CN106683109B - SAR image segmentation method based on semantic facility random field models - Google Patents
SAR image segmentation method based on semantic facility random field models Download PDFInfo
- Publication number
- CN106683109B CN106683109B CN201611237232.4A CN201611237232A CN106683109B CN 106683109 B CN106683109 B CN 106683109B CN 201611237232 A CN201611237232 A CN 201611237232A CN 106683109 B CN106683109 B CN 106683109B
- Authority
- CN
- China
- Prior art keywords
- pixel
- subspace
- sar image
- function
- follows
- 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
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/10—Image acquisition modality
- G06T2207/10032—Satellite or aerial image; Remote sensing
- G06T2207/10044—Radar image
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/20—Special algorithmic details
- G06T2207/20112—Image segmentation details
Landscapes
- Image Analysis (AREA)
Abstract
The invention discloses a kind of SAR image segmentation methods based on semantic facility random field models.It mainly solves the problems, such as not to be able to maintain image detail information in the prior art.Implementation step are as follows: 1. are divided into mixing aggregated structure atural object subspace, structural region subspace and homogenous region subspace according to the administrative division map of SAR image, by SAR image；2. feature is extracted using bag of words in pair mixing aggregated structure atural object subspace, it is split with the method that AP is clustered；3. building semantic facility random field models are split structural region subspace and homogenous region subspace；4. the segmentation result for mixing aggregated structure atural object subspace, structural region subspace and homogenous region subspace is merged, the segmentation result of SAR image is obtained.Present invention obtains the good segmentation effects of SAR image, can be used for the semantic segmentation of SAR image.
Description
Technical field
The invention belongs to technical field of image processing, in particular to SAR image segmentation method, can be used for image classification, know
Not and detect.
Background technique
Random Fields Method is a kind of relatively popular method in SAR image segmentation.Typical Random Fields Method is Ma Er
Can husband's random field MRF model, it is a kind of generative probabilistic model.In MRF model, posterior probability is equal to likelihood probability and priori
The product of probability.Likelihood probability describes the feature of SAR image, is usually indicated with the statistical distribution of SAR image, the choosing of distribution
Select the characteristic mainly according to SAR image.Prior probability describes the spatial context information of image, usually with Gibbs distribution come
It indicates.However, needing strong dependence in the hypothesis of MRF model, and do not account for seeing in the prior model of MRF model
Correlation between measured data.
For the above-mentioned deficiency of MRF model, condition random field CRF model is applied and is given birth to, mainly by unitary potential function and
Binary potential function composition.It is a kind of model of identification, and posterior probability is directly defined as Gibbs distribution.The CRF mould
Type not only captures the information of single pixel point and the information of neighborhood territory pixel, but also captures the phase interaction between image category
With the interaction between image observation.Since CRF model has the advantage than other dividing methods, by extensively through being used for
SAR image segmentation.
But since original CRF model does not account for the characteristic of SAR image itself, in response to this problem, Zhang Peng et al. is mentioned
Go out the general CRF model improved to the unitary potential function of original CRF model, incorporates the textural characteristics of SAR image
With the statistical property of SAR image.The binary potential function of the general CRF model is using this base of a fruit function of traditional multilayer logic capture figure
As the information of context.However, the binary potential function only captures isotropic relationship in image space context, ignore
The anisotropic relationship of SAR image itself causes the loss of detailed information in segmentation result, and segmentation result does not have semanteme
Consistency influences subsequent classification, identification and detection to SAR image.
Summary of the invention
It is an object of the invention to propose a kind of based on semantic facility random field models for above-mentioned existing methods deficiency
SAR image segmentation method, with promoted SAR image segmentation effect.
Technical thought of the invention is: by being improved to general CRF model, the effect of improvement SAR image segmentation,
The anisotropic relationship of SAR image is captured according to the semantic space that the sketch map of SAR image and administrative division map form, it will be semantic empty
Between information be embedded into general CRF model, construct semantic facility random field models, implementation step is as follows:
(1) according to the administrative division map of SAR image, SAR image is divided into mixing aggregated structure atural object subspace, structural region
Subspace and homogenous region subspace；
(2) to mixing aggregated structure atural object subspace, the feature in region is extracted using bag of words, and with affine propagation AP
The method of cluster is split the atural object subspace；
(3) semantic facility random field models are constructed:
(3a) defines unitary potential function are as follows:
Wherein, us(xs,ys) it is unitary potential function, Z={ 1,2 ..., N } is the set of entire SAR image pixel, and N is
The total number of SAR image pixel；ysIt is s-th of pixel in SAR image, s ∈ Z；xsIt is the category of s-th of pixel in SAR image；
fs(ys) be pixel category feature function, p (ys) be pixel category feature probability, p (ys|xs) it is likelihood probability；K is image
The total number of classification, l ∈ { 1,2 ..., K }, δ (xs, l) and it is the first indicator function, if xs=l, then δ (xs, l)=1, if
xs≠ l, then δ (xs, l)=0；
(3b) defines binary potential function are as follows:
Wherein, ytIt is t-th of pixel in SAR image, t ∈ Z, and s ≠ t；xtIt is the category of t-th of pixel in SAR image；l
∈ { 1,2 ..., K }, m ∈ { 1,2 ..., K }, δ (xt, m) and it is the second indicator function, if xt=m, then δ (xt, m)=1, if
xt≠ m, then δ (xt, m)=0；ψst(ys,yt,φ(ys,yt)) it is mixed kernel function, φ (ys,yt) it is pixel ysWith pixel ytBetween
Semantic function, p (xs,xt|ψst(ys,yt,φ(ys,yt))) it is prior probability；
(3c) according to (3a) and (3b), the objective function for defining semantic facility random field models is as follows:
p(xs|ys)=p (ys)×p(ys|xs)×p(xs,xt|ψst(ys,yt,φ(ys,yt))) <3>
Wherein, p (ys) be pixel category feature probability, p (ys|xs) it is likelihood probability, p (xs,xt|ψst(ys,yt,φ
(ys,yt))) it is prior probability；
(4) using the objective function of (3c) obtained semantic facility random field models to structural region subspace and homogeneous area
Domain subspace is split, i.e. each pixel to structural region subspace and homogenous region subspace, the maximum of modus ponens<3>
Value, obtains the category of each pixel are as follows:
(5) segmentation result of aggregated structure atural object subspace, structural region subspace and homogenous region subspace will be mixed
It merges, obtains the segmentation result of SAR image.
The invention has the following advantages over the prior art:
The first, the present invention combines the information of SAR image pixel space and the information of semantic space, carries out SAR image
Segmentation can effectively complete the task of SAR image segmentation.
The second, the present invention captures the relationship of anisotropic in SAR image by establishing semantic facility random field models,
The region consistency of segmentation result is not only increased, and effectively remains the detailed information of image.
Detailed description of the invention
Fig. 1 is implementation flow chart of the present invention to SAR image segmentation；
Fig. 2 is the division result figure in the present invention to SAR image subspace；
Fig. 3 is the SAR image segmentation result figure for being 1 meter to Ku wave band resolution ratio with the present invention and existing method；
Fig. 4 is the SAR image segmentation result figure for being 3 meters to C-band resolution ratio with the present invention and existing method.
Specific embodiment
Referring to Fig.1, the present invention is the administrative division map according to SAR image, and SAR image is divided into mixing aggregated structure atural object
Space, structural region subspace and homogenous region subspace；Mixing aggregated structure atural object subspace is extracted using bag of words
Then the feature in region obtains the segmentation result of the atural object subspace with the method that AP is clustered；To structural region subspace and even
Matter region subspace, building semantic facility random field are split；The unitary potential function of semantic facility random field uses multinomial
The statistical property of logistic regression function and SAR image indicates；The binary potential function of semantic facility random field is used based on mixed
The multinomial logistic regression function representation of synkaryon function；Aggregated structure atural object subspace, structural region subspace will be mixed
Merge with the segmentation result of homogenous region subspace, obtain the segmentation result of SAR image, specific implementation step is as follows:
Step 1, according to the administrative division map of SAR image, SAR image is divided into mixing aggregated structure atural object subspace, structure
Region subspace and homogenous region subspace.
(1.1) IEEE Transactions on was published in 2014 according to Jie-Wu and Fang-Liu et al.
