CN101923714A - Texture image segmentation method based on spatial weighting membership fuzzy c-mean value - Google Patents

Abstract

The invention discloses a texture image segmentation method based on a spatial weighting membership fuzzy c-mean value, which mainly solves the local mis-segmentation problem of texture image segmentation and improves the segmentation effect of edge binding sites. The method comprises the following steps of: (1) inputting an image to be segmented; (2) extracting the gray level co-occurrence feature and the wavelet feature of the image to be segmented; (3) adjusting the membership of a fuzzy c-mean value cluster according to spatial information and clustering; (4) optimizing the adjustment parameters during the clustering process by using a particle swarm optimization method; and (5) judging whether the set circulation stop condition is achieved or not, if not, then returning to the step (3) for the next circulation, or else exiting the circulation to obtain the final membership value, namely the final segmentation result of the image. Compared with the prior art, the method remarkably improves the segmentation performance of the image, and can effectively segment texture images and SAR (Synthetic Aperture Radar) images.

Description

Texture image segmenting method based on spatial weighting membership fuzzy c-mean value

Technical field

The invention belongs to image processing field, relate to image segmentation, can be used for Study Of Segmentation Of Textured Images and cutting apart the SAR image.

Background technology

Along with science and technology development, people more and more obtain various information with the form of image.Image partition method also becomes the focus of people's research.

Shen S and Liew A W have proposed based on fuzzy c-average (Fuzzy C-Means, FCM) improvement algorithm also is applied to medical image respectively and natural image is cut apart, obtained the better image segmentation result, but therefore can not obtain desirable segmentation result the tangible image of textural characteristics owing to only consider gray feature.

Oskoei MA and Omran MG are respectively with genetic algorithm (Genetic Algorithm, GA) and particle group optimizing (Particle Swarm Optimization, PSO) algorithm is introduced fuzzy clustering to realize the optimization of objective function, be applied to obtain when natural image is cut apart comparatively desirable segmentation result, improve the data clusters result of fuzzy clustering and be applied to image segmentation obtaining extensive concern at present by optimized Algorithm.

The SAR image segmentation that people such as Tian Xiaolin propose based on PSO optimization space constraint cluster, in the cutting apart of part SAR image, obtained ideal results, because it only uses multiple dimensioned gray feature, although still considered that spatial information can not obtain desirable segmentation result in the tangible SAR image of textural characteristics.The extraction of textural characteristics becomes the key of improving above algorithm in the image segmentation.

The spatial weighting degree of membership model that people such as Swagatam Das propose does not embody space adjacent pixel location and texture information and spatial information influence degree fully.

Above-mentioned method is not utilized textural characteristics and spatial information simultaneously, therefore can not obtain desirable segmentation result in the tangible image segmentation of textural characteristics.

Utilizing textural characteristics to carry out in the process of cluster, (Fuzzy C-Means, FCM) dividing method is not considered spatial information to traditional fuzzy c-average, therefore exists serious local wrong branch phenomenon, especially edge part misclassification to divide serious.

Summary of the invention

The objective of the invention is to overcome above-mentioned existing methods shortcoming, proposed a kind of texture image segmenting method,, improve the edge segmentation effect to improve the local wrong branch phenomenon of texture image based on spatial weighting membership fuzzy c-mean value.

For achieving the above object, the present invention includes following process:

(1) imports image to be split, and extract image texture features to be split;

(2) degree of membership of regulating this pixel fuzzy c-mean cluster with the textural characteristics information of the spatial information of the neighborhood territory pixel of each pixel and extraction, the degree of membership after adjusted is:

$u_{\mathrm{ij}}^{\′}=u_{\mathrm{ij}}{h}_{\mathrm{ij}}/{\mathrm{\Σ}}_{c=1}^C{u}_{\mathrm{cj}}{h}_{\mathrm{cj}}$

Wherein: u _IjPixel j before expression is regulated is under the jurisdiction of the degree of membership numerical value of i class;

C represents clusters number, u _CjPixel j before expression is regulated is under the jurisdiction of the degree of membership numerical value of c class;

U ' _IjBe with the degree of membership value of the pixel j after spatial information and the adjusting of neighborhood texture information with respect to the i class,

U is the matrix that the degree of membership after all pixels of image are regulated with spatial information and neighborhood texture information is formed;

h _CjRemarked pixel point j is under the jurisdiction of the spatial information and the neighborhood texture information adjustment factor of c class;

h _IjRemarked pixel point j is under the jurisdiction of the spatial information and the neighborhood texture information adjustment factor of i class, and its representation formula is: h _Ij=α u _{(D) ij}+ β u _{(T) ij}

In the formula, α is u _{(D) ij}The regulation and control parameter;

u _{(D) ij}Remarked pixel point j is under the jurisdiction of the spatial information weighting degree of membership of i class, and it is defined as:

$u_{\left(D\right)\mathrm{ij}}={\mathrm{\Σ}}_{k=1}^S{u}_{\mathrm{ik}}{d}_{\mathrm{jk}}^2/{\mathrm{\Σ}}_{k=1}^S{d}_{\mathrm{jk}}^2$

In the formula, s represents neighborhood territory pixel number, u _IkK the neighborhood of remarked pixel point j is for the degree of membership value of i class; d _JkSpace length between remarked pixel point j and its k the neighborhood;

β is u _{(T) ij}The regulation and control parameter;

u _{(T) ij}Remarked pixel point j is under the jurisdiction of the neighborhood texture information weighting degree of membership of i class, and it is defined as:

$u_{\left(T\right)\mathrm{ij}}={\mathrm{\Σ}}_{k=1}^S{u}_{\mathrm{ik}}{X}_{\mathrm{jk}}/{\mathrm{\Σ}}_{k=1}^S{X}_{\mathrm{jk}}$

In the formula, X _JkBe the Euclidean distance of the textural characteristics between pixel j and its k the neighborhood, X _Jk=|| X _j-X _k||, the X here _jAnd X _kThe texture feature vector of k the neighborhood of difference remarked pixel point j and pixel j;

(3) according to the degree of membership value u ' after regulating _Ij, utilize following formula to treat split image and carry out cluster:

$\min J_m(U,V)={\mathrm{\Σ}}_{j=1}^N{\mathrm{\Σ}}_{i=1}^C{\left(u_{\mathrm{ij}}^{\′}\right)}^m{d}^2(X_j,Z_i)$

Wherein: J _mThe objective function of expression fuzzy c-mean cluster, and m ∈ (1, ∞) control blur level weight index;

N presentation video pixel count, C represents clusters number;

X _jRepresent j pixel position textural characteristics;

Z _iThe cluster centre of representing the i class

The matrix that V is made up of cluster centre;

Be similarity measure, its representation formula is: d ²(X _j, Z _i)=|| X _j-Z _i||;

(4) by particle group optimizing regulation and control parameter alpha and β are upgraded, the maximum algebraically of optimization is 60;

(5) judge whether to reach the loop ends condition of setting,, then cut apart end, current degree of membership value as image segmentation result, is circulated otherwise forward step 2 to next time if reach the loop ends condition,

Described loop ends condition is set in 5 suboptimization and will satisfies

Wherein

The target function value in t generation that expression is optimized,

The target function value in t+1 generation that expression is optimized, the span of t is [1,59].

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

1) the present invention has been owing to utilized image texture features, and is more effective compared with carry out image segmentation with gray feature;

2) the present invention carries out simple cluster with textural characteristics, but by spatial information and neighborhood texture information the degree of membership of FCM is regulated, and has eliminated local wrong branch phenomenon substantially, has improved the image border segmentation effect;

3) the present invention regulates formula h owing to adopt particle group optimizing PSO algorithm to degree of membership in the FCM clustering algorithm _Ij=α u _{(D) ij}+ β u _{(T) ij}Middle u _{(D) ij}Factor alpha and u _{(T) ij}Factor beta regulate and control, help to obtain good segmentation result.

Description of drawings

Fig. 1 is realization flow figure of the present invention;

Fig. 2 is that the present invention synthesizes texture image SYN1 for three width of cloth, SYN2, SYN3 segmentation result and existing three kinds of method segmentation result comparison diagrams;

Fig. 3 be the present invention for three width of cloth SAR image SAR1, SAR2, SAR3 segmentation result and existing three kinds of method segmentation result comparison diagrams.

Embodiment

With reference to Fig. 1, specific implementation process of the present invention is as follows:

Step 1. input image to be split extracts image texture features to be split.

The textural characteristics that the present invention treats the split image extraction comprises gray scale symbiosis feature and wavelet character:

1.1) gray scale symbiosis feature extraction

In gray scale symbiosis Feature Extraction, at first calculate in the image to be split pixel with feature to be extracted and be center, the gray level co-occurrence matrixes P (u of the image subblock in the characteristic window of setting, v, d, θ), wherein u and v represent the gray scale of two pixels of statistics in the gray level co-occurrence matrixes calculating; θ represents to extract the direction of gray level co-occurrence matrixes, and θ is chosen as 4 discrete directions: 0 °, and 45 °, 90 °, 135 °; D represents the distance of two pixels of statistics in the gray level co-occurrence matrixes calculating, and value is 1 among the present invention.Texture feature extraction amount again on the basis of gray level co-occurrence matrixes is called second degree statistics.

These second degree statisticses below extracting from the gray level co-occurrence matrixes of image among the present invention are as the characteristic coefficient of Classification and Identification: (1) angle second moment:

(2) homogeney:

(3) contrast: L-1 represents the gray shade scale that quantizes.Wherein, the angle second moment is at 0 °, and 45 °, 90 °, the value on 135 ° is with (g ₁₀, g ₁₁, g ₁₂, g ₁₃) expression, homogeney is used (g on four direction ₂₀, g ₂₁, g ₂₂, g ₂₃) expression, contrast is used (g on four direction ₃₀, g ₃₁, g ₃₂, g ₃₃) expression.

1.2) the wavelet character extraction

When extracting wavelet character, at first treat in the split image pixel with feature to be extracted and be the center, image subblock in the characteristic window of setting carries out one-dimensional filtering respectively along x direction and y direction, each yardstick is resolved into four subband LL, HL, LH and HH, respectively the details on the low-frequency information of token image and level, the vertical and tilted direction.

By formula

Obtain the L1 norm of subband respectively, in the formula, w represents the L1 norm of subband, and M is the line number of sub-band coefficients, and N is the columns of sub-band coefficients, and M * N is the subband size, and m, n represent the index of coefficient in the subband, and (m n) is the coefficient value of the capable n row of m in this subband to coef.Through image is carried out L layer wavelet transformation, extract and obtain the L1 norm of 3L+1 subband, thereby obtain the proper vector (w that a 3L+1 ties up ₀, w ₁..., w _3L+1).

Experimental section of the present invention adopts 12 dimension gray scale symbiosis features for synthetic texture image and treats split image and decompose the 19 dimensional feature (g that the two-layer 7 dimension wavelet characters that obtain combine and obtain ₁₀..., g ₁₃, g ₂₀..., g ₂₃, g ₃₀..., g ₃₃, w ₀..., w ₆), abbreviate the G_W feature as.Only use for the SAR image and to treat split image and decompose the two-layer 7 dimension wavelet character (w that obtain ₀, w ₁..., w ₆), abbreviate the W feature as.

The degree of membership that step 2. is regulated this pixel fuzzy c-mean cluster with the textural characteristics information of the spatial information of the neighborhood territory pixel of each pixel and extraction, the adjusting formula of degree of membership is:

$u_{\mathrm{ij}}^{\′}=u_{\mathrm{ij}}{h}_{\mathrm{ij}}/{\mathrm{\Σ}}_{c=1}^C{u}_{\mathrm{cj}}{h}_{\mathrm{cj}}$

In the formula: u _IjPixel j before expression is regulated is under the jurisdiction of the degree of membership numerical value of i class;

C represents clusters number, u _CjPixel j before expression is regulated is under the jurisdiction of the degree of membership numerical value of c class;

U ' _IjBe with the degree of membership value of the pixel j after spatial information and the adjusting of neighborhood texture information with respect to the i class,

U is the matrix that the degree of membership after all pixels of image are regulated with spatial information and neighborhood texture information is formed;

h _CjRemarked pixel point j is under the jurisdiction of the spatial information and the neighborhood texture information adjustment factor of c class;

h _IjRemarked pixel point j is under the jurisdiction of the spatial information and the neighborhood texture information adjustment factor of i class, and its representation formula is:

h _ij＝α·u _(D)ij+β·u _(T)ij

In the formula: α is u _{(D) ij}The regulation and control parameter;

u _{(D) ij}Remarked pixel point j is under the jurisdiction of the spatial information weighting degree of membership of i class, u _{(D) ij}Expression be: In the formula, s represents neighborhood territory pixel number, u _IkRepresent that k neighborhood with spatial information and neighborhood texture information adjusting preceding pixel point j is under the jurisdiction of the degree of membership value of i class, d _JkBe the relative position between pixel j and its k the neighborhood, its representation formula is:

(ζ in the formula _j, η _j) be the coordinate of the pixel j that studied, (ζ _k, η _k) be the coordinate of k the neighborhood of pixel j;

β is u _{(T) ij}The regulation and control parameter;

u _{(T) ij}Remarked pixel point j is under the jurisdiction of the neighborhood texture information weighting degree of membership of i class, u _{(T) ij}Expression be: In the formula, X _Jk=|| X _j-X _k|| be the Euclidean distance of the textural characteristics between pixel j and its k the neighborhood, the X here _jAnd X _kThe texture feature vector of k the neighborhood of difference remarked pixel point j and pixel j.

Step 3. is utilized improved degree of membership value to treat split image by following formula to carry out cluster:

$\min J_m(U,V)={\mathrm{\Σ}}_{j=1}^N{\mathrm{\Σ}}_{i=1}^C{\left(u_{\mathrm{ij}}^{\′}\right)}^m{d}^2(X_j,Z_i)$

In the formula: J _mThe objective function of expression fuzzy c-mean cluster, and m ∈ (1, ∞) the weight index of control blur level;

N presentation video pixel count, C represents clusters number;

X _jRepresent j pixel position textural characteristics;

Z _iThe cluster centre of representing the i class

The matrix that V is made up of cluster centre;

d ²(X _j, Z _i) be similarity measure, the textural characteristics of remarked pixel point j and the Euclidean distance of the cluster centre, its representation formula is: d ²(X _j, Z _i)=|| X _j-Z _i||, X in the formula _jAnd X _kThe texture feature vector of k the neighborhood of difference remarked pixel point j and pixel j.

Step 4. is upgraded regulation and control parameter alpha and β by particle group optimizing PSO algorithm.

If the population scale of PSO is p _s, maximum population evolutionary generation is G _MaxIf i particle is expressed as k _i=(α _i, β _i), α _iBe the numerical value of α in i particle, β _iBe the numerical value of β in i particle, describe k for convenient _iUnified with Y _iExpression, individual body position Y _iBe exactly spatial information regulation and control parameters, Y _i=(y _I1, y _I2), the desired positions of its experience is designated as p _i=(p _I1, p _I2), be also referred to as P _BestThe desired positions that lives through at all particles of colony is called g _BestThe speed V of particle i _i=(v _I1, v _I2) expression, V _iBy a maximal rate V _MaxLimit.The built-up pattern of PSO algorithm is:

v _id＝wv _id+c ₁×rand(·)×(p _id-y _id)+c ₂×Rand(·)×(p _gd-y _id)

y _id＝y _id+v _id

Wherein: v _IdThe speed of the d of i particle of tabular form dimension is if current acceleration to particle causes its speed v in the d dimension _IdThe maximal rate v that surpasses this dimension _{Max, d}, then the speed of this dimension is restricted to this dimension maximal rate v _{Max, d}W is an inertia weight, and w is big, and algorithm has stronger ability of searching optimum, and the less then algorithm of w tends to Local Search, and the present invention is to make it by maximum initial value v to the obtaining value method of w _MaxWith the increase linear decrease of iterations to w _Minc ₁And c ₂Be aceleration pulse; p _IdThe desired positions of representing the d dimension of i particle; y _IdThe current location of representing the d dimension of i particle; Rand () and Rand () are two random functions that change in [0,1] scope; p _GdThe desired positions of representing the d dimension of all particles.

The step that PSO carries out spatial information regulation and control parameter optimization is as follows:

4a) if FCM is optimized first, for i individual, given position Y at random then _iAnd speed V _i, otherwise, get each individual position Y ' that upgrades _iWith speed V ' _i

4b) by formula

Calculate each individual target function value;

4c) the individual optimal objective function value that writes down in the target function value that each individuality is tried to achieve and its experience compares, if optimal result is better before for present target function value, the individual optimal objective function value of with replacement then, if the optimal objective function value of trying to achieve at present is better than colony's optimal objective function value, then colony's optimal objective function value is reset to present result;

4d) revise each individual position and speed in the population, the position Y ' that obtains upgrading according to the built-up pattern of PSO algorithm _iWith the speed V ' that upgrades _i, this built-up pattern is:

v _id＝wv _id+c ₁×rand(·)×(p _id-y _id)+c ₂×Rand(·)×(p _gd-y _id)

y _id＝y _id+v _id?。

The value of PSO optimization major parameter is as shown in table 1:

Table 1.PSO optimized Algorithm major parameter

Step 5. judges whether to reach the loop ends condition of setting, if reach the loop ends condition, then cuts apart end, current degree of membership value as image segmentation result, is circulated otherwise forward step 2 to next time,

Described loop ends condition is set in 5 suboptimization and will satisfies

Wherein The target function value in t generation that expression is optimized, The target function value in t+1 generation that expression is optimized, the span of t is [1,59].

Effect of the present invention can further specify by following experiment:

1. experiment condition and content

The experiment simulation environment is: MATLAB 7.0.4, Intel (R) Pentium (R) 1CPU 2.4GHz, WindowXP Professional.

Experiment content comprises: the present invention has done test experiments with the synthetic texture image of 3 width of cloth respectively, composograph has 3 classes, 4 classes, 5 classes, three width of cloth, size all is 256 * 256, difference called after: SYN1, SYN2, SYN3, feature adopts the G_W feature, gray scale symbiosis characteristic window size is 9 * 9, and the wavelet character window is 8 * 8, and the spatial neighborhood window size is 15 * 15.

The present invention has also done test experiments to three width of cloth SAR images, is divided into 3 classes respectively, 2 classes, 2 classes, the image size all is 256 * 256, difference called after: SAR1, SAR2, SAR3, adopt the W feature in the experiment,, under the prerequisite that guarantees segmentation effect, reduced calculated amount with respect to the G_W feature.Among the SAR1 in order to protect the runway detailed information to select less characteristic window 4 * 4 and less spatial neighborhood window 5 * 5.SAR2 and SAR3 adopt the general standard of SAR image segmentation, and the characteristic window size is 16 * 16, spatial neighborhood window 21 * 21.

2. experimental result

(1) with the present invention and Kmeans, FCM, three kinds of methods of PSO-SCFCM to SYN1, SYN2, the segmentation result of the synthetic texture image of SYN3 three width of cloth as shown in Figure 2, wherein figure (2a) is the original image of SYN1; Figure (2b) is the original image of SYN2; Figure (2c) is the original image of SYN3; Figure (2d) is the template of cutting apart of figure (2a); Figure (2e) is the template of cutting apart of figure (2b); Figure (2f) is the template of cutting apart of figure (2c); Figure (2g) is the segmentation result of existing Kmeans algorithm to figure (2a); Figure (2h) is the segmentation result of existing Kmeans algorithm to figure (2b); Figure (2i) is the segmentation result of existing Kmeans algorithm to figure (2c); Figure (2j) is the segmentation result of existing FCM algorithm to figure (2a); Figure (2k) is the segmentation result of existing FCM algorithm to figure (2b); Figure (2l) is the segmentation result of existing FCM algorithm to figure (2c); Figure (2m) is the segmentation result of existing PSO-SCFCM algorithm to figure (2a); Figure (2n) is the segmentation result of existing PSO-SCFCM algorithm to figure (2b); Figure (2o) is the segmentation result of existing PSO-SCFCM algorithm to figure (2c); Figure (2p) is that algorithm of the present invention is to figure (2a) segmentation result; Figure (2q) is the segmentation result of algorithm of the present invention to figure (2b); Figure (2r) is the segmentation result of algorithm of the present invention to figure (2c).

From the segmentation result of figure (2g), figure (2h) and figure (2i) as seen, though the Kmeans algorithm has utilized textural characteristics, owing to do not consider spatial information, so can not obtain desirable segmentation result.

From the segmentation result of figure (2j), figure (2k) and figure (2l) as seen, though the FCM algorithm has utilized textural characteristics, owing to do not consider spatial information, so can not obtain desirable segmentation result.

From figure (2m), figure (2n) and figure (2o) segmentation result as seen, the PSO-SCFCM algorithm since the employing be gray feature, so the tangible image of textural characteristics is not had correct segmentation effect.

From the segmentation result of figure (2p), figure (2q) and figure (2r) as seen, the inventive method has more satisfactory branch result to synthesizing texture image.

Table 2 has provided algorithms of different to SYN1, SYN2, the SYN3 segmentation result, the number percent of data representation misclassification pixel number and total number of image pixels in the table 2, mistake is divided number of pixels/total number of image pixels * 100%, as can be known from Table 2, segmentation result of the present invention is compared with three kinds of existing algorithm segmentation results, and mistake branch rate obviously reduces.

The synthetic texture image mistake of table 2. divides rate relatively

(2) with the present invention and Kmeans, FCM, three kinds of methods of PSO-SCFCM to SAR1, SAR2, the segmentation result of SAR3 three width of cloth SAR images as shown in Figure 3, wherein figure (3a) is the original image of SAR1; Figure (3b) is the original image of SAR2; Figure (3c) is the original image of SAR3; Figure (3d) is the segmentation result of existing Kmeans algorithm to figure (3a); Figure (3e) is the segmentation result of existing Kmeans algorithm to figure (3b); Figure (3f) is the segmentation result of existing Kmeans algorithm to figure (3c); Figure (3g) is the segmentation result of existing FCM algorithm to figure (3a); Figure (3h) is the segmentation result of existing FCM algorithm to figure (3b); Figure (3i) is the segmentation result of existing FCM algorithm to figure (3c); Figure (3j) is the segmentation result of existing PSO-SCFCM algorithm to figure (3a); Figure (3k) is the segmentation result of existing PSO-SCFCM algorithm to figure (3b); Figure (3l) is the segmentation result of existing PSO-SCFCM algorithm to figure (3c); Figure (3m) is that algorithm of the present invention is to figure (3a) segmentation result; Figure (3n) is the segmentation result of algorithm of the present invention to figure (3b); Figure (3o) is the segmentation result of algorithm of the present invention to figure (3c).

From the segmentation result of figure (3d), figure (3e) and figure (3f) as seen, though the Kmeans algorithm has utilized textural characteristics, owing to do not consider spatial information, so can not obtain desirable segmentation result.

From the segmentation result of figure (3g), figure (3h) and figure (3i) as seen, though the FCM algorithm has utilized textural characteristics, owing to do not consider spatial information, so can not obtain desirable segmentation result.

From figure (3j), figure (3k) and figure (3l) segmentation result as seen, the PSO-SCFCM algorithm since the employing be gray feature, so the tangible SAR image of textural characteristics is not had correct segmentation effect.

From figure (3m), figure (3n) and the segmentation result of scheming (3o) as seen, the inventive method has more satisfactory branch result to the tangible SAR image of textural characteristics.

To sum up, the texture image segmenting method that the present invention proposes based on spatial weighting membership fuzzy c-mean value, by the degree of membership of regulating fuzzy C-average FCM clustering algorithm, eliminated the local wrong branch phenomenon of image substantially with the spatial information that relative position and texture information constituted of neighbor.Spatial information is finished by particle group optimizing PSO algorithm the regulation and control of FCM clustering algorithm, and the regulation and control parameter of optimization helps to obtain good segmentation result.

Claims (2)

1. texture image segmenting method based on spatial weighting membership fuzzy c-mean value comprises following steps:

(1) imports image to be split, and extract image texture features to be split;

(2) degree of membership of regulating this pixel fuzzy c-mean cluster with the textural characteristics information of the spatial information of the neighborhood territory pixel of each pixel and extraction, the degree of membership after adjusted is:

$u_{\mathrm{ij}}^{\′}=u_{\mathrm{ij}}{h}_{\mathrm{ij}}/{\mathrm{\Σ}}_{c=1}^C{u}_{\mathrm{cj}}{h}_{\mathrm{cj}}$

Wherein: u _IjPixel j before expression is regulated is under the jurisdiction of the degree of membership numerical value of i class;

C represents clusters number, u _CjPixel j before expression is regulated is under the jurisdiction of the degree of membership numerical value of c class;

U ' _IjBe with the degree of membership value of the pixel j after spatial information and the adjusting of neighborhood texture information with respect to the i class,

U is the matrix that the degree of membership after all pixels of image are regulated with spatial information and neighborhood texture information is formed;

h _CjRemarked pixel point j is under the jurisdiction of the spatial information and the neighborhood texture information adjustment factor of c class;

h _IjRemarked pixel point j is under the jurisdiction of the spatial information and the neighborhood texture information adjustment factor of i class, and its representation formula is: h _Ij=α u _{(D) ij}+ β u _{(T) ij}

In the formula, α is u _{(D) ij}The regulation and control parameter;

u _{(D) ij}Remarked pixel point j is under the jurisdiction of the spatial information weighting degree of membership of i class, and it is defined as:

In the formula, s represents neighborhood territory pixel number, u _IkK the neighborhood of remarked pixel point j is for the degree of membership value of i class; d _JkSpace length between remarked pixel point j and its k the neighborhood;

β is u _{(T) ij}The regulation and control parameter;

u _{(T) ij}Remarked pixel point j is under the jurisdiction of the neighborhood texture information weighting degree of membership of i class, and it is defined as:

In the formula, X _JkBe the Euclidean distance of the textural characteristics between pixel j and its k the neighborhood, X _Jk=|| X _j-X _k||, the X here _jAnd X _kThe texture feature vector of k the neighborhood of difference remarked pixel point j and pixel j;

(3) according to the degree of membership value u ' after regulating _Ij, utilize following formula to treat split image and carry out cluster:

$\min J_m(U,V)={\mathrm{\Σ}}_{j=1}^N{\mathrm{\Σ}}_{i=1}^C{\left(u_{\mathrm{ij}}^{\′}\right)}^m{d}^2(X_j,Z_i)$

Wherein: J _mThe objective function of expression fuzzy c-mean cluster, and m ∈ (1, ∞) control blur level weight index;

N presentation video pixel count, C represents clusters number;

X _jRepresent j pixel position textural characteristics;

Z _iThe cluster centre of representing the i class

The matrix that V is made up of cluster centre;

d ²(X _j, Z _i) be similarity measure, its representation formula is: d ²(X _j, Z _i)=|| X _j-Z _i||;

(4) by particle group optimizing regulation and control parameter alpha and β are upgraded, the maximum algebraically of optimization is 60;

(5) judge whether to reach the loop ends condition of setting,, then cut apart end, current degree of membership value as image segmentation result, is circulated otherwise forward step 2 to next time if reach the loop ends condition.

2. according to claims 1 described method, wherein the described loop ends condition of step (5) is set in 5 suboptimization and will satisfies

Wherein

The target function value in t generation that expression is optimized,

The target function value in t+1 generation that expression is optimized, the span of t is [1,59].

Images