CN105261004A - Mean shift and neighborhood information based fuzzy C-mean image segmentation method - Google Patents

Abstract

The invention discloses a mean shift and neighborhood information based fuzzy C-mean image segmentation method, and mainly solves the problems of low segmentation accuracy and poor robustness of an existing image segmentation method. The method comprises the steps of (1) inputting a to-be-segmented image; (2) calculating a cluster number and an initial cluster center by adopting a mean shift algorithm; (3) performing initialization; (4) calculating a weight of an neighborhood image block in the to-be-segmented image; (5) calculating a weighted fuzzy factor weight of each pixel point in the to-be-segmented image; (6) performing cluster iteration; (7) judging whether an iterative stop condition is met or not; and (8) generating a segmented image. According to the method, the neighborhood information of the image is fully utilized, so that the method is high in noise robustness and the accuracy of image segmentation is greatly improved.

Description

Based on the fuzzy C-mean algorithm image partition method of average drifting and neighborhood information

Technical field

The invention belongs to technical field of image processing, further relate to a kind of fuzzy C-mean algorithm image partition method based on average drifting and neighborhood information in technical field of image segmentation.The present invention obtains initial clusters number and cluster centre by mean shift process, and in cluster iteration, use the degree of membership of neighborhood information to pixel smoothing, realizes the segmentation to image, can be used for the extraction to characteristics of image target.

Background technology

Fuzzy cluster analysis is one of major technique of data mining, and wherein fuzzy C-means clustering method is a kind of most widely used fuzzy clustering method.Fuzzy clustering being applied to Iamge Segmentation is in recent years in a popular research direction in Iamge Segmentation field.The process of Iamge Segmentation is exactly using each pixel as a data point, and the result of segmentation is for these data points give a class mark.There is same class target pixel and be divided into a class, thus realize the segmentation to image.

Traditional image partition method based on fuzzy clustering, owing to being subject to the impact of initial cluster center, and more responsive to noise spot, cause the precision of Iamge Segmentation low, poor robustness.

Z.Ji, Y.Xiab, Q.Chena, Q.Suna, a kind of image partition method of fuzzy clustering of improvement is proposed in paper " Fuzzyc-meansclusteringwithweightedimagepatchforimagesegm entation " (AppliedSoftComputing121659 – 1667.2012) that D.Xiaa, D.D.Feng deliver at it.The method is compared with the image partition method of traditional fuzzy clustering, its key be make use of pixel neighborhood image block to replace the pixel in traditional fuzzy C-means clustering image partition method, make use of the neighborhood information of image fully, improve the segmentation precision of image.But the method still Shortcomings part is, owing to the process employs the Euclidean distance of non-robust, cause cutting procedure more responsive to noise spot, poor to the robustness of noise.

A kind of fuzzy C-mean algorithm image partition method is disclosed in patent " a kind of fuzzy C-mean algorithm image partition method " (number of patent application 201310072342.X, publication number CN103150731A) that Nanjing Aero-Space University applies at it.The method utilizes K-means algorithm to carry out cluster to initial pictures, obtains K cluster centre; Again the K of an acquisition cluster centre is carried out cluster as the initial cluster center of fuzzy C-means clustering method again to image, realize the segmentation of image.Although the method solves random selecting initial cluster center in traditional fuzzy C means clustering method and makes the high defect of its computation complexity also improve segmentation precision simultaneously.But the weak point that the method still exists is, owing to have employed traditional fuzzy C-means clustering method, do not consider the impact of neighborhood information on cluster process of pixel, cause the unbalanced data set of segmentation Density Distribution that the method can not be correct.

Summary of the invention

The object of the invention is to the deficiency overcoming above-mentioned prior art, propose a kind of fuzzy C-mean algorithm image partition method based on average drifting and neighborhood information.The present invention takes full advantage of the neighborhood information of image to be split, achieves the Accurate classification to image, improves the segmentation effect of image and the robustness to noise.

The basic ideas realizing the object of the invention are: first, use mean shift process to carry out iteration to the pixel of image and obtain initial clusters number and cluster centre; Then, the neighborhood image block of use weighting replaces the pixel in traditional fuzzy C-mean algorithm image partition method, and the Weighted Fuzzy factor is introduced in the objective function of this classic method, consider space length constraint and the spatial-intensity constraint of pixel simultaneously, ensure that the integrality of space neighborhood information; Finally, in cluster iteration, use neighborhood information smoothing to degree of membership, better realize the segmentation to image.

To achieve these goals, specific implementation step of the present invention is as follows:

(1) image to be split is inputted;

(2) adopt mean shift algorithm, calculate clusters number and initial cluster center:

(3) initialization:

Initial cycle number of times is set to 0, maximum iteration time is set to 500, the subordinated-degree matrix of random initializtion image to be split;

(4) weights of neighborhood image block in image to be split are calculated:

(4a) centered by the pixel in image to be split, with 1 pixel unit for radius, obtain the neighborhood image block of 3*3 pixel unit of 9 pixel compositions within the scope of this, obtain the neighborhood image block of all pixels;

(4b) variance yields of pixel gray-scale value in each neighborhood image block is calculated;

(4c) by the variance projection of pixel gray-scale values all in each neighborhood image block to gaussian kernel space;

(4d) by the weights normalization of each neighborhood image block;

(5) the Weighted Fuzzy factor weights of each pixel in image to be split are calculated:

(5a) the space length weights of pixel in pixel and its neighborhood image block in image to be split are calculated;

(5b) the spatial-intensity weights of pixel in pixel and its neighborhood image block in image to be split are calculated;

(5c) the Weighted Fuzzy factor weights of each pixel in image to be split are calculated;

w _i＝w ₁·w ₂

Wherein, w _irepresent the Weighted Fuzzy factor weights of i-th pixel in image to be split, i represents the label of i-th pixel in image to be split, w ₁to represent in image to be split the space length weights of pixel in i-th pixel and its neighborhood image block, w ₂to represent in image to be split the spatial-intensity weights of pixel in i-th pixel and its neighborhood image block;

(6) cluster iteration:

(6a) degree of membership of each pixel in image to be split is calculated, by the degree of membership of obtained all pixels composition subordinated-degree matrix;

(6b) to the smoothing process of degree of membership of each pixel in image to be split;

1st step, according to the following formula, the pixel in calculating neighborhood image block is to the control coefrficient of the central pixel point of neighborhood image block:

$p(x_i,x_j)=\frac{d(x_i,x_j)}{\underset{\left\{j\right\}\∈N_i}{\Σ}d(x_i,x_j)}$

Wherein, p () represents that pixel in neighborhood image block is to the control coefrficient of the central pixel point of neighborhood image block, and its value, more close to 1, shows that the correlativity of pixel in neighborhood image block and central pixel point is larger, x _irepresent the gray-scale value of i-th pixel in image to be split, i represents the label of i-th pixel in image to be split, x _jrepresent the gray-scale value of a jth pixel in the neighborhood image block of i-th pixel, j represents the label of a jth pixel in the neighborhood image block of i-th pixel in image to be split, d () represents the Euclidean distance of two pixels, ∑ represents sum operation, { } represents element set symbol, ∈ represents and belongs to symbol, N _irepresent the set of all pixels in i-th neighborhood of pixel points image block in image to be split;

2nd step, according to the following formula, calculates the weight coefficient of all pixels in each neighborhood image block:

$a_{ij}=\left\{\begin{array}{cc}p(x_i,x_j),& |x_i-x_m|<{\mathrm{\&delta;}}_i\\ 0,& |x_i-x_m|>{\mathrm{\&delta;}}_i\end{array}\right.$

Wherein, a _ijrepresent the weight coefficient of a jth pixel in the neighborhood image block of i-th pixel in image to be split, i represents the label of i-th pixel in image to be split, j represents the label of a jth pixel in the neighborhood image block of i-th pixel in image to be split, pixel in p () expression neighborhood image block is to the control coefrficient of the central pixel point of neighborhood image block, || represent and ask absolute value operation, x _irepresent the gray-scale value of i-th pixel in image to be split, x _mrepresent the mean value of all pixel gray-scale values in the neighborhood image block of i-th pixel in image to be split, δ _irepresent the variance yields of all pixel gray-scale values in i-th neighborhood of pixel points image block in image to be split;

3rd step, according to the following formula, the smoothing process of degree of membership to each pixel in image to be split:

$u_{ki}^{\&prime;}=\underset{\left\{j\right\}\&Element;N_i}{\&Sigma;}{a}_{ij}{u}_{ki}$

Wherein, u ' _kirepresent i-th pixel in image to be split belong to the degree of membership smoothing processing of kth class in initial cluster center after value, ∑ represents summation symbol, i represents the label of i-th pixel in image to be split, j represents the label of a jth pixel in the neighborhood image block of i-th pixel in image to be split, { } represents element set symbol, ∈ represents and belongs to symbol, N _irepresent the set of all pixels in i-th neighborhood of pixel points image block in image to be split, u _kito represent in image to be split that i-th pixel belongs to the degree of membership of kth class in initial cluster center, a _ijrepresent the weight coefficient of a jth pixel in the neighborhood image block of i-th pixel in image to be split, meet a _ij∈ [0,1], a _ijbe used for controlling the size that the pixel in image to be split in i-th neighborhood of pixel points image block affects neighborhood image block central pixel point;

(6c) the image clustering center to be split of current iteration number of times according to the following formula, is calculated:

$v_{2k}=\frac{\underset{i=1}{\overset{n}{\mathrm{\&Sigma;}}}{\left(u_{ki}^{\&prime;}\right)}^m\underset{\left\{j\right\}\&Element;N_i}{\&Sigma;}{w}^i{x}_{ij}}{\underset{i=1}{\overset{n}{\mathrm{\&Sigma;}}}{\left(u_{ki}^{\&prime;}\right)}^m\underset{\left\{j\right\}\&Element;N_i}{\&Sigma;}{w}^i}$

Wherein, v _2krepresent the center gray-scale value of the image clustering center to be split kth class of current iteration number of times, n represents the pixel number of image to be split, and ∑ represents sum operation, and i represents the label of i-th pixel in image to be split, u ' _kito represent in image to be split i-th pixel belong to the degree of membership of kth class in initial cluster center level and smooth after value, k represents the label of kth class in cluster centre, m represents Fuzzy Exponential, value is 2, j represents the label of a jth pixel in the neighborhood image block of i-th pixel in image to be split, { } represents element set symbol, and ∈ represents and belongs to symbol, N _irepresent the set of all pixels in the neighborhood image block of i-th pixel in image to be split, w ⁱrepresent the weights of the Weighted Fuzzy factor of i-th pixel in image to be split, x _ijrepresent the gray-scale value of a jth pixel in i-th pixel neighborhood of a point in image to be split;

(6d) replace image clustering center to be split gray-scale value with the image clustering center to be split gray-scale value of current iteration number of times, obtain new image clustering center to be split gray-scale value;

(6e) deduct the image clustering center to be split gray-scale value before replacement with new image clustering center to be split gray-scale value, obtain a matrix of differences, then element each in matrix of differences is taken absolute value, obtain absolute difference matrix;

(7) judge whether the element value in absolute difference matrix meets iteration stopping condition, if so, then perform step (8), otherwise, the iterations in cluster iteration is added 1, performs step (6);

(8) segmentation image is produced:

(8a) from the subordinated-degree matrix of image slices vegetarian refreshments to be split, find out the maximum membership degree in image to be split in each pixel column, by the line label of this maximum membership degree position in subordinated-degree matrix, as the class label of the pixel corresponding to this maximum membership degree;

(8b) using the gray-scale value of the gray-scale value of each class in cluster centre as all pixels in such;

(8c) show all classes in image to be split, complete Iamge Segmentation.

The present invention compared with prior art has the following advantages:

First, because the present invention is in the process of Iamge Segmentation, have employed mean shift algorithm, obtain initial cluster centre and clusters number, overcome prior art due to the randomness of cluster centre cause very sensitive and be easily absorbed in the shortcoming of local optimum to initial cluster center point, make cluster process of the present invention to converge on global optimum, reasonably split image.

Second, in cluster process, replace the pixel in traditional fuzzy C-means clustering method with image pixel neighborhood of a point image block to be split due to the present invention, and in the objective function of the method, introduce the Weighted Fuzzy factor comprising neighborhood information, neighborhood information can be utilized fully, the details of image is taken into full account, thus the unbalanced data set of Density Distribution can be split, also improve the precision of Iamge Segmentation of the present invention.

3rd, because the present invention uses image pixel neighborhood of a point information to be split smoothing to the degree of membership of pixel in image to be split in cluster iterative process, reduce the impact of noise on segmentation result, the present invention is declined to noise sensitivity in cutting procedure, improves the robustness of cutting procedure to noise.

Accompanying drawing explanation

Fig. 1 is process flow diagram of the present invention;

Fig. 2 is the process flow diagram of mean shift algorithm of the present invention;

Fig. 3 is analogous diagram of the present invention.

Embodiment

Below in conjunction with accompanying drawing, the present invention is described in further detail.

With reference to accompanying drawing 1, the concrete steps that the present invention realizes are described in further detail.

Step 1. inputs an image to be split.

In embodiments of the present invention, input the image to be split that a width size is 244*244 pixel unit, obtain the intensity profile matrix of this image slices vegetarian refreshments.

Step 2. adopts mean shift algorithm, calculates clusters number and initial cluster center.

With reference to accompanying drawing 2, the concrete steps of mean shift algorithm of the present invention are described in further detail.

The weights of each pixel in the image to be split of input are set to-1 by the 1st step.

2nd step, from the image to be split of input, an optional unlabelled pixel is as cluster centre point.

3rd step, according to the following formula, calculates the gray-scale value of the cluster centre after current cluster centre point drift.

$m\left(x\right)=\frac{\underset{i=1}{\overset{n}{\mathrm{\&Sigma;}}}{x}_{i}g\left(\right|\left|\right(x-x_i)/h|{|}^2)}{\underset{i=1}{\overset{n}{\mathrm{\&Sigma;}}}g\left(\right|\left|\right(x-x_i)/h|{|}^2)}$

Wherein, m (x) represents the gray-scale value of the cluster centre after current cluster centre point drift, and i represents the label of i-th pixel in image to be split, ∑ represents sum operation, n represents the number of pixel in image to be split, and x represents the gray-scale value of initial cluster center point, x _irepresent the gray-scale value of i-th pixel in image to be split, k () represents gaussian kernel function _,g () represents the opposite number of gaussian kernel function gradient, || || represent and ask Euclidean distance to operate, h represents the bandwidth matrices being proportional to unit matrix, and the coefficient of bandwidth matrices is 20.

4th step, deducts the gray-scale value of the point of the cluster centre after drift, obtains a difference, then difference taken absolute value with the gray-scale value of current cluster centre point.

5th step, judges whether absolute value is less than iteration stopping threshold value 0.01, if so, then performs the 6th step, otherwise, replace current cluster centre point with the central point after drift, perform the 3rd step.

6th step, the weights of the pixel after pixel gray-scale value in image to be split being in current cluster centre point gray-scale value and drift between cluster centre point gray-scale value are labeled as 1, the pixel be in by pixel gray-scale value in image to be split between cluster centre point gray-scale value after current cluster centre point gray-scale value and drift is divided into same class, using the cluster centre point of the cluster centre point after drift as such.

7th step, adds 1 by clusters number, and adding up pixel weights in all images to be split is the number of the pixel of 1.

8th step, is set to the number being less than 200 than the number of pixel in image to be split by pixel number predetermined value.

9th step, judges whether the pixel number added up is more than or equal to pixel number predetermined value, if so, then performs the 10th step, otherwise, perform the 1st step.

10th step, exports the gray-scale value of clusters number and cluster centre point.

Example of the present invention is 4 by the clusters number that mean shift algorithm draws, initial cluster center V1=[0,85,170,255].

Step 3. initialization.

Initial cycle number of times is set to 0, maximum iteration time is set to 500, the subordinated-degree matrix of random initializtion image to be split.

Embodiments of the invention are that clusters number is set to 4, and initial cluster center is set to V1=[0,85,170,255].

Step 4. calculates the weights of neighborhood image block in image to be split.

Centered by pixel in image to be split, with 1 pixel unit for radius, obtain the neighborhood image block of 3*3 pixel unit of 9 pixel compositions within the scope of this, each pixel in image to be split is asked to the operation of neighborhood image block, obtain the neighborhood image block of all pixels.

According to the following formula, the variance yields of pixel gray-scale value in each neighborhood image block is calculated:

${\mathrm{\&delta;}}_i={\&lsqb;\frac{\underset{\left\{j\right\}\&Element;N_i}{\&Sigma;}{(x_j-x_i)}^2}{9}\&rsqb;}^{\frac{1}{2}}$

Wherein, δ _irepresent the variance yields of all pixel gray-scale values in the neighborhood image block of i-th pixel in image to be split, i represents the label of i-th pixel in image to be split, j represents the label of a jth pixel in i-th pixel neighborhood of a point in image to be split, ∑ represents summation symbol, { } represents element set symbol, ∈ represents and belongs to symbol, N _irepresent the set of all pixels in i-th pixel neighborhood of a point in image to be split, x _jrepresent the gray-scale value of a jth pixel in i-th pixel neighborhood of a point in image to be split, x _irepresent the gray-scale value of i-th pixel in image to be split.

According to the following formula, by the variance projection of pixel gray-scale values all in each neighborhood image block to gaussian kernel space:

Wherein, represent the value of variance projection behind gaussian kernel space of pixel gray-scale values all in the neighborhood image block of i-th pixel in image to be split, i represents the label of i-th pixel in image to be split, and exp () represents index operation, δ _irepresent the variance yields of all pixel gray-scale values in the neighborhood image block of i-th pixel in image to be split, ∑ represents sum operation, and j represents the label of a jth pixel in the neighborhood image block of i-th pixel in image to be split, and { } represents element set symbol, ∈ represents and belongs to symbol, N _irepresent the set of all pixels in the neighborhood image block of i-th pixel in image to be split, δ _jrepresent the variance yields of all pixel gray-scale values in the neighborhood image block of a jth pixel in the neighborhood image block of i-th pixel in image to be split.

According to the following formula, by the weights normalization of each neighborhood image block:

Wherein, w ⁱrepresent the weights of the neighborhood image block of i-th pixel in image to be split, i represents the label of i-th pixel in image to be split, represent the value of variance projection behind gaussian kernel space of pixel gray-scale values all in the neighborhood image block of i-th pixel in image to be split, ∑ represents sum operation, j represents the label of a jth pixel in the neighborhood image block of i-th pixel in image to be split, { } represents element set symbol, ∈ represents and belongs to symbol, N _irepresent the set of all pixels in the neighborhood image block of i-th pixel in image to be split, represent the value of variance projection behind gaussian kernel space of all pixel gray-scale values in the neighborhood image block of a jth pixel in the neighborhood image block of i-th pixel in image to be split.

Step 5. calculates the Weighted Fuzzy factor weights of each pixel in image to be split.

According to the following formula, the space length weights of pixel in pixel and its neighborhood image block in image to be split are calculated:

$w_1=\frac{1}{d_{ij}+1}$

Wherein, w ₁to represent in image to be split the space length weights of pixel in i-th pixel and its neighborhood image block, i represents the label of i-th pixel in image to be split, j represents the label of a jth pixel in the neighborhood image block of i-th pixel in image to be split, d _ijrepresent the Euclidean distance of a jth pixel in i-th pixel and neighborhood image block in image to be split.

According to the following formula, the spatial-intensity weights of pixel in pixel and its neighborhood image block in image to be split are calculated:

$w_2=1-log\left(\frac{1}{9}\underset{q=1}{\overset{9}{\&Sigma;}}\frac{x_q^{N_i}}{x_q^{N_j}}\right)$

Wherein, w ₂to represent in image to be split the spatial-intensity weights of pixel in i-th pixel and its neighborhood image block, log () represents denary logarithm operation, and ∑ represents sum operation, and q represents the label of q pixel in neighborhood image block, to represent in image to be split centered by i-th pixel the gray-scale value of q pixel in neighborhood of a point image block, N _irepresent the set of all pixels in i-th neighborhood of pixel points image block in image to be split, i represents the label of i-th pixel in image to be split, to represent in the neighborhood image block of i-th pixel in image to be split the gray-scale value of q pixel in neighborhood of a point image block centered by a jth pixel, N _jrepresent the set of all pixels in a jth neighborhood of pixel points image block in image to be split, j represents the label of a jth pixel in i-th neighborhood of pixel points image block in image to be split.

According to the following formula, the Weighted Fuzzy factor weights of each pixel in image to be split are calculated:

w _i＝w ₁·w ₂

Wherein, w _irepresent the Weighted Fuzzy factor weights of i-th pixel in image to be split, i represents the label of i-th pixel in image to be split, w ₁to represent in image to be split the space length weights of pixel in i-th pixel and its neighborhood image block, w ₂to represent in image to be split the spatial-intensity weights of pixel in i-th pixel and its neighborhood image block.

Step 6. cluster iteration.

Calculate the degree of membership of each pixel in image to be split, by the degree of membership of obtained all pixels composition subordinated-degree matrix.

According to the following formula, the pixel calculated in image to be split belongs to the Weighted Fuzzy factor of cluster centre kth class:

$G_{ki}=\underset{j\&Element;N_i,j\&NotEqual;i}{\&Sigma;}{w}_i{(1-u_{kj})}^m\mathrm{||}{x}_i-v_k{\mathrm{||}}^2$

Wherein, G _kirepresent that i-th pixel in image to be split belongs to the blur level of kth class, k represents the label of kth class in cluster centre, and i represents the label of i-th pixel in image to be split, and j represents the label of a jth pixel in neighborhood image block, ∑ represents sum operation, N _irepresent the set of all pixels in the neighborhood image block of i-th pixel in image to be split, ≠ expression is not equal to symbol, w _irepresent the Weighted Fuzzy factor weights of i-th pixel in image to be split, u _kjrepresent that in the neighborhood image block of i-th pixel in image to be split, a jth pixel belongs to the degree of membership of kth class, m represents Fuzzy Exponential, and value is 2, || || represent and ask Euclidean distance to operate, x _irepresent the gray-scale value of i-th pixel in image to be split, v _krepresent the gray-scale value of a kth cluster centre point.

According to the following formula, the degree of membership of each pixel in image to be split is calculated:

$u_{ki}={\{\underset{t=1}{\overset{c}{\&Sigma;}}\&lsqb;{\left(\frac{\underset{\left\{j\right\}\&Element;N_i}{\&Sigma;}{w}^i{(x_{ij}-v_{1k})}^2+G_{ki}}{\underset{\left\{j\right\}\&Element;N_i}{\&Sigma;}{w}^i{(x_{ij}-v_{1t})}^2+G_{ti}}\right)}^{\frac{1}{m-1}}\&rsqb;\}}^{-1}$

Wherein, u _kito represent in image to be split that i-th pixel belongs to the degree of membership of kth class in cluster centre, u _kimeet constraint condition: k is the label of cluster centre kth class, and c is the number of cluster, and ∑ represents sum operation, N _irepresent the set of all pixels in the neighborhood image block of i-th pixel in image to be split, j represents the label of a jth pixel in neighborhood image block, and ∈ represents and belongs to symbol, w ⁱrepresent the weights of i-th neighborhood of pixel points image block in image to be split, x _ijrepresent the gray-scale value of the jth pixel in image to be split in i-th neighborhood of pixel points image block, v _1krepresent the center gray-scale value of kth class in initial cluster center, k=1,2 ..., c, G _kito represent in image to be split that i-th pixel belongs to the blur level of kth class, G _tito represent in image to be split that i-th pixel belongs to the blur level of t class, m represents Fuzzy Exponential, and value is 2.

To the smoothing process of degree of membership of each pixel in image to be split.

According to the following formula, the pixel in calculating neighborhood image block is to the control coefrficient of the central pixel point of neighborhood image block:

$p(x_i,x_j)=\frac{d(x_i,x_j)}{\underset{\left\{j\right\}\&Element;N_i}{\&Sigma;}d(x_i,x_j)}$

Wherein, p () represents that pixel in neighborhood image block is to the control coefrficient of the central pixel point of neighborhood image block, and its value, more close to 1, shows that the correlativity of pixel in neighborhood image block and central pixel point is larger, x _irepresent the gray-scale value of i-th pixel in image to be split, i represents the label of i-th pixel in image to be split, x _jrepresent the gray-scale value of a jth pixel in the neighborhood image block of i-th pixel, j represents the label of a jth pixel in the neighborhood image block of i-th pixel in image to be split, d () represents the Euclidean distance of two pixels, ∑ represents sum operation, { } represents element set symbol, ∈ represents and belongs to symbol, N _irepresent the set of all pixels in i-th neighborhood of pixel points image block in image to be split.

According to the following formula, the weight coefficient of all pixels in each neighborhood image block is calculated:

$a_{ij}=\left\{\begin{array}{cc}p(x_i,x_j),& |x_i-x_m|<{\mathrm{\&delta;}}_i\\ 0,& |x_i-x_m|>{\mathrm{\&delta;}}_i\end{array}\right.$

Wherein, a _ijrepresent the weight coefficient of a jth pixel in the neighborhood image block of i-th pixel in image to be split, i represents the label of i-th pixel in image to be split, j represents the label of a jth pixel in the neighborhood image block of i-th pixel in image to be split, pixel in p () expression neighborhood image block is to the control coefrficient of the central pixel point of neighborhood image block, || represent and ask absolute value operation, x _irepresent the gray-scale value of i-th pixel in image to be split, x _mrepresent the mean value of all pixel gray-scale values in the neighborhood image block of i-th pixel in image to be split, δ _irepresent the variance yields of all pixel gray-scale values in i-th neighborhood of pixel points image block in image to be split.

According to the following formula, to the smoothing process of degree of membership of each pixel in image to be split:

$u_{ki}^{\&prime;}=\underset{\left\{j\right\}\&Element;N_i}{\&Sigma;}{a}_{ij}{u}_{ki}$

Wherein, u ' _kirepresent i-th pixel in image to be split belong to the degree of membership smoothing processing of kth class in initial cluster center after value, ∑ represents summation symbol, i represents the label of i-th pixel in image to be split, j represents the label of a jth pixel in the neighborhood image block of i-th pixel in image to be split, { } represents element set symbol, ∈ represents and belongs to symbol, N _irepresent the set of all pixels in i-th neighborhood of pixel points image block in image to be split, u _kito represent in image to be split that i-th pixel belongs to the degree of membership of kth class in initial cluster center, a _ijrepresent the weight coefficient of a jth pixel in the neighborhood image block of i-th pixel in image to be split, meet a _ij∈ [0,1], a _ijbe used for controlling the size that the pixel in image to be split in i-th neighborhood of pixel points image block affects neighborhood image block central pixel point.

According to cluster centre more new formula, calculate the image clustering center to be split of current iteration number of times:

$v_{2k}=\frac{\underset{i=1}{\overset{n}{\mathrm{\&Sigma;}}}{\left(u_{ki}^{\&prime;}\right)}^m\underset{\left\{j\right\}\&Element;N_i}{\&Sigma;}{w}^i{x}_{ij}}{\underset{i=1}{\overset{n}{\mathrm{\&Sigma;}}}{\left(u_{ki}^{\&prime;}\right)}^m\underset{\left\{j\right\}\&Element;N_i}{\&Sigma;}{w}^i}$

Wherein, v _2krepresent the center gray-scale value of the image clustering center to be split kth class of current iteration number of times, n represents pixel number in image to be split, and ∑ represents sum operation, and i represents the label of i-th pixel in image to be split, u ' _kito represent in image to be split i-th pixel belong to the degree of membership of kth class in initial cluster center level and smooth after value, k represents the label of kth class in cluster centre, m represents Fuzzy Exponential, value is 2, j represents the label of a jth pixel in the neighborhood image block of i-th pixel in image to be split, { } represents element set symbol, and ∈ represents and belongs to symbol, N _irepresent the set of all pixels in the neighborhood image block of i-th pixel in image to be split, w ⁱrepresent the weights of the Weighted Fuzzy factor of i-th pixel in image to be split, x _ijrepresent the gray-scale value of a jth pixel in i-th pixel neighborhood of a point in image to be split.

Replace image clustering center to be split gray-scale value with the image clustering center to be split gray-scale value of current iteration number of times, obtain new image clustering center to be split gray-scale value.

Deduct the image clustering center to be split gray-scale value before replacement with new image clustering center to be split gray-scale value, obtain a matrix of differences, then element each in matrix of differences is taken absolute value, obtain absolute difference matrix.

Step 7. judges whether the element value in absolute difference matrix meets iteration stopping condition, and described iteration stopping condition refers to the situation of one of following condition: condition 1, and any one element value in absolute difference matrix is less than iteration stopping threshold value 0.01; Condition 2, cycle index reaches maximum iteration time 500 times.If so, then step 8 is performed, otherwise, the iterations in cluster iteration is added 1, performs step 6.

Step 8. produces segmentation image.

The maximum membership degree in image to be split in each pixel column is found out from the subordinated-degree matrix of image slices vegetarian refreshments to be split, by the line label of this maximum membership degree position in subordinated-degree matrix, as the class label of the pixel corresponding to this maximum membership degree.

Using the gray-scale value of the gray-scale value of each class in cluster centre as all pixels in such.

Show all classes in image to be split, complete Iamge Segmentation.

Below in conjunction with accompanying drawing 3, simulated effect of the present invention is further described.

1. simulated conditions:

Emulation experiment of the present invention is being configured to corei32.30GHZ, and the computing machine of internal memory 2G, WINDOWS7 system uses MatlabR2009a to carry out.

2. emulate content:

Fig. 3 is analogous diagram of the present invention.Wherein Fig. 3 (a) is the Prof. Du Yucang SAR image that a width is conventional in the emulation experiment of technical field of image processing.Fig. 3 (b) is that to add standard deviation to the Prof. Du Yucang SAR image of Fig. 3 (a) be the image obtained after the Gaussian noise of 0.05.Adopt the classical fuzzy C-means clustering image partition method of prior art to split Fig. 3 (b), the result after segmentation is as shown in Fig. 3 (c).Adopt the fuzzy C-mean algorithm image partition method based on average drifting and neighborhood information of the present invention to split Fig. 3 (b), the result after segmentation is as shown in Fig. 3 (d).

3. analysis of simulation result:

As can be seen from Fig. 3 (c) and Fig. 3 (d), when using classical fuzzy C-means clustering image partition method and method of the present invention to split Fig. 3 (b), correct segmentation number can be obtained.But comparison diagram 3 (c) and Fig. 3 (d) two figure can find, in Fig. 3 (c), the noise of segmentation image still exists, and the image border of Fig. 3 (c) does not have the image edge clear of Fig. 3 (d); In Fig. 3 (d), picture noise is little, edge clear, region consistency is high, illustrate that the fuzzy C-mean algorithm image partition method based on average drifting and neighborhood information in this paper can obtain sharp-edged segmentation result, and the robustness of image segmentation process to noise is better.

The accuracy of separation SA (%) of the different noise of table 1 Prof. Du Yucang image

Noise	Fuzzy C-means clustering dividing method	The inventive method
			Gauss 1	66.60	99.08
Gauss 2	55.90	98.94
			Gauss 3	50.87	98.91
The spiced salt 1	96.27	99.33
			The spiced salt 2	92.66	98.90
The spiced salt 3	88.73	96.41
			Mixing	64.33	98.95

Table 1 is the Gaussian noise that Fig. 3 (a) adds varying strength, the accuracy of separation that during mixed noise of salt-pepper noise and two kinds of noises, control methods and the inventive method obtain.Gaussian noise 1, Gaussian noise 2, Gaussian noise 3 is averages is 0, and standard deviation is the Gaussian noise of 0.05,0.1 and 0.15.Salt-pepper noise 1, salt-pepper noise 2, salt-pepper noise 3 to be intensity be 0.05,0.1 and 0.15 noise, mixed noise is the mixing of Gaussian noise 1 and salt-pepper noise 1.As can be seen from Table 1, when image adds salt-pepper noise 1, the method that control methods and the present invention propose all can obtain desirable segmentation accuracy, but along with the increase of noise intensity, the fall of the segmentation accuracy of the method that the present invention proposes is less than control methods; When image adds standard deviation different Gaussian noise, the segmentation accuracy of control methods is very low, and the accuracy of separation of the method that the present invention proposes is all more than 98%.So the method that the present invention proposes can obtain desirable result to the segmentation of Noise image, better to the robustness of noise in image segmentation process.

Claims (9)

1., based on a fuzzy C-mean algorithm image partition method for average drifting and neighborhood information, performing step is as follows:

(1) image to be split is inputted;

(2) adopt mean shift algorithm, calculate clusters number and initial cluster center:

(3) initialization:

Initial cycle number of times is set to 0, maximum iteration time is set to 500, the subordinated-degree matrix of random initializtion image to be split;

(4) weights of neighborhood image block in image to be split are calculated:

(4a) centered by the pixel in image to be split, with 1 pixel unit for radius, obtain the neighborhood image block of 3*3 pixel unit of 9 pixel compositions within the scope of this, obtain the neighborhood image block of all pixels;

(4b) variance yields of pixel gray-scale value in each neighborhood image block is calculated;

(4c) by the variance projection of pixel gray-scale values all in each neighborhood image block to gaussian kernel space;

(4d) by the weights normalization of each neighborhood image block;

(5) the Weighted Fuzzy factor weights of each pixel in image to be split are calculated:

(5a) the space length weights of pixel in pixel and its neighborhood image block in image to be split are calculated;

(5b) the spatial-intensity weights of pixel in pixel and its neighborhood image block in image to be split are calculated;

(5c) the Weighted Fuzzy factor weights of each pixel in image to be split according to the following formula, are calculated;

w _i＝w ₁·w ₂

Wherein, w _irepresent the Weighted Fuzzy factor weights of i-th pixel in image to be split, i represents the label of i-th pixel in image to be split, w ₁to represent in image to be split the space length weights of pixel in i-th pixel and its neighborhood image block, w ₂to represent in image to be split the spatial-intensity weights of pixel in i-th pixel and its neighborhood image block;

(6) cluster iteration:

(6a) degree of membership of each pixel in image to be split is calculated, by the degree of membership of obtained all pixels composition subordinated-degree matrix;

(6b) to the smoothing process of degree of membership of each pixel in image to be split;

1st step, according to the following formula, the pixel in calculating neighborhood image block is to the control coefrficient of the central pixel point of neighborhood image block:

$p(x_i,x_j)=\frac{d(x_i,x_j)}{\underset{\left\{j\right\}\&Element;N_i}{\&Sigma;}d(x_i,x_j)}$

Wherein, p () represents that pixel in neighborhood image block is to the control coefrficient of the central pixel point of neighborhood image block, and its value, more close to 1, shows that the correlativity of pixel in neighborhood image block and central pixel point is larger, x _irepresent the gray-scale value of i-th pixel in image to be split, i represents the label of i-th pixel in image to be split, x _jrepresent the gray-scale value of a jth pixel in the neighborhood image block of i-th pixel, j represents the label of a jth pixel in the neighborhood image block of i-th pixel in image to be split, d () represents the Euclidean distance of two pixels, ∑ represents sum operation, { } represents element set symbol, ∈ represents and belongs to symbol, N _irepresent the set of all pixels in i-th neighborhood of pixel points image block in image to be split;

2nd step, according to the following formula, calculates the weight coefficient of all pixels in each neighborhood image block:

$a_{ij}=\left\{\begin{array}{cc}p(x_i,x_j),& |x_i-x_m|<{\mathrm{\&delta;}}_i\\ 0,& |x_i-x_m|>{\mathrm{\&delta;}}_i\end{array}\right.$

Wherein, a _ijrepresent the weight coefficient of a jth pixel in the neighborhood image block of i-th pixel in image to be split, i represents the label of i-th pixel in image to be split, j represents the label of a jth pixel in the neighborhood image block of i-th pixel in image to be split, pixel in p () expression neighborhood image block is to the control coefrficient of the central pixel point of neighborhood image block, || represent and ask absolute value operation, x _irepresent the gray-scale value of i-th pixel in image to be split, x _mrepresent the mean value of all pixel gray-scale values in the neighborhood image block of i-th pixel in image to be split, δ _irepresent the variance yields of all pixel gray-scale values in i-th neighborhood of pixel points image block in image to be split;

3rd step, according to the following formula, the smoothing process of degree of membership to each pixel in image to be split:

$u_{ki}^{\&prime;}=\underset{\left\{j\right\}\&Element;N_i}{\&Sigma;}{a}_{ij}{u}_{ki}$

Wherein, u' _kirepresent i-th pixel in image to be split belong to the degree of membership smoothing processing of kth class in initial cluster center after value, ∑ represents sum operation, i represents the label of i-th pixel in image to be split, j represents the label of a jth pixel in the neighborhood image block of i-th pixel in image to be split, { } represents element set symbol, ∈ represents and belongs to symbol, N _irepresent the set of all pixels in i-th neighborhood of pixel points image block in image to be split, u _kito represent in image to be split that i-th pixel belongs to the degree of membership of kth class in initial cluster center, a _ijrepresent the weight coefficient of a jth pixel in the neighborhood image block of i-th pixel in image to be split, a _ij∈ [0,1], a _ijbe used for controlling the size that the pixel in image to be split in i-th neighborhood of pixel points image block affects neighborhood image block central pixel point;

(6c) the image clustering center to be split of current iteration number of times according to the following formula, is calculated:

$v_{2k}=\frac{\underset{i=1}{\overset{n}{\mathrm{\&Sigma;}}}{\left(u_{ki}^{\&prime;}\right)}^m\underset{\left\{j\right\}\&Element;N_i}{\&Sigma;}{w}^i{x}_{ij}}{\underset{i=1}{\overset{n}{\mathrm{\&Sigma;}}}{\left(u_{ki}^{\&prime;}\right)}^m\underset{\left\{j\right\}\&Element;N_i}{\&Sigma;}{w}^i}$

Wherein, v _2krepresent the center gray-scale value of the image clustering center to be split kth class of current iteration number of times, n represents the pixel number of image to be split, and ∑ represents sum operation, and i represents the label of i-th pixel in image to be split, u' _kito represent in image to be split i-th pixel belong to the degree of membership of kth class in initial cluster center level and smooth after value, k represents the label of kth class in cluster centre, m represents Fuzzy Exponential, value is 2, j represents the label of a jth pixel in the neighborhood image block of i-th pixel in image to be split, { } represents element set symbol, and ∈ represents and belongs to symbol, N _irepresent the set of all pixels in the neighborhood image block of i-th pixel in image to be split, w ⁱrepresent the weights of the Weighted Fuzzy factor of i-th pixel in image to be split, x _ijrepresent the gray-scale value of a jth pixel in i-th pixel neighborhood of a point in image to be split;

(6d) replace image clustering center to be split gray-scale value with the image clustering center to be split gray-scale value of current iteration number of times, obtain new image clustering center to be split gray-scale value;

(6e) deduct the image clustering center to be split gray-scale value before replacement with new image clustering center to be split gray-scale value, obtain a matrix of differences, then element each in matrix of differences is taken absolute value, obtain absolute difference matrix;

(7) judge whether the element value in absolute difference matrix meets iteration stopping condition, if so, then perform step (8), otherwise, the iterations in cluster iteration is added 1, performs step (6);

(8) segmentation image is produced:

(8a) from the subordinated-degree matrix of image slices vegetarian refreshments to be split, find out the maximum membership degree in image to be split in each pixel column, by the line label of this maximum membership degree position in subordinated-degree matrix, as the class label of the pixel corresponding to this maximum membership degree;

(8b) using the gray-scale value of the gray-scale value of each class in cluster centre as all pixels in such;

(8c) show all classes in image to be split, complete Iamge Segmentation.

2. the fuzzy C-mean algorithm image partition method based on average drifting and neighborhood information according to claim 1, is characterized in that: the concrete steps of the mean shift algorithm described in step (2) are as follows;

The weights of each pixel in the image to be split of input are set to-1 by the 1st step;

2nd step, from the image to be split of input, an optional unlabelled pixel is as cluster centre point;

3rd step, according to the following formula, calculates the gray-scale value of the cluster centre after current cluster centre point drift:

$m\left(x\right)=\frac{\underset{i=1}{\overset{n}{\mathrm{\&Sigma;}}}{x}_{i}g\left(\right|\left|\right(x-x_i)/h|{|}^2)}{\underset{i=1}{\overset{n}{\mathrm{\&Sigma;}}}g\left(\right|\left|\right(x-x_i)/h|{|}^2)}$

Wherein, m (x) represents the gray-scale value of the cluster centre after current cluster centre point drift, and i represents the label of i-th pixel in image to be split, ∑ represents sum operation, n represents the number of pixel in image to be split, and x represents the gray-scale value of initial cluster center point, x _irepresent the gray-scale value of i-th pixel in image to be split, k () represents gaussian kernel function, and g () represents the opposite number of gaussian kernel function gradient, || || represent and ask Euclidean distance to operate, h represents the bandwidth matrices being proportional to unit matrix, and the coefficient of bandwidth matrices is 20;

4th step, deducts the gray-scale value of the point of the cluster centre after drift, obtains a difference, then difference taken absolute value with the gray-scale value of current cluster centre point;

5th step, judges whether absolute value is less than iteration stopping threshold value 0.01, if so, then performs the 6th step, otherwise, replace current cluster centre point with the central point after drift, perform the 3rd step;

6th step, the weights of the pixel after pixel gray-scale value in image to be split being in current cluster centre point gray-scale value and drift between cluster centre point gray-scale value are labeled as 1, the pixel be in by pixel gray-scale value in image to be split between cluster centre point gray-scale value after current cluster centre point gray-scale value and drift is divided into same class, using the cluster centre point of the cluster centre point after drift as such;

7th step, adds 1 by clusters number, and adding up pixel weights in all images to be split is the number of the pixel of 1;

8th step, is set to the number being less than 200 than the number of pixel in image to be split by pixel number predetermined value;

9th step, judges whether the pixel number added up is more than or equal to pixel number predetermined value, if so, then performs the 10th step, otherwise, perform the 1st step;

10th step, exports the gray-scale value of clusters number and cluster centre point.

3. the fuzzy C-mean algorithm image partition method based on average drifting and neighborhood information according to claim 1, is characterized in that: the variance yields calculating pixel gray-scale value in each neighborhood image block described in step (4b) obtains according to the following formula;

${\mathrm{\&delta;}}_i={\&lsqb;\frac{\underset{\left\{j\right\}\&Element;N_i}{\&Sigma;}{(x_j-x_i)}^2}{9}\&rsqb;}^{\frac{1}{2}}$

Wherein, δ _irepresent the variance yields of all pixel gray-scale values in the neighborhood image block of i-th pixel in image to be split, i represents the label of i-th pixel in image to be split, j represents the label of a jth pixel in i-th pixel neighborhood of a point in image to be split, ∑ represents summation symbol, { } represents element set symbol, ∈ represents and belongs to symbol, N _irepresent the set of all pixels in i-th pixel neighborhood of a point in image to be split, x _jrepresent the gray-scale value of a jth pixel in i-th pixel neighborhood of a point in image to be split, x _irepresent the gray-scale value of i-th pixel in image to be split.

4. the fuzzy C-mean algorithm image partition method based on average drifting and neighborhood information according to claim 1, is characterized in that: completed according to the following formula to gaussian kernel space by the variance projection of pixel gray-scale values all in each neighborhood image block described in step (4c);

Wherein, represent the value of variance projection behind gaussian kernel space of pixel gray-scale values all in the neighborhood image block of i-th pixel in image to be split, i represents the label of i-th pixel in image to be split, and exp () represents index operation, δ _irepresent the variance yields of all pixel gray-scale values in the neighborhood image block of i-th pixel in image to be split, ∑ represents sum operation, and j represents the label of a jth pixel in the neighborhood image block of i-th pixel in image to be split, and { } represents element set symbol, ∈ represents and belongs to symbol, N _irepresent the set of all pixels in the neighborhood image block of i-th pixel in image to be split, δ _jrepresent the variance yields of all pixel gray-scale values in the neighborhood image block of a jth pixel in the neighborhood image block of i-th pixel in image to be split.

5. the fuzzy C-mean algorithm image partition method based on average drifting and neighborhood information according to claim 1, is characterized in that: the weights normalization of each neighborhood image block realized according to the following formula described in step (4d);

Wherein, w ⁱrepresent the weights of the neighborhood image block of i-th pixel in image to be split, i represents the label of i-th pixel in image to be split, represent the value of variance projection behind gaussian kernel space of pixel gray-scale values all in the neighborhood image block of i-th pixel in image to be split, ∑ represents sum operation, j represents the label of a jth pixel in the neighborhood image block of i-th pixel in image to be split, { } represents element set symbol, ∈ represents and belongs to symbol, N _irepresent the set of all pixels in the neighborhood image block of i-th pixel in image to be split, represent the value of variance projection behind gaussian kernel space of all pixel gray-scale values in the neighborhood image block of a jth pixel in the neighborhood image block of i-th pixel in image to be split.

6. the fuzzy C-mean algorithm image partition method based on average drifting and neighborhood information according to claim 1, is characterized in that: in the calculating image to be split described in step (5a), in pixel and its neighborhood image block, the space length weights of pixel calculate according to the following formula;

$w_1=\frac{1}{d_{ij}+1}$

Wherein, w ₁to represent in image to be split the space length weights of pixel in i-th pixel and its neighborhood image block, i represents the label of i-th pixel in image to be split, j represents the label of a jth pixel in the neighborhood image block of i-th pixel in image to be split, d _ijrepresent the Euclidean distance of a jth pixel in i-th pixel and neighborhood image block in image to be split.

7. the fuzzy C-mean algorithm image partition method based on average drifting and neighborhood information according to claim 1, is characterized in that: in the calculating image to be split described in step (5b), in pixel and its neighborhood image block, the spatial-intensity weights of pixel calculate according to the following formula;

$w_2=1-log\left(\frac{1}{9}\underset{q=1}{\overset{9}{\&Sigma;}}\frac{x_q^{N_i}}{x_q^{N_j}}\right)$

Wherein, w ₂to represent in image to be split the spatial-intensity weights of pixel in i-th pixel and its neighborhood image block, log () represents denary logarithm operation, and ∑ represents sum operation, and q represents the label of q pixel in neighborhood image block, to represent in image to be split centered by i-th pixel the gray-scale value of q pixel in neighborhood of a point image block, N _irepresent the set of all pixels in i-th neighborhood of pixel points image block in image to be split, i represents the label of i-th pixel in image to be split, to represent in the neighborhood image block of i-th pixel in image to be split the gray-scale value of q pixel in neighborhood of a point image block centered by a jth pixel, N _jrepresent the set of all pixels in a jth neighborhood of pixel points image block in image to be split, j represents the label of a jth pixel in i-th neighborhood of pixel points image block in image to be split.

8. the fuzzy C-mean algorithm image partition method based on average drifting and neighborhood information according to claim 1, is characterized in that: in the calculating image to be split described in step (6a), the step of the degree of membership of each pixel is as follows:

1st step, according to the following formula, the pixel calculated in image to be split belongs to the Weighted Fuzzy factor of cluster centre kth class:

$G_{ki}=\underset{j\&Element;N_i,j\&NotEqual;i}{\&Sigma;}{w}_i{(1-u_{kj})}^m\left|\right|x_i-v_k|{|}^2$

Wherein, G _kirepresent that i-th pixel in image to be split belongs to the blur level of kth class, k represents the label of kth class in cluster centre, and i represents the label of i-th pixel in image to be split, and j represents the label of a jth pixel in neighborhood image block, ∑ represents sum operation, N _irepresent the set of all pixels in the neighborhood image block of i-th pixel in image to be split, ≠ expression is not equal to symbol, w _irepresent the Weighted Fuzzy factor weights of i-th pixel in image to be split, u _kjrepresent that in the neighborhood image block of i-th pixel in image to be split, a jth pixel belongs to the degree of membership of kth class, m represents Fuzzy Exponential, and value is 2, || || represent and ask Euclidean distance to operate, x _irepresent the gray-scale value of i-th pixel in image to be split, v _krepresent the gray-scale value of a kth cluster centre point;

2nd step, according to following degree of membership more new formula, calculates the degree of membership of each pixel in image to be split:

$u_{ki}={\{\underset{t=1}{\overset{c}{\&Sigma;}}\&lsqb;{\left(\frac{\underset{\left\{j\right\}\&Element;N_i}{\&Sigma;}{w}^i{(x_{ij}-v_{1k})}^2+G_{ki}}{\underset{\left\{j\right\}\&Element;N_i}{\&Sigma;}{w}^i{(x_{ij}-v_{1t})}^2+G_{ti}}\right)}^{\frac{1}{m-1}}\&rsqb;\}}^{-1}$

Wherein, u _kito represent in image to be split that i-th pixel belongs to the degree of membership of kth class in cluster centre, u _kimeet constraint condition: k is the label of cluster centre kth class, and c is the number of cluster, and ∑ represents sum operation, N _irepresent the set of all pixels in the neighborhood image block of i-th pixel in image to be split, j represents the label of a jth pixel in neighborhood image block, and ∈ represents and belongs to symbol, w ⁱrepresent the weights of i-th neighborhood of pixel points image block in image to be split, x _ijrepresent the gray-scale value of the jth pixel in image to be split in i-th neighborhood of pixel points image block, v _1krepresent the center gray-scale value of kth class in initial cluster center, k=1,2 ..., c, G _kito represent in image to be split that i-th pixel belongs to the blur level of kth class, G _tito represent in image to be split that i-th pixel belongs to the blur level of t class, m represents Fuzzy Exponential, and value is 2.

9. the fuzzy C-mean algorithm image partition method based on average drifting and neighborhood information according to claim 1, is characterized in that: described in step (7), iteration stopping condition refers to the situation of one of following condition;

Condition 1, any one element value in absolute difference matrix is less than iteration stopping threshold value 0.01;

Condition 2, cycle index reaches maximum iteration time 500 times.