CN103699900B - Building horizontal vector profile automatic batch extracting method in satellite image - Google Patents
Building horizontal vector profile automatic batch extracting method in satellite image Download PDFInfo
- Publication number
- CN103699900B CN103699900B CN201410003206.XA CN201410003206A CN103699900B CN 103699900 B CN103699900 B CN 103699900B CN 201410003206 A CN201410003206 A CN 201410003206A CN 103699900 B CN103699900 B CN 103699900B
- Authority
- CN
- China
- Prior art keywords
- building
- image
- pixel
- area
- profile
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
Landscapes
- Image Analysis (AREA)
- Image Processing (AREA)
Abstract
The invention provides building horizontal vector profile automatic batch extracting method in a kind of satellite image, carry out image classification first with K mean algorithm, obtain the trunk portion of building, take each construction zone barycenter and solve the On The Choice of initial seed point.After growing all seed points regions, by image edge information, construction zone is separated with peripheral region, and get rid of non-construction zone according to characteristic parameters such as rectangular degree, banding indexes, thus realize automatically extracting of building horizontal pixel profile.Then utilizing the technology such as Hough transform, block process that horizontal pixel profile carries out Line vectorization process, final batch obtains the horizontal vector profile of owned building.Present invention top view be applicable to satellite image is the batch rapid extraction of the horizontal vector profile of linear section structure polygonal common building thing.
Description
Technical field
The present invention relates to a kind of from single width groups of building satellite image automatic batch extract building horizontal vector profile
Method, is particularly that the automatic batch of the horizontal vector profile of the polygonal building of linear structure extracts to top view.
Background technology
Utilizing Mono-satellite image to realize three-dimensional scenic Virtual Reconstruction is a research topic the most active, and it is mainly applied
At aspects such as physical construction planning, military scene simulation, resource management, earthquake relief work simulations.Three-dimensional at real scene
In reconstruction, the overwhelming majority is linear section structure polygonal common building thing for simple in construction, top view, how to realize this kind of greatly
The rapid modeling of the common building thing that amount exists is the key efficiently rebuilding three-dimensional building group, and how to realize from groups of building image
It is the basis efficiently rebuilding three-dimensional building group that middle automatic batch ground extracts the horizontal profile of building, and decides following reconstruction
Virtual scene in the matching degree of Distribution Pattern and real scene of groups of building.
In current research, for different application purposes, there has been proposed various building image border and carry
Taking algorithm, relatively common has snake modelling, level set curve evolvement method, region-growing method etc..The research of above method is right
As being gray level image, it is not suitable for coloured image, therefore cannot utilize colouring information abundant in coloured image.Except this it
Outward, snake modelling is sensitive to initial position, needs to rely on other mechanism and initial profile is placed on characteristics of image interested
Near, otherwise contours extract can be failed, and the manual way chosen of employing is to arrange initial boundary the most mostly, the most numerous
Trivial, and automatically generate to contour line and cause difficulty.And more conventional region-growing method also exists 2 deficiencies: first,
It it is the On The Choice of initial seed point.It is manually to choose that seed points is chosen great majority by current method, need substantial amounts of manually
Interfering, time and effort consuming, efficiency is the lowest.Second, it is the On The Choice of growing threshold.Growing threshold is excessive it would appear that over-segmentation,
The target area area i.e. generated is often big than real area;And the too small segmentation deficiency that can cause of growing threshold, i.e. target area
Territory growth is imperfect.Therefore different buildings need to choose different growing thresholds, and this job demand is artificial one by one according to warp
Test.It addition, snake modelling and region-growing method are directed to single target region, namely can be only generated one every time
Target area, result in the poor efficiency of method.
The edge contour that above method is extracted, is formed by pixel, referred to herein as pixel profile.This patent research building
The purpose of the horizontal contours extract of thing, is to rebuild for the three-dimensional of follow-up groups of building to provide modeling data, needs extracting
Building pixel profile is made Line vectorization further and is processed, and the profile after being processed by Line vectorization here is referred to as vector wheel
Wide.Current contour vectorization method is to be extracted by edge pixel and form list, then this list is carried out straightway
Matching, is provided with a distance threshold, is defined as deviateing the ultimate range of straight line, after exceeding this distance threshold, straightway
Will be divided into two in proportion.The result of this method is chosen by distance threshold to be affected relatively greatly, in order to obtain preferable effect,
Each building in image is required for corresponding different distance threshold, and the method in the past arranging distance threshold is often according to warp
Testing manual setting, operating efficiency is low.
Summary of the invention
In order to overcome the deficiencies in the prior art, the present invention provides a kind of and quickly carries in batches from single width groups of building satellite image
Take the automatic mode of building horizontal pixel profile;It is further proposed that a kind of new choosing without distance threshold and other threshold values
Line vectorization automatic processing method, the building horizontal pixel profile extracted is carried out Line vectorization process, obtains
Meet the horizontal vector profile needed for follow-up building three-dimensional is rebuild.The method is particularly well-suited to top view in satellite image
For the batch rapid extraction of the horizontal vector profile of linear section structure polygonal common building thing, to taking photo by plane in groups of building image
This kind of groups of building equally applicable.
The technical solution adopted for the present invention to solve the technical problems is: carries out image first with K-mean algorithm and divides
Class, obtains the trunk portion of building, takes each construction zone barycenter and solves the On The Choice of initial seed point.Grow all
Behind seed points region, by image edge information, construction zone is separated with peripheral region, and according to rectangular degree, banding
The characteristic parameters such as index get rid of non-construction zone, thus realize automatically extracting of building horizontal pixel profile.Then profit
By technology such as Hough transform, block process, horizontal pixel profile being carried out Line vectorization process, final batch obtains all buildings
The horizontal vector profile of thing.
In the present invention, the initial point of image coordinate system is positioned at the image upper left corner, x-axis positive direction straight down, y-axis positive direction water
Put down to the right.Specifically comprise the following steps that
Groups of building satellite image is smoothed by step one, employing Gaussian filter, and described Gaussian smoothing is filtered
Variances sigma=0.8 of ripple device, window size is 7*7 pixel;Then, in Hsv color space, the tone of each pixel is kept not
Become, adjust saturation according to step value Δ s, adjust brightness according to step value Δ v, satellite image is carried out image enhancement processing,
Obtain source images Iorig, wherein Δ s is 0.06-0.08, and Δ v is 0.05-0.07;
Step 2, profile automatically extract, and comprise the following steps:
1. the automatically generating of different building seed points, comprises the following steps:
1) K-mean algorithm is utilized at Lab color space, to source images on a, b color subspace represent colouring information
IorigCarrying out built-up area regional partition, the class categories number that K-mean algorithm needs in performing is equal to image IorigAt a, b color
Peak value number in the two-dimensional histogram in space, is stored as I by the image being partitioned into buildingseg;
2) by image IsegBe converted to gray level image, then be binary image I by greyscale image transitionsbw, wherein, building
Region is white, and background is black, to binary image IbwCarry out lower column processing successively:
(1) median filter using window to be 5*5 pixel removes pixel isolated in image;
(2) use the square structure operator of 5*5 pixel to carry out opening and closing operation, remove area and cover less than square structure operator
The region of scope;
(3) area is removed less than sareaThe connected region of individual pixel, sareaSpan be whole in [700,2000]
Number;
(4) area is removed more than IareaThe connected region of individual pixel, IareaSpan be whole in [8000,10000]
Number;
3) construction zone is numbered respectively, and record its barycenter initial seed point as each building;If seed points
Position coordinates is (xk,yk), k is building numbering, xkAnd ykBe respectively kth building seed points line number in the picture and
Row number,I represents the line number of image, and j represents the row of image
Number, (i j) represents that (i, j) gray value of place's pixel, D represents UNICOM region, symbol to fRepresent downward rounding operation;
2. extract construction zone, comprise the following steps:
1) utilize region-growing method, generate the region of each candidate architecture thing, specifically comprise the following steps that
(1) one and source images I is set uporigThe same size, the image I of pixel grey scale all 0outIf growing threshold is
Thresh, its span is 8-10 pixel;
(2) an initial seed point (x is orientedk,yk), k is building numbering, arranges Iout(xk,yk)=1;
(3) with (xk,ykCentered by), it is considered to (xk,yk) four neighborhood territory pixel (xk-1,yk), (xk+1,yk), (xk,yk-1),
(xk,yk+ 1), in Lab color space, the colour-difference of pixel and seed points in seed points neighborhood is calculated respectively according to Euclidean distance Wherein, L (x, y), a (x, y), b (x,
Y) current pixel (x, L, a, b component value y), and L in neighborhood is represented respectivelyseed、aseed、bseedThen represent current seed point
L, a, b component value;If current pixel (x, I y) in neighborhoodout(x, y)=0, and d (x, y) < thresh then arranges Iout
(x, y)=1, simultaneously will (x, y) press-in storehouse;
(4) from storehouse, one pixel of ejection, as new seed points, is assigned to (xk,yk), return to step (3);Work as storehouse
During for sky, growth terminates;
(5) return to step (2), orient next seed points, repeat step (3) and (4), until all seed points scan
Completing, whole growth course terminates;
2) to source images IorigDetect that strong edge, the pixel on strong edge are white with Canny operator after gray processing, its
His pixel is black, and the image that this is contained strong marginal information is stored as Iedge;
3) traversing graph is as Iout, work as IoutIn pixel be in white construction zone time, check image IedgeMiddle correspondence
Whether the pixel of position is in the edge of white, the most then by IoutMiddle respective pixel becomes black;
4) at image IoutIn, calculate rectangular degreeLength-width ratioWith banding indexWherein A, l
It is area and the girth in each region respectively, Ae, h, the w area of corresponding each region minimum enclosed rectangle, length and width respectively;When
<when 5, determining that this region is building, the region being identified as building will remain, and otherwise will for d>0.7 and s<3 and r
This region becomes black;
Step 3, utilize the method for labelling connected region by image IoutIn single building split, store respectively
For Iout_k, k is building numbering;To the single building image I splitout_kDo and process as follows:
(1) at Iout_kIn, utilize Hough transform to detect the longest corresponding in building horizontal pixel profile straightway,
Main shaft p as building;With main shaft one end for fixed point, that horizontal pixel profile is rotated to main shaft and image coordinate system
Coordinate axes is parallel, and specific practice is as follows:
If theta angle is building main shaft p and the angle of x-axis positive direction in image coordinate system, just it is counterclockwise, single
Position is degree;When | cos (theta) | > cos (45 °), building horizontal pixel profile is rotated to main shaft and image coordinate system
X-axis parallel;When | cos (theta) |≤cos (45 °), building horizontal pixel profile is rotated to main shaft and image coordinate
The y-axis of system is parallel;
(2) by postrotational horizontal pixel profile by its extraneous rectangle piecemeal at equal intervals on x, y direction, every block size is
5*3 pixel, is defined as cell block, is denoted as unit;On the direction at main shaft place, cell block is defined as 3 pictures after rotation
Element, other direction is defined as 5 pixels;Last column and last string calculating by a cell block less than a cell block;
(3) at image Iout_kIn, from top to bottom, from left to right progressive scan each unit block, calculate in each unit block and build
Object area area and the percentage ratio of cell block areaWherein, AbuildIt it is construction zone in cell block
Area, AunitFor the area of cell block, square measure is number of pixels;When Percentage is more than or equal to 0.45, recognize
It is construction zone for this cell block, retains this cell block and be stuffed entirely with into white, otherwise abandoning this cell block;
(4) the linearization(-sation) profile obtained is gone back to home position around the fixed point rotary-theta angle adjustment of main shaft, obtain list
The linearization(-sation) profile of individual building;
(5) respectively all of construction zone is carried out the operation of above-mentioned steps (1)-(4), and each building after processing
Thing linearization(-sation) profile is merged in same image, and is stored as vectogram, is finally completed all in groups of building on raw video
The batch extracting of building horizontal vector profile.
The invention has the beneficial effects as follows: solve existing groups of building horizontal pixel contours extract based on satellite image and picture
Semi-automatic and the defect of manual methods during element profile Line vectorization, it is proposed that Intelligent treatment technology automatically.First,
Propose a kind of method automatically generating different building initial seed point, and utilize over-segmentation and the image edge of region growing
Information separation building and surrounding objects, obtain the horizontal pixel profile of building, overcomes in existing region-growing method, initially
Seed points and growing threshold need the drawback manually the most rule of thumb chosen.Secondly, for meeting what follow-up three-dimensional was rebuild
Need, the horizontal pixel profile extracted is carried out Line vectorization process automatically, utilizes Hough transform to detect building picture
Main shaft in element profile, coordinating block treatment technology carries out Line vectorization, thus obtains the horizontal vector profile of building, avoids
In traditional method, zones of different needs the rule of thumb manual deficiency arranging different distance threshold value.The present invention can be to satellite shadow
In Xiang, a large amount of horizontal vector profiles that top view is the polygonal building of linear structure existed realize batch rapid extraction, for
Follow-up groups of building three-dimensional reconstruction provides modeling data efficiently.Equally applicable to this kind of building taken photo by plane in groups of building image.Can
It is widely used in many three-dimensional modelings and the virtual emulation field such as physical construction planning, military scene simulation.
Accompanying drawing explanation
Fig. 1 is the angle theta schematic diagram of building horizontal pixel profile main shaft p and x-axis, and p is detected by Hough transform
Arrive.
Fig. 2 is horizontal pixel profile Line vectorization schematic diagram.
Fig. 3 is the flow chart of the inventive method.
In figure, 1 building horizontal pixel profile, the main shaft p(left upper end herein of 2 building horizontal pixel profiles is p's
Fixed point), 3 main shaft p and the angle theta of x-axis, 4 imagination image coordinate systems move to initial point and overlap, to show with main shaft fixed point
Theta angle is constituted, the extraneous rectangle of 5 horizontal pixel profiles, 6 cell block unit, 7 5*3 pixels.
Detailed description of the invention
The present invention is further described with embodiment below in conjunction with the accompanying drawings, and the present invention includes but are not limited to following enforcement
Example.
The present invention specifically comprises the following steps that
Step one, Image semantic classification
Being smoothed groups of building satellite image initially with Gaussian filter, smoothed filtering in image is deposited
Some speckle noises.Here using variances sigma=0.8, window size is that former image is carried out by the Gaussian filter of 7*7
Filtering.Then in Hsv color space, the hue h value keeping each pixel is constant, adjusts saturation according to s=s+ Δ s, presses
Adjust brightness according to v=v+ Δ v, satellite image is carried out image enhancement processing so that building color and the color of periphery scenery
Contrast is strengthened, it is simple to distinguishing building and non-construction zone, wherein the scope of Δ s is between 0.06-0.08, Δ v
Scope between 0.05-0.07.The image being used for subsequent treatment after pretreatment is referred to as source images, is stored as Iorig。
Step 2, profile automatically extract
1. the automatically generating of different building seed points
1) K-mean algorithm is utilized at Lab color space, to image on a, b color subspace represent colouring information
IorigCarrying out built-up area regional partition, the class categories number that K-mean algorithm needs in performing is equal to image IorigAt a, b color
Peak value number in the two-dimensional histogram in space.The image being partitioned into building is stored as Iseg.Why select Lab color
Space, is because it and is more suitable for human eye sensation, and color is more rich.
2), completed the rough segmentation in region by K-mean algorithm after, on satellite image, some are close with building color
Ground, roof, road etc. are also divided in the classification of building, need to process further.First by image IsegBy cromogram
As being converted to gray level image, then it is binary image by greyscale image transitions, is stored as Ibw.At binary image IbwIn, building
Object area is white, and background is black.To image IbwCarry out lower column processing successively:
(1) medium filtering: using window is 5*5(unit: pixel) median filter, remove in image isolated pixel
Point.
(2) opening and closing operation: use 5*5(unit: pixel) square structure operator carry out opening and closing operation, remove area and be less than
The region of square structure operator coverage.
(3) little connected region is removed: remove area less than sareaThe connected region of individual pixel, main target is to remove to be similar to
The connected region of roof, ground etc.Here sareaSpan be the integer in [700,2000].
(4) big connected region is removed: remove area more than IareaThe connected region of individual pixel;Herein for object main
Ground, region that road homalographic is bigger, IareaSpan be the integer in [8000,10000].
After above-mentioned a series of process, most of non-building can be got rid of with the trunk portion of preserved building thing simultaneously
Thing part, the non-partial building wherein remained will be in 4 in subsequent step 2) time remove.
3) finally construction zone is numbered respectively, and record its barycenter initial seed point as each building.If planting
Son point position coordinates is (xk,yk), k is building numbering, xkAnd ykIt is respectively kth building seed points row in the picture
Number and row number, here using the position coordinates of initial seed point as (xk,yk) initial value, computational methods are as follows:
In above formula, i represents the line number of image, and j represents the row number of image, f (i, j) represent (i, j) gray value of place's pixel,
D represents UNICOM region, symbolRepresent downward rounding operation.
2. extract construction zone
Mainly include two parts: use the building initial seed point initiated area growth obtained in above-mentioned steps 1;Profit
Use source images IorigMarginal information separation construction zone and non-construction zone, avoid conventional region-growing method need the most right
The deficiency of different growing threshold chosen by different buildings.It is specifically divided into the following steps:
1) utilize region-growing method, generate the region of each candidate architecture thing, ensured by the selection of growing threshold thresh
Candidate architecture object area presents the situation of over-segmentation, so that candidate architecture object area is completely retained.Specifically comprise the following steps that
(1) one and source images I is initially set uporigThe same size, the image I of pixel grey scale all 0outIf growing threshold
Value is thresh, and its span is 8-10 pixel.
(2) an initial seed point (x is oriented by formula (1) and formula (2)k,yk), k is building numbering, arranges Iout
(xk,yk)=1.
(3) with (xk,ykCentered by), it is considered to (xk,yk) four neighborhood territory pixel (xk-1,yk), (xk+1,yk), (xk,yk-1),
(xk,yk+ 1), in Lab color space, the colour-difference of pixel and seed points in seed points neighborhood is calculated respectively according to Euclidean distance
D (x, y), computational methods are:
In above formula, L (x, y), a (x, y), b (x, y) represent respectively current pixel in neighborhood (x, L, a, b component value y),
And Lseed、aseed、bseedThen represent L, a, b component value of current seed point.
If current pixel (x, I y) in neighborhoodout(x, y)=0, and d (x, y) < thresh then arranges Iout(x,y)
=1, simultaneously will (x, y) press-in storehouse.
(4) from storehouse, eject a pixel, it as new seed points, be assigned to (xk,yk), return to previous step
(3).When storehouse is empty, growth terminates.Result makes IoutConstruction zone has been gradually formed (white near initial seed point
Color part).
(5) return to step (2), orient next seed points, repeat above step (3) and (4), until all seed points
Scanning completes, and whole growth course terminates.The mode of choosing of period growing threshold and seed points neighborhood keeps constant.
With source images IorigContrast mutually, the image I obtained in said processoutShow as candidate architecture object area to occur
Segmentation, the non-construction zone of some of edges of regions generates the most simultaneously, and this mode ensure that the complete of candidate architecture object area
Property, it can be ensured that the subsequent operation complete extraction to candidate architecture object area.
2) to source images IorigDetect that strong edge, the pixel on strong edge are white with Canny operator after gray processing, its
His pixel is black, and the image that this is contained strong marginal information is stored as Iedge。
3) traversing graph is as Iout, work as IoutIn pixel be in white construction zone time, check image IedgeMiddle correspondence
Whether the pixel of position is in the edge of white, the most then by IoutMiddle respective pixel becomes black, i.e. the color of background.Pass through
After above-mentioned process, at image IoutMiddle candidate architecture thing will separate with peripheral region, and the candidate showing as sheet in the picture builds
Build object area along with trifling, scattered non-construction zone.
4) at image IoutIn, utilize the characteristic parameters such as rectangular degree d, length-width ratio r, banding index s to get rid of non-built-up area
Non-building in territory and candidate architecture object area.The computational methods of each characteristic parameter are respectively as follows:
Wherein A, l are area and the girth in each region respectively, Ae, the face of h, w corresponding each region minimum enclosed rectangle respectively
Long-pending, length and width.
Owing to, in groups of building satellite image, the closing that the top view of common building thing is typically all linear section structure is polygon
Shape, has regular rectangular shape or the feature of similar rectangle, therefore can have compared with high rectangle degree, relatively low banding index and certain
Length-width ratio, the most satisfied to building: rectangular degree d be more than 0.7, banding index s be less than 3.0, length-width ratio r be less than 5.0.According to
Above three characteristic parameter can get rid of the non-building in non-construction zone and candidate architecture object area.Judgment criterion
For:
As d>0.7 and s<3 and r<when 5, determine that this region is building.
The region being identified as building will remain, and otherwise this region be become black, i.e. the color of background.
Step 3, the Line vectorization of horizontal pixel profile process
The construction zone marginal portion pixel obtained after previous step can be uneven, originally should be for the limit of straight line
Edge, becomes the broken line being made up of many trifling little line segments, and such situation is unfavorable for entering of follow-up three-dimensional modeling work
OK.In three-dimensional modeling, the medium and small line segment of the horizontal profile of building increases can cause the three-dimensional building model unilateral number of intermediate cam shape
Increase sharply, directly affect the real-time that dummy emulation system runs, especially in simulation on wide scene, affect more serious.Therefore originally
Part utilizes the technology such as Hough transform, block process that the horizontal pixel profile of building is carried out Line vectorization process.
First, on the basis of step 2, utilize the method for labelling connected region by image IoutIn single building divide
Cut out, be stored as I respectivelyout_k, k is building numbering.To the single building image I splitout_kDo following place
Reason:
(1) at Iout_kIn, utilize Hough transform to detect the longest corresponding in building horizontal pixel profile straightway,
Main shaft p as building.With main shaft one end for fixed point, that horizontal pixel profile is rotated to main shaft and image coordinate system
Coordinate axes is parallel, and specific practice is as follows:
If theta angle is building main shaft p and the angle of x-axis positive direction in image coordinate system, just it is counterclockwise, single
Position is degree.
When | cos (theta) | > cos (45 °), building horizontal pixel profile is rotated to main shaft and image coordinate system
X-axis parallel;
When | cos (theta) |≤cos (45 °), building horizontal pixel profile is rotated to main shaft and image coordinate system
Y-axis parallel.
