CN101667292A - SAR image segmentation system and segmentation method based on immune clone and projection pursuit - Google Patents

Abstract

The invention discloses an SAR image segmentation system and a segmentation method based on immune clone and projection pursuit. The system comprises an image characteristic-extracting module, an initial label-selecting submodule, a projection direction-selecting submodule and a subspace clustering submodule, wherein the image characteristic-extracting module extracts the gray co-occurrence matrixcharacteristics, the wavelet characteristics, the brushlet characteristics and the contourlet characteristics of an input image; the initial label-selecting submodule clusters the image characteristics to acquire and transmit an initial label to the projection direction-selecting submodule for calculating a linear judgment analysis projection index and acquiring an optimal projection direction; the subspace clustering submodule projects the image characteristics in the optimal projection direction, acquires and clusters an optimal subspace to acquire a clustering label, returns the clusteringlabel to the initial label-selecting submodule for iteration, acquires a final clustering label corresponding to image pixels and acquires a final image segmentation result. The invention has the advantage of high segmentation accuracy and can be applied to civil and industrial fields or as martial reconnaissance means.

Description

SAR image segmentation system and dividing method based on immune clone and projection pursuit

Technical field

The invention belongs to technical field of image processing, relate to the SAR image segmentation, can be used for military surveillance means and civilian and industrial circle.

Background technology

SAR is as a kind of indispensable military surveillance means, civilian and industrially also have been widely used.Growing along with the SAR imaging technique, the SAR view data that is obtained is more and more, and the machine decipher has progressively replaced artificial decipher, thus the SAR Flame Image Process also becomes the research focus.The SAR image segmentation is as one of basic problem of SAR Flame Image Process, it is in research and application to the SAR image, can find interested target area by image segmentation, thereby for the classification and the identification in SAR image later stage lays the foundation, and the accuracy of Target Recognition depends on the quality of image segmentation to a great extent.For the target in identification and the analysis SAR image need be come out they separation and Extraction from image, just might further target be measured and the SAR image be utilized on this basis.For the SAR image, because the non-stationary degree of the similar atural object scene of influence of speckle noise is stronger, be difficult to choose representational sample effectively and be used for parameter training, clustering algorithm is as unsupervised learning method, the advantage of its self-adaptation and practicality makes it become the Study of Image Segmentation focus.Present stage, the SAR image partition method based on clustering algorithm mainly contained K means clustering algorithm and spectral clustering algorithm, wherein K mean cluster SAR image partition method unstable result and segmentation precision are not high, can not effectively handle extensive image segmentation problem based on the SAR image partition method of spectral clustering.

Summary of the invention

The object of the invention is to overcome the shortcoming of above-mentioned prior art, has proposed a kind of SAR image segmentation system and dividing method based on immune clone and projection pursuit, to realize directly characteristics of image being operated raising image segmentation precision.

For achieving the above object, SAR image segmentation system of the present invention comprises:

The image characteristics extraction module is finished gray level co-occurrence matrixes, small echo, brushlet and contourlet feature extraction to the image of input, and its extraction result is sent to the PROJECTION PURSUIT CLUSTER ON module;

The PROJECTION PURSUIT CLUSTER ON module adopts immune clone algorithm to carry out the self-adaptation PROJECTION PURSUIT CLUSTER ON to the characteristics of image of importing, and the cluster label is corresponding with image pixel, obtains image segmentation result.

Described PROJECTION PURSUIT CLUSTER ON module comprises:

Initial labels chooser module is used for carrying characteristics of image, carries out cluster and obtains initial label, and give projecting direction chooser module with tag transmits;

Projecting direction chooser module is calculated linear discriminant analysis projection index in the projection pursuit algorithm according to input label, and optimizes this projection index by immune clone algorithm, obtains optimum projecting direction and reaches the subspace clustering submodule;

The subspace clustering submodule, characteristics of image is mapped to optimal projection direction, obtain optimal subspace, optimal subspace is carried out cluster, obtain the cluster label, the cluster label is back to initial labels chooser module, iteration for several times, obtain final cluster label, final cluster label is corresponding with image pixel, obtain the final image segmentation result.

For achieving the above object, SAR image partition method of the present invention comprises the steps:

(1) image to input extracts gray level co-occurrence matrixes, small echo, brushlet and contourlet feature, and the characteristics of image that extracts is defined as X _Ij, wherein, i represents the number of sample, is expressed as i pixel in image, j represents the dimension of sample, is expressed as the j dimensional feature of pixel in image;

(2) the characteristics of image X to extracting _IjCarry out normalization, set label iterations g;

(3) with the characteristics of image X ' after the normalization _Ij, accidental projection to a low n-dimensional subspace n U, and this subspace U carried out the K mean cluster, obtain an initial labels H;

(4) according to the initial labels H that obtains, calculate the linear discriminant analysis projection index of projection pursuit, and optimize this projection index by immune clone algorithm, obtain optimum projecting direction;

(5) with the characteristics of image X ' after the normalization _Ij, project to optimal projection direction, obtain optimum subspace U ';

(6) subspace U ' is carried out the K mean cluster, obtain new label H ';

(7) with new label H ', return step (4) and carry out iteration, through after the iteration for several times as initial labels H, if obtain new label H ' in iterative process, reach convergence, output convergent label H ", if in the not convergence of g back of iteration, then the label H of the last acquisition of output ';

(8) with the convergent label H " or the last label H that obtains ' corresponding with the pixel of input picture, the output image segmentation result.

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

1. the present invention is owing to adopt immune clone algorithm to optimize the linear discriminant analysis projection index of projection pursuit, thereby can obtain optimum projecting direction and optimal subspace;

2. the present invention has improved the precision of image segmentation owing to adopt the linear discriminant analysis projection index and the K means clustering algorithm of projection pursuit to carry out adaptive iteration;

Simulation result shows that the present invention improves than K mean algorithm precision.

Description of drawings

Fig. 1 SAR image segmentation system of the present invention synoptic diagram;

Fig. 2 is a SAR image partition method process flow diagram of the present invention;

Fig. 3 is the former figure of SAR image that the present invention adopts;

Fig. 4 is the figure as a result that adopts the K means clustering algorithm that Fig. 3 is cut apart;

Fig. 5 is the figure as a result that adopts the present invention that Fig. 3 is cut apart.

Embodiment

With reference to Fig. 1, the SAR image segmentation system that the present invention is based on immune clone and projection pursuit comprises: image characteristics extraction module and PROJECTION PURSUIT CLUSTER ON module.Wherein the PROJECTION PURSUIT CLUSTER ON module comprises initial labels chooser module, projecting direction chooser module and subspace clustering submodule.

The image characteristics extraction module is finished gray level co-occurrence matrixes, small echo, brushlet and contourlet feature extraction to the image of input, and it is extracted the result reaches initial labels chooser module in the PROJECTION PURSUIT CLUSTER ON module; Initial labels chooser module to carrying characteristics of image, is carried out simple clustering and is obtained initial label, and give projecting direction chooser module with tag transmits; Projecting direction chooser module, according to the linear discriminant analysis projection index in the input label calculating projection pursuit algorithm, and, obtain optimum projecting direction, and the gained projecting direction is reached the subspace clustering submodule by immune clone algorithm optimization projection index; The subspace clustering submodule, characteristics of image is mapped to optimal projection direction, obtain optimal subspace, optimal subspace is carried out cluster, obtain the cluster label, the cluster label is back to initial labels chooser module, iteration for several times, obtain final cluster label, final cluster label is corresponding with image pixel, obtain the final image segmentation result.

With reference to Fig. 2, SAR image segmentation system and dividing method based on immune clone and projection pursuit of the present invention comprise the steps:

Step 1: gray level co-occurrence matrixes, small echo, brushlet and the contourlet feature of extracting input picture.

1a) the original SAR image of input, its size is M * M ', as shown in Figure 3.This input picture is four width of cloth SAR images, and its size is respectively 256 * 256;

1b) the original SAR image to input extracts gray level co-occurrence matrixes feature 8 dimensions, two-layer wavelet character 7 dimensions, three layers of contourlet feature of two-layer brushlet feature 8 peacekeepings, 17 dimensions.The characteristics of image that extracts is defined as X _Ij, wherein, i represents the number of sample, is expressed as i pixel in image, j represents the dimension of sample, is expressed as the j dimensional feature of pixel in image.

Step 2: to the characteristics of image X that extracts _IjCarry out normalization, set label iterations g.

2a) characteristics of image X to extracting _IjAdopt following formula to carry out normalization:

X′ _ij＝(X _ij-X _jmin)/(X _jmax-X _jmin)

X wherein _JmaxAnd X _JminBe respectively the maximin of j dimension, X ' _IjBe the characteristics of image after the normalization;

2b) setting label iterations g is 10 times.

Step 3: with the characteristics of image X ' after the normalization _Ij, accidental projection to a low n-dimensional subspace n U, and this subspace U carried out the K mean cluster, obtain an initial labels H.

