CN105469408A - Building group segmentation method for SAR image - Google Patents

Abstract

The invention discloses a building group segmentation method for an SAR image, and the method comprises the steps: firstly carrying out the preprocessing of an original SAR image, and enabling the characteristics of the SAR image to be consistent; secondly carrying out the variation function feature extraction of the SAR image after preprocessing, obtaining a variation function feature graph, wherein a building group in the variation function feature graph has gray scale homogeneity; thirdly carrying out the segmentation of the building group based on the variation function feature graph through employing a CV model image segmentation method, and obtaining a binary image which can discriminate building regions and non-building regions; communicating adjacent building regions in the binary image through employing a communication region detection algorithm, enabling the adjacent discrete building regions to be integrated, and then removing false alarm segmentation caused by other ground objects through area screening; and finally extracting the contour of the building regions in the binary image, and obtaining an explicit image segmentation result. The method solves a problem that a segmentation effect of the building group in the SAR image is not good, and achieves an effect of effective segmentation.

Description

A kind of SAR image groups of building dividing method

Technical field

The invention belongs to SAR image process and automatic interpretation field, relate to a kind of SAR image groups of building dividing method.

Background technology

Groups of building segmentation in complex scene SAR image, is the major issue of terrain classification, military and civilian has extremely positive meaning.Because the building area in SAR image is textured, shade is obvious, does not have neat edge, if with traditional image partition method, be difficult to process.

In recent years, the Iamge Segmentation of based upon activities skeleton pattern starts to rise.Movable contour model is roughly divided into parametric active contour model and geometric active contour model, and level set movable contour model is the important one in geometric active contour model.The outline line of parametric active contour model operation parameter shows and carries out curve evolvement.Level set movable contour model uses Level Set Method, outline line is converted into a level set of three-dimension curved surface, carries out the curve evolvement of implicit expression.Hiddenly expression is shown, so level set movable contour model can be good at the change in topology processing curve evolvement, as Abruption and mergence etc. because curve evolvement have employed.

In level set movable contour model, the CV models applying proposed by Chan and Vese is extensive.This model mainly employs Level Set Method and Mumford-Shah model theory, has the detection and positioning ability of weak edge, perception edge and internal edge.In recent years, there is a large amount of achievements in research of carrying out SAR image segmentation based on CV model method both at home and abroad, the multiple atural objects such as maneuvering target, lake, farmland can be partitioned into.

In the groups of building segmentation of SAR image, CV model method is directly used to acquire a certain degree of difficulty, because CV model method requires that two class regions to be split have gray scale homogeney in respective region.In the SAR image of scene complexity, include groups of building and other complicated atural objects, also noisy interference, region homogeney is very poor, cannot meet the requirement of CV model method.If directly CV model method to be applied to the segmentation of SAR image groups of building, then cannot obtain desirable segmentation effect.

Because groups of building are usually expressed as periodic textural characteristics in SAR image, the textural characteristics of image can be utilized to carry out groups of building segmentation, and the means of feature extraction comprise a lot of methods of the texture measure such as variogram, MRF model analysis, wavelet analysis, FRACTAL DIMENSION feature.When variogram characteristic pattern is wherein for describing SAR image, built-up area can be made to occur good gray scale homogeney, being suitable for using CV model method to carry out the segmentation of SAR image groups of building.

Therefore, variogram feature being combined with CV model image dividing method, carrying out the groups of building segmentation of SAR image, is a problem needing research.

Summary of the invention

Goal of the invention: in order to overcome the deficiencies in the prior art, the invention provides a kind of method splitting groups of building in SAR image, the method extracts the variogram characteristic pattern of actual measurement SAR image, achieve the gray scale homogeney of construction area in SAR image, and as the input data that CV model image is split, efficiently solve the problem that groups of building in SAR image are difficult to split.

Technical scheme: for achieving the above object, the SAR image groups of building dividing method based on variogram CV model of the present invention, comprises the steps:

1) pre-service is carried out to original SAR image, comprise links such as eliminating coherent spot, adjustment brightness and contrast, picture element interpolation or extraction, its properties and characteristics is reached unanimity;

2) based on step 1) the pre-service SAR image that obtains, carry out variogram feature extraction, form variogram characteristic pattern, realize the gray scale homogeney of built-up area in SAR image;

3) based on step 2) the variogram characteristic pattern that obtains, adopt CV model image dividing method to carry out groups of building segmentation, obtain the binary map that can be distinguished construction area and territory, non-building area;

4) based on step 3) binary map that obtains, adopt connected region detection algorithm to be communicated with adjacent construction area, adjacent scrappy construction area is integrated, then remove by area screening the false-alarm segmentation that other atural object causes;

5) based on step 4) binary map that obtains, extract the outline line of construction area, obtain explicit image segmentation result.

The step 2 of the inventive method) in, when calculating variogram textural characteristics figure, adopt FAST RECURSIVE ALGORITHM to calculate textural characteristics, significantly to reduce calculated amount, the mean value simultaneously getting the textural characteristics on 0 °, 45 °, 90 °, 135 ° four directions is final textural characteristics value.

The step 3 of the inventive method) in, using the driving-energy of variogram feature as level set movements, set up corresponding CV model energy functional, specifically comprise the following steps:

(1) initialization level set function: zero level collection profile is initialized as equally distributed circle in SAR image, according to length and the width of image, determine centre coordinate and the radius of each circle, initial profile is made to pass through each building area, by level set function φ (x, y) value is defined as symbolic measurement, on each circular contour, level set function φ (x, y)=0, in the inside of each circular contour, φ (x, y) >0; Other regions on image, level set function φ (x, y) <0;

(2) zones of different gray average is calculated: calculate the curved exterior gray average C formed by level set function φ (x, y) ₀with with interior intensity average C ₁;

(3) level set function upgrades: utilize described curved exterior gray average C ₀with with interior intensity average C ₁according to the energy functional of CV model determination level set function, set up the level set function recursion formula in adjacent two moment;

(4) level set function border is revised: on border, apply Neumann condition, revises the level set function φ (x, y) of current time;

(5) level set function convergence judges: first judge whether the level set function φ (x, y) of current time restrains, if convergence just stops iteration; Otherwise, will judge whether to reach the maximum iteration time preset, just stop iteration if reached, otherwise enter step (2) continuation next round iteration.

Beneficial effect: the present invention is compared with traditional CV model SAR image groups of building dividing method, and tool has the following advantages:

(1) not directly using the process data of SAR image as CV model image dividing method, but using the process data of the variogram characteristic pattern of SAR image as CV model image dividing method, achieve the gray scale homogeney of construction area in process data, significantly improve the effect of groups of building segmentation;

(2) in the variogram feature extraction of SAR image, have employed FAST RECURSIVE ALGORITHM, under the prerequisite keeping feature extraction precision, significantly reduce computing time.

Accompanying drawing explanation

Fig. 1 is the process flow diagram of groups of building dividing method in the SAR image that proposes of the present invention;

Fig. 2 is the schematic diagram that when calculating variogram, pixel is right; Fig. 2 (a) is when calculating variogram, the schematic diagram that on 0 ° of direction, pixel is right; Fig. 2 (b) is when calculating variogram, the schematic diagram that on 45 ° of directions, pixel is right; Fig. 2 (c) is when calculating variogram, the schematic diagram that on 90 ° of directions, pixel is right; Fig. 2 (d) is when calculating variogram, the schematic diagram that on 135 ° of directions, pixel is right;

Fig. 3 is the redundancy schematic diagram of window movement when calculating variogram; Fig. 3 (a) is when calculating variogram, the redundancy schematic diagram that on 0 ° of direction, window moves down; Fig. 3 (b) is when calculating variogram, the redundancy schematic diagram that on 0 ° of direction, window moves right;

Fig. 4 is the algorithm flow chart carrying out Iamge Segmentation based on CV model;

Fig. 5 is the original SAR image of experiment 1 and the SAR image after splitting; Fig. 5 (a) is the original SAR image comprising groups of building in experiment 1; Fig. 5 (b) is SAR image buildings segmentation result in experiment 1;

Fig. 6 is the original SAR image of experiment 2 and the SAR image after splitting; Fig. 6 (a) is the original SAR image comprising groups of building in experiment 2; Fig. 6 (b) is the SAR image buildings segmentation result in experiment 2.

Embodiment

The present invention will be described below with reference to accompanying drawings.Process flow diagram of the present invention as shown in Figure 1, provides a kind of SAR image groups of building dividing method.First, pre-service is carried out to the original SAR image comprising groups of building, then calculate its variogram feature based on SAR image.By variogram feature input CV model image partitioning algorithm, distinguish according to the feature of groups of building with other scene, Iamge Segmentation is carried out to construction area.From the PRELIMINARY RESULTS of segmentation, reject false-alarm region, finally at groups of building edge delineate line, namely obtain final groups of building image segmentation result.

The SAR image groups of building dividing method operand that the present invention proposes is little, and strong to the recognition capability of groups of building, segmenting edge is accurate.Implementation step is as follows:

Step 1): the pre-service of original SAR image

Carrying out groups of building segmentation to carry out SAR image better, needing to carry out pre-service to original SAR image.This is mainly because differ in SAR image source, and feature of image is different, and such as the coherent spot order of severity is different, and brightness and contrast is also different, and resolution height is different.In order to allow these SAR image can adapt to follow-up image segmentation routine, needing carry out coherent spot elimination, brightness and contrast's adjustment, picture element interpolation or extract process, its visual signature is reached unanimity.

Step 2): the textural characteristics figure of SAR image calculates

For a width SAR image I (x, y), the mode adopting traversal is each pixel windowing, and in this window, calculate the variogram textural characteristics of this pixel, then obtain the variogram characteristic pattern of SAR image, its size is the same with SAR image.Variogram feature has directivity, and this method is by the variogram feature of computed image on 0 °, 45 °, 90 ° and 135 ° of directions.

For any one pixel I (x ₀, y ₀), list the definition of 0 ° of direction upper variation Function feature:

${\mathrm{\&gamma;}}^0(x_0,y_0)=\frac{1}{W(W-h)}\underset{y=y_0-d}{\overset{y_0+d}{\&Sigma;}}\underset{x=x_0-d}{\overset{x_0+d-h}{\&Sigma;}}{\&lsqb;I(x,y)-I(x+h,y)\&rsqb;}^2---\left(1\right)$

In formula, γ represents variogram; W represents window width, and this method requires that W must be odd number; Parameter d=(W-1)/2; H represents the distance that window mid point is right.Fig. 2 (a), Fig. 2 (b), Fig. 2 (c) and Fig. 2 (d) be respectively on 0 °, 45 °, 90 ° and 135 ° of directions W=5 and h=1 time point to schematic diagram.

The present invention have employed FAST RECURSIVE ALGORITHM, significantly to reduce operand when calculating variogram characteristic pattern.Here for 0 ° of direction, then composition graphs 3 (a) and Fig. 3 (b), explain the FAST RECURSIVE ALGORITHM of texture variogram feature calculation.First according to formula (1), current pixel point I (x is calculated ₀, y ₀) variogram eigenwert γ ⁰(x ₀, y ₀).As calculating pixel I (x ₀, y ₀+ 1) γ during variogram eigenwert ⁰(x ₀, y ₀+ 1), in fact window is the distance having moved down 1 pixel.By observing Fig. 3 (a), can find out that to there is point identical in a large number between two windows right, only having last column of the first row of first window and second window to there is different points right, recursive algorithm therefore can be adopted to calculate fundamental function value γ ⁰(x ₀, y ₀+ 1).Recursion formula is as follows:

$\begin{array}{c}{\mathrm{\&gamma;}}^0(x_0,y_0+1)={\mathrm{\&gamma;}}^0(x_0,y_0)+\frac{1}{2(W-h)}(\underset{x=x_0-d}{\overset{x_0+d-h}{\&Sigma;}}{\&lsqb;I(x,y_0+d+1)-I(x+h,y_0+d+1)\&rsqb;}^2\\ -\underset{x=x_0-d}{\overset{x_0+d-h}{\&Sigma;}}{\&lsqb;I(x,y_0-d)-I(x+h,y_0-d)\&rsqb;}^2)\end{array}---\left(2\right)$

As shown in Fig. 3 (b), when window center is by (x ₀, y ₀) move to (x ₀+ 1, y ₀) time, there is point identical in a large number between the left window and right side window right, only have the point on the edge, the right of the left margin of the left window and right side window to being not identical.In like manner, recursion formula can be derived:

$\begin{array}{c}{\mathrm{\&gamma;}}^0(x_0+1,y_0)={\mathrm{\&gamma;}}^0(x_0,y_0)+\frac{1}{2W}(\underset{y=y_0-d}{\overset{y_0+d}{\&Sigma;}}{\&lsqb;I(x_0+d+1-h,y)-I(x_0+d+1,y)\&rsqb;}^2\\ -\underset{y=y_0-d}{\overset{y_0+d}{\&Sigma;}}{\&lsqb;I(x_0-d,y)-I(x_0-d+h,y)\&rsqb;}^2)\end{array}---\left(3\right)$

After adopting FAST RECURSIVE ALGORITHM, as long as calculate the variogram feature in 0 ° of direction at (0,0) place, just can be obtained the variogram characteristic pattern γ in 0 ° of direction on full figure by recursion ⁰(x, y).Take same recurrence thought, calculate 45 °, (0,0) place, 90 °, after the variogram eigenwert on 135 ° of directions, just recursion can obtain on these three directions characteristic pattern γ ⁴⁵(x, y), γ ⁹⁰(x, y), γ ¹³⁵(x, y).Replace omnidirectional variogram eigenwert with the average of the variogram on above-mentioned four direction, just can obtain variogram textural characteristics figure, that is:

$\mathrm{\&gamma;}(x,y)=\frac{1}{4}\&lsqb;{\mathrm{\&gamma;}}^0(x,y)+{\mathrm{\&gamma;}}^{45}(x,y)+{\mathrm{\&gamma;}}^{90}(x,y)+{\mathrm{\&gamma;}}^{135}(x,y)\&rsqb;---\left(4\right)$

Step 3): split based on the SAR image of CV model

Image segmentation algorithm based on CV model is applicable to background and target area gray scale homogeneity, the obvious image of contrast separately.In the variogram characteristic pattern of SAR image, groups of building and other background characteristics have respective gray scale homogeney, so using the variogram characteristic pattern of SAR image as input, are suitable for using CV model image partitioning algorithm to carry out groups of building segmentation.

CV model image partitioning algorithm process flow diagram in the present invention as shown in Figure 4, comprises initialization level set function, calculating zones of different gray average, level set function upgrades, level set function border is revised, level set function restrains the steps such as judgement.

(1) the zero level collection of initialization level set function φ (x, y), the namely set of the pixel of φ (x, y)=0.Zero level collection profile is initialized as equally distributed circle on image, according to length and the width of image, determines centre coordinate and the radius of each circle, guarantee that initial profile can, through each building area, avoid successive iterations to be absorbed in Local Search.The value of level set function φ (x, y) is defined as symbolic measurement, equals the minimum distance of each pixel to profile.So on each circular contour, level set function φ (x, y)=0; In the inside of each circular contour, φ (x, y) >0; Other regions on image, level set function φ (x, y) <0.

(2) the gray average C that calculated curve is outside and inner ₀and C ₁, computing formula is as follows:

$C_0=\frac{\underset{\mathrm{\&Omega;}}{\&Integral;\&Integral;}\mathrm{\&gamma;}(x,y)(1-H(\mathrm{\&phi;}\left(x,y\right)\left)\right)dxdy}{\underset{\mathrm{\&Omega;}}{\&Integral;\&Integral;}(1-H(\mathrm{\&phi;}\left(x,y\right)\left)\right)dxdy}---\left(5\right)$

$C_1=\frac{\underset{\mathrm{\&Omega;}}{\&Integral;\&Integral;}\mathrm{\&gamma;}(x,y)H\left(\mathrm{\&phi;}\right(x,y\left)\right)dxdy}{\underset{\mathrm{\&Omega;}}{\&Integral;\&Integral;}H\left(\mathrm{\&phi;}\right(x,y\left)\right)dxdy}---\left(6\right)$

Wherein C ₀and C ₁be the gray average in φ (x, y) <0 and φ (x, y) on image>=0 region respectively, Ω represents whole image integration region, and Heaviside function H (z) is defined as:

$H\left(z\right)=\{\begin{array}{cc}1,& ifz\&GreaterEqual;0\\ 0,& ifz\&GreaterEqual;0\end{array}---\left(7\right)$

(3) in order to determine the EVOLUTION EQUATION of level set function φ (x, y), the energy functional of level set function must be defined.According to the definition of CV model, the energy functional E (C of level set form ₀, C ₁, φ) be defined as:

$\begin{array}{c}E(C_0,C_1,\mathrm{\&phi;})=\mathrm{\&mu;}\underset{\mathrm{\&Omega;}}{\&Integral;\&Integral;}\mathrm{\&delta;}\left(\mathrm{\&phi;}\right(x,y\left)\right)|\&dtri;\mathrm{\&phi;}(x,y)|dxdy+\underset{\mathrm{\&Omega;}}{\&Integral;\&Integral;}{|\mathrm{\&gamma;}(x,y)-C_1|}^{2}H\left(\mathrm{\&phi;}\right(x,y\left)\right)dxdy\\ +\underset{\mathrm{\&Omega;}}{\&Integral;\&Integral;}{|\mathrm{\&gamma;}(x,y)-C_0|}^2(1-H(\mathrm{\&phi;}\left(x,y\right)\left)\right)dxdy\end{array}---\left(8\right)$

This is one and adopts variogram feature as the functional of level set movements driving-energy, and wherein μ is weighting coefficient, represent the gradient of level set function, Dirac function δ (z) is then defined as:

$\mathrm{\&delta;}\left(z\right)=\left\{\begin{array}{cc}1,& z=0\\ 0,& z\&NotEqual;0\end{array}\right.---\left(9\right)$

According to variational principle and gradient descent method, the EVOLUTION EQUATION of level set function φ can be derived:

$\frac{\&part;\mathrm{\&phi;}}{\&part;t}=\mathrm{\&delta;}\left(\mathrm{\&phi;}\right)\&lsqb;\mathrm{\&mu;}\&dtri;\&CenterDot;\left(\frac{\&dtri;\mathrm{\&phi;}}{\left|\&dtri;\mathrm{\&phi;}\right|}\right)-{(\mathrm{\&gamma;}-C_1)}^2+{(\mathrm{\&gamma;}-C_0)}^2\&rsqb;+\mathrm{\&nu;}\&dtri;\&CenterDot;\left(\right(1-\frac{1}{\left|\&dtri;\mathrm{\&phi;}\right|}\&dtri;\mathrm{\&phi;}\left)\right)---\left(10\right)$

Wherein represent the derivative of level set function, represent that divergence calculates, v is weighting coefficient.According to the EVOLUTION EQUATION of level set function φ, can by the level set function φ in n moment ⁿobtain the level set function φ in n+1 moment ⁿ⁺¹.

(4) on border, apply Neumann condition, revise level set function φ ⁿ⁺¹.For partial differential equation problem, only just can solve providing under suitable starting condition and border condition, these conditions i.e. definite condition.By step (1), give starting condition.Suppose that n is the normal direction of image boundary, border is applied Neumann condition, that is:

$\frac{\&part;\mathrm{\&phi;}}{\&part;n}=0---\left(11\right)$

After level set function discretize, for certain some p={ (i, j) | 0≤i<M, 0≤j<N}, M and N are respectively columns and the line number of image, revise level set function φ ⁿ⁺¹method can sketch and be:

On level set border, the value of angle point is respectively:

For point (0,0), ${\mathrm{\&phi;}}_1^{n+1}(0,0)={\mathrm{\&phi;}}_1^{n+1}(2,2);$

For point (0, N-1), ${\mathrm{\&phi;}}_1^{n+1}(0,N-1)={\mathrm{\&phi;}}_1^{n+1}(2,N-3);$

For point (M-1,0), ${\mathrm{\&phi;}}_1^{n+1}(M-1,0)={\mathrm{\&phi;}}_1^{n+1}(M-3,2);$

For point (M-1, N-1), ${\mathrm{\&phi;}}_1^{n+1}(M-1,N-1)={\mathrm{\&phi;}}_1^{n+1}(M-3,N-3);$

On border, the process of remaining point is such:

For point (0, j) | 1≤j<N-1},

For point (i, 0) | 1≤i<M-1},

For point (i, N-1) | 1≤i<M-1}, ${\mathrm{\&phi;}}_1^{n+1}(i,N-1)={\mathrm{\&phi;}}_1^{n+1}(i,N-3);$

For point (M-1, j) | 1≤j<N-1}, ${\mathrm{\&phi;}}_1^{n+1}(M-1,j)={\mathrm{\&phi;}}_1^{n+1}(M-3,j);$

Can prevent the gradient of level set function from departing from 1 to the process on level set border, follow-up level set function iteration refresh routine be run more sane.

(5) whether determined level set function φ restrains or reaches maximum iteration time.First whether determined level set function restrains, if convergence just stops iteration; Otherwise, will judge whether to reach maximum iteration time, just stop iteration if reached, otherwise continue next round iteration.

During level set function convergence, its zero level collection is exactly the border of Iamge Segmentation, and such border can allow energy functional reach minimum, is also the optimum division of zones of different.

Step 4): Regional Integration and removal false-alarm

Input based on the image segmentation routine of CV model is variogram characteristic pattern, after level set function φ iteration completes, its output is a binary map, and comprise two regions of level set function φ >=0 and φ <0, the size of binary map is consistent with original image.There is much scrappy division due to normal in binary map, so need to adopt the connected region detection algorithm based on eight neighborhood to extract connected region, scrappy division is integrated.Because groups of building are all appearance in blocks, general area is all larger, can arrange certain threshold value, and the region visual that area is less than threshold value is that false-alarm is removed.

Step 5): building area outline line extracts

Based on the binary map after Regional Integration and removal false-alarm, extract groups of building outline line, obtain explicit segmentation result.Then outline line can be drawn in original SAR image, just intuitively can arrive the segmentation result of groups of building.

In order to prove the validity of the present invention in the groups of building segmentation of SAR image, provide two emulation experiments herein.All have an original SAR image in each experiment, wherein with groups of building, but the resolution of two width SAR image is different, and gray scale and contrast are also different.Example needs by image by pre-service, calculates variogram and Iamge Segmentation again after making image proterties consistent.

Experiment 1: Fig. 5 (a) is actual measurement SAR image, and its resolution is relatively high, and it is more clearly that groups of building are seen.With the interference such as the woods and farmland scenery in figure, especially the woods easily cause false-alarm.The segmentation result that Fig. 5 (b) is groups of building, as seen from the figure, although the luminance difference of shade and buildings is larger in construction area, the outline line obtained still accurately can orient the border of whole groups of building, avoids aim curve and is absorbed in local minimum.Groups of building are defined well, and other atural objects such as the such as woods, farmland, road, lake obtain and effectively suppress.

Experiment 2: Fig. 6 (a) is actual measurement SAR image, and its resolution is relatively low, and adjacent buildings mixes, and not easily differentiates.Also with other atural objects such as farmland, road, rivers in figure.The segmentation result that Fig. 6 (b) is groups of building, as seen from the figure, groups of building are accurately segmented, and the profile definition separatrix of groups of building and other atural object, illustrates that the present invention can split the SAR image groups of building of low resolution very well.

Two embodiments show, the SAR image groups of building dividing method that the present invention proposes, and can distinguish groups of building and other interference atural object, be partitioned into the groups of building in SAR image exactly.

Claims (4)

1. a SAR image groups of building dividing method, is characterized in that, the method comprises the following steps:

1) carrying out pre-service to original SAR image makes the performance parameters of image reach the index preset, and obtains pretreated SAR image;

2) variogram feature extraction is carried out to described pretreated SAR image and go out textural characteristics, form variogram characteristic pattern, realize the gray scale homogeney of construction area;

3) adopt CV model image dividing method to carry out groups of building segmentation to described variogram characteristic pattern, obtain the binary map that can be distinguished construction area and territory, non-building area;

4) adopt connected region detection algorithm to be communicated with adjacent construction area to described binary map, integrated by adjacent scrappy construction area, the false-alarm then caused by other atural object of area screening removal is split;

5) to the binary map obtained after described Iamge Segmentation, extract the outline line of construction area, obtain explicit SAR image segmentation result.

2. SAR image groups of building dividing method according to claim 1, is characterized in that, step 1) described in pre-service comprise and eliminate coherent spot, regulate brightness and contrast, picture element interpolation or extraction.

3. SAR image groups of building dividing method according to claim 1, is characterized in that, described textural characteristics value is the mean value of the textural characteristics on 0 °, 45 °, 90 °, 135 ° four directions.

4. SAR image groups of building dividing method according to claim 1, is characterized in that, described step 3) in adopt CV model image dividing method to carry out groups of building segmentation to comprise the following steps:

(1) initialization level set function: zero level collection profile is initialized as equally distributed circle in SAR image, according to length and the width of image, determine centre coordinate and the radius of each circle, initial profile is made to pass through each building area, by level set function φ (x, y) value is defined as symbolic measurement, on each circular contour, level set function φ (x, y)=0, in the inside of each circular contour, φ (x, y) >0; Other regions on image, level set function φ (x, y) <0;

(2) zones of different gray average is calculated: calculate the curved exterior gray average C formed by level set function φ (x, y) ₀with with interior intensity average C ₁;

(3) level set function upgrades: utilize described curved exterior gray average C ₀with with interior intensity average C ₁according to the energy functional of CV model determination level set function, set up the level set function recursion formula in adjacent two moment;

(4) level set function border is revised: on border, apply Neumann condition, revises the level set function φ (x, y) of current time;

(5) level set function convergence judges: first judge whether the level set function φ (x, y) of current time restrains, if convergence just stops iteration; Otherwise, will judge whether to reach the maximum iteration time preset, just stop iteration if reached, otherwise enter step (2) continuation next round iteration.