CN105138992A - Coastline detection method based on regional active outline model - Google Patents

Abstract

The invention proposes a coastline detection method based on a regional active outline model, and the method has the following steps: employing the difference between a neighborhood pixel value of a false point in an image and the pixel value of the false point to calculate a correction value for correcting the false point, and obtaining a corrected remote sensing image; enabling the corrected remote sensing image to be imported into a mean value decreasing energy functional model, and obtaining an improved mean value decreasing energy functional; calculating the parameters of the remote sensing image through an improved active outline model according to the improved energy functional after the false point is modified, and obtaining a level set updating equation; enabling the improved active outline model to automatically expand or withdraw according to the level set updating equation, and completing the detection of a coastline when the absolute value of two adjacent iteration differences of an outline is less than a specific value and the outline movement stops. The whole process is not affected by an initial outline, and the method solves a problem of false point interference through a false point removal method during the segmenting of a remote sensing image through an existing active outline model, and enables the outline of the coastline to be clear and unique.

Description

A kind of coastline Detection Method method based on region active contour model

Technical field

The present invention relates to coastline Detection Method technical field, particularly relate to a kind of coastline Detection Method method based on region active contour model.Relate to Patent classificating number G06 to calculate; Calculate; Counting G06K data identification; Data representation; Record carrier; The process G06K9/00 of record carrier is used for reading or identify printing or written character or for identifying figure, such as, and the method for fingerprint or the extraction of device G06K9/46 characteristics of image or characteristic.

Background technology

Shore line is located in the boundary on Yu Haiyang and land, and the geography superiority by feat of its uniqueness becomes the important area that the mankind competitively develop.Silt along with seashore is risen, corrode, climate warming causes that sea level rise and the mankind build the impact of the social factor such as dykes and dams, landfill soil stone at coastal zone, shore line is in constantly expansion or stretch, have complicacy and polytrope, the detection in shore line is that one of object paid close attention to detects in environment department.The detection in shore line has great significance in renewal map, ocean resources management, marine navigation field, and the detection in shore line can realize people to the sustainable utilization of coastal zone resources and scientific management.

People usually use the method for visual interpretation to carry out the detection in shore line, but visual interpretation needs the manpower of at substantial, extremely expends time in.So people more and more focus on using the method for machine interpretation to carry out the detection in shore line.Nineteen ninety, scholar Jone-senLee and IgorJurkevich proposes the algorithm that a kind of boundary tracking detects shore line, and this algorithm is first by remote sensing images binaryzation, and definition search rule, uses the method for boundary tracking to mark and draw shore line.This use is first to image binaryzation, and then the method for boundary tracking reasonably can detect shore line, so extensively quote from the scholar proposing to start just to be subject to studying remote sensing images tidal saltmarsh.But this algorithm is very sensitive to noise, and require higher to initial image segmentation result, comparatively serious to the dependence of image.Through the arrangement to domestic and international remote sensing images coastline Detection Method method in recent ten years, the detection method in shore line in remote sensing images is summarized as: the method for rim detection, the method for threshold value, the method for region segmentation and the method for active contour model, wherein, the method for active contour model can be divided into: the active contour model based on edge and the active contour model based on region.

2010, Liu Yongxue etc. proposed a kind of OO remote sensing image tidal saltmarsh method, first carry out filtering, then selected dividing method to split, extracted shore line according to segmentation result region growing methods.Filtering method and general dividing method all have certain artificial participation, and the accuracy of filtering and dividing method will directly affect the extraction accuracy in later stage, have determinacy scarcely.

2012, Gu Dandan etc. proposed land, a kind of SAR image based on wavelet transformation and OTSU threshold value sea dividing method, carry out coarse segmentation, then extract shore line with small echo and Edge Following with threshold method.There is the inaccurate problem of coarse segmentation in threshold method, needs artificially to participate in, and aftertreatment is complicated.

In the same year, Cheng Liang etc. propose a kind of tidal saltmarsh method based on Satellite microwave remote sensing image, adopt anisotropy parameter to carry out filtering and complete pre-service, then with multiple dimensioned to image block, carry out coastline Detection Method by selecting appropriate threshold.Filtering can lose the part detailed information at edge, and Threshold selection needs artificial participation.

2011, the people such as Huang Kuihua proposed a kind of partial statistics active contour model based on G0 distribution and detect shore line, and this algorithm is mainly for synthetic-aperture radar (SAR) image, and its implementation is: (1) Image semantic classification.In order to the consistance of optimum configurations, first raw image data interval is set to [0,255].(2) C-V model segmentation.Use C-V model to carry out coarse segmentation to SAR image, the result of coarse segmentation is as the initial profile line of next step fine segmentation.(3) set the sliding window size of neighborhood, to its sample average of calculating each on outline line and sample mean value of square, estimate the local parameter of G0 distribution.(4) level set function is upgraded, realize the accurate detection to SAR image shore line.There is two problems in this coastline Detection Method method: 1, coastline Detection Method is responsive to initial profile.If initial profile is on sea, coastline Detection Method is ideal, if initial profile is selected in land, coastline Detection Method there will be multiple contour area.2, its probability Distribution Model is not suitable for remote sensing image.

Summary of the invention

The present invention is directed to the problem that traditional dividing method based on active contour model is difficult to split complicated remote sensing images, and the pseudo-point that remote sensing images are too much because the strong unevenness of density exists, the regional average values such as land vegetation, hills present similarity with the regional average value in sea, the situation of the coastline Detection Method result of remote sensing images that what these factors were serious have impact on, propose a kind of coastline Detection Method method based on region active contour model, comprise the steps:

First, the pseudo-point in remote sensing images to be analyzed is detected; To the pseudo-point confirmed after testing, use this pseudo-neighborhood of a point pixel value and pseudo-point described in pseudo-mathematic interpolation correction correction of putting pixel value, obtain revised remote sensing images.

By adopting above-mentioned technical scheme, effectively can eliminate the impact that the puppet point in remote sensing images brings, and relative to the method adopting Largest Mean point (causing sea to occur land feature) and minimum mean point (sea feature appears in Sea continental margin) to eliminate pseudo-point, decrease and correct mistakes.

Then revised remote sensing images are imported average to successively decrease in energy functional model, the average be improved is successively decreased energy functional; According to the energy functional of described improvement, regional average value diminishing method is adopted to generate the active contour model improved;

Use the active contour model of described improvement to calculate the parameter of the revised remote sensing images of described puppet point, obtain level set renewal equation;

According to described level set renewal equation, make the active contour model of described improvement automatically expand or shrink, when before and after profile, the absolute value of adjacent twice iteration difference is less than specified value, contour motion stops; Now contour curve maintenance is stablized constant, completes the detection in shore line.

As preferred embodiment, the step of described " detecting the pseudo-point in remote sensing images to be analyzed " is specific as follows:

First to the doubtful pseudo-point in remote sensing images to be analyzed, this doubtful pixel value of puppet point and pixel average in the multiple homalographic region of this puppet point periphery is contrasted; The pixel value of puppet point as doubtful in this is less than periphery arbitrary region pixel average, then judge that this doubtful puppet point is as pseudo-point.

Further, described " multiple homalographic region " is:

Four areas are the square subregion of α × α the area of composition is the square area W of 2 α × 2 α _x; Wherein, x represents described doubtful pseudo-point, i.e. central point, and α represents the positive integer of image window size, value between 1 to 10;

Detect whether doubtful puppet point is pseudo-point according to following formula:

$Index\left(x\right)=\left\{\begin{array}{c}1,I\left(x\right)<min\left({\mathrm{\&mu;}}^k\right(x),k=1,2,3,4)\\ 0,else\end{array}\right.,$

Wherein, μ ^kx () represents the pixel average of a kth sub-square area, Index (x) represents detective operators, and I (x) represents the pixel value of described central point/doubtful puppet point; When the pixel value of the described central point on observed image is less than the minimum value in the pixel average in four regions built centered by described central point, Index (x)=1, described central point is pseudo-point; Otherwise Index (x)=0, described central point is not pseudo-point.

As preferred embodiment, use this pseudo-neighborhood of a point pixel value and pseudo-point described in pseudo-mathematic interpolation correction correction of putting pixel value " specifically comprise the steps:

By formula D ^k(x)=| I (x)-μ ^k(x) |, k=1,2,3,4 computing parameters

Wherein, represent the absolute value of the pixel value of described central point and the difference of a kth sub-square window pixel average;

By formula ε (x)=μ ^k(x)-I (x), k=argmin _k(D ^k(x), k=1,2,3,4) determine μ ^kx correction that () puts as described puppet with the minor increment of I (x), wherein, ε (x) represents the correction of pseudo-some x; Use the pseudo-point through confirming described in described correction correction.

Further, revised remote sensing images I ^newx the expression formula of () is:

I ^new(x)＝I(x)+index(x)ε(x)

The pass of revised remote sensing images and initial remote sensing images is:

I(x)+index(x)ε(x)≈b(y)c _i+n(x)

Wherein, b (y) indicates inclined field, c _irepresent image-region average, n (x) represents picture noise.

Consider, improve the impact that energy model (pixel value in pseudo-some periphery homalographic region) can revise pseudo-point, but still there is the region of some easy over-segmentations, simple puppet point correction cannot be worked to these regions, the area pixel of these easy over-segmentations is positioned at two class regions, namely, between ocean and the average of land area, called after weak-strong test, these weak-strong test cause the model improved still to be difficult to distinguish just.

By analysis, in an iterative process, the regional average value c that two class regions are excessive _icause segmentation errors, can be expected by this point, must in energy model minimization process control band average, namely must reduce regional average value c _ithe segmentation that guarantee is correct.Pretend as preferred embodiment, the average of the improvement " successively decrease energy functional " is specially:

$\begin{array}{c}E=\&Integral;\&lsqb;\underset{outside\left(C\right)}{\&Integral;}{K}_{\mathrm{\&sigma;}}\left(y-x\right){\left(I\left(x\right)+index\left(x\right)\mathrm{\&epsiv;}\left(x\right)-b\left(y\right)c_1\right)}^{2}dx\&rsqb;dy\\ +\&Integral;\&lsqb;\underset{inside\left(C\right)}{\&Integral;}{K}_{\mathrm{\&sigma;}}\left(y-x\right){\left(I\left(x\right)+index\left(x\right)\mathrm{\&epsiv;}\left(x\right)-b\left(y\right)c_2\right)}^{2}dx\&rsqb;dy\end{array}$

Wherein, E represents energy functional, and y represents that image is in x neighborhood of a point pixel, K _σrepresent gaussian kernel, c ₁represent land area average, c ₂represent sea region average, outside (C) represents outline line perimeter, and inside (C) represents outline line interior zone; Accordingly, the energy functional of described improvement is:

$\begin{array}{c}{E}^{MLIC}=\\ \&Integral;\&lsqb;\underset{outside\left(C\right)}{\&Integral;}{K}_{\mathrm{\&sigma;}}\left(y-x\right){\left(I\left(x\right)+index\left(x\right)\mathrm{\&epsiv;}\left(x\right)-b\left(y\right)c_1\right)}^{2}dx\&rsqb;dy\\ +\&Integral;\&lsqb;\underset{inside\left(C\right)}{\&Integral;}{K}_{\mathrm{\&sigma;}}\left(y-x\right){\left(I\left(x\right)+index\left(x\right)\mathrm{\&epsiv;}\left(x\right)-b\left(y\right)c_2\right)}^{2}dx\&rsqb;dy\\ +{\mathrm{\&omega;}}_1{c}_1^2\&Integral;Sdy+{\mathrm{\&omega;}}_2{c}_2^2\&Integral;Sdy\end{array},$

Wherein, ω ₁, ω ₂represent weight coefficient, and meet 0≤ω ₁< 1,0≤ω ₂< 1, E ^mILCrepresent the energy functional of average, represent the first bound term that average is successively decreased, represent that average successively decreases second successively decreases bound term, c ₁and c ₂represent the regional average value in Liang Ge region, land, image sea respectively.

Accordingly, the active contour model of described improvement is:

Wherein, represent that the average with model parameter is successively decreased energy functional;

represent regularization term, represent Heaviside function,

μ represents regularization term regulating parameter, is arithmetic number, represent profile gradients, represent that the gradient of profile Heaviside function smears value, v represents regularization term regulating parameter, is arithmetic number;

Described step "-use the active contour model of described improvement to calculate the parameter of the revised remote sensing images of described puppet point, obtain level set renewal equation; " specifically comprise the steps:

The active contour model of the improvement that-basis obtains, obtains c ₁, c ₂, b:

$c_1=\frac{\&Integral;(b*K_{\mathrm{\&sigma;}})\left(I\right(x)+index(x\left)\mathrm{\&epsiv;}\right(x\left)\right)u_{1}dy}{\&Integral;(b^2*K_{\mathrm{\&sigma;}})u_{1}dy+{\mathrm{\&omega;}}_1\&Integral;Sdy},$

$c_2=\frac{\&Integral;(b*K_{\mathrm{\&sigma;}})\left(I\right(x)+index(x\left)\mathrm{\&epsiv;}\right(x\left)\right)u_{2}dy}{\&Integral;(b^2*K_{\mathrm{\&sigma;}})u_{2}dy+{\mathrm{\&omega;}}_2\&Integral;Sdy},$

$b=\frac{\left\{\left(I\left(x\right)+index\left(x\right)\mathrm{\&epsiv;}\left(x\right)\right)\underset{i=1}{\overset{2}{\&Sigma;}}{c}_i{u}_i\right\}*K}{\left(\underset{i=1}{\overset{2}{\&Sigma;}}{c}_i^2{u}_i\right)*K},$

Wherein, * convolution algorithm is represented, represent Heaviside function;

-according to gradient descent flow method, obtain representing that level set is measured over time and be:

Wherein, e _i(x)=∫ K (y-x) (I (x)+index (x) ε (x)-b (y) c _i) ²dy, wherein represent the variable quantity of level set, represent Heaviside function derivative, e _irepresent one of front two differentiate results about level set on the right side of step 402 equal sign, K (y-x) is pixel y and x distance gaussian kernel, v and μ is regulating parameter, is arithmetic number. represent the curvature of level set, for item second from the bottom on the right side of step 402 equal sign is for the derivative of level set, for item last on the right side of step 402 equal sign is for the derivative of level set;

-the renewal equation that obtains level set is:

Wherein represent the level set of n-th iteration, represent the level set of (n-1)th iteration.

Accompanying drawing explanation

In order to the technical scheme of clearer explanation embodiments of the invention or prior art, introduce doing one to the accompanying drawing used required in embodiment or description of the prior art simply below, apparently, accompanying drawing in the following describes is only some embodiments of the present invention, for those of ordinary skill in the art, under the prerequisite not paying creative work, other accompanying drawing can also be obtained according to these accompanying drawings.

The process schematic of the coastline Detection Method method based on region active contour model that Fig. 1 provides for the embodiment of the present invention;

The realization flow figure of the coastline Detection Method method based on region active contour model that Fig. 2 provides for the embodiment of the present invention;

Fig. 3 is the Application of Splitting Window design sketch based on observed image in the coastline Detection Method method of region active contour model that the embodiment of the present invention provides;

The Detection results figure of the coastline Detection Method method based on region active contour model that Fig. 4 a-c provides for the embodiment of the present invention.

Embodiment

For making the object of embodiments of the invention, technical scheme and advantage clearly, below in conjunction with the accompanying drawing in the embodiment of the present invention, clear complete description is carried out to the technical scheme in the embodiment of the present invention:

As shown in Figure 2, the coastline Detection Method method based on region active contour model that the embodiment of the present invention provides comprises:

First remote sensing images are gathered.

The remote sensing images size collected is 200 × 200 pixels, and remote sensing images that Landsat, Radarsat or Spot tri-kinds is different can be used as data source.The method gathering remote sensing images comprises: utilize spaceborne multiband optical imaging method to gather optical imagery; Or utilize satellite-borne synthetic aperture radar formation method to gather diameter radar image.

Then, carry out puppet point to the remote sensing images collected to detect.

Wherein, pseudo-point is the given window average of remote sensing images current point, than given window edge pixel in all little pixel region of four, given window size region border average.Pseudo-some position is in the depressed portions in neighborhood, and can well put position to puppet by neighborhood information positions, concrete:

Then, using the remote sensing images that collect as observed image, observed image positions puppet point.

Point centered by puppet point, it is that the square window of 2 α × 2 α is set to W that observed image is done area _x, centrally put and do horizontal and vertical two line segments respectively, Application of Splitting Window is become four onesize sub-square window for α × α, name four sub-square window to be respectively according to order from left to right, from top to bottom wherein, x represents central point, and α represents the doubtful pseudo-point that x represents described, i.e. central point, α represents the positive integer of image window size, value between 1 to 10.

Fig. 3 is the Application of Splitting Window design sketch based on observed image in the coastline Detection Method method of region active contour model that the embodiment of the present invention provides.After window is separated, the Application of Splitting Window effect of observed image is with reference to Fig. 3.

Step 203, calculates four sub-square window respectively pixel average, use μ respectively ¹(x), μ ²(x), μ ³(x) and μ ⁴x () represents.

Detect whether current point is pseudo-point by following formula:

$Index\left(x\right)=\left\{\begin{array}{c}1,I\left(x\right)<min\left({\mathrm{\&mu;}}^k\right(x),k=1,2,3,4)\\ 0,else\end{array}\right.$

Wherein, μ ^kx () represents the pixel average of a kth sub-square window, Index (x) represents detective operators, and I (x) represents the pixel value of described central point; When the pixel value of the described central point on observed image is less than the minimum value in the pixel average in four regions built centered by described central point, Index (x)=1, described central point is pseudo-point; Otherwise Index (x)=0, described central point is not pseudo-point.Wherein, whether detecting device Index (x) can well be that pseudo-point detects to an x.

If detect that described central point is pseudo-point, then by formula D ^k(x)=| I (x)-μ ^k(x) |, k=1,2,3,4 computing parameters wherein, represent the absolute value of the pixel value of described central point and the difference of a kth sub-square window pixel average.

The puppet point detected is revised.

By formula ε (x)=μ ^k(x)-I (x), k=argmin _k(D ^k(x), k=1,2,3,4) determine μ ^kx correction that () puts as described puppet with the minor increment of I (x), wherein, ε (x) represents the correction of pseudo-some x.

Detailed process: the present invention defines pseudo-some x correction ε (x), and ε (x) is expressed as ε (x)=μ ^k(x)-I (x), k=argmin _k(D ^k(x), k=1,2,3,4), find and determine μ ^kx correction ε (x) that () puts as described puppet with the minor increment of I (x).The present invention uses minor increment to define correction both effectively can revise the generation that existing puppet point also can avoid new pseudo-point.

Utilize the described correction determined to revise described puppet point, obtain the relation of new observed image and observed image and true picture, have modified the new observed image of pseudo-point by formula I ^newx ()=I (x)+index (x) ε (x) represents, wherein, and I ^newx () represents the new observed image that have modified pseudo-point, the observed image obtained and the pass of true picture are: I (x)+index (x) ε (x) ≈ b (y) c _i+ n (x), wherein, b (y) indicates inclined field, c _irepresent image-region average, n (x) represents picture noise.

Concrete, after solving the problem of how to locate pseudo-point and how to revise pseudo-point, if the new observed image that have modified pseudo-point is I ^newx (), new observed image is by formula I ^newx ()=I (x)+index (x) ε (x) represents.The correction of the described puppet point determined is utilized to revise puppet point, the change of epigraph can have been followed in the inclined field that has of slow change, and the pass obtaining observed image and true picture is: I (x)+index (x) ε (x) ≈ b (y) c _i+ n (x).

Substituted in energy functional by described new observed image, the energy functional be improved is:

$\begin{array}{c}E=\&Integral;\&lsqb;\underset{outside\left(C\right)}{\&Integral;}{K}_{\mathrm{\&sigma;}}\left(y-x\right){\left(I\left(x\right)+index\left(x\right)\mathrm{\&epsiv;}\left(x\right)-b\left(y\right)c_1\right)}^{2}dx\&rsqb;dy\\ +\&Integral;\&lsqb;\underset{inside\left(C\right)}{\&Integral;}{K}_{\mathrm{\&sigma;}}\left(y-x\right){\left(I\left(x\right)+index\left(x\right)\mathrm{\&epsiv;}\left(x\right)-b\left(y\right)c_2\right)}^{2}dx\&rsqb;dy\end{array},$

Wherein, E represents energy functional, and y represents that image is in x neighborhood of a point pixel, K _σrepresent gaussian kernel, c ₁represent land area average, c ₂represent sea region average, outside (C) represents outline line perimeter, and inside (C) represents outline line interior zone.

The method utilizing regional average value to successively decrease obtains the active contour model improved.

According in the energy functional of improvement that obtains, the average the be improved energy functional that successively decreases is:

$\begin{array}{c}{E}^{MLIC}=\\ \&Integral;\&lsqb;\underset{outside\left(C\right)}{\&Integral;}{K}_{\mathrm{\&sigma;}}\left(y-x\right){\left(I\left(x\right)+index\left(x\right)\mathrm{\&epsiv;}\left(x\right)-b\left(y\right)c_1\right)}^{2}dx\&rsqb;dy\\ +\&Integral;\&lsqb;\underset{inside\left(C\right)}{\&Integral;}{K}_{\mathrm{\&sigma;}}\left(y-x\right){\left(I\left(x\right)+index\left(x\right)\mathrm{\&epsiv;}\left(x\right)-b\left(y\right)c_2\right)}^{2}dx\&rsqb;dy\\ +{\mathrm{\&omega;}}_1{c}_1^2\&Integral;Sdy+{\mathrm{\&omega;}}_2{c}_2^2\&Integral;Sdy\end{array}.$

Detailed process is: the first bound term that definition average is successively decreased and the second bound term, wherein, represent the first bound term that average is successively decreased, represent average second bound term of successively decreasing, the first bound term and the second bound term are introduced in the energy functional of described improvement, the average the be improved energy functional that successively decreases is:

$\begin{array}{c}{E}^{MLIC}=\\ \&Integral;\&lsqb;\underset{outside\left(C\right)}{\&Integral;}{K}_{\mathrm{\&sigma;}}\left(y-x\right){\left(I\left(x\right)+index\left(x\right)\mathrm{\&epsiv;}\left(x\right)-b\left(y\right)c_1\right)}^{2}dx\&rsqb;dy\\ +\&Integral;\&lsqb;\underset{inside\left(C\right)}{\&Integral;}{K}_{\mathrm{\&sigma;}}\left(y-x\right){\left(I\left(x\right)+index\left(x\right)\mathrm{\&epsiv;}\left(x\right)-b\left(y\right)c_2\right)}^{2}dx\&rsqb;dy\\ +{\mathrm{\&omega;}}_1{c}_1^2\&Integral;Sdy+{\mathrm{\&omega;}}_2{c}_2^2\&Integral;Sdy\end{array},$

Wherein, ω ₁, ω ₂represent weight coefficient, and meet 0≤ω ₁< 1,0≤ω ₂< 1, E ^mILCrepresent the energy functional that average is successively decreased.ω ₁, ω ₂larger, energy functional is more responsive to weak-strong test.Such as, initialization ω _i, i=1,2, parameter ω ₁=0, ω ₂=0.2.

First bound term and the second bound term are introduced in the energy functional of the improvement obtained in step 303, when in the minimized iterative process of energy functional, regional average value can reduce along with iteration, finally stablizes, and the regional average value of reduction can ensure correct segmentation.

In the average of improvement that obtains to successively decrease energy functional, the active contour model be improved is:

Detailed process is: introduce one and the onesize all 1's matrix S of described observed image, introduce level set sum fanction item the active contour model be improved is:

Wherein, represent that the average with model parameter is successively decreased energy functional, represent regularization term, represent Heaviside function, μ represents regularization term regulating parameter, is arithmetic number, represent profile gradients, represent that the gradient of profile Heaviside function smears value, v represents regularization term regulating parameter, is arithmetic number.

The present invention successively decreases in the average improved and to introduce one and the onesize all 1's matrix S of described observed image in energy functional, is to ensure that the average of bound term and the improvement energy functional that successively decreases keeps same magnitude.

Utilize the active contour model of the improvement obtained to carry out model parameter calculation to remote sensing images, obtain the renewal equation of level set.

According to the active contour model of the improvement obtained, obtain c ₁, c ₂, b:

$c_1=\frac{\&Integral;(b*K_{\mathrm{\&sigma;}})\left(I\right(x)+index(x\left)\mathrm{\&epsiv;}\right(x\left)\right)u_{1}dy}{\&Integral;(b^2*K_{\mathrm{\&sigma;}})u_{1}dy+{\mathrm{\&omega;}}_1\&Integral;Sdy},$

$c_2=\frac{\&Integral;(b*K_{\mathrm{\&sigma;}})\left(I\right(x)+index(x\left)\mathrm{\&epsiv;}\right(x\left)\right)u_{2}dy}{\&Integral;(b^2*K_{\mathrm{\&sigma;}})u_{2}dy+{\mathrm{\&omega;}}_2\&Integral;Sdy},$

$b=\frac{\left\{\left(I\left(x\right)+index\left(x\right)\mathrm{\&epsiv;}\left(x\right)\right)\underset{i=1}{\overset{2}{\&Sigma;}}{c}_i{u}_i\right\}*K}{\left(\underset{i=1}{\overset{2}{\&Sigma;}}{c}_i^2{u}_i\right)*K}.$

Detailed process is: by the active contour model two ends of improvement that obtain simultaneously to c ₁, c ₂, b differentiate, when derivative meets with time, energy functional obtains minimal value.

The energy functional of the improvement obtained is substituted into formula respectively with can obtain:

Final acquisition c ₁, c ₂, b expression formula be:

$c_1=\frac{\&Integral;(b*K_{\mathrm{\&sigma;}})\left(I\right(x)+index(x\left)\mathrm{\&epsiv;}\right(x\left)\right)u_{1}dy}{\&Integral;(b^2*K_{\mathrm{\&sigma;}})u_{1}dy+{\mathrm{\&omega;}}_1\&Integral;Sdy},$

$c_2=\frac{\&Integral;(b*K_{\mathrm{\&sigma;}})\left(I\right(x)+index(x\left)\mathrm{\&epsiv;}\right(x\left)\right)u_{2}dy}{\&Integral;(b^2*K_{\mathrm{\&sigma;}})u_{2}dy+{\mathrm{\&omega;}}_2\&Integral;Sdy},$

$b=\frac{\left\{\left(I\left(x\right)+index\left(x\right)\mathrm{\&epsiv;}\left(x\right)\right)\underset{i=1}{\overset{2}{\&Sigma;}}{c}_i{u}_i\right\}*K}{\left(\underset{i=1}{\overset{2}{\&Sigma;}}{c}_i^2{u}_i\right)*K},$

Wherein, * convolution algorithm is represented, represent Heaviside function.

Step 502, according to gradient descent flow method, obtains representing that level set is measured over time and is:

Gradient descent flow method is an optimization algorithm, usually also referred to as method of steepest descent.Method of steepest descent solves one of the simplest and the most ancient method of unconstrained optimization problem, and many efficient algorithms all carry out improving and revising and obtain based on it.Method of steepest descent negative gradient direction is the direction of search, and method of steepest descent is more close to desired value, and step-length is less, advances slower.

The renewal equation obtaining level set is:

According to the renewal equation of the level set obtained in above-mentioned, make profile automatically expand or shrink, before and after profile, the absolute value of adjacent twice iteration difference is less than specified value, and contour motion stops; At this moment contour curve maintenance is stablized constant, completes the detection in shore line.

Concrete, repeat according to the active contour model improved to above-mentioned steps, until γ is less constant, can preset.Such as, maximum iteration time 100 times or level set curvilinear motion amount are less than 2 as stopping criterion for iteration.The Detection results figure of the coastline Detection Method method that Fig. 4 a-c provides for the embodiment of the present invention.Can find out that the method adopting the present embodiment to provide is to the initial profile of Landsat remote sensing images and check result from Fig. 4 a.Can find out that the method adopting the present embodiment to provide is to the initial profile of Radarsat remote sensing images and check result from Fig. 4 b.Can find out that the method adopting the present embodiment to provide is to the initial profile of Spot remote sensing images and testing result from Fig. 4 c.

The coastline Detection Method method based on region active contour model that the present embodiment provides, for remote sensing images sea even density, the situation of land Density inhomogeneity, the present invention is based on region active contour model feature, have and do not affect by initial profile, solved the problem of the puppet point interference that existing region active contour model segmentation remote sensing images occur simultaneously by pseudo-some minimizing technology, make shore line clear-cut unique; And the present invention adopts regional average value diminishing method can eliminate the problem in the wrong point shore line of existing method.The present invention can solve the coastline Detection Method of optical imagery, is also applicable to meeting SAR image coastline Detection Method.The present invention does not need artificial participation, can realize automatic detection.

The above; be only the present invention's preferably embodiment; but protection scope of the present invention is not limited thereto; anyly be familiar with those skilled in the art in the technical scope that the present invention discloses; be equal to according to technical scheme of the present invention and inventive concept thereof and replace or change, all should be encompassed within protection scope of the present invention.

Claims (9)

1., based on a coastline Detection Method method for region active contour model, it is characterized in that there are following steps:

-detect doubtful pseudo-point in remote sensing images to be analyzed; To the pseudo-point confirmed after testing, use this pseudo-neighborhood of a point pixel value and pseudo-point described in pseudo-mathematic interpolation correction correction of putting pixel value, obtain revised remote sensing images;

-revised remote sensing images are imported average to successively decrease in energy functional model, and the average be improved is successively decreased energy functional; According to the energy functional of described improvement, regional average value diminishing method is adopted to generate the active contour model improved;

-use the active contour model of described improvement to calculate the parameter of the revised remote sensing images of described puppet point, obtain level set renewal equation;

-according to described level set renewal equation, make the active contour model of described improvement automatically expand or shrink, when before and after profile, the absolute value of adjacent twice iteration difference is less than specified value, contour motion stops; Now contour curve maintenance is stablized constant, completes the detection in shore line.

2. to go a kind of coastline Detection Method method based on region active contour model described in 1 according to right, be further characterized in that the step of described " detecting the pseudo-point in remote sensing images to be analyzed " is specific as follows:

-to the doubtful pseudo-point in remote sensing images to be analyzed, contrast this doubtful pixel value of puppet point and pixel average in the multiple homalographic region of this puppet point periphery; The pixel value of puppet point as doubtful in this is less than periphery arbitrary region pixel average, then judge that this doubtful puppet point is as pseudo-point;

Described doubtful puppet point is the darker point of some and surrounding contrast's color that remote sensing images land part exists.

3. a kind of coastline Detection Method method based on region active contour model according to claim 2, is further characterized in that described " multiple homalographic region " is:

Four areas are the square subregion of α × α the area of composition is the square area W of 2 α × 2 α _x; Wherein, x represents described doubtful pseudo-point, i.e. central point, and α represents the positive integer of image window size, is the distance between two neighbors, value between 1 to 10;

Detect whether doubtful puppet point is pseudo-point according to following formula:

Wherein, μ ^kx () represents the pixel average of a kth sub-square area, Index (x) represents detective operators, and I (x) represents the pixel value of described central point/doubtful puppet point; When the pixel value of the described central point on observed image is less than the minimum value in the pixel average in four regions built centered by described central point, Index (x)=1, described central point is pseudo-point; Otherwise Index (x)=0, described central point is not pseudo-point.

4. a kind of coastline Detection Method method based on region active contour model according to claim 3, is further characterized in that described " using this pseudo-neighborhood of a point pixel value and pseudo-point described in pseudo-mathematic interpolation correction correction of putting pixel value " specifically comprises the steps:

-by formula D ^k(x)=| I (x)-μ ^k(x) |, k=1,2,3,4 computing parameters

Wherein, represent the absolute value of the pixel value of described central point and the difference of a kth sub-square window pixel average;

-by formula ε (x)=μ ^k(x)-I (x), k=argmin _k(D ^k(x), k=1,2,3,4) determine μ ^kx correction that () puts as described puppet with the minor increment of I (x);

Wherein, ε (x) represents the correction of pseudo-some x; Calculate in four square area when when value is minimum, the value of k; After k value is got and is determined, computing formula first half, namely minor increment is the average in a kth region and the difference of central point pixel value;

Correction described in-use, the correction of pseudo-point is added to the pixel value of former puppet point, revises the described pseudo-point through confirming.

5. a kind of coastline Detection Method method based on region active contour model according to claim 4, is further characterized in that the revised remote sensing images I of definition ^newx the expression formula of () is:

I ^new(x)＝I(x)+index(x)ε(x)

The pass of revised remote sensing images and initial remote sensing images is:

I(x)+index(x)ε(x)≈b(y)c _i+n(x)

Wherein, b (y) indicates inclined field, c _irepresent image-region average, n (x) represents picture noise..

6. a kind of coastline Detection Method method based on region active contour model according to claim 5, is further characterized in that described the average of the improvement " successively decrease energy functional " is specially:

Wherein, E represents energy functional, and y represents that image is in x neighborhood of a point pixel, K _σrepresent gaussian kernel, c ₁represent land area average, c ₂represent the pixel average of sea region, outside (C) represents outline line perimeter, and inside (C) represents outline line interior zone.

7. a kind of coastline Detection Method method based on region active contour model according to claim 6, is further characterized in that the energy functional of described improvement is:

Wherein, ω ₁, ω ₂represent weight coefficient, and meet 0≤ω ₁< 1,0≤ω ₂< 1, E ^mILCrepresent the energy functional of average, represent the first bound term that average is successively decreased, represent that average successively decreases second successively decreases bound term, c ₁and c ₂represent the regional average value in Liang Ge region, land, image sea respectively.

8. a kind of coastline Detection Method method based on region active contour model according to claim 1, is further characterized in that the active contour model of described improvement is:

Wherein, represent that the average with model parameter is successively decreased energy functional;

represent regularization term, represent Heaviside function,

μ represents regularization term regulating parameter, is arithmetic number, represent profile gradients, represent that the gradient of profile Heaviside function smears value, v represents regularization term regulating parameter, is arithmetic number.

9. a kind of coastline Detection Method method based on region active contour model according to claim 8, be further characterized in that described step "-use the active contour model of described improvement calculate described puppet point revised remote sensing images parameter, obtain level set renewal equation; " specifically comprise the steps:

The active contour model of the improvement that-basis obtains, obtains c ₁, c ₂, b:

Wherein, * convolution algorithm is represented, represent Heaviside function;

-according to gradient descent flow method, obtain representing that level set is measured over time and be:

Wherein, e _i(x)=∫ K (y-x) (I (x)+index (x) ε (x)-b (y) c _i) ²dy, wherein represent the variable quantity of level set, represent Heaviside function derivative, e _irepresent one of front two differentiate results about level set on the right side of step 402 equal sign, K (y-x) is pixel y and x distance gaussian kernel, v and μ is regulating parameter, is arithmetic number. represent the curvature of level set, for item second from the bottom on the right side of step 402 equal sign is for the derivative of level set, for item last on the right side of step 402 equal sign is for the derivative of level set;

-the renewal equation that obtains level set is:

Wherein represent the level set of n-th iteration, represent the level set of (n-1)th iteration.