CN103106658A - Island or reef coastline rapid obtaining method - Google Patents
Island or reef coastline rapid obtaining method Download PDFInfo
- Publication number
- CN103106658A CN103106658A CN201310024820XA CN201310024820A CN103106658A CN 103106658 A CN103106658 A CN 103106658A CN 201310024820X A CN201310024820X A CN 201310024820XA CN 201310024820 A CN201310024820 A CN 201310024820A CN 103106658 A CN103106658 A CN 103106658A
- Authority
- CN
- China
- Prior art keywords
- image
- island
- point
- reef
- mean shift
- 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.)
- Pending
Links
Images
Abstract
The invention relates to an island or reef coastline rapid obtaining method. The method includes: firstly constructing low-resolution images for raw images, conducting Mean Shift processing, obtaining the processed low-resolution images, making tags, and obtaining the coastline by a region growing method. The Mean Shift algorithm is introduced to use color information of color images to obtain island (reef) coastline effectively and lower images of noise information such as waves in the images. Coastline obtaining speed and effect are improved due to good connectivity of ocean and pixel tags. The Mean Shift algorithm is capable of making full use of the color information and suitable for coastline obtaining of remote-sensing images.
Description
Technical field
The present invention relates to a kind of island, reef water front rapid extracting method.
Background technology
1, prior art situation
Yang Qi etc. have carried out identification to Zhejiang Sea Area TM data islands information and have extracted, and have obtained good effect.Zhijun Liu etc. utilizes TM fused data and spot data to carry out the extraction experiment on unoccupied island, and has calculated the area on island.
The technology such as Lee and Jurkevich utilize that image is level and smooth, Threshold segmentation and border tracking are extracted the shore line in the SAR image.The people such as Giancarlo utilize the gray scale of image and texture to realize reaching cutting apart of image the purpose of extracting the shore line, but because bearing accuracy is not high slower with processing speed, thereby practicality is bad.The method that Hongxing Liu etc. are cut apart by adaptive threshold has accurately been extracted the shore line in the Antarctica, provides benchmark for follow-up variation detects research.
The interferometry of at first utilizing Marcus Schwabisch etc. realizes the correction of SAR image, detects on this basis the position obtain larger interference values gradient on image, thus extract land and and water body between the border.
AndreasNiedermeier etc. utilize area tracking, Wavelet Edge Detection and Snake algorithm to realize the extraction of flowage line, and use on this basis identical tidal conditions and the SAR image under mutually when identical, the dynamic change in shore line, estuary zone is detected.
The high resolution spatial panchromatic image that Zhang Yong continues etc. and to utilize remote sensing satellite IKONOS to provide considers spectral information and spatial information, has proposed a kind of based on related information of neighborhood and based on the fast automatic extraction shore line method of two-dimentional inter-class variance principle.
Open towards the male engagement aberration theoretical, Canny algorithm and comprehensive multiple dimensioned morphology extraction algorithm have been improved, proposed shore line extraction algorithm based on aberration Canny, Morphological Chromatic Aberration for RS Color Image, and the relative merits of the algorithm realized have been carried out analyzing relatively.
Recklessly positive equality proposes the region growing Image Segmentation algorithm described based on interactive support vector territory, and this algorithm utilizes the training sample of the sub-block structure SVDD of the sub-block of target area and nontarget area; Then utilize these known training sample training support vector domain classifiers; By a Seed Points in interactive mode select target zone, in the region growing process, utilization training gained builds and increases rule, has overcome to a certain extent the rule of growing in classical region growing algorithm and has chosen hard problem.
Xie Minghong etc. have provided a kind of shore line fast automatic Extraction Algorithm based on the thought of seeds growing, and this algorithm utilizes pixel value statistical information automatically one of location initial seed point zone, and calculates initial average and initial threshold.Then carry out the growth of marine site point based on the initial average of constantly updating and initial threshold.After increase finishing, follow the tracks of and determine concrete position, shore line thereby the profile border is carried out in the connection marine site that obtains.
Jing Hao etc. are for the instability of gray feature in remote sensing image, especially the SAR(synthetic-aperture radar) the weak contrast of land and water gray scale in image, use for reference the initiatively thought of profile, realized the extraction in shore line.
Wang Changying has proposed the shore line extracting method based on correlation rule, its basic thought is: at first extract the characteristic attribute (spectral properties, texture properties and basic statistics amount attribute) of sample in the survey region remote sensing image, then the applicating category association rule algorithm carries out association rule mining; The correlation rule that will excavate out at last is applied to whole survey region data centralization and carries out the separation of extra large land, thereby realizes the extraction in shore line.