3a) to the characteristics of image X ' after the normalization _IjAdopt following formula to carry out accidental projection:

U＝a _rX′ _ij

A wherein _rIt is the projecting direction that produces at random;

3b) the low n-dimensional subspace n U to obtaining behind the accidental projection, adopt following formula to carry out the K mean cluster:

$\min J_K=\underset{i=1}{\overset{K}{\mathrm{\Σ}}}\underset{j=1}{\overset{n_r}{\mathrm{\Σ}}}{\left|\right|x_j-m_i\left|\right|}^2$

Wherein, J _KBe the interior divergence of class of pixel characteristic among the low n-dimensional subspace n U that obtains behind the K class accidental projection, K is the number of classification, x _jBe the pixel of k class, m _iBe the average of k class pixel, n _iThe number of pixel in each class;

3c) calculate each pixel respectively to m _iEuclidean distance, guaranteeing J _KMinimum prerequisite is little, and the classification number that will have minimum euclidean distance is given the label of pixel, obtains an initial labels H.

Step 4: setup parameter also generates population according to stochastic parameter, the target that projecting direction is optimized as immune clone algorithm.

4a) setting evolutionary generation gn is 500, and population size N is 10, the variation Probability p _mBe 0.9, clone's scale n _cBe 2;

4b) according to population size N, produce initial population A size at random and be m * N, wherein m is the characteristics of image dimension.

Step 5:, calculate the ideal adaptation degree among the initial population A according to acquisition initial labels H.

5a) according to acquisition initial labels H, adopt following formula to calculate between class scatter S _bWith divergence S in the class _w:

$S_b=\underset{i=1}{\overset{c}{\mathrm{\Σ}}}{n}_i({\stackrel{\‾}{X}}_i-\stackrel{\‾}{X}){({\stackrel{\‾}{X}}_i-\stackrel{\‾}{X})}^T$

$S_w=\underset{i=1}{\overset{c}{\mathrm{\Σ}}}\underset{j=1}{\overset{n_i}{\mathrm{\Σ}}}(X_{\mathrm{ij}}^{\′}-{\stackrel{\‾}{X}}_i){(X_{\mathrm{ij}}^{\′}-{\stackrel{\‾}{X}}_i)}^T$

X wherein _iBe i class characteristics of image X ' _IjAverage, promptly X is the characteristics of image X ' after the normalization _IjGrand mean, promptly

Wherein

n _iBe i class characteristics of image X ' _IjThe number of middle pixel, n is the characteristics of image X ' after the normalization _IjIn the number of total pixel, c is the classification number.

5b) according to the between class scatter S that obtains _bWith divergence S in the class _w, adopt each the individual fitness among the following formula calculating initial population A:

$f\left(a\right)=\left\{\begin{array}{cc}1-\frac{\left|a^T{S}_{w}a\right|}{\left|a^T{S}_{t}a\right|}& \mathrm{for}\left|a^T{S}_{t}a\right|\≠0\\ 0& \mathrm{for}\left|a^T{S}_{t}a\right|=0\end{array}\right.$

S wherein _t=S _b+ S _wBe total divergence, a is an any individual among the initial population A.

Step 6: according to clone's scale n _c, initial population A is carried out clone operations, the population A after obtaining to clone _c

6a) according to the clone's scale n that sets _c, adopt following formula to the initial population clone operations:

$T_c^C\left(a\right)=[{a,a}_1,a_2,...,a_{n_c}]$

Wherein, T _c ^CExpression clone operator, a is any one individuality among the initial population A, a ₁, a ₂...,

Being respectively a scale of carrying out is n _cClone operations after clone's individuality;

6b) each individuality among the initial population A is carried out clone operations, the population A after obtaining to clone _c

Step 7: according to the variation Probability p _m, to cloning the back population A _cCarry out mutation operation, the population A after obtaining to make a variation _m

7a) according to the variation Probability p of setting _m, adopt following formula to cloning the back population A _cThe individuality of middle variation carries out mutation operation:

β′＝β·exp(τ′N(0，1)+τN(0，1))

s′＝s+β′N(0，1)

Wherein, T _m ^CThe expression mutation operator, s is a Probability p _mThe variation individuality of selecting, s ' are individual after the individual corresponding variation of selecting of variation, and τ and τ are respectively

With

β and β ' are setup parameter, and N (0,1) is that to satisfy average be 0, and variance is 1 normal random variable;

7b) to cloning the back population A _cIn all variation individualities finish mutation operation, not have the individuality of variation to remain unchanged, the population A after making a variation _m

Step 8: calculate the population A after making a variation _mThe fitness function value, with population A _mIn among ideal adaptation degree and the initial population A ideal adaptation degree compare, select the bigger individuality of fitness successively, obtain new population A '.

