CN105513060A - Visual perception enlightening high-resolution remote-sensing image segmentation method - Google Patents
Visual perception enlightening high-resolution remote-sensing image segmentation method Download PDFInfo
- Publication number
- CN105513060A CN105513060A CN201510862868.7A CN201510862868A CN105513060A CN 105513060 A CN105513060 A CN 105513060A CN 201510862868 A CN201510862868 A CN 201510862868A CN 105513060 A CN105513060 A CN 105513060A
- Authority
- CN
- China
- Prior art keywords
- image
- gradient
- texture
- segmentation
- filtering
- 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
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
Abstract
The invention discloses a visual perception enlightening high-resolution remote-sensing image segmentation method, and the method enables an object boundary in an image to be divided into an intensity boundary corresponding to a spectrum homogeneous area, and a texture boundary corresponding to a spectrum change area, and respectively extracts two types of visual information which serve as the main basis of segmentation. The method comprises the steps: filtering noise and texture information in the image through a nonlinear filtering method, and obtaining the intensity gradient through a gradient operator; analyzing the texture features of the image through employing a Gabor filter, enabling the filtering output of a plurality of channels to be merged and processed through the gradient operator, and obtaining the texture gradient of the image; carrying out the fusion of the intensity gradient and the texture gradient, carrying out the conversion of the gradient image after fusion through watershed conversion, and achieving the segmentation of the image. Compared with the prior art, the method improves the boundary accuracy of image segmentation, reduces the over-segmentation and under-segmentation phenomena, and can be effectively used in the field of high-resolution remote-sensing image information processing.
Description
Technical field
The invention belongs to technical field of image processing, further relate to the dividing method of a kind of analog vision perception mechanism in high-resolution remote sensing image segmentation technology.The present invention is mainly used in the segmentation of high-resolution satellite remote sensing images, aerial remote sensing images, to reach the object of ground object information extraction in image.
Background technology
High-resolution remote sensing image can provide abundant ground object details, but too increase the inside SPECTRAL DIVERSITY of homogeneous region and present various texture features, in addition, sensor can introduce noise in imaging process, each of which increases the difficulty that remote sensing images are accurately split.In recent years, high-resolution remote sensing image is mainly based on graphical analysis (GEOBIA) method towards geographic object, first be the region of non-overlapping copies by Iamge Segmentation, be called object, the accuracy of object bounds determines subsequent characteristics and extracts and the quality of classification.In order to produce object, need by means of image partition method, the current multi-resolution segmentation of employing more than in GEOBIA algorithm, it is a kind of region growing method considering object shapes and spectral information in essence, the shortcoming of this method is that yardstick is correlated with, and need multiple scale parameter comprehensively to select excellent, and cutting procedure does not consider texture factor, be difficult to obtain object bounds accurately describe, easily produce over-segmentation and less divided phenomenon.
Show the research of visual perception mechanism, scattered, single visual information is mainly organized into the process of large unit, significant object and their mutual relationship by the early process of Vision information processing.Here object refers to the region possessing consistent visual information (spectral intensity, texture etc.), according to visual masking effect, when there is complex texture in image-region, HVS is insensitive to the visual signal distortion in this region, is also considered as one " entirety " by this region.The development trend of present image segmentation is the decipher process of simulating human vision system (HVS) to image, consider various visual information, as edge, spectral intensity, texture and spatial relationship attribute etc., be the not overlapping subregion corresponding with real world object by Iamge Segmentation, how to simulate the process of HVS system to image information better in algorithm aspect, be a urgently open question.
The deficiency of spectral information is only considered in order to improve high-resolution remote sensing image cutting procedure, some researchs attempt more visual information to be incorporated into cutting procedure, as added marginal information to improve region merging technique effect, marginal information is utilized to carry out watershed divide mark and segmentation, considering texture information when splitting, texture feature extraction, merging the composite character of spectral information and to represent and research is launched in the aspect such as region growing merging.But these methods mainly adopt competition mechanism to determine the gradient information of image, only a category information is adopted in the deterministic process of border, with there is texture gradient and spectral intensity gradient contradiction in natural image simultaneously, analyze image process with HVS not to be inconsistent, also there is the border of texture object on the border that in high-resolution remote sensing images, between adjacent objects, existing spectral intensity difference produces, too much consideration texture easily produces Texture Boundaries effect on spectral intensity border, and the noise in image is easy to cause over-segmentation phenomenon and the inaccurate problem of boundary alignment.
Summary of the invention
The present invention institute, for above-mentioned the deficiencies in the prior art, proposes a kind of method for segmentation of high resolution remote sensing image of analog vision perception mechanism.Texture region and spectrum homogeneous area can be considered as entirety to process this characteristic according to human visual system by method, extract the border of spectrum homogeneous area and the texture region border of spectrum change in image respectively, then the two effective integration is realized the segmentation of image, solve in existing remote sensing image segmentation method the problem such as use single spectral information, boundary alignment inaccurate, improve the segmentation precision of remote sensing segmentation, reduce the post-processed complexity in sensor information leaching process.
The thinking that the present invention realizes above-mentioned purpose is: extract the spectrum homogeneous area in image and texture region respectively, first by means of the nonlinear filter method with edge retention performance, input picture is processed, noise in filtering image and texture information, obtain the intensity gradient of spectrum homogeneous area under the effect of gradient operator; Then use and the Gabor filter group of analog vision receptive field direction and frequency selective characteristic can extract the texture energy feature of image, these textures are exported after carrying out suitable Filtering and smoothing process, obtain texture region information, and then obtain texture gradient image, use watershed transform to obtain final segmentation result after being merged by two class boundary informations.
Basic step of the present invention is:
Step S1, extracts the intensity gradient of image, and adopt the non-linear filtering method with edge retention performance, the noise in filtering image and texture information, obtain intensity gradient image by means of gradient detective operators;
Step S2, extracts the texture gradient of image, uses Gabor filter group to extract different directions and texture energy image corresponding to wavelength, carries out filtering, smoothing processing, obtain the texture gradient of each passage by means of gradient detective operators to output image;
Step S3, intensity gradient and texture gradient are merged, first expansive working is carried out to each subchannel texture gradient, then cumulative sum normalization is carried out, and be added with the intensity gradient after normalization, obtain final fusion gradient image, watershed algorithm is used to fusion gradient image, obtains final segmentation result.
The nonlinear filtering wave pattern used in described step S1 is for improving bilateral filtering model, and the expression formula of bilateral filtering model is:
Wherein, I is input picture, f is output image, and p, q are location of pixels, and η () represents Neighbourhood set operator, w () is space length weight function, g () for spectral value difference weight function, the g () in traditional two-sided filter be Gaussian function, in the present invention, use Tukey ' sbiweight function to replace Gaussian function, expression formula is:
In formula, σ is level and smooth scale parameter, uses σ
grepresent overall scale parameter, this parameter is obtained by the spectral distribution property of entire image, and use median absolute deviation to represent here, expression formula is:
Wherein,
for gradient operator, median is median operator.Adopt the level and smooth scale parameter of the overall situation to have ignored local difference interregional in remote sensing images, especially at the boundary that texture and non-grain have a common boundary, use partial gradient statistics intermediate value to represent local scale parameter, to distinguish texture and edge, expression formula is for this reason:
Then,
constant is wherein Grads threshold correction.
In described step S2, use the real part of Gabor function to extract texture energy feature, the Gabor image filtering model in m wavelength and the n-th direction is:
Wherein, according to 1 octave wavelength, adjacent filter wavelength meets the multiple proportion of 2, is simultaneously the character of π/6 according to the direction half amplitude responder of mammiferous V1 cell to bandwidth, gets half amplitude pattern bandwidth π/6 of Gabor filter, then obtain
In formula, h () is Gabor filter function, and λ is the wavelength of Gabor filter, and θ is the anglec of rotation.
In described step S2, in order to eliminate the texture leakage effect each passage being exported to texture image, use the improvement two-sided filter in step S1 to carry out filtering to texture image, the filter window range set of two-sided filter is 2 times of Gabor filter wavelength.
In described step S2, the texture of each passage exports at spectral distribution homogeneous area intersection along with Texture Boundaries effect, namely there is Multiple edge phenomenon in the border of objects in images, for eliminating Texture Boundaries effect, adopt median filter method to process result, the window width of medium filtering is set as the Gabor filter wavelength width of 4 times.
In described step S3, before texture gradient and intensity gradient merge, pre-service is carried out to texture gradient.At image texture region and spectrum homogeneous area intersection, the expression of texture gradient and intensity gradient can produce inconsistent phenomenon, easily generate border object in watershed transform process, the method for this problem of process of the present invention carries out expansive working to texture gradient image, and formula is:
Texture Boundaries after expansion is unaffected in segmentation result, and meanwhile, texture and homogenous area intersection, based on intensity gradient information, ensure that the accuracy of segmentation result.
In described step S3, the warm method of the intensity gradient IG after normalization and the texture gradient TG after normalization is carried out according to the following formula:
Wherein, (x, y) is pixel coordinate in image, and median is median operator, and HG is that the final gradient after merging represents.
In described step S3, use standard watershed transform to realize the segmentation of image, the minimizing impact in order to avoid region, introducing the h-minimal value suppressing method minimal value that decapacitation causes redundancy split of making a return journey affects.
The invention has the beneficial effects as follows:
(1) consider spectral information and texture information simultaneously, with tradition only based on spectrum dividing method compared with, better can split texture region, reduce the problem of over-segmentation in texture region;
(2) extract the information of spectrum homogeneous area border and Texture Boundaries respectively, with the image interpretation similar process of visually-perceptible system, higher object bounds Position location accuracy can be obtained;
(3) textural characteristics processing procedure adopt nonlinear filtering and median filter method, enhance Texture Boundaries extract validity;
(4) present invention reduces the scales dependence in remote sensing image information interpretation process and select complexity, reducing over the problem of segmentation and less divided, can be well used in high-resolution remote sensing image process field.
Accompanying drawing explanation
Fig. 1 is the overview flow chart of the method for segmentation of high resolution remote sensing image that visually-perceptible proposed by the invention inspires;
Fig. 2 is that the texture gradient in the present invention extracts process flow diagram;
Fig. 3 is based on the segmentation result that different gradients (intensity gradient, texture gradient and fusion gradient) obtains in the present invention;
Fig. 4 is the present invention and existing typical separator algorithm (multi-resolution segmentation algorithm and super-pixel partitioning algorithm) the segmentation result comparison diagram on high-resolution remote sensing image.
Embodiment
Below in conjunction with accompanying drawing and example, the present invention is described in more detail.
Fig. 1 is overview flow chart of the present invention, is described in detail below in conjunction with the implementation procedure of process flow diagram to the present embodiment.
Step S1, extract the intensity gradient of input picture, process is as follows:
(S1-1) the level and smooth scale parameter σ of input high-resolution remote sensing image is calculated, obtain the level and smooth size distribution of image, the value that the window width calculating local scale in the present embodiment is set to 11 to 15 all can reach good effect, and window width is set to 13 here.
(S1-2) use is based on the bilateral filtering method of Tukey ' sbiweight function to the smoothing process of image, and the scale parameter in Tukey ' sbiweight function is the result that step (S1-1) calculates.
For true color remote sensing image (RGB image), bilateral filtering process is implemented respectively at three Color Channels, use the smoothing computing of form of coupling vector, namely when calculating neighborhood spectral differences, same position place pixel value is used as Vector Processing, and the result of spectral differences represents by the norm of the vector differentials of one dimension.For the remote sensing images of multispectral (more than 3 passages), the method for principal component analysis (PCA) or independent component analysis can be used to carry out dimension-reduction treatment, then on 3 major components export, carry out smoothing processing.
(S1-3) use gradient operator to calculate the image gradient smoothly, for the gray level image of single spectrum, directly use Sobel operator compute gradient amplitude as gradient information, to multispectral image, use color gradient detective operators compute gradient amplitude information.The method of gradient operator with reference to the Digital Image Processing translations of Paul Gonzales, can not repeat here.
