CN103077499B - SAR (Synthetic Aperture Radar) image pre-processing method based on similar block - Google Patents

Abstract

The invention discloses a SAR image preprocessing method based on similar blocks, which mainly overcomes the problems that the existing methods cannot maintain image texture details and compress strong reflection point targets. The implementation process is: (1) For the input SAR image, calculate the variance coefficient matrix of the image; (2) Use the variance coefficient to divide the image pixels into a smooth area and a texture detail area; (3) Use the mean value for the smooth area Filter to obtain the processed pixel value; (4) Rotate the texture detail area; (5) Obtain the similarity measurement formula according to the ratio probability distribution, and calculate the similarity between the point to be processed and its 8 neighboring pixels; (6 ) Take the average value of the point to be processed and the three points most similar to it as the gray value of the pixel. (7) Calculate the restored values of all pixels to obtain the preprocessed image. The invention can better protect the texture and structure information of the image while suppressing the noise, can well keep the strong reflection point target, and can be used for the preprocessing before the application of the image.

Description

SAR image preprocessing method based on similar blocks

technical field

The invention belongs to the technical field of image processing, and relates to a SAR image preprocessing method based on homogeneous and similar blocks, which can be used for preprocessing before the application of the SAR image.

Background technique

The image formed by synthetic aperture radar SAR has the characteristics of all-weather, all-time, high resolution and strong penetrating ability. Therefore, this image is widely used in the fields of target recognition, transformation detection and water surface surveillance. However, SAR images are easily corrupted by multiplicative noise, which comes from the continuous interference of backscattered radar reflections. This speckle noise destroys the radiometric resolution of SAR images and affects the performance of background analysis and understanding tasks.

With the widespread application of SAR images in military and civilian applications, SAR image processing has become another research focus of SAR technology. SAR images contain a large number of coherent spots caused by the random interference of coherent echoes from scattering points on the imaging interface, and these speckle noises will greatly reduce the effectiveness of image segmentation, edge detection, feature extraction, target recognition and other information processing techniques , so that the SAR image processing cannot achieve the expected goal. Therefore, the preprocessing before processing the SAR image is an indispensable process.

The purpose of preprocessing is to suppress the image noise while retaining the feature information of the image, such as image texture, edge and point target. This is very different from image denoising. Denoising is to remove noise as much as possible, so that the texture of the image will be smoothed, and the image structure information such as edges, linear bodies, points and other objects will be blurred to a certain extent. Or filter out, which is not conducive to subsequent image processing. At present, there are mainly two kinds of SAR image preprocessing methods: 1) 3×3 block mean method; 2) 3×3MMSE method. The preprocessed image obtained by the 3×3 block mean method has a good smoothing ability in the homogeneous area, but the brightness of the strong reflection point target is severely compressed, and the edge is also largely blurred; the 3×3MMSE method can better Maintaining the smoothing ability of homogeneous areas will also compress the brightness of strong reflection point targets, and the texture details of the image will become blurred, which is not conducive to subsequent processing.

Contents of the invention

The purpose of the present invention is to address the deficiencies in the prior art above, and propose a SAR image preprocessing method based on homogeneous similar blocks, so that the processed image can well maintain the smoothness of the homogeneous region, while protecting the image The strong reflection point target and the image texture structure information are beneficial to the subsequent processing of the image.

The technical key to realize the object of the present invention is to rotate the image block when calculating the similarity with the block, so that the similarity calculation of the similar block is more accurate; at the same time, the ratio distribution probability is introduced on the basis of the homogeneous similar block, and its technical scheme Including the following steps:

(1) For an L-view SAR image v with an input size of (m, n), calculate the variance coefficient CV of all pixels, and obtain the variance coefficient matrix K;

(2) Set the variance coefficient classification threshold T _cv to classify the input SAR image v, if the variance coefficient of the pixel point x _i,j in the image v in the variance coefficient matrix K is less than the threshold T _cv , then perform the step ( 3), otherwise execute step (4);

(3) The mean value of the pixels in the 3×3 block centered on the pixel point x _i,j is used as the pixel value after preprocessing of the pixel point;

(4) For pixel x _l in the 8 neighborhoods of pixel x _{i, j} , l=1, 2,...,8, take a 3×3 block v _l centered on pixel x _{l ,} and take The rotation operation makes the homogeneous area in the 3×3 block v _i,j centered on the pixel point x _i ,j and the homogeneous area in the 3×3 block v _l centered on the pixel point x _l be in the same position, and v _l The rotated block is denoted as ;

(5) Calculate the similarity distance between pixel x _i,j and its 8 neighbor pixel x _l , l=1,2,...,8 based on block ratio probability:

5a) Take the 3×3 block v _i,j centered on the pixel point x _i,j , and obtain its neighborhood pixel point x _l by step (4), where l=1,2,…,8 flipped 3× 3 blocks ;

5b) Calculate the above two pixel blocks v _i,j and The ratio r _i,k of :

$r_{i,k}=\min \{\frac{v_{i,k}}{v_{l,k}},\frac{v_{l,k}}{v_{i,k}}\},$ r _{i, k} ∈ [0,1], l=1,2,...,8, k=1,2,...,9,

Among them, v _i,k represents the gray value of the kth pixel of the 3×3 block v _i,j centered on the pixel point x _i,j to be processed, and v _l,k represents the 3× pixel point x _l as the center 3 blocks v _l rotated blocks The gray value of the kth pixel of ;

5c) Using the ratio distribution probability formula to calculate the probability p(r _i,k ) of the ratio r _i,k of the pixel point x _i,j ;

5d) Define the similarity distance d _l between the pixel x _i,j to be processed and its 8 neighbor pixel x _l as:

$d_l=\underset{k=1}{\overset{3\×3}{\mathrm{\Σ}}}\log \left(p\left(r_{i,k}\right)\right),$ k=1,2,...,9, l=1,2,...,8;

(6) Sort the similarity distances d ₁ , d ₂ ,..., d ₈ obtained in step (5) in ascending order, and the sorted result is Take the sorted distance as The pixel point of is used as the similar point of pixel point x _{i , j,} and the average value of pixel point x i, j and the gray value of these three similar points is taken as the gray value of pixel point x _{i, j} after preprocessing.

(7) Repeat steps (2)~(6) to calculate the restored values of all pixels and obtain the preprocessed image.

Compared with the prior art, the present invention has the following advantages:

1. The present invention uses the variance coefficient to divide the SAR image into two types of regions: homogeneous region and texture region, and adopts different methods for different regions, thereby improving the processing accuracy.

2. The present invention adopts the rotation operation, so that the homogeneous area in the block is at the same position, so the calculated similarity is more accurate, so that the current point can be repaired with a pixel with a higher similarity, and the texture of the image can be better maintained and details;

3. The present invention adopts the probability similarity distance based on the direct distribution, which can more accurately calculate the similarity between SAR image pixels;

Description of drawings

Fig. 1 is a flow chart of the present invention;

Fig. 2 is the test image that the present invention uses;

Fig. 3 is the result figure that Fig. 2 is preprocessed with existing MMSE method;

Fig. 4 is the result figure that Fig. 2 is preprocessed with existing average method;

Fig. 5 is the result figure that Fig. 2 is preprocessed with the method of the present invention

Fig. 6 is an enlarged view of the details of the preprocessed graph of Fig. 2 using the existing MMSE method.

Fig. 7 is the detailed enlargement figure of the result figure that Fig. 2 is preprocessed with existing average method;

Fig. 8 is a detailed enlarged view of the preprocessed result graph of Fig. 2 by the method of the present invention.

detailed description

With reference to accompanying drawing 1, the present invention comprises the steps:

Step 1. For an L-view SAR image v with an input size of (m, n), calculate the variance coefficient CV of all pixels, and obtain the variance coefficient matrix K.

1.1) Calculate the variance coefficient CV _i,j of pixel x _i,j , i∈[1,m],j∈[1,n]:

${\mathrm{<span\; class="notranslate">\; cv</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; j</span>}}<span\; class="notranslate">\; =</span>\frac{{\mathrm{<span\; class="notranslate">\; \&sigma;</span>}}_{{\mathrm{<span\; class="notranslate">\; x</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; j</span>}}}}{{\mathrm{<span\; class="notranslate">\; \&mu;</span>}}_{{\mathrm{<span\; class="notranslate">\; x</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; j</span>}}}}$

in, is the standard deviation of the gray value of all pixels in the 7×7 neighborhood centered on pixel x _i,j , is the mean value of the gray value of all pixels in the 7×7 neighborhood centered on pixel x _i,j ;

1.2) Calculate the variance coefficient of each pixel in the SAR image v, and obtain the variance coefficient matrix K ₀

K ₀ ={CV _i,j },i∈[1,m],j∈[1,n]

1.3) Perform 3×3 mean value filtering on the calculated variance coefficient matrix K ₀ to obtain the filtered variance coefficient matrix K.

Step 2, set the classification threshold T _cv , and classify the input SAR image v.

2.1) Set the threshold T _cv according to the input L-view image category:

If the input is an L-magnitude SAR image, the threshold T _cv is set as:

${\mathrm{<span\; class="notranslate">\; T</span>}}_{\mathrm{<span\; class="notranslate">\; cv</span>}}<span\; class="notranslate">\; =</span>\sqrt{\frac{\mathrm{<span\; class="notranslate">\; 1</span>}}{\mathrm{<span\; class="notranslate">\; L</span>}}<span\; class="notranslate">\; (</span>\frac{\mathrm{<span\; class="notranslate">\; 4</span>}}{\mathrm{<span\; class="notranslate">\; \&pi;</span>}}<span\; class="notranslate">\; -</span>\mathrm{<span\; class="notranslate">\; 1</span>}<span\; class="notranslate">\; )</span>}<span\; class="notranslate">\; ,</span>$

If the input is an L apparent intensity SAR image, the threshold T _cv is set as:

${\mathrm{<span\; class="notranslate">\; T</span>}}_{\mathrm{<span\; class="notranslate">\; cv</span>}}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 0.82</span>}\sqrt{\frac{\mathrm{<span\; class="notranslate">\; 1</span>}}{\mathrm{<span\; class="notranslate">\; L</span>}}}<span\; class="notranslate">\; ,</span>$

2.2) Compare the variance coefficient CV _i,j of the pixel point x _i, j with the threshold T _cv to divide the image into two types of processing. If the variance coefficient CV _i,j is smaller than the threshold T _cv , execute step 3, otherwise execute Step 4;

Step 3, take the mean value of the gray value of the pixel points in the 3×3 block centered on the pixel point x _{i ,j} as the preprocessed gray value of the pixel point x i,j.

Step 4, for pixel point x _l in the 8 neighbors of pixel point x _i,j , l=1,2,…,8, take a 3×3 block v _l centered on pixel point x _{l ,} and take The rotation operation makes the homogeneous area in the 3×3 block v _i,j centered on the pixel point x _i ,j and the homogeneous area in the 3×3 block v _l centered on the pixel point x _l be in the same position, and v _l The rotated block is denoted as .

Step 5, calculate the similarity distance between pixel x _{i, j} and its 8 neighbor pixel x _l :

5.1) Take the 3×3 block v _i,j centered on the pixel point x _i,j , and obtain the 3×3 block of its neighborhood pixel point x _l by step 4, l=1,2,…,8 after flipping

5.2) Calculate the above two pixel blocks v _{i, j} and The ratio r _i,k of :

$r_{i,k}=\min \{\frac{v_{i,k}}{v_{l,k}},\frac{v_{l,k}}{v_{i,k}}\},$ r _{i, k} ∈ [0,1], l=1,2,...,8, k=1,2,...,9,

Among them, v _i,k represents the gray value of the kth pixel of the 3×3 block v _i,j centered on the pixel point x _i,j to be processed, and v _l,k represents the 3× pixel point x _l as the center 3 blocks v _l rotated blocks The gray value of the kth pixel of ;

5.3) Use the ratio distribution probability formula to calculate the probability p(ri _,k ) of the ratio r _i,k of the pixel point x _i,j , which is calculated in two cases:

If the input SAR image v is a magnitude image, then:

$\mathrm{<span\; class="notranslate">\; p</span>}<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; r</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; k</span>}}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; =</span>\frac{\mathrm{<span\; class="notranslate">\; 4</span>}\mathrm{<span\; class="notranslate">\; \&Gamma;</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; 2</span>}\mathrm{<span\; class="notranslate">\; L</span>}<span\; class="notranslate">\; )</span>}{\mathrm{<span\; class="notranslate">\; \&Gamma;</span>}{<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; L</span>}<span\; class="notranslate">\; )</span>}^{\mathrm{<span\; class="notranslate">\; 2</span>}}}\frac{{{\mathrm{<span\; class="notranslate">\; r</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; k</span>}}}^{\mathrm{<span\; class="notranslate">\; 2</span>}\mathrm{<span\; class="notranslate">\; L</span>}<span\; class="notranslate">\; -</span>\mathrm{<span\; class="notranslate">\; 1</span>}}}{{<span\; class="notranslate">\; (</span>{{\mathrm{<span\; class="notranslate">\; r</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; k</span>}}}^{\mathrm{<span\; class="notranslate">\; 2</span>}}<span\; class="notranslate">\; +</span>\mathrm{<span\; class="notranslate">\; 1</span>}<span\; class="notranslate">\; )</span>}^{\mathrm{<span\; class="notranslate">\; 2</span>}\mathrm{<span\; class="notranslate">\; L</span>}}}<span\; class="notranslate">\; ,</span>$

If the input SAR image v is an intensity image, then:

$\mathrm{<span\; class="notranslate">\; p</span>}<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; r</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; k</span>}}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; =</span>\frac{\mathrm{<span\; class="notranslate">\; 2</span>}\mathrm{<span\; class="notranslate">\; \&Gamma;</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; 2</span>}\mathrm{<span\; class="notranslate">\; L</span>}<span\; class="notranslate">\; )</span>}{\mathrm{<span\; class="notranslate">\; \&Gamma;</span>}{<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; L</span>}<span\; class="notranslate">\; )</span>}^{\mathrm{<span\; class="notranslate">\; 2</span>}}}\frac{{{\mathrm{<span\; class="notranslate">\; r</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; k</span>}}}^{\mathrm{<span\; class="notranslate">\; L</span>}<span\; class="notranslate">\; -</span>\mathrm{<span\; class="notranslate">\; 1</span>}}}{{<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; r</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; k</span>}}<span\; class="notranslate">\; +</span>\mathrm{<span\; class="notranslate">\; 1</span>}<span\; class="notranslate">\; )</span>}^{\mathrm{<span\; class="notranslate">\; 2</span>}\mathrm{<span\; class="notranslate">\; L</span>}}}$

5.4) Define the similarity distance d _l between the pixel point x _i,j to be processed and its 8 neighbor pixel points x _l as:

$d_l=\underset{k=1}{\overset{3\&times;3}{\mathrm{\&Sigma;}}}\log \left(p\left(r_{i,k}\right)\right),$ k=1,2,...,9, l=1,2,...,8

Calculate the similarity distance d _l between the pixel point x _i,j and its 8 neighbor pixel point x _l respectively.

Step 6, sort the similarity distances d ₁ , d ₂ ,..., d ₈ obtained in step 5 in ascending order, and the sorted result is Take the sorted distance as The pixel point of is used as the similar point of pixel point x _{i , j,} and the average value of pixel point x i, j and the gray value of these three similar points is taken as the gray value of pixel point x _{i, j} after preprocessing.

Step 7, repeat steps 2 to 6, calculate the restored values of all pixels, and obtain the preprocessed image.

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

1. Experimental conditions and content

Experimental conditions: The input image used in the experiment is a two-view X-band amplitude SAR image (Bedfordshire), as shown in Figure 2.

Experimental method: under the above-mentioned experimental conditions, select two typical algorithms of current image preprocessing and the method of the present invention to carry out comparative experiment, they are respectively: (1) minimum mean square error (MMSE) method; (2) mean value method.

two. Experimental content

Experiment 1: Process Figure 2 with the existing MMSE method, where the block size is 3×3, and the experimental results are shown in Figure 3 and Figure 6, where Figure 6 is an enlarged view of the details of Figure 3. It can be seen from Figure 3 and Figure 6 that the MMSE method has a good ability to suppress noise in homogeneous areas, but the edges and details become blurred, and the brightness of strong reflection points is compressed.

Experiment 2: Process Figure 2 with the existing mean value method, where the block size is 3×3, and the experimental results are shown in Figure 4 and Figure 7, where Figure 7 is a detailed enlarged view of Figure 4. It can be seen from the results in Figure 7 and Figure 4 that the stability of the noise suppression ability of the mean method is better than that of the MMSE method, but it cannot preserve the edge and texture information of the image very well, and the brightness of the strong reflection point target is severely compressed.

Experiment three: process Fig. 2 with the inventive method, wherein block size is 3 * 3, experimental result is as shown in Fig. 5, Fig. 8, and wherein Fig. 8 is the detailed enlarged view of Fig. 5, by Fig. 8 and Fig. 5 It can be seen from the result figure that the present invention can effectively suppress image noise, and well preserve the edge and texture information of the image, and has good retention for strong reflection point targets.

The above experimental results show that the overall performance of the present invention is superior to other two similar preprocessing methods, and it can smooth the noise well while maintaining details such as edges and textures of natural images, and it has a good preservation of strong reflection point targets. sex.

Claims (4)

1. A SAR image preprocessing method based on similar blocks comprises the following steps:

(1) calculating the variance coefficient CV of all pixel points for the L-view SAR image v with the input size of (m, n) to obtain a variance coefficient matrix K;

(2) setting a variance coefficient classification threshold T_cvClassifying the input SAR image v, if the pixel point x in the image v_i,jThe coefficient of variance in the coefficient of variance matrix K is less than a threshold T_cvIf not, executing the step (3), otherwise, executing the step (4);

(3) will be provided withWith the pixel point x_i,jTaking the average value of the pixels in the 3 x 3 blocks as the center as the pixel value of the pixel point after preprocessing;

(4) for pixel point x_i,j8 neighborhood pixel point x_l1,2, …,8, and takes pixel x_lCentered 3 x 3 block v_lTo block v_lBy rotating to make pixel point x_i,jCentered 3 x 3 block v_i,jAnd with pixel point x_lCentered 3 x 3 block v_lThe medium homogeneous region is at the same position, v_lThe rotated block is marked as

(5) Calculating a pixel point x_i,jAnd 8 neighborhood pixel points x thereof_l1,2, …,8 similarity distance based on block ratio probability:

5a) taking pixel x_i,jCentered 3 x 3 block v_i,jObtaining the neighborhood pixel point x in the step (4)_l1,2, …,8 inverted 3 × 3 blocks

5b) Calculating the two pixel blocks v_i,jAndratio r of_i,k：

Wherein v is_i,kRepresenting a pixel point x to be processed_i,jCentered 3 x 3 block v_i,jThe gray value v of the kth pixel point_l,kRepresenting a pixel point x_lCentered 3 x 3 block v_lRotated blockThe gray value of the kth pixel point;

5c) calculating pixel point x by using ratio distribution probability formula_i,jRatio r of_i,kProbability of occurrence p (r)_i,k)；

5d) Defining a pixel point x to be processed_i,jAnd 8 neighborhood pixel points x thereof_lThe distance d of similarity therebetween_lComprises the following steps:

(6) for the similarity distance d obtained in the step (5)₁,d₂,…,d₈Sorted in ascending order, the sorted result isGet the distance after the sequence asAs pixel point x_i,jLike the pixel point x_i,jAnd the mean value of the gray values of the 3 similar points is taken as a pixel point x_i,jPreprocessing the gray value;

(7) and (5) repeating the steps (2) to (6), and calculating the recovery values of all the pixel points to obtain the preprocessed image.

2. The SAR image preprocessing method based on similar blocks as claimed in claim 1, wherein said calculating variance coefficient CV of all pixel points in step (1) to obtain variance coefficient matrix K is performed according to the following steps:

1a) pixel point x_i,j，i∈[1,m],j∈[1,n]Coefficient of variance CV of_i,jThe calculation formula of (2) is as follows:

wherein,is a pixel point x_i,jThe standard deviation of the gray values of all the pixel points in the 7 x 7 neighborhood of the center,is a pixel point x_i,jThe mean value of gray values of all pixel points in a 7 multiplied by 7 neighborhood which is taken as a center;

1b) calculating the variance coefficient of each pixel point in the SAR image v to obtain a variance coefficient matrix K₀：

K₀＝{CV_i,j},i∈[1,m],j∈[1,n]；

1c) For the calculated variance coefficient matrix K₀And carrying out 3 × 3 mean filtering to obtain a filtered variance coefficient matrix K.

3. The SAR image preprocessing method based on similar blocks as claimed in claim 1, wherein said setting variance coefficient classification threshold T in step (2)_cvThreshold value T_cvCalculated according to the following formula:

2a) if the input is an L-view amplitude SAR image, the threshold value T_cvCalculated by the following formula:

2b) if the input is the L apparent intensity SAR image, the threshold value T_cvCalculated by the following formula:

4. the SAR image preprocessing method based on similar blocks as claimed in claim 1, wherein said calculating pixel point x using ratio distribution probability formula in step (5c)_iRatio r of_i,kProbability of occurrence p (r)_i,k) The calculation is performed in two cases:

if the input SAR image v is a magnitude image, its probability p (r)_i,k) Calculated by the following ratio distribution probability formula:

if the input SAR image v is an intensity image, its probability p (r)_i,k) Calculated by the following ratio distribution probability formula: