CN104794723A - Remote-sensing image building location detection method based on probability - Google Patents

Abstract

The invention relates to a remote-sensing image building location detection method based on probability. The remote-sensing image building location detection method includes the steps of (1) performing smoothing filter to acquire the image 2; (2) constructing a controllable filter; (3) convoluting; (4) binaryzating; (5) extracting structure features of twelve directions; (6) classifying straight lines and curves; (7) calculating probability of central points of candidate buildings by means of probability density functions; and (8) defining thresholds to determine location of the central points of the buildings. By determining the location of the central points of the buildings according to the probability density functions, the problem of low building-location detection efficiency caused by the conventional methods which adopt modeling and classifiers and the like is solved, accuracy and efficiency of building detection are both improved, and the effect of full automation is achieved.

Description

A kind of remote sensing image building method for detecting position based on probability

Technical field

The present invention relates to a kind of remote sensing image process field, is a kind of remote sensing image building method for detecting position based on probability specifically.

Background technology

Utilize remote sensing image to detect to carry out buildings change, have the impact of several factors, the difficulty that these changes all determining buildings detect.Influence factor mainly contains the following aspects:

(1) different shooting angle: different shooting angle to cause in aerial image the projections such as atural object to change.

(2) different illumination condition: because the time of image taking is different, therefore different during imaging illumination conditions can cause the difference of the different atural object intensity profile of image, and meanwhile, different sun irradiation angles also can make shadow-casting direction change.

(3) the different moment in season: the simplest example be exactly winter in summer vegetation distribution---summer dense vegetation, relatively dense trees may block buildings, and winter then can not.

(4) some artificial factors: such as, the buildings of non-recurring structure change may be judged to be the change etc. of the change of recurring structure as roof gray scale due to radiation characteristic by different people.

Summary of the invention

The invention provides a kind of remote sensing image building method for detecting position based on probability, utilize the position of probability density function determination buildings, without the need to the support in side information data storehouse, automaticity is high.

The technical scheme adopted for realizing target of the present invention is: method comprises the following steps:

Step 1: to the smoothing filtering of the panchromatic remote sensing image image1 of input, obtain image2;

Step 2: structure controllable filter;

Step 3: carrying out convolution with controllable filter and image2, obtaining the architectural feature with the highest response perpendicular on controllable filter direction, obtaining image3;

Step 4: split image3, obtains binary image image4;

Step 5: the architectural feature SF extracting 12 directions in image4;

Step 6: classification architectural feature SF being carried out to straight line and curve;

Step 7: with the probability of probability density function calculated candidate buildings central point;

Step 8: definition threshold value, determines buildings center position.

The method of described smothing filtering is for adopting bi-linear filter.

Described controllable filter carries out linear combination by two basis filters, and wherein two basis filters ask local derviation to obtain respectively by both direction:

$G_{p0}=\frac{\∂}{\∂x}G(x,y)=-2x{e}^{-(x^2+y^2)}$

$G_{p\frac{\mathrm{\π}}{2}}=\frac{\∂}{\∂y}G(x,y)=-2y{e}^{-(x^2+y^2)}$

Controllable filter $G_{\mathrm{p\θ}}=\cos \left(\mathrm{\θ}\right)G_{p0}+\sin \left(\mathrm{\θ}\right)G_{p\frac{\mathrm{\π}}{2}}.$

Described dividing method is: carry out binaryzation using 0.8 of the maximal value of image3 times as threshold value, and utilizes 8 neighborhood concatenate rules to process binary image, deletes short-term.

The architectural feature SF in 12 described directions respectively value is: θ=[0, π/12 ..., 11 π/12].

Described straight line and the sorting technique of curve are: choose each architectural feature SF, and along architectural feature SF search, extract mid point, its coordinate is designated as (x _m, y _m); Join two endpoints and make straight line, the middle point coordinate of note straight line is (x ₀, y ₀), ask (x _m, y _m) and (x ₀, y ₀) between Euclidean distance, if be 0, be then be judged to be straight line, if be greater than 0, be then judged to be curve.

Described the formula of probability density function is:

$p_b(x,y)=\frac{1}{12}\underset{i=1}{\overset{12}{\mathrm{\Σ}}}\frac{1}{\sqrt{2\mathrm{\π}{\mathrm{\σ}}_0}}{e}^{-\frac{{(x-x_{\mathrm{ic}})}^2+{(y-y_{\mathrm{ic}})}^2}{2{\mathrm{\σ}}_0}}$

Wherein, (x _ic, y _ic) be the middle point coordinate of the architectural feature in i-th direction, σ ₀for the weights of architectural feature, when architectural feature is straight line, σ ₀=0.4; When architectural feature is curve, σ ₀=1.

The threshold value of described determination buildings center position is set as 40% of probability density function maximal value, and the coordinate being greater than threshold value is defined as the center of buildings.

The invention has the beneficial effects as follows: solve the problem utilizing the method such as modeling and sorter to cause buildings position detection efficiency low in classic method, be conducive to the accuracy and efficiency improving buildings detection, reach full automatic effect.

Accompanying drawing explanation

Fig. 1 is overall process flow figure of the present invention.

Embodiment

The specific embodiment of the present invention is described in detail below in conjunction with accompanying drawing.

In step 101, the pending remote sensing image image1 of input is Quick bird panchromatic image, is of a size of 1000 × 1000, and carries out the pre-service such as radiant correction and geometry correction.

In step 102, adopt bi-linear filter to the smoothing filtering of image1, obtain image2.

In step 103, utilize two basis filters to carry out linear combination structure controllable filter, wherein two basis filters ask local derviation to obtain respectively by both direction:

$G_{p0}=\frac{\∂}{\∂x}G(x,y)=-2x{e}^{-(x^2+y^2)}$

$G_{p\frac{\mathrm{\π}}{2}}=\frac{\∂}{\∂y}G(x,y)=-2y{e}^{-(x^2+y^2)}$

Controllable filter $G_{\mathrm{p\θ}}=\cos \left(\mathrm{\θ}\right)G_{p0}+\sin \left(\mathrm{\θ}\right)G_{p\frac{\mathrm{\π}}{2}}.$

In step 104, the controllable filter G utilizing step 103 to obtain _{p θ}convolution is carried out, perpendicular to filter direction obtaining the architectural feature with the highest response to processing the image2 obtained in step 102.

In step 105, carry out binaryzation using 0.8 of the maximal value of image3 times as threshold value, and utilize 8 neighborhood concatenate rules to process binary image, delete the line that length is less than 20.

In step 106, utilize the controllable filter G of structure in step 103 _{p θ}, another θ=0 respectively, π/12 ..., 11 π/12 are totally 12 directions, corresponding extraction architectural feature.

In step 107, utilize step 106 to extract architectural feature, for one of them architectural feature, sorting technique is as follows: along each architectural feature SF search, extract mid point, its coordinate is designated as (x _m, y _m); Join two endpoints and make straight line, the middle point coordinate of note straight line is (x ₀, y ₀), ask (x _m, y _m) and (x ₀, y ₀) between Euclidean distance, if be 0, be then be judged to be straight line, if be greater than 0, be then judged to be curve.

In step 108, utilize probability density function:

$p_b(x,y)=\frac{1}{12}\underset{i=1}{\overset{12}{\mathrm{\Σ}}}\frac{1}{\sqrt{2\mathrm{\π}{\mathrm{\σ}}_0}}{e}^{-\frac{{(x-x_{\mathrm{ic}})}^2+{(y-y_{\mathrm{ic}})}^2}{2{\mathrm{\σ}}_0}}$

Computing center's point parameter probability valuing.Wherein, (x _ic, y _ic) be the middle point coordinate of the architectural feature in i-th direction, σ ₀for the weights of architectural feature, when architectural feature is straight line, σ ₀=0.4; When architectural feature is curve, σ ₀=1.

In step 109, export buildings center position.

Claims (8)

1., based on a remote sensing image building method for detecting position for probability, it is characterized in that comprising the following steps:

Step 1: to the smoothing filtering of the panchromatic remote sensing image image1 of input, obtain image2;

Step 2: structure controllable filter;

Step 3: carrying out convolution with controllable filter and image2, obtaining the architectural feature with the highest response perpendicular on controllable filter direction, obtaining image3;

Step 4: split image3, obtains binary image image4;

Step 5: the architectural feature SF extracting 12 directions in image4;

Step 6: classification architectural feature SF being carried out to straight line and curve;

Step 7: with the probability of probability density function calculated candidate buildings central point;

Step 8: definition threshold value, determines buildings center position.

2. a kind of remote sensing image building method for detecting position based on probability according to claim 1, is characterized in that the method for smothing filtering is for adopting bi-linear filter.

3. a kind of remote sensing image building method for detecting position based on probability according to claim 1, it is characterized in that controllable filter carries out linear combination by two basis filters, wherein two basis filters ask local derviation to obtain respectively by both direction:

$G_{p0}=\frac{\∂}{\∂x}G(x,y)=-2x{e}^{-(x^2+y^2)}$

$G_{p\frac{\mathrm{\π}}{2}}=\frac{\∂}{\∂x}G(x,y)=-2y{e}^{-(x^2+y^2)}$

Controllable filter $G_{\mathrm{p\θ}}=\cos \left(\mathrm{\θ}\right)G_{p0}+\sin \left(\mathrm{\θ}\right)G_{p\frac{\mathrm{\π}}{2}}.$

4. a kind of remote sensing image building method for detecting position based on probability according to claim 1, it is characterized in that dividing method is: carry out binaryzation using 0.8 of the maximal value of image3 times as threshold value, and utilize 8 neighborhood concatenate rules to process binary image, delete short-term.

5. a kind of remote sensing image building method for detecting position based on probability according to claim 1, the architectural feature SF that it is characterized in that 12 directions respectively value is: θ=[0, π/12 ..., 11 π/12].

6. a kind of remote sensing image building method for detecting position based on probability according to claim 1, it is characterized in that the sorting technique of straight line and curve is: choose each architectural feature SF, along architectural feature SF search, extract mid point, its coordinate is designated as (x _m, y _m); Join two endpoints and make straight line, the middle point coordinate of note straight line is (x ₀, y ₀), ask (x _m, y _m) and (x ₀, y ₀) between Euclidean distance, if be 0, be then be judged to be straight line, if be greater than 0, be then judged to be curve.

7. a kind of remote sensing image building method for detecting position based on probability according to claim 1, is characterized in that the formula of probability density function is:

$p_b(x,y)=\frac{1}{12}\underset{i=1}{\overset{12}{\mathrm{\Σ}}}\frac{1}{\sqrt{2\mathrm{\π}{\mathrm{\σ}}_0}}{e}^{-\frac{{(x-x_{\mathrm{ic}})}^2+{(y-y_{\mathrm{ic}})}^2}{{2\mathrm{\σ}}_0}}$

Wherein, (x _ic, y _ic) be the middle point coordinate of the architectural feature in i-th direction, σ ₀for the weights of architectural feature, when architectural feature is straight line, σ ₀=0.4; When architectural feature is curve, σ ₀=1.

8. a kind of remote sensing image building method for detecting position based on probability according to claim 1, it is characterized in that determining that the threshold value of buildings center position is set as 40% of probability density function maximal value, the coordinate being greater than threshold value is defined as the center of buildings.