(S1-4) gradient obtained is normalized, obtains the intensity gradient IG of image.
Step S2, extracts the texture gradient of image, and its flow process is as shown in Figure 2, and process is as follows:
(S2-1) high-resolution remote sensing image is input in each subfilter of Gabor filter group, obtains mn sub-channel filtering and export.The wavelength chooses of wave filter distributes according to the textural characteristics of image, and the texture image for large period gets the Gabor filter of large wavelength value, in the present embodiment, uses the Gabor filtering of front 4 wavelength to export, can meet most of remote sensing image processing demand, namely
the deflection of bank of filters is set to { 0, π/6, π/3, pi/2,2 π/3,5 π/6}.
(S2-2) use the improvement two-sided filter in step S1 to carry out filtering to texture image, the filter window range set in bilateral filtering method is 2 times of Gabor filter wavelength, and in removal of images, the texture of each passage texture information reveals effect.
(S2-3) process the output resume in (S2-2), adopt the Texture Boundaries effect in median filter method filtering image, the window width of medium filtering is set as the Gabor filter wavelength width of 4 times.
(S2-4) to result after the smoothing processing of each passage, use gradient operator compute gradient amplitude, the texture gradient obtaining each passage exports TG
i, wherein i=1 ..., mn.
Step S3, carry out gradient fusion and use watershed transform to complete the segmentation of image, process is as follows:
(S3-1) morphological dilation is carried out to texture gradient image, the radius of the structural element that the expansive working of each subchannel texture gradient uses is set to wavelength corresponding to this texture subchannel, the shape of structural element adopts collar plate shape structure, and after expanding, the texture gradient of each subchannel is DTG
i.
(S3-2) DTG is exported to each passage
ibe normalized and add up, obtain final texture gradient image TG, normalization and totalization formula are expressed as:
(S3-3) intensity gradient IG and texture gradient TG is merged, obtain total gradient and represent HG.
(S3-4) h-minima minimal value conversion process is carried out to HG gradient image, eliminate the degree of depth in gradient image and be less than the region of h.
(S3-5) on gradient HG, use watershed transform, realize the segmentation of image.
Content of the present invention is described further by following simulation result.
(1) content is emulated: emulation is divided into two steps, and the first step verifies rationality and the validity of the inventive method, emulates the segmentation result obtained based on different visual information; Second step verifies the advantage of the inventive method, method of the present invention and current typical partitioning algorithm is compared, and adopts the multi-resolution segmentation algorithm in remote sensing fields and the JSEG dividing method in computer vision field as comparison other in the present embodiment.
(2) the simulation experiment result
Emulation experiment one, on the remote sensing images of 0.38 meter of every pixel resolution, intensity gradient, texture gradient and fusion gradient that the inventive method proposes carry out split-run test respectively, the result of three kinds of gradients represents the segmentation effect based on different visual information, Fig. 3 is the instantiation result figure of the inventive method, and wherein h-minima minimal value is set as 0.015.
In the result, Fig. 3 (a) is the high-resolution remote sensing image of input; The level and smooth size distribution image of Fig. 3 (b) for calculating, in figure, visible texture region correspond to high level and smooth scale parameter, ensures that texture region can by effective smoothing processing; Fig. 3 (c) be bilateral filtering level and smooth after the gradient image that obtains, noise and texture are effectively suppressed, and obtain intensity gradient; The segmentation result that Fig. 3 (d) is intensity gradient, the border of non-grain region as aquiclude etc. in figure is effectively split, and boundary alignment is accurate, and at texture region, as the woods, meadow etc., can not effectively split; Fig. 3 (e) is the texture gradient after morphological dilations; The watershed segmentation of Fig. 3 (f) for realizing in texture gradient, can find out, the limb recognition of texture object is effective, but to the non-grain such as buildings, road edges of regions poor location; Fig. 3 (g) is the gradient after intensity gradient and texture gradient being merged; Fig. 3 (h) is for utilizing the segmentation result merging rear gradient and obtain, good to the limb recognition segmentation effect of object in figure, its segmentation result combines the advantage of intensity gradient and texture gradient, is obviously better than the gradient dividing method only adopting a kind of visual information.
Emulation experiment two, in the city remote sensing image of 0.3 meter of every pixel resolution, multiresolution method, JSEG (JointSystemsEngineeringGroup) method and the inventive method is used to carry out split-run test respectively, limit the cut zone that each method produces equal number, the result of comparative approach, obtains the instantiation result figure of Fig. 4.
Three kinds of methods generate 250 and 500 cut zone respectively in the diagram, and Fig. 4 (a) is multi-resolution segmentation method 250 region segmentation result; Fig. 4 (b) is multi-resolution segmentation method 500 region segmentation result; Fig. 4 (c) is JSEG method 250 region segmentation result; Fig. 4 (d) is JSEG method 500 region segmentation result; Fig. 4 (e) is the inventive method 250 region segmentation result; Fig. 4 (f) is the inventive method 500 region segmentation result.
As seen from Figure 4, many resolutions dividing method easily produces over-segmentation in atural object edge, and boundary alignment accuracy is lower, JSEG method also can produce over-segmentation phenomenon in non-grain region after increase cut zone, when the inventive method cut zone increases, only can increase the segmentation of texture region, the boundary alignment accuracy of the inventive method is apparently higher than multi-resolution segmentation and JSEG method simultaneously, and over-segmentation and less divided effect minimum in three kinds of methods.
Below by reference to the accompanying drawings the specific embodiment of the present invention is described; but the not restriction of scope; those skilled in the art do not need to pay to the present invention the various amendment or distortion that creative work can make, and are all encompassed within protection scope of the present invention.
Claims (7)
1. the method for segmentation of high resolution remote sensing image of a visually-perceptible inspiration, it is characterized in that, Texture Boundaries two class corresponding to the intensity border being divided into spectrum homogenous area corresponding according to vision system mechanism the information in image and spectrum change region, then two kinds of visual informations in image are extracted respectively, and using gradient to represent border, method comprises the steps:
Step S1, extracts the intensity gradient of image, and adopt the non-linear filtering method with edge retention performance, the noise in filtering image and texture information, obtain intensity gradient image by means of gradient detective operators;
Step S2, extract the texture gradient of image, Gabor filter group is used to extract different directions and texture energy image corresponding to wavelength, filtering, smoothing processing are carried out to output image, obtain the texture gradient of each passage by means of gradient detective operators, then these gradients are added and obtain the total texture gradient of image;
Step S3, is normalized rear addition respectively by intensity gradient and texture gradient, obtains final fusion gradient image, uses watershed algorithm, obtain final segmentation result to fusion gradient image.
2. image partition method according to claim 1, it is characterized in that, in described step S1, adopted non-linear filtering method be improve two-sided filter method, the filtering core function of this wave filter uses Tukey ' sbiweight function, scale parameter in function uses the maximal value in the neighborhood intermediate value of image pixel and overall deviation median to represent, the corresponding scale parameter of each pixel.
3. image partition method according to claim 1, it is characterized in that, in described step S2, the wild feature of the mammiferous visual experience of Setting Up Parameters of Gabor filter group, the design of subfilter adopts 1 octave wavelength, adjacent filter wavelength meets the multiple proportion of 2, and minimum wavelength is set to
meanwhile, direction half amplitude bandwidth is set to π/6, deflection is chosen { 0, π/6, π/3, pi/2,2 π/3,5 π/6}.
4. image partition method according to claim 1, it is characterized in that, in described step S2, use the improvement two-sided filter in step S1 to carry out filtering to texture image, eliminate texture and reveal effect, the filter window range set of two-sided filter is 2 times of Gabor filter wavelength.
5. image partition method according to claim 1, it is characterized in that, in described step S2, adopt median filter method to eliminate the Multiple edge phenomenon of texture object border appearance, the window width of medium filtering is set as the Gabor filter wavelength width of 4 times.
6. image partition method according to claim 1, is characterized in that, in described step S3, gradient carries out morphological dilation to texture gradient before merging.
7. image partition method according to claim 1, is characterized in that, in described step S3, the warm method of intensity gradient IG and texture gradient TG is carried out according to the following formula:
Wherein, (x, y) is pixel coordinate in image, and median is median operator, and HG is the gradient after merging.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201510862868.7A CN105513060A (en) | 2015-11-30 | 2015-11-30 | Visual perception enlightening high-resolution remote-sensing image segmentation method |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201510862868.7A CN105513060A (en) | 2015-11-30 | 2015-11-30 | Visual perception enlightening high-resolution remote-sensing image segmentation method |
Publications (1)
Publication Number | Publication Date |
---|---|
CN105513060A true CN105513060A (en) | 2016-04-20 |
Family
ID=55721017
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201510862868.7A Pending CN105513060A (en) | 2015-11-30 | 2015-11-30 | Visual perception enlightening high-resolution remote-sensing image segmentation method |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN105513060A (en) |
Cited By (6)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN108024795A (en) * | 2016-07-05 | 2018-05-11 | 株式会社日立制作所 | Frequency spectrum resolver, frequency spectrum analytic method and ultrasonic imaging apparatus |
CN108171258A (en) * | 2017-12-12 | 2018-06-15 | 西安电子科技大学 | Independent component analysis Multiple Shape prior level diversity method, image segmentation system |
CN108596139A (en) * | 2018-05-03 | 2018-09-28 | 武汉大学 | A kind of remote sensing image urban area extracting method based on Gabor characteristic conspicuousness |
CN109919960A (en) * | 2019-02-22 | 2019-06-21 | 西安工程大学 | A kind of image continuous boundary detection method based on Multiscale Gabor Filters device |
CN110309781A (en) * | 2019-07-01 | 2019-10-08 | 中国科学院遥感与数字地球研究所 | Damage remote sensing recognition method in house based on the fusion of multi-scale spectrum texture self-adaption |
CN113902759A (en) * | 2021-10-13 | 2022-01-07 | 自然资源部国土卫星遥感应用中心 | Space-spectrum information combined satellite-borne hyperspectral image segmentation and clustering method |
Citations (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN101510310A (en) * | 2009-02-19 | 2009-08-19 | 上海交通大学 | Method for segmentation of high resolution remote sensing image based on veins clustering constrain |
CN102024258A (en) * | 2010-12-15 | 2011-04-20 | 中国科学院遥感应用研究所 | Multi-scale segmentation method for remote sensing image with boundary maintenance characteristics |
CN102496151A (en) * | 2011-12-08 | 2012-06-13 | 南京大学 | Method for multi-scale segmentation of high-resolution remote sensing images |
-
2015
- 2015-11-30 CN CN201510862868.7A patent/CN105513060A/en active Pending
Patent Citations (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN101510310A (en) * | 2009-02-19 | 2009-08-19 | 上海交通大学 | Method for segmentation of high resolution remote sensing image based on veins clustering constrain |
CN102024258A (en) * | 2010-12-15 | 2011-04-20 | 中国科学院遥感应用研究所 | Multi-scale segmentation method for remote sensing image with boundary maintenance characteristics |
CN102496151A (en) * | 2011-12-08 | 2012-06-13 | 南京大学 | Method for multi-scale segmentation of high-resolution remote sensing images |
Non-Patent Citations (2)
Title |
---|
JIANTING ZHANG 等: "Image Segmentation Using a Hybrid Gradient Based Watershed Transform", 《2013 INTERNATIONAL CONFERENCE ON MECHATRONIC SCIENCE, ELECTRIC ENGINEERING AND COMPUTER(MEC)》 * |
PADRAIG CORCORAN 等: "Complementary texture and intensity gradient estimation and fusion for watershed segmentation", 《MACHINE VISION AND APPLICATIONS》 * |
Cited By (9)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN108024795A (en) * | 2016-07-05 | 2018-05-11 | 株式会社日立制作所 | Frequency spectrum resolver, frequency spectrum analytic method and ultrasonic imaging apparatus |
CN108171258A (en) * | 2017-12-12 | 2018-06-15 | 西安电子科技大学 | Independent component analysis Multiple Shape prior level diversity method, image segmentation system |
CN108596139A (en) * | 2018-05-03 | 2018-09-28 | 武汉大学 | A kind of remote sensing image urban area extracting method based on Gabor characteristic conspicuousness |
CN108596139B (en) * | 2018-05-03 | 2020-05-08 | 武汉大学 | Remote sensing image urban area extraction method based on Gabor feature saliency |
CN109919960A (en) * | 2019-02-22 | 2019-06-21 | 西安工程大学 | A kind of image continuous boundary detection method based on Multiscale Gabor Filters device |
CN110309781A (en) * | 2019-07-01 | 2019-10-08 | 中国科学院遥感与数字地球研究所 | Damage remote sensing recognition method in house based on the fusion of multi-scale spectrum texture self-adaption |
CN110309781B (en) * | 2019-07-01 | 2021-03-02 | 中国科学院空天信息创新研究院 | House damage remote sensing identification method based on multi-scale spectrum texture self-adaptive fusion |
CN113902759A (en) * | 2021-10-13 | 2022-01-07 | 自然资源部国土卫星遥感应用中心 | Space-spectrum information combined satellite-borne hyperspectral image segmentation and clustering method |
CN113902759B (en) * | 2021-10-13 | 2022-04-22 | 自然资源部国土卫星遥感应用中心 | Space-spectrum information combined satellite-borne hyperspectral image segmentation and clustering method |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN105513060A (en) | Visual perception enlightening high-resolution remote-sensing image segmentation method | |
CN103208001B (en) | In conjunction with shape-adaptive neighborhood and the remote sensing image processing method of texture feature extraction | |
CN108492260B (en) | Gelolgical lineament extracting method based on Tensor Voting coupling Hough transformation | |
CN103413151B (en) | Hyperspectral image classification method based on figure canonical low-rank representation Dimensionality Reduction | |
CN101520894B (en) | Method for extracting significant object based on region significance | |
CN102609726B (en) | Method for classifying remote sensing images blended with high-space high-temporal-resolution data by object oriented technology | |
CN102110227B (en) | Compound method for classifying multiresolution remote sensing images based on context | |
CN103839267B (en) | Building extracting method based on morphological building indexes | |
CN104240244B (en) | A kind of conspicuousness object detecting method based on communication mode and manifold ranking | |
CN101510309B (en) | Segmentation method for improving water parting SAR image based on compound wavelet veins region merge | |
CN106250895A (en) | A kind of remote sensing image region of interest area detecting method | |
CN104318569A (en) | Space salient region extraction method based on depth variation model | |
Zheng et al. | Large-scale oil palm tree detection from high-resolution remote sensing images using faster-rcnn | |
CN103198479B (en) | Based on the SAR image segmentation method of semantic information classification | |
CN105956557A (en) | Object-oriented timing sequence remote sensing image cloud coverage area automatic detection method | |
CN103761742B (en) | A kind of high-spectrum remote sensing sparse solution mixing method based on homogeneity index | |
CN103971115A (en) | Automatic extraction method for newly-increased construction land image spots in high-resolution remote sensing images based on NDVI and PanTex index | |
CN106846322B (en) | The SAR image segmentation method learnt based on curve wave filter and convolutional coding structure | |
CN104346608A (en) | Sparse depth map densing method and device | |
CN103927741A (en) | SAR image synthesis method for enhancing target characteristics | |
CN104299232B (en) | SAR image segmentation method based on self-adaptive window directionlet domain and improved FCM | |
CN105184779A (en) | Rapid-feature-pyramid-based multi-dimensioned tracking method of vehicle | |
CN106446936A (en) | Hyperspectral data classification method for spectral-spatial combined data and oscillogram conversion based on convolution neural network | |
CN106228130B (en) | Remote sensing image cloud detection method of optic based on fuzzy autoencoder network | |
CN106683102A (en) | SAR image segmentation method based on ridgelet filters and convolution structure model |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
C06 | Publication | ||
PB01 | Publication | ||
C10 | Entry into substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
WD01 | Invention patent application deemed withdrawn after publication |
Application publication date: 20160420 |
|
WD01 | Invention patent application deemed withdrawn after publication |