CN102163325A - Edge feature detection method of multi-spectral image - Google Patents

Abstract

The invention discloses an edge feature detection method of a multi-spectral image, which comprises the following steps: using a first-order differential operator to respectively compute the gradients of N wave bands of the multi-spectral image in horizontal and vertical directions; adding the square of gradients in the horizontal and vertical directions one phase element by one phase element to obtain the feature elements in the horizontal and vertical directions; adding the multiplication of the gradients in the horizontal and vertical directions to obtain the feature element in two directions and to form a feature matrix; and computing a feature value of the feature matrix one phase element by one phase element, and computing a gradient value of the image to obtain the edge feature of the multi-spectral image. The gradient information at each wave band is used for obtaining the edge feature of the multi-spectral image in an intuitive and rapid manner. While detecting, all wave bands can be selected to compute the edge feature, or a plurality of optimum wave bands can be selected according to the application requirement to compute the edge feature; and the method which directly uses the multi-spectral information to detect the edge feature is more rational and efficient in comparison with a method which uses an algebraic operation to synthesize the gradient information of a plurality of wave bands.

Description

A kind of multispectral image edge feature detection method

Technical field

The present invention relates to a kind of image processing method, particularly a kind of multispectral image edge feature detection method.

Background technology

Edge of image has been concentrated the most information of image, and the detection of edge feature is one of core content of Digital Image Processing, and its accuracy of detection directly has influence on the effect of follow-up image recognition and understanding.Image can roughly be divided into gray level image, coloured image and multispectral image three classes.

At present, the gray-scale Image Edge characteristic detection method mainly depends on the single order or the second-order differential computing of image, the former differentiates according to level, vertical or crisscross single order and obtains the gradient feature of image, as Roberts operator, Prewitt operator, Sobel operator etc.; The latter determines the position at edge according to the zero cross point of second-order differential computing, as the Laplacian operator, and LoG operator etc.Coloured image has three components of RGB, and its rim detection mainly depends on color spatial alternations such as IHS (I refers to brightness, and H refers to tone, and S refers to saturation degree) conversion, strengthens the acquisition edge feature by the luminance component after the conversion being carried out the edge.

A main difference part of multispectral image and gray level image, coloured image is that it often has the multi light spectrum hands more than three, therefore is difficult to utilize the method for colour space transformation to detect edge feature.The multispectral image edge feature detects does not at present still have good solution, method is at first to use gradient operator respectively each wave band to be detected comparatively intuitively, then by carrying out dimension-reduction treatment with algebraically computings such as, mould, maximal value, minimum value or principal component transform.Such dimension-reduction treatment lacks physical basis, and because the atural object border on the multispectral image at different-waveband and inconsistent, must make edge detection results bigger uncertainty occur.

Summary of the invention

Goal of the invention: at the problem and shortage of above-mentioned existing existence, the purpose of this invention is to provide a kind of dimension-reduction treatment that need not, directly utilize the multispectral image edge feature detection method of multispectral information calculations edge feature.

Technical scheme: for achieving the above object, the technical solution used in the present invention is a kind of multispectral image edge feature detection method, comprises following steps:

(1) gradient G of the horizontal direction of N wave band of calculating multispectral image _IxGradient G with vertical direction _Iy, wherein N is a natural number, i=1, and 2,3 ..., N;

(2), obtain the characteristic element G of horizontal direction by the gradient summed square of pixel with the horizontal direction of N wave band in the described step (1) _Xx,, obtain the characteristic element G of vertical direction by the gradient summed square of pixel with the vertical direction of N wave band in the described step (1) _Yy, by the gradient G of pixel with the horizontal direction of N wave band in the described step (1) _IxGradient G with vertical direction _IyThe product addition, obtain the characteristic element G of level and vertical twocouese _XyAnd G _Yx, by G _Xx, G _Yy, G _Xy, G _YxComposition characteristic matrix G (x, y);

(3) calculate eigenmatrix G (x, eigenvalue y) in the described step (2) by pixel ₊With λ _-

(4) utilize the eigenvalue that obtains in the step (3) ₊With λ _-(x y), thereby obtains the edge feature of multispectral image to the Grad g of computed image.

In the described step (1), first order differential operator is the simplest first order difference, and is convenient and swift.

Described wave band can be all wave bands of described image, also can select the best wave band of part according to application purpose.As when detecting the edge of vegetation, can select vegetation indication the most responsive ruddiness and near-infrared band are detected.

The eigenvalue of calculating in the described step (3) ₊With λ _-Indicated the amplitude at multispectral edge.

(x, method y) can be for calculating described eigenvalue for the Grad g of computed image in the described step (4) ₊With λ _-The square root of difference, can also be for calculating described eigenvalue ₊With λ _-And or other algebraic operation such as ratio.

Beneficial effect: the present invention has utilized the gradient information of each wave band simultaneously, with intuitively, mode obtains the edge feature of multispectral image fast.Both can select all wave band edge calculation features during detection, also can select only certain several wave band edge calculation feature according to application need.This method of directly utilizing multispectral information to carry out the edge feature detection, more reasonable more and efficient than the multiwave gradient information of method synthesis with algebraic operation.

Description of drawings

Fig. 1 is a process flow diagram of the present invention;

Fig. 2 is high-resolution QuickBird satellite multispectral image;

Fig. 3 is a result schematic diagram of the present invention of utilizing computer programming language to realize.

Embodiment

Below in conjunction with the drawings and specific embodiments, further illustrate the present invention, should understand these embodiment only is used to the present invention is described and is not used in and limit the scope of the invention, after having read the present invention, those skilled in the art all fall within the application's claims institute restricted portion to the modification of the various equivalent form of values of the present invention.

Basic ideas of the present invention are: if regard the multispectral data of each pixel as a vector, then multispectral image can be considered as a two dimension, multiattribute vector field, thereby the detection problem on atural object border is converted to the neighborhood difference problem of hypersurface, but the first fundamental form in the application of differential geometry is found the solution.

Make I (x, y): R ²→ R ^NThe multispectral image of representing a width of cloth N wave band, any one wave band wherein can be expressed as I _i(x, y): R ²→ R, i=1 ..., N.Then image I is at arbitrfary point (x ₀, y ₀) on value can regard R as ^NA vector in space, so any adjacent 2 P=(x on the image ₀, y ₀) and Q=(x ₁, y ₁) between variation can be expressed as:

ΔI＝I(P)-I(Q)

When between 2 of P, the Q apart from d (P, when Q) being tending towards infinitely small, can represent with differential form:

$\mathrm{dI}=\frac{\∂I}{\∂x}\mathrm{dx}+\frac{\∂I}{\∂y}\mathrm{dy}$

Accordingly, its second-order differential form is:

${\mathrm{dI}}^2={\left(\begin{array}{c}\mathrm{dx}\\ \mathrm{dy}\end{array}\right)}^T\left(\begin{array}{cc}{\left(\frac{\∂I}{\∂x}\right)}^2& \frac{\∂I}{\∂x}\·\frac{\∂I}{\∂y}\\ \frac{\∂I}{\∂y}\·\frac{\∂I}{\∂x}& {\left(\frac{\∂I}{\∂y}\right)}^2\end{array}\right)\left(\begin{array}{c}\mathrm{dx}\\ \mathrm{dy}\end{array}\right)={\left(\begin{array}{c}\mathrm{dx}\\ \mathrm{dy}\end{array}\right)}^T\left(\begin{array}{cc}{G}_{\mathrm{xx}}& G_{\mathrm{xy}}\\ G_{\mathrm{yx}}& G_{\mathrm{yy}}\end{array}\right)\left(\begin{array}{c}\mathrm{dx}\\ \mathrm{dy}\end{array}\right)$

Following formula just is called first fundamental form, and it has reflected the variable gradient of spectrum vector in the multispectral image.By G _Xx, G _Xy, G _Yx, G _YyFormed one 2 * 2 eigenmatrix, the proper vector of this matrix has determined the direction of gradient, feature λ ₊Value has determined the amplitude of gradient.Eigenwert can be simplified calculating with following formula:

${\mathrm{\λ}}_{\±}=\frac{G_{\mathrm{xx}}+G_{\mathrm{yy}}\±\sqrt{{(G_{\mathrm{xx}}-G_{\mathrm{yy}})}^2+{4G}_{\mathrm{xy}}^2}}{2}$

Like this, the edge feature of multispectral image just can pass through function f=f (λ ₊, λ _-), i.e. λ ₊And λ _-Algebraic operation realize detecting, as:

$f=\sqrt{{\mathrm{\λ}}_{+}-{\mathrm{\λ}}_{-}}$

As shown in Figure 1,, calculate the gradient of each wave band respectively, promptly utilize the difference of horizontal direction and vertical direction to calculate the pending multispectral image that N wave band arranged.

Utilize the gradient G of the first order difference calculated level direction of horizontal direction _Ix:

G _ix＝f _i(x+1，y)-f _i(x，y)

Utilize the gradient G of the first order difference calculating vertical direction of vertical direction _Iy:

G _iy＝f _i(x，y+1)-f _i(x，y)

After obtaining the horizontal gradient and VG (vertical gradient) of N wave band,, obtain G by the gradient summed square of pixel with the horizontal direction of each wave band _Xx:

$G_{\mathrm{xx}}=\underset{i=1}{\overset{N}{\mathrm{\Σ}}}{G}_{ix}^2(x,y)$

By the gradient summed square of pixel, obtain G with the vertical direction of each wave band _Yy:

$G_{\mathrm{yy}}=\underset{i=1}{\overset{N}{\mathrm{\Σ}}}{G}_{\mathrm{iy}}^2(x,y)$

By the product addition of pixel, obtain G with the gradient of the horizontal direction of each wave band and vertical direction _XyAnd G _Yx:

$G_{\mathrm{xy}}=G_{\mathrm{yx}}=\underset{i=1}{\overset{N}{\mathrm{\Σ}}}{G}_{\mathrm{ix}}(x,y)G_{\mathrm{iy}}(x,y)$

At last by G _Xx, G _Yy, G _Xy, G _YxComposition characteristic matrix G (x, y):

$G(x,y)=\left(\begin{array}{cc}{G}_{\mathrm{xx}}& G_{\mathrm{xy}}\\ G_{\mathrm{yx}}& G_{\mathrm{yy}}\end{array}\right)$

Utilize following formula to pursue the pixel simplification and find the solution eigenmatrix G (x, eigenvalue y) ₊:

${\mathrm{\λ}}_{\±}=\frac{G_{\mathrm{xx}}+G_{\mathrm{yy}}\±\sqrt{{(G_{\mathrm{xx}}-G_{\mathrm{yy}})}^2+{4G}_{\mathrm{xy}}^2}}{2}$

Calculate the square root of the difference of two eigenwerts by pixel, as the Grad g of this pixel (x, y):

$g(x,y)=\sqrt{{\mathrm{\λ}}_{+}-{\mathrm{\λ}}_{-}}$

Thereby obtain the edge feature of multispectral image.

An example of the present invention realizes that on the PC platform through experimental verification, this multispectral image rim detection flow process can obtain comparatively ideal testing result, and the edge accuracy is higher.As shown in drawings, Fig. 2 is high-resolution QuickBird satellite multispectral image, Fig. 3 is the result of the present invention who utilizes computer programming language to realize, the edge feature of detection is clear, and the profile of main atural object such as house, vegetation, road and the consistance of original image are better.

Claims (3)

1. multispectral image edge feature detection method is characterized in that comprising following steps:

(1) utilize first order differential operator to calculate the gradient G of the horizontal direction of a multispectral image N wave band respectively _IxGradient G with vertical direction _Iy, wherein N is a natural number, i=1, and 2,3 ..., N;

(2), obtain the characteristic element G of horizontal direction by the gradient summed square of pixel with the horizontal direction of N wave band in the described step (1) _Xx,, obtain the characteristic element G of vertical direction by the gradient summed square of pixel with the vertical direction of N wave band in the described step (1) _Yy, by the gradient G of pixel with the horizontal direction of N wave band in the described step (1) _IxGradient G with vertical direction _IyThe product addition, obtain the characteristic element G of level and vertical twocouese _XyAnd G _Yx, by G _Xx, G _Yy, G _Xy, G _YxComposition characteristic matrix G (x, y);

(3) calculate eigenmatrix G (x, eigenvalue y) in the described step (2) by pixel ₊And λ _-

(4) utilize the eigenvalue that obtains in the step (3) ₊With λ _-(x y), thereby obtains the edge feature of multispectral image to the Grad g of computed image.

2. according to the described a kind of multispectral image edge feature detection method of claim 1, it is characterized in that: all wave bands that described wave band is described image or the subband of selection.

3. according to the described a kind of multispectral image edge feature detection method of claim 1, it is characterized in that: (x, method y) is for calculating described eigenvalue for the Grad g of computed image in the described step (4) ₊With λ _-Difference square root or calculate described eigenvalue ₊With λ _-And or ratio.

Images