Article " Local maximal homogenous region on Geoscience and Remote Sensing magazine
search for SAR speckle reduction with sketch-based geometrical kernel
Function " proposed in model obtain SAR image sketch map；
(1.2) according to the concentration class of sketch line segment in SAR image sketch map, by sketch line be divided into aggregation sketch line and
The sketch line of non-agglomerated；
(1.3) the round primitive that a radius is the optimal concentration class section upper bound is constructed, with the circle primitive to aggregation
Sketch line is expanded, and corrodes to the line segment aggregate ecto-entad after expansion, aggregation zone is obtained in sketch map；
(1.4) to the sketch line of non-agglomerated, the geometry window that size is 5 × 5 is constructed, structural region is obtained；
(1.5) part other than aggregation zone and structural region will be removed in sketch map as can not sketch region；
(1.6) by sketch map aggregation zone, structural region and can not sketch region, be mapped in SAR image, will
SAR image is divided into mixing aggregated structure atural object subspace, structural region subspace and homogenous region subspace, as shown in Fig. 2,
Fig. 2 (a) is original SAR image, and Fig. 2 (b) is SAR image sketch map, and white area is with mixing aggregated structure in Fig. 2 (c)
Object subspace, gray area are structural region subspace, and black region is homogenous region subspace.
Step 2, the feature for extracting region using bag of words to mixing aggregated structure atural object subspace, then with affine biography
The method for broadcasting AP cluster is split the atural object subspace.
To each pixel y in mixing aggregated structure atural object subspacea, It is that mixing is poly-
The total number for collecting structure atural object subspace pixel obtains pixel y using the method that AP is clusteredaCategory xa。
Step 3, semantic facility random field models are constructed.
(3a) defines unitary potential function are as follows:
Wherein, us(xs,ys) it is unitary potential function, Z={ 1,2 ..., N } is the set of entire SAR image pixel, and N is
The total number of SAR image pixel；ysIt is s-th of pixel in SAR image, s ∈ Z；xsIt is the category of s-th of pixel in SAR image；
fs(ys) be pixel category feature function, p (ys) be pixel category feature probability, p (ys|xs) it is likelihood probability；K is image
The total number of classification, l ∈ { 1,2 ..., K }, δ (xs, l) and it is the first indicator function, if xs=l, then δ (xs, l)=1, if
xs≠ l, then δ (xs, l)=0；
(3b) defines binary potential function are as follows:
Wherein, ytIt is t-th of pixel in SAR image, t ∈ Z, and s ≠ t；xtIt is the category of t-th of pixel in SAR image；l
∈ { 1,2 ..., K }, m ∈ { 1,2 ..., K }, δ (xt, m) and it is the second indicator function, if xt=m, then δ (xt, m)=1, if
xt≠ m, then δ (xt, m)=0；ψst(ys,yt,φ(ys,yt)) it is mixed kernel function, φ (ys,yt) it is pixel ysWith pixel ytBetween
Semantic function, p (xs,xt|ψst(ys,yt,φ(ys,yt))) it is prior probability；
(3c) according to (3a) and (3b), the objective function for defining semantic facility random field models is as follows:
p(xs|ys)=p (ys)×p(ys|xs)×p(xs,xt|ψst(ys,yt,φ(ys,yt))) <3>
Wherein, p (ys) be pixel category feature probability, p (ys|xs) it is likelihood probability, p (xs,xt|ψst(ys,yt,φ
(ys,yt))) it is prior probability, it respectively indicates as follows:
Wherein, wlIt is the weight parameter of pixel,ηsIt is pixel ysNeighborhood territory pixel；Γ(αk) it is gamma letter
Number, αkFor scale parameter,μkIt is displacement parameter,ψst(ys,yt,φ(ys,yt)) be
Mixed kernel function is expressed as follows:
Wherein, ρ is weight parameter, ρ ∈ { 0,1 }, if ysIt is the pixel of homogenous region subspace, then ρ=1, if ysIt is
The pixel of structural region subspace, then ρ=0；σ is variance parameter, σ=3；λ is long and narrow factor parameter, λ=1；θ is direction ginseng
Number generates sketch map according to SAR image sketch model, and the value of θ is the direction of sketch line in sketch map；For first direction function, it is expressed as follows:
Wherein, (nxs, nys) it is pixel ysCoordinate, (nxt,nyt) it is pixel ytCoordinate；
For second direction function, it is expressed as follows:
Step 4, empty to structural region subspace and homogenous region using the objective function of semantic facility random field models
Between be split.
