CN106934805A - SAR image superpixel segmentation method based on Gamma filtering - Google Patents

Abstract

The present invention proposes a kind of SAR image superpixel segmentation method based on Gamma filtering, the technical problem low for solving super-pixel segmentation result precision present in the existing superpixel segmentation method based on filtering, realizes that step is：The original SAR image being input into is filtered using mean filter, obtains the average value and standard deviation of pixel grey scale；The variation coefficient of the coefficient of variation, the coefficient of variation of coherent speckle noise and SAR image of pixel grey scale average value is calculated respectively；Judge the relation of each coefficient, it is determined whether carry out Gamma filtering；Coherent speckle noise noise reduction is carried out to the original SAR image being input into using Gamma filtering methods；Super-pixel segmentation is carried out to the SAR image after noise reduction, multiple super-pixel block is obtained and is exported.Present invention reduces the influence of coherent speckle noise in SAR image, the degree of accuracy of SAR image super-pixel segmentation is improve, can be used for the target detection to SAR image, identification and classify.

Description

SAR image superpixel segmentation method based on Gamma filtering

Technical field

The invention belongs to technical field of image processing, it is related to a kind of SAR image superpixel segmentation method, and in particular to a kind of Based on the SAR image superpixel segmentation method of Gamma filtering, can be used for the target detection to SAR image, identification and classify.

Background technology

Synthetic aperture radar (Synthetic aperture radar, SAR) is a kind of to be operated in the relevant of microwave band Imaging radar, it turns into current remote sensing observations with its high-resolution and round-the-clock, round-the-clock, the data retrieval capabilities of large area Important means, be widely used at aspects such as resource, environment, archaeology and military affairs.Ground target is to radar emission electricity The back scattering of magnetic wave forms SAR image.With a large amount of acquisitions of SAR image, the image understanding of intelligence turns into Interpretation Technology Current study hotspot.Used as image understanding and the critical task for interpreting, the super-pixel segmentation of image is by an auxiliary input Adjacent pixel is divided into image block by SAR image according to different attributes, content, feature etc..Piece image is carried out effectively And accurate super-pixel segmentation, it is possible to increase further to the degree of accuracy of image understanding, more effective informations are obtained, and it is very big The complexity of follow-up SAR image process task is reduced in degree.But the intrinsic coherent imaging mechanism of SAR image makes SAR image In the presence of the coherent speckle noise that cannot be eliminated, cause the gray value of each pixel that random change occurs, so as to influence SAR image Super-pixel segmentation effect and follow-up SAR image result.

At present, Most scholars are mainly solved the above problems using following three kinds of methods：1st, it is super based on statistical model Pixel dividing method；2nd, the superpixel segmentation method based on yardstick.3rd, the superpixel segmentation method based on filtering.

First method is to carry out probability statistics to the coherent speckle noise in SAR image, it is assumed that probability distribution meets Rayleigh is distributed, Gamma distributions, K distributions etc., and probability statistics model is fused to the model of conventional super-pixel segmentation algorithm In, so as to obtain being suitable to the super-pixel segmentation result of SAR image, the weak point of the method is, although make use of SAR image Coherent speckle noise information carries out the super-pixel segmentation of SAR image, but it is based on statistical information to be calculated, so as to increase The complexity of calculating process, reduces the operational efficiency of algorithm.

The core concept of second method is that SAR image is regarded as the texture of different scale, according to different classes of target Different texture this property is presented, by the super-pixel segmentation of SAR image, the identification to different texture in SAR image is converted into. But the process that such method is realized is complicated, the result for obtaining compared to other superpixel segmentation methods, not significantly Advantage.

The third method is that the coherent speckle noise in SAR image is filtered using wave filter first, reduces coherent spot The influence that noise is processed SAR image, so as to obtain the SAR image after noise reduction, then carries out super picture to the SAR image after noise reduction Element segmentation.The method realizes simple, the super-pixel segmentation obtained in the case where complexity is smaller than the complexity of first two method The degree of accuracy it is identical.In such method, it is important to carry out coherent speckle noise to SAR image and filter, and this technology is non- It is often ripe, which includes using median filter, local wave filter, Lee wave filters, Sigma wave filters, Frost wave filters Coherent speckle noise is removed Deng pretreatment is carried out to SAR image.But these wave filters are filtered to the coherent speckle noise of SAR image When, although can filter coherent speckle noise, but marginal information, texture information and linear character information in image etc. can not Kept well.

The content of the invention

It is an object of the invention to overcome the shortcomings of that above-mentioned prior art is present, it is proposed that a kind of based on Gamma filtering SAR image superpixel segmentation method, in the case where coherent speckle noise is disturbed it in reducing SAR image, effectively maintains figure The marginal information of picture, texture information and linear character information, for solving the existing superpixel segmentation method based on filtering in The low technical problem of the super-pixel segmentation degree of accuracy of presence.

To achieve the above object, the technical scheme that the present invention takes comprises the following steps：

(1) it is input into original SAR image；

(2) the original SAR image being input into is filtered using mean filter, obtains original SAR image pixel grey scale Average valueWith standard deviation sigma (i, j)；

(3) using the average value of original SAR image pixel grey scaleWith standard deviation sigma (i, j), original SAR image is calculated Pixel grey scale average valueCoefficient of variation C_I：

(4) the coefficient of variation C of coherent speckle noise in original SAR image is calculated_u：

Wherein, L represent original SAR image regarding number；

(5) the variation coefficient C of original SAR image is calculated_max:

(6) C is worked as_u≤C_I≤C_maxWhen, using Gamma filtering methods, the original SAR image to being input into carries out coherent speckle noise Noise reduction, obtains the SAR image I after noise reduction_Gam(i,j)：

Wherein, α is heterogeneous parameter, and

(7) to the SAR image I after noise reduction_Gam(i, j) carries out super-pixel segmentation, obtains multiple super-pixel block and exports.

The present invention compared with prior art, with advantages below：

The present invention schemes due to filtering original SAR using Gamma first before super-pixel segmentation is carried out to SAR image Coherent speckle noise as in, obtain the SAR image after noise reduction carries out super-pixel segmentation to it again, compared with prior art, passes through The present invention when the influence that coherent speckle noise is processed SAR image is reduced, can effectively maintain image marginal information, Texture information and linear character information, improve the degree of accuracy of super-pixel segmentation.

Brief description of the drawings

Fig. 1 realizes FB(flow block) for of the invention；

Fig. 2 is the super-pixel image that present invention simulation SAR image is produced；

Fig. 3 is the super-pixel image that the present invention is produced containing textured simulation SAR image；

Fig. 4 is the super-pixel image that true SAR image of the invention is produced.

Specific embodiment

With reference to the accompanying drawings and examples, the present invention is described in further detail.

Embodiment 1

Reference picture 1：The present invention comprises the following steps：

Step 1：It is input into original SAR image.

Original SAR image is divided into simulation SAR image and the major class of true SAR image two, in the present embodiment, original SAR figures As using the simulation SAR image without texture.

Step 2：The original SAR image being input into is filtered using mean filter, obtains original SAR image pixel grey scale Average valueWith standard deviation sigma (i, j)；

Mean filter is typical linear filtering algorithm, when mean filter is carried out, it is necessary to a Filtering Template, uses template In the average value of entire pixels replace original pixel value.For the selection of Filtering Template, the mould of 3*3 is generally comprised The template of plate, the template of 5*5, or more big window, the in the present invention selection for template is not limited, but due to window It is bigger, can be influenceed for filtered result, the weight coefficient matrix template of 3*3 is selected in an embodiment of the present invention, its Expression formula is as follows：

Wherein, s is the row of pattern matrix expression formula, and t is the row of pattern matrix expression formula；

(2a) is filtered using weight coefficient matrix template w (s, t), the original SAR image to being input into, and obtains original SAR The average value of image pixel gray level

(2b) calculates the standard deviation sigma (i, j) of original SAR image pixel grey scale：

Wherein, N represents the number of pixels on the abscissa of the original SAR image of input, and M represents the original SAR figures of input Number of pixels on the ordinate of picture.

Step 3：Using the average value of original SAR image pixel grey scaleWith standard deviation sigma (i, j), original SAR is calculated Image pixel gray level average valueCoefficient of variation C_I：

Step 4：Calculate the coefficient of variation C of coherent speckle noise in original SAR image_u：

Wherein, L represent original SAR image regarding number；

Step 5：Calculate the variation coefficient C of original SAR image_max:

Step 6：Judge the coefficient of variation C of original SAR image pixel grey scale average value_IWith coherent spot in original SAR image The coefficient of variation C of noise_uAnd original SAR image variation coefficient C_maxRelation, work as C_u≤C_I≤C_maxWhen, using Gamma filtering sides Method, the original SAR image to being input into carries out coherent speckle noise noise reduction, obtains the SAR image I after noise reduction_Gam(i,j)：

Wherein, α is heterogeneous parameter, and

Work as C_I＜ C_uWhen, then the grey scale pixel value of SAR image is the average value of original SAR image pixel grey scale

Work as C_I＞ C_maxWhen, then the grey scale pixel value of SAR image is grey scale pixel value I (i, j) of original SAR image.And it is right In both of these case, Gamma filtering is not carried out to it.

Step 7：To the SAR image I after noise reduction_Gam(i, j) carries out super-pixel segmentation, obtains multiple super-pixel block and exports. This step is implemented as follows：

(7a) calculates the SAR image I after noise reduction_GamThe poor W (i, j) of the pixel value of adjacent pixel in (i, j) horizontal direction：

W (i, j)=| I_Gam(i,j)-I_Gam(i-1,j)|

Wherein, I_Gam(i-1 j) represents pixel value I_GamThe pixel value of adjacent pixel in (i, j) horizontal direction；

(7b) calculates the SAR image I after noise reduction_GamThe poor V (i, j) of the pixel value of (i, j) vertically adjacent pixel：

V (i, j)=| I_Gam(i,j)-I_Gam(i,j-1)|

Wherein, I_Gam(i, j-1) represents pixel value I_GamThe pixel value of (i, j) vertically adjacent pixel；

(7c) calculates the SAR image I after noise reduction_GamGrad G (i, j) of (i, j), computing formula is：

G (i, j)=W (i, j)+V (i, j)

(7d) sets the SAR image I after noise reduction_GamThe number K of super-pixel block in (i, j), and calculate super-pixel initial seed The average distance d of point：

Wherein, S is the SAR image I after noise reduction_GamThe gross area of (i, j), and S=MN；

(7e) is to SAR image I_GamThe position of the super-pixel initial seed point on (i, j) edge is disturbed, and is caused a deviation from SAR image I_GamGrad at the edge of (i, j), the i.e. initial seed point is less than SAR image I_GamThe gradient of (i, j) edge Value G (i, j), obtains super-pixel seed point；

(7f) enters row bound growth to super-pixel seed point, is collided until neighbouring super pixels seed point is expanded to border When, stop border and increase, obtain multiple super-pixel block and export.

Embodiment 2

Embodiment 2 is identical with other steps of embodiment 1, and the species of the only original SAR image to being used in step 1 is carried out Adjustment, the original SAR image of the present embodiment is using containing textured simulation SAR image.

Embodiment 3

Embodiment 3 is identical with other steps of embodiment 1, and the species of the only original SAR image to being used in step 1 is carried out Adjustment, the original SAR image of the present embodiment uses true SAR image.

Below in conjunction with emulation experiment, effect of the invention is further described.

1st, simulated conditions：

Emulation of the invention be Intel (R) Core (TM) 2Duo of dominant frequency 2.4GHZ, the hardware environment of internal memory 4GB and Carried out under the software environment of MATLAB R2015a.In the present invention, the SAR image of use is and regards coherent speckle noise containing 1 SAR image, the super-pixel block of generation is set to 2500, super-pixel segmentation is carried out for SAR image, generally by super-pixel block Number is arranged between 1000 to 5000.

2nd, emulation content and interpretation of result：

Emulation one, super-pixel segmentation is produced with the present invention to simulation SAR image, produces the super-pixel point as shown in Fig. 2 (c) Cut result.

Reference picture 2,

Fig. 2 (a) is the simulation SAR image of input, and size is 512 × 512 pixels；Fig. 2 (b) is the simulation SAR to being input into Image carries out the SAR image after the noise reduction obtained after Gamma filtering, be can be seen that from Fig. 2 (b) after carrying out Gamma filtering, phase Dry spot noise is largely filtered out, and effectively maintains the edge of image；Fig. 2 (c) is to the SAR figures after noise reduction As producing super-pixel image, from Fig. 2 (c) as can be seen that the super-pixel segmentation obtained after Gamma is filtered effectively is maintained Edge compatible degree, so that the degree of accuracy of super-pixel segmentation is also improved.

Emulation two, with the present invention to producing super-pixel segmentation containing textured simulation SAR image, produces as shown in Fig. 3 (c) Super-pixel segmentation result.

Reference picture 3,

Fig. 3 (a) is to be input into containing textured simulation SAR image, and size is 512 × 512 pixels；Fig. 3 (b) is to input The SAR image after the noise reduction obtained after Gamma filtering is carried out containing textured simulation SAR image, from Fig. 3 (b) as can be seen that entering After row Gamma filtering, coherent speckle noise is largely filtered out, and it is thin effectively to maintain the edge and texture of image Section information；Fig. 3 (c) is to produce super-pixel image to the SAR image after noise reduction, from Fig. 3 (c) as can be seen that being filtered by Gamma The super-pixel segmentation for obtaining afterwards effectively maintains the marginal information and grain details information of image, so that the standard of super-pixel segmentation Exactness is also improved.

Emulation three, super-pixel segmentation is produced with the present invention to true SAR image, produces the super-pixel point as shown in Fig. 4 (c) Cut result.

Reference picture 4,

Fig. 4 (a) is the true SAR image of input, and the image is from U.S. Sandia National Labs (Sandia National Laboratories) real-time processing carried SAR, which show California, USA China Lake airports Area, its resolution ratio is 3 meters, and size is 522 × 466 pixels；Fig. 4 (b) is to carry out Gamma filters to the true SAR image being input into SAR image after the noise reduction obtained after ripple, from Fig. 4 (b) as can be seen that after carrying out Gamma filtering, coherent speckle noise is in very great Cheng It is filtered out on degree, and effectively maintains the marginal information of image, grain details information and linear character information；Fig. 4 (c) It is that super-pixel image is produced to the SAR image after noise reduction, from Fig. 4 (c) as can be seen that the super picture obtained after Gamma is filtered Element segmentation effectively maintains marginal information, grain details information and the linear character information of image, so that super-pixel segmentation The degree of accuracy be also improved.

Can be seen that the present invention from above the simulation experiment result can effectively keep the edge of image, when containing in SAR image When linear feature and texture information, it is also possible to linear character and the effective holding of grain details information, so as to improve super The degree of accuracy of pixel segmentation result.

Claims (3)

1. a kind of SAR image superpixel segmentation method based on Gamma filtering, comprises the following steps：

(1) it is input into original SAR image；

(2) the original SAR image being input into is filtered using mean filter, obtains the average of original SAR image pixel grey scale ValueWith standard deviation sigma (i, j)；

(3) using the average value of original SAR image pixel grey scaleWith standard deviation sigma (i, j), original SAR image pixel is calculated Average grayCoefficient of variation C_I：

$C_I=\frac{\mathrm{\σ}(i,j)}{\stackrel{\‾}{I}(i,j)};$

(4) the coefficient of variation C of coherent speckle noise in original SAR image is calculated_u：

$C_u=\frac{1}{L}$

Wherein, L represent original SAR image regarding number；

(5) the variation coefficient C of original SAR image is calculated_max:

$C_{max}=\sqrt{2/L-1};$

(6) C is worked as_u≤C_I≤C_maxWhen, using Gamma filtering methods, the original SAR image to being input into carries out coherent speckle noise drop Make an uproar, obtain the SAR image I after noise reduction_Gam(i,j)：

$I_{Gam}(i,j)=\frac{(\mathrm{\α}-L-1)\stackrel{\‾}{I}(i,j)+\sqrt{{(\mathrm{\α}-L-1)}^2{\stackrel{\‾}{I}}^2(i,j)+4\mathrm{\α}L\stackrel{\‾}{I}(i,j)I(i,j)}}{2\mathrm{\α}}$

Wherein, α is heterogeneous parameter, and

(7) to the SAR image I after noise reduction_Gam(i, j) carries out super-pixel segmentation, obtains multiple super-pixel block and exports.

2. it is according to claim 1 based on Gamma filtering SAR image superpixel segmentation method, it is characterised in that：Step (2) the use mean filter described in is filtered to the original SAR image being input into, and realizes that step is as follows：

(2a) is filtered using weight coefficient matrix template w (s, t), the original SAR image to being input into, and obtains original SAR image The average value of pixel grey scale

$\stackrel{\‾}{I}(i,j)=\underset{s=-1}{\overset{1}{\Σ}}\underset{t=-1}{\overset{1}{\Σ}}w(s,t)I(i+s,j+t)$

Wherein,S is the row of pattern matrix expression formula, and t is the row of pattern matrix expression formula；

(2b) calculates the standard deviation sigma (i, j) of original SAR image pixel grey scale：

$\mathrm{\σ}(i,j)=\sqrt{\frac{1}{N}\frac{1}{M}\underset{i=1}{\overset{N}{\Σ}}\underset{j=1}{\overset{M}{\Σ}}{\left(I\right(i,j)-\stackrel{\‾}{I}(i,j\left)\right)}^2}$

Wherein, N represents the number of pixels on the abscissa of the original SAR image of input, and M represents the original SAR image of input Number of pixels on ordinate.

3. it is according to claim 1 based on Gamma filtering SAR image superpixel segmentation method, its feature in：Step (7) described in the SAR image I after noise reduction_Gam(i, j) carries out super-pixel segmentation, realizes that step is as follows：

(7a) calculates the SAR image I after noise reduction_GamThe poor W (i, j) of the pixel value of adjacent pixel in (i, j) horizontal direction：

W (i, j)=| I_Gam(i,j)-I_Gam(i-1,j)|

Wherein, I_Gam(i-1 j) represents pixel value I_GamThe pixel value of adjacent pixel in (i, j) horizontal direction；

(7b) calculates the SAR image I after noise reduction_GamThe poor V (i, j) of the pixel value of (i, j) vertically adjacent pixel：

V (i, j)=| I_Gam(i,j)-I_Gam(i,j-1)|

Wherein, I_Gam(i, j-1) represents pixel value I_GamThe pixel value of (i, j) vertically adjacent pixel；

(7c) calculates the SAR image I after noise reduction_GamGrad G (i, j) of (i, j), computing formula is：

G (i, j)=W (i, j)+V (i, j)

(7d) sets the SAR image I after noise reduction_GamThe number K of super-pixel block in (i, j), and calculate super-pixel initial seed point Average distance d：

$d=\frac{S}{K}$

Wherein, S is the SAR image I after noise reduction_GamThe gross area of (i, j), and S=MN；

(7e) is to SAR image I_GamThe position of the super-pixel initial seed point on (i, j) edge is disturbed, and causes a deviation from SAR figures As I_GamGrad at the edge of (i, j), the i.e. initial seed point is less than SAR image I_GamThe Grad G of (i, j) edge (i, j), obtains super-pixel seed point；

(7f) enters row bound growth to super-pixel seed point, when neighbouring super pixels seed point is expanded to border to collide, Stop border increasing, obtain multiple super-pixel block and export.