CN105976379A - Fuzzy clustering color image segmentation method based on cuckoo optimization - Google Patents

Abstract

The invention discloses a fuzzy clustering color image segmentation method based on cuckoo optimization. An image to be segmented is input, and color feature is extracted from the image; a cuckoo algorithm is used to optimize a clustering center of a fuzzy clustering algorithm; an improved fuzzy clustering algorithm is used to cluster pixel points in a color space of the image; the clustering center is output, and a membership degree matrix is calculated; and pixels of the image are divided according to the output clustering center and the membership degree matrix, and the image is segmented. An HSV color space which is suitable for sensing of human eyes, the segmentation effect can be improved, an iteration process that the cuckoo algorithm is used to optimize the fuzzy clustering center is provided for overcoming the defect that a traditional fuzzy clustering algorithm tends to fall into the optimum, the operation speed and the convergence speed of the clustering algorithm are improved, the problem that the initial value of the clustering center has much influence on the clustering algorithm, and the clustering effect is good.

Description

A kind of fuzzy clustering color image segmentation method optimized based on cuckoo

Technical field

The present invention relates to technical field of image processing, be specifically related to a kind of fuzzy clustering cromogram optimized based on cuckoo As dividing method.

Background technology

Image segmentation is a vital technology in technical field of image processing, because it is as image procossing Prior step, the quality of segmentation directly influences the result of subsequent processes, such as feature extraction, target recognition etc..Image divides The scope cut is more and more extensive, such as communication, military affairs, remote Sensing Image Analysis, medical diagnosis, intelligent bus, agricultural modernization and work The numerous areas such as industry automatization all be unable to do without the figure of image segmentation.The most either at practical application area or academic neck Territory, image segmentation is all a forward position and far reaching problem.

For the segmentation of coloured image, we to choose suitable color space.RGB (Red, Green, Blue) is a kind of Conventional color space representation, but its widely different with human eye perception.Therefore consider to be transformed into from rgb color space HSV (Hue, Saturation, Value) space.HSV color space is made up of tone, saturation and brightness.Wherein, tone Being referred to as colourity with saturation, it had both illustrated the wavelength components distribution of color, added the hue of clear this color light. These more meet the aware space of human eye.

Fuzzy C-Means Clustering Algorithm (fuzzy c-means algorithm, FCM) is to be subordinate to by ceaselessly iterative computation Genus degree matrix and bunch center, solve the process of object function minima.This algorithm has simply, processing speed fast, without supervision etc. Plurality of advantages.In recent years, a popular application is fuzzy clustering to be applied on image is split.Its main thought be by These data points as a data point, are then divided in different bunches, it is possible to obtain the result of segmentation by each pixel. Such as Patent No. CN103150731B, the patent utilization K-means of entitled " a kind of fuzzy clustering image partition method " is calculated Initial pictures is clustered by method, it is thus achieved that K cluster centre；Again using K cluster centre of acquisition as Fuzzy C-Means Clustering Image is clustered by the initial cluster center of algorithm again, it is achieved the segmentation of image, and the method solves tradition to a certain extent The problem that the computation complexity that randomly selects initial cluster center in Fuzzy C-Means Clustering Algorithm and cause is high.But it is traditional FCM algorithm is affected relatively big by initial point, and is easily trapped into local optimum, it is therefore desirable to be optimized it.

Cuckoo algorithm (Cuckoo Search is called for short CS) is a kind of swarm intelligence algorithm based on Levy flights.It Some solve generations by progressivelying reach optimum around levy migration, thus accelerate Local Search.By partially The a part of solution randomly generated from remote position is remote from current optimal solution, and the system that so may insure that is not absorbed in Excellent solution.This advantage of cuckoo algorithm just can solve FCM algorithm and be easily trapped into the defect of local optimum.

Summary of the invention

The technical problem to be solved in the present invention is to overcome present in above-mentioned traditional fuzzy clustering algorithm segmentation image technique Affected relatively big by initial point, the problem being easily trapped into local optimum.

In consideration of it, the technical scheme that the present invention proposes is a kind of fuzzy clustering color images optimized based on cuckoo Method, comprises the steps of

Step 101: input image to be split, extracts the color character of this image；

Step 102: utilize cuckoo algorithm that the cluster centre of fuzzy clustering algorithm is optimized；

Step 103: the fuzzy clustering algorithm of application enhancements, clusters pixel in the color space of image；

Step 104: according to the cluster centre of step 103 output, calculates subordinated-degree matrix；

Step 105: cluster centre and the calculated subordinated-degree matrix of step 104 according to step 103 output are to image Pixel divide, thus realize the segmentation of image.

As preferably, in a step 101, according to equation below, the color value of image is transformed into HSV from rgb space empty Between:

$H_1={\cos }^{-1}\left\{\frac{0.5\[R-G+(R-B)\]}{\sqrt{{(R-G)}^2+(R-G)(G-B)}}\right\}$

$V=\frac{max(R,G,B)}{255}$

$V=\frac{max(R,G,B)}{255}$

Wherein H represents that tone, S represent that saturation, V represent brightness.

Further, step 102 comprises the steps:

Step 301: parameter initialization, including scale N of cuckoo algorithm_max, Bird's Nest finds the general of external cuckoo bird egg Rate P, number n of cluster centre, algorithm Termination Threshold G；

Step 302: first with the cluster centre that traditional FCM stochastic generation is initial, as the initial position of bird's nest；

Step 303: utilize the cluster centre that cuckoo algorithm optimization step 302 generates, final by not stopping grey iterative generation Cluster centre；

Step 304: utilize the final cluster centre that step 303 generates, calculates FCM object function；

Step 305: if meeting given maximum iteration time G, algorithm terminates, picture is finally split in output, otherwise goes to Step 302.

Further, described step 303 comprises the steps:

Step 401: select fitness function, calculate the target function value of each bird's nest, select optimal value therein, target Function formula is:

$J(U,A)={\mathrm{\Σ}}_{j=1}^m{\mathrm{\Σ}}_{i=1}^n{u}_{ij}^b{d}_{ij}^2$

Wherein u_ijRepresent that i-th sample point belongs to the degree of membership of jth class, u_ij∈ [0,1], has for arbitrary i

d_ij=| | X_i-A_i| | it is sample x_iTo the Euclidean distance of jth cluster centre, b represents fuzzy finger Number,

General 1 ＜ b≤5；

Step 402: retain the optimal value calculated in step 401, and utilize equation below to be moved the position of bird's nest Dynamic:

$x_i^{(t+1)}=x_i^{\left(t\right)}+\mathrm{\α}\⊕L\left(\mathrm{\λ}\right),i=1,2,...,n$

WhereinRepresent the i-th bird's nest bird's nest position in t generation,Representing point-to-point multiplication, α represents step size controlling Amount, its value Normal Distribution, L (λ) is Levy random search path, and arbitrary width is Levy distribution, L (s, λ)～s^-λ,s It is that levy flies the arbitrary width that obtains；

Step 403: according to the target function value that new bird's nest position calculation is new, then with current target function value ratio Relatively, if new value is better than current, just with new replacement currency；

Step 404；After step 402 obtains new position, generate a random number r (r ∈ [0,1]), and compare with P Relatively, if r > P, bird's nest position of the most random change, the most constant, finally retain desired positions；

Step 405: if iterations exceedes algorithm scale, just export operation result, otherwise forward step 402 to.

Especially, step 304 comprises the steps:

Step 501: utilize equation below to calculate subordinated-degree matrix,

$u_{ij}=\frac{{(1/||X_j-A_i|{|}^2)}^{1/(m-1)}}{{\mathrm{\Σ}}_{i=1}^m{(1/||X_j-A_i|{|}^2)}^{1/(m-1)}}$

This formula is solved by object function introducing Lagrange multiplier and obtains；

Step 502: utilize equation below to calculate bunch central point,

$A_i=\frac{{\mathrm{\Σ}}_{j=1}^n{u}_{ij}^m{X}_j}{{\mathrm{\Σ}}_{j=1}^n{u}_{ij}}$

This formula is solved by object function introducing Lagrange multiplier and obtains；

Step 503: utilize equation below to calculate FCM target function value,

$J(U,A)={\mathrm{\Σ}}_{j=1}^m{\mathrm{\Σ}}_{i=1}^n{u}_{ij}^b{d}_{ij}^2$

Wherein u_ijRepresent that i-th sample point belongs to the degree of membership of jth class, u_ij∈ [0,1], has for arbitrary i d_ij=| | X_i-A_i| | it is sample x_iTo the Euclidean distance of jth cluster centre, b represents Fuzzy Exponential, span be 1 ＜ b≤ 5。

Beneficial effect: first the present invention chooses the color space HSV of suitable human eye perception, is so conducive to improving segmentation Effect, is then easily trapped into the defect of local optimum for traditional fuzzy clustering algorithm, proposes to utilize cuckoo algorithm optimization The iterative process at fuzzy clustering center, ultimately generates the cluster centre optimized.This method improves the computing speed of clustering algorithm Degree and convergence rate, efficiently solve the problem that the initial value of cluster centre is excessive on clustering algorithm impact, have good Clustering Effect.

Accompanying drawing explanation

Fig. 1 is the flow chart of the present invention.

Detailed description of the invention

In conjunction with accompanying drawing, the present invention is further detailed explanation.

As it is shown in figure 1, a kind of fuzzy clustering color image segmentation method optimized based on cuckoo that the present invention proposes, bag Include following steps:

Step one: input image to be split, then extracts the color character of this image, here by the color value of image from Rgb space is transformed into HSV space.Wherein H represents that tone, S represent that saturation, V represent brightness.

Conversion is carried out according to equation below:

$H_1={\cos }^{-1}\left\{\frac{0.5\[R-G+(R-B)\]}{\sqrt{{(R-G)}^2+(R-G)(G-B)}}\right\}$

$S=\frac{max(R,G,B)-min(R,G,B)}{max(R,G,B)}$

$V=\frac{max(R,G,B)}{255}$

Step 2: c cluster centre of random initializtion, and using them as bird's nest position.First parameter is initial Change, including scale N of cuckoo algorithm_max, Bird's Nest finds the probability P of external cuckoo bird egg, number n of cluster centre, algorithm Termination Threshold G.Then the cluster centre that traditional FCM stochastic generation is initial is utilized, as the initial position of bird's nest.

Step 3: according to the cluster centre calculating target function value generated, and compare and draw wherein optimal bird's nest position.? Function to be chosen a fitting goal, we select equation below as object function:

$J(U,A)=\underset{j=1}{\overset{m}{\Σ}}\underset{i=1}{\overset{n}{\Σ}}{u}_{ij}^b{d}_{ij}^2$

Wherein u_ijRepresent that i-th sample point belongs to the degree of membership of jth class, u_ij∈ [0,1], has for arbitrary i d_ij=| | X_i-A_i| | it is sample x_iTo the Euclidean distance of jth cluster centre, b represents Fuzzy Exponential, general 1 ＜ b≤5。

Step 4: update bird's nest position.Here it is that optimal bird's nest position step 3 calculated moves, mobile Mode is carried out according to equation below:

$x_i^{(t+1)}=x_i^{\left(t\right)}+\mathrm{\α}\⊕L\left(\mathrm{\λ}\right),i=1,2,...,n$

WhereinRepresent the i-th bird's nest bird's nest position in t generation,Representing point-to-point multiplication, α represents step size controlling Amount, its value Normal Distribution, L (λ) is Levy random search path, and arbitrary width is Levy distribution, L (s, λ)～s^-λ,s It is that levy flies the arbitrary width that obtains.

Step 5: according to the target function value that the bird's nest position calculation after updating is new, then with current target function value Relatively, if new value is better, just with new replacement currency than current.

Step 6: generate a random number r (r ∈ [0,1]), and compare with P, if r > P, the most random change is once Bird's nest position, the most constant, finally retain desired positions.

Step 7: if iterations exceedes algorithm scale, just export operation result.Otherwise, step 3 is forwarded to.

Step 8: according to the cluster centre that optimum bird's nest position calculation is final.Utilize equation below calculate cluster centre:

$A_i=\frac{{\mathrm{\Σ}}_{j=1}^n{u}_{ij}^m{X}_j}{{\mathrm{\Σ}}_{j=1}^n{u}_{ij}}$

This formula is solved by object function introducing Lagrange multiplier and obtains.

Step 9: divide pixel, output segmentation image according to cluster centre.

Claims (5)

1. the fuzzy clustering color image segmentation method optimized based on cuckoo, it is characterised in that comprise the steps of

Step 101: input image to be split, extracts the color character of this image；

Step 102: utilize cuckoo algorithm that the cluster centre of fuzzy clustering algorithm is optimized；

Step 103: the fuzzy clustering algorithm of application enhancements, clusters pixel in the color space of image；

Step 104: according to the cluster centre of step 103 output, calculates subordinated-degree matrix；

Step 105: according to cluster centre and the step 104 calculated subordinated-degree matrix picture to image of step 103 output Element divides, thus realizes the segmentation of image.

A kind of fuzzy clustering color image segmentation method optimized based on cuckoo the most according to claim 1, its feature It is: in a step 101, according to equation below, the color value of image is transformed into HSV space from rgb space:

$H_1={\cos }^{-1}\left\{\frac{0.5\[R-G+(R-B)\]}{\sqrt{{(R-G)}^2+(R-G)(G-B)}}\right\}$

$V=\frac{max(R,G,B)}{255}$

$V=\frac{max(R,G,B)}{255}$

Wherein H represents that tone, S represent that saturation, V represent brightness.

3., according to a kind of fuzzy clustering color image segmentation method optimized based on cuckoo described in claims 1, it is special Levy and be that step 102 comprises the steps:

Step 301: parameter initialization, including scale N of cuckoo algorithm_max, Bird's Nest finds the probability P of external cuckoo bird egg, Number n of cluster centre, algorithm Termination Threshold G；

Step 302: first with the cluster centre that traditional FCM stochastic generation is initial, as the initial position of bird's nest；

Step 303: utilize the cluster centre that cuckoo algorithm optimization step 302 generates, by not stopping final the gathering of grey iterative generation Class center；

Step 304: utilize the final cluster centre that step 303 generates, calculates FCM object function；

Step 305: if meeting given maximum iteration time G, algorithm terminates, picture is finally split in output, otherwise goes to step 302。

4., according to a kind of fuzzy clustering color image segmentation method optimized based on cuckoo described in claims 3, it is special Levy and be that described step 303 comprises the steps:

Step 401: select fitness function, calculate the target function value of each bird's nest, select optimal value therein, object function Formula is:

$J(U,A)={\mathrm{\Σ}}_{j=1}^m{\mathrm{\Σ}}_{i=1}^n{u}_{ij}^b{d}_{ij}^2$

Wherein u_ijRepresent that i-th sample point belongs to the degree of membership of jth class, u_ij∈ [0,1], has for arbitrary i d_ij=| | X_i-A_i| | it is sample x_iTo the Euclidean distance of jth cluster centre, b represents Fuzzy Exponential, general 1 ＜ b≤5；

Step 402: retain the optimal value calculated in step 401, and utilize equation below to be moved the position of bird's nest:

$x_i^{(t+1)}=x_i^{\left(t\right)}+\mathrm{\α}\⊕L\left(\mathrm{\λ}\right),i=1,2,...,n$

WhereinRepresent the i-th bird's nest bird's nest position in t generation,Representing point-to-point multiplication, α represents step size controlling amount, it Value Normal Distribution, L (λ) is Levy random search path, arbitrary width be Levy distribution, L (s, λ)～s^-λ, s is levy The arbitrary width that flight obtains；

Step 403: according to the target function value that new bird's nest position calculation is new, then compares with current target function value, if New value is better than current, just with new replacement currency；

Step 404；After step 402 obtains new position, generate a random number r (r ∈ [0,1]), and compare with P, If r > P, bird's nest position of the most random change, the most constant, finally retain desired positions；

Step 405: if iterations exceedes algorithm scale, just export operation result, otherwise forward step 402 to.

5., according to a kind of fuzzy clustering color image segmentation method optimized based on cuckoo described in claims 3, it is special Levy and be that step 304 comprises the steps:

Step 501: utilize equation below to calculate subordinated-degree matrix,

$u_{ij}=\frac{{(1/||X_j-A_i|{|}^2)}^{1/(m-1)}}{{\mathrm{\Σ}}_{i=1}^m{(1/||X_j-A_i|{|}^2)}^{1/(m-1)}}$

This formula is solved by object function introducing Lagrange multiplier and obtains；

Step 502: utilize equation below to calculate bunch central point,

$A_i=\frac{{\mathrm{\Σ}}_{j=1}^n{u}_{ij}^m{X}_j}{{\mathrm{\Σ}}_{j=1}^n{u}_{ij}}$

This formula is solved by object function introducing Lagrange multiplier and obtains；

Step 503: utilize equation below to calculate FCM target function value,

$J(U,A)={\mathrm{\Σ}}_{j=1}^m{\mathrm{\Σ}}_{i=1}^n{u}_{ij}^b{d}_{ij}^2$

Wherein u_ijRepresent that i-th sample point belongs to the degree of membership of jth class, u_ij∈ [0,1], has for arbitrary i d_ij=| | X_i-A_i| | it is sample x_iTo the Euclidean distance of jth cluster centre, b represents Fuzzy Exponential, span be 1 ＜ b≤ 5。