsequent treatment.

Claims (6)

1. a kind of image partition method based on Multiobjective Intelligent body evolution clustering algorithm, it is characterised in that include following step Suddenly：

Step 1 inputs image to be split, extracts the half-tone information of image to be split, and half-tone information is labeled as data；

Step 2 sets initial clustering number c and iteration upper limit number of times T, and the random first of clustering prototype is carried out to gradation of image information Beginningization, the clustering prototype of each half-tone information is as an image intelligent body, then for image intelligent body L_y, L subscript y is Refer to the numbering of image intelligent body, y=1,2 ..., L_size×L_size, L_y={ e_p, e_pIt is image intelligent body L_yIn element, p is figure As the subscript of element in intelligent body, 1≤p≤c, setting image intelligent body existence grid L^tSize be L_size×L_size, competition is generally Rate is P₀, mutation probability is P_m, make current evolutionary generation t=0；

Step 3 calculates image intelligent body L_yENERGY E rengyL_y=(E₁,E₂), ENERGY E rengyL_yIncluding the component E of energy 1₁With The component E of energy 2₂；

Step 4 is according to the energy value of all image intelligent bodies, and survive grid L to image intelligent body^tIn image intelligent body L_yCarry out Non-dominated ranking is operated, and the image intelligent body that order is not dominated by any other image intelligent body is first layer, is saved in set In level1, remaining all image intelligent body is made to be saved in set level2；

Step 5 acts on neighborhood competition operator on the image intelligent body in set level2 successively, and by obtained new figure New image intelligent body existence grid L is collectively constituted as intelligent body merges with the image intelligent body in set level1^t+1/2；

Step 6 actively produces the random number R (0,1) of one 0 to 1, by mutation probability P_mCompared with R (0,1), if mutation probability P_mMore than R (0,1), then Gaussian mutation operator is acted on to image intelligent body existence grid L successively^t+1/2In intelligent body on, obtain To image intelligent body of future generation existence grid L^t+1；

Step 7 pair next generation's image intelligent body existence grid L^t+1In image intelligent body perform step 3 arrive step 4, obtain one group Non-dominant image intelligent body L_a, a refers to the numbering of non-dominant image intelligent body, a=1,2 ..., A, 1≤A≤L_size×L_size；

Step 8 is to non-dominant image intelligent body L_aSelf-learning operator operation is carried out successively, obtains second generation non-dominant disaggregation, comprising Second generation non-dominant image intelligent body L_b, b refers to the numbering of second generation non-dominant image intelligent body, b=1,2 ..., A；

Step 9 judges whether iteration terminates：If evolutionary generation t is not less than termination algebraically T, jump procedure 10, if evolutionary generation t is small In termination algebraically T, t=t+1, return to step 3；

Step 10 is according to the second generation non-dominant image intelligent body L obtained in step 8_b, the crowded of each image intelligent body is calculated successively Apart from dis,

$dis\left(L_b\right)=\sqrt{{\left(L_{b-1}\right(E_1)-L_{b+1}(E_1\left)\right)}^2+{\left(L_{b-1}\right(E_2)-L_{b+1}(E_2\left)\right)}^2}$

Optimum image intelligent body L is selected according to maximum crowding distance principle to crowding distance_BestAnd export image intelligent body correspondence Optimum cluster label；

Step 11 carries out pixel classifications according to optimum cluster label to input picture, using similar pixel as a block of pixels, All block of pixels composition segmentation figure pictures, export the segmentation result of input picture.

2. the image partition method according to claim 1 based on Multiobjective Intelligent body evolution clustering algorithm, its feature exists In the calculating of image intelligent physical efficiency amount, is carried out as follows in step 3：

(3a) calculates image intelligent body L_yENERGY E rengyL_y=(E₁,E₂) in the component E of energy 1₁,

$E_1=\frac{1}{\underset{j=1}{\overset{c}{\&Sigma;}}\underset{x\&Element;S_j}{\&Sigma;}\left|\right|x-z_j|{|}^2}$

X is the half-tone information of input picture, S in formula_jFor correspondence cluster set, z_jFor the cluster centre of correspondence cluster set；

(3b) calculates image intelligent body L_yENERGY E rengyL_y=(E₁,E₂) in the component E of energy 2₂,

I is the label of ith pixel point in formula, and m is the number of pixel, and j is the mark of the nearest neighbor pixels point of ith pixel point Number, s is nearest neighbor pixels point number, x_i,jFor the relation value of j-th of nearest neighbor pixels point of ith pixel point, wherein when i-th Individual pixel and j-th of pixel then value 0 when belonging to same cluster set, are not belonging to same cluster set and then take 1/j.

3. the image partition method according to claim 1 based on Multiobjective Intelligent body evolution clustering algorithm, its feature exists In the non-dominated ranking operation in step 4 is carried out as follows：

(4a) initiation parameter, order dominates intelligent body L_yIntelligent body number n_Ly=0；

(4b) updates n as follows_Ly：

For image intelligent body L_y, there is image intelligent body L_i, i refers to the numbering of image intelligent body, i=1,2 ..., L_size×L_size, And L_iCan not be L_yIf, image intelligent body L_yEnergy meetThen image intelligent body L_yBy image intelligent body L_iDominated, make n_Ly=n_Ly+ 1, if meetingThen claim image intelligent body L_yDominating figure is as intelligent body L_i；

(4c) is if image intelligent body L_yN_Ly=0, then make image intelligent body L_yInto first layer, it is saved in set level1, has n_level1Individual, if its n_Ly≠ 0, make L_yInto the second layer, it is saved in level2.

4. the image partition method according to claim 1 based on Multiobjective Intelligent body evolution clustering algorithm, its feature exists In the neighborhood competition operator in step 5, is to produce an image intelligent body by one of the following two kinds strategyStrategy 1：Press Formula produces new image intelligent body

Wherein,x _pFor the lower bound of element value in all image intelligent bodies,For the upper bound of element value in all image intelligent bodies, m_p For image intelligent bodyIn element be located at p at element value, l_pFor image intelligent body L_ii,jjIn be located at p at element value, R (- 1,1) be -1 to+1 between random number；

Strategy 2, produces new image intelligent body as follows

The first step, as the following formula willIn all elements value m_pIt is mapped on interval [0,1], obtains new element value:Register map is constituted as intelligent body with these element values

$L_{ii,jj}^{{\mathrm{new}}^{\&prime;}}=(m_1^{\&prime;},m_2^{\&prime;},...,m_{i_1-1}^{\&prime;},m_{i_2-1}^{\&prime;},...,m_{i_1+1}^{\&prime;},m_{i_1}^{\&prime;},m_{i_2+1}^{\&prime;},m_{i_2+2}^{\&prime;},...,m_p^{\&prime;}),$

Wherein, 1<i₁<C, 1<i₂<C, i₁<i₂；

Second step, according to following formula by image intelligent bodyMap back intervalOn, obtain new image intelligent body

$e_p={\underset{\&OverBar;}{x}}_p+m_p^{\&prime;}({\stackrel{\&OverBar;}{x}}_p-{\underset{\&OverBar;}{x}}_p)$

Two kinds of Different Strategies are according to competition probability P₀Selected：

By 0 to 1 random number R (0,1) and competition probability P₀Compare, if R (0,1) ＞ P₀, then selection strategy 1, otherwise, is selected Select strategy 2.

5. the image partition method according to claim 1 based on Multiobjective Intelligent body evolution clustering algorithm, its feature exists In the Gaussian mutation operator in step 6 is to produce a new variation image intelligent body by following formula

Wherein, l_pFor intelligent body L_ii,jjIn be located at p at element value；G (0,1/t) is the random number of Gaussian Profile.

6. the image partition method according to claim 1 based on Multiobjective Intelligent body evolution clustering algorithm, its feature exists In the self-learning operator in step 8 refers to produce a new image intelligent body as follows：

(8a) produces a self study image intelligent body existence grid sL, its size using the method for image intelligent volume mesh generation For sL_size×sL_size, sL_sizeFor integer, all image intelligent body sL thereon_i′,j′, i '=1,2 ..., sL_size, j '= 1,2,...,sL_sizeProduced according to following formula：

Wherein,x _pFor the lower bound of element value in all image intelligent bodies；For the upper bound of element value in all image intelligent bodies；l_p For intelligent body L_aIn be located at p locate element value, R (1-sR, 1+sR) expression 1-sR to 1+sR between random number, sR ∈ [0,1] Represent search radius；

(8b) neighborhood is competed into operator and mutation operator iteration is acted on self study image intelligent body existence grid sL, and maximum changes Number is sG from generation to generation, and present image intelligence is substituted with the maximum image intelligent body of energy in self study image intelligent body existence grid sL Body L_a。