8a) relatively the variation after population A _mIn corresponding original ideal adaptation degree among ideal adaptation degree and the initial population A, as if f (T _m ^C(a))＞and f (a), then select T _m ^C(a) be new population A ', otherwise select a be new population A ', wherein,

Be fitness function, T _s ^COperator, T are selected in expression _m ^C(a) be population A _mIn individual, a be an individuality among the corresponding initial population A.

8b) compare population A _mWith corresponding all ideal adaptation degree among the initial population A, select the bigger individuality of fitness and be new population A '.

Step 9:, carry out iteration according to the evolutionary generation gn that sets with new population A ', return step 5 as initial population A.

Step 10: in evolutionary generation gn, if the convergence of the maximum adaptation degree functional value in each generation, then termination of iterations is exported maximum adaptation degree functional value.

Step 11: with the characteristics of image X ' after the normalization _Ij, adopt following formula to project to optimum projecting direction, obtain optimum subspace U ':

U′＝a′X′ _ij

Wherein a ' is the projecting direction by the optimum of immune clone algorithm selection.

Step 12: the low n-dimensional subspace n U ' that obtains after the projection is carried out the K mean cluster, obtain new label H '.

12a) the low n-dimensional subspace n U ' to obtaining after the projection, adopt following formula to carry out the K mean cluster:

$\min J_K=\underset{i=1}{\overset{K}{\mathrm{\Σ}}}\underset{j=1}{\overset{n_i}{\mathrm{\Σ}}}{\left|\right|x_j-m_i\left|\right|}^2$

Wherein, J _KBe the interior divergence of class of the low middle pixel characteristic of n-dimensional subspace n U ' of K class, K is the number of classification, x _jBe the pixel of k class, m _iBe the average of k class pixel, n _iThe number of pixel in each class;

12b) calculate each pixel respectively to m _iEuclidean distance, guaranteeing J _KMinimum prerequisite is little, and the classification number that will have minimum euclidean distance is given the label of pixel, obtain new label H '.

Step 13: with new label H ', return step 5 and carry out iteration, through after the iteration repeatedly as initial labels H, if obtain new label H ' in iterative process, reach convergence, output convergent label H ", if not convergence behind the iterations g that sets, then the last label H that obtains of output ';

Step 14: with the convergent label H " or the last label H that obtains ' corresponding with the pixel of input picture, the output image segmentation result.

Effect of the present invention can further specify the SAR image segmentation by following:

1, simulated conditions

Emulation of the present invention is at windows XP, and SPI, CPU Pentium (R) 4, basic frequency 2.4GHZ, software platform are the Matlab7.0.1 operation.The former figure of four width of cloth SAR images is selected in emulation for use, and as Fig. 3, wherein Fig. 3 (a) is trees, bridge, field and river, and Fig. 3 (b) is land and lake, and Fig. 3 (c) is land and river, and Fig. 3 (d) is trees and field.

2, emulation content

(1) respectively Fig. 3 (a), Fig. 3 (b), Fig. 3 (c) and Fig. 3 (d) are carried out image segmentation with existing K Mean Method.Simulation result as shown in Figure 4.Wherein, Fig. 4 (a) is the segmentation result figure to Fig. 3 (a), and Fig. 4 (b) is the segmentation result figure to Fig. 3 (b), and Fig. 4 (c) is the segmentation result figure to Fig. 3 (c), and Fig. 4 (d) is the segmentation result figure to Fig. 3 (d);

(2) respectively Fig. 3 (a), Fig. 3 (b), Fig. 3 (c) and Fig. 3 (d) are carried out image segmentation with method of the present invention.Simulation result as shown in Figure 5 wherein, Fig. 5 (a) is the segmentation result figure to Fig. 3 (a), Fig. 5 (b) is the segmentation result figure to Fig. 3 (b), Fig. 5 (c) is the segmentation result figure to Fig. 3 (c), Fig. 5 (d) is the segmentation result figure to Fig. 3 (d).

3, analysis of simulation result

More as can be seen, Fig. 5 (a) can be partitioned into the shaded side of trees below the bridge from the segmentation result figure of Fig. 4 (a) and Fig. 5 (a), and Fig. 4 (a) is not partitioned into.

More as can be seen, the wrong branch in the lower left corner, Fig. 5 (b) image land will be less than the segmentation result of Fig. 4 (b) from the segmentation result figure of Fig. 4 (b) and Fig. 5 (b).

