CN103996039B - SAR image channel extraction method combining gray-level threshold-value segmentation and contour shape identification - Google Patents

Abstract

The invention discloses an SAR image channel extraction method which combines gray-level threshold-value segmentation and contour shape identification. The method includes step1: adopting gray-level threshold-value segmentation to carry out a first background segmentation on an SAR image; step2: according to channel contour shape characteristics, carrying out sectioned modeling on channel areas, wherein the channel areas are represented as combination of a plurality of minimum external-connection rectangular windows; step3: combining minimum external-connection rectangular windows in the same area so as to form a plurality of second rectangular windows; step4: according to the shape and connectivity of a channel contour, splicing second rectangular windows which satisfy a condition into a rough channel area; and step5: carrying out gray-level threshold-value segmentation again so as to obtain a channel extraction image. The SAR image channel extraction method which combines gray-level threshold-value segmentation and contour shape identification inhibits background noises, which are approximate to both a water-body gray level and a shape characteristic in the SAR image, through combination of an image gray-level characteristic and a shape contour characteristic of the channel and a multi-mode SAR image segmentation decision so that channel areas are extracted precisely and accurately.

Description

Joint gray level threshold segmentation and the SAR image river extraction method of outline shape identification

Technical field

The present invention relates to the segmentation of synthetic aperture radar (SAR, Synthetic Aperture Rader) river course image and Technology of identification, more particularly, to a kind of based on SAR image using joint gray level threshold segmentation and outline shape identification technology carry The method taking river course, belongs to technical field of computer vision.

Background technology

In terms of remote sensing images river course object identification and extraction, synthetic aperture radar image-forming has it compared with optical imagery Unique advantage, is mainly reflected in, one, the spatial resolution of synthetic aperture radar (SAR) higher, can accurately describe River course target.Additionally, it can also be round-the-clock, the realizing earth observation and possess certain penetration capacity of round-the-clock；2nd, river course Water body in region has significant reflection of electromagnetic wave characteristic, is reflected in and shows as the relatively low image spy of gray value in SAR image Levy, with background between there is notable contrast.Therefore, it is widely used in river extraction based on the remote sensing images of SAR imaging.

In SAR image river extraction, gray level threshold segmentation and river course outline identification are the methods more commonly used, these sides Method has certain reference significance for the river course Objective extraction research in SAR image, but is single use single method to river Road identification, extraction aspect there is a problem of certain, are mainly reflected in the following aspects：

At present, SAR image resolution ratio has developed to sub-meter grade, and increasing Small object is same due to its reflection of electromagnetic wave rate Water body is close, easily forms stronger ambient noise and is difficult to realize filtering by gray threshold.Additionally, gray level threshold segmentation Cannot be distinguished by the difference between river course region and fragmentary water body, river course refers to the route that river flows through, be often referred to the water route that can open the navigation or air flight, Scattered water body region is simultaneously not belonging to the target of this research, for such target single cannot be effective by gray level threshold segmentation Suppression is realized on ground, therefore defines serious interference in segmentation result.Finally, intrinsic coherent speckle noise in SAR image Also river extraction can be produced and have a strong impact on.

River course profile is different from the bulk targets such as battlebus aircraft, also different from the regular linear target of bridge, road etc., river Road profile is relative complex, can be considered as roughly the ribbon region to institute's envelope for the parallel lines, but this belt-like zone not pen Straight, and the thickness in different section river course also there were significant differences, artificial works such as bridge in addition, the impact of dock etc., river course The priori that profile contains parallel lines pair is also often inapplicable, and therefore, the accurate modeling for river course outline shape is very difficult, The river course identification based on outline shape feature can be had a strong impact on.At present, conventional Snake modeling method is built to river course profile Mould, but because it is sensitive to initial profile location comparison, also more sensitive to noise, it is subject to certain restricting in its application.

River course target in SAR image is not suitable for extracting using the senior mode identification method such as cluster, because different river course Cosmetic variation larger and complex.It is difficult to river course and non-river course region are entered according to existing feature extraction and clustering method Row accurately clusters, and can not search out the classifying face of optimum.

In view of the above problems, the outline shape feature in the imaging characteristicses according to SAR image and river course, the invention discloses one Plant new joint gray level threshold segmentation and the SAR image river extraction method of outline shape identification.

Content of the invention

The main object of the present invention is the SAR image river extraction using joint gray level threshold segmentation and outline shape identification Method is extracted to river course, is combined by the gradation of image feature and form contour feature in river course and multi-modal SAR figure As segmentation decision-making, the ambient noise being all closer to water body gray scale and morphological feature in suppression SAR image, thus accurately carry Take out river course region.

For reaching above-mentioned purpose, the present invention provides a kind of joint gray level threshold segmentation and the SAR image of outline shape identification River extraction method, described method includes：

The first step, does first time background segment using gray level threshold segmentation to SAR image；

Second step, according to river course outline shape feature, the river course region in the image that the first step is obtained carries out segmentation and builds Mould, specifically adopts the image partition method based on graph theory, described river course region representation is some minimum enclosed rectangle windows Combination；

3rd step, carries out form identification to aforementioned each minimum enclosed rectangle window, will be external for the minimum in the same area Rectangular window merges, and forms some second rectangular windows；

4th step, the shape according to river course profile and connectedness, screen to aforementioned each second rectangular window, will meet bar Second rectangular window of part is spliced into rough river course region；

5th step, does background segment again using the image that gray level threshold segmentation obtains to the 4th step, obtains river extraction Image.

In the above-mentioned first step or the 5th step, the detailed content using gray level threshold segmentation is：Optimal grey is carried out according to following formula Degree threshold value k^*Calculate：

Wherein, L represents the grade of grey level histogram；K represent differentiation water body target with and water body target differ greatly target Gray threshold；Represent the overall class inherited of view picture SAR image,Wherein, w₁For target water The probability that body occurs, w₂Be with target water body differ greatly target appearance probability, μ₁For the average gray of target water body, μ₂For The average gray of the target that differs greatly with target water body.

In above-mentioned second step, river course region representation is the combination of m minimum enclosed rectangle window, and all minimum enclosed rectangle Window is satisfied by following condition：

λ(C_i)=| C_i|/|R_i| ＞ τ

Wherein, λ (C_i) represent i-th minimum enclosed rectangle window dutycycle, i=1,2 ..., m；|C_i| represent i-th The image-region pixel count of little boundary rectangle window；|R_i| represent the pixel count of i-th minimum enclosed rectangle window institute envelope；τ represents Default duty cycle threshold.

Whether the detailed content of upper 3rd step is, judge the side of each minimum enclosed rectangle window as second by the use of following criterion The border of rectangular window：

Wherein, i=1,2 ..., m, j=1,2 ..., m, and i ≠ j；Dif(C_i,C_j) be expressed as connecting two zoness of different C_i And C_jThe minimal weight on summit；MInt(C_i,C_j)=min (Int (C_i)+η(C_i),Int(C_j)+η(C_j)), wherein, Int (C_i)、 Int(C_j) represent C respectively_i、C_jWeight, using relevance function tolerance, determined by image self information；η(C_i)、η(C_j) point Biao Shi not C_i、C_jModulation parameter, and η (C)=a/ | C |, wherein | C | is the pixel count in region, and a is default modulation amplitude； λ(C_i)、λ(C_j) represent region C respectively_iAnd C_jDutycycle, λ (C_i,C_j) it is region C_iAnd C_jThe duty of the second rectangular window after merging β is modulation parameter to ratio, and scope is：0 ＜ β ＜ 1.

The condition that above-mentioned 4th step is screened is：

Corresponding length-width ratio L of each second rectangular window_R/W_RMore than threshold value T, that is,：

L_R/W_R＞ Τ,

And the distance of the nearest pixel of different second rectangular window is less than threshold gamma, that is,：

d(R_i,R_j) ＜ γ.

The 6th step is also included after above-mentioned 5th step：Using the burr to the image that the 5th step obtains for the region dilation erosion method Spot noise and hole are optimized process.

The present invention adopts technique scheme, has the advantages that：

1st, anti-ambient noise ability is strong.By the design using SAR image multi-stage division, can effectively suppress in background To river course, there is similar gray scale and the close ambient noise target of profile, it is to avoid the strong jamming that river extraction is formed.

2nd, river extraction integrality is high.The river course region being extracted by joint gray feature and outline shape feature, Loss is relatively low, and integrality is higher, and accuracy is preferable.

3rd, algorithm complex is remarkably decreased.The complexity of algorithm forms linear relationship with the pixel count of image, and complexity is relatively Low, it is easy to accomplish online process.

In view of above feature, the SAR image river extraction that the present invention can stablize, be reliably used in complex scene.

Brief description

The present invention is further described with reference to the accompanying drawings and examples.

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

Fig. 2 is the original SAR image in the embodiment of the present invention；

Fig. 3 is the result exemplary plot that in the embodiment of the present invention, first time gray level threshold segmentation obtains；

Fig. 4 is the rough river course example region figure in the embodiment of the present invention after the identification segmentation of river course outline shape；

Fig. 5 is final river extraction result exemplary plot in the embodiment of the present invention.

Specific embodiment

In order to more fully hereinafter understand feature and the technology contents of the present invention, illustrate with reference to specific embodiment Embodiments of the present invention, but embodiments of the present invention not limited to this.

In the present embodiment, using spaceborne or carried SAR to target river course field emission electromagnetic wave, and receive target river course area The echo-signal of domain feedback, the spaceborne or airborne SAR image according to echo signal form target river course region, this SAR image is For described original SAR image, as shown in Fig. 2 the present invention is by processing to this SAR image, therefrom accurately extract river course.

In a preferred embodiment of the invention, in conjunction with shown in Fig. 1, described joint Threshold segmentation and outline shape are known Other SAR image river extraction method comprises the following steps：

Step one, first time background segment is done to SAR image using gray level threshold segmentation, specifically adopt Otsu method to choose figure Optimum gray threshold k as segmentation^*；According to optimum gray threshold k^*SAR image is done with first time background segment；

Build the histogram of [1,2 ..., L] tonal gradation first according to SAR image, wherein belong to the pixel of tonal gradation b Quantity is n_b.Therefore, the pixel count of entire imageThus be in a certain gray scale pixels in the images and occur Probability can be described as：

p_b=n_b/N (1)

Wherein, p_b>=0,

Secondly, water body and the target larger with water body gray scale are classified, water body target class is defined as c₁, with water body The larger target class of mesh gray difference is defined as c₂, it is k for distinguishing this two classifications target gray threshold.Therefore, two class target The probability w being occurred₁, w₂It is respectively：

Classification c₁, c₂Corresponding average gray μ₁, μ₂And the average gray μ of entire image_TIt is respectively：

The overall gray variance of view picture SAR imageAnd class inheritedIt is respectively：

The method turns to optimal threshold selection criterion with class inherited maximum, sets up the error function ξ with regard to gray threshold k K () is：

Finally, draw optimum gray threshold k^*It is calculated as：

According to optimum gray threshold k^*First time background segment is carried out to original SAR image, obtains first time gray threshold The result schematic diagram of segmentation, as shown in Figure 3.

Step 2, according to river course outline shape feature, the river course region in the image that step one is obtained carries out segmentation and builds Mould, using the image partition method based on graph theory, the combination that described river course region representation is some minimum enclosed rectangle windows；

Apply to the river of complexity by the characteristic that nonlinear curved lines all can be approximately linear straight-line segment combination In road outline shape, therefore river course profile can approximately be identified using the combination of minimum enclosed rectangle window, thus proposing one Plant segmentation modeling method.Compacted every section of river course of envelope region according to minimum enclosed rectangle window, introduce dutycycle λ (C_i), in minimum On the premise of the dutycycle of adjacent rectangular window is more than certain threshold tau, obtains river course skeleton pattern and be all piecewise rectangular windows Combination, and accordingly image is processed；

λ(C_i)=| C_i|/|R_i| (11)

Wherein

Wherein | C_i| by the image-region pixel count being partitioned into, | R_i| for this region minimum enclosed rectangle window institute envelope Pixel count, C_rFor minimum enclosed rectangle window, R_rFor the river course region of institute's envelope, τ represents default duty cycle threshold.

Step 3, aforementioned each minimum enclosed rectangle window is carried out with form identification, introduce the merging constraints pair of new region Minimum enclosed rectangle window merges, and forms some new regions；

Gained image will be processed through step 2 and be mapped as weighted-graph G=(V, E), wherein v_x∈ V is every in image Individual pixel；(v_x,v_y) ∈ E be image contour connection adjoin summit；Each border (v in the picture_x,v_y) ∈ E is all corresponding A non-negative connection weight w ((v_x,v_y)), represent vertex v_xAnd v_ySimilitude, the connection between belonging to the summit in region Weight w ((v_x,v_y)) should be less than the connection weight between summit between zones of different；

The same areaInterior difference is defined as the weight limit on this region summit minimum spanning tree MST (C, E)：

Between zones of differentDifference be connect two regions summit minimal weight：

Two interregional threshold value MInt (C with the presence or absence of border₁,C₂) may be defined as internal weight Int of two adjacent domains (C) and modulation parameter η (C) sum minimum of a value：

MInt(C₁,C₂)=min (Int (C₁)+η(C₁),Int(C₂)+η(C₂)) (15)

Wherein, Int (C) embodies the similitude between adjacent interval pixel, using relevance function tolerance, by image itself Information such as intensity, color etc. determine；Design η (C)=a/ | C |, wherein | C | is the pixel count in region, and a is default modulated amplitude Degree, span is between 200 to 300；By this design, for the region that area is too small, it usually needs larger weight Just can determine that the presence of zone boundary.

Dutycycle according to combined region is more than or equal to threshold value：

λ(C_i,C_j)≥β(λ(C_i)+λ(C_j)) (16)

Wherein, λ (C_i,C_j) it is region C_iAnd C_jThe dutycycle of new region after merging, λ (C_i) it is region C_iDutycycle, λ (C_j) it is region C_jDutycycle, β be modulation parameter, 0 ＜ β ＜ 1；β=0.75 in the present embodiment, according to the judgement of zone boundary Carry out image segmentation according to (17).

Step 4, the shape according to river profile and connectedness, screen to aforementioned each new region, because river course is divided Become different sections, each section is modeled as second rectangular window as the property of the contour feature in this region, according to river course The shape of profile and the connective river course region fusion rule that defines are identified to river course region, meet the second square of filter criteria The image-region of shape window institute envelope is river course, and the criterion of concrete screening is to meet：

Corresponding length-width ratio L of second rectangular window_R/W_RMore than threshold value T, that is,：

L_R/W_R＞ Τ (18)

And the distance of nearest pixel is less than threshold gamma, that is, between the second rectangular window of zones of different：

d(R_i,R_j) ＜ γ (19)

Concrete value T=4.5, γ=10 in the present embodiment, the identifying processing result obtaining is as shown in Figure 4.

Step 5, again adopt Otsu gray level threshold segmentation method, image is split again；

On the basis of the rough river course region that first time gray level threshold segmentation and river course outline shape identifying processing obtain, Calculate the optimum gray threshold k* that (formula 10) goes out this image again according to optimum gray scale threshold calculations formula, according to optimum gray scale threshold Value k* carries out the secondary segmentation that becomes more meticulous and extracts to rough river course area image.

Step 6, using region dilation erosion algorithm, the burr point in image and hole are suppressed and filled, finally River extraction result is as shown in Figure 5.

Claims (6)

1. a kind of SAR image river extraction method of joint gray level threshold segmentation and outline shape identification is it is characterised in that described Method include：

The first step, does first time background segment using gray level threshold segmentation to SAR image；

Choose the optimum gray threshold k of image segmentation using Otsu method^*：According to optimum gray threshold k^*Do background to SAR image to divide Cut；Build the histogram of [1,2 ..., L] tonal gradation first according to SAR image, wherein belong to the pixel quantity of tonal gradation b For n_b, therefore, the pixel count of entire imageThus it is in the general of a certain gray scale pixels appearance in the images Rate is described as：

p_b=n_b/N (1)

Wherein, p_b>=0,

Secondly, water body and the target larger with water body gray scale are classified；Water body target class is defined as c₁, with water body mesh ash Spend the target class differing greatly and be defined as c₂, it is k for distinguishing this two classifications target gray threshold, therefore, two class targets are gone out Existing probability w₁, w₂It is respectively：

$w_1=\underset{b=1}{\overset{k}{\Σ}}{p}_b---\left(2\right)$

$w_2=\underset{b=k+1}{\overset{L}{\Σ}}{p}_b---\left(3\right)$

Classification c₁, c₂Corresponding average gray μ₁, μ₂And the average gray μ of entire image_TIt is respectively：

${\mathrm{\μ}}_1=\underset{b=1}{\overset{k}{\Σ}}{\mathrm{bp}}_b/w_1---\left(4\right)$

${\mathrm{\μ}}_2=\underset{b=k+1}{\overset{L}{\Σ}}{\mathrm{bp}}_b/w_2---\left(5\right)$

${\mathrm{\μ}}_T=\underset{b=1}{\overset{L}{\Σ}}{\mathrm{bp}}_b---\left(6\right)$

The overall gray variance of view picture SAR imageAnd class inheritedIt is respectively：

${\mathrm{\σ}}_T^2=\underset{b=1}{\overset{L}{\Σ}}{(b-{\mathrm{\μ}}_T)}^2{p}_b---\left(7\right)$

${\mathrm{\σ}}_B^2=w_1{w}_2{({\mathrm{\μ}}_1-{\mathrm{\μ}}_2)}^2---\left(8\right)$

The method turns to optimal threshold selection criterion with class inherited maximum, sets up error function ξ (k) with regard to gray threshold k For：

$\mathrm{\ξ}\left(k\right)={\mathrm{\σ}}_B^2\left(k\right)/{\mathrm{\σ}}_T^2---\left(9\right)$

Finally, show that optimum gray scale threshold calculations are：

$k^{*}=\underset{k}{\arg }\left(\underset{1\≤k\≤L}{max}{\mathrm{\σ}}_B^2\right(k\left)\right)---\left(10\right)$

According to optimum gray threshold k^*Background segment is carried out to original SAR image；

Second step, according to river course outline shape feature, the river course region in the image that the first step is obtained carries out segmentation modeling, tool Body is using the image partition method based on graph theory, the combination that described river course region representation is some minimum enclosed rectangle windows；

3rd step, carries out form identification to aforementioned each minimum enclosed rectangle window, by the minimum enclosed rectangle in the same area Window merges, and forms some second rectangular windows；

4th step, the shape according to river course profile and connectedness, screen to aforementioned each second rectangular window, will meet condition Second rectangular window is spliced into rough river course region；

5th step, does background segment again using the image that gray level threshold segmentation obtains to the 4th step, obtains river extraction image.

2. the SAR image river extraction method of joint gray level threshold segmentation according to claim 1 and outline shape identification, It is characterized in that, in the described first step or the 5th step, the detailed content using gray level threshold segmentation is：Optimum is carried out according to following formula Gray threshold k^*Calculate：

$k^{*}=\underset{k}{\arg }\left(\underset{1\≤k\≤L}{max}{\mathrm{\σ}}_B^2\right(k\left)\right)$

Wherein, L represents the grade of grey level histogram；K represent distinguish water body target with and water body target differ greatly the ash of target Degree threshold value；Represent the overall class inherited of view picture SAR image,Wherein, w₁Occur for target water body Probability, w₂Be with target water body differ greatly target appearance probability, μ₁For the average gray of target water body, μ₂It is and target Water body differs greatly the average gray of target.

3. the SAR image river extraction method of joint gray level threshold segmentation according to claim 1 and outline shape identification, It is characterized in that, in described second step, river course region representation is the combination of m minimum enclosed rectangle window, and all minimums are external Rectangular window is satisfied by following condition：

λ(C_i)=| C_i|/|R_i| ＞ τ

Wherein, λ (C_i) represent i-th minimum enclosed rectangle window dutycycle, i=1,2 ..., m；|C_i| represent that i-th minimum is outer Connect the image-region pixel count of rectangular window, | R_i| represent the pixel count of i-th minimum enclosed rectangle window institute envelope；τ represents default Duty cycle threshold.

4. the SAR image river extraction method of joint gray level threshold segmentation according to claim 3 and outline shape identification, It is characterized in that, whether the detailed content of described 3rd step is to be made using the side that following criterion judges each minimum enclosed rectangle window Border for the second rectangular window：

$D(C_1,C_2)=\left\{\begin{array}{cc}true& \begin{array}{cccc}if& Dif(C_i,C_j)>MInt(C_i,C_j)& or& \mathrm{\λ}(C_i,C_j)\≤\mathrm{\β}\left(\mathrm{\λ}\right(C_i)+\mathrm{\λ}(C_j\left)\right)\end{array}\\ false& otherwise\end{array}\right.$

Wherein, i=1,2 ..., m, j=1,2 ..., m, and i ≠ j；Dif(C_i,C_j) be expressed as connecting two zoness of different C_iAnd C_j The minimal weight on summit；MInt(C_i,C_j)=min (Int (C_i)+η(C_i),Int(C_j)+η(C_j)), wherein, Int (C_i)、Int (C_j) represent C respectively_i、C_jWeight, using relevance function tolerance, determined by image self information；η(C_i)、η(C_j) difference table Show C_i、C_jModulation parameter, and η (C)=a/ | C |, wherein | C | is the pixel count in region, and a is default modulation amplitude；λ (C_i)、λ(C_j) represent region C respectively_iAnd C_jDutycycle, λ (C_i,C_j) it is region C_iAnd C_jThe duty of the second rectangular window after merging β is modulation parameter to ratio, and scope is：0 ＜ β ＜ 1.

5. the SAR image river extraction method of joint gray level threshold segmentation according to claim 4 and outline shape identification, It is characterized in that, the condition that described 4th step is screened is：

Corresponding length-width ratio L of each second rectangular window_R/W_RMore than threshold value T, that is,：

L_R/W_R＞ T,

And the distance of the nearest pixel of different second rectangular window is less than threshold gamma, that is,：

d(R_i,R_j) ＜ γ.

6. the SAR image river extraction method of joint gray level threshold segmentation according to claim 1 and outline shape identification, It is characterized in that, also include the 6th step after described 5th step：Using region dilation erosion method to the image that the 5th step obtains Burr spot noise and hole are optimized process.