To each pixel of structural region subspace and homogenous region subspace, the category feature Probability p of pixel is calculated
(ys), likelihood probability p (ys|xs) and prior probability p (xs,xt|ψst(ys,yt,φ(ys,yt))), then this 3 probability multiplications are obtained
To the value of objective function, the maximum value of objective function is taken, obtains the category of each pixel are as follows:
Step 5, the segmentation knot of aggregated structure atural object subspace, structural region subspace and homogenous region subspace will be mixed
Fruit merges, and obtains the segmentation result of SAR image.
To the category x of each pixel in mixing aggregated structure atural object subspacea, structural region subspace and homogenous region are sub
The category x of each pixel in spaces, take xaWith xsUnion, obtain the segmentation result of SAR image.
Advantages of the present invention is further illustrated by the data and image of following emulation.
1. simulated conditions
The hardware condition that the present invention emulates are as follows: Intellisense and image understanding laboratory graphics workstation；
The present invention emulate used in SAR image are as follows: the SAR image and C-band resolution ratio that Ku wave band resolution ratio is 1 meter be
3 meters of SAR image.
2. emulation content and result
Emulation 1: the SAR image for being 1 meter using Ku wave band resolution ratio, with the present invention and existing Markov random field mould
Type and conditional random field models are split SAR image, and as a result such as Fig. 3, it is 1 meter that wherein Fig. 3 (a), which is Ku wave band resolution ratio,
Original SAR image, Fig. 3 (b) are the segmentation result of Markov random field model, and Fig. 3 (c) is the segmentation of conditional random field models
As a result, Fig. 3 (d) is segmentation result of the invention.
Emulation 2: the SAR image for being 3 meters using C-band resolution ratio, with the present invention and existing Markov random field mould
Type and conditional random field models are split SAR image, and as a result such as Fig. 4, it is 3 meters that wherein Fig. 4 (a), which is C-band resolution ratio,
Original SAR image, Fig. 4 (b) are the segmentation result of Markov random field model, and Fig. 4 (c) is the segmentation of conditional random field models
As a result, Fig. 4 (d) is segmentation result of the invention.
Simulation result: from figs. 3 and 4 it can be seen that Markov random field model is to image boundary and detailed information
It keeps preferably, but results in over-segmentation phenomenon, region consistency is poor；Method based on conditional random field models cannot retain
The detailed information of SAR image, this is because conditional random field models cannot capture the anisotropic relationship of SAR image；The present invention
The anisotropic relationship of SAR image is captured by semantic facility random field models, segmentation result not only has preferable region one
Cause property and the detailed information for remaining image.
In conclusion the present invention realizes the holding of region consistency and detailed information in SAR image segmentation simultaneously, obtain
SAR image good segmentation effect.
The part that the present embodiment does not specifically describe belongs to the common knowledge and well-known technique of the art, if any need
Us are wanted to can provide reference！The foregoing examples are only illustrative of the present invention, does not constitute to protection of the invention
The limitation of range, it is all with the present invention it is the same or similar design all belong to the scope of protection of the present invention within.
Claims (5)
1. the SAR image segmentation method based on semantic facility random field models, characterized by the following steps:
(1) according to the administrative division map of SAR image, SAR image is divided into mixing aggregated structure atural object subspace, structural region sky
Between and homogenous region subspace；
(2) to mixing aggregated structure atural object subspace, the feature in region is extracted using bag of words, and is clustered with affine propagation AP
Method the atural object subspace is split；
(3) semantic facility random field models are constructed:
(3a) defines unitary potential function are as follows:
Wherein, us(xs,ys) it is unitary potential function, Z={ 1,2 ..., N } is the set of entire SAR image pixel, and N is SAR figure
As the total number of pixel；ysIt is s-th of pixel in SAR image, s ∈ Z；xsIt is the category of s-th of pixel in SAR image；fs(ys)
It is the category feature function of pixel, p (ys) be pixel category feature probability, p (ys|xs) it is likelihood probability；K is image category
Total number, l ∈ { 1,2 ..., K }, δ (xs, l) and it is the first indicator function, if xs=l, then δ (xs, l)=1, if xs≠
L, then δ (xs, l)=0；
(3b) defines binary potential function are as follows:
Wherein, ytIt is t-th of pixel in SAR image, t ∈ Z, and s ≠ t；xtIt is the category of t-th of pixel in SAR image；l∈
{ 1,2 ..., K }, m ∈ { 1,2 ..., K }, δ (xt, m) and it is the second indicator function, if xt=m, then δ (xt, m)=1, if xt
≠ m, then δ (xt, m)=0；ψst(ys,yt,φ(ys,yt)) it is mixed kernel function, φ (ys,yt) it is pixel ysWith pixel ytBetween
Semantic function, p (xs,xt|ψst(ys,yt,φ(ys,yt))) it is prior probability；
(3c) according to (3a) and (3b), the objective function for defining semantic facility random field models is as follows:
p(xs|ys)=p (ys)×p(ys|xs)×p(xs,xt|ψst(ys,yt,φ(ys,yt))) <3>
Wherein, p (ys) be pixel category feature probability, p (ys|xs) it is likelihood probability, p (xs,xt|ψst(ys,yt,φ(ys,
yt))) it is prior probability；
(4) using the objective function of (3c) obtained semantic facility random field models to structural region subspace and homogenous region
Space is split, i.e. each pixel to structural region subspace and homogenous region subspace, and the maximum value of modus ponens<3>obtains
To the category of each pixel are as follows:
(5) segmentation result for mixing aggregated structure atural object subspace, structural region subspace and homogenous region subspace is carried out
Merge, obtains the segmentation result of SAR image.
2. according to the method described in claim 1, the method wherein in step (2) with affine propagation AP cluster is empty to atural object
Between be split, as follows carry out:
To each pixel y in mixing aggregated structure atural object subspacea, It is mixing aggregation knot
The total number of structure atural object subspace pixel obtains pixel y using the method that AP is clusteredaCategory xa。
3. according to the method described in claim 1, the wherein category feature Probability p (y of the pixel in step (3c) formulas), it indicates
It is as follows:
Wherein, wlIt is the weight parameter of pixel,ηsIt is pixel ysNeighborhood territory pixel.
4. according to the method described in claim 1, the wherein likelihood probability p (y in step (3c) formulas|xs), it is expressed as follows:
Wherein, Γ (αk) it is gamma function, αkFor scale parameter,μkIt is displacement parameter,
5. according to the method described in claim 1, the wherein prior probability in step (3c) formula, p (xs, xt|ψst(ys,yt,φ
(ys,yt))), it is expressed as follows:
Wherein, ψst(ys,yt,φ(ys,yt)) it is mixed kernel function, it is expressed as follows:
Wherein, ρ is weight parameter, ρ ∈ { 0,1 }, if ysIt is the pixel of homogenous region subspace, then ρ=1, if ysIt is structure
The pixel of region subspace, then ρ=0；σ is variance parameter, σ=3；λ is long and narrow factor parameter, λ=1；θ is directioin parameter, root
Sketch map is generated according to SAR image sketch model, the value of θ is the direction of sketch line in sketch map；For
First direction function, is expressed as follows:
Wherein, (nxs, nys) it is pixel ysCoordinate, (nxt,nyt) it is pixel ytCoordinate；
For second direction function, it is expressed as follows:
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201611237232.4A CN106683109B (en) | 2016-12-28 | 2016-12-28 | SAR image segmentation method based on semantic facility random field models |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201611237232.4A CN106683109B (en) | 2016-12-28 | 2016-12-28 | SAR image segmentation method based on semantic facility random field models |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN106683109A CN106683109A (en) | 2017-05-17 |
| CN106683109B true CN106683109B (en) | 2019-05-03 |
Family
ID=58871950
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201611237232.4A Active CN106683109B (en) | 2016-12-28 | 2016-12-28 | SAR image segmentation method based on semantic facility random field models |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN106683109B (en) |
Families Citing this family (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107977624A (en) * | 2017-11-30 | 2018-05-01 | 国信优易数据有限公司 | A kind of semantic segmentation method, apparatus and system |
| CN109345537B (en) * | 2018-08-17 | 2021-02-12 | 西安电子科技大学 | SAR image segmentation method based on high-order multi-scale CRF semi-supervision |
| CN109272515B (en) * | 2018-08-17 | 2020-10-09 | 西安电子科技大学 | Unsupervised SAR image segmentation method based on high-order multi-scale CRF |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103198479A (en) * | 2013-03-27 | 2013-07-10 | 西安电子科技大学 | SAR image segmentation method based on semantic information classification |
| CN104036491A (en) * | 2014-05-14 | 2014-09-10 | 西安电子科技大学 | SAR image segmentation method based on area division and self-adaptive polynomial implicit model |
| CN104346814A (en) * | 2014-11-20 | 2015-02-11 | 西安电子科技大学 | SAR (specific absorption rate) image segmentation method based on hierarchy visual semantics |
-
2016
- 2016-12-28 CN CN201611237232.4A patent/CN106683109B/en active Active
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103198479A (en) * | 2013-03-27 | 2013-07-10 | 西安电子科技大学 | SAR image segmentation method based on semantic information classification |
| CN104036491A (en) * | 2014-05-14 | 2014-09-10 | 西安电子科技大学 | SAR image segmentation method based on area division and self-adaptive polynomial implicit model |
| CN104346814A (en) * | 2014-11-20 | 2015-02-11 | 西安电子科技大学 | SAR (specific absorption rate) image segmentation method based on hierarchy visual semantics |
Non-Patent Citations (5)
| Title |
|---|
| Hierarchical Conditional Random Fields Model for Semisupervised SAR Image Segmentation;Peng Zhang et al.;《IEEE Transactions on Geoscience and Remote Sensing》;20150930;第53卷(第9期);第4933-4951页 |
| Semantic Segmentation of Polarimetric SAR Imagery Using Conditional Random Fields;Wen Yang et al.;《IGARSS 2010》;20101231;第1593-1596页 |
| Unsupervised SAR Image Segmentation Based on Conditional Triplet Markov Fields;Xiaojie Lian et al.;《IEEE Geoscience and Remote sensing Letters》;20140731;第11卷(第7期);第1185-1189页 |
| 基于统计模型的SAR图像降斑和分割方法研究;张鹏;《中国博士学位论文全文数据库 信息科技辑》;20131115;正文全文 |
| 面向图像标记的条件随机场模型研究;徐丽;《中国有些博士学位论文全文数据库 信息科技辑》;20140615;正文全文 |
Also Published As
| Publication number | Publication date |
|---|---|
| CN106683109A (en) | 2017-05-17 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Wei et al. | Toward automatic building footprint delineation from aerial images using CNN and regularization | |
| CN115861135B (en) | Image enhancement and recognition method applied to panoramic detection of box body | |
| Marques et al. | SAR image segmentation based on level set approach and {\cal G} _A^ 0 model | |
| Xue et al. | Median-based image thresholding | |
| WO2018023734A1 (en) | Significance testing method for 3d image | |
| CN101493520B (en) | SAR image variation detecting method based on two-dimension gamma distribution | |
| CN109446894B (en) | Multispectral image change detection method based on probability segmentation and Gaussian mixture clustering | |
| CN106651865B (en) | A kind of optimum segmentation scale automatic selecting method of new high-resolution remote sensing image | |
| CN106991686B (en) | A kind of level set contour tracing method based on super-pixel optical flow field | |
| CN106683109B (en) | SAR image segmentation method based on semantic facility random field models | |
| EP3073443A1 (en) | 3D Saliency map | |
| CN102945378A (en) | Method for detecting potential target regions of remote sensing image on basis of monitoring method | |
| CN109583493A (en) | A kind of credit card detection and digit recognition method based on deep learning | |
| CN111666856A (en) | High-resolution single-polarization SAR image building target detection method based on structural characteristics | |
| CN104346814B (en) | Based on the SAR image segmentation method that level vision is semantic | |
| CN112699717A (en) | SAR image generation method and generation device based on GAN network | |
| Zafari et al. | Resolving overlapping convex objects in silhouette images by concavity analysis and Gaussian process | |
| Khandare et al. | A survey paper on image segmentation with thresholding | |
| Liang et al. | Adaptive multiple kernel fusion model using spatial-statistical information for high resolution SAR image classification | |
| Mehltretter et al. | Aleatoric uncertainty estimation for dense stereo matching via CNN-based cost volume analysis | |
| Zhang et al. | Segmentation algorithm of complex ore images based on templates transformation and reconstruction | |
| CN107862701B (en) | Remote sensing image segmentation method based on markov random file and mixed kernel function | |
| Daudt et al. | Learning to understand earth observation images with weak and unreliable ground truth | |
| Zhou et al. | Estimating the size of densely packed and complex grains using valley edge detection with a one-pixel width and grain shape information | |
| Frosio et al. | Adaptive segmentation based on a learned quality metric. |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |