CN101540042B - SAR image speckle suppression method based on second generation curvilinear wave transformation - Google Patents

Abstract

The invention discloses an SAR image speckle suppression method based on second generation curvilinear wave transformation, which mainly overcomes the defect of scratch effect and point target loss brought by a curvilinear wave to the SAR image speckle suppression. The SAR image speckle suppression method comprises the following steps: performing the second generation curvilinear wave transformation to a selected test image and partitioning the selected test image into 5 layers of subbands; keeping coefficients of the first layer unchangeable and zero-setting coefficients of the fifth layer; respectively evaluating parameter vectors of hybrid Gaussian models from the second layer to the fourth layer by an EM method; marking the coefficients from the second layer to the fourth layer; reconstructing the image, detecting the edge of the reconstructed image, and performing the average filtering to the uniform area of the reconstructed image to obtain the filtered image; and performing the nonlinear anisotropy dispersion iteration to a difference image obtained by the original image and the filtered image to obtain a speckle suppressed image. The invention has the advantages of keeping clean edge of the image, removing the scratch effect and remaining the point target characteristic information of the image, and can be used for preprocessing scene analysis and image understanding in the SAR image.

Description

Based on the relevant speckle suppression method of the SAR image of second generation curve wave conversion

Technical field

The invention belongs to digital image processing field, relate to a kind of relevant speckle suppression method of SAR image, this method can be used for the scene analysis in the SAR image and the pre-service of image understanding.

Background technology

The SAR image field that is widely used, but during the SAR image imaging, the scatter echo of imaging scatterer has relevant effect, makes image can not effectively react the scattering properties of ground object target, and this interference is called speckle noise in the SAR image.The existence of speckle noise all is very disadvantageous for the analysis of scene in the image and the understanding of image, and therefore suppressing speckle noise is a part very important in the SAR Flame Image Process.The main target that suppresses speckle noise keeps minutias such as edge of image again when being the inhibition speckle noise.

SAR image coherent spot suppresses and can carry out in the spatial domain, as Lee filtering, based on method of partial differential equation etc., the method of Lee filtering is carried out filtering by the window that slides to image, the ability that its coherent spot suppresses is directly proportional with the size of window, but increasing along with window, can cause the fuzzy of image border, lose the detailed information of image, based on the method for partial differential equation and the coefficient of diffusion of being constructed very big relation is arranged, and need to consider the special processing of data fitting item and characteristic area, along with the increase of iterations can cause image object and edge more and more fuzzyyer; Also can carry out at transform domain, as wavelet transformation and curve wave conversion, be by the threshold process of conversion coefficient or the method for setting up the coefficient model are removed noise.Small echo only has a singularity in two-dimensional space, can only effectively handle the homogeneous area in the image, and can not effectively handle the zone that edge in the image etc. has the line singularity, and the curve wave conversion has yardstick, three parameters of direction and locus, and its odd function satisfies anisotropic shrinkage rates relation: width ≈ length ², make the curve ripple have very strong directivity and anisotropy, the edge that can optimum rarefaction representation has straight line and curve singularity.The hard-threshold denoising method that people such as Candes and Donoho proposes: promptly, be set to zero as long as the coefficient littler than threshold value is used as noise processed without exception based on the curve wave conversion.The maximum deficiency of this method is " cross and strangle " to coefficient, thereby it is fuzzy to cause the image border to occur, and because curve wave conversion itself lacks translation invariance, the image after coherent spot suppresses has the cut effect.

Summary of the invention

The objective of the invention is to overcome the deficiency of above-mentioned prior art, propose the relevant speckle suppression method of a kind of SAR image,, eliminate the cut effect that the curve wave conversion brings to keep the edge of image sharpness effectively based on second generation curve wave conversion.