From the segmentation result figure of Fig. 4 (c) and Fig. 5 (c) more as can be seen, Fig. 5 (c) the land part assorted put and the river width close with former figure, and the image top of Fig. 4 (c) has the river width of a spot of assorted point and segmentation result greater than original graph 3 (c).

More as can be seen, the segmentation result that wrong branch is less than Fig. 4 (d) is located in Fig. 5 (d) upper left corner and field, the lower right corner, and the lower limb of the trees of Fig. 5 (d) is smooth than Fig. 4 (d) from the segmentation result figure of Fig. 4 (d) and Fig. 5 (d).

By Fig. 4 and Fig. 5 more as can be seen the present invention have advantage of high segmentation accuracy.

This example is being to implement under the prerequisite with the technical solution of the present invention, provided detailed embodiment and concrete operating process, but protection scope of the present invention is not limited to the foregoing description.

Claims (4)

1, a kind of SAR image segmentation system based on immune clone and projection pursuit comprises:

The image characteristics extraction module is finished gray level co-occurrence matrixes, small echo, brushlet and contourlet feature extraction to the image of input, and its extraction result is reached the PROJECTION PURSUIT CLUSTER ON module;

The PROJECTION PURSUIT CLUSTER ON module adopts immune clone algorithm to carry out the self-adaptation PROJECTION PURSUIT CLUSTER ON to the characteristics of image of importing, and the cluster label is corresponding with image pixel, obtains image segmentation result.

2. according to claims 1 described SAR image segmentation system, wherein the PROJECTION PURSUIT CLUSTER ON module comprises:

Initial labels chooser module is used for carrying characteristics of image, carries out cluster and obtains initial label, and give projecting direction chooser module with tag transmits;

Projecting direction chooser module is calculated linear discriminant analysis projection index in the projection pursuit algorithm according to input label, and optimizes this projection index by immune clone algorithm, obtains optimum projecting direction and reaches the subspace clustering submodule;

The subspace clustering submodule, characteristics of image is mapped to optimal projection direction, obtain optimal subspace, optimal subspace is carried out cluster, obtain the cluster label, the cluster label is back to initial labels chooser module, iteration for several times, obtain final cluster label, final cluster label is corresponding with image pixel, obtain the final image segmentation result.

3. the SAR image partition method based on immune clone and projection pursuit comprises the steps:

(1) image to input extracts gray level co-occurrence matrixes, small echo, brushlet and contourlet feature, and the characteristics of image that extracts is defined as X _Ij, wherein, i represents the number of sample, is expressed as i pixel in image, j represents the dimension of sample, is expressed as the j dimensional feature of pixel in image;

(2) the characteristics of image X to extracting _IjCarry out normalization, set label iterations g;

(3) with the characteristics of image X ' after the normalization _Ij, accidental projection to a low n-dimensional subspace n U, and this subspace U carried out the K mean cluster, obtain an initial labels H;

(4) according to the initial labels H that obtains, calculate the linear discriminant analysis projection index of projection pursuit, and optimize this projection index by immune clone algorithm, obtain optimum projecting direction;

(5) with the characteristics of image x ' after the normalization _Ij, project to optimum projecting direction, obtain optimum subspace U ';

(6) subspace U ' is carried out the K mean cluster, obtain new label H ';

(7) with new label H ', return step (4) and carry out iteration, through after the iteration for several times as initial labels H, if obtain new label H ' in iterative process, reach convergence, output convergent label H ", if in the not convergence of g back of iteration, then the label H of the last acquisition of output ';

(8) with the convergent label H " or the last label H that obtains ' corresponding with the pixel of input picture, the output image segmentation result.

4. according to claims 3 described SAR image partition methods, wherein step (4) is described optimizes this projection index by immune clone algorithm, carries out according to the following procedure:

4a) set population size N, evolutionary generation gn, clone's scale n _c, the variation Probability p _m

4b), produce initial population A at random according to population size N;

4c) the ideal adaptation degree of calculating initial population A;

4d) according to clone's scale n _c, initial population is cloned the population A after obtaining to clone _c

4e) to cloning the back population A _c, according to the variation Probability p _m, carry out mutation operation, the population A after obtaining to make a variation _m

4f) the population A behind the calculating clonal vaviation _mThe fitness function value, with population A _mIn among ideal adaptation degree and the initial population A ideal adaptation degree compare, select the bigger individuality of fitness successively, obtain new population A ' and export maximum adaptation degree functional value, with new population A ', return step 4c as initial population A);

4g) in generation, restrain for the maximum adaptation degree functional value of output as if each, then termination of iterations at evolution gn.

Images