2, the technical matters and the defective that exist
In the last thirty years, fast development along with spationautics, sensor technology, computer technology and related discipline thereof, the satellite imagery Intelligence Technology is rapidly improved, it has all obtained huge progress at aspects such as spectral resolution, spatial resolution, temporal resolutions, has formed high spectrum, high spatial resolution, round-the-clock, round-the-clock, real-time/quasi real time earth observation ability.
The increase of remote sensing image data amount, the raising of resolution provides better data to prepare to remote sensing image island (reef) water front; Simultaneously quantity of information is greatly abundant, makes when sufficient useful information is provided, and disturbing factor also further increases, and as the wave in the waters, boats and ships etc., this correct extraction to island (reef) and water front thereof is had higher requirement.
Current disposal route is based on panchromatic image mostly, is also to process after being converted into panchromatic image for the processing of chromatic image.This is for chromatic image, and its chromatic information just is not fully utilized, and the removal that the needs of the result of extracting according to rim detection carry out false water front edge is connected with the edge, the process more complicated.
Summary of the invention
The purpose of this invention is to provide a kind of island, reef water front rapid extracting method, can not take full advantage of the problem of chromatic image information in order to solve existing extracting method.
For achieving the above object, the solution of the present invention is: a kind of island, reef water front rapid extracting method comprise that step is as follows:
Step 1) is for raw video G
0, build low resolution image G
1,
In formula (15), 0≤i<R
1, 0≤j<C
1, R
1And C
1Be respectively line number and the columns of low resolution image;
Adopt 5 * 5 window functions,
Step 2), select a kind of kernel function, to low resolution image G
1Carry out Mean Shift and process, the low resolution image G after being processed
1';
Step 4) is to raw video G
0Connected domain estimate: if certain 1 a in image
0With the some a around it
i(i=1 ~ 8) all satisfy condition: || a
i-a
0‖
2≤ h
r, think a
iHomogeney is satisfied in the zone that (i=0 ~ 8) surround, and regional location is carried out mark; After mark is completed, at raw video G
0In the pixel that is not labeled is carried out Mean Shift calculation process, search its modal point, and the value of modal point be assigned to the image G that interpolation builds
0The pixel of ' relevant position is until all pixels that are not labeled are disposed;
Step 5) is carried out the marine site with the region growing method and is extracted: determines initial seed point, selects adjacent with Seed Points eight to be communicated with or four connected regions, carry out region growing, extraction island, reef water front.
In step 1), at first raw video is carried out Gassian low-pass filter, and then build.
Step 2) in, described kernel function comprises: homogeneous nucleus function, normal state kernel function, Epanechnikov kernel function.
Carry out the extraction in shore line, island (reef) for effectively using the color information of chromatic image, and the image of the noise information such as wave in the reduction image, Mean Shift algorithm is introduced into, the connectedness of utilizing simultaneously the marine site is characteristics preferably, by the mode of element marking, shore line extraction rate and effect have been improved.The characteristic of Mean Shift algorithm can take full advantage of color information, is fit to the extraction in remote sensing image shore line.
About understanding the application of concepts more of the present invention and Mean Shift algorithm, the below is introduced:
The general introduction of Mean Shift algorithm
Mean Shift algorithm is a kind of nonparametric statistical method, modal point by the phase plesiotype of search and the probability distribution of samples points is realized, needs hardly priori, mainly relies on sample analysis, and its speed of convergence is very fast, is a focus of image processing area research.Mean Shift algorithm is proposed in 1975 by Fukunaga and Hostetler at first, and uses it for the estimation of density gradient function, by continuous iterative process, searches its modal point.The initial implication of Mean Shift represents the mean vector that is offset, but the development along with Mean Shift theory, variation has also occured its implication: generally its representative is the process of an iteration, namely first calculate the mean-shift vector of current point, then this point is moved to the position that this offset vector points to, then proceed offset operation as new starting point, until satisfy corresponding end condition.
Mean Shift vector and kernel function
Suppose given d dimension space X={x
i| i=1.2....n}, sample x take from d dimension space X, and the citation form of the Mean Shift vector of x can be expressed as follows by 1.1 formulas so:
S wherein
hBe the sample set zone of satisfying formula (2), k represents to fall into regional S
hThe number of middle sample, x
i-x representation feature vector is with respect to the side-play amount of x.We claim vector M
h(x) be the average drifting vector, its expression falls into regional S
hA middle k sample point is with respect to the mean deviation amount of x.
