CN107368851B - A Fast Fuzzy C-Means Clustering Image Segmentation Method with Neighborhood Selection Strategy - Google Patents

Abstract

The invention discloses a fast fuzzy C-means clustering image segmentation method with neighborhood selection strategy. It is introduced into the calculation process of local similarity and new graph ξ, and the particle swarm algorithm is used to obtain the appropriate clustering center, which improves the traditional fuzzy C-means clustering method, so that the image segmentation details are better protected.

Description

A Fast Fuzzy C-Means Clustering Image Segmentation Method with Neighborhood Selection Strategy

Technology neighborhood

The invention relates to an image segmentation neighborhood, in particular to a fast fuzzy C-means clustering image segmentation method with a neighborhood selection strategy.

Background technique

Image detail protection has always been one of the key and difficult points of image segmentation. The traditional Fast Fuzzy C-Means Clustering Image Segmentation (FGFCM) does not have a good effect of preserving edges, and does not take into account the selection of window neighborhoods and edge information. relationship, only select a fixed-size window and its neighbor points corresponding to the target pixel.

SUMMARY OF THE INVENTION

In order to overcome the shortcomings and deficiencies of the prior art, the present invention provides a fast fuzzy C-means clustering image segmentation method with a neighborhood selection strategy.

The present invention mainly utilizes the edge extraction of the image, and the extracted edge information is used to select a window of suitable size and its neighborhood, the window neighborhood selection strategy is introduced into the calculation process of the local similarity and the new graph ξ, and the particle swarm algorithm is adopted. (PSO) obtains the appropriate clustering center, improves the traditional fuzzy C-means clustering method, and makes the image segmentation details better protected.

The present invention adopts following technical scheme:

A fast fuzzy C-means clustering image segmentation method with neighborhood selection strategy, comprising the following steps:

S1 inputs the image to be segmented, and extracts the edge of the image;

S2 generates a new graph ξ based on the neighborhood selection strategy of the image edge;

S3 obtains the initial cluster center according to the generated new graph;

S4 fast fuzzy C-means clustering obtains the best clustering partition.

The S2 generates a new graph ξ based on the neighborhood selection strategy of the image edge, specifically:

S2.1 Window neighborhood selection based on image edges;

S2.2 combines spatial and gray level information to obtain a local similarity measure. The calculation formula is as follows:

i pixel is the center of the local window, k pixel represents the pixel in the window neighborhood of i pixel, the window neighborhood is obtained based on the window neighborhood selection strategy described in S2.1, p _i , q _i are the pixel i coordinates, x _i is the gray value of the window neighborhood, λ _s and λ _g are two scale factors;

_σi is defined as:

S2.3 Calculate and generate a new graph ξ

ξ is calculated as follows

Among them, ξ _i represents the gray value of the ith pixel in the image ξ, x _k represents the gray value of the pixel in the neighborhood of _xi in the original image, and the window neighborhood is based on the window neighborhood selection strategy described in S2.1 Obtained, N _i is the neighborhood set of _xi and S _ik is the local similarity measure between the ith pixel and the kth pixel.

S2.1 performs window neighborhood selection based on image edges, specifically:

Set the initial window size to 5*5, if no edge falls in the window, select the window as the local window, and the pixels in the window are the target pixel neighborhood;

If there is an edge in the window, expand the window to 7*7, and select the pixel on the same side as the edge of the target pixel as the neighborhood.

The _λs and _λg are both set to 2.

The particle swarm algorithm was used to obtain the initial cluster center parameters.

