CN101515366B - Watershed SAR image segmentation method based on complex wavelet extraction mark - Google Patents

Abstract

The invention discloses a watershed SAR image segmentation method based on complex wavelet extraction mark, which is used for solving the problem of over-segmentation of the watershed SAR image. The invention integrates the advantages of a feature extraction method designated to texture information and of a watershed method on the segmentation of the SAR image, and can excellently inhibit the over-segmentation. The method comprises the specific implementing steps of: (1) obtaining a mark for watershed transform in a manner of complex wavelet energy feature extraction; (2) implementing Gaussian low-pass filter on original images; (3) using Priwitt operator for the filtered images to calculate gradient; (4) correcting the gradient by using inner and outer marks and utilizing a mandatory minimum technology in order that a local minimum area appears at a marked location only; and (5) subjecting the corrected gradient map to the watershed transform to obtain an initial segmentation image of the SAR image. A set of experiments prove that the inventive image segmentation effect accords with the standards basically. The invention not only reduces the over-segmentation, but also guarantees the accuracy of edges.

Description

Watershed SAR image segmentation method based on complex wavelet extraction mark

Technical field

The invention belongs to technical field of image processing, be specifically related to a kind of Watershed SAR image segmentation method based on complex wavelet extraction mark.Be used in synthetic-aperture radar (synthetic aperture radar, the SAR) over-segmentation of inhibition SAR image segmentation in the image segmentation field.

Background technology

Synthetic-aperture radar SAR image segmentation is a committed step that realizes that the SAR image is handled automatically, its objective is the subregion or the object that the SAR image segmentation are become to have strong correlation.Be convenient to further to the SAR image analyze, discern, etc., the accuracy of cutting apart directly influences the validity of follow-up work.

The object of observation of SAR image is atural object and the target of highly not having under the constraint scene, and radar return is responsive to reactions such as type of ground objects, orientation, degree of uniformity, spatial relationships, is reflected as texture information on image greatly.In the process of SAR image segmentation, texture information is considered to distinguish the important decipher information of type of ground objects, and the object of observation that out of Memory is difficult to differentiate can be distinguished easily with texture probably.As far back as 1981, the Shanmugan of the U.S. has just recognized the importance that texture is understood for radar image, and after two more than ten years in a large amount of researchists this is studied, existing now many documents show introduces the precision that textural characteristics helps to improve the SAR image segmentation.At present, what the analytical approach of SAR image texture information was mainly adopted is the statistics texture method, methods such as the textural characteristics that mainly comprises autocorrelation function, Fourier's power spectrum method, changes based on small echo, gray level co-occurrence matrixes, probabilistic model.But this class kind method is not accurate enough to the location, edge.Address this problem the method that to introduce the edge accurate positioning.

The morphology watershed algorithm is the extracted region images partitioning algorithm of a kind of edge accurate positioning, because the algorithm implementation procedure is simply effective, has now become widely used image segmentation instrument.Watershed algorithm made full use of the marginal information that the gradient operation is caught, the edge accurate positioning, and resulting zone boundary formed the set of a sealing, connection, has Space Consistency.Yet the gradient operation in the watershed algorithm is to texture and noise-sensitive, and being used for the SAR image will produce serious over-segmentation because of its abundant texture information.Therefore at the characteristics of SAR image, the combination that explores multiple dividing method has been a kind of trend that present image is handled to the SAR image segmentation.

Summary of the invention

The technical problem to be solved in the present invention is: the deficiency that overcomes the conventional images cutting techniques, promptly the segmentation result edge that adopts multiple wavelet character to extract cluster separately is inaccurate, adopt the over-segmentation problem of classical watershed transform or mark watershed transform separately, a kind of Watershed SAR image segmentation method based on complex wavelet extraction mark is proposed, with the mark source of the image after the multiple wavelet character extraction cluster as the watershed divide, combine specially and the SAR image is cut apart at the feature extracting method of texture information and the method for watershed divide, the advantage of two comprehensive methods, eliminate the influence of texture information, effectively solve the over-segmentation problem of Watershed SAR image segmentation.

For achieving the above object, according to SAR image partition method provided by the invention, at first extract the needed mark of watershed segmentation method with multiple small echo, simultaneously in order to have eliminated the influence of SAR picture noise and texture information to a certain extent, carry out the gradient map of trying to achieve again behind the gaussian filtering with such mark correction original image, carry out the computing of classical watershed divide at this gradient map, then obtain the result of the fine solution over-segmentation of final energy problem.

The realization of concrete technical scheme comprises the steps:

(1) uses the inner marker LImg of the mode of multiple wavelet energy feature from original image Img extraction mark watershed transform;

(2) inner marker LImg is carried out classical watershed transform and get external label WLImg;

(3) original image Img is carried out Gauss's low-pass filtering and get image GImg after filtering;

(4) filtered image GImg is asked figure PGImg with the Priwitt operator;

(5) with inside and outside mark gradient map PGImg is carried out the gradient correction, the concrete utilization forced the minimum technology correction, makes the local minimum area of gradient map appear at mark position;

(6) revised gradient map is carried out watershed transform, obtain final image segmentation figure RImg.

The step that the mode of the multiple wavelet energy feature of above-mentioned steps (1) usefulness obtains inner marker LImg is as follows:

[1] to the multiple wavelet energy feature of each pixel extraction of original image;

[2] each pixel is carried out the K-mean cluster;

[3] result images after the cluster is carried out Gauss's low-pass filtering;

[4] filtered image is asked gradient with the Priwitt operator;

[5] gradient map is extracted inner marker, the process of choosing inner marker is exactly to find the process of local minimum, and local minimum is meant the continuum of gray-scale value in a tonal range, and near the value of the pixel this zone is all greater than the value in this zone.

Multiple small echo in the multiple wavelet energy Feature Extraction Technology in the above-mentioned step [1] is a kind of TIME SHIFT INVARIANCE that has, the small echo of characteristics such as more than the many and phase information of directional information.To complex field, the small echo that constructs not only has traditional wavelet time-frequency localization feature, and has good directivity with the structure spatial spread of small echo for it.The algorithm of the multiple wavelet energy feature among the present invention is as follows: it is two-layer that the video in window of each pixel is carried out multiple wavelet decomposition, obtain a low frequency sub-band and every layer of 6 high-frequency sub-band, totally 13 subbands ask wavelet energy as eigenvector with following formula respectively to each subband, and the normalization eigenvector.

$e=\frac{1}{M\·N}\underset{x=1}{\overset{M}{\mathrm{\Σ}}}\underset{y=1}{\overset{N}{\mathrm{\Σ}}}\left|s(x,y)\right|$

Wherein, (x y) is coefficient of dissociation to s, and M * N is the size of band image, and x and y be the row and column of presentation video respectively, and e is the energy feature vector.

Above-mentioned steps (2) and the used Gauss's low-pass filtering technique of step [3] are a kind of low frequency enhancement techniques, and it can strengthen some frequecy characteristic of image, change the gray scale contrast between ground object target and neighborhood or the background.When analysis image signal frequency-domain characteristic, the high fdrequency component of the edge of piece image, jump part and grain noise representative image signal, large-area background area is the low frequency component of representative image signal then.If suppress high-frequency information,, keep and outstanding even subject image in flakes then with level and smooth image detail by low-pass filtering enhancing low-frequency information.The present invention removes the details or the noise section of image with it.

Above-mentioned steps (3) and step [4] image is asked in the gradient method, said gradient is meant the place of the marked change of gradation of image value.Can regard image as the two-dimensional discrete function from the angle of mathematics, image gradient is exactly the differentiate of two-dimensional discrete function, different operator correspondences the different methods of asking gradient.The present invention has adopted the Priwitt operator, and it is to define it to differentiate in an odd number template.

The applied watershed divide of the present invention image partition method derives from topography, the watershed divide image segmentation algorithm is regarded image as natural feature awash on the topography, the gray-scale value of each pixel in the image is represented the sea level elevation of this point, its each local minimum and range of influence thereof are called the catchment basin, and the border of catchment basin then is the watershed divide.Traditional dividing ridge method extracts the local minimum of gradient image just on gradient image, the intra-zone that Grad is less is regarded a catchment basin as, and the object boundary that Grad is bigger is as the watershed divide line; Carry out region growing according to Grad then, when water logging there was not the basin, the pixel of the low gradient that the watershed divide line is following joined together gradually, when water arrives the gradient local maximum point, erects the watershed divide line, and two adjacent catchment basins are separated.Like this, the watershed divide line just becomes image segmentation several catchment basin, thereby obtains different target areas.The watershed divide image segmentation algorithm is simply quick, and segmentation area sealing and consistance are strong, and the edge is accurate.

The present invention has the following advantages compared with prior art:

1, the present invention adopted the watershed divide after a kind of improve method to the SAR image segmentation, the zone that can be sealed and be communicated with.

2. the present invention has adopted multiple small echo marker extraction, has effectively eliminated the over-segmentation that texture information produces.

3, the present invention's image of method after to Threshold Segmentation of having adopted Gauss's low-pass filtering carried out filtering, eliminated the over-segmentation that radio-frequency component brings.

4, gradient correction of the present invention is that filtered gradient image at original image carries out, and is not the filtering image after the Threshold Segmentation, has so both reduced over-segmentation, has guaranteed the accuracy at edge again.

Description of drawings

Fig. 1 is the Watershed SAR image segmentation method process flow diagram that the present invention is based on complex wavelet extraction mark

Fig. 2 is the process flow diagram that the multiple wavelet algorithm of the present invention extracts mark

Fig. 3 is to the experiment and the contrast and experiment figure of SAR image I among the present invention.Wherein (a) is the SAR original image; (b) result of multiple wavelet character extraction+K-mean cluster; (c) algorithm of the present invention; (d) edge that obtains of algorithm of the present invention (b) result on the figure as a result that is added to

Fig. 4 is to experiment and the contrast and experiment figure of SAR image I I among the present invention.Wherein (a) is the SAR original image; (b) result of multiple wavelet character extraction+K-mean cluster; (c) algorithm of the present invention; (d) edge that obtains of algorithm of the present invention (b) result on the figure as a result that is added to

Fig. 5 is to experiment and the contrast and experiment figure of SAR image I II among the present invention.Wherein (a) is the SAR original image; (b) result of multiple wavelet character extraction+K-mean cluster; (c) algorithm of the present invention; (d) edge that obtains of algorithm of the present invention (b) result on the figure as a result that is added to

Embodiment

With reference to above-mentioned accompanying drawing, the present invention is described in detail.

As depicted in figs. 1 and 2, the present invention's method of carrying out image segmentation comprises the steps:

1, extracts the inner marker LImg of mark watershed transform from original image Img with the mode of multiple wavelet energy feature.

(1) to the multiple wavelet energy feature of each pixel extraction of original image;

Multiple small echo has TIME SHIFT INVARIANCE, characteristics such as the many and phase information of directivity information.To complex field, the small echo that constructs not only has traditional wavelet time-frequency localization feature, and has good directivity with the structure spatial spread of small echo for it.With this pixel is the center, gets a square wicket, and this video in window is carried out two-layer multiple wavelet decomposition, obtains every layer of 6 high-frequency sub-band and a low frequency sub-band, and totally 13 subbands ask wavelet energy as eigenvector with following formula respectively to them.And normalization eigenvector.

$e=\frac{1}{M\×N}\underset{x=1}{\overset{M}{\mathrm{\Σ}}}\underset{y=1}{\overset{N}{\mathrm{\Σ}}}\left|s(x,y)\right|$

Wherein, (x y) is coefficient of dissociation to s, and M * N is the size of band image, and x and y be the row and column of presentation video respectively, and e is the energy feature vector.With the energy feature of this video in window energy feature as this pixel.

(2) K-means clustering algorithm

The K-means clustering algorithm belongs to a kind of basic division methods in the clustering technique, has simply, advantage fast.Its basic thought is to choose k data object as initial cluster center, by iteration data object is divided in different bunches, makes the similarity between bunch internal object very big, and bunch between the similarity of object very little.The value of parameter k is in advance given in the algorithm, and concentrates picked at random k data object as initial cluster center at data object.The step of K-mean algorithm is as follows:

A) from n data object, select k object as initial cluster center arbitrarily;

B) (3) to (4) are carried out in circulation, till each cluster no longer changes;

C), calculate the distance of each object and these center object in the sample set, and again corresponding object is divided according to minor increment according to the average (center object) of all objects in each cluster;

D) recomputate the average (center object) of each (changing) cluster.

(3) cluster result figure is carried out Gauss's low-pass filtering.

Use some frequecy characteristic that Gauss's low-pass filtering strengthens image, to change the gray scale contrast between ground object target and neighborhood or the background.When analysis image signal frequency-domain characteristic, the high fdrequency component of the edge of piece image, jump part and grain noise representative image signal, large-area background area is the low frequency component of representative image signal then.The form of Gauss's low-pass filter function on frequency domain is:

$H(u,v)=e^{-D^{2(u,v)/{2\mathrm{\σ}}^2}}$

$D(u,v)={(u^2+v^2)}^{\frac{1}{2}}$

(u, the initial point that v) is frequency field is to (u, distance v) for D.σ is used for weighing the range of Gaussian curve, order

σ＝D ₀

$H(u,v)=e^{{-D}^{2(u,v)/2{D_0}^2}}$

D ₀It is cutoff frequency.When D (u, v)=D ₀, wave filter drops to peaked 0.607.

Why will carry out Gauss's low-pass filtering operation to image, be because suppress high-frequency information by low-pass filtering enhancing low-frequency information, then with level and smooth image detail, keeps and outstanding even subject image in flakes.Gauss's low-pass filtering treatment image can reduce The noise, reduces over-segmentation to a certain extent.

(4) filtered image is asked gradient with the Priwitt operator.

Gradient is meant the place of the marked change of gradation of image value.Can regard image as the two-dimensional discrete function from the angle of mathematics, the image gradient different method of asking gradient that has been exactly the different operator correspondence of the differentiate of this two-dimensional discrete function in fact.The Priwitt operator is to define it to differentiate in an odd number template.The Priwitt differentiating operator is defined as follows:

D _X＝[f(x+1，y-1)-f(x-1，y-1)]+[f(x+1，y)-f(x-1，y)]+[f(x+1，y+1)-f(x-1，y+1)]

D _Y＝[f(x+1，y-1)-f(x-1，y-1)]+[f(x，y+1)-f(x，y-1)]+[f(x+1，y+1)-f(x+1，y-1)]

$\▿f(x,y)=\sqrt{{D_X}^2+{D_Y}^2}$

The gradient of being tried to achieve exactly.

(5) gradient map is extracted inner marker, the process of choosing inner marker is exactly to find the process of local minimum, and local minimum is meant the continuum of gray-scale value in a tonal range, and near the value of the pixel this zone is all greater than the value in this zone.

2, external label is extracted: inner marker LImg is carried out classical watershed transform get external label WLImg; Inner marker and external label are formed the mark of mark watershed divide.

3, original image is done the filtering of Gauss's low pass ripple.

4, filtered image is asked gradient with the Priwitt operator.

5, gradient correction: the gradient to the filtering image after step 3 operation is revised, and specifically is with forcing minimum technology to the gradient correction, so that local minimum area only appears at mark position, mark is to extract with multiple wavelet character.

6, revised gradient map is carried out classical watershed transform, the SAR image segmentation result figure that gained is final.

The watershed divide image partition method derives from topography, the watershed divide image segmentation algorithm is regarded image as natural feature awash on the topography, the gray-scale value of each pixel in the image is represented the sea level elevation of this point, its each local minimum and range of influence thereof are called the catchment basin, and the border of catchment basin then is the watershed divide.Traditional dividing ridge method extracts the local minimum of gradient image just on gradient image, the intra-zone that Grad is less is regarded a catchment basin as, and the object boundary that Grad is bigger is as the watershed divide line; Carry out region growing according to Grad then, when water logging there was not the basin, the pixel of the low gradient that the watershed divide line is following joined together gradually, when water arrives the gradient local maximum point, erects the watershed divide line, and two adjacent catchment basins are separated.Like this, the watershed divide line just becomes image segmentation several catchment basin, thereby obtains different target areas.The watershed divide image segmentation algorithm is simply quick, and segmentation area sealing and consistance are strong, and the edge is accurate.

Analysis of simulation result

Be the validity of checking the inventive method, carry out that this group experiment is as next group experiment.Because this algorithm computation amount is bigger, adopts 256 * 256 image in the experiment.In this experiment 4 width of cloth SAR images are extracted multiple wavelet energy feature (13), used 256 * 256 image here.The window of being got is 20*20's.With the result of multiple wavelet energy feature extraction and K-mean cluster experimental result as a comparison, and the final edge that the inventive method obtains is added to verifies the accuracy of the inventive method on the feature extraction cluster result the location, edge.

Experimental result from Fig. 3 (c), 4 (c), 5 (c), can find out the inventive method of the half-tone information, marginal information and the texture information that have made full use of the SAR image, obtain reflecting SAR image objectives and consistance cut zone preferably, and object boundary comparatively accurate, continuous, a pixel size, well solved the over-segmentation problem of watershed divide.

Experimental result from Fig. 3 (b), Fig. 4 (b), Fig. 5 (b), multiple wavelet character extraction cluster segmentation method can be told several zones roughly, the multiple wavelet character that has been added to from the edge with the inventive method of Fig. 4 (d), Fig. 5 (d), Fig. 6 (d) extracts on the clustering result, it is inaccurate to find out that multiple wavelet character extracts cluster segmentation method edge, the inventive method has solved the inaccurate problem in edge that multiple wavelet character extracts the cluster segmentation method, has also solved the over-segmentation problem of watershed divide simultaneously.

Claims (3)

1. based on the Watershed SAR image segmentation method of complex wavelet extraction mark, it is characterized in that, comprise the steps:

1.1 the mode with multiple wavelet energy feature is extracted the inner marker LImg that is used for watershed transform from original image Img;

1.2 being carried out classical watershed transform, inner marker LImg gets external label WLImg;

1.3 being carried out Gauss's low-pass filtering, original image Img gets image GImg after filtering;

1.4 filtered image GImg is asked gradient map PGImg with the Priwitt operator;

1.5 with inside and outside mark gradient map PGImg is carried out the gradient correction, makes the local minimum area of gradient map appear at mark position;

1.6 revised gradient map is carried out watershed transform, obtains final image segmentation figure RImg.

2. the Watershed SAR image segmentation method based on complex wavelet extraction mark according to claim 1 is characterized in that described step 1.1 specifically comprises following substep:

2.1 to the multiple wavelet energy feature of each pixel extraction of original image;

2.2 each pixel is carried out the K-mean cluster;

2.3 the image after the cluster is carried out Gauss's low-pass filtering;

2.4 filtered image is asked gradient map with the Priwitt operator;

2.5 extract inner marker from gradient map, find local minimum.

3. the Watershed SAR image segmentation method based on complex wavelet extraction mark according to claim 2, it is characterized in that in step 2.1 the multiple wavelet energy feature of each pixel extraction of original image being adopted following method: the video in window to each pixel carries out two-layer multiple wavelet decomposition, obtain a low frequency sub-band and every layer of 6 high-frequency sub-band, totally 13 subbands, ask multiple wavelet energy as eigenvector with following formula to each subband, the normalization eigenvector

$e=\frac{1}{M\·N}\underset{x=1}{\overset{M}{\mathrm{\Σ}}}\underset{y=1}{\overset{N}{\mathrm{\Σ}}}\left|s(x,y)\right|$

Wherein, (x y) is coefficient of dissociation to s, and M * N is the size of band image, and x and y be the row and column of presentation video respectively, and e is the energy feature vector.

Images