CN103761526A - Urban area detecting method based on feature position optimization and integration - Google Patents
Urban area detecting method based on feature position optimization and integration Download PDFInfo
- Publication number
- CN103761526A CN103761526A CN201410038043.9A CN201410038043A CN103761526A CN 103761526 A CN103761526 A CN 103761526A CN 201410038043 A CN201410038043 A CN 201410038043A CN 103761526 A CN103761526 A CN 103761526A
- Authority
- CN
- China
- Prior art keywords
- point
- city
- weighting
- unique point
- threshold value
- 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.)
- Granted
Links
Images
Landscapes
- Image Analysis (AREA)
Abstract
The invention provides an urban area detecting method based on feature position optimization and integration, prior learning is not needed, calculation is simple, and the urban area detecting method is more suitable for being implemented in practical application. The urban area detecting method includes the steps that step1, images are preprocessed, and the image processing process includes RGB color images conversion to gray level images and Gaussian pyramid generation; step2, urban position feature points are selected preliminarily; step3, the urban position feature points are screened; step4, regional integration is performed on urban areas based on Gaussian rendering weighting; step5, partition threshold values are obtained through a self-adaptive iteration method, binaryzation is performed on a weighting matrix, connected domains of binary images are marked, and the connected domains with the area smaller than the preset threshold value are rejected; step6, from the step2 to the step5 are repeated on all layers of a Gaussian pyramid generated in the step1, after results of all the layers are expanded to the size of the original images, a union set of the results is obtained to obtain an urban area candidate range, color features of the candidate range in the RGB color images, and pixel regions of which the color features do not meet the conditions are rejected to obtain a final detecting result.
Description
Technical field
The invention belongs to image object detection technique field, be specifically related to a kind of city detection method of preferably integrating based on feature locations.
Background technology
Along with the development of remote sensing technology, the resolution of remote sensing image is more and more higher, and the information that can obtain is also more and more.Wherein, by remote sensing images, obtaining urban area information becomes the focus of Chinese scholars research gradually, and this will bring huge help to aspects such as nation-building and territory explorations.First, the first step as city monitoring is detected in urban area, can effectively detect which regional extent and belong to city, contributes to government or urban planning and construction department to formulate urban construction and development plan; Secondly, urban area is detected and be can be used for multiple fields such as urban change detection, digitalized city construction, military surveillance; Finally, because remote sensing images overlay area is wide, by manually sentencing figure mode, to carry out testing amount large, the automatic detection in city with detect and will become the trend of development in real time.
Over nearly 10 years, Chinese scholars makes in all sorts of ways and model detects automatically to urban area, and its technology path is roughly divided into two classes.The first kind is the urban area detection method based on structure, Texture eigenvalue.Wherein, there is scholar to use SIFT algorithm or Harris Robust Algorithm of Image Corner Extraction to extract the unique point within the scope of city, according to the distribution density of unique point, be divided into different subgraphs again, by many subgraph matchs, determine city scope, feasible in this theory of algorithm, but calculation of complex and operation time are longer, are not suitable for practical application; There is scholar by image texture characteristic, to obtain the density of buildings, according to density information, carry out cutting apart of urban area.Equations of The Second Kind is the detection method based on pattern-recognition.There is scholar to use mathematical morphology operation to extract structural information, and by neural network or support vector machine, it is classified and obtains urban area; There is scholar to use architectural feature, as the method that gradient, straight line or various features merge is mutually classified to city by statistical sorter.But the whole bag of tricks based on pattern-recognition all needs priori data to carry out parameter training.
Summary of the invention
Given this, the present invention proposes a kind of city detection method of preferably integrating based on feature locations on the basis of improving the urban area detection method based on structural texture feature, without carrying out priori study, calculates simply, is more suitable for realizing in actual applications.
In order to solve the problems of the technologies described above, the present invention is achieved in that
A city detection method of preferably integrating based on feature locations, comprises the following steps:
Beneficial effect of the present invention:
The present invention carries out priori study or training without mass data, can complete the automatic detection in city; The present invention calculates simply, operation time is short, so can on payload platform, process in real time, only by passing to ground monitoring personnel under the testing result of urban area, has reduced data transfer bandwidth, has shortened the delay of data loop.Can be applicable to the continuous detecting of spaceborne or airborne platform.
Accompanying drawing explanation
Fig. 1 is the city testing process schematic diagram of preferably integrating based on feature locations;
Fig. 2 is feature point extraction template schematic diagram.
Embodiment
Below in conjunction with the accompanying drawing embodiment that develops simultaneously, describe the present invention.
As shown in Figure 1, the city detection method of preferably integrating based on feature locations, concrete steps are:
Wherein, Igray is gray level image, and Ir, Ig, Ib represent respectively red in coloured image, green, blue component.
Then, gray level image is set up to gaussian pyramid.Use Gauss's template to carry out filtering to gray level image, afterwards former figure is carried out respectively taking out a down-sampling and taking out a down-sampling every four every two, gray-scale map itself is as pyramid ground floor like this, and every two, taking out a rear image is the pyramid second layer, and every four, taking out a rear image is the 3rd layer, pyramid.
1) set up signature Matrix C, each element initial value is zero, and size equals this layer of pyramid picture size;
2) as shown in Figure 2, cunning window center P is placed in to image pixel I to be judged (i, j), set up 16 one-dimension array a[n], by comparing the gray-scale value of mid point P and its 16 discrete points that radius is 3 around, array is carried out to assignment, be shown below:
Wherein, I
p → nrepresent the gray-scale value of the point of n on circular arc in template, I
pthe gray-scale value that represents template mid point P, T is similarity measurement threshold value, selection range is generally 15 to 30.
3) to array a[n] summation, and if be more than or equal to 15 or be less than or equal to-15, think that this point is candidate feature point, is set to 1 by C (i, j) in signature matrix.And if equal 0, think that this point is not candidate feature point, carries out next pixel judgement.
4) if do not meet the 3rd step condition, the first two number a[1 of array relatively] and a[16], if both identical ring shifts of carrying out a time compare after displacement again, until both enter next step after not identical.
5) judge whether the array after displacement exists continuous 9 " 1 " or continuous 9 " 1 ", if array meets above-mentioned situation and thinks that this point is candidate feature point, C (i, j) in signature matrix is set to 1, otherwise P is moved to next pixel repeating step 2-5.
According to the method described above each full figure pixel is traveled through one time, the position that in Matrix C (i, j), element is 1 is characteristic point position in figure.
First, construct a scoring function V as criterion, for rejecting the unique point of non-maximum value in subrange.Scoring function V is defined as:
Wherein,
S
bright={x|I
p→x≥I
p+T} (7)
S
dark={x|I
p→x≤I
p-T} (8)
I
prepresent the gray-scale value of any candidate feature point P, I
p → xexpression centered by P in Fig. 2 template corresponding label be the gray-scale value of x, T is similarity measurement threshold value.
Secondly, in all candidate feature point set M that extract, choose arbitrarily a candidate feature point in the first step, might as well establish this point is P, and its scoring function response is V
pif the 3*3 scope internal memory centered by P is at any point q ∈ M, its scoring function response is V
q, work as V
p>=V
qtime, P retains as unique point, otherwise P is just disallowable.
Again, suppose arbitrary characteristics point P
ineighborhood N
piexpression is with P
icentered by the radius border circular areas that is M, define constraint criterion and be:
Wherein card{.} represents to gather the number of interior element, with P
icentered by neighborhood N
pitotal number of middle unique point, δ is given threshold value.Meet the unique point P of above-mentioned criterion
ito be retained, ungratified disallowable.
First, use two-dimensional Gaussian function that each unique point is expanded into weighting submatrix.Setting Gauss's weighting submatrix size is M × M, uses following formula to calculate weighting submatrix:
Wherein, G
c(x, y) is dimensional Gaussian matrix, (x
c, y
c) be unique point position, in above formula, the value of x, y is with (x
c, y
c) centered by the scope of M × M.Choosing σ is fixed value, and at this moment Gauss's weighting submatrix of each unique point expansion is identical.The superimposed weighting matrix that obtains view picture figure of weighting submatrix that each unique point is expanded:
Wherein, total number that N is unique point, the weighting matrix that G (x, y) is whole figure, size is identical with this layer of picture size.
Finally, to G (x, y), use the window of 5*5 to carry out medium filtering to weighting matrix.
1) intermediate value of choosing high-high brightness and minimum brightness in image is as initial value Th;
2) use Th to cut apart image, all pixel composition I of brightness value>=Th
1, all pixel composition I of brightness value L EssT.LTssT.LT Th
2;
3) the average brightness value μ of pixel within the scope of calculating I1 and I2
1and μ
2;
4) calculate according to the following formula new threshold value:
5) repeat above-mentioned steps 2 to 4, until the Th that subsequent iteration obtains meets following formula, stop, wherein ε is predetermined threshold value, is set as 0.1.
Th
n-Th
n-1|≤ε (13)
6) by the some assignment that is greater than threshold value Th, being 1, is 0 by the some assignment that is less than threshold value.And to binary map carry out connected component labeling and by connected domain area be less than certain threshold value connected domain reject.
In RGB coloured picture, check the color component of candidate's scope, when G (green) component average is greater than certain threshold value, this connected domain is rejected, the region obtaining is afterwards final detection result.
Visible, the present invention is by using Feature Points Extraction fast, and carry out unique point screening in conjunction with unique point local restriction and global restriction, can effectively guarantee the degree of confidence of unique point, and the integration that the method that these unique points are played up to weighting by Gauss is carried out urban area extracts, thereby carry out the detection in city sooner more accurately.
In sum, these are only preferred embodiment of the present invention, be not intended to limit protection scope of the present invention.Within the spirit and principles in the present invention all, any modification of doing, be equal to replacement, improvement etc., within all should being included in protection scope of the present invention.In sum, these are only preferred embodiment of the present invention, be not intended to limit protection scope of the present invention.Within the spirit and principles in the present invention all, any modification of doing, be equal to replacement, improvement etc., within all should being included in protection scope of the present invention.
Claims (4)
1. a city detection method of preferably integrating based on feature locations, is characterized in that, comprises the following steps:
Step 1, image pre-service: comprise that RGB coloured picture turns gray level image and gaussian pyramid generates;
Step 2, city position feature point are tentatively chosen: in the image generating from step 1, choose the unique point that can indicate city;
Step 3, city position feature point screening: by the local maximum screening and the constraint of global characteristic point quantity of the unique point chosen in step 2, reject unsettled unique point;
Step 4, based on Gauss, play up the city Regional Integration of weighting: use two-dimensional Gaussian function that the each unique point after rejecting in step 3 is expanded into weighting submatrix, by the superimposed weighting matrix that obtains view picture figure of weighting submatrix of each unique point expansion, and weighting matrix is carried out to medium filtering;
Step 5, weighting matrix step 4 being obtained by adaptive iteration method are obtained segmentation threshold and are carried out binaryzation operation, and then to binary map carry out connected component labeling and by connected domain area be less than predetermined threshold value connected domain reject;
Each layer of repeating step two of step 6, the gaussian pyramid to step 1 generation is to step 5, and each layer of result expanded to and get union after former figure size and obtain city candidate's scope, check the color character of candidate's scope in RGB coloured picture, reject the pixel region that color character does not satisfy condition, obtain final detection result.
2. a kind of city detection method of preferably integrating based on feature locations as claimed in claim 1, is characterized in that, the following method of employing of choosing of city unique point in step 2:
1) set up signature Matrix C, each element initial value is zero, and size equals this layer of pyramid picture size;
2) cunning window center P is placed in to image pixel I to be judged (i, j), sets up 16 one-dimension array a[n], by comparing the gray-scale value of mid point P and its 16 discrete points that radius is 3 around, array is carried out to assignment;
3) to array a[n] summation, and if be more than or equal to 15 or be less than or equal to-15, think that this point is candidate feature point, C (i, j) in signature matrix is set to 1, and if equal 0, think that this point is not candidate feature point, carries out next pixel judgement;
4) if do not meet the 3rd) condition of step, the first two number a[1 of array relatively] and a[16], if both identical ring shifts of carrying out a time compare after displacement again, until both enter next step after not identical;
5) judge whether the array after displacement exists continuous 9 " 1 " or continuous 9 " 1 ", if array meets above-mentioned situation and thinks that this point is candidate feature point, by C (i in signature matrix, j) be set to 1, otherwise P moved to next pixel repeating step 2)-5);
Repeating step 1)-5) each full figure pixel is traveled through one time, the position that in Matrix C (i, j), element is 1 is characteristic point position in figure.
3. a kind of city detection method of preferably integrating based on feature locations as claimed in claim 1 or 2, is characterized in that, in step 3, position feature point screening in city adopts following methods:
First, construct a scoring function V as criterion, for rejecting the unique point of non-maximum value in subrange;
Secondly, at all candidate feature points that extract, concentrate, choose arbitrarily a candidate feature point, establishing this point is P, and its scoring function response is V
pif the 3*3 scope internal memory centered by P is at any point q ∈ M, its scoring function response is V
q, work as V
p>=V
qtime, P retains as unique point, otherwise P is just disallowable;
Finally, establish arbitrary characteristics point P
ineighborhood N
piexpression is with P
icentered by the radius border circular areas that is M, define constraint criterion and be:
card{p
j|p
j∈N
Pi}>δ
Wherein card{.} represents to gather the number of interior element, with P
icentered by neighborhood N
pitotal number of middle unique point, δ is given threshold value, meets the unique point P of above-mentioned criterion
ito be retained, ungratified disallowable.
4. a kind of city detection method of preferably integrating based on feature locations as claimed in claim 1 or 2, is characterized in that, the method for carrying out binaryzation in step 5 is as follows:
1) intermediate value of choosing high-high brightness and minimum brightness in image is as initial value Th;
2) use Th to cut apart image, all pixel composition I of brightness value>=Th
1, all pixel composition I of brightness value L EssT.LTssT.LT Th
2;
3) the average brightness value μ of pixel within the scope of calculating I1 and I2
1and μ
2;
4) calculate according to the following formula new threshold value:
5) repeat above-mentioned steps 2 to 4, until the Th that subsequent iteration obtains meets following formula, stop:
| Th
n-Th
n-1|≤ε, wherein ε is predetermined threshold value, is set as 0.1;
6) by the some assignment that is greater than threshold value Th, being 1, is 0 by the some assignment that is less than threshold value, and to binary map carry out connected component labeling and by connected domain area be less than certain threshold value connected domain reject.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201410038043.9A CN103761526B (en) | 2014-01-26 | 2014-01-26 | Urban area detecting method based on feature position optimization and integration |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201410038043.9A CN103761526B (en) | 2014-01-26 | 2014-01-26 | Urban area detecting method based on feature position optimization and integration |
Publications (2)
Publication Number | Publication Date |
---|---|
CN103761526A true CN103761526A (en) | 2014-04-30 |
CN103761526B CN103761526B (en) | 2017-04-12 |
Family
ID=50528762
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201410038043.9A Active CN103761526B (en) | 2014-01-26 | 2014-01-26 | Urban area detecting method based on feature position optimization and integration |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN103761526B (en) |
Cited By (10)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN105069463A (en) * | 2015-07-17 | 2015-11-18 | 重庆交通大学 | Object-oriented multiple scale mountainous city land coverage information obtaining method |
CN106067172A (en) * | 2016-05-27 | 2016-11-02 | 哈尔滨工程大学 | A kind of underwater topography image based on suitability analysis slightly mates and mates, with essence, the method combined |
CN106447675A (en) * | 2016-10-09 | 2017-02-22 | 福州大学 | Target segmentation method based on prior shape and cyclic shift |
CN108009556A (en) * | 2017-12-23 | 2018-05-08 | 浙江大学 | A kind of floater in river detection method based on fixed point graphical analysis |
CN108230240A (en) * | 2017-12-31 | 2018-06-29 | 厦门大学 | It is a kind of that the method for position and posture in image city scope is obtained based on deep learning |
CN108898603A (en) * | 2018-05-29 | 2018-11-27 | 北京佳格天地科技有限公司 | Plot segmenting system and method on satellite image |
CN109060285A (en) * | 2018-09-10 | 2018-12-21 | 西安交通大学 | A kind of detection device and method of helical spring dynamic vibration characteristics |
CN111598101A (en) * | 2020-05-25 | 2020-08-28 | 中国测绘科学研究院 | Urban area intelligent extraction method, system and equipment based on remote sensing image scene segmentation |
CN111986300A (en) * | 2020-09-03 | 2020-11-24 | 贝壳技术有限公司 | Rendering point determination method and device for house decoration, storage medium and electronic equipment |
CN112734844A (en) * | 2021-01-08 | 2021-04-30 | 河北工业大学 | Monocular 6D pose estimation method based on octahedron |
Family Cites Families (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN102176014B (en) * | 2011-01-19 | 2013-02-13 | 西安理工大学 | Method for detecting urban region change based on multi-temporal SAR (synthetic aperture radar) images |
CN103258324B (en) * | 2013-04-02 | 2015-09-30 | 西安电子科技大学 | Based on the method for detecting change of remote sensing image that controlled kernel regression and super-pixel are split |
-
2014
- 2014-01-26 CN CN201410038043.9A patent/CN103761526B/en active Active
Cited By (15)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN105069463A (en) * | 2015-07-17 | 2015-11-18 | 重庆交通大学 | Object-oriented multiple scale mountainous city land coverage information obtaining method |
CN106067172A (en) * | 2016-05-27 | 2016-11-02 | 哈尔滨工程大学 | A kind of underwater topography image based on suitability analysis slightly mates and mates, with essence, the method combined |
CN106447675A (en) * | 2016-10-09 | 2017-02-22 | 福州大学 | Target segmentation method based on prior shape and cyclic shift |
CN106447675B (en) * | 2016-10-09 | 2019-05-10 | 福州大学 | Target Segmentation method based on prior shape and cyclic shift |
CN108009556B (en) * | 2017-12-23 | 2021-08-24 | 浙江大学 | River floating object detection method based on fixed-point image analysis |
CN108009556A (en) * | 2017-12-23 | 2018-05-08 | 浙江大学 | A kind of floater in river detection method based on fixed point graphical analysis |
CN108230240A (en) * | 2017-12-31 | 2018-06-29 | 厦门大学 | It is a kind of that the method for position and posture in image city scope is obtained based on deep learning |
CN108230240B (en) * | 2017-12-31 | 2020-07-31 | 厦门大学 | Method for obtaining position and posture in image city range based on deep learning |
CN108898603A (en) * | 2018-05-29 | 2018-11-27 | 北京佳格天地科技有限公司 | Plot segmenting system and method on satellite image |
CN109060285A (en) * | 2018-09-10 | 2018-12-21 | 西安交通大学 | A kind of detection device and method of helical spring dynamic vibration characteristics |
CN111598101A (en) * | 2020-05-25 | 2020-08-28 | 中国测绘科学研究院 | Urban area intelligent extraction method, system and equipment based on remote sensing image scene segmentation |
CN111986300A (en) * | 2020-09-03 | 2020-11-24 | 贝壳技术有限公司 | Rendering point determination method and device for house decoration, storage medium and electronic equipment |
CN111986300B (en) * | 2020-09-03 | 2024-04-19 | 贝壳技术有限公司 | Rendering point determining method and device for house decoration, storage medium and electronic equipment |
CN112734844A (en) * | 2021-01-08 | 2021-04-30 | 河北工业大学 | Monocular 6D pose estimation method based on octahedron |
CN112734844B (en) * | 2021-01-08 | 2022-11-08 | 河北工业大学 | Monocular 6D pose estimation method based on octahedron |
Also Published As
Publication number | Publication date |
---|---|
CN103761526B (en) | 2017-04-12 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN103761526A (en) | Urban area detecting method based on feature position optimization and integration | |
Lee et al. | Robust concrete crack detection using deep learning-based semantic segmentation | |
CN106127204B (en) | A kind of multi-direction meter reading Region detection algorithms of full convolutional neural networks | |
CN103839065B (en) | Extraction method for dynamic crowd gathering characteristics | |
CN108038445B (en) | SAR automatic target identification method based on multi-view deep learning framework | |
CN110310264A (en) | A kind of large scale object detection method, device based on DCNN | |
CN103049763B (en) | Context-constraint-based target identification method | |
CN109409263A (en) | A kind of remote sensing image city feature variation detection method based on Siamese convolutional network | |
CN108596108B (en) | Aerial remote sensing image change detection method based on triple semantic relation learning | |
CN104408481B (en) | Classification of Polarimetric SAR Image method based on depth wavelet neural network | |
CN103177458B (en) | A kind of visible remote sensing image region of interest area detecting method based on frequency-domain analysis | |
CN104392228A (en) | Unmanned aerial vehicle image target class detection method based on conditional random field model | |
CN108647602B (en) | A kind of aerial remote sensing images scene classification method determined based on image complexity | |
CN105975929A (en) | Fast pedestrian detection method based on aggregated channel features | |
CN112560675B (en) | Bird visual target detection method combining YOLO and rotation-fusion strategy | |
CN109117836A (en) | Text detection localization method and device under a kind of natural scene based on focal loss function | |
CN104537647A (en) | Target detection method and device | |
CN104217438A (en) | Image significance detection method based on semi-supervision | |
CN107507170A (en) | A kind of airfield runway crack detection method based on multi-scale image information fusion | |
CN103218831A (en) | Video moving target classification and identification method based on outline constraint | |
CN112232328A (en) | Remote sensing image building area extraction method and device based on convolutional neural network | |
CN104361351A (en) | Synthetic aperture radar (SAR) image classification method on basis of range statistics similarity | |
CN103927759A (en) | Automatic cloud detection method of aerial images | |
CN106709515A (en) | Downward-looking scene matching area selection criteria intervention method | |
CN103268498A (en) | Method for area-of-interest blurred image semantic comprehension |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
C06 | Publication | ||
PB01 | Publication | ||
C10 | Entry into substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
GR01 | Patent grant | ||
GR01 | Patent grant |