The S4 fast fuzzy C-means clustering obtains the best clustering division, and specifically makes the J value reach the minimum to obtain the best clustering division

Among them, c _j is the center of the jth class, u _ij is the membership degree of the pixel with the gray value i belonging to the jth class, M is the gray level of the graph ξ, γ _i is the pixel with the same gray value as i The number of points, m is the fuzzy index factor, ξ _i represents the gray value of the ith pixel of the graph ξ;

The iterative relationship between u _ij and c _j is as follows:

The specific process is:

S4.1 Initialization, set the fuzzy index factor m, the initial iteration count b2, the maximum iteration number t2, the iteration threshold ε parameter, set the initial value of the cluster membership matrix U ⁽⁰⁾ = U ^best , the initial cluster center

S4.2 uses c _j to update the class center c _j , and uses u _ij to update the membership matrix U(b+1);

S4.3 If max{U(b)-U(b+1)}<ε or b2>t2, the iteration stops, otherwise, b=b+1, and go to step S4.2.

The fuzzy index factor m is set to 2, the iteration count is initialized to b2=0, the maximum iteration number t2 is set to 100 times, and the iteration termination threshold ε is set to 1e-5.

Beneficial effects of the present invention:

The invention adds an edge-based window neighborhood selection strategy to the traditional fast fuzzy C-means clustering image segmentation method, which enhances the detail protection of image segmentation.

Description of drawings

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

Fig. 2 is the flow chart of search class center of the present invention;

Fig. 3 is the flow chart of window neighborhood selection strategy;

Fig. 4 is a clustering flow chart;

Fig. 5(a) is the original image, Fig. 5(b) is the standard image, Fig. 5(c) is the effect image using the traditional method FGFCM, and Fig. 5(d) is the image segmented by this method.

Detailed ways

The present invention will be described in further detail below with reference to the embodiments and the accompanying drawings, but the embodiments of the present invention are not limited thereto.

Example

As shown in Figures 1-4, a fast fuzzy C-means clustering image segmentation method with a neighborhood selection strategy includes the following steps:

S1 inputs the image to be segmented, and extracts the edge of the image;

Because the Canny edge detection operation is more convenient and the effect is better, the Canny edge detection is selected to extract the image edge.

S2 generates a new graph ξ based on the neighborhood selection strategy of the image edge, specifically:

S2.1 selects the neighborhood through the relationship between the window and the edge, which can enhance the protection of image details. The specific strategy for window neighborhood selection is as follows:

(1) Set the initial window size to 5*5. If no edge falls within the window, select the window as a local window, and the pixels in the window are the target pixel neighborhood;

(2) Set the initial window size to 5*5. If there is an edge in the window, expand the window to 7*7, and select the pixel on the same side as the edge of the target pixel as the neighborhood.

S2.2 Calculate local similarity;

The local similarity measure is obtained by combining the spatial and gray level information. The calculation formula is as follows, and the scale factors λ _s and λ _g in the formula are set;

i pixel is the center of the local window, k pixel represents the pixel in the window neighborhood of i pixel, the window neighborhood is obtained based on the window neighborhood selection strategy described in S2.1, p _i , q _i are the pixel i coordinates, x _i is the gray value of the window neighborhood, λ _s and λ _g are two scale factors;

_σi is defined as:

In formula (2), N _i is the set of window neighborhood pixels of i pixel, _NR is the total number of neighborhood pixels in the i pixel window, and the window neighborhood is obtained based on the window neighborhood selection strategy described in step 1.

In the embodiment of the present invention, both λ _s and λ _g are set to 2.

S2.3 Calculate and generate a new graph ξ

ξ is calculated as follows

Among them, ξ _i represents the gray value of the i-th pixel of the graph ξ. x _k represents the gray value of the pixel in the neighborhood of _xi in the original image, the window neighborhood is obtained based on the window neighborhood selection strategy described in step 1, and N _i is the neighborhood set of _xi . S _ik is the local similarity measure between the ith pixel and the kth pixel.

S2.4 Obtain the initial cluster center according to the generated new graph;

Selecting different cluster centers will have different effects on image segmentation, and in order to avoid falling into local optimum, the present invention uses particle swarm algorithm to obtain better initial cluster center parameters, and particle swarm algorithm updates particles according to equations 7 and 8. position and speed.

v _ij =w · v _ij +c ₁ ·rand1 _ij ·(pbest _ij -x _ij )+c ₂ ·rand2 _ij ·(gbest _j -x _ij ) (7)

x _ij = x _ij +v _ij (8)

_{where i} = 1, 2, . Position, pbes _ij and gbest _j are the best position of the i-th particle in j cycles and the global best particle position of j cycles respectively; rand1ij and rand2ij are random numbers between (0, 1); c1 and c2 are learning factors, and w is an inertia factor.

The evaluation criterion of the particle swarm algorithm is represented by a fitness function. The fitness function used in the present invention is set as shown in the following formula 9. When the fitness value is larger, it means that the particle position is better.

Fitness=1/(1+J) (9)

As shown in Figure 2, the initial cluster center search process, the specific search process is as follows:

S2.4.1 Parameter setting

Set the population size, population size, learning factors c1 and c2, the highest iteration number t1, iteration count initialization b1, and the number of clusters c.

In Fig. 5 of the simulation example of the present invention, the population size is set to 50, the population size is set to 0.5, the learning factors c1 and c2 are both set to 0.5, the maximum iteration number t1 is 1000 times, the iteration count initialization b1=1, the number of clusters c is set to 3.

S2.4.2 Randomly set the starting position of particles

S2.4.3 Calculate the membership degree of each particle according to the formula 5, calculate the fitness according to the formula 9, find the optimal position of each particle in the b1 cycle and the global optimal particle position in the b1 cycle, and use the formulas 10 and 11 to update the particle position .

S2.4.4 If the number of iterations b1<t1, set the number of iterations b1=b1+1, and continue with step (3); when b1≥t1, that is, when the number of iterations reaches the limit of the maximum number of iterations, stop the iteration, and the obtained global optimum particle

Location as initial cluster center for fast fuzzy C-means clustering

And obtain its membership degree matrix U ^best by formula 5.

S4 fast fuzzy C-means clustering to obtain the best clustering division;

Based on the new graph, fast fuzzy C-means clustering obtains the best clustering partition by minimizing Equation 4.

Among them, cj is the center of the jth class, uij is the membership degree of the pixel with the gray value i belonging to the jth class, M is the gray level of the graph ξ, γ _i is the number of pixels with the same gray value as i, m is the fuzzy exponent factor, and the iterative relationship between uij and cj is shown in Equations 5 and 6 below:

The specific steps of clustering are as follows:

(1) Initialization

Set the fuzzy index factor m, the initial iteration count b2, the maximum iteration number t2, and the iteration threshold ε parameters in formulas (4), (5), and (6). On the basis of the new graph obtained in step 3, U ^best and c ^vest obtained in step 4 are used to set the initial value of the cluster membership matrix U ⁽⁰⁾ = U ^best , the initial cluster center

In the present invention, the blur index factor m is set to 2. The iteration count is initialized with b2=0, the maximum iteration number t2 is set to 100 times, and the iteration termination threshold ε is set to 1e-5.

(2) The class center c _j is updated according to formula 6, and the membership degree matrix U(b+1) is updated by formula 5.

(3) If max{U(b)-U(b+1)}<ε or b2>t2, the iteration stops, otherwise, b=b+1, continue to step (2)

The standard for judging the effect of image segmentation, namely the calculation formula of segmentation accuracy (SA), compared with the traditional method, it shows that the accuracy of this method is higher. The specific results are shown in Table 1, which is the average result of 10 experiments.

In the formula, C is the number of clusters, Aj is the set of pixels belonging to the Jth category found by a certain segmentation method, and Cj is the set of pixels belonging to the Jth category in the standard image. If the value of SA is higher, It means that the segmentation effect is better. Ideally, this value is 1.

Table 1

FGFCM The method of this paper 0.9770 0.9820

Using the method of the present invention to perform image segmentation matlab simulation effect diagram, Figure 5 (a) the original image, Figure 5 (b) is the standard image, Figure 5 (c) is the effect image of using the traditional method FGFCM, Figure 5 (d) ) is the picture segmented by this method, and it can be found that the image segmentation effect of the method of the present invention is better.

The above-mentioned embodiments are preferred embodiments of the present invention, but the embodiments of the present invention are not limited by the described embodiments, and any other changes, modifications, substitutions, and combinations made without departing from the spirit and principle of the present invention , simplification, all should be equivalent replacement modes, and are all included in the protection scope of the present invention.

Claims (7)

1. a fast fuzzy C-means clustering image segmentation method with neighborhood selection strategy, is characterized in that, comprises the steps: S1 inputs the image to be segmented, and extracts the edge of the image; S2 generates a new graph ξ based on the neighborhood selection strategy of the image edge; S3 obtains the initial cluster center according to the generated new graph; S4 fast fuzzy C-means clustering to obtain the best clustering division; The S2 generates a new graph ξ based on the neighborhood selection strategy of the image edge, specifically: S2.1 Window neighborhood selection based on image edges; S2.2 combines spatial and gray level information to obtain a local similarity measure. The calculation formula is as follows:

i pixel is the center of the local window, k pixel represents the pixel in the window neighborhood of i pixel, the window neighborhood is obtained based on the window neighborhood selection strategy described in S2.1, p _i , q _i are the pixel i coordinates, x _i is the gray value of the window neighborhood, λ _s and λ _g are two scale factors; _σi is defined as:

Among them, _NR is the total number of neighbor pixels in the i pixel window; S2.3 Calculate and generate a new graph ξ ξ is calculated as follows

Among them, ξ _i represents the gray value of the ith pixel in the image ξ, x _k represents the gray value of the pixel in the neighborhood of _xi in the original image, N _i is the neighborhood set of _xi , and S _ik is the ith pixel. and the local similarity measure between the kth pixel. 2. fast fuzzy C-means clustering image segmentation method according to claim 1, is characterized in that, S2.1 carries out window neighborhood selection based on image edge, is specifically: Set the initial window size to 5*5, if no edge falls in the window, select the window as the local window, and the pixels in the window are the target pixel neighborhood; If there is an edge in the window, expand the window to 7*7, and select the pixel on the same side as the edge of the target pixel as the neighborhood. 3 . The fast fuzzy C-means clustering image segmentation method according to claim 1 , wherein the λ _s and λ _g are both set to 2. 4 . 4 . The fast fuzzy C-means clustering image segmentation method according to claim 1 , wherein the initial cluster center parameter is obtained by adopting a particle swarm algorithm. 5 . 5. fast fuzzy C-means clustering image segmentation method according to claim 1, is characterized in that, described S4 fast fuzzy C-means clustering obtains optimal clustering division, specifically makes J value reach minimum to obtain optimal clustering divide

Among them, c _j is the center of the jth class, u _ij is the membership degree of the pixel with the gray value i belonging to the jth class, M is the gray level of the graph ξ, γ _i is the pixel with the same gray value as i The number of points, m is the fuzzy index factor, ξ _i represents the gray value of the ith pixel of the graph ξ; The iterative relationship between u _ij and c _j is as follows:

6. fast fuzzy C-means clustering image segmentation method according to claim 5, is characterized in that, concrete process is: S4.1 Initialization, set the fuzzy index factor m, the initial iteration count b2, the maximum iteration number t2, the iteration threshold ε parameter, set the initial value of the cluster membership matrix U ⁽⁰⁾ = U ^best , the initial cluster center

S4.2 uses c _j to update the class center c _j , and uses u _ij to update the membership matrix U(b+1); S4.3 If max{U(b)-U(b+1)}<ε or b2>t2, the iteration stops, otherwise, b=b+1, and go to step S4.2. 7 . The fast fuzzy C-means clustering image segmentation method according to claim 6 , wherein the fuzzy index factor m is set to 2, the iteration count initialization b2 = 0, the maximum iteration number t2 is set to 100 times, and the iteration termination threshold is set to 100. 8 . ε is set to 1e-5.

Images