S
h(x)≡{y:(y-x)
T(y-x)≤h
2} (2)
Can find out from formula (1), the mean-shift vector that it is represented is for falling into regional S
hIn each sample point, no matter how many sample points is from the distance of x, they are the same to the contribution of mean-shift vector.Usually introduce the concept of kernel function in the nonparametric probability density is estimated, be used for the sample point of different distance in the Expressive Features space to calculating M
h(x) time, impact, that is to say, the sample point from the distance of x close to more is more effective to the statistical property of estimating x.In addition, in feature space, the importance of different sample points is different, therefore needs to introduce a weight coefficient and is used for representing that the sample point of same distance is to M
h(x) difference of contribution.The essence of kernel function is that contribution according to each sample is weighted so, and Selection of kernel function satisfies K (x)=c usually
k,dK (x ‖
2) round symmetric kernel, kernel function commonly used has following several form:
The homogeneous nucleus function;
The normal state kernel function;
Epanechnikov core;
In addition kernel function also can by normal state kernel function and homogeneous nucleus function multiply each other and truncation obtain, its form is expressed as follows:
Consider the contribution of each sample, introduce kernel function and be different sample tax with weight factor after, the average drifting vector can be expressed as:
G in formula
H(x
i-x)=| H|
-1/2G(H
-1/2(x
i-x)).
Wherein H represents bandwidth matrices, is the symmetric matrix of a positive definite, and it has determined the coverage of kernel function.In order to reduce the complexity of calculating, generally select the ratio battle array H=h of diagonal matrix or unit matrix
2I is as bandwidth matrices; w(x
iThe weight that) 〉=0 expression is given each sample point.
After introducing single fixed coefficient h and kernel function, the extend type of Mean Shift vector can be expressed as:
Principle and the computation process of Mean Shift algorithm
The basic thought of Mean Shift algorithm is: given one initial sample, carry out the local stationary point of the core estimation of iterative search sample distribution along normalized probability density gradient direction, when being used for mode and selecting, the mode that the point that utilizes this algorithm to restrain to obtain will be selected exactly; When being used for cluster analysis, the point that needs to converge to identical or close stable point is classified as a class; Iteration convergence to stable point can also regard optimized result as.
Mean shift algorithm is mainly used to carry out searching and the cluster of modal point.This method is estimated as the basis with probability density, generally we need to seek the place (being modal point) of probability density maximum, it is zero position that modal point is present in density gradient, therefore do not need estimated probability density in actual the execution, calculating Mean Shift vector but be translated into, is that use in zero place by seeking density gradient.Its principle is: regard feature space as posterior probability density function, the local maximum place of probability density function (i.e. unknown modal point) is corresponding with the dense zone in feature space so.If determined the position of modal point, just relative cluster can be depicted by the partial structurtes of feature space.
A kind of probability density method of estimation relatively more commonly used of Density Estimator is called again Parzen window method, can converge on progressively probability density function arbitrarily in the sufficient situation of sampling.For the estimation of certain sampled point probability density function, it utilizes the local average value of a function centered by this sampled point to replace.
The first step that probability density f (x) carries out feature space analysis is exactly the searching modes point, and modal point is that probability density f (x) is the place of extreme point, and its density gradient is zero herein, namely
Can utilize so the gradient calculation of density Estimation as the estimation of density gradient, the gradient function of its density Estimation is as follows:
Defined function g (x)=-k'(x), wherein suppose section function k x ∈ (0, ∞) can lead, g (x) as new section function, is made G (x)=c
d,g(|| x ‖
2), G (x) can regard new section function, wherein c as
d,gBe normalization coefficient, substitution formula so (10) can obtain:
Wherein suppose
Be positive number, in following formula, first is proportional to x and sentences the density Estimation that G is core
And second be exactly Mean Shift vector, can regard the poor of weighted mean centered by x and x as, that is:
So easily draw:
The mean shift vector that can find out the x place always is proportional to Density gradient estimation, offset vector points to the larger direction of core probability density, local mean value always moves towards the zone at great majority point place, that is to say the sample for the probability density function of obeying certain distribution, at regional S
hInterior sample point, in the zone of high probability density, sample distribution is more intensive, therefore the average larger zone of offset vector sensing density.
First of modus ponens (12) the right is designated as m
H, G(x), that is:
A given initial point x, kernel function G (x) and allowable error
Can utilize in the following way so Mean Shift algorithm to search its modal point:
Step 1: calculate m
H, G(x) value;
Step 2: with m
H, G(x) be assigned to x, and again calculate m
H, G(x) value;
Know m by formula (12) and formula (14)
H, G(x)=M
h,G(x)+x, therefore above process is exactly the continuous process that moves along the gradient direction of probability density, can converge near the peak value (modal point) of this point under certain condition.
The region growing method:
The region growing method can be used for extracting the marine site of island (reef) image, and then island (reef) extracted.Thereby its aggregates into larger zone according to the criterion of definition in advance with the pixel in the image waters or subregion and realizes cutting apart.Its basic thought is exactly at first to choose certain point (certain zone) in the waters as known Seed Points (seed region), search for the point of adjacent with Seed Points (four are communicated with or eight connections), if the not accessed mistake of these consecutive point and satisfy condition of growth, so it is carried out stacked processing (state that will put changes to consistent with initial seed point), and with this point as new Seed Points, then continuing cyclic search goes down, until during the pixel that does not satisfy condition in stack, region growing finishes.
Usually face three problems when the practical application region-growing method:
1) one of selection or one group of point that can correctly represent area-of-interest are as Seed Points or seed region.Sub pixel is decided according to particular problem usually, but it is chosen and generally should satisfy following three requirements: must be within the target area, and keep clear of the border of target; For growing larger zone, its core and gray scale on every side thereof should be relatively even; The gray scale of core can reflect the feature of target area.When there is no priori for concrete problem, can utilize the growth criterion that each pixel in image is calculated accordingly, the target area can produce the effect of cluster so, and this moment, sub pixel was got the pixel near cluster centre.
2) how threshold value is chosen.The region growing threshold value is extremely important in area growth process, it choose whether proper size and the homogeneity that can directly affect growth district.
Determining of growth rule namely determined rule or the condition that neighbor can be comprised the criterion of coming in and allow growth course stop in growth course.The general problem that needs to consider three aspects: of the choosing of rule of growing: the kind of particular problem itself, view data used, the connectedness between pixel and propinquity, otherwise insignificant segmentation result appears sometimes.
Gaussian filtering:
Gaussian filter is that a class is selected the linear smoothing wave filter of weights according to the shape of Gaussian function actually.The thought of linear filter is exactly to realize the smothing filtering of image with the weighted sum of pixel in the continuous lights function.As the linear smoothing wave filter of a quasi-representative, the weights of Gaussian filter template are to select according to the shape of Gaussian function, and its shape, namely the precipitous degree of peak value is to be decided by scale parameter σ.Obviously, σ is larger, and level and smooth degree is larger, otherwise less.Desirable filter effect be filter smoothing be only noise signal, therefore, the scale parameter σ that noise pixel signal point is corresponding should get greatly as far as possible, and smoothly distinguishes and the σ of marginarium goes to zero as far as possible.
Wherein, S (x, y) for image through the signal after gaussian filtering, the i.e. convolution of picture signal and Gaussian filter; I (x, y) is picture signal, is combined by original signal f (x, y) and noise signal n (x, y).
Description of drawings
Fig. 1 is homogeney zone marker schematic diagram;
Fig. 2 is raw video 1;
Fig. 3 is raw video 2;
Fig. 4 is that image 1 is directly used the extraction result of Mean Shift;
Fig. 5 is that the Mean Shift after image 1 improves extracts result;
Fig. 6 is that image 1 improves front profile edge stack result;
Fig. 7 is the contour edge stack result after image 1 improves;
Fig. 8 is that image 2 is directly used the result of Mean Shift;
Fig. 9 is that the Mean Shift after image 2 improves extracts result;
Figure 10 is water front stack result before image 2 improves;
Figure 11 is the water front stack result after image 2 improves;
Figure 12 is that image 1 is directly used the demonstration of MeanShift methods and results details;
Figure 13 is that after image 1 improves, the stack result details shows;
Figure 14 is that image 2 is directly used the demonstration of MeanShift methods and results details;
Figure 15 is that after image 2 improves, the stack result details shows;
Figure 16 is process flow diagram of the present invention;
Figure 17 is the situation schematic diagram of two kinds of search neighborhoods, and a is cubic to situation, and b is from all directions to situation.
Embodiment
The present invention will be further described in detail below in conjunction with accompanying drawing.
Mean Shift island (reef) water front rapid extracting method based on element marking
Because Mean Shift algorithm is a kind of statistics alternative manner, often need to carry out repeatedly interative computation in order to obtain higher computational accuracy, suppose that the calculation cost to the iteration each time of each data is O (N), therefore, the calculation cost of the whole data acquisition of cluster is O (kN
2), wherein N is the pixel count in image, and k is the mean iterative number of time of each pixel, and as seen its calculation cost is sizable.In order better to adapt to the extraction in shore line, this method is first carried out Mean Shift interative computation and is processed on the low resolution image, utilize marine site regional connectivity characteristic preferably characteristics high resolution image is carried out mark, do further processing, thereby improved speed and stability that island (reef) water front extracts.Extract flow process and see accompanying drawing 16, main process is described as follows:
Build the low resolution image
Much smaller than original high resolution image, but it can keep the overall profile of raw video to the low resolution image preferably on data volume, therefore deals with very fast.For Enhancement Method noiseproof feature, raw video is first made Gassian low-pass filter, then builds.
If raw video is G
0, newly-generated low resolution image is G
1, low resolution image building process is
0≤i<R wherein
1, 0≤j<C
1, R
1And C
1Be respectively line number and the columns of low resolution image;
Be 5 * 5 window functions, expression formula is as follows:
The structure of low resolution image is actually raw video is first carried out Gassian low-pass filter one time so, then with all even number line even column deletions, resampling generates, and the size of easily calculating newly-generated low resolution image is 1/4 of raw video size.
Utilize Mean Shift algorithm to carry out filtering on the low resolution image
Select corresponding kernel function, the low resolution image is carried out Mean Shift process.Due to the reduction of resolution, make the processing number of pixel greatly reduce, so also can have substantial degradation accordingly in the processing time.Simultaneously when setting up the low resolution image, raw video is carried out Gassian low-pass filter, therefore made the noiseproof feature of algorithm further promote.
The filtering that low resolution image after comprehensive utilization is processed and raw video are realized raw video
1), with image after the low resolution image interpolarting structure filtering after processing
If the low resolution image after processing is G
1', the image that interpolation builds is G
0', at first to image G
1' expand as original twice on each dimension, newly-increased row (even number line) and row (even column) are filled with 0, then carry out convolution algorithm with corresponding wave filter, and former image is estimated.Here adopt Gaussian filter, the image G that interpolation builds
0' and filtered low resolution image G
1' between the pass be:
2), connected domain is estimated:
By filtered low resolution image G
1', to raw video G
0Connected domain estimate (connected domain refers to the identical zone of character in image).As shown in Figure 1:
For certain the some z in image and some a, definition:
If certain 1 a in image
0With the some a around it
i(i=1 ~ 8) all satisfy condition: || a
i-a
0‖
2≤ h
r, we think a so
iHomogeney is satisfied in the zone that (i=0 ~ 8) surround, and regional location is carried out mark.
3), further process:
After estimation end to connected region, we think that the pixel of these connected regions meets the demands, and these pixels will be not processed.According to the mark situation of regional location, at raw video G
0In the pixel that is not labeled is carried out Mean Shift calculation process, search its modal point, and the value of modal point be assigned to the image G that interpolation builds
0The pixel of ' relevant position until the pixel that all are not labeled search is complete, obtains image F after processing.
Utilizing region growing to carry out the marine site extracts
The region growing method generally has two kinds of methods to determine initial seed point: (1) determines initial seed point (zone) by user-interactive, and the method has the characteristics of quicklook; (2) follow the relevant gray-scale statistical characteristics of the concrete image of certificate by computing machine automatic searching initial seed point (zone).Through island (reef) image that previous methods was processed, the gray-scale value of silhouette target tends towards stability, and edge details is able to good reservation, has passed through again filtering operation, and noiseproof feature is better, and inside, marine site is relatively level and smooth, so the selection of Seed Points is relatively free.
Take in the marine site as example, because the marine site has connectedness preferably, the gray-scale value that presents in image is basic identical.Select adjacent with Seed Points eight to be communicated with or four connected regions, carry out region growing.For Seed Points a and certain neighborhood point z, the region growing rule adopts following condition to represent:
T
0Be growing threshold.If neighborhood point z satisfies above condition, z is carried out stacked processing.The search of neighborhood territory pixel is divided into the four directions to search with from all directions to search.The four directions thinks in search, only have center pixel up and down the neighbor of four direction be connected with center pixel, upper left, lower-left, upper right, the pixel that the bottom right four direction is adjacent are not communicated with center pixel.And think in search from all directions, on 8 directions of center pixel, adjacent pixel all is connected with center pixel.The four directions is 0,2,4,6 to middle communication direction if describe with Direction code, and from all directions in the search communication direction be 0-7.Be the situation schematic diagram of two kinds of search neighborhoods as shown in figure 17, wherein X represents disconnected pixel, and a is cubic to situation, and b is from all directions to situation.
After the neighborhood territory pixel judgement of Seed Points a finishes, centered by the pixel of new merging, proceed the neighborhood judgement as new Seed Points, check the neighborhood of new pixel, until the zone can not further be expanded, when in stack, all Seed Points judgements finish, complete the extraction in marine site, realize cutting apart of marine site and island (reef), obtain cutting apart image G.
Experiment and result
For the validity of checking this method, to choose two width images and be analyzed respectively as experimental data, image 1 size is: 822 * 1076, image 2 sizes are: 880 * 890.
Wherein Fig. 2 and Fig. 3 are for being respectively raw video 1 and image 2, Fig. 4, Fig. 5 and Fig. 8, Fig. 9 are respectively image 1 and image 2 is directly used the method for Mean Shift and utilizes and carry out based on the Mean Shift method of element marking the result that Hai Lu is cut apart acquisition, from result, directly use the method for Mean Shift, can realize correctly cutting apart of Hai Lu, but still have more isolated area in segmentation result, noiseproof feature relatively a little less than; And utilize when carrying out correctly realizing that Hai Lu is cut apart with the Mean Shift method of element marking, the impact that the noises such as wave in image are caused has good noise immunity.
Can see from the details of Figure 12, Figure 13, Figure 14, Figure 15 stack result shows, utilize the MeanShift shore line rapid extracting method of element marking better to process details, but all can have " burr " in two kinds of testing results.This is that impact due to image noise and image in small, broken bits makes always adhere to " branch " in the contour edge that extracts, and therefore also needs to be further processed.
Working time, contrast was as follows:
Time contrast used before and after table 1 improves
Contrast from top table 1 can be found out due to the introducing of low resolution image and the mark of homogeneous region, greatly shortened the detection time of the Mean Shift water front rapid extracting method that utilizes element marking, and in the testing result after improving, isolated area obviously reduces, method noiseproof feature after explanation improves on the one hand is better, and the minimizing of isolated area is highly beneficial to the elimination of follow-up isolated area on the other hand.
Claims (3)
1. an island, reef water front rapid extracting method, is characterized in that, comprises that step is as follows:
Step 1) is for raw video G
0, build low resolution image G
1,
In formula (15), 0≤i<R
1, 0≤j<C
1, R
1And C
1Be respectively line number and the columns of low resolution image;
Adopt 5 * 5 window functions,
Step 2), select a kind of kernel function, to low resolution image G
1Carry out Mean Shift and process, the low resolution image G after being processed
1';
Step 4) is to raw video G
0Connected domain estimate: if certain 1 a in image
0With the some a around it
i(i=1 ~ 8) all satisfy condition: || a
i-a
0‖
2≤ h
r, think a
iHomogeney is satisfied in the zone that (i=0 ~ 8) surround, and regional location is carried out mark; After mark is completed, at raw video G
0In the pixel that is not labeled is carried out Mean Shift calculation process, search its modal point, and the value of modal point be assigned to the image G that interpolation builds
0The pixel of ' relevant position is until all pixels that are not labeled are disposed;
Step 5) is carried out the marine site with the region growing method and is extracted: determines initial seed point, selects adjacent with Seed Points eight to be communicated with or four connected regions, carry out region growing, extraction island, reef water front.
2. island according to claim 1, reef water front rapid extracting method, is characterized in that, in described step 1), at first raw video carried out Gassian low-pass filter, and then build.
3. island according to claim 1, reef water front rapid extracting method, is characterized in that described step 2) in, described kernel function comprises: homogeneous nucleus function, normal state kernel function, Epanechnikov kernel function.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201310024820XA CN103106658A (en) | 2013-01-23 | 2013-01-23 | Island or reef coastline rapid obtaining method |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201310024820XA CN103106658A (en) | 2013-01-23 | 2013-01-23 | Island or reef coastline rapid obtaining method |
Publications (1)
Publication Number | Publication Date |
---|---|
CN103106658A true CN103106658A (en) | 2013-05-15 |
Family
ID=48314485
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201310024820XA Pending CN103106658A (en) | 2013-01-23 | 2013-01-23 | Island or reef coastline rapid obtaining method |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN103106658A (en) |
Cited By (12)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN104732215A (en) * | 2015-03-25 | 2015-06-24 | 广西大学 | Remote-sensing image coastline extracting method based on information vector machine |
CN106875408A (en) * | 2017-02-27 | 2017-06-20 | 网易(杭州)网络有限公司 | Method, device and terminal device for sectional drawing |
CN106875399A (en) * | 2017-01-04 | 2017-06-20 | 努比亚技术有限公司 | A kind of method for realizing interactive image segmentation, device and terminal |
CN107292882A (en) * | 2017-08-02 | 2017-10-24 | 国网电力科学研究院武汉南瑞有限责任公司 | One kind is based on the adaptive electrical equipment malfunction detection methods of Meanshift |
CN109190538A (en) * | 2018-08-24 | 2019-01-11 | 华北水利水电大学 | A kind of fully-loaded stream river deposition coast band evolution analysis method based on remote sensing technology |
CN109284663A (en) * | 2018-07-13 | 2019-01-29 | 上海大学 | A kind of sea obstacle detection method based on normal state and uniform Mixture Distribution Model |
CN110956182A (en) * | 2019-09-23 | 2020-04-03 | 四创科技有限公司 | Method for detecting water area shoreline change based on deep learning |
CN111046772A (en) * | 2019-12-05 | 2020-04-21 | 国家海洋环境监测中心 | Multi-temporal satellite remote sensing island shore line and development and utilization information extraction method |
CN111879292A (en) * | 2020-07-27 | 2020-11-03 | 深圳大学 | Coastline dynamic monitoring method, coastline dynamic monitoring equipment and storage medium |
CN112183473A (en) * | 2020-10-28 | 2021-01-05 | 电子科技大学 | Geological curved surface visual semantic feature extraction method |
CN112365469A (en) * | 2020-11-11 | 2021-02-12 | 中国船舶科学研究中心 | Two-dimensional wave spectrum segmentation method based on Otsu threshold value |
CN113158840A (en) * | 2021-04-01 | 2021-07-23 | 中国科学院南京地理与湖泊研究所 | Inland river shoreline resource utilization type extraction method based on high-definition remote sensing image |
Citations (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN102589528A (en) * | 2012-01-19 | 2012-07-18 | 中国测绘科学研究院 | Multi-temporal imaging island shoreline surveying method |
-
2013
- 2013-01-23 CN CN201310024820XA patent/CN103106658A/en active Pending
Patent Citations (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN102589528A (en) * | 2012-01-19 | 2012-07-18 | 中国测绘科学研究院 | Multi-temporal imaging island shoreline surveying method |
Non-Patent Citations (1)
Title |
---|
张宏伟: "基于遥感影像的海岛礁提取技术研究", 《中国优秀硕士学位论文全文数据库》, 15 July 2012 (2012-07-15), pages 4 - 2 * |
Cited By (16)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN104732215A (en) * | 2015-03-25 | 2015-06-24 | 广西大学 | Remote-sensing image coastline extracting method based on information vector machine |
CN106875399A (en) * | 2017-01-04 | 2017-06-20 | 努比亚技术有限公司 | A kind of method for realizing interactive image segmentation, device and terminal |
CN106875408A (en) * | 2017-02-27 | 2017-06-20 | 网易(杭州)网络有限公司 | Method, device and terminal device for sectional drawing |
CN107292882B (en) * | 2017-08-02 | 2019-09-17 | 国网电力科学研究院武汉南瑞有限责任公司 | One kind being based on the adaptive electrical equipment malfunction detection method of Meanshift |
CN107292882A (en) * | 2017-08-02 | 2017-10-24 | 国网电力科学研究院武汉南瑞有限责任公司 | One kind is based on the adaptive electrical equipment malfunction detection methods of Meanshift |
CN109284663A (en) * | 2018-07-13 | 2019-01-29 | 上海大学 | A kind of sea obstacle detection method based on normal state and uniform Mixture Distribution Model |
CN109190538B (en) * | 2018-08-24 | 2021-06-08 | 华北水利水电大学 | Sediment-laden river delta coastal zone evolution analysis method based on remote sensing technology |
CN109190538A (en) * | 2018-08-24 | 2019-01-11 | 华北水利水电大学 | A kind of fully-loaded stream river deposition coast band evolution analysis method based on remote sensing technology |
CN110956182A (en) * | 2019-09-23 | 2020-04-03 | 四创科技有限公司 | Method for detecting water area shoreline change based on deep learning |
CN110956182B (en) * | 2019-09-23 | 2023-04-07 | 四创科技有限公司 | Method for detecting water area shoreline change based on deep learning |
CN111046772A (en) * | 2019-12-05 | 2020-04-21 | 国家海洋环境监测中心 | Multi-temporal satellite remote sensing island shore line and development and utilization information extraction method |
CN111879292A (en) * | 2020-07-27 | 2020-11-03 | 深圳大学 | Coastline dynamic monitoring method, coastline dynamic monitoring equipment and storage medium |
CN112183473A (en) * | 2020-10-28 | 2021-01-05 | 电子科技大学 | Geological curved surface visual semantic feature extraction method |
CN112183473B (en) * | 2020-10-28 | 2023-03-24 | 电子科技大学 | Geological curved surface visual semantic feature extraction method |
CN112365469A (en) * | 2020-11-11 | 2021-02-12 | 中国船舶科学研究中心 | Two-dimensional wave spectrum segmentation method based on Otsu threshold value |
CN113158840A (en) * | 2021-04-01 | 2021-07-23 | 中国科学院南京地理与湖泊研究所 | Inland river shoreline resource utilization type extraction method based on high-definition remote sensing image |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN103106658A (en) | Island or reef coastline rapid obtaining method | |
CN108573276B (en) | Change detection method based on high-resolution remote sensing image | |
CN103400151B (en) | The optical remote sensing image of integration and GIS autoregistration and Clean water withdraw method | |
WO2017071160A1 (en) | Sea-land segmentation method and system for large-size remote-sensing image | |
CN110119728A (en) | Remote sensing images cloud detection method of optic based on Multiscale Fusion semantic segmentation network | |
CN102013015B (en) | Object-oriented remote sensing image coastline extraction method | |
CN103679675B (en) | Remote sensing image fusion method oriented to water quality quantitative remote sensing application | |
CN103578110B (en) | Multiband high-resolution remote sensing image dividing method based on gray level co-occurrence matrixes | |
CN104361582B (en) | Method of detecting flood disaster changes through object-level high-resolution SAR (synthetic aperture radar) images | |
Shu et al. | Shoreline extraction from RADARSAT-2 intensity imagery using a narrow band level set segmentation approach | |
CN103198480B (en) | Based on the method for detecting change of remote sensing image of region and Kmeans cluster | |
CN104361590A (en) | High-resolution remote sensing image registration method with control points distributed in adaptive manner | |
CN102254319A (en) | Method for carrying out change detection on multi-level segmented remote sensing image | |
CN101901343A (en) | Remote sensing image road extracting method based on stereo constraint | |
CN103984946A (en) | High resolution remote sensing map road extraction method based on K-means | |
CN107247927B (en) | Method and system for extracting coastline information of remote sensing image based on tassel cap transformation | |
CN103236063A (en) | Multi-scale spectral clustering and decision fusion-based oil spillage detection method for synthetic aperture radar (SAR) images | |
CN103473755B (en) | Based on the sparse denoising method of SAR image that change detects | |
Chen et al. | A new process for the segmentation of high resolution remote sensing imagery | |
CN104217436B (en) | SAR image segmentation method based on multiple features combining sparse graph | |
Chen et al. | An improved multi-resolution hierarchical classification method based on robust segmentation for filtering ALS point clouds | |
CN109389062A (en) | Utilize the method for High Resolution Spaceborne SAR image zooming-out lake land and water cut-off rule | |
CN103679734A (en) | Method for eyed typhoon two-dimensional surface wind field inversion on basis of SVM and PDE | |
CN103871039A (en) | Generation method for difference chart in SAR (Synthetic Aperture Radar) image change detection | |
CN102830404A (en) | Method for identifying laser imaging radar ground target based on range profile |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
C06 | Publication | ||
PB01 | Publication | ||
C10 | Entry into substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
C12 | Rejection of a patent application after its publication | ||
RJ01 | Rejection of invention patent application after publication |
Application publication date: 20130515 |