Realize that technical scheme of the present invention is is transformation tool with second generation curve ripple, adopt mixed Gauss model that curve wave system number is carried out modeling, with Bayes's contraction factor noise figure is shunk on this basis, homogeneous area to reconstructed image carries out mean filter, come the error image of image behind level and smooth former figure and the mean filter again with the method for Anisotropic Nonlinear diffusion, error image and the image addition behind the mean filter after level and smooth are obtained falling the spot image, and concrete steps are as follows:

(1) the test pattern I that chooses is carried out second generation curve wave conversion, be divided into 5 straton bands;

(2) keep the 1st layer coefficient constant, the 5th layer the whole zero setting of coefficient;

(3) estimate the parameter vector θ of the 2nd～4 layer mixed Gauss model respectively with the EM method;

(4) the 2nd～4 layer coefficient is carried out mark, be labeled as 1 coefficient and remain unchanged, be labeled as 0 coefficient zero setting, be labeled as 2 coefficient and shrink with Bayes's contraction factor;

(5) obtain reconstructed image R to carrying out second generation curve ripple inverse transformation through the coefficient after step (2) and step (4) processing;

(6) detect the edge of reconstructed image R, and its homogeneous area is carried out mean filter, obtain image U after the filtering;

(7) original image and filtered image subtraction are obtained error image: V=I-U, carry out the Anisotropic Nonlinear diffusion to this error image and iterate to v k time ^k

(8) with v ^kWith image U addition after the filtering, obtain the image D=U+v after coherent spot suppresses ^k

The present invention compared with prior art has following advantage:

1, the present invention can keep edge of image preferably.

Because the curve wave conversion has yardstick, three parameters of direction and locus, and its odd function satisfies anisotropic shrinkage rates relation: width ≈ length ²Make the curve ripple have very strong directivity and anisotropy, the edge that can optimum rarefaction representation has straight line and curve singularity, therefore, the present invention can more effective processing edge of image than existing method such as Lee filtering and small echo, make that image border after coherent spot suppresses keeps better.

2, the present invention can eliminate the cut effect that existing curve wave method is produced.

Because the curve wave conversion lacks translation invariance, the SAR image is carried out can producing the cut effect after coherent spot suppresses with the curve wave method.The present invention carries out mean filter to the homogeneous area of image after by rim detection, can eliminate the not enough cut effect of being brought of curve ripple itself.

3, the present invention can effectively keep the detailed information such as point target of image.

Because the direction of curve ripple itself is more, energy to point target behind the point target rarefaction representation in the image can be dispersed, therefore the point target of image can cause fuzzyly after the curve ripple inverse transformation, and the used method to the Anisotropic Nonlinear diffusion of error image of the present invention can effectively keep the point target of image.

Description of drawings

Fig. 1 is a main operating process synoptic diagram of the present invention;

Fig. 2 is 12 direction synoptic diagram setting during rim detection among the present invention;

Fig. 3 is with the present invention and the existing method coherent spot inhibition effect contrast figure to test pattern Bedfordshire 256 * 256;

Fig. 4 is a details amplification effect comparison chart after using the present invention and existing method to the coherent spot inhibition of test pattern Bedfordshire 256 * 256;

Fig. 5 is with the present invention and the existing method coherent spot inhibition effect contrast figure to test pattern Horse track one 256 * 256;

Fig. 6 is with the present invention and the existing method coherent spot inhibition effect contrast figure to test pattern Horse track two 256 * 256;

Fig. 7 is with the present invention and the existing method coherent spot inhibition effect contrast figure to test pattern Stanwick 512 * 512.

Embodiment

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

Step 1: choose testing SA R image I, it is carried out second generation curve wave conversion.

The mathematical model that the present invention chooses testing SA R image is: y=x+n

Wherein y={y (i, j) | i, j=1,2 ... N} represents the SAR image, x={x (i, j) | i, j=1,2 ... N} represents the backscatter intensity of SAR image real scene, n={n (i, j) | i, j=1,2 ... N} represents that zero-mean and variance equal σ ²Gaussian noise, the size of N presentation video;

SAR image to be processed is carried out second generation curve wave conversion, is divided into 5 straton bands, obtain the SAR image in W subband position of S layer (i, the curve wave system number of j) locating:

C(S，W，i，j)＝s(S，W，i，j)+ε(S，W，i，j)

Wherein s (S, W, i, j) and ε (i is respectively that backscatter intensity and noise are at W subband position of S layer (i, the curve wave system number of j) locating j) for S, W.

Step 2: keep the 1st layer coefficients of SAR image curve wave system number constant, putting the 5th layer coefficients all is zero.

For size is the SAR image of N * N, and the size of its layer 5 matrix of coefficients also is N * N, and these coefficient overwhelming majority are the noise information in the presentation video all, therefore all coefficients in this matrix of coefficients all is set to 0.

Step 3: the parameter vector θ that estimates the 2nd～4 layer mixed Gauss model with the EM method respectively.

Matrix of coefficients C{s}{w} on each direction W of S layer is integrated into delegation, and it is capable as the W of matrix of coefficients Y,, obtains the mixed Gaussian function and be with the coefficient histogram of the mixed Gaussian approximation of function Y of two zero-means:

$f\left(y_n\right)={\mathrm{\Σ}}_{m=1}^{m=2}{\mathrm{\ϵ}}_m{f}_m\left(y_n\right)$

ε wherein ₁+ ε ₂=1, $f_m\left(y_n\right)=\exp (-{\left(y_n\right)}^2/\left({2\mathrm{\σ}}_m^2\right))/\sqrt{{2\mathrm{\π\σ}}_m^2},$ ε _m, σ _m ²Be respectively f _m(y _n) weights and variance, with parameter ε ₁, ε ₂, σ ₁ ², σ ₂ ²Be designated as the parameter vector θ of mixed Gauss model, promptly $\mathrm{\θ}=\{{\mathrm{\ϵ}}_1,{\mathrm{\ϵ}}_2,{\mathrm{\σ}}_1^2,{\mathrm{\σ}}_2^2\};$

Select the initial value of parameter to be: ε ₁=ε ₂=1/2, ${\mathrm{\σ}}_1^2=(\max -\min )/3,$ ${\mathrm{\σ}}_2^2=2(\max -\min )/3,$ Calculate each the sample y among the Y _nThe posterior probability of Gaussian probability-density function:

$E\left(h_{\mathrm{nm}}\right)=\frac{f_m\left(y_n\right){\mathrm{\ϵ}}_m}{{\mathrm{\Σ}}_{m=1}^{m=2}{f}_m\left(y_n\right){\mathrm{\ϵ}}_m}$

Use formula then ${\widehat{\mathrm{\&sigma;}}}_m^2=\frac{{\mathrm{\&Sigma;}}_{n=1}^N\left[E\left(h_{\mathrm{nm}}\right)y_n^2\right]}{{\mathrm{\&Sigma;}}_{n=1}^{N}E\left(h_{\mathrm{nm}}\right)},$ ${\widehat{\mathrm{\&epsiv;}}}_m=\frac{{\mathrm{\&Sigma;}}_{n=1}^{N}E\left(h_{\mathrm{nm}}\right)}{N}$ Parameter vector θ is carried out the iteration training, when $\mathrm{\&Sigma;}\left|\frac{{\mathrm{\&theta;}}_j^i-{\mathrm{\&theta;}}_j^{i-1}}{{\mathrm{\&theta;}}_j^i}\right|<{10}^{-4}$ The time, stop iteration, obtain parameter vector θ, θ _j ⁱJ element representing the parameter vector that the i time iteration obtains.

Step 4: the 2nd～4 layer coefficient is carried out mark, be labeled as 1 coefficient and remain unchanged, be labeled as 0 coefficient zero setting, be labeled as 2 coefficient and shrink with Bayes's contraction factor.

Matrix of coefficients C{s}{w} to centre each layer different sub-band of SAR image curve wave system number estimates its noise criteria difference σ with didactic prior estimate _Nc, establish threshold value T ₁=3 σ _Nc, T ₂=0.5 σ _Ncσ _Sw, to each the coefficient C among the C{s}{w} (i, j), if | C (i, j) |＞T ₁, then (i j) gets its 3 * 3 field window, if two above absolute values are arranged greater than T in this window for the center with C ₁Coefficient, then these coefficients are labeled as 1, remaining is labeled as 2; To being labeled as 2 coefficient, if | C (i, j) |＜T ₂, then (i j) is labeled as 0 C.After all coefficient marks are intact, are labeled as 1 coefficient and remain unchanged, be labeled as 0 coefficient zero setting, be labeled as 2 coefficient and shrink with Bayes's contraction factor and obtain not having the estimation of coefficient of making an uproar

As follows:

$\hat{x}=\underset{m=\mathrm{1,2}}{\mathrm{\&Sigma;}}\frac{{\mathrm{\&epsiv;}}_m{f}_m\left(y_n\right)}{f\left(y_n\right)}\frac{|{\mathrm{\&sigma;}}_m^2-{\mathrm{\&sigma;}}_n^2|}{{\mathrm{\&sigma;}}_m^2}{y}_n.$

Step 5: carry out curve ripple inverse transformation and obtain reconstructed image R.

After all coefficient processing were finished, (i j) carried out curve ripple inverse transformation, obtains reconstructed image R for S, W to matrix of coefficients C.

Step 6: detect the edge of reconstructed image R, and its homogeneous area is carried out mean filter, obtain image U after the filtering.

(i j), is that 3 * 3 field window is got at the center with it, calculates the pixel average P on 12 directional ray both sides in this window respectively to each pixel R among the reconstructed image R ₁And P ₂, establish R=P ₁/ P ₂, if R≤1, then ratio detects factor r=R, otherwise r=R ^-1, find out 12 minimum rates on the direction and detect factor r _Min

If threshold value T ₀, if r _Min＞T ₀, think that then (i j) is marginal point to R, is labeled as 1, otherwise be labeled as 0, after all element markings are intact, if pixel is labeled as 1, be that 3 * 3 field window is got at the center then, be less than 4, then this pixel is labeled as 0 if this field window internal labeling is 1 number with it;

If (i j) is marked as 0 to pixel R, and then (i j) gets its 5 * 5 field W for the center with R ₁, if W ₁In pixel only be less than 4 point and be labeled as 1, then (i, pixel value j) is taken as W to R ₁The mean value of interior all pixels; (i j) is the center, gets its 3 * 3 field W otherwise with R ₂, if W ₂In pixel only be less than 4 point and be labeled as 1, then (i, pixel value j) is taken as W to R ₂In the mean value of all pixels, otherwise R (i, pixel value j) is taken as W ₂In all are labeled as the mean value of 0 pixel,

All are labeled as after 0 pixel carries out above-mentioned handling, obtain filtered image U.

Step 7: this error image is carried out the Anisotropic Nonlinear diffusion iterate to v k time ^k, obtain the image D=U+v after coherent spot suppresses ^k

Original image I and filtered image U are subtracted each other, obtain error image V=I-U;

Note v ⁰=V, carry out iteration according to following formula to error image:

$v_{i,j}^{k+1}=v_{i,j}^k+t\underset{r,s=-1}{\overset{1}{\mathrm{\&Sigma;}}}\frac{g\left({\sqrt{2}}^{1-\left|r\right|-\left|s\right|}\right|v_{i+r,j+s}^k-v_{i,j}^k\left|\right)(v_{i+r,j+s}^k-v_{i,j}^k)}{r^2+s^2}$

G (x)=1/ (1+x wherein ²/ ρ ²), (r, s) ≠ (0,0), t=1/8, ρ=6, v _{I, j}For the i of v is capable, the pixel of j row;

Step 8: obtain nonlinear diffusion figure v as a result after iteration k time ^k, v ^kWith filtered image U addition, obtain the image D=U+v after coherent spot suppresses ^k

Effect of the present invention further specifies by following emulation.

1, simulated conditions

Adopt image commonly used in the SAR image coherent spot inhibition test: 1. size is 256 * 256, resolution is that 3m, equivalent number are 2 X-band Britain Bedfordshire area SAR magnitude image, shown in Fig. 3 (a); 2. size is 256 * 256, resolution is that 1m, equivalent number are 4 Ku wave band New Mexico Horsetrack one area SAR intensity image, shown in Fig. 5 (a); 3. size is 256 * 256, resolution is that 1m, equivalent number are 4 Ku wave band New Mexico Horse track two area SAR intensity image, shown in Fig. 6 (a); 4. size is 512 * 512, resolution is that 3m, equivalent number are 4 X-band Britain Stanwick area SAR intensity image, shown in Fig. 7 (a), 4 width of cloth SAR images are as tested object altogether, the non-downsampling Contourlet method of using enhanced Lee filtering, stationary wavelet soft-threshold, curve ripple hard-threshold, statistics priori to instruct respectively, abbreviate NSCT as, and carry out coherent spot based on the relevant speckle suppression method of the SAR image of second generation curve wave conversion and suppress.

2, analysis of simulation result

To above-mentioned image simulation result 1. as shown in Figure 3, wherein Fig. 3 (b) is the coherent spot inhibition result of enhanced Lee filtering method, Fig. 3 (c) is that the coherent spot of stationary wavelet soft-threshold method suppresses the result, Fig. 3 (d) is that the coherent spot of curve ripple hard-threshold method suppresses the result, Fig. 3 (e) is that the coherent spot of NSCT method suppresses the result, and Fig. 3 (f) is that the coherent spot of the inventive method suppresses the result;

The partial enlarged drawing of image simulation result 1. as shown in Figure 4, wherein Fig. 4 (a) is the details enlarged drawing of Fig. 3 (a), Fig. 4 (b) is the details enlarged drawing that the coherent spot of enhanced Lee filtering method suppresses the result, Fig. 4 (c) is the details enlarged drawing that the coherent spot of stationary wavelet soft-threshold method suppresses the result, Fig. 4 (d) is the details enlarged drawing that the coherent spot of curve ripple hard-threshold method suppresses the result, Fig. 4 (e) is the details enlarged drawing that the coherent spot of NSCT method suppresses the result, and Fig. 4 (f) is the details enlarged drawing that the coherent spot of the inventive method suppresses the result;

To above-mentioned image simulation result 2. as shown in Figure 5, wherein Fig. 5 (b) is the coherent spot inhibition result of enhanced Lee filtering method, Fig. 5 (c) is that the coherent spot of stationary wavelet soft-threshold method suppresses the result, Fig. 5 (d) is that the coherent spot of curve ripple hard-threshold method suppresses the result, Fig. 5 (e) is that the coherent spot of NSCT method suppresses the result, and Fig. 5 (f) is that the coherent spot of the inventive method suppresses the result;

To above-mentioned image simulation result 3. as shown in Figure 6, wherein Fig. 6 (b) is the coherent spot inhibition result of enhanced Lee filtering method, Fig. 6 (c) is that the coherent spot of stationary wavelet soft-threshold method suppresses the result, Fig. 6 (d) is that the coherent spot of curve ripple hard-threshold method suppresses the result, Fig. 6 (e) is that the coherent spot of NSCT method suppresses the result, and Fig. 6 (f) is that the coherent spot of the inventive method suppresses the result;

To above-mentioned image simulation result 4. as shown in Figure 7, wherein Fig. 7 (b) is the coherent spot inhibition result of enhanced Lee filtering method, Fig. 7 (c) is that the coherent spot of stationary wavelet soft-threshold method suppresses the result, Fig. 7 (d) is that the coherent spot of curve ripple hard-threshold method suppresses the result, Fig. 7 (e) is that the coherent spot of NSCT method suppresses the result, and Fig. 3 (f) is that the coherent spot of the inventive method suppresses the result.

As can be seen from Figure 3, after coherent spot of the present invention suppressed, homogeneous area was smoother, and detailed information such as edge and point target also keep better; Enhanced Lee filtering has suppressed speckle noise to a certain extent, but the edge is fuzzyyer, has lost the detailed information of image; Stationary wavelet soft-threshold method has caused the fuzzy of edge, and has lost point target information; The edge of curve ripple hard-threshold method keeps better, but homogeneous area has cut effect clearly, and has lost point target; The point target of NSCT method keeps better, but homogeneous area is level and smooth inadequately, and has lost a part of marginal information.

As can be seen from Figure 4, preceding 4 kinds of methods have been lost the point target in the white circle substantially, and the edge is also by fuzzy; The point target of NSCT method keeps better, but has lost a part of edge in the square region, and homogeneous area is level and smooth inadequately; The present invention has kept detailed information such as point target and edge preferably, and has improved the flatness of homogeneous area than additive method.

From Fig. 5, Fig. 6 and Fig. 7, all can draw similar conclusion, in a word, the present invention can not only effectively suppress speckle noise, and can effectively keep image detail features such as edge of image and point target.

Coherent spot rejection for concrete comparison the whole bag of tricks, provided the index of estimating the relevant speckle suppression method performance of Bedfordshire area SAR image: the equivalent number of homogeneous area and the average of image, equivalent number is to weigh a common counter of the relevant speckle suppression method of SAR image, equivalent number is big more, the ability that the coherent spot inhibition is described is strong more, the mean value of pixel in the average representative image, the image average that generally requires coherent spot to suppress front and back is consistent, both differ big more, illustrate that the distortion of backward radiation degree is big more, as shown in table 1.

The index contrast of table 1Bedfordshire area SAR image coherent spot inhibition test

Index	Zone	1	Zone 2	Zone 3	Average
						Former figure	3.1199	2.937	2.6746	106.8689
Enhanced Lee filtering	28.153	19.289	38.485	106.3758
					The stationary wavelet soft-threshold	51.088	26.443	45.795	98.3034
Curve ripple hard-threshold	38.519	19.003	22.809	101.1162
					NSCT	44.792	28.583	71.877	106.8689
Method of the present invention	81.745	33.279	87.609	106.8476

The equivalent number of the homogeneous area in the table 1 in zone 1, zone 2, regional 3 three white box that are respectively among Fig. 3 (a), method of the present invention all greater than enhanced Lee filtering, stationary wavelet soft-threshold, curve ripple hard-threshold, NSCT method, illustrates the advantage of the inventive method on coherent spot inhibition ability in trizonal equivalent number; The average fundamental sum original image unanimity of the method for the present invention in the table 1 has kept the backward radiation degree.

The equivalent number of the homogeneous area in three white box in the table 2 in zone 1, zone 2, the zone 3 difference presentation graphs 5 (a), compare with enhanced Lee filtering, stationary wavelet soft-threshold, curve ripple hard-threshold, NSCT method, the present invention has remarkable advantages, the advantage of the inventive method on coherent spot inhibition ability is described, though the image average of the inventive method is in full accord not as the average and the original image average of NSCT method, but also Shi Fen the average near original image illustrates that the distortion of backward radiation degree is very little.

The SAR image coherent spot inhibition test index contrast of table 2Horse track one area

Index	Zone	1	Zone 2	Zone 3	Average
						Former figure	12.327	18.485	10.659	74.2231
Enhanced Lee filtering	66.511	222.85	69.615	73.7958
					The stationary wavelet soft-threshold	84.085	411.04	76.927	66.6424
Curve ripple hard-threshold	80.802	135.6	61.12	68.8308
					NSCT	90.093	354.09	99.258	74.2231
Method of the present invention	102.36	433.51	133.54	74.2147

The equivalent number of the homogeneous area in three white box in the table 3 in zone 1, zone 2, the zone 3 difference presentation graphs 6 (a), the equivalent number of the inventive method also equivalent number than enhanced Lee filtering, stationary wavelet soft-threshold, curve ripple hard-threshold, NSCT method is big, and homogeneous is also very near the average of original image.

The SAR image coherent spot inhibition test index contrast of table 3Horse track two area

Index	Zone	1	Zone 2	Zone 3	Average
						Former figure	14.692	13.202	9.5038	83.8504
Enhanced Lee filtering	217.19	122.21	90.753	83.6654
					The stationary wavelet soft-threshold	304.56	141.67	108.97	77.7831
Curve ripple hard-threshold	170.21	156.85	52.489	79.2257
					NSCT	210.4	127.77	173.46	83.8504
Method of the present invention	610.1	229.42	267.55	83.9305

The equivalent number of the homogeneous area in three white box in the table 4 in zone 1, zone 2, the zone 3 difference presentation graphs 7 (a), the equivalent number of the inventive method equivalent number than enhanced Lee filtering, stationary wavelet soft-threshold, curve ripple hard-threshold, NSCT method equally is big, illustrates that the inventive method can effectively suppress the coherent speckle noise of SAR image.

The SAR image coherent spot inhibition test index contrast of table 4Stanwick area

Index	Zone	1	Zone 2	Zone 3	Average
						Former figure	18.142	15.974	11.026	128.3673
Enhanced Lee filtering	163.44	134.09	97.811	128.0123
					The stationary wavelet soft-threshold	262.18	211.61	89.685	121.4968
Curve ripple hard-threshold	188.76	148.42	51.123	123.1420
					NSCT	242.27	220.1	186.08	106.8689
Method of the present invention	337.7	275.27	207.64	128.5128

Claims (6)

1. the relevant speckle suppression method of the SAR image based on second generation curve wave conversion comprises the steps:

(1) the test pattern I that chooses is carried out second generation curve wave conversion, be divided into 5 straton bands;

(2) keep the 1st layer coefficient constant, the 5th layer the whole zero setting of coefficient;

(3) estimate the parameter vector θ of the 2nd～4 layer mixed Gauss model respectively with the EM method;

(4) the 2nd～4 layer coefficient is carried out mark, be labeled as 1 coefficient and remain unchanged, be labeled as 0 coefficient zero setting, be labeled as 2 coefficient and shrink with Bayes's contraction factor;

(5) obtain reconstructed image R to carrying out second generation curve ripple inverse transformation through the coefficient after step (2) and step (4) processing;

(6) detect the edge of reconstructed image R, and its homogeneous area is carried out mean filter, obtain image U after the filtering;

(7) original image and filtered image subtraction are obtained error image: V=I-U, carry out the Anisotropic Nonlinear diffusion to this error image and iterate to v k time ^k

(8) with v ^kWith image U addition after the filtering, obtain the image D=U+v after coherent spot suppresses ^k

2. the SAR image according to claim 1 speckle suppression method that is concerned with, the described parameter vector θ that estimates the 2nd～4 layer mixed Gauss model with the EM method respectively of step (3) wherein, carry out as follows:

(2a) establish the matrix of coefficients that Y is the S layer, S=2,3,4 with the coefficient histogram of the mixed Gaussian approximation of function Y of two zero-means, obtains the mixed Gaussian function and is:

$f\left(y_n\right)={\mathrm{\&Sigma;}}_{m=1}^{m=2}{\mathrm{\&epsiv;}}_m{f}_m\left(y_n\right)$

Wherein

Be respectively f _m(y _n) weights and variance,

With parameter

Be designated as the parameter vector θ of mixed Gauss model, promptly

(2b) select the initial value of parameter to be: ε ₁=ε ₂=1/2,

${\mathrm{\&sigma;}}_2^2=2(\max -\min )/3;$

(2c) calculate each sample y _nThe posterior probability of Gaussian probability-density function:

$E\left(h_{\mathrm{nm}}\right)=\frac{f_m\left(y_n\right){\mathrm{\&epsiv;}}_m}{{\mathrm{\&Sigma;}}_{m=1}^{m=2}{f}_m\left(y_n\right){\mathrm{\&epsiv;}}_m};$

(2d) use formula

Parameter vector θ is carried out the iteration training, and N is the number of samples among the Y;

(2e) when

The time, stop iteration, obtain parameter vector θ, J element representing the parameter vector that the i time iteration obtains.

3. the SAR image according to claim 1 speckle suppression method that is concerned with, wherein the said coefficient to the 2nd～4 layer of step (4) carries out mark, carries out as follows:

(3a) structure, average identical with original image I size is 0, variance is 1 stochastic matrix E, it is carried out 5 layers of curve wavelength-division separates, obtain the matrix of coefficients Ct{s}{w} on different scale S and the direction W, poor with the noise criteria that didactic prior estimate method is estimated this matrix of coefficients respectively:

And the noise criteria difference σ of the matrix of coefficients C{s}{w} of SAR image curve wave system number on same scale and direction _Nc

(3b) establish threshold value T ₁=3 σ _Nc, T ₂=0.5 σ _Ncσ _Sw, to each the coefficient C among the above-mentioned C{s}{w} (i j) judges, if | C (i, j) |＞T1, then (i j) gets its 3 * 3 field window for the center, two above absolute values are arranged greater than T as if in this window with C ₁Coefficient, then these coefficients are labeled as 1, remaining is labeled as 2;

(3c) to being labeled as 2 coefficient, if | C (i, j) |＜T ₂, then (i j) is labeled as 0 C.

4. the SAR image according to claim 1 speckle suppression method that is concerned with, wherein said 2 the coefficient of being labeled as of step (4) shrinks with Bayes's contraction factor, is to obtain not having the estimation of coefficient of making an uproar by following formula

$\hat{x}=\underset{m=\mathrm{1,2}}{\mathrm{\&Sigma;}}\frac{{\mathrm{\&epsiv;}}_m{f}_m\left(y_n\right)}{f\left(y_n\right)}\frac{|{\mathrm{\&sigma;}}_m^2-{\mathrm{\&sigma;}}_n^2|}{{\mathrm{\&sigma;}}_m^2}{y}_n.$

5. the SAR image according to claim 1 speckle suppression method that is concerned with, the edge of the said detection reconstructed image of step (6) R wherein, and its homogeneous area carried out mean filter obtains image U after the filtering, carries out as follows:

(5a) to each pixel among the reconstructed image R, get 3 * 3 fields at the center, in this field, set 12 directions, note P ₁And P ₂Be respectively the mean value of directional ray both sides pixel, establish R=P ₁/ P ₂, if R≤1, then ratio detects factor r=R, otherwise r=R ^-1, remember that the minimum value of the ratio detection factor on 12 directions is r _Min

(5b) establish threshold value T ₀, T ₀Span be: 0.5≤T ₀≤ 1, if r _Min＞T ₀, think that then current point is a marginal point, be labeled as 1, otherwise be labeled as 0, after all element markings are intact, if pixel is labeled as 1, be that 3 * 3 field window is got at the center then, be less than 4, then this pixel is labeled as 0 if this field window internal labeling is 1 number with it;

If (i j) is marked as 0 to the pixel R (5c) among the R, and then (i j) gets its 5 * 5 field W for the center with R ₁, if W ₁In pixel only be less than 4 point and be labeled as 1, then (i, pixel value j) is taken as W to R ₁The mean value of interior all pixels; (i j) is the center, gets its 3 * 3 field W otherwise with R ₂, if W ₂In pixel only be less than 4 point and be labeled as 1, then (i, pixel value j) is taken as W to R ₂In the mean value of all pixels, otherwise R (i, pixel value j) is taken as W ₂In all are labeled as the mean value of 0 pixel, all are labeled as 0 processes pixel and obtain image U after the filtering after intact.

6. the SAR image according to claim 1 speckle suppression method that is concerned with, wherein step (7) is described carries out the Anisotropic Nonlinear diffusion to error image and iterates to v k time ^k, carry out iteration according to following formula:

$v_{i,j}^{k+1}=v_{i,j}^k+t\underset{r,s=-1}{\overset{1}{\mathrm{\&Sigma;}}}\frac{g\left({\sqrt{2}}^{1-\left|r\right|-\left|s\right|}\right|v_{i+r,j+s}^k-v_{i,j}^k\left|\right)(v_{i+r,j+s}^k-v_{i,j}^k)}{r^2+s^2}$

G (x)=1/ (1+x wherein ²/ ρ ²), (r, s) ≠ (0,0), t=1/8, ρ=6, v ⁰=V, v _{I, j}For the i of v is capable, the pixel of j row.

Images