(2) by postrotational horizontal pixel profile by its extraneous rectangle piecemeal at equal intervals on x, y direction, every block size is
5*3 pixel, is defined as cell block, is denoted as unit.Owing to having the thin of these classes of outthrust such as balcony on usual major axes orientation
Joint information, is defined as 3 pixels the most after rotation by cell block on the direction at main shaft place, in order to avoid losing detailed information, separately
One direction is defined as 5 pixels.Last column and last string calculating by a cell block less than a cell block.
(3) at image Iout_kIn, from top to bottom, from left to right progressive scan each unit block, calculate in each unit block and build
Object area area and the percentage ratio of cell block area:
Wherein, AbuildIt is the area of construction zone, A in cell blockunitFor the area of cell block, square measure is picture
Element number.
When Percentage is more than or equal to 0.45, it is believed that this cell block is construction zone, retain this cell block the most complete
Portion is filled to white, otherwise abandons this cell block.After the end of scan, all cell blocks finally retained constitute new building
Region, the edge of tracing area can be obtained by the profile of linearization(-sation).
(4) the linearization(-sation) profile obtained is gone back to home position around the fixed point rotary-theta angle adjustment of main shaft, obtain list
The linearization(-sation) profile of individual building.
(5) respectively all of construction zone is carried out the operation of above-mentioned (1)-(4), and each building after processing is straight
Line profile is merged in same image, and is stored as vectogram, is finally completed buildings all in groups of building on raw video
The batch extracting of thing horizontal vector profile.
In this example, the extraction of satellite image groups of building includes three below step:
Step one, Image semantic classification
It is the Gaussian filter of 7*7 initially with variances sigma=0.8, window size, smoothed filters groups of building satellite
Some speckle noises present in image.Then, in Hsv color space, make satellite image to adjust as follows: each pixel color
Adjusting h constant, saturation s=s+ Δ s, brightness v=v+ Δ v, wherein the scope of Δ s is between 0.06-0.08, and the scope of Δ v exists
Between 0.05-0.07, complete the image enhancement processing to image.The image being used for subsequent treatment after pretreatment is referred to as source figure
Picture, is stored as Iorig。
Step 2, profile automatically extract
Detailed operating procedures is as follows:
1. the automatically generating of different building seed points
1) K-mean algorithm is utilized at Lab color space, to image on a, b color subspace represent colouring information
IorigCarry out built-up area regional partition, the image obtained is stored as Iseg.The class categories number etc. that wherein K-mean algorithm needs
In image IorigPeak value number in the two-dimensional histogram of a, b color subspace.
2) by image IsegBe converted to gray level image by coloured image, then be binary image by greyscale image transitions, storage
For Ibw, wherein construction zone is white, and background is black.To image IbwCarry out lower column processing successively, get rid of most
Territory, non-building area:
(1) medium filtering: using window is 5*5(unit: pixel) median filter, remove in image isolated pixel
Point.
(2) opening and closing operation: use 5*5(unit: pixel) square structure operator carry out opening and closing operation, remove area and be less than
The region of square structure operator coverage.
(3) little connected region is removed: remove area less than sareaThe connected region of individual pixel, main target is to remove to be similar to
The connected region of roof, ground etc.Here sareaSpan be the integer in [700,2000].
(4) big connected region is removed: remove area more than IareaThe connected region of individual pixel;Herein for object main
Ground, region that road homalographic is bigger, IareaSpan be the integer in [8000,10000].
After above-mentioned a series of process, minority residual non-partial building will be in 4 in subsequent step 2) time go
Remove.
3) finally construction zone is numbered respectively, and record its barycenter initial seed point as each building.If planting
Son point position coordinates is (xk,yk), k is building numbering, xkAnd ykIt is respectively kth building seed points row in the picture
Number and row number, here using the position coordinates of initial seed point as (xk,yk) initial value, computational methods are as follows:
In above formula, i represents the line number of image, and j represents the row number of image, f (i, j) represent (i, j) gray value of place's pixel,
D represents UNICOM region, symbolRepresent downward rounding operation.
2. extract construction zone
1) based on the building initial seed point obtained in above-mentioned steps 1, utilize region-growing method, generate each candidate architecture
The region of thing, presents the situation of over-segmentation by the guarantee candidate architecture object area that selects of growing threshold thresh, so that candidate
Construction zone is completely retained.Specifically comprise the following steps that
(1) one and source images I is initially set uporigThe same size, the image I of pixel grey scale all 0outIf growing threshold
Value is thresh, and its span is 8-10 pixel.
(2) an initial seed point (x is oriented by formula (1) and formula (2)k,yk), k is building numbering, arranges Iout
(xk,yk)=1.
(3) with (xk,ykCentered by), it is considered to (xk,yk) four neighborhood territory pixel (xk-1,yk), (xk+1,yk), (xk,yk-1),
(xk,yk+ 1), in Lab color space, the colour-difference of pixel and seed points in seed points neighborhood is calculated respectively according to Euclidean distance
D (x, y), computational methods are:
In above formula, L (x, y), a (x, y), b (x, y) represent respectively current pixel in neighborhood (x, L, a, b component value y),
And Lseed、aseed、bseedThen represent L, a, b component value of current seed point.
If current pixel (x, I y) in neighborhoodout(x, y)=0, and d (x, y) < thresh then arranges Iout(x,y)
=1, simultaneously will (x, y) press-in storehouse.
(4) from storehouse, eject a pixel, it as new seed points, be assigned to (xk,yk), return to previous step
(3).When storehouse is empty, growth terminates.Result makes IoutConstruction zone has been gradually formed (white near initial seed point
Color part).
(5) return to step (2), orient next seed points, repeat above step (3) and (4), until all seed points
Scanning completes, and whole growth course terminates.The mode of choosing of period growing threshold and seed points neighborhood keeps constant.
With source images IorigContrast mutually, the image I obtained in said processoutShow as candidate architecture object area to occur
Segmentation, to guarantee the subsequent operation complete extraction to candidate architecture object area.
2) to source images IorigGoing out strong edge by Canny rim detection after gray processing, the pixel on strong edge is white, its
His pixel is black.The image that this is contained strong marginal information is stored as Iedge。
3) traversing graph is as Iout, work as IoutIn be white construction zone time, check image IedgeWhether middle correspondence position
For white edge, when for white edge, by IoutMiddle respective pixel becomes black, i.e. the color of background.So, at image
IoutMiddle candidate architecture thing will separate with peripheral region, shows as the candidate architecture object area of sheet in the picture along with trivial
Broken, scattered non-construction zone.
4) at image IoutIn, utilize the characteristic parameters such as rectangular degree d, length-width ratio r, banding index s to get rid of non-built-up area
Non-building in territory and candidate architecture object area.The computational methods of each characteristic parameter are respectively as follows:
Wherein A, l are area and the girth in each region respectively, Ae, the face of h, w corresponding each region minimum enclosed rectangle respectively
Long-pending, length and width.
According to criterion " when the d of connected domain>0.7 and s<3 and r<when 5, determine that this region is building." judge IoutIn
Whether each region is building, if building, then retains this region;Otherwise this region is become black, i.e. the face of background
Color.
Step 3, the Line vectorization of horizontal pixel profile process
This part utilizes the technology such as Hough transform, block process that the horizontal pixel profile of building is carried out Line vectorization
Process.First, at image IoutIn, utilize the method for labelling connected region by image IoutIn single building split,
It is stored as I respectivelyout_k, k is building numbering.To image Iout_kDo respectively and process as follows:
(1) at Iout_kIn, utilize Hough transform to detect the longest corresponding in building horizontal pixel profile straightway,
Main shaft p as building.With main shaft one end for fixed point, that horizontal pixel profile is rotated to main shaft and image coordinate system
Coordinate axes is parallel.If theta angle is building main shaft p and the angle of x-axis positive direction in image coordinate system, it is counterclockwise
Just, unit is degree.Theta represents this building horizontal positioned in the picture equal to 90 °, and theta is disposed vertically equal to 0 ° of expression.
As theta neither 0 °, when being not 90 °, building shows as slant setting in the picture, it is therefore desirable to rotate it
Operation so that it is horizontally or vertically place:
When | cos (theta) | >=cos (45 °), building horizontal pixel profile is rotated to main shaft and image coordinate system
X-axis parallel;
When | cos (theta) |≤cos (45 °), building horizontal pixel profile is rotated to main shaft and image coordinate system
Y-axis parallel.
(2) by postrotational horizontal pixel profile by its extraneous rectangle piecemeal at equal intervals on x, y direction, every block size is
5*3 pixel, is defined as cell block, is denoted as unit.Owing to having the thin of these classes of outthrust such as balcony on usual major axes orientation
Joint information, is defined as 3 pixels the most after rotation by cell block on the direction at main shaft place, in order to avoid losing detailed information, separately
One direction is defined as 5 pixels.Last column and last string calculating by a cell block less than a cell block.
(3) at image Iout_kIn, from top to bottom, from left to right progressive scan each unit block, calculate in each unit block and build
Object area area and the percentage ratio of cell block area:
Wherein, AbuildIt is the area of construction zone, A in cell blockunitFor the area of cell block, square measure is picture
Element number.
When Percentage is more than or equal to 0.45, it is believed that this cell block is construction zone, retain this cell block the most complete
Portion is filled to white, otherwise abandons this cell block.The all cell blocks retained after the end of scan constitute new construction zone,
The edge of tracing area can be obtained by the profile of linearization(-sation).
(4) the linearization(-sation) profile obtained is gone back to home position around the fixed point rotary-theta angle adjustment of main shaft, obtain list
The linearization(-sation) profile of individual building.
(5) respectively all of construction zone is carried out the operation of above-mentioned (1)-(4), and each building after processing is straight
Line profile is merged in same image, and is stored as vectogram, is finally completed buildings all in groups of building on raw video
The batch extracting of thing horizontal vector profile.
Embodiment 1: somewhere, Beijing satellite image that original image is a 702*902 pixel is carried out contour of building
Extraction.
Using variances sigma=0.8, window size is that the Gaussian filter of 7*7 carries out smothing filtering to former image;?
In Hsv space, the h value keeping each pixel is constant, and v=v+0.08, s=s+0.07 carry out image enhancement processing so that building
Strengthen with the color contrast of background.Image after processing is stored as Iorig, as the source images of subsequent treatment.
Image IorigIn a, b color subspace of Lab color space, peak value number is 3, as by K-mean algorithm pair
It carries out the classification number during regional partition of built-up area, and the image after segmentation saves as Iseg, in this embodiment, construction zone is deposited
It is the 3rd class.Then to image IsegCarry out gray processing and binary conversion treatment, obtain its binary image Ibw, now IbwIn
Connected region number is 100, which includes building and non-construction zone.Then travel through each connected region, get rid of
Area is less than sarea=2000 and area more than IareaThe region of=10000, utilizes size for 5*5(unit: pixel) square structure
Operator carries out opening operation to image, removes the area region less than square structure operator coverage, now bianry image IbwIn
The number of connected region becomes 9, i.e. the trunk portion of building has 9.Calculate the center-of-mass coordinate of each connected region respectively
As the initial seed point coordinates of each building, as shown in table 1:
The initial seed point coordinates of each building of table 1
Zone number | Seed points x coordinate | Seed points y-coordinate |
1 | 100 | 352 |
2 | 245 | 479 |
3 | 319 | 107 |
4 | 345 | 700 |
5 | 352 | 217 |
6 | 463 | 230 |
7 | 540 | 739 |
8 | 563 | 314 |
9 | 655 | 366 |
Utilize 9 initial seed point tried to achieve above, choose growing threshold thresh=10, utilize region-growing method to obtain 9
Individual candidate architecture object area, image saves as Iout.In conjunction with Canny operator from source images IorigIn the strong edge that detects, by image
IoutIn candidate architecture object area separate with territory, non-building area.At image IoutIn, calculate the rectangular degree d in each region, band respectively
Shape index s, tri-characteristic parameters of length-width ratio r, wherein the characteristic parameter of candidate architecture object area is as shown in table 2 (due to non-building
Object area is trifling, scattered, irregular and easily distinguishable, the most no longer enumerates.):
The characteristic parameter of 2 nine candidate architecture object areas of table
Zone number | Rectangular degree d | Banding index s | Length-width ratio r |
1 | 0.75 | 0.0029 | 1.9 |
2 | 0.77 | 0.0027 | 2.26 |
3 | 0.83 | 0.0035 | 1.98 |
4 | 0.71 | 0.003 | 2.83 |
5 | 0.76 | 0.004 | 2.49 |
6 | 0.82 | 0.0031 | 1.91 |
7 | 0.76 | 0.0028 | 2.53 |
8 | 0.85 | 0.003 | 2.06 |
9 | 0.55 | 0.0093 | 1.12 |
According to criterion " as d>0.7 and s<3 and r<when 5, determine that this region is building." judge whether each region is building
Thing.Wherein region 9 is unsatisfactory for index, therefore casts out this region, has the most tentatively obtained the water of all 8 buildings in image
Flat pixel profile.
Utilize the method for labelling connected region by image IoutIn single building split, be stored as I respectivelyout_k,
K is building numbering.Hough transform is utilized to detect each Iout_kThe main shaft p of middle building horizontal pixel profile, calculates respectively
Horizontal pixel profile main shaft and the angle theta, | the cos (theta) | >=cos (45 °) and | cos (theta) | according to condition of x-axis
Horizontal pixel profile is rotated to horizontally or vertically position by≤cos (45 °), facilitates next step to operate.
Table 3 each building horizontal pixel profile main shaft and the angle theta of x-axis
Zone number | Main shaft and the angle theta of x-axis |
1 | -42.02° |
2 | -41.28° |
3 | -41.11° |
4 | -30.37° |
5 | -22.38° |
6 | -40.99° |
7 | -30.04° |
8 | -41.66° |
In this example, in table 3, building is all in obliquity, is therefore required to rotate it.The size is utilized to be
Single postrotational construction zone is divided by the cell block unit of 5*3 pixel respectively, it is judged that build in each cell block
Build object area (white portion) and account for the percentage ratio Percentage of whole cell block region area, as Percentage > 0.45 time,
I.e. think that this cell block is construction zone and is stuffed entirely with as white, after having processed whole cell block, by building level picture
Element profile rotates back to initial position.Operating as above owned building region successively, the edge of tracing area can be obtained by
Contour of building after linearization(-sation), is finally merged into each building linearization(-sation) profile in same image, and is stored as vectogram.
Embodiment 2: the satellite image of the dormitory area, Purdue University that original image is a 775*401 pixel is built
The extraction of thing profile.
Using variances sigma=0.8, window size is that the Gaussian filter of 7*7 carries out smothing filtering to former image;?
In Hsv space, keeping the h of each pixel constant, v=v+0.06, s=s+0.05 carry out image enhancement processing so that building with
The color contrast of background is strengthened.Image after processing is stored as Iorig, as the source images of subsequent treatment.
Image IorigIn a, b color subspace of Lab color space, peak value number is 3, as by K-mean algorithm pair
It carries out the classification number during regional partition of built-up area, and the image after segmentation saves as Iseg, in this embodiment, construction zone is deposited
It is the 2nd class.Then to image IsegCarry out gray processing and binary conversion treatment, obtain its binary image Ibw, now IbwIn
Connected region number is 16, which includes building and non-construction zone.Then travel through each connected region, get rid of face
Long-pending less than sarea=700 and area more than IareaThe region of=8000, utilizes size for 5*5(unit: pixel) square structure operator
Image is carried out opening operation, removes the area region less than square structure operator coverage, this bianry image IbwMiddle connected region
The number in territory becomes 9, i.e. the trunk portion of building has 9.The center-of-mass coordinate calculating each connected region the most respectively is made
For the initial seed point coordinates of each building, as shown in table 4:
The initial seed point coordinates of each building of table 4
Zone number | Seed points x coordinate | Seed points y-coordinate |
1 | 108 | 45 |
2 | 96 | 194 |
3 | 182 | 366 |
4 | 207 | 209 |
5 | 359 | 179 |
6 | 437 | 58 |
7 | 542 | 333 |
8 | 513 | 179 |
9 | 665 | 118 |
Utilize 9 initial seed point tried to achieve above, choose growing threshold thresh=8, utilize region-growing method to obtain 9
Individual candidate architecture object area, image saves as Iout.In conjunction with Canny operator from source images IorigIn the strong edge that detects, by image
IoutIn candidate architecture object area separate with territory, non-building area.At image IoutIn, calculate the rectangular degree d in each region, band respectively
Shape index s, tri-characteristic parameters of length-width ratio r, wherein the characteristic parameter of candidate architecture object area is as shown in table 5 (due to non-building
Object area is trifling, scattered, irregular and easily distinguishable, the most no longer enumerates.):
The characteristic parameter of 5 nine candidate architecture object areas of table
Zone number | Rectangular degree d | Banding index s | Length-width ratio r |
1 | 0.85 | 0.003 | 4.75 |
2 | 0.83 | 0.0036 | 4.35 |
3 | 0.92 | 0.0035 | 5.32 |
4 | 0.87 | 0.0032 | 4.46 |
5 | 0.82 | 0.0032 | 4.11 |
6 | 0.97 | 0.006 | 1.93 |
7 | 0.84 | 0.0033 | 4.7 |
8 | 0.86 | 0.0032 | 4.37 |
9 | 0.94 | 0.0032 | 4.83 |
According to criterion " as d>0.7 and s<3 and r<when 5, determine that this region is building." judge whether each region is building
Thing.Above-mentioned 9 connected regions all meet the parameter index of construction area, have the most tentatively obtained all 9 buildings in image
Horizontal pixel profile.
Utilize the method for labelling connected region by image IoutIn single building split, be stored as I respectivelyout_k,
K is building numbering.Hough transform is utilized to detect each Iout_kThe main shaft p of middle building horizontal pixel profile, calculates respectively
Horizontal pixel profile main shaft and the angle theta, | the cos (theta) | >=cos (45 °) and | cos (theta) | according to condition of x-axis
Horizontal pixel profile is rotated to horizontally or vertically position by≤cos (45 °), facilitates next step to operate.
Table 6 each building horizontal pixel profile main shaft and the angle theta of x-axis
Zone number | Main shaft and the angle theta of x-axis |
1 | 90° |
2 | 0° |
3 | 90° |
4 | 0° |
5 | 90° |
6 | 90° |
7 | 89.5° |
8 | 0° |
9 | 90° |
In this example, in table 6, building is all in horizontally or vertically position, therefore need not rotate it.Profit
Respectively single postrotational construction zone is divided with the cell block unit that size is 5*3 pixel, it is judged that Mei Gedan
In unit's block, construction zone (white portion) accounts for the percentage ratio Percentage of whole cell block region area, works as Percentage
> 0.45 time, i.e. think that this cell block is construction zone and is stuffed entirely with as white, after having processed whole cell block, will building
Thing horizontal pixel profile rotates back to initial position.Operating as above owned building region successively, the edge of tracing area is just
The contour of building after linearization(-sation) can be obtained, finally each building linearization(-sation) profile is merged in same image, and stores
For vectogram.
Embodiment 3: the satellite image in the somewhere, Xi'an that original image is a 874*383 pixel is carried out building wheel
Wide extraction.
Using variances sigma=0.8, window size is that the Gaussian filter of 7*7 carries out smothing filtering to former image;?
In Hsv space, keeping the h of each pixel constant, v=v+0.07, s=s+0.06 carry out image enhancement processing so that building with
The color contrast of background is strengthened.Image after processing is stored as Iorig, as the source images of subsequent treatment.
Image IorigIn a, b color subspace of Lab color space, peak value number is 5, as by K-mean algorithm pair
It carries out the classification number during regional partition of built-up area, and the image after segmentation saves as Iseg, in this embodiment, construction zone is deposited
It is the 1st class.Then to image IsegCarry out gray processing and binary conversion treatment, obtain its binary image Ibw, now IbwIn
Connected region number is 84, which includes building and non-construction zone.Then travel through each connected region, get rid of face
Long-pending less than sarea=1000 and area more than IareaThe region of=9000, utilizes size for 5*5(unit: pixel) square structure calculate
Son carries out opening operation to image, removes the area region less than square structure operator coverage, this bianry image IbwMiddle connection
The number in region becomes 12, i.e. the trunk portion of building has 12.The center-of-mass coordinate calculating each connected region respectively is made
For the initial seed point coordinates of each building, as shown in table 7:
The initial seed point coordinates of each building of table 7
Zone number | Seed points x coordinate | Seed points y-coordinate |
1 | 15 | 39 |
2 | 37 | 146 |
3 | 93 | 72 |
4 | 152 | 168 |
5 | 167 | 346 |
6 | 239 | 195 |
7 | 404 | 192 |
8 | 406 | 310 |
9 | 356 | 10 |
10 | 566 | 253 |
11 | 778 | 308 |
12 | 764 | 191 |
Utilize 12 initial seed point tried to achieve above, choose growing threshold thresh=9, utilize region-growing method to obtain
12 candidate architecture object areas, image saves as Iout.In conjunction with Canny operator from source images IorigIn the strong edge that detects, will figure
As IoutIn candidate architecture object area separate with territory, non-building area.At image IoutIn, calculate respectively each region rectangular degree d,
Banding index s, tri-characteristic parameters of length-width ratio r, wherein the characteristic parameter of candidate architecture object area (is built due to non-as shown in table 8
Build object area trifling, scattered, irregular and easily distinguishable, the most no longer enumerate.):
The characteristic parameter of 80 two candidate architecture object areas of table
Zone number | Rectangular degree d | Banding index s | Length-width ratio r |
1 | 0.71 | 0.0083 | 9.81 |
2 | 0.59 | 0.011 | 5.99 |
3 | 0.65 | 0.005 | 2.35 |
4 | 0.85 | 0.0036 | 1.93 |
5 | 0.7 | 0.0093 | 1.18 |
6 | 0.45 | 0.0046 | 2.5 |
7 | 0.81 | 0.004 | 4.65 |
8 | 0.83 | 0.0039 | 4.62 |
9 | 0.55 | 0.0093 | 1.12 |
10 | 0.89 | 0.0037 | 4.74 |
11 | 0.89 | 0.0036 | 4.5 |
12 | 0.79 | 0.0046 | 3.36 |
According to criterion " as d>0.7 and s<3 and r<when 5, determine that this region is building." judge whether each region is building
Thing.Wherein region 1,2,3,5,6,9 is unsatisfactory for index, therefore casts out this 6 regions, and remaining 6 connected regions all meet builds
Build the parameter index in region, the most tentatively obtain the horizontal pixel profile of all 6 buildings in image.
Utilize the method for labelling connected region by image IoutIn single building split, be stored as I respectivelyout_k,
K is building numbering.Hough transform is utilized to detect each Iout_kThe main shaft p of middle building horizontal pixel profile, calculates respectively
Horizontal pixel profile main shaft and the angle theta, | the cos (theta) | >=cos (45 °) and | cos (theta) | according to condition of x-axis
Horizontal pixel profile is rotated to horizontally or vertically position by≤cos (45 °), facilitates next step to operate.
Table 9 each building horizontal pixel profile main shaft and the angle theta of x-axis
Zone number | Main shaft and the angle theta of x-axis |
1 | 45° |
2 | -80.5° |
3 | 90° |
4 | 0° |
5 | -48.8° |
6 | 90° |
In this example, in table 9, building 1,2,5 is in obliquity, needs to rotate it, and remaining construction thing is the most not
Need to carry out rotation processing.Utilize cell block unit that size is 5*3 pixel respectively to after single rotation (the need to)
Construction zone divides, it is judged that in each cell block, construction zone (white portion) accounts for whole cell block region area
Percentage ratio Percentage, as Percentage > 0.45 time, i.e. think that this cell block is construction zone and is stuffed entirely with as in vain
Color, after having processed whole cell block, rotates back to initial position again by the building horizontal pixel profile rotated through.Successively to all
Construction zone operates as above, and the edge of tracing area can be obtained by the contour of building after linearization(-sation), finally will respectively build
Build thing linearization(-sation) profile to be merged in same image, and be stored as vectogram.
Claims (1)
1. building horizontal vector profile automatic batch extracting method in a satellite image, it is characterised in that include following step
Rapid:
Groups of building satellite image is smoothed by step one, employing Gaussian filter, described Gaussian filter
Variances sigma=0.8, window size is 7*7 pixel;Then, in Hsv color space, the tone keeping each pixel is constant, presses
Adjust saturation according to step value Δ s, adjust brightness according to step value Δ v, satellite image is carried out image enhancement processing, obtains source
Image Iorig, wherein Δ s is 0.06-0.08, and Δ v is 0.05-0.07;
Step 2, profile automatically extract, and comprise the following steps:
<1>. automatically generating of different building seed points, comprise the following steps:
1) K-mean algorithm is utilized at Lab color space, to source images I on a, b color subspace represent colouring informationorig
Carrying out built-up area regional partition, the class categories number that K-mean algorithm needs in performing is equal to image IorigEmpty at a, b color
Between two-dimensional histogram in peak value number, the image being partitioned into building is stored as Iseg;
2) by image IsegBe converted to gray level image, then be binary image I by greyscale image transitionsbw, wherein, construction zone
For white, background is black, to binary image IbwCarry out lower column processing successively:
(1) median filter using window to be 5*5 pixel removes pixel isolated in image;
(2) use the square structure operator of 5*5 pixel to carry out opening and closing operation, remove area less than square structure operator coverage
Region;
(3) area is removed less than sareaThe connected region of individual pixel, sareaSpan be the integer in [700,2000];
(4) area is removed more than IareaThe connected region of individual pixel, IareaSpan be the integer in [8000,10000];
3) construction zone is numbered respectively, and record its barycenter initial seed point as each building;If seed point location
Coordinate is (xk,yk), k is building numbering, xkAnd ykIt is respectively kth building seed points line number in the picture and row number,I represents the line number of image, and j represents the row number of image, f
(i j) represents that (i, j) gray value of place's pixel, D represents UNICOM region, symbolRepresent downward rounding operation;
<2>. extract construction zone, comprise the following steps:
1) utilize region-growing method, generate the region of each candidate architecture thing, specifically comprise the following steps that
(1) one and source images I is set uporigThe same size, the image I of pixel grey scale all 0outIf growing threshold is
Thresh, its span is 8-10 pixel;
(2) an initial seed point (x is orientedk,yk), k is building numbering, arranges Iout(xk,yk)=1;
(3) with (xk,ykCentered by), it is considered to (xk,yk) four neighborhood territory pixel (xk-1,yk), (xk+1,yk), (xk,yk-1), (xk,
yk+ 1), in Lab color space, the colour-difference of pixel and seed points in seed points neighborhood is calculated respectively according to Euclidean distanceWherein, L (x, y), a (x, y), b (x,
Y) current pixel (x, L, a, b component value y), and L in neighborhood is represented respectivelyseed、aseed、bseedThen represent current seed point
L, a, b component value;If current pixel (x, I y) in neighborhoodout(x, y)=0, and d (x, y) < thresh then arranges Iout
(x, y)=1, simultaneously will (x, y) press-in storehouse;
(4) from storehouse, one pixel of ejection, as new seed points, is assigned to (xk,yk), return to extract construction zone step
In step (3);When storehouse is empty, growth terminates;
(5) return to extract the step (2) in construction zone step, orient next seed points, repeat step (3) and (4),
Until all seed points have scanned, whole growth course terminates;
2) to source images IorigDetect that strong edge, the pixel on strong edge are white with Canny operator after gray processing, other pictures
Element is black, and the image that this is contained strong marginal information is stored as Iedge;
3) traversing graph is as Iout, work as IoutIn pixel be in white construction zone time, check image IedgeMiddle correspondence position
Whether pixel is in the edge of white, the most then by IoutMiddle respective pixel becomes black;
4) at image IoutIn, calculate rectangular degreeLength-width ratioWith banding indexWherein A, l are respectively
The area in each region and girth, Ae, h, the w area of corresponding each region minimum enclosed rectangle, length and width respectively;Work as d > 0.7
And s < 3 and r < when 5, determining that this region is building, the region being identified as building will remain, otherwise by this region
Become black;
Step 3, utilize the method for labelling connected region by image IoutIn single building split, be stored as respectively
Iout_k, k is building numbering;To the single building image I splitout_kDo and process as follows:
(1) at Iout_kIn, utilize Hough transform to detect the longest corresponding in building horizontal pixel profile straightway, as
The main shaft p of building;With main shaft one end for fixed point, horizontal pixel profile is rotated a coordinate to main shaft Yu image coordinate system
Axle is parallel, and specific practice is as follows:
If theta angle is building main shaft p and the angle of x-axis positive direction in image coordinate system, being just counterclockwise, unit is
Degree;When | cos (theta) | > cos (45 °), building horizontal pixel profile is rotated the x-axis to main shaft Yu image coordinate system
Parallel;When | cos (theta) |≤cos (45 °), building horizontal pixel profile is rotated the y to main shaft Yu image coordinate system
Axle is parallel;
(2) by postrotational horizontal pixel profile by its extraneous rectangle piecemeal at equal intervals on x, y direction, every block size is 5*3
Individual pixel, is defined as cell block, is denoted as unit;On the direction at main shaft place, cell block is defined as 3 pixels after rotation, separately
One direction is defined as 5 pixels;Last column and last string calculating by a cell block less than a cell block;
(3) at image Iout_kIn, from top to bottom, from left to right progressive scan each unit block, calculate built-up area in each unit block
Territory area and the percentage ratio of cell block areaWherein, AbuildIt it is the face of construction zone in cell block
Long-pending, AunitFor the area of cell block, square measure is number of pixels;When Percentage is more than or equal to 0.45, it is believed that should
Cell block is construction zone, retains this cell block and is stuffed entirely with into white, otherwise abandoning this cell block;
(4) the linearization(-sation) profile obtained is gone back to home position around the fixed point rotary-theta angle adjustment of main shaft, obtain single building
Build the linearization(-sation) profile of thing;
(5) respectively all of construction zone is carried out the operation of above-mentioned steps (1)-(4), and each building after processing is straight
Line profile is merged in same image, and is stored as vectogram, is finally completed buildings all in groups of building on raw video
The batch extracting of thing horizontal vector profile.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201410003206.XA CN103699900B (en) | 2014-01-03 | 2014-01-03 | Building horizontal vector profile automatic batch extracting method in satellite image |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201410003206.XA CN103699900B (en) | 2014-01-03 | 2014-01-03 | Building horizontal vector profile automatic batch extracting method in satellite image |
Publications (2)
Publication Number | Publication Date |
---|---|
CN103699900A CN103699900A (en) | 2014-04-02 |
CN103699900B true CN103699900B (en) | 2016-10-05 |
Family
ID=50361422
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201410003206.XA Active CN103699900B (en) | 2014-01-03 | 2014-01-03 | Building horizontal vector profile automatic batch extracting method in satellite image |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN103699900B (en) |
Families Citing this family (12)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN103927541B (en) * | 2014-04-21 | 2017-02-22 | 西北工业大学 | Target edge line detection method applicable to space tethered robot capture process |
CN108665489B (en) * | 2017-03-27 | 2023-03-21 | 波音公司 | Method for detecting changes in geospatial images and data processing system |
CN107330979B (en) * | 2017-06-30 | 2020-09-18 | 电子科技大学中山学院 | Vector diagram generation method and device for building house type and terminal |
CN108323389B (en) * | 2018-01-18 | 2019-11-08 | 华南农业大学 | The rice transplanting rice shoot spacing in the rows of rice transplanter and the detection method and device of cave rice shoot number |
CN109034073B (en) * | 2018-07-30 | 2021-05-25 | 深圳大学 | Method, system, equipment and medium for predicting generation amount of building demolition waste |
CN109345454B (en) * | 2018-09-18 | 2023-01-06 | 徐庆 | Bitmap image vectorization method, storage medium and system |
CN110619282B (en) * | 2019-08-26 | 2023-01-10 | 海南撰云空间信息技术有限公司 | Automatic extraction method for unmanned aerial vehicle orthoscopic image building |
CN110674735B (en) * | 2019-09-23 | 2022-05-24 | 中国科学院地理科学与资源研究所 | Agricultural facility remote sensing extraction method and device based on fine classification |
CN111523391B (en) * | 2020-03-26 | 2021-11-05 | 上海刻羽信息科技有限公司 | Building identification method, system, electronic device and readable storage medium |
CN111652892A (en) * | 2020-05-02 | 2020-09-11 | 王磊 | Remote sensing image building vector extraction and optimization method based on deep learning |
CN114581786B (en) * | 2021-12-28 | 2022-11-25 | 深圳市城市产业发展集团有限公司 | Method and device for estimating building area according to ground image |
CN114998664A (en) * | 2022-07-18 | 2022-09-02 | 中国科学院烟台海岸带研究所 | Rapid detection method and device for micro-plastic in seawater by multiple optical platforms |
Citations (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN101978395A (en) * | 2008-04-23 | 2011-02-16 | 株式会社博思科 | Building roof outline recognizing device, building roof outline recognizing method, and building roof outline recognizing program |
CN103020342A (en) * | 2012-12-04 | 2013-04-03 | 南京大学 | Method for extracting contour and corner of building from ground LiDAR data |
CN103198494A (en) * | 2013-04-03 | 2013-07-10 | 武汉海达数云技术有限公司 | Building elevation extraction method based on building outer contour lines and laser-point cloud |
CN103218618A (en) * | 2013-01-09 | 2013-07-24 | 重庆交通大学 | Highway route automatic extraction method based on remote-sensing digital image |
Family Cites Families (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US8422825B1 (en) * | 2008-11-05 | 2013-04-16 | Hover Inc. | Method and system for geometry extraction, 3D visualization and analysis using arbitrary oblique imagery |
-
2014
- 2014-01-03 CN CN201410003206.XA patent/CN103699900B/en active Active
Patent Citations (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN101978395A (en) * | 2008-04-23 | 2011-02-16 | 株式会社博思科 | Building roof outline recognizing device, building roof outline recognizing method, and building roof outline recognizing program |
CN103020342A (en) * | 2012-12-04 | 2013-04-03 | 南京大学 | Method for extracting contour and corner of building from ground LiDAR data |
CN103218618A (en) * | 2013-01-09 | 2013-07-24 | 重庆交通大学 | Highway route automatic extraction method based on remote-sensing digital image |
CN103198494A (en) * | 2013-04-03 | 2013-07-10 | 武汉海达数云技术有限公司 | Building elevation extraction method based on building outer contour lines and laser-point cloud |
Also Published As
Publication number | Publication date |
---|---|
CN103699900A (en) | 2014-04-02 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN103699900B (en) | Building horizontal vector profile automatic batch extracting method in satellite image | |
CN107578418B (en) | Indoor scene contour detection method fusing color and depth information | |
CN107230202B (en) | Automatic identification method and system for road surface disease image | |
CN109558806B (en) | Method for detecting high-resolution remote sensing image change | |
CN105354865B (en) | The automatic cloud detection method of optic of multispectral remote sensing satellite image and system | |
CN103077384B (en) | A kind of method and system of vehicle-logo location identification | |
CN106296666B (en) | A kind of color image removes shadow method and application | |
CN110232389B (en) | Stereoscopic vision navigation method based on invariance of green crop feature extraction | |
CN104392212B (en) | The road information detection and front vehicles recognition methods of a kind of view-based access control model | |
CN102968782B (en) | In a kind of coloured image, remarkable object takes method automatically | |
CN106651872A (en) | Prewitt operator-based pavement crack recognition method and system | |
CN104835164B (en) | A kind of processing method and processing device of binocular camera depth image | |
CN108537239B (en) | Method for detecting image saliency target | |
CN106846339A (en) | A kind of image detecting method and device | |
CN102800094A (en) | Fast color image segmentation method | |
CN103177451B (en) | Based on the self-adapting window of image border and the Stereo Matching Algorithm of weight | |
CN105740945A (en) | People counting method based on video analysis | |
CN103198479B (en) | Based on the SAR image segmentation method of semantic information classification | |
CN110544300B (en) | Method for automatically generating three-dimensional model based on two-dimensional hand-drawn image characteristics | |
CN103198319B (en) | For the blurred picture Angular Point Extracting Method under the wellbore environment of mine | |
CN107154026B (en) | Method for eliminating road surface shadow based on self-adaptive brightness elevation model | |
CN106204494A (en) | A kind of image defogging method comprising large area sky areas and system | |
CN104966285A (en) | Method for detecting saliency regions | |
CN102930538B (en) | The self-adaptive projection method method that Gauss potential and spatial histogram merge | |
CN104134198A (en) | Method for carrying out local processing on image |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
C06 | Publication | ||
PB01 | Publication | ||
C10 | Entry into substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
C14 | Grant of patent or utility model | ||
GR01 | Patent grant |