CN102081799B - Method for detecting change of SAR images based on neighborhood similarity and double-window filtering - Google Patents

Abstract

The invention discloses a method for detecting the change of synthetic aperture radar (SAR) images based on neighborhood similarity and double-window filtering, and belongs to the field of change detection of the SAR images. The method comprises the following steps of: firstly, performing geometric correction and registration on two SAR images at different time in the same area; secondly, constructing a difference image graph by neighborhood similarity, and processing the difference image graph by double-window filtering; and finally, performing threshold processing on the filtered difference image graph by a maximum between-class variance (Ostu) threshold method to acquire a change detection result graph. By the change detection method based on the neighborhood similarity and the double-window filtering, the problem that the structure of the difference graph is severely influenced by speckle noise of the SAR images during the detection of the change of the SAR images is solved, so that change detection accuracy is improved.

Description

SAR image change detection method based on neighborhood similarity and the filtering of double window mouth

Technical field

The present invention relates to SAR Image Change Detection field, is a kind of not SAR image difference figure building method of phase simultaneously.This method can be suitable for the precision that multiple complicated SAR image is improved the SAR Image Change Detection with solving the not high problem of accuracy of detection in the SAR Image Change Detection field.

Background technology

The SAR Image Change Detection is exactly from the multi-temporal remote sensing image of the same geographic area that different time obtains, the technology of qualitative or quantitative analysis and definite face of land variation characteristic and process.Because system compares with remote optical sensing, the SAR system has round-the-clock, the round-the-clock ability of obtaining data, so SAR Image Change Detection technology just is being widely used in every field; For example environmental monitoring; Agricultural research, urban area research, aspects such as forest monitoring.

The SAR Image Change Detection is one of main application of remote sensing technology.Through the various images of an identical simultaneously scene are not compared analysis, obtain the change information of needed atural object of people or target according to the difference between the image.Change detection techniques can detect the variation between different times gradation of image value or the local grain, and the target that needs of acquisition is in shape, position, quantity, and the situation of change of other attribute on this basis.

Yet the SAR image is when having above-mentioned advantage, and the speckle noise influence is a main bottleneck of its application, and how from SAR image own characteristic, effectively suppressing speckle noise is a problem must considering in the practical application of SAR image.Along with the continuous development of computing machine and imaging technique, SAR Image Change Detection technology also in improving constantly and improving, several kinds of change detection algorithm comparatively commonly used below having formed gradually in recent years:

The image difference method: its main process is that the gray-scale value with 2 o'clock phase SAR image corresponding pixel points subtracts each other and obtains disparity map, chooses 0～255 threshold value then and cuts apart, and obtains changing and non-region of variation.Image difference method algorithm is simple, and is convenient and easy, but the shortcoming of this method is also apparent in view: influenced by the not equal objective condition of SAR image quality, wave spectrum characteristic, be prone to produce " the pseudo-variation " information.

The image ratioing technigue: its main process is that the ratio that calculates 2 o'clock phase SAR image corresponding pixel points gray-scale values obtains disparity map, if a pixel does not change, then ratio should be near with 1, otherwise, much larger than or much smaller than 1.This method is insensitive to the multiplicative noise of SAR image, but that the disparity map degree of accuracy that obtains tends to receive change type to influence is bigger.

Image average ratioing technigue: its main process be 2 o'clock phase SAR image corresponding pixel points of utilization with and the ratio of the average of neighborhood territory pixel point obtain disparity map; Similar and the ratioing technigue of the method; Multiplicative noise to the SAR image is insensitive; But also well eliminated speckle noise, loss is reduced greatly.

Above three kinds of methods are change detecting methods comparatively commonly used, also have change vector analytic approach, principal component analysis method or the like in addition.Noise effect is bigger but these methods all receive the SAR Image Speckle; Must carry out filtering and noise reduction to 2 o'clock phase SAR images in advance; So just might change raw image data; Change among the differentia influence figure of these method constructs in addition type all obvious inadequately with non-variations type difference, make cut apart after accuracy of detection not high.

Summary of the invention

The objective of the invention is to: in order to overcome the faults rate deficiency of existing method; A kind of differentia influence figure building method based on neighborhood similarity has been proposed; And, remedied the not high defective of some other difference image figure building method accuracy of detection to the difference image figure filtering method that own characteristic designs.

Technical scheme of the present invention is: based on the principle of neighborhood similarity; In conjunction with the advantage of ratio approach and average ratio approach; A kind of method of structural differences striograph has been proposed; The difference image figure DI of phase SAR image during according to neighborhood similarity method construct two is again through obtaining change-detection figure as a result to Ostu threshold value after the double window mouth Filtering Processing of DI.Its concrete performing step is following:

(1) two width of cloth SAR image A and the B to the same area different time carries out geometry correction and registration;

(2) according to the neighborhood similarity method, according to the difference image figure DI of following steps structure image A and B;

2a) obtain 2 o'clock phase SAR image A and the B grey scale pixel value neighborhood set N on same position i respectively ₁(i) and N ₂(i), its size is N * N, N ∈ 3,5,7,9};

2b) relatively two neighborhoods are gathered N ₁(i) and N ₂(i) similarity obtains the pixel similarity value DI (i) of difference image figure DI on the i of position:

$\mathrm{DI}\left(i\right)=\frac{\min (N_1\left(i\right),N_2\left(i\right))}{\max (N_1\left(i\right),N_2\left(i\right))}+\frac{\underset{j=1,j\≠i}{\overset{N\×N}{\mathrm{\Σ}}}\min (N_1\left(j\right),N_2\left(j\right))}{\underset{j=1,j\≠i}{\overset{N\×N}{\mathrm{\Σ}}}\max (N_1\left(j\right),N_2\left(j\right))}$

Wherein, N ₁(j) and N ₂(j) difference presentation video A and the neighborhood pixel of B on the i of position, set N ₁(i) and N ₂(i) similar more, then DI (i) value is big more, and promptly the value of DI (i) approaches 2 more, and the possibility that image A and the B pixel on the i of position belongs to non-region of variation is big more, otherwise the possibility that belongs to region of variation is big more;

2c) on duty with 128 to the pixel similarity, obtain i order grey scale pixel value.

2b repeating step 2a from top to bottom)) and 2c), obtain difference image figure DI 2d) from left to right, to each position i of image A and B;

(3) difference image figure DI is carried out double window mouth Filtering Processing, obtains new difference image figure FDI,

The step of double window mouth filtering is following:

The grey scale pixel value neighborhood of 3a) getting on the difference image figure position x is gathered I _i(x) and I (x), its size is respectively n * n, N * N, n ∈ 3,5,7}, N ∈ 11,13 ..., 27};

3b) try to achieve the set I at the central pixel point i place in set I (x) _i(x) and the set I at pixel j place _jSimilarity d (x) (i, j):

$d(i,j)=\frac{{|\frac{1}{n\×n}\underset{k=1}{\overset{n\×n}{\mathrm{\Σ}}}{I}_i\left(k\right)-\frac{1}{n\×n}\underset{k=1}{\overset{n\×n}{\mathrm{\Σ}}}{I}_j\left(k\right)|}^2}{L}$

The gray level of L presentation video, the number of pixel during k representes to gather;

3c) the similarity factor is carried out normalization, the normalized factor Z that j is ordered (i j) is:

$Z(i,j)=\frac{\exp (-d(i,j)/h^2)}{\underset{j=1}{\overset{N\×N}{\mathrm{\Σ}}}\exp (-d(i,j)/h^2)}$

H is the exponential damping controlling elements, the degree of decision filtering;

3d) obtain the value FDI (i) of new difference image figure FDI on the i of position:

$\mathrm{FDI}\left(i\right)=\underset{j=1}{\overset{N\×N}{\mathrm{\Σ}}}\{Z(i,j)\·[\frac{1}{n\×n}\underset{k=1}{\overset{n\×n}{\mathrm{\Σ}}}{I}_j\left(k\right)\left]\right\}$

3c 3b repeating step 3a from top to bottom))) and 3d), obtain new difference image figure FDI 3e) from left to right, to difference image figure DI;

(4) utilize the Ostu threshold method that new difference image figure FDI is carried out threshold process, obtain change-detection figure as a result.

The present invention compared with prior art has following characteristics:

1, the present invention adopts the difference image figure of neighborhood similarity method construct, has suppressed the part speckle noise of SAR image, need well not keep the information of original image to the denoising of original SAR image filtering;

2, the present invention is directed to the difference image figure of neighborhood similarity method construct; In conjunction with double window mouth filtering method; Further weakened the influence of noise among the difference image figure, increased simultaneously variation class center and the distance at non-variation class center among this difference image figure effectively, and the histogram distribution of difference image figure has been adjusted into bimodal distribution; Help the cutting apart of difference image figure, improved and changed the precision that detects;

3, simulation result shows, neighborhood similarity method construct difference image drawing method that the present invention adopts and double window mouth filtering method are than the ratio method, and the accuracy of detection of average ratio approach structural differences striograph method and Lee filter processing method is higher.

Description of drawings

Fig. 1 is the process flow diagram of the inventive method.

Fig. 2 is Ottawa two width of cloth SAR striograph and the standard drawings that the present invention relates to, and wherein (a) expression in May, 1997 this area's geomorphology information (b) is represented in August, 1997 this area's geomorphology information, (c) the standard results figure of expression change-detection.

Fig. 3 is the inventive method to disparity map and the filtered disparity map of double window mouth of Ottawa structure, and wherein (a) representes the disparity map that the neighborhood similarity method obtains, and (b) disparity map is carried out resulting new disparity map after the filtering of double window mouth.

Fig. 4 is the histogram of disparity map before and after the Ottawa image filtering among the present invention, and the histogram of the disparity map that obtains of (a) expression neighborhood similarity method wherein (b) carries out the histogram of resulting new disparity map after the filtering of double window mouth to disparity map.

Fig. 5 is the comparative test result figure of the inventive method and two species diversity figure building methods, and wherein (a) representes the figure as a result of average ratio approach, (b) figure as a result of expression ratio approach, and (c) expression is based on neighborhood similarity methods and results figure.

Fig. 6 is territory, MEXICO CITY suburb two width of cloth SAR striograph and the standard drawings that the present invention relates to, and wherein (a) expression in May, 2002 this area's geomorphology information (b) is represented in April, 2005 this area's geomorphology information, (c) the standard results figure of expression change-detection.

Fig. 7 is disparity map and the double window mouth filtered disparity map of the inventive method to MEXICO CITY suburb domain construction, and the disparity map that obtains of (a) expression neighborhood similarity method wherein (b) is carried out resulting new disparity map after the filtering of double window mouth to disparity map.

Fig. 8 is the histogram of disparity map before and after the filtering of Mexico's suburban area area image among the present invention, and the histogram of the disparity map that obtains of (a) expression neighborhood similarity method wherein (b) carries out the histogram of resulting new disparity map after the filtering of double window mouth to disparity map.

Fig. 9 is to the filtering of double window mouth, the Lee filtering of disparity map and does not have the figure as a result of filtering; Wherein (a) expression does not add the figure as a result of Filtering Processing to disparity map; (b) expression is carried out the figure as a result after the Lee Filtering Processing to disparity map, and (c) figure as a result of double window mouth Filtering Processing is carried out in expression to difference image figure.

Embodiment

Like accompanying drawing 1, the present invention includes following steps:

Step 1: two width of cloth SAR image A and B to the same area different time carry out geometry correction and registration;

Step 2: according to the neighborhood similarity method, according to the difference image figure DI of following steps structure image A and B;

2a) obtain 2 o'clock phase SAR image A and the B grey scale pixel value neighborhood set N on same position i respectively ₁(i) and N ₂(i), its size is N * N, N ∈ 3,5,7,9};

2b) relatively two neighborhoods are gathered N ₁(i) and N ₂(i) similarity obtains the grey scale pixel value DI (i) of difference image figure DI on the i of position:

$\mathrm{DI}\left(i\right)=\frac{\min (N_1\left(i\right),N_2\left(i\right))}{\max (N_1\left(i\right),N_2\left(i\right))}+\frac{\underset{j=1,j\≠i}{\overset{N\×N}{\mathrm{\Σ}}}\min (N_1\left(j\right),N_2\left(j\right))}{\underset{j=1,j\≠i}{\overset{N\×N}{\mathrm{\Σ}}}\max (N_1\left(j\right),N_2\left(j\right))}$

Wherein, N ₁(j) and N ₂(j) difference presentation video A and the neighborhood pixel of B on the i of position, set N ₁(i) and N ₂(i) similar more, then DI (i) value is big more, and promptly the value of DI (i) approaches 2 more, and the possibility that image A and the B pixel on the i of position belongs to non-region of variation is big more, otherwise the possibility that belongs to region of variation is big more;

2c) on duty with 128 to the pixel similarity, obtain i order grey scale pixel value.

2b repeating step 2a from top to bottom)) and 2c), obtain difference image figure DI 2d) from left to right, to each position i of image A and B;

Step 3: difference image figure DI is carried out double window mouth Filtering Processing, obtain new difference image figure FDI,

The step of double window mouth filtering is following:

The grey scale pixel value neighborhood of 3a) getting on the difference image figure position x is gathered I _i(x) and I (x), its size is respectively n * n, N * N, n ∈ 3,5,7}, N ∈ 11,13 ..., 27};

3b) try to achieve the set I at the central pixel point i place in set I (x) _i(x) and the set I at pixel j place _jSimilarity d (x) (i, j):

$d(i,j)=\frac{{|\frac{1}{n\×n}\underset{k=1}{\overset{n\×n}{\mathrm{\Σ}}}{I}_i\left(k\right)-\frac{1}{n\×n}\underset{k=1}{\overset{n\×n}{\mathrm{\Σ}}}{I}_j\left(k\right)|}^2}{L}$

The gray level of L presentation video.

3c) the similarity factor is carried out normalization, the normalized factor Z that j is ordered (i j) is:

$Z(i,j)=\frac{\exp (-d(i,j)/h^2)}{\underset{j=1}{\overset{N\×N}{\mathrm{\Σ}}}\exp (-d(i,j)/h^2)}$

H is the exponential damping controlling elements, the degree of decision filtering.

3d) obtain the value FDI (i) of new difference image figure FDI on the i of position:

$\mathrm{FDI}\left(i\right)=\underset{j=1}{\overset{N\×N}{\mathrm{\Σ}}}\{Z(i,j)\·[\frac{1}{n\×n}\underset{k=0}{\overset{n\×n}{\mathrm{\Σ}}}{I}_{}\left(k\right)\left]\right\}$

3e) from left to right, repeating step 3a from top to bottom) to difference image figure DI, 3b), 3c), 3d), obtain new difference image figure FDI;

Step 4: utilize the Ostu threshold method that new difference image figure FDI is carried out threshold process, obtain change-detection figure as a result:

The gray-scale value of (4a) establishing new difference image figure FDI is 0 to the m-1 level, and the pixel of gray-scale value i is n _i, the sum of all pixels that obtain this moment does

Each is worth probability

Be divided into two groups of C with threshold value T then ₀={ 0～T-1} and C ₁={ T～m-1};

(4b) probability of each group generation is following:

C ₀Produce probability

C ₁Produce probability

C ₀Mean value

C ₁Mean value

Wherein,

Be the average gray of new difference image figure FDI,

Be threshold value average gray when being T, so all the average gray of sampling is μ=w ₀μ ₀+ w ₁μ ₁

(4c) variance between two groups is obtained with following formula:

${\mathrm{\δ}}^2\left(T\right)=w_0{({\mathrm{\μ}}_0-\mathrm{\μ})}^2+w_1{({\mathrm{\μ}}_1-\mathrm{\μ})}^2=\frac{{[\mathrm{\μ}\·w\left(T\right)-\mathrm{\μ}\left(T\right)]}^2}{w\left(T\right)[1-w\left(T\right)]}$

Change T between 1～m-1, work as δ ²T value when (T) being maximal value is threshold value, and new difference image figure FDI is cut apart according to threshold value T, is made as pixel 0 less than T, is made as pixel 255 greater than T, obtains change-detection figure as a result.

Effect of the present invention can further specify through following emulation:

1, simulation parameter

Two groups of experiment simulation figure for having reference diagram, carry out quantitative change-detection interpretation of result:

1. calculate the omission number: change among the figure as a result number of pixels in zone of statistical experiment; Compare with the number of pixels of region of variation in the reference diagram; Detect changing in the reference diagram but in the experimental result picture and be unchanged number of pixels, be called omission number FN;

2. calculate the false retrieval number: do not change among the figure as a result number of pixels in zone of statistical experiment; With in the reference diagram not the number of pixels of region of variation compare; Detect number of pixels not changing in the reference diagram but in the experimental result picture, be called false retrieval number FP for changing;

3. total faults count OE equal omission number and false retrieval number and.

2, emulation content

1. with the difference image figure of neighborhood similarity method construct, and ratio approach, difference image figure gained binary map after the filtering of double window mouth of average ratio approach structure compares, and qualitative, quantitative analyzes experimental result;

2. the histogram data before and after the double window mouth filtering of difference image figure is compared analysis;

3. to the difference image figure of neighborhood similarity method construct, the effect after no filtering algorithm, Lee filtering algorithm and double window mouth filtering algorithm are handled compares analysis.

3, The simulation experiment result and analysis

1. to the histogram data analysis before and after the double window mouth filtering of difference image figure

Fig. 4 representes the histogram before and after the double window mouth filtering of Ottawa difference image figure, and Fig. 8 representes the histogram before and after the double window mouth filtering of difference image figure in territory, MEXICO CITY suburb.

With the difference image figure of neighborhood similarity method construct Ottawa emulated data such as difference image figure such as Fig. 8 of Fig. 3 and territory, MEXICO CITY suburb emulated data; Wherein Fig. 3 (a) and Fig. 8 (a) are the difference image figure before the double window mouth Filtering Processing; Fig. 3 (b) and Fig. 8 (b) are the new difference image figure after the double window mouth Filtering Processing, from Fig. 3 (a) and (b) and Fig. 8 (a) and contrast (b) can find out that the histogram distribution of filtered difference image figure is more obvious; And centre distance has increased; The assorted point of testing result after the difference image figure process double window mouth Filtering Processing effectively reduces, and the edge is more level and smooth, helps the Threshold Segmentation in later stage.

2. with the Ottawa region S AR view data of reference diagram

Use ratioing technigue, average ratioing technigue and neighborhood similarity method divergence figure building method on true Ottawa region S AR view data.The experiment test former figure mutually and reference diagram as shown in Figure 2.Wherein (a) expression in May, 1997 this area's geomorphology information (b) expression in August, 1997 this area's geomorphology information (c) is represented the standard results figure of change-detection.

The change-detection result that ratioing technigue, average ratioing technigue and neighborhood similarity method construct disparity map obtain is as shown in Figure 5, and the comparative result of accuracy of detection is listed in the table 1.Wherein figure as a result (c) expression of figure as a result (b) the expression ratio approach of (a) expression average ratio approach is based on neighborhood similarity methods and results figure

The change-detection interpretation of result of the different difference image figure building methods of table 1

Can find out the figure as a result that from change-detection this method is apparent in view for the treatment effect of assorted point.This has not only considered the influence of neighborhood information to picture dot just because of this method, has been also noted that the influence of pixel can not be equal to the center pixel in the neighborhood to treat.Treat because average ratio is equal to neighborhood pixel and center pixel, can make the pixel that does not change be mistaken as the variation pixel like this; And ratioing technigue direct object unit makes ratio, has ignored neighborhood information, and disparity map receives the influence of speckle noise can make the variation pixel by omission.

Can find out that from the listed accuracy of detection statistics of table 1 in the experiment to the Ottawa data, the flase drop number of average ratio is the highest; And the omission number of ratioing technigue is more; This method is got its chief, makes total flase drop number minimum, makes the precision of change-detection more accurate.

3. with the regional SAR view data in the true Mexico suburbs of reference diagram

The difference of territory, MEXICO CITY suburb SAR image configuration is compared in Lee filtering, the filtering of double window mouth and the figure as a result that do not have filtering.Experiment original image and reference diagram are as shown in Figure 6.Wherein (a) expression in May, 2002 this area's geomorphology information (b) expression in April, 2005 this area's geomorphology information (c) is represented the standard results figure of change-detection.

To such as table 2, experimental result picture is as shown in Figure 9 to the change-detection precision of the Lee filtering of difference image figure, the filtering of double window mouth, no filtering.Wherein (a) expression figure as a result (b) expression of disparity map not being added Filtering Processing is carried out double window mouth Filtering Processing is carried out in figure as a result (c) expression after the Lee Filtering Processing to difference image figure figure as a result to disparity map.

Territory, table 2 MEXICO CITY suburb change-detection performance comparison result

Can find out that from difference image figure and segmentation result speckle noise can be obviously removed in filtering, level and smooth homogenous area is obviously distinguished region of variation and non-region of variation.

Table 2 be double window mouth filtering method, Lee filtering method and the testing result that does not add filtering method relatively, from table, can see, can make our testing result approach actual change more for the subsequent treatment of difference image figure; We can find out that also double window mouth filtering method is relatively good for the treatment effect of difference image figure with the comparison of Lee filtering.

Claims (1)

1. based on the SAR image change detection method of neighborhood similarity and the filtering of double window mouth, it is characterized in that following steps:

(1) two width of cloth SAR image A and the B to the same area different time carries out geometry correction and registration;

(2) according to the neighborhood similarity method, the difference image figure DI of structure image A and B, realize as follows:

(2a) obtain 2 o'clock phase SAR image A and the B grey scale pixel value neighborhood set N on same position i respectively ₁(i) and N ₂(i), its size is N * N, N ∈ 3,5,7,9};

(2b) relatively two neighborhoods are gathered N ₁(i) and N ₂(i) similarity obtains the pixel similarity value DI (i) of difference image figure DI on the i of position:

Wherein, N ₁(j) and N ₂(j) difference presentation video A and the neighborhood pixel of B on the i of position, set N ₁(i) and N ₂(i) similar more, then DI (i) value is big more, and promptly the value of DI (i) approaches 2 more, and the possibility that image A and the B pixel on the i of position belongs to non-region of variation is big more, otherwise the possibility that belongs to region of variation is big more;

(2c) pixel similarity value DI (i) multiply by 128, obtain the grey scale pixel value that i is ordered;

(2d) to each position i of image A and B from left to right, repeating step (2a), (2b) and (2c) obtain difference image figure DI from top to bottom;

(3) difference image figure DI is carried out double window mouth Filtering Processing, obtains new difference image figure FDI, realize as follows:

The grey scale pixel value neighborhood of (3a) getting on the difference image figure position x is gathered I _i(x) and I (x), its size is respectively n * n, N * N, n ∈ 3,5,7}, N ∈ 11,13 ..., 27};

(3b) try to achieve the set I at the central pixel point i place in set I (x) _i(x) and the set I at pixel j place _jSimilarity d (x) (i, j):

The gray level of L presentation video, the number of pixel during k representes to gather;

(3c) the similarity factor is carried out normalization, the normalized factor Z that j is ordered (i j) is:

H is the exponential damping controlling elements, the degree of decision filtering;

(3d) obtain the value FDI (i) of new difference image figure FDI on the i of position:

3c repeating step 3b from top to bottom)) and 3d), obtain new difference image figure FDI (3e) from left to right, to difference image figure DI;

(4) utilize the Ostu threshold method that new difference image figure FDI is carried out threshold process, obtain change-detection